Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Dynamic modeling and process design of a membrane enhanced biological phosphorus removal process Al-Atar, Eman 2007

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
831-ubc_2007-317273.pdf [ 39.51MB ]
Metadata
JSON: 831-1.0228857.json
JSON-LD: 831-1.0228857-ld.json
RDF/XML (Pretty): 831-1.0228857-rdf.xml
RDF/JSON: 831-1.0228857-rdf.json
Turtle: 831-1.0228857-turtle.txt
N-Triples: 831-1.0228857-rdf-ntriples.txt
Original Record: 831-1.0228857-source.json
Full Text
831-1.0228857-fulltext.txt
Citation
831-1.0228857.ris

Full Text

D Y N A M I C MODELING A N D PROCESS DESIGN OF A M E M B R A N E ENHANCED BIOLOGICAL PHOSPHORUS R E M O V A L PROCESS B y E M A N A L - A T A R B . A . S c , The University of British Columbia, 1997 M . A . S c . , The University of British Columbia, 2000 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L M E N T OF T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F D O C T O R OF P H I L O S O P H Y in T H E F A C U L T Y OF G R A D U A T E S T U D I E S (Chemical and Biological Engineering) T H E U N I V E R S I T Y OF B R I T I S H C O L U M B I A August 2007 © Eman Al-Atar, 2007 A B S T R A C T The design and operation of the membrane enhanced biological phosphorus removal ( M E B P R ) process was studied in the current research to utilize the utmost capacity of the membrane system for operating the process under high influent loads. The study was conducted in two parts. In the first part, a dynamic model was calibrated to predict data collected from the U B C M E B P R pilot plant. Then the calibrated model was utilized in simulation studies to develop guidelines for the design and operation of a UCT-type M E B P R process under high flowrates without jeopardizing the effluent quality. The Technical University of Delft model combined with A S M 2 d model (TUDP) which is developed for conventional biological phosphorus systems was found sufficient to describe the process behavior of the M E B P R process. The trend of the measured concentration profiles were reasonably predicted, but the exact concentration values for the anoxic nitrate and the effluent ortho-phosphate were not predicted. The calibrated model for the M E B P R process was able to predict the measured data collected from the U B C conventional enhanced biological phosphorus removal ( C E B P R ) process without changing any of the model parameters except for the rate of poly-phosphate formation, kpp, which was increased from 0.1 to 0.2 g P/(g C O D • d) to better predict the anoxic ortho-phosphate concentrations. Simulation studies for the UCT-type M E B P R process showed that the sludge mass distribution in the bioreactor zones of the anaerobic and the aerobic zone are critical for the bio-P removal and the nitrification processes respectively. Appropriate design of the bioreactor zone volumes is important to ensure proper sludge mass distribution in the biological zones. A constant influent volatile fatty acid to total phosphorus concentration was also found important for an efficient bio-P removal process. The aerobic recycle flow was found to be most important for reducing the effluent nitrate concentration while minimizing nitrate leakage to the anaerobic zone. Based on the experimental results and the simulation studies carried out in the current project, a set of guidelines for the design and operation of a UCT-type M E B P R process and the application of process control were i i developed to achieve stable process performance for' nutrient removal under high flowrate operation. i i i Table of Contents Abstract " Table of Contents iv List of Tables viii List of Figures x List of Abbreviations xix Preface xxiii Acknowledgments xxiv 1 INTRODUCTION 1 1.1 Background 1 1.2 Literature Review 3 1.2.1 Enhanced Biological Phosphorus Removal Process 3 1.2.1.1 Involved Mechanisms 3 1.2.1.2 Process Description 6 1.2.1.3 Key Parameters Affecting E B P R Process 8 1.2.2 Modeling of Enhanced Biological Phosphorus Removal Process 11 1.2.2.1 Mechanistic and Black-Box Modeling 12 1.2.2.2 Mechanistic Models for the E B P R Process 12 1.2.2.3 The Technical University of Delft Phosphorus Model 13 1.2.2.4 Applications of T U D P Model 14 1.2.2.5 Data Reconciliation for Activated Sludge Modeling 21 1.2.3 Membrane Bio-Reactor System 22 1.2.3.1 Advantages of M B R System 23 1.2.3.2 Impacts of Membrane Process on Bioreactor Stage 24 1.2.3.3 M B R Applications in E B P R Process 25 1.2.3.4 Modeling of M B R Systems 26 1.2.4 Calibration of Dynamic Models for Activated Sludge Systems 28 1.2.4.1 Model Calibration Challenges 29 1.2.4.2 Obtaining Real Dynamic Data in Activated Sludge Systems 30 1.2.4.3 Dynamic Calibration Techniques for Activated Sludge Systems ... 33 iv 1.2.4.4 Model Validation for Activated Sludge Systems 35 1.2.5 Process Design and Control Studies in E B P R Systems 36 1.2.5.1 Control Challenges of Unit Operation in E B P R Process 37 1.2.5.2 Process Control Applications in Activated Sludge Systems 40 1.3 Research Objectives 45 2 MATERIALS AND METHODS 48 2.1 U B C Wastewater Treatment Pilot Plant Process 48 2.2 Process Operating Conditions 49 2.3 Sampling and Analysis 56 2.4 Quality Assurance and Quality Control of Measurements 57 2.5 T U D P Model 59 2.6 Influent Characterization 59 2.6.1 Influent Characterization of C O D Fractions 59 2.6.2 Characterization of N and P Fractions in the Influent 61 2.7 Model Set-up in A Q U A S I M 2.0 63 2.7.1 M E B P R Process Model in A Q U A S I M 63 2.7.2 C E B P R Process Model in A Q U A S I M 64 3 STEADY STATE MODELING OF THE MEBPR USING THE TUDP MODEL 65 3.1 M E B P R and C E B P R Process Data and Behavior 65 3.2 Data Set Selection for the Steady State Modeling of M E B P R Process 74 3.3 Error Diagnosis and Data Reconciliation of the M E B P R Process Data 77 3.4 Steady State Model Calibration Results for the M E B P R Process 82 3.5 Comparison to Steady State Modeling Results of C E B P R Process 86 4 DYNAMIC MODELING OF THE MEBPR PILOT PLANT PROCESS AT UBC USING THE TUDP MODEL 89 4.1 Application of a New Practical Protocol for the Dynamic Modeling of Activated Sludge Systems for the Purpose of Process Control and and Optimization Studies 89 4.1.1 Dynamic Experimental Design 91 v 4.1.2 Dynamic Data Collection and Screening 96 4.1.3 Dynamic Simulation Using the T U D P Model Parameters Determined During the Steady State Calibration of the M E B P R Process 104 4.1.4 Selection o f the T U D P Model Parameters to be Used for Calibration in the Dynamic Modeling of the M E B P R Process 108 4.1.5 Parameter Calibration 120 4.1.5.1 Manual Model Parameter Estimation Method 122 4.1.5.2 Parameter Estimation in A Q U A S I M 2.0 134 4.1.6 Residual Analysis 139 4.1.7 Model Validation 147 4.2 Effectiveness of the Util ized Dynamic Modeling Protocol 154 4.2.1 Steady State versus Dynamic Modeling 154 4.2.2 Steady State Modeling 155 4.2.3 Experimental Design and Dynamic Data 155 4.2.4 Influent Characterization 157 4.2.5 Initial Biomass Characterization 158 4.2.6 Model Calibration 158 4.2.7 Summary 160 4.3 Comparison of Dynamic Modeling Results of M E B P R to C E B P R Process .. 161 4.4 T U D P Model Prediction Capabilities 170 5 DEVELOPMENT OF GUIDELINES FOR MEBPR PROCESS DESIGN AND OPERATION USING THE SIMULATION MODEL 173 5.1 Summary of Experimental Results for the Effect of Different Operating Conditions on Process Performance for the U B C M E B P R Pilot Plant Process 176 5.2 T U D P Model Predictions for the Effect of Different Design and Operating Conditions on M E B P R Process Behavior 178 5.2.1 Effect o f Biological Mass Distribution 184 5.2.2 Effect of SRT 188 5.2.3 Effect o f Recycle Flows 191 5.2.4 Effect of Temperature 194 5.2.5 Effect of Influent V F A / T P Ratio 198 5.3 Recommended Design Conditions Based on Simulation Results 204 5.4 Proposed Guidelines for the Design and Operation of a UCT-type M E B P R Process 209 5.5 Proposed Guidelines for the Application of Process Control for the Operation of a UCT-type M E B P R Process 211 5.6 Summary 212 6 CONCLUSIONS AND DIRECTIONS FOR FUTURE RESEARCH 213 6.1 Conclusions 213 6.1.1 Dynamic Calibration of Activated Sludge Models 214 6.1.2 Modeling o f M E B P R and C E B P R Processes 215 6.1.3 Guidelines for M E B P R Process Design and Operation 215 6.2 Research Significance 216 6.3 Research Needs 217 REFERENCES 220 APPENDIX I: TUDP Model Rate Equations and Parameters 230 v i i List of Tables Table 2.1. Design specifications of the biological zones of the U B C wastewater treatment pilot plant 49 Table 2.2. Experimental operating conditions for the C E B P R and M E B P R Processes 53 Table 2.3. Sample collection and analysis during the course of the experiment 57 Table 2.4. Coefficient of variation of measurement data 58 Table 2.5. Results of Q A / Q C for TP and T K N measurements 58 Table 2.6. Influent concentration ranges of different components 62 Table 3.1. Date range used for averaging the concentration data for the steady state modeling o f the M E B P R process 75 Table 3.2. Error diagnosis and data reconciliation of averaged flow and total phosphorus measurements 81 Table 3.3. Results of the error diagnosis and data reconciliation of averaged flow and total phosphorus measurements 81 Table 3.4. T U D P model parameters used for the M E B P R process for fitting the averaged measurements 84 Table 3.5. Averaged measurements and simulation results of the calibrated T U D P model using pseudo-steady state data collected from the M E B P R process 84 Table 3.6. T U D P model parameters used for the C E B P R process for fitting the averaged measurements 87 Table 3.7. Averaged measurements and simulation results of the calibrated T U D P model for the C E B P R process 88 Table 4.1. Summary of calibration parameters reported in the literature for A S M 2 d and T U D P models I l l Table 4.2. Selected parameters for calibration o f the T U D P model for the dynamic M E B P R process data 120 Table 4.3. List o f variables and standard deviations used in parameter estimation algorithm in A Q U A S I M 137 Table 4.4. T U D P model parameters estimated using the simplex parameter estimation method and manual calibration for fitting the measured data collected from the U B C M E B P R process 139 Table 4.5. The sum of squares results for the model fit of the dynamic measured data of the M E B P R process for different calibration methods 141 Table 4.6. T U D P model parameters used for model validation for the M E B P R process 148 Table 4.7. T U D P model parameters used in the modeling of the M E B P R and C E B P R processes 162 Table 5.1. Summary of the U B C M E B P R process performance under various operating conditions as reported by Mont i (2006a) 177 Table 5.2. Influent concentrations used for process design and operation studies . 179 Table 5.3. Simulation test run conditions for the initial process design and operation studies for the M E B P R process 180 Table 5.4. Process conditions used in the M E B P R process design and operation studies 184 Table 1.1. Stoichiometric matrix of the T U D P model (Meijer, 2004) 231 Table 1.2. Stoichiometric coefficients for S N H , SPO, SHCO and X T S S (Meijer, 2004) 232 Table 1.3. Component composition factors (Meijer, 2004) 233 Table 1.4. Stoichiometric parameters o f the T U D P model (Meijer, 2004) 234 Table I.5a. Kinetic parameters for Hydrolysis, X H and X A (Meijer, 2004) 235 Table I.5b. Kinetic parameters for X P A o (Meijer, 2004) 236 Table 1.6. Kinetic rate equations of the T U D P model (Meijer, 2004) 237 ix List of Figures Figure 1.1. Metabolic processes of organisms involved in E B P R under anaerobic, aerobic and anoxic conditions (Brdjanovic, 1998) 5 Figure 1.2. Schematic representation of E B P R process. The graph gives a schematic representation of the change in concentrations in the process (Brdjanovic, 1998) 6 Figure 1.3. The University o f Cape Town wastewater treatment process (Adapted from Metcalfand Eddy, 2003) 8 Figure 1.4. The UCT-type B N R process exhibit operational challenges (Adapted from Metcalfand Eddy, 2003) 38 Figure 2.1. U B C wastewater treatment pilot plant process 50 Figure 2.2. Total suspended solids in all biological zones during the experimental phase in the M E B P R and C E B P R processes 52 Figure 2.3. Temperatures in the M E B P R and C E B P R processes during the course of the study 54 Figure 2.4. p H in the M E B P R and C E B P R processes during the course of the study 55 Figure 2.5. Dissolved oxygen concentrations in the M E B P R and C E B P R processes during the course of the study 55 Figure 2.6. M E B P R process model in A Q U A S I M 63 Figure 2.7. C E B P R process model in A Q U A S I M 64 Figure 3.1. Influent and effluent C O D concentrations of the M E B P R and C E B P R processes during the course of the experiment 66 Figure 3.2. Ammonium profiles in the M E B P R process during the course of the experiment 67 Figure 3.3. Ammonium profiles in the C E B P R process during the course o f the experiment 68 Figure 3.4. Nitrate profiles in the M E B P R process during the course of the experiment 69 x Figure 3.5. Nitrate profiles in the C E B P R process during the course of the experiment 69 Figure 3.6. Ortho-phosphate profiles in the M E B P R process during the course of the experiment 71 Figure 3.7. Ortho-phosphate profiles in the C E B P R process during the course of the experiment 71 Figure 3.8. Influent V F A / T P ratio and effluent P O 4 - P concentrations for the M E B P R and C E B P R process during the course o f the experiment 72 Figure 3.9. Mixed liquor TP concentrations for the M E B P R process during the course of the experiment , 75 Figure 3.10. Mixed liquor T K N concentrations for the M E B P R process during the course of the experiment 76 Figure 3.11. Mixed liquor C O D t o t concentrations for the M E B P R process during the course of the experiment 76 Figure 3.12. Outlier analysis o f the M E B P R PO4 -P concentrations in the effluent stream and the aerobic zone during the course o f the experiment 78 Figure 3.13. The M E B P R process flow diagram as identified in Macrobal 79 Figure 4.1. M E B P R process dynamic simulation results for the TP profile in response to a step increase in the influent PO4-P concentration (PO4-P concentration increased from 2.5 to 8 g P /m 3 on day 201 at process SRT = 12 days and H R T = 10 hours) 95 Figure 4.2. Estimated missing data for influent COD t o t concentration used for the dynamic simulation of the M E B P R process 97 Figure 4.3. Power spectrum of the anaerobic zone PO4-P concentrations with respect to the influent PO4-P concentrations for the data used in the dynamic modeling o f the M E B P R process 99 Figure 4.4. Power spectrum of the anaerobic zone TP concentrations with respect to the influent TP concentrations for the data used in the dynamic modeling of the M E B P R process 100 Figure 4.5. Choosing the proper cut-off frequency for the filter design to filter the anaerobic zone TP concentrations of the M E B P R process 103 xi Figure 4.6. Filtered and unfiltered measured dynamic data for the TP concentration in the anaerobic zone of the M E B P R process 103 Figure 4.7. Frequency content of the filtered and unfiltered measured dynamic data for the TP concentration in the anaerobic zone of the M E B P R process 104 Figure 4.8. T U D P model predictions for N O 3 - N concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined in the steady state calibration stage 106 Figure 4.9. T U D P model prediction of the TP concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined in the steady state calibration stage 107 Figure 4.10. T U D P model prediction of the PO4 -P concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined in the steady state calibration stage 107 Figure 4.11. Dynamic sensitivity function values for PO4 -P in the anaerobic zone of the M E B P R process for Y ° p , Y P o 4 and fxsin parameters o f the T U D P model (HRT = 10 hours, SRT = 12 days and QINF = 5.1 m 3/day) 113 Figure 4.12. Averaged sensitivity analysis values of the PO4 -P concentration profile in the anaerobic zone to T U D P model parameters for the M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, SRT = 12 days and QrNF = 5.1 m 3/day) 115 Figure 4.13. Averaged sensitivity analysis values o f the PO4-P concentration profile in the anoxic zone to T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, SRT = 12 days and QrNF = 5.1 m 3/day) 116 Figure 4.14. Averaged sensitivity analysis values of the PO4 -P concentration profile in the aerobic zone to the T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. ( H R T = 10 hours, S R T = 12 days and QJNF = 5.1 m 3/day) 117 Figure 4.15. Averaged sensitivity analysis values of the NO3 -N concentration profile in the anoxic zone to the T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, SRT = 12 days and QINF = 5.1 m 3/day) 118 Figure 4.16. Averaged sensitivity analysis values of the V F A concentration profile in the anaerobic zone to the T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, SRT = 12 days and QINF = 5.1 m 3/day) 119 Figure 4.17. T U D P model prediction for the C O D t o t concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 123 Figure 4.18. T U D P model prediction for the TSS concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 124 Figure 4.19. T U D P model prediction for C O D s o i concentration profile in the effluent of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 125 Figure 4.20. T U D P model prediction of the filtered measured T N concentration profiles in the biological zones, influent and the effluent of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 126 Figure 4.21. T U D P model prediction for the N H 4 - N concentration profiles in the biological zones and the effluent o f the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 127 Figure 4.22. T U D P model prediction for the N O 3 - N concentration profiles in the biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 129 Figure 4.23. T U D P model prediction of the filtered measured TP concentration profiles in the biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 132 Figure 4.24. T U D P model prediction for the PO4-P concentration profiles in the anaerobic and anoxic zones and the effluent of the U B C M E B P R process using the T U D P model parameters determined during model calibration using dynamic data 133 Figure 4.25. T U D P model prediction for the V F A concentration profile in the anaerobic zone of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 133 Figure 4.26. T U D P model prediction results for the U B C M E B P R process for (a) effluent PO4-P (b) anoxic NO3 -N (c) aerobic TP and (d) anaerobic V F A for the period between August 22 - September 26 134 Figure 4.27. Cross-correlation function of the effluent C O D s o i residuals for the T U D P model predictions and the measured influent C O D s o i concentrations of the M E B P R process 142 Figure 4.28. Cross-correlation function of the effluent C O D s o i residuals for the T U D P model predictions and the measured influent C O D t o t concentrations of the M E B P R process 142 Figure 4.29. Cross-correlation function of the effluent N H 4 - N residuals for the T U D P model predictions and the measured influent N H 4 - N concentrations for the M E B P R process 144 Figure 4.30. Cross-correlation function of the effluent T N residuals for the T U D P model predictions and the measured influent T N concentrations for the M E B P R process 144 Figure 4.31. Cross-correlation function of the effluent PO4-P residuals for the T U D P model predictions and the measured influent PO4-P concentrations for the M E B P R process 145 Figure 4.32. Cross-correlation function of the anaerobic PO4 -P residuals for the T U D P model predictions and the measured influent PO4 -P concentrations for the M E B P R process 145 Figure 4.33. Cross-correlation function of the anoxic PO4-P residuals for the T U D P model predictions and the measured influent PO4-P concentrations for the M E B P R process 146 Figure 4.34. Cross-correlation function of the effluent TP residuals for the T U D P model predictions and the measured influent TP concentrations for the M E B P R process 146 xiv Figure 4.35. Model prediction of the C O D s o i concentration profiles in the anaerobic and aerobic zones for model validation of the calibrated T U D P using data collected from the M E B P R process in May of 2003 149 Figure 4.36. Model prediction of the C O D t o t concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 149 Figure 4.37. Model prediction of the T N concentration profiles in the influent and biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 150 Figure 4.38. Model prediction o f the N O 3 - N concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 151 Figure 4.39. Model prediction of the N H 4 - N concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 151 Figure 4.40. Model prediction of the TP concentration profiles in the influent and all biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 152 Figure 4.41. Model prediction of the PO4-P concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 153 Figure 4.42. Model prediction of the V F A concentration profiles in the anaerobic and anoxic zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 153 Figure 4.43. T U D P model prediction of the C O D s o i concentration profile of the effluent of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 163 Figure 4.44. T U D P model prediction of the C O D t o t concentration profiles in the biological zones o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 163 Figure 4.45. T U D P model prediction of the T N concentration profiles in the biological zones and the effluent C E B P R process using the T U D P model parameters calibrated for the M E B P R process 165 xv Figure 4.46. T U D P model prediction of the N H 4 - N concentration profiles in the biological zones of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 166 Figure 4.47. T U D P model prediction of the N O 3 - N concentration profiles in the biological zones o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 167 Figure 4.48. T U D P model prediction of the TP concentration profiles in the biological zones and the effluent o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 168 Figure 4.49. T U D P model prediction of the PO4-P concentration profiles in the biological zones of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 170 Figure 4.50. T U D P model prediction of the V F A concentration profile in the anaerobic zone o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 171 Figure 5.1. Effluent concentrations for the steady state simulation results of the different run conditions (simulated process T= 21.9 °C, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) 181 Figure 5.2. Effect of the aerobic biomass fraction on the nitrification process (simulated process temperature = 21.9 °C, H R T = 4 hours, S R T =10 days and influent V F A / T P = 6.2) 186 Figure 5.3. Effect of the anaerobic biomass fraction on the Bio-P removal process (simulated process temperature = 21.9 C, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) 187 Figure 5.4. Effect of the anoxic biomass fraction on the anoxic N O 3 - N concentration (simulated process temperature = 21.9, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) 188 Figure 5.5. Effect of process SRT and aerobic biomass fraction on the effluent N H 4 - N concentration (simulated Process temperature = 20 °C, H R T = 4 hours and influent V F A / T P = 10) 190 Figure 5.6. Effect of process SRT and anaerobic biomass fraction on the effluent PO4 -P concentration (simulated process temperature = 20 °C, H R T = 4 hours and influent V F A / T P = 5) ; 191 xvi Figure 5.7. Effect of the aerobic recycle ratio on the effluent NO3 -N concentration (simulated process temperature = 21.9 °C, H R T = 4 hours, and influent V F A / T P = 6.2) 193 Figure 5.8. Effect of the aerobic recycle ratio on the anoxic N O 3 - N concentration (simulated process temperature = 21.9 °C, H R T = 4 hours, and influent V F A / T P = 6.2) . 1 9 4 Figure 5.9. Effect of the temperature and aerobic biomass fraction on the effluent N H 4 - N concentration (simulated process SRT =10 days, H R T = 4 hours and influent V F A / T P = 5) 196 Figure 5.10. Effect o f the temperature on the effluent PO4-P concentration (simulated process SRT = 20 days, H R T = 4 hours, influent V F A / T P = 10 and biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic) 197 Figure 5.11. Effect of the temperature and S R T on the effluent PO4-P concentration (simulated process H R T = 4 hours, influent V F A / T P = 10 and biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic) 198 Figure 5.12. Effect of the S R T and influent V F A / T P ratio on the effluent P O 4 - P concentration (simulated process H R T = 4 hours, temperature = 20 °C and biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic) 200 Figure 5.13. Effect of SRT and influent V F A / T P ratio on the effluent P O 4 - P concentration (simulated process H R T = 4 hours, temperature = 20 °C and biomass distribution of 4% anaerobic, 21% anoxic and 75% aerobic) 201 Figure 5.14. Effect of the anaerobic biomass fraction and influent V F A / T P ratio on the effluent PO4 -P concentration (simulated process H R T = 4 hours, SRT = 15 days and temperature = 20 °C) 202 Figure 5.15. Effect of the influent V F A / T P ratio on the effluent N H 4 - N concentration (simulated process H R T = 4 hours, SRT = 10 days, temperature = 20 °C and sludge mass distribution of 4% anaerobic, 21% anoxic and 75% aerobic) 203 Figure 5.16. Recommended biomass distributions for a UCT-type M E B P R process for (a) aerobic zone and (b) anaerobic zone for process H R T = 4, temperature = 20 °C and different process SRTs and influent V F A / T P ratios to ensure effective P and N removal 206 xvn Figure 5.17. Recommended biomass fraction distributions for a UCT-type M E B P R process for (a) aerobic zone and (b) anaerobic zone for process H R T = 4, temperature = 15 °C and different process SRTs and influent V F A / T P ratios to ensure effective P and N removal 207 Figure 5.18. Recommended biomass fraction distributions for a UCT-type M E B P R process for (a) aerobic zone and (b) anaerobic zone for process H R T = 4, temperature = 10 °C and different process SRTs and influent V F A / T P ratios to ensure effective P and N removal 208 Figure 6.1. Different Process Configuration for an M E B P R Process 218 xv i i i List of Abbreviations A E R Aerobic bioreactor A N A Anaerobic bioreactor A N O Anoxic bioreactor A R G M C Adaptive robust generic model control A S M Activated sludge model A S M 1 Activated Sludge Model No. 1 A S M 2 Activated Sludge Model No. 2 A S M 2 d Activated Sludge Model No. 2d A S M 3 Activated Sludge Model No. 3 A U T Autotroph b A uT Autotrophic decay rate [g COD/(g C O D • d)] b H Heterotrophic decay rate [g COD/(g C O D • d)] bio-P Biological P removal B N R Biological nutrient removal B O D Biochemical oxygen demand B P N R Biological phosphorus and nitrogen removal C E B P R Conventional enhanced biological phosphorus removal C O D Chemical oxygen demand C O D p a r t Particulate C O D C 0 D S 0 | Soluble C O D COD t o t Total C O D C O V Coefficient of variation D O Dissolved oxygen E A W A G Swiss Federal Institute for Environmental Science and E B P R Enhanced biological phosphorus removal E F F Effluent of biological wastewater treatment process fxsin fraction of X s in the influent particulate C O D (g C O D / g C O D ) G D P M C Generic distributed parameter control G L Y Glycogen G M C Generic model control gpp Saturation reduction factor for poly-P formation H A c Acetic acid H R T Hydraulic retention time I A W Q International association on water quality INF Influent to biological wastewater treatment process (primary effluent) iNBM Nitrogen fraction of biomass (g N / g C O D ) iNSF Nitrogen content of soluble fermentable C O D (g N / g C O D ) JNSI Nitrogen content of inert particulate C O D (g N / g C O D ) iNXI Nitrogen content of inert particulate C O D (g N / g C O D ) iNXS Nitrogen content of particulate substrate (g N / g C O D ) IPBM Phosphorus fraction of biomass (g VI g C O D ) IPSF Phosphorus content of soluble fermentable C O D (g P/g C O D ) ipsi Phosphorus content of inert particulate C O D (g P/g C O D ) ipxi Phosphorus content of inert particulate C O D (g P/g C O D ) ipxs Phosphorus content of particulate substrate (g P/g C O D ) k G L Y Glycogen formation rate [g COD/(g C O D • d )] K N H Saturation coefficient for ammonium (g N / m 3 ) K 0 2 Saturation/inhibition coefficient for oxygen (g 02 /m 3 ) kpHA P H A degradation rate g COD/(g C O D • d) kpp Poly-phosphate formation rate [g P/(g C O D • d)] M B R Membrane bioreactor M E B P R Membrane enhanced biological phosphorus removal M L S S M i x e d liquor suspended solids N Nitrogen N H 4 - N Ammonium-nitrogen N O 2 - N Nitrite-nitrogen N O 3 - N Nitrate-nitrogen O E D Optimal experimental design P Phosphorus P A O s Phosphate-accumulating organisms P H A Poly-hydroxy-alkanoates P H B Poly-hydroxy-butyrate PI Proportional integral controller PID Proportional and integral and derivative controller PO4 -P Ortho-phosphate Poly-P Poly-phosphate q f e Maximum fermentation rate [g COD/(g C O D • d)] q s m a x Maximum anaerobic acetate uptake rate [g COD/ (g C O D • d )] P v A S Return activated sludge S Sensitivity function S F Readily biodegradable fermentable organic substrate S i Inert soluble organic material SISO Single-input-single-output S M P Soluble microbial products SRT Sludge retention time S T O W A Dutch Foundation for Applied Water Research S W O T Strength, Weakness, Opportunities, Threats analysis t Time (days) T K N Total Kjeldahl Nitrogen T N Total nitrogen T O C Total organic carbon TP Total Phosphorus TSS Total suspended solids T U D P The Technical University of Delft Phosphorus model U B C University of British Columbia U C T University o f Cape Town V F A Volatile fatty acids W W T Wastewater treatment W W T P s Wastewater treatment plants X A U T Autotrophic nitrifying organisms X G L Y Cell-internal storage o f glycogen in P O A s xxi X H Heterotrophic organisms X i Inert particulate organic material X s Slowly biodegradable particulate substrate Y A Autotrophic yield for growth g COD/(g C O D ) Y H Heterotrophic yield for growth on substrate (g C O D / g C O D ) Ypo4 Anaerobic yield for phosphate release (g C O D / g C O D ) P A U T Autotrophic growth rate [g COD/ (g C O D • d )] P H Maximum heterotrophic growth rate [g COD/(g C O D • d )] r|fe Anaerobic hydrolysis reduction factor r | H N 0 3 Reduction factor for denitrification r | L N 0 3 Anoxic hydrolysis reduction factor TI?NO3 Reduction factor for denitrifying P removal Preface This Ph.D. thesis is organized as follows: Chapter 1 presents the introduction a comprehensive review of the reported studies in the literature that serve as a background for this work. Then Chapter 2 describes the methodology adapted in this research including the process description of the pilot plant system, data collection, process behavior and the modeling environment. Chapter 3 is dedicated to the steady state calibration of the M E B P R and the C E B P R process data. In Chapter 4, a new dynamic calibration protocol is used to model the dynamic process data of the M E B P R and the results are compared to the modeling results of the C E B P R process. Guidelines for the design and operation of M E B P R processes under high flowrates and some proposed process control strategies for the process are discussed in Chapter 5. xx i i i Acknowledgments I would like to first express my sincere gratitude to my supervisors, Dr. Ezra Kwok and Dr. Eric Hal l , for their excellent guidance and continuous support throughout this challenging research without which this work would not have been possible. I would also like to extend my gratitude to the external examiner, Dr. Peter Vanrolleghem, for his thorough review o f the dissertation and the values comments. I would like to acknowledge my colleague and good friend Dr. Alessandro Monti for the valuable discussions and for the great effort in operating the U B C pilot plant. I also wish to thank Fred Koch, the U B C pilot plant manager, for his time and for sharing his wealth of knowledge gained over years of experience in the field. I would like to extend my gratitude to my colleague Dr. Zuohong Geng for helping out with the sampling at the plant and the great time we spent on this project. I also wish to thank my friend (Jeff) Guan Tien Tan for his advice and support throughout the project. Special thanks to Dr. Mark van Loosdrecht of the Technical University o f Delft for his informative discussions and for welcoming me to his lab during my short visit to the University of Delft. I also wish to thank Bas Meijer for the valuable assistance with the modeling and the great friendship. I would like to acknowledge the staff of The Chemical and Biological Engineering Department, Lor i Tanaka, Helsa Leong and Amber Lee, for their help and friendship. Special thanks to the staff of The C i v i l Engineering Environmental lab, Susan Harper and Paula Parkinson, for their assistance with the analytical work. The funding from the National Sciences and Engineering Research Council ( N S E R C ) is much appreciated. Also , I would like to extend my gratitude to the participation of the industrial partners of the N S E R C strategic project, Stantec Consulting Ltd., Zenon Environmental Inc., and Dayton and Knight Ltd. for their financial support and advice during the course o f the project. xxiv I am grateful to my fellow graduate students and friends for their love and support. M y special thanks to my dear friend, Dr. Poupak Mehrani, for her ongoing support and encouragement throughout my Ph.D. Many thanks to my dear friends, Sevan Bedrossian, A n a Stefano and Erik van Lier who made my life at U B C such an enjoyable experience. This thesis is dedicated to my family for their ongoing support and encouragement and to my husband, Farghad, for his patience and love. xxv 1 I N T R O D U C T I O N 1.1 Background The treatment o f wastewater for removal of phosphorus and nitrogen compounds to protect receiving water bodies against eutrophication has become a standard requirement in many countries in the world. Furthermore, population growth, environmental pressure, privatization pressure, stringent effluent discharge limits and shortage of land for conventional treatment systems result in the need for expanding the capacity of existing municipal treatment plants while minimizing capital expenditures for land and equipment. These incentives are motivating industry and research organizations to strive for innovative, effective and efficient alternatives to conventional treatment methods. These efforts led to the introduction of the membrane bioreactor ( M B R ) system, which combines the biological treatment process with membrane filtration for the final separation of suspended solids and liquid. The M B R system has been utilized for biological carbon and nitrogen removal processes in a number o f cases (Fan et al., 1996; Davies et al., 1998; Cote et al., 1998; de Silva et al., 1998). In these studies, the M B R system was found to produce superior effluent quality, when compared to the conventional clarifier system, with complete nitrification and essentially zero effluent suspended solids. M B R systems often offer simpler operation and maintenance and a smaller footprint. Furthermore, the membrane unit allows for operation at higher flowrates without being limited by the clarifier capacity. Nonetheless, the application of the M B R system for biological phosphorus and nitrogen removal is limited, despite its potential operational and economical benefits, since the design and operational parameters are not well known for this system. Only recently have researchers tackled this area in bench-scale studies where promising preliminary results were obtained (Lesjean et al., 2002; Lesjean et al. 2003; A h n et al., 2003). Therefore, this area required further investigation to enable the full-scale application of membrane enhanced biological phosphorus removal ( M E B P R ) to become wide spread. 1 In response to these needs, the environmental research group at the C i v i l Engineering Department of the University of British Columbia ( U B C ) carried out an inter-disciplinary research project focusing on various aspects of the M E B P R process, including a study on the process behavior of the system under various operating conditions, a comparison between the M E B P R process and the conventional system with a clarifier, a modeling and process design of the M E B P R process and a membrane fouling study. The focus o f this thesis is on the process modeling, design and operation of the M E B P R process. The main objective was to develop guidelines for the process design and operation of the M E B P R process based on simulation studies and experimental results to treat high flowrates of wastewater while also producing high quality effluents. Real time process dynamic data collected from the U B C M E B P R pilot plant process were modeled using the Technical University of Delft (TUDP) model. Then the model was utilized in simulation studies aimed at obtaining the set of process design and operating conditions that exploit the capacity of the M E B P R process for treating high loads without affecting the process performance and jeopardizing the effluent quality. Some recommendations for process control strategies aimed at ensuring effective process performance while meeting effluent requirements were also made. 2 1.2 Literature Review 1.2.1 Enhanced Biological Phosphorus Removal Process The enhanced biological phosphorus removal (EBPR) process is an activated sludge system that results in the removal of carbonaceous, nitrogenous and phosphorus-containing compounds from wastewater in a single process. A n activated sludge process is a complex system in which several bacterial conversion and transport processes occur. Kinetics, stoichiometry and transport processes play important roles in the conversion of contaminants. This section is dedicated to explaining different mechanisms taking place in the biological carbon, nitrogen and phosphorus removal systems, followed by a description of common process configurations for the E B P R process and ending by indicating the key operating parameters for achieving efficient E B P R . These sections provide an understanding of the E B P R process mechanism, design, and operation which are key elements in this project. 1.2.1.1 Involved Mechanisms Biological removal of carbonaceous material and nitrogen proceeds in two steps. The first step involves the biological oxidation of the organic carbonaceous material and the nitrification process, which requires an aerobic environment (presence of abundant molecular oxygen). A commonly used total measure for organic compounds in wastewater is the chemical oxygen demand (COD), which is a standardized method for wastewater characterization. The treatment steps under aerobic conditions can be written as follows. Step 1 - Aerobic environment: C O D removal by heterotrophic organisms: C O D + 0 2 • C 0 2 + H 2 0 + activated sludge Nitrification by autotrophic organisms: N H 4 + + 0 2 • N 0 3 " + 2 H + + activated sludge 3 The second step is the reduction of nitrate to molecular nitrogen, referred to as denitrification, requiring abundant C O D . In this step, the environment is anoxic (without oxygen but with nitrate). The biological process for this step can be written as follows. Step 2 - Anoxic environment: The organisms involved in E B P R have a complex physiology in which the formation and consumption of storage polymers such as poly-phosphate (poly-p), glycogen ( G L Y ) , and poly-hydroxy-alkanoates ( P H A ) play a dominant role. E B P R from wastewater is based on the enrichment of activated sludge with phosphate-accumulating organisms (PAOs). These microorganisms, also called P-removing organisms or poly-P bacteria, are able to store intracellular phosphorus as poly-phosphate (poly-P). Achieving a phosphorus-removing bacterial population in an activated sludge system requires exposure of activated sludge to anaerobic (without oxygen or oxidized nitrogen) and aerobic (aerobic E B P R ) or anoxic conditions (denitrifying E B P R ) . Under anaerobic conditions, P-removing bacteria transport volatile fatty acids ( V F A ) , e.g. acetic acid (HAc) , into the cell and subsequently convert and store these as P H A , e.g. poly-hydroxy-butyrate (PHB). The energy for this transport and storage is supplied by the hydrolysis of intracellularly stored poly-P to ortho-phosphate, which is released from the cell to the liquid (Figure 1.1). The reduction equivalents required for the conversion of acetate to P H B are supplied by the conversion of intracellularly stored glycogen through the glycolytic pathway to P H B and CO2. When oxygen (aerobic conditions) or nitrate (anoxic conditions) are present in the absence of organic substrate, the anaerobically formed P H B is used as substrate to generate energy for cell growth (poly-P synthesis) and glycogen formation and maintenance, resulting in the uptake of ortho-phosphate. During the aerobic or anoxic phases, P A O s accumulate more phosphate than is released during the anaerobic phase, resulting in a net uptake of phosphate. Phosphate is removed from the system by sludge wasting (Brdjanovic, 1998). The main criterion for obtaining good Denitrification by heterotrophic organisms (includes C O D removal) 4 removal efficiency in a biological phosphorus removal process is the amount of bio-P bacteria formed. These are needed to accumulate the phosphate. ANAEROBIC PHASE Acetate A Acetate Maintenance 0 \ PHB V NADH Jlycogen [Poly-PJ Phosphate AEROBIC PHASE Maintenance PHB Biomass r ^ J L t NADH -^ |ATP L ^ V c o g e n V ^ (Poly-p) <2> r Phosphate ANOXIC PHASE Nitrate , V f • 4 f p u n \ 1> NADH I ATP tT J Glycogen Poly-PK Phosphate Figure 1.1. Metabolic processes of organisms involved in E B P R under anaerobic, aerobic and anoxic conditions (Brdjanovic, 1998) 5 1.2.1.2 Process Description In practice, there are many different E B P R process configurations in which the removal o f organic matter, nitrogen and phosphorus (chemically and/or biologically) can be achieved. A l l these process configurations can however, be grouped into two basic types of processes, full biological processes and combined biological-chemical processes. Since the membrane enhanced biological phosphorus removal ( M E B P R ) process used in this study was solely a biological system, only the biological E B P R process is described. A basic scheme for a biological phosphorus removal process is shown in Figure 1.2. In the anaerobic phase, substrate, V F A is taken up by P A O s . This results in ortho-phosphate release to the liquid phase. The P A O s have a clear competitive advantage due to the ability to accumulate the VFA-substrate in the cells without the need o f an external electron acceptor. Therefore, the availability o f sufficient amounts o f substrates with absolutely no oxygen or nitrite present in the anaerobic zone is the essential requirement for an efficient phosphorus removal process. When the sludge is subsequently aerated, or when nitrite is available, other heterotrophic organisms wi l l have no substrate left while the P A O s can grow at the expense of their stored substrate. Anaerobic tank Aerobic tank Settled s e w a g e i l i i i i i i ^ mmmmmsi LV.V.V.V.V.V.V.V. [ i \ A a A J I . A ft A A t * . » j •>>>>>>>>>>>>>>%'Settling tank Return sludge Anaerobic tank Aerobic tank PfiA Glycogen V Gtifcoaen ft?ff^ f —— mn g M "3 ca I 0 CQ Effluent Excess sludge *-u u i t i n a i i L n^picai^iiiauuii u i jj/jjr rv piuL;css. i nc gliipil gives a schematic representation of the change in concentrations in the process (Brdjanovic, 1998) 6 To achieve E B P R in addition to carbon and nitrogen removal from wastewater, a treatment process consists generally of three stages through which activated sludge flows. • Anaerobic compartment: needed for selection of bio-P bacteria. It is crucial that no oxygen or nitrate is present in this compartment. The retention time in this compartment wi l l generally depend on the rate of fermentation of the complex organic carbon present in the wastewater to V F A . • Anoxic compartment: needed for denitrification and P-uptake. It is advantageous to remove phosphorus by organisms that can also denitrify. This w i l l save significant amounts of organic carbon. Accumulation of these denitrifying E B P R organisms can best be achieved in a process configuration in which the sludge is continuously recycled between the anaerobic and anoxic stage. The retention time in the anoxic reactor is determined by the denitrification rate or the hydrolysis rate by which particulate organic matter is converted into soluble substrate available for denitrification. • Aerobic compartment: needed for nitrification and P-uptake. The aerobic compartment is needed for the conversion of ammonium to nitrate and the P-uptake by the P A O s . The formed nitrate is then recycled with the sludge to the anoxic compartment. The retention time in the aerobic compartment is determined by the nitrification rate. A common process configuration to achieve E B P R and nitrogen removal is the University o f Cape Town (UCT) treatment process (Metcalf and Eddy, 2003). The U C T process includes the three biological zones mentioned above. The influent enters the anaerobic zone, then overflows to the anoxic zone followed by the aerobic zone. Nitrate or nitrite is introduced to the anoxic zone by recycling nitrified mixed liquor from the aerobic zone. For increased organic utilization in the anaerobic stage, there is an internal recycle from the anoxic zone to the anaerobic zone. The mixed liquor from the anoxic stage contains a substantial amount of soluble biochemical oxygen demand (BOD), but little nitrate, so the recycle of the anoxic mixed liquor provides the optimal conditions for fermentation in the anaerobic zone. In the conventional process, in which a secondary clarifier is used for the solids-liquid separation, the return activated sludge is recycled to 7 the anoxic zone to eliminate the introduction of nitrate to the anaerobic zone improving the uptake of phosphorus in the anaerobic zone. The University o f British Columbia ( U B C ) wastewater treatment pilot plant is based on the U C T process configuration which is presented in Figure 1.3. Anoxic Recycle Influent Return Activated Sludge ^ F i g u r e 1.3. The University o f Cape Town wastewater treatment process (Adapted from Metcalfand Eddy, 2003) 1.2.1.3 K e y P a r a m e t e r s A f f e c t i n g E B P R Process This section reviews the key parameters that influence the E B P R from wastewaters. The focus is on wastewater composition parameters and operational parameters affecting successful nutrient removal in wastewater treatment plants. A review of these factors was important to better understand the process behavior o f the M E B P R and the C E B P R (conventional enhanced biological phosphorus removal) process o f the U B C pilot plant. S u d d e n D i s t u r b a n c e s In general, the E B P R process is sensitive to sudden disturbances to the system and therefore, should be kept as stable as possible with minimal fluctuations in influent wastewater flows and compositions (Shehab et al., 1996). Okada et al. (1992) reported that prolonged disturbances to the E B P R process can lead to recovery times o f over 4 weeks. 8 Volatile Fatty Acids A s described above, V F A is essential for an effective E B P R process operation. Mulkerrins et al. (2004) reported that 7-9 mg of V F A is needed to remove 1 mg of phosphorus. It has also been shown that supplementation by each additional 7.5 mg acetate/L of influent contributed to an extra removal of 1.0 mg P / L (Manoharan, 1988). A paper by Oldham and Rabinowitz (2001) emphasized the need for primary sludge fermentation, or the provision of alternative sources of V F A , especially in North American wastewater treatment plants (WWTPs) . They further indicated that where fermenter supernatant is added to the anaerobic zone to provide V F A for the phosphorus removal process, the remainder of the fermentation products spill over into the main anoxic zone, thereby increasing the rate of denitrification that is realized in that zone. Moreover, they reported that this rate increase can be as much as 30 or 40% above that measured in an anoxic zone that must rely solely upon the organic carbon resources of the raw sewage to fuel the denitrification reactions. This improved denitrification rate resulting from prefermentation was also reported in other studies (McCue et al., 2003; M c C u e et al., 2004). Nitrate Load in The Anoxic Zone One of the major factors influencing the occurrence of denitrifying P A O s and associated anoxic P-uptake is the nitrate load to the anoxic reactor, which should be equal to or greater than the denitrification potential of ordinary heterotrophic organisms, since the latter w i l l out-compete P A O s for use o f the limited nitrate (Hu et al., 2002). Furthermore, sufficient concentration of nitrate is also required in the anoxic zone to avoid secondary P-release when all nitrate is denitrified before the end of the anoxic zone (Meinhold etal., 1999). On the other hand, excess concentrations of nitrate in the anoxic zone should be avoided to eliminate the presence of residual nitrate in the anaerobic phase, since it negatively impacts the anaerobic P-release as it results in consumption of influent organic compounds by denitrifiers, thus decreasing the availability of organic matter for PAOs . 9 Effect of Nitrite Work presented by Saito et al. (2004) indicated that the presence o f nitrite (NCV) , formed and accumulated through both nitrification and denitrification, in an E B P R system inhibited both aerobic and anoxic ortho-phosphate uptake while aerobic P-uptake was more affected. Cations Sufficient amounts of both potassium and magnesium are simultaneously required for E B P R since in any poly-P chain, one positive charge is required to stabilize each phosphate group and the expulsion of each phosphate molecule from the cell needs one cationic charge from either potassium or magnesium (Pattarkine and Randall, 1999). Temperature A t low temperatures, denitrification becomes the limiting factor in the overall nutrient removal process, leading to elevated levels of nitrate in the effluent. Therefore in E B P R -B N R (biological nutrient removal) combined systems, the E B P R process can be negatively affected by low temperatures since it can lead to higher nitrate concentrations in the return sludge, thereby impacting on E B P R (Mulkerrins et al., 2004). Dissolved Oxygen Requirements Maintaining an oxygen concentration between 3 and 4 mg/L in the aerobic zone of an E B P R - B N R combined system is recommended. Shehab et al. (1996) indicated that a dissolved oxygen (DO) concentration of 2 mg/L is sufficient for E B P R , but when nitrification is also necessary, a D O of 3 - 4 mg/L is essential. Phosphorus Load The ratio of phosphorus to total organic carbon (P/COD) in the wastewater is important in the selection of the P A O s bacteria and in giving them a competitive advantage (Mulkerrins et al., 2004). Research carried out by L i u et al. (1998) showed that at low influent P / T O C (total organic carbon) ratio, the growth of P A O s was suppressed while higher P / T O C ratios encouraged the growth of P A O s . 10 pH Effect Mulkerrins et al. (2004) reported that the optimal nitrification rate occurs between p H 7.5 and 9.0 and becomes zero at approximately 6.0. They also reported that the p H optimum for denitrification appears to be between p H 7.0 and 8.0 while an optimum p H of 6.8 ±0.7 was proposed for anaerobic acetate metabolism. SRT Oldham and Rabinowitz (2001) indicated that for typical North American wastewaters, a main bioreactor sludge retention time (SRT) of about 10 days in conjunction with a hydraulic retention time (HRT) of 8 hours is capable of good nitrogen and phosphorus removal with careful attention paid to the design of the secondary clarifiers to avoid sludge blanket overflows. Moreover, other studies reported that an SRT of 10 days gave the best P-removal rate (Chang et al., 1996; Choi et al., 1996). The large number of factors affecting the E B P R process indicates the complexity of the process, which further emphasizes the need for a process model that can describe the process behavior of the system to be used in process design and control studies. The following section describes the various models available in literature for describing the E B P R process. 1.2.2 Modeling of Enhanced Biological Phosphorus Removal Process Models that are based on the actual or believed physics, chemistry and microbiology that govern the system are referred to as mechanistic models. There are other types of models which are commonly used in the analysis of process data that are based on a convenient mathematical function that reasonably represents available data from the system. These models are usually referred to as black-box models. Examples of such models are: linear time series and artificial neural network models which have proved popular for fitting nonlinear data (Olsson and Newell , 1999). The differences between the two types of models are described in the next section. 11 1.2.2.1 Mechanistic and Black-Box Modeling Mechanistic models are based on real physical principles and therefore, they tend to be complicated, nonlinear in many cases and require extensive process knowledge. Furthermore, since it is not possible to account for all physical processes as it adds to the model complexity, when mechanistic models are used to predict real plant data, most often they do not match well and require extensive calibration and validation. On the other hand, mechanistic models are more reliable for long term applications as they do not tend to deviate over time and so they require less maintenance in the long term and can be extrapolated to predict process behavior under different conditions (Ogunnaike and Ray, 1994). Black-box models on the other hand are less complicated (require less process knowledge) and have a better chance to match real plant data well , since they are developed based on data dynamics. However to ensure model validity, large sets of reliable process data collected at various process conditions are required. In addition, black-box models require continuous maintenance, as processes tend to deteriorate over time. Furthermore, due to the lack of mechanistic interpretability o f black-box models, extrapolations to similar systems and different driving conditions are not meaningful, especially for complex systems such as activated sludge processes. A more detailed description o f various black-box models available for wastewater treatment processes is given by Olsson and Newell (1999). Since it was important in this project to use the model to predict the process behavior under different design and operating conditions, it was decided to use a mechanistic model for modeling the M E B P R process. Therefore, black-box models are not discussed further in this thesis. The next section describes some of the models presented in the literature for the modeling of the E B P R process. 1.2.2.2 Mechanistic Models for the EBPR Process Modeling of the activated sludge process for C O D and nitrogen removal has become a standard practice and provides a valuable tool for the design and operation of activated sludge processes. 12 The International Association on Water Quality ( IAWQ) task group has proposed the Activated Sludge Model No. 1 (ASM1) (Henze et al., 1987), which has been used for more than two decades as a valuable tool for modeling the removal o f organic matter, and for the nitrification and denitrification processes. However, E B P R is a much more complex process, as a range of bacterial conversion and transport process occur as shown earlier. There are a number of model structures available for describing the E B P R process (Henze et al., 1999; Barker and Dold, 1997; and Rieger et al., 2001). A l l these models use more or less a black-box description of the E B P R process, as only one lumped organic storage compound is modeled. However in the metabolic E B P R model proposed by Smodlers et al., (1994a, b) and Kuba et al, (1996) and extended by Murnleitner et al. (1997) to include denitrifying phosphorus removal, a full account of the metabolism of P A O s is given, by modeling all storage compounds explicitly. The model was integrated with the heterotrophic, hydrolytic and autotrophic reactions of A S M 2 d , proposed by the I A W Q task group (Henze et al., 1999), and further referred to as the Technical University of Delft Phosphorus (TUDP) model (Van Veldhuizen et al., 1999). Subsequently, several improvements of the model have been proposed. Meijer (2004) presented a detailed history of the model development along with a complete and updated version of the T U D P model. In his work, Meijer concluded that the stoichiometry of this version of the metabolic model can be used without calibration. Since the T U D P model gives a full account of the bacterial metabolism, it was considered as a potential candidate for the modeling work in this research work and therefore further research was carried out to evaluate its applicability for modeling the M E B P R process. The T U D P model (as presented by Meijer, 2004) is described in the following section. 1.2.2.3 The Technical University of Delft Phosphorus Model The T U D P model is based on combining the metabolic E B P R model with the A S M 2 d model from the I A W Q task group. The A S M 2 d model (Henze et al., 1999) is an extension of the A S M 2 (Henze et al. 1994) and the A S M 1 (Henze et al. 1987) models, and uses the concepts incorporated in these models. The A S M 1 model has long since proved to be an excellent tool for modeling nitrification-denitrification processes. 13 In the metabolic model, eight cell-internal processes are described involving three storage polymers: P H A , G L Y and poly-P. The processes are anaerobic P H A storage, anaerobic poly-P degradation, aerobic and anoxic poly-P and G L Y formation and aerobic and anoxic P H A degradation. The yield coefficients of these processes are determined by two metabolic yield coefficients (Smolders et al., 1994a,b). The concept of endogenous respiration or maintenance is used (Henze et al., 1999). The formation of biomass (PAOs) is calculated as the net result of conversions of P H A , G L Y and poly-P. In the model version proposed by Meijer (2004), a new process was added for the anoxic storage of P H A to be able to handle simultaneous presence of V F A substrate for storage reactions and electron acceptors. The T U D P model as presented by Meijer (2004) consists of 22 processes (each represented by a rate equation) and 18 components (soluble and particulate). The model also has 10 component composition factors for nitrogen ( i N ) and phosphorus (ip) and 13 stoichiometric parameters in addition to 50 kinetic parameters. A detailed stoichiometric matrix, kinetic expressions, and all model parameters for the T U D P model as given by Meijer (2004) are presented in Appendix I. A major limitation of the current kinetic model structure of the T U D P model, as concluded by Meijer (2004), is that it is not suited to describe the complex cell-internal mechanism that balances the formation and composition of P A O s storage compounds and therefore, more research is needed to determine the precise mechanism. The following section describes the different applications of the T U D P model to simulate data collected from full-scale W W T P s . The review of these studies provides a good understanding of the model's prediction power, range of applicability and the calibration procedures used to fit the experimental data. 1.2.2.4 Applicat ions of T U D P M o d e l The parameters Of the metabolic E B P R model have been extensively validated by many independent experiments using sequencing batch reactors at laboratory scale (Smodlers et al., 1994 a,b; Kuba et al., 1996; Murnleitner et al., 1997). The T U D P model was used in 14 modeling full-scale wastewater treatment plants in many cases (Van Veldhuizen et al., 1999; Brdjanovic et al. 2000; Meijer et al., 2001; Meijer et al., 2002a, and Meijer et al., 2002b). These cases are discussed in more detail below, focusing on the calibration methodology and results. In the work by Van Veldhuizen et al. (1999), the metabolic E B P R model was incorporated in the A S M 2 model structure and the entire model was implemented by means of the computer software package S I M B A 3.0 (based on M A T L A B / S I M U L I N K ) . The integrated model was used to simulate the measurements collected from the Holten W W T P in the Netherlands. The authors indicated that model calibration, validation and sensitivity analysis were performed. In the calibration, they started with a static influent making use of the dynamic model (stoichiometery and kinetics), but used average constant values for the influent flow and composition, for the 2-day sampling period, giving an approximate steady state solution. Samples were collected at four different places in the activated sludge tank. In the calibration stage, parameters were changed based on mechanistic reasoning, rather than on their sensitivity. Only three kinetic parameters, the fermentation rate (qfe), poly-P uptake rate (k p p ) and the reduction factor under anoxic conditions (r|PN03), were changed, based on the causality of the parameters on the predicted concentrations, and the relative uncertainty in the original parameter values. The authors indicated that the simulated changes in effluent concentrations were in good agreement with the measured values. After this static influent calibration, the authors claimed validation of the model under dynamic conditions by introducing the daily influent flow profile for the 2-day sampling program of the same period used for calibration. It was further indicated that there was no significant difference between the predicted concentration values using the static or the dynamic influent profiles (the maximum difference observed in the predicted concentrations was 0.5 mg/L). It should be noted that, in their study, model kinetic dynamics were not investigated since the data only showed hydraulic dynamic variation while almost constant concentration profiles were observed. Furthermore, in order to evaluate the model's kinetic dynamics, dynamic data sampled over a period exceeding twice the dominant time constant Of the process (which is in this case the process sludge retention time, SRT) are required for calibration. 15 Moreover, according to the standard system identification techniques indicated by Soderstrom and Stoica (1989) and discussed later in section 1.2.4.4, model validation should be performed using a different set of data than that utilized in the model calibration step. In the sensitivity analysis stage of the work reported by Van Veldhuizen et al. (1999), the concentrations of ammonium, nitrate and ortho-phosphate in the effluent and the sludge production for all parameters and process choices were analyzed. The sensitivity (S) of the model parameters (x) with respect to y (sludge production and effluent ammonium, nitrate and ortho-phosphate concentrations) was calculated by: S = fc (1.1) dx y Results of their sensitivity analysis showed that in addition to a limited number of model parameters (12%), the influent characterization influenced the model output significantly. Based on this result and the calibration procedure, the authors found that since a standard sensitivity analysis-based calibration procedure can not differentiate between more or less defined parameter values (since some parameters are more defined as they are based on known reaction stoichiometry), a calibration based on process knowledge is more sensible. However, they further indicated that since the latter requires more experience, therefore, possibly a combination of parameter sensitivity analysis and process knowledge, is in general, the best approach. In another study performed by Brdjanovic et al. (2000), the integrated metabolic E B P R model with the A S M 2 model was used to simulate the measured data collected at the Haarlem Waarderpolder full-scale W W T P in the Netherlands under steady state conditions. In their study, model calibration was based on a proper influent and sludge characterization and detailed evaluation of the flow scheme of the treatment plant. Calibration was done by adjusting the model parameters to fit the steady state data collected from the treatment plant while some kinetic parameters were estimated using separate batch experiments performed with sludge and wastewater. Detailed sampling revealed that the daily variation in the plant concentrations was marginal. A stepwise 16 calibration procedure was applied in which just a few specific parameters were calibrated on specific plant data. The calibration was also based on mechanistic reasoning in this study as well . Only three parameters were adjusted, two of which were the same as the ones adjusted by Van Veldhuizen et al. (1999). Batch tests, some of which were used for estimating some parameters in the model calibration stage, were satisfactorily described by the model when used for model validation. In that study, batch test data were used for model calibration and validation as well , which is not a recommended procedure in system identification practice. A more recent study was carried out by Meijer et al. (2001), in which the T U D P model was used to simulate data collected from the Hardenberg W W T P in the Netherlands. During a 60-h sampling period, with a 2-h sampling frequency, the pseudo-steady state of the plant was recorded. The plant was sampled at five different internal locations together with the measurement of influent and effluent. The authors indicated that since the variation in influent concentrations was relatively small, it was decided to use the actual influent flowrate profile with the average concentrations for influent modeling. A t the start of the simulation, they were unable to simulate effluent and process internal concentrations correctly with the model and parameters estimated by V a n Veldhuizen et al. (1999). It was concluded that the operational data or measurements contained errors. Therefore, the measurements and operational data were checked and corrected by solving mass balance equations. Then they performed model calibration using a stepwise approach in which only three specific parameters were calibrated. Each parameter was changed to better fit a specific concentration measurement. It was indicated that a heuristic approach for model calibration was used for the following reasons. (i) Since sensitivity analysis showed that the predicted concentrations are more sensitive to operational data than to model parameters, evaluation of operational data was preferred over calibration of the model parameters. 17 (ii) Mathematical calibration procedures usually neglect the influence of faulty operational data and instead focus on model kinetic and stoichiometric parameters. (iii) After correcting the faulty operational data and measurements, no major calibration was needed to fit the model. A s a result of the calibration, the authors found that only the kinetic rates that are limiting were sensitive to calibration. Therefore, pseudo-steady state systems are not suitable for evaluation of kinetic parameters. They further concluded that the calibration procedure should be focused on checking operational data rather than on calibrating stoichiometric and kinetic parameters. It was also stated that two parameters, the inert C O D fraction in the influent, calibrated by the fraction of slowly biodegradable C O D in the influent particulate C O D (fxsin) and the actual anoxic sludge fraction, calibrated using the saturation/inhibition coefficient for oxygen (K02), would always need calibration for all systems. It should however be noted that this conclusion was based on steady state calibration of the dynamic model and therefore needs further verification under dynamic conditions. Another study by Meijer et al. (2002a) was focused on describing a method for gross error detection and data reconciliation, which wi l l be described in the next section. This method was used to check data collected from the Katwoude W W T P in the Netherlands. After error detection and data reconciliation, model calibration was completed using the stepwise approach proposed in the previous study of Meijer et al. (2001). Solids were fitted on the basis of yearly average measurements, while soluble concentrations were calibrated using averaged measurements of data collected during an 8-day sampling period, with 24-h sampling frequency. In their study, the SRT was calibrated based on the overall phosphorus balance the phosphorus sludge fraction was fitted by adjusting the fxsin ratio, and the nitrogen sludge fraction was fitted using the factor for nitrogen content of inert particulate C O D (iNxi) and the factor for nitrogen content of particulate substrate ONXS)- Nitrification was fitted by adjusting the oxygen set-points. Denitrification was fitted by increasing K02, which 18 increases the anoxic sludge fraction as i f the penetration depth of oxygen in the sludge floe is decreased, causing an anoxic zone in the center of the floe (Pochana and Keller, 1999). A l l previous studies were focused on calibrating the T U D P model using more or less steady state concentration profiles. However, the next study described below was performed by Meijer et al. (2002b) using data collected during the start-up of an upgraded E B P R full-scale plant. The study by Meijer et al. (2002b) was carried out to evaluate the kinetics of the metabolic E B P R model under dynamic conditions imposed by the start-up of the full-scale upgraded biological phosphorus and nitrogen removal (BPNR) process of the Hardenberg W W T P in the Netherlands. The steady state data of this plant had been modeled previously using the T U D P model in a study by the same group (Meijer et al., 2001). However, the data used in the (2002b) study were collected prior to that, during the start-up of the upgraded W W T P . The modeling results showed that the start-up was simulated with identical calibration parameter values for K02, gpp (the saturation reduction factor for poly-P formation) and fxsin as for the steady state modeling work presented earlier (Meijer et al., 2001). The authors further indicated that glycogen formation is a key process during the start-up of a biological phosphorus and nutrient removal (BPNR) process and that the model showed particular sensitivity for temperature as well . From a careful examination of the fitted concentration profiles presented in the work by Meijer et al. (2002b) for the start-up, the following observations can be noted. • For the total phosphorus profile in reactor 4, the aerobic zone, no measurements were reported until day 20, when the process had reached 25% of its final value. • The predicted ortho-phosphate profile in reactor 2 (the anoxic zone) was shown to be always higher than the measured concentrations. A constant offset of about 1 g P /m 3 was observed throughout the period. • A n over-prediction was observed of the ammonium concentration, with a difference of about 15 g N / m 3 , in the anoxic zone (reactor 2) in the first 5 days o f the simulation. 19 • The predicted effluent nitrate profile for the first 10 days was also over-estimated by the model. These observations, which were not addressed in the paper, indicate that more effort is required to examine the T U D P model kinetics and kinetic parameters. This was also noted in a review paper of existing available simulation models for B N R activated sludge systems by H u et al. (2003). The authors indicated that systems used for model validation o f the "Delft" Bio-P were phosphorus-limited, which may influence the E B P R results when the model is applied to carbon-limited systems and therefore, the predictive power of this model needs to be evaluated against more experimental data. Moreover, in all reported studies on fitting the T U D P model, calibration was performed manually by adjusting the parameters to achieve the best fit as judged by eye. This approach does not guarantee the best fit as discussed later in section 1.2.4. Furthermore, the T U D P model has been used primarily in the. modeling of W W T P s in the Netherlands, and hence, the effect of different wastewater characteristics and process environments on model kinetics has not yet been investigated. More research is therefore required in light o f these observations to further investigate the T U D P model predictive power under real dynamic conditions, to identify possible missing components of the model and areas that are in need of attention. The calibration techniques used to date require careful evaluation as well . In general, these studies showed promising results for the ability of the T U D P model to predict data collected from batch results and full-scale W W T P s and these were considered encouraging for the modeling work in the present project. However, the studies also demonstrated the need to further evaluate the predictive power of the T U D P model using dynamic data collected for an extended period to ensure capturing the process dynamics required for the dynamic calibration. Furthermore, these studies also presented the need for a reliable model calibration technique that takes into account process knowledge and standard mathematical calibration techniques to ensure a good fit of the experimental data. These findings were considered during the development of the research approach and required tasks. 20 1.2.2.5 Data Reconciliation for Activated Sludge Modeling Two important aspects in modeling of activated sludge system are data reliability and integrity, which are the focus of this section. Data measurements in activated sludge systems can be very difficult, tedious and time consuming as they require lengthy laboratory analytical procedures with large error margins that can go as high as 25% in some cases, especially for sludge concentrations. The lack of reliable on-line sensors and the costs associated with them adds to the problem. Hence, sampling is usually done manually and less frequently than needed for modeling purposes to reduce the effort, time and costs associated with the sampling exercise. Furthermore, some model components of activated sludge models are difficult to measure on a daily basis during sampling, such as biodegradable and inert C O D fractions for example, and so they are estimated, which introduces more data errors. Therefore, data reconciliation is essential for modeling of activated sludge systems. There are a number o f techniques to be used for data cleaning. The first step in any data reconciliation exercise is outlier detection. This step consists of removing readings associated with error measurements. Then other data reconciliation methods can be applied such as performing mass balances around the biological reactors. In the work presented by Meijer et al. (2002a), they indicated the importance of checking the prime data for errors and consistency to eliminate unjustified model calibrations resulting from fitting a model to erroneous or unbalanced data. In their work, the gross error detection and data reconciliation techniques for biochemical processes implemented in the free-domain software "Macrobal" by Hellinga (1992) were used to balance the mass flows of the Katwoude full-scale W W T P in the Netherlands. The technique used in Macrobal is based on the fact that all systems can be described by a set of linear relations based on mass balances (e.g. C O D , nitrogen, phosphorus and flows). However, i f more measurements are available than the minimum required to solve the set of linear equations, the system would be over determined. In that case, gross errors for systems can be detected by evaluating the balance residuals (s). When there is 21 no statistical proof for gross errors, the data can be balanced in a data reconciliation procedure resulting in a balanced data set without gross errors and with an improved overall accuracy. Detailed description of the Macrobal software and techniques is given byHell inga(1992) Meijer et al. (2002a) formulated a set of linear equations on the basis of flow for phosphorus and ammonium balances over different reactors of the process and implemented the equations in Macrobal. The program was found to be suitable for checking raw operational data and process measurements. Erroneous flow measurements were detected and corrected based on the results obtained. Details of the application of Macrobal to wastewater treatment systems is described in Meijer et al. (2002a). A l l literature studies discussed so far have dealt with the C E B P R system. The following sections describe related studies in the literature about M B R systems and their application to the E B P R process. Furthermore, the modeling of the M E B P R system is also described. 1.2.3 Membrane Bio-Reactor System The operation of the conventional activated sludge process is restricted to fairly low biomass concentrations because of the final settlement limitations. In addition, the required reactor volumes are large and the conversion rates are low. Over recent years, attempts have been made to increase the working biomass concentrations. These developments have brought improvement, but there is still plenty of scope for highly efficient systems to be developed (Davies et al., 1998). There is a need to deliver processes that meet legislative requirements more cost effectively in light of increased demands on industry to deliver effluents to higher standards and more reliable quality. This led to the introduction of membrane bioreactor ( M B R ) systems in activated sludge processes. Membrane technology is making an impact on a number of wastewater treatment areas with its versatile separation capability (Davies et al., 1998). M B R technology offers an innovative approach to the treatment of municipal and industrial wastewaters. The M B R approach couples biodegradative treatment to a physical solids-liquid separation process in a manner that harnesses the advantages of 22 both unit operations to produce ultra-high quality treated effluent. There are two major ways of deployment of the membrane unit in the biological system. Membranes are either submerged into the bioreactor or connected as a sidestream process. The immersed membrane module usually consists of hollow fibers or hollow panes that are immersed in the aerated tank while effluent is produced by applying suction in the inner part of the fibers. A i r flow to the aeration zone is used to create the turbulence required to reduce membrane fouling. The following sections describe the main advantages o f M B R , anticipated impact of the membrane unit on the bioreactor stage and the reported modeling attempts for M B R systems. 1.2.3.1 Advantages of M B R System Membranes offer a reliable, high rate filtration process with absolute removal of sub-micron sized particles and coliform bacteria. The small footprint of membranes ensures minimum land space requirement and provides excellent scope for retrofitting existing wastewater treatment works. The following are some of the reported advantages of M B R systems. • The membrane separation system leads to excellent effluent quality capable o f meeting stringent discharge requirements with the possibility of eliminating extensive disinfections and the corresponding hazards related to disinfection by-products (Cicek, 2003). • Operation with higher hydraulic loads without the fear o f clarifier overflows. • Operation with higher mixed liquor suspended solids ( M L S S ) concentrations compared to the conventional process. • Since suspended solids are not lost in the clarification step, total separation and independent control of the SRT and H R T are possible, enabling optimum control of the microbial population and flexibility in operation. • Microfiltration membranes with defined pore diameters ensure that microorganisms and all relevant particles are retained in the biological process, enabling slow-growing species, such as nitrifying bacteria, to develop in the system even under short SRTs. 23 • The system is able to handle fluctuations in nutrient concentrations due to extensive biological acclimation and retention of decaying biomass (Cicek, 2003). • M B R s require less land space compared to the conventional process. The main disadvantages of the M B R system on the other hand, are high capital cost and membrane fouling, which are being addressed by many manufacturing companies as they strive for innovative approaches to making M B R systems more attractive. Membrane fouling is also gaining more attention and becoming an active area of research to elucidate fouling mechanisms in attempts to reduce this problem. Another disadvantage to membrane systems is the increased energy related to aeration, which is required to prevent fouling. Furthermore, the literature reports some concerns about anticipated impacts of the M B R on the activated sludge system, which are discussed in the next section. 1.2.3.2 Impacts of Membrane Process on Bioreactor Stage It is important to examine possible negative impacts of the membrane process on the bioreactor stage since it is essential to maintain adequate biological activity under sequential anaerobic, anoxic and aerobic conditions. In fact, very little is known about the physiological state o f microorganisms in M B R systems operating at long S R T conditions present in M E B P R processes (Manem and Sanderson, 1996). Cicek et al. (1999) indicated that when M B R systems are operated at high SRTs, inorganic compounds accumulating in the bioreactor can reach concentration levels that can be harmful to the microbial population. Moreover, results presented by Canales et al. (1994) showed that the viability and sludge production yield decreased when sludge age increased. Furthermore, M B R systems exhibit reduced particle size distribution compared to conventional activated sludge processes. This reduction in particle size is attributed to the turbulence required in M B R systems to reduce membrane fouling, which in turn results in sludge floe break-down (Tay et al., 2003). However, the smaller sludge particles are desirable as they are believed to enhance mass transfer, thereby inducing higher organic 24 removal rates and better oxygen utilization (Tay et al., 2003). The effect of this anticipated impact of M B R requires further investigation, especially when integrated in the complex microbial physiology of the E B P R process. Studies investigating the application of M B R systems in E B P R process are discussed in the following section. 1.2.3.3 M B R Applicat ions i n E B P R Process There are a number of applications of M B R in wastewater treatment which showed promising results for the application of the membrane unit in a wastewater treatment process (Fan et al., 1996; Davies et al., 1998; Cote et a l , 1998; de Silva et al., 1998). In all these studies, M B R s were used for nitrogen removal. Other studies were carried out with M B R s for nitrogen removal while achieving phosphorus removal by chemical precipitation (Engelhardt et al., 1998; Cote and Thompson, 2000). There is a great potential to couple the M B R concept to E B P R processes to result in a M E B P R process with low effluent ortho-phosphate content with operation at high flowrates. Bench scale studies of M E B P R applications performed by Lesjean et al. (2002) and Lesjean et al. (2003) showed that efficient Bio-P removal can be achieved with M B R systems in both pre-denitrification and post-denitrification configurations. Furthermore, the authors indicated that under similar operation conditions of sludge age and organic mass load, an M B R system achieved slightly higher P-removal than a conventional technology due to the rejection of particles and colloids through the microfiltration membrane. In another study by A h n et al. (2003), an innovative process of a sequencing anoxic/anaerobic zone and an aerobic zone membrane bioreactor was used for E B P R resulting in 93% phosphorus removal efficiency. Preliminary pilot-scale testing at the University of British Columbia in the 90's, in which a submerged hollow fiber membrane module was inserted into the aerobic zone of a UCT-Type E B P R system, indicated that an M E B P R process could produce high quality treated effluents in the most efficient manner possible (Koch, 2002). These studies provide the base foundation for further research on M E B P R systems to better understand the impact o f the membrane unit on the E B P R process operation, mechanisms and kinetics. These results can then be utilized in modeling the M E B P R 25 process and utilizing the model to study the process behavior under various design and operating conditions. Some of the studies in the literature that have attempted to model the process behavior of M B R systems are discussed in the next section. 1.2.3.4 Mode l ing of M B R Systems Modeling o f M B R systems has been the focus of a number of research groups. Wen et al . (1999) studied the kinetic properties of an M B R and derived an equation to calculate the sludge concentration in the M B R from the material balance of substrate and biomass. The equation was then satisfactorily employed to predict the trend of activated sludge growth during an experiment. The same research group then utilized this equation to introduce a formula to predict the excess sludge production in an M B R and verified it against experimental data (Xing et al., 2003). Urbain et al . (1998) performed a complete characterization o f an M B R activated sludge system using nucleic acid probe analysis to determine the heterotrophic and nitrifier populations and the results were compared to the output from a multispecies model that integrated substrate removal kinetics and soluble microbial products (SMP) production/consumption. The model was based on the model presented by Furumai and Rittmann (1992). They were able to describe the M B R system in terms of C O D removal and nitrification as well as sludge production and population dynamics through the ratio of active nitrifiers/bacterial population. Both steady state and transient conditions could be described accurately by the model. Another modeling attempt for a pilot-scale M B R wastewater treatment process with aerobic-anoxic cycles to quantify the relationships among heterotrophic bacteria, autotrophic bacteria and key inorganic, ammonia and nitrate, and organic C O D compounds was presented by de Si lva et al. (1998). Again in their work, the foundation for the M B R model was based on the work of Furumai and Rittmann (1992). The presented model was able to accurately predict the sludge concentration and the aggregate effluent quality obtained through experiments. Key features of the M B R model 26 were zero biomass in the effluent with partial removal of biomass-associated products by the membrane. A more recent study by L u et al. (2001) showed that a model of S M P formation-degradation incorporated into the A S M N o . 1 model was able to describe the performance of a membrane bioreactor process for C O D and nitrogen removal, with a high concentration of activated sludge, and the results predicted by the model were in good agreement with the measured data. The default parameters of the A S M No. 1 model, as reported by Henze et al. (1987), were used; however the denitrification reduction factor ( r | H N 0 3 ) was increased to 0.90 due to the high concentration of activated sludge, in the reactor. In this case, it was assumed that due to membrane filtration, a fraction of SMPs was retained in the reactor while no particulates or biomass were wasted in the effluent. Gehlert and Hapke (2003) presented an approach to modeling of a continuous aerobic M B R for C O D removal from municipal wastewater. The model was based on the biochemical processes of the A S M 3 model in conjunction with mass balances typical of an M B R running at a constant total suspended solids (TSS) concentration. Stoichiometric parameters were similar to the default values of A S M 3 (Gujer et al., 1999, Henze et al., 2000) while the kinetic parameter values differed significantly as expected by the authors, since kinetic parameter values are closely related to the nature o f the substrate. The model was then validated by comparing the predicted effluent C O D , sludge production and CO2 concentration in the exhaust gas to the experimental data. Research was conducted by Zhang and Hal l (2006) using the wastewater treatment pilot plant at the University of British Columbia ( U B C ) , during which the heterotrophic kinetic parameters of M E B P R and C E B P R UCT-type processes were measured and compared to the values reported in the literature. In general, minor differences were reported between measured parameter values for both processes and they were found to be within the ranges reported in the literature. The study was considered to be the first attempt in the literature to examine some kinetic properties of sludge taken from an M E B P R process. 27 Results presented in the above studies indicate that C O D and nitrogen removal kinetic models available in the literature for conventional systems are capable of representing model kinetics of M B R systems with some changes to model parameters. However, modeling of M B R systems, when integrated with an E B P R process, has not been investigated. It is possible that conventional microbial kinetics may not apply under M E B P R conditions. In turn, more research is needed to examine the validity of conventional E B P R process models in describing microbial kinetic dynamics of M E B P R processes under various operating conditions. In general, the modeling of activated sludge and M B R systems has been the focus of researchers for decades, resulting in many models for these systems. However, complex model structure and limited process data availability have motivated some research attempts on developing standard model calibration techniques for activated sludge models. These techniques are intended to provide some guidelines for practical applications of the complex activated sludge models, even by the inexperienced modeler, to achieve quality calibration. The following section describes the main challenges associated with the dynamic calibration of activated sludge systems and provides a summary o f the different research studies in this area. 1.2.4 Calibration of Dynamic Models for Activated Sludge Systems Dynamic model calibration of activated sludge systems has gained the interest of researchers for a number of years, due to the many challenges associated with it. A number of calibration protocols have been established and used to ensure reliable model calibration. This section describes the challenges and the dynamic characteristics of activated sludge systems and describes some of the calibration techniques presented in literature. 28 1.2.4.1 M o d e l C a l i b r a t i o n C h a l l e n g e s There are a number of challenges faced in calibrating activated sludge dynamic models. Some of these challenges include the following. • The complexity of the system is reflected in models developed for activated sludge processes, resulting in over-parameterized models, e.g. 60 parameters for the T U D P model, making it difficult to calibrate without a thorough knowledge of the model. • It is recommended to measure some kinetic parameters using batch tests which could be a source of error in interpretation and extrapolation of these results to continuous full-scale systems. • Models are usually not well identifiable, so that different combinations of parameter values give almost identical model behavior. Therefore the data must contain enough information to make all the parameters identifiable (Weijers, 2000). • Models usually incorporate state variables that are not easily measurable in the lab, which adds to the complexity of model calibration due to limited data versus complex model structure. • Initial conditions for dynamic simulations of activated sludge systems have to be known or estimated accurately, most importantly for suspended components since they exhibit slow dynamics. Incorrect estimates may result in biased estimates of the biological parameters. • The hydraulics of biological reactors needs to be determined to detect non-ideal behavior of the system. • Activated sludge systems usually operate under pseudo-steady state conditions, making it difficult to collect real dynamic data for model calibration. • In spite of the availability of on-line in situ nutrient sensors for soluble nitrate, ammonium and ortho-phosphate, they are not used frequently. Furthermore, the determination of sludge concentration of total phosphorus and nitrogen content is still based on manual analysis of samples in the laboratory. A low sampling frequency is usually used for data measured off-line due to the laborious analytical techniques involved. 29 • Manual analytical techniques are very complex, resulting in large measurement errors. A s a result of these challenges, most modeling studies reported in the literature for full-scale activated sludge plants were usually performed using data collected at pseudo-steady state. Model calibration was mainly based on expert knowledge and driven by ad hoc calibration approaches. Influent wastewater characterization and parameter estimation of the biological processes mainly rely on respirometric measurements, which are laborious and time consuming. Dynamic modeling of activated sludge systems is usually limited to batch scale processes or using plant data collected over a short period of time exhibiting daily dynamics of the plant, rather than slower biological process dynamics. In addition to the literature studies sited in the previous sections, there have been a number of developments concurrent to this research work in the field of membrane bioreactor modeling and performance. A summary of those work can be found in A h n et al. (2006), Al -Malack (2006), Jiang et al. (2005), Jiang (2007), Grelier et al. (2006), Manser et al. (2005a), Manser et al. (2005b), Manser et a. (2006), Masse et al. (2006), Sperandio et al. (2005). 1.2.4.2 O b t a i n i n g R e a l D y n a m i c D a t a i n A c t i v a t e d S l u d g e Sys tems Process evaluation for the purpose of process control design and optimization is usually performed in a simulation environment using dynamic mechanistic nonlinear models of high order to examine process dynamics under several conditions. The mechanistic model should reliably reflect real plant behavior under these conditions. In order to satisfy these requirements, a model should adequately capture the relevant process characteristics for the intended purpose and it must be accurately calibrated to the real plant dynamic behavior (Olsson and Newell , 1999). In order to capture the real plant dynamic behavior, it is critical to collect process data that reflect the main process dynamics. The following section describes the main dynamic conditions imposed in activated sludge systems. 30 There are two kinds of dynamics imposed in W W T P : 1- daily dynamics imposed by the daily variation in the influent flowrate and concentration profiles (short-term dynamics) and 2- slower process dynamics due to biomass microbial population changes imposed by different factors such as temperature, p H , substrate availability, etc. (long-term dynamics). The distinction between these two process dynamics is not very clear and too often researchers claim to use dynamic data to calibrate and validate kinetic parameters of activated sludge models when the data exhibit short-term dynamics, only while the population dynamics remain at pseudo-steady state conditions. Vanrolleghem et al. (2003) emphasize this by indicating that a dynamic calibration o f the model necessitates the availability of dynamic influent concentration profiles. The authors further state that a larger amount of data and optimized sampling, both provide sufficient information for the understanding of the system behavior and allow a precise estimation of parameters which should be considered when collecting dynamic system data. Some of the conditions required for collecting proper dynamic data for calibrating and validating dynamic models according to system identification theory are listed below. Proper Excitat ion In order to achieve proper calibration of dynamic models, dynamic data capturing the dynamic changes of the kinetics are needed. A s stated by Meijer et al. (2001), pseudo-steady state systems are not suitable for evaluation of kinetic parameters in activated sludge models. Therefore proper excitation of the model dynamics is required. This is achieved by a change in the input signal, e.g. a step change, while sampling the process over a period o f time, at least twice the dominant time constant of the process, to ensure capturing all process dynamics as a new steady state condition is reached. Spectral analysis can be used to ensure proper excitation of the process. For a given signal, the power spectrum gives a plot of the portion of a signal's power (energy per unit time) falling within a given frequency range. The most common way of generating a power spectrum is by using a discrete Fourier transform. 31 When a dynamic process is properly excited, the process input/output spectrum, computed at frequencies between (0 - n), contains large amplitude values over a wide range of frequencies. However, a system at steady state or pseudo-steady state would only be excited, or show large amplitude values, at a narrow range of frequencies and therefore, the data collected in this case would not be useful for calibrating dynamic models (Ljung, 1999). Details on spectral analysis can be found in Soderstrom and Stoica (1989). In activated sludge systems, the influent is the major source of dynamic disturbance to the process kinetics. However, using data collected over a short time period at pseudo-steady state for model calibration, which is the case in most studies reported previously, would not ensure identifiability of kinetic model parameters. Therefore, cases of low substrate supply or limited process conditions (such as V F A or P limitation) wi l l result in proper excitation o f E B P R process kinetics. Only in the study by Meijer et al. (2002b) were suitable dynamic data used in model calibration. More research is clearly required to further verify and validate model applications under other dynamic conditions. S a m p l i n g F r e q u e n c y Another critical consideration when collecting dynamic data is the selection of a proper sampling frequency. Data collected at a low frequency wi l l result in losing important dynamic information, referred to as aliasing, while high sampling frequency may result in noisy data not suitable for calibration. In activated sludge systems, as shown earlier, data collection is troublesome and usually occurs at low frequencies. Recommended sampling frequencies should be in the range of one-tenth to one-fourth of the dominant time constant of the process of interest (Ogunnaike and Ray, 1994). A s a result, careful examination of the proper sampling frequency is required prior to sampling. S a m p l i n g D u r a t i o n The sampling duration refers to the period of time that samples are collected from the process in order to collect the data required for process modeling. A n acceptable sampling duration is in the range of 2 to 3 times the dominant time constant of the 32 process. Since the SRT is the dominant time constant in activated sludge systems, the process needs to be sampled for 2 to 3 times the process S R T to ensure capturing the slow sludge dynamics over time. However, in some of the studies reviewed previously, the process sampling was carried on for about 2 days when the process SRT was in the range of days. This means that the process dynamics were not captured in these studies and so the data presented the pseudo-steady state behavior of the process. 1 . 2 . 4 . 3 D y n a m i c C a l i b r a t i o n T e c h n i q u e s fo r A c t i v a t e d S l u d g e Sys tems Research efforts have been directed towards developing systematic dynamic calibration protocols of activated sludge systems to provide a standardized procedure for simulation studies. These protocols are aimed to increase the quality and reliability o f mathematical modeling in wastewater treatment. Some of the proposed systematic calibration protocols in the literature include: the B I O M A T H protocol by Vanrolleghem et al. (2003), the S T O W A protocol by Hulsbeek et al. (2002), the Hochschulgruppe (HSG) guidelines by Langergraber et al. (2004) and the W E R F protocol by Melcer et al. (2003). Sin et al. (2005) presented a critical comparison of these systematic calibration protocols referred to as a S W O T (Strength, Weakness, Opportunities, Threats) analysis. The authors provide a comprehensive description and analysis of the calibration protocols. A s a result of the S W O T analysis, the authors indicated that overall, the S T O W A protocol, developed in the Netherlands and proposed by Hulsbeek et al. (2002), appeared to be the most straightforward, practical and easy to follow and implement protocol. The S T O W A protocol was found to be most suited for practical applications to perform a good quality calibration. Furthermore, the authors identified few areas of deficiency in the S T O W A protocol. Some of the points raised include the following. 1. The influent characterization in the S T O W A protocol is based on combined B O D and physical-chemical measurement which gives reproducible/consistent results, but the B O D method leads to a major uncertainty with the determination of the inert particulate fraction. However, Meijer et al. (2001) addressed this issue by introducing the calibration parameter, fxs in, for the inert particulate fraction so the B O D method would not be required. 33 2. Steady state calibration o f the model, to obtain the initial set of model parameters to be used for the dynamic calibration of the model, was not acknowledged in the main structure of the model calibration of the S T O W A protocol. 3. The biomass characterization, i.e., the determination of the initial autotrophic and heterotrophic biomass concentrations in the W W T P for the dynamic model calibration, was not specified in the protocol. 4. The protocol did not provide any guidelines or remarks concerning the design of the dynamic measurement campaigns, even though this is the most expensive aspect of a model calibration study. 5. The protocol puts little emphasis on mathematical and/or statistical methods such as optimal experimental design (OED), used in the B I O M A T H protocol, that can be used for better design of measurement campaigns and lab-scale batch tests. 6. The manual calibration procedure proposed by the protocol was generalized to all full-scale applications. Sin et al. (2005) indicated that it is too difficult to expect that the fixed model parameter subsets, proposed by the S T O W A protocol to be used for dynamic model calibration, remain valid for different W W T P s . It was suggested by Sin et al (2005) to check whether these proposed parameter subsets remain indeed the most sensitive in each calibration study. Sin et al. (2005) provided some recommendations to contribute to the further advancement of the calibration practice of activated sludge models. These suggestions include the following. • The limitation in transferring of results obtained from lab-scale experiments to full-scale models is a significant problem. Therefore, there is a need to further develop and standardize the experimental methodologies for data collection at lab-scale. • The design of data collection (measurement campaigns) needs to be based on more advanced mathematical tools rather than expert knowledge. The authors indicated 34 that by designing clever data collection campaigns (information rich but low cost), a considerable cost reduction can be achieved for the overall calibration study. • Full-scale activated sludge models are rather complex models to identify with limited and information-poor data. This aspect of model calibration may be improved by applying partially automated calibration procedures such as the selection of an identifiable subset and the estimation of the identifiable parameters using mathematical/statistical approaches (Weijers and Vanrolleghem, 1997; Brun et al., 2001; Brun etal., 2002). In general, most of the calibration studies dealing with activated sludge models are done manually, due to the over-parameterized nature of these models with respect to availability of dynamic data (Vanrolleghem et al., 2003). This method relies solely on process knowledge and reasoning; however, it does not guarantee obtaining the set of parameters resulting in the best fit o f measured data and does not take all possible parameter interdependency into consideration when performing these calibrations. Beyond manual calibration, there are a number of nonlinear parameter estimation algorithms based on minimizing a nonlinear objective function. Dochain and Vanrolleghem (2001) provide a good description of common algorithms used for this purpose. However they indicate that no perfect minimization algorithm for nonlinear objective functions exists and consequently, finding the global minimum for nonlinear problems cannot be guaranteed. Therefore, care should be taken when choosing the initial starting values for the parameters to avoid ending up with a local minimum of the objective function. 1.2.4.4 Model Validation for Activated Sludge Systems Limited attention is usually paid to model validation when modeling activated sludge systems. Model validation is required to ensure that the calibrated dynamic model is capable o f predicting the behavior of the system under different conditions. According to system identification theory (Soderstrom and Stoica, 1989), when performing system modeling using measured data from a process, two-thirds of the data set should be used 35 for model calibration while one third should be utilized for model validation. However when reviewing the different studies reported in previous sections, it was noted that usually, model validation was not considered. Furthermore, in cases where model validation was performed, for example in the work by Brdjanovic (2000), batch test data were used to validate the calibrated T U D P model o f a full-scale plant. In conclusion, the literature review reflects the need for practical, stable and robust model calibration procedures suitable for complex activated sludge systems which address dynamic data requirements within process limitations. 1.2.5 Process Design and Control Studies in EBPR Systems The E B P R and nitrogen removal processes are competing processes, which adds to the complexity o f activated sludge systems. A n example o f the competing nature o f the two processes is the nitrate load to the anoxic zone. For better nitrogen removal, the nitrate load to the anoxic zone needs to be increased to remove as much nitrogen as possible via denitrification. However, high nitrate concentrations in the anoxic zone may be carried over to the anaerobic zone, via the recycle flow, and negatively impact the phosphorus release process. In order to overcome these operational challenges in biological nutrient removal (BNR) processes, usually these systems are oversized to avoid system failures. In other cases they are designed based on process knowledge and experience rather than based on modeling and simulation results. Therefore, there is a need to qualitatively and quantitatively study the effect of various operational and design parameters such as zone volume ratios, H R T , SRT, and recycle flows on process performance o f E B P R and B N R systems. When dealing with M E B P R systems, another dimension of complexity is added since very little is known about this system and the affect of introducing the membrane unit on the process operation and stability. Ramphao et al. (2005) presented the impact of the membrane system on the B N R process in terms of mass distribution variations that affect the design of membrane enhanced B N R systems. However, the effect o f the membrane 36 system on the process operation and the required design and operational parameters to ensure stable process performance, are yet to be investigated. Process control of B N R processes for removal of organic carbon and nitrogen compounds has gained the attention of researchers recently and it is slowly making its way to practice as well (Ingildsen, 2002a). However automatic control of E B P R processes is rather sparse, due to the operation-related difficulties of applying process control resulting from poor process understanding. It is particularly difficult to implement automatic control in E B P R plants due to the competing nature of the phosphorus and nitrogen removal processes, as described earlier, making it challenging to control both simultaneously. 1.2.5.1 Control Challenges of Unit Operation in EBPR Process Some of the operational objectives and challenges of E B P R system control design are described below. VFA Concentration Control The key mechanism in the anaerobic zone, as described above, is the uptake of the volatile fatty acids ( V F A ) by the bio-P bacteria which store them as P H A . The energy to do this comes from the simultaneous release of ortho-phosphate into solution from stored polyphosphate. The concentration and the type of carbon source available to the bio-P bacteria may be a limiting factor. V F A promote biological phosphorus uptake and therefore, a sufficient concentration o f V F A is important for E B P R . Usually when the concentration of V F A in the influent stream is insufficient, an external supply of acetate is added to compensate. However the determination of this amount is dependant on different factors such as: the influent phosphorus concentration, V F A in the influent stream, and the amount of stored phosphorus in the cells of sludge recycled from the anoxic to the anaerobic zone. Obtaining accurate online measurements of these variables and determining the required external acetate concentration can be very challenging. 37 Aerobic Recycle Con t ro l The aerobic recycle is the recycle of the mixed liquor from the aerobic zone to the anoxic zone o f the UCT-type E B P R process, as shown in Figure 1.4. A conflict exists between the phosphorus and the nitrogen removal process with regard to this recycle flow, which makes controlling o f this flow the most challenging and complicated control problem in the system. Anoxic Recycle Influent Return Activated Sludge y Figure 1.4. The UCT-type B N R process exhibit operational challenges (Adapted from Metcalf and Eddy, 2003) A s mentioned earlier, ammonia is oxidized to nitrate in the aerobic zone and the nitrate is recycled to the anoxic zone to be reduced to nitrogen gas by the oxidation o f organic carbon; via biological denitrification. In order to reduce effluent nitrate concentration, and hence total effluent nitrogen, it is desirable to increase the aerobic recycle rate. On the other hand, since the dominating organisms in the denitrification process are the heterotrophs which reduce nitrate in an (ideally) oxygen-free environment, the dissolved oxygen level has to be kept at a minimal in the anoxic zone. Given that the aerobic recycle contains dissolved oxygen, then to reduce the negative effect of oxygen on the denitrification rate it is desirable to decrease the recycle flow. Furthermore, nitrate, which is recirculated back to the anaerobic zone with the return sludge in the conventional system, or with the anoxic recycle in the M E B P R system, may have a detrimental effect on the E B P R process. It is usually assumed that bio-P 38 organisms are not able to compete successfully with other organisms for the readily degradable substrates when oxygen, nitrate or both are present. Hence, in an M E B P R system, the aerobic recycle flow rate should be chosen carefully to ensure that complete denitrification is achieved in the anoxic zone to avoid nitrate leaks to the anaerobic zone. It is apparent from the above description that the aerobic recycle flow affects different processes and it is challenging to control this variable to optimize these processes. In order to reduce this effect, other variables can be considered such as the volume ratio between the anoxic and aerobic zones or the dissolved oxygen concentration in the aerobic zone. DO Control In order to achieve an adequate environment for nitrification and phosphorus uptake in the aerobic zone, a sufficient concentration of dissolved oxygen is required. Furthermore, aeration is used to provide the turbulence necessary to reduce fouling of the membrane unit. A sufficient concentration of dissolved oxygen is also required to avoid the formation of filamentous growth. However, as mentioned above, excessive aeration wi l l introduce high concentrations of oxygen in the anoxic zone, which may in turn hinder the denitrification rate. Thus D O control is a necessity in activated sludge systems to optimize these processes. Control of Carbon Source for Denitrification The denitrification organisms need carbon as an energy source, so it is important to have sufficient concentrations of organic carbon in the system. Insufficient carbon not only limits denitrification, but also negatively influences the biomass floe formation. Carbon is usually delivered with the influent flow. However during low load periods, extra carbon may have to be added, in the form of fermented waste slduge. Furthermore, obtaining a proper carbon dosage is a crucial operational task since a reliable measure of the carbon concentration is required and any excess carbon wi l l have to be removed in the aerobic zone. 39 Control of Waste Sludge Usually the sludge wasting from the system controls the solids concentration in the system and in turn the process SRT. In conventional E B P R systems with the presence of clarifiers, this control becomes more complicated due to the unknown amount of suspended solids leaving the system with the effluent. However in M E B P R systems, a better control over the process SRT can be achieved since all suspended solids are essentially retained in the system. Volume Ratio Control The volume ratio between the anoxic and aerobic zone creates an interesting control problem. In periods of low load the volume of the aeration tank can be reduced while the volume of the anoxic zone can be increased to achieve better denitrification, thus lowering the total effluent nitrogen. However, in periods of high loads, larger aeration tanks are required to ensure efficient nitrification and hence low effluent ammonia. It is apparent that satisfying all these requirements is a true operational challenge, and obviously there are some competing and conflicting goals. There remains no obvious way to uniquely define how to optimize an M E B P R system. A number of studies have been carried out to address process control in activated sludge systems and some of these studies are reviewed in the following section. 1.2.5.2 Process Control Applications in Activated Sludge Systems Most of the reported process control studies in the literature are for carbon-only or carbon and nitrogen removal systems. Only a few applications looked at process control of biological nitrogen and phosphorus removal systems. In carbon-only removal activated sludge processes, the automatic control of dissolved oxygen (DO) concentration has been recognized as rewarding and meaningful, both from an economic and biological point of view. Some successful conventional D O control schemes, such as proportional-only and proportional, integral and derivative (PID) 40 control by manipulating the air flowrate in full-scale activated sludge plants have been reported (Lee et al., 1999). Other conventional control strategies reported in the literature include proportional and integral (PI) or PID control of sludge concentration in the aeration tank and solids inventory in the system by manipulating waste activated sludge flow or return activated sludge flow from the clarifier. Process control of activated sludge systems with biological carbon and nitrogen removal becomes more complex due to the low food/micro-organism ratios maintained. Vrecko et al. (2001) performed an analysis of activated sludge systems using the benchmark process developed by Alex et al. (1999). The benchmark process includes two anoxic zones, three aerobic zones and a settler. The analysis has shown that the benchmark process is a non-linear and multivariable process with strong interactions between process variables. The benchmark therefore seems to require complex control. However, it was suggested by the authors that the control-problems can be simplified since some sub-processes in the benchmark have main time constants of different orders of magnitude. These sub-processes are weakly coupled and can therefore be controlled independently from each other i f the controllers are tuned in such a way as to preserve the time constants of different orders in the closed-loops as well (O 'Rei l ly and Leithead, 1991). The study also suggested that the control problem can be additionally simplified i f sub-processes are chosen so that they can be successfully controlled with linear controllers. Moreover, the study indicated that the most appropriate linear controllers for controlling the plant are PI controllers because they are very simple and most often used in practice. Based on the these results, Vrecko et al. (2002) applied a simple control strategy to the benchmark proposed by Alex et al. (1999) and assessed it in simulation. The control strategy was used to control the dissolved oxygen concentration in the aerobic zone at 2 mg/L by manipulating the air flow using a PI controller with a time constant of a few minutes. The purpose o f this control loop was to supply enough oxygen that maximum nitrification and organic matter biodegradation was achieved. The objective of the next control loop was to retain the desired amount of biomass in the basin of the plant using the suspended solids concentration in the aerobic zone as the controlled variable with a 41 set-point of 4500 mg/L. The waste sludge flow was a manipulated variable employing a PI controller. The main time constant of this sub-process was several days. The third control loop was also based on a PI controller which was used to supply the required nitrate from the aerobic zone to the anoxic zone so that maximum denitrification was achieved. This loop controlled the nitrate concentration in the anoxic zone at 2 mg/L by manipulating the recycle flow from the aerobic to the anoxic zone, which has a main time constant of about an hour. The sludge recycle flow from the clarifier, was manipulated by feed-forward control, which set the sludge recycle flow at a chosen ratio of 1.5 of the influent flow to additionally adjust the amount of biomass in the zones depending on the influent flow. The authors of this study indicated that this control strategy resulted in a lower number of effluent violations as well as lower energy consumption and lower sludge production when compared to results published up to that date. This control strategy is quite simple and appealing. However, care should be taken when applying it to an E B P R system, since a nitrate concentration o f 2 mg/1 in the anoxic zone, for example, would definitely affect the anaerobic phosphorus release in the anaerobic zone. Furthermore i f this set-point is lowered, it w i l l be difficult to obtain reliable nitrate concentration measurements using an on-line sensor and this may cause the control to be unstable. Dissolved oxygen concentration also needs to be determined carefully for an E B P R system. A s for model-based controllers, which utilize a process model to determine the controller action, Ingildsen et al. (2002b) developed a feed-forward control strategy using on-line measurements of influent ammonia to control aeration in a nitrogen removal system by controlling the D O set-point. The model used treated ammonia as a tracer (no reaction) for simplicity. Work by Galluzzo et al. (2001) indicated that one important limitation to D O control in a nitrogen removal system is the use of only one manipulated variable, the air flowrate. They indicated that the possibility of using an additional manipulative variable like the recycle flowrate to the anoxic reactor or the aerator volume could make the control more effective. A study by Samuelsson and Carlsson (2002) confirmed that a combination of 42 D O control and volume control could be an interesting alternative to pure D O control in an activated sludge process for nitrogen removal. In their study, a method for controlling the aerobic volume in a nitrification/ denitrification process employing a dynamic model-based strategy, using both traditional feedback and feed-forward information was suggested. The objective of the control loop was to allow the process to quickly respond to changes in influent load variations. During periods of high ammonium load, a larger aeration volume could be used, and during periods of low load, the aeration volume could be decreased, allowing a larger volume to be used for denitrification. It was further stated that it would be interesting to implement the suggested strategy in a pilot plant after performing more complete analysis and simulation experiments. For biological phosphorus and nitrogen removal systems, Galluzzo et al. (2001) designed an expert control structure for controlling the D O in the aeration tanks using a supervisory fuzzy controller that determined the set point of an inner D O control loop where an adaptive robust generic model control ( A R G M C ) was used. The discrete process model suggested by Lindberg and Carlsson (1996) was used for the D O dynamics. The control scheme was tested in simulation. Lee et al. (1999) designed an overall control system for a biological phosphorus and nitrogen removal activated sludge process system using a multi-loop control system. They extended the generic model control ( G M C ) control strategy to include a class of distributed parameter models which resulted in the generic distributed parameter control ( G D P M C ) law which was based on a modified, reduced-order A S M No. 2 presented by Henze et al. (1994) resulting in nonlinear controllers. The control strategy consisted of three loops, one for each of anaerobic, anoxic and aerobic zones. In the anaerobic zone, a multivariable G D P M C design was used to maintain the soluble phosphorus and acetate concentrations required to ensure a low effluent phosphorus concentration. The sludge recycle flow was manipulated to control the concentration o f ortho-phosphate and an external acetate flow was used to control the V F A concentration. In the anoxic zone, a single-input-single-output (SISO) G D P M C controller was designed to control the nitrate concentration by manipulating the aerobic recycle flow. In the 43 aerobic zone, a supervisory-type PI feedback controller was used to adjust the D O set-points in the multivariable G D P M C D O profile controller such that the outlet ammonia was controlled. The G D P M C controller was used to adjust the air flowrate. When these loops were run together to perform total control of nutrients for the B N R activated sludge plant using multi-loop control, results showed that the system was robust against plant-model mismatch of up to 25% changes in three uncertain kinetic parameters. Performance of the controller deteriorated as the plant-model mismatch increased. Despite the results, the authors indicated that difficulties may arise in practical implementation o f this strategy in view of the limitation on the manipulated variables (spatial air supply) and the conflicting requirements of the controlled variables (soluble nitrate, ammonia and phosphorus concentrations). A more recent study by the same group (Lee et al., 2004) presented a robust multivariable controller design for robustly stabilizing the effluent nutrient concentrations of a biological phosphorus and nitrogen removal system. In general, very few studies can be found in the literature which address the issue of overall process control of continuous biological phosphorus and nitrogen removal systems, despite the potential operational and economical benefits of it. However the proposed strategies depend on complex control algorithms that are yet to be tested for applicability at a full-scale level. More research is needed to develop simple control strategies that are easy to implement in real applications, but yet are complex enough to ensure reliable, efficient and robust operation o f E B P R systems. The challenge becomes even greater when dealing with control design for M E B P R processes, since little is known about the impact of the membrane unit on the overall system performance and consequently, its effect on the control system design. 44 1.3 Research Objectives The presented literature review clearly shows that the enhanced biological phosphorus removal (EBPR) process with nitrogen removal is a complex system with competing and conflicting goals and slow dynamics. Since it is difficult to study the system behavior via an experimental set-up, a reliable process model that reflects the real plant behavior under various operating conditions is required. Once this model becomes available, it can be used to closely study the process behavior, variable interactions and the effects of different design and operational conditions on the system performance. The result of these simulation studies can then be used to develop a set of recommended design and operating conditions that result in stable and efficient process performance. The model can also be used in process control studies to develop and test different control strategies aimed at achieving a reliable and robust process operation under various process disturbances. A number of activated sludge models are available in the literature, however, little is known about the applicability of these models to describe the M B R system behavior and kinetics. Furthermore, careful investigation of reported modeling studies revealed that most of these models have not been validated under real dynamic conditions in continuous flow enhanced biological phosphorus removal systems. Moreover, the complexity of these systems is reflected in these models, resulting in over-parameterization and highly complicated models that are very difficult to calibrate against experimental data. Therefore, almost all of the calibration studies dealing with activated sludge models have been done manually (on ad hoc basis). Hence there is a need to develop a model calibration and validation protocol in line with standard system identification techniques, taking into account system complexity and yet ensuring reliable model predictions under different conditions. Furthermore, reports on studies addressing the design and operation of the E B P R process for effective process performance under various conditions are sparse in the literature due to system complexity and conflicting priorities. The M E B P R adds another dimension of complexity to the system, making such studies very rare. 45 The main objective of the present study was to develop guidelines for the design and operation of the M E B P R process to fully utilize the available process capacity, while producing high quality effluents. In order to achieve this objective, the following tasks were identified. 1. Complete a comprehensive review of activated sludge models available in the literature to determine their suitability and applicability for the M E B P R process and select the most reliable one for use in this study. 2. Perform steady state simulation of the M E B P R process using pseudo-steady state data, after checking for data reconciliation, using the selected model and the chosen simulation environment to ensure that the model structure is capable o f presenting the process behavior of the M E B P R process. Furthermore, this step was carried out to obtain a set of initial model parameters to be used in the dynamic modeling stage of the study. 3. Develop a reliable protocol for dynamic calibration and validation of activated sludge systems in line with standard system identification techniques, while taking into account model complexity and data collection limitations o f these systems. The calibration technique should be consistently applicable to any system, resulting in reliable model predictions under various conditions. 4. Test the developed protocol in modeling the M E B P R process using the selected activated sludge model to ensure the practicality and validity of the proposed method and compare the results of the dynamic modeling to that of steady state modeling. 5. Use the model parameters calibrated for the M E B P R process to predict the process behavior o f the conventional enhanced biological phosphorus removal ( C E B P R ) process. This step was carried out to identify i f there was a difference in biological system between the M E B P R and the C E B P R processes due to the difference in the final liquid-solid separation technology employed. 46 6. Use the calibrated model in simulation studies for the M E B P R process behavior under high flowrates and various design and operating conditions to understand the effect of these variables on system performance. Then develop a set of guidelines based on the results of the simulation studies for the design and operation of the M E B P R process under high loads to achieve high quality effluents. 7. Propose a control strategy suitable for the M E B P R process, based on simulation results and experience gained during the operation of the U B C M E B P R process. The control strategy should be simple enough to ensure its applicability in industry but yet addresses the needs of the system to achieve reliable and robust process operation. 47 2 M A T E R I A L S A N D M E T H O D S 2.1 U B C W a s t e w a t e r T r e a t m e n t P i l o t P l a n t Process The biological phosphorus and nitrogen removal wastewater treatment pilot plant at the University of British Columbia has been operated and maintained by the Department of C i v i l Engineering for a number of years. The pilot plant was used for conducting various environmental research projects. It consists of two parallel trains of the University of Cape Town (UCT) type wastewater treatment process. One of the trains was changed to a membrane enhanced biological phosphorus removal process ( M E B P R ) process by substituting the secondary clarifier with a ZeeWeed ultrafiltration membrane unit (GE Water and Process Technologies - Z E N O N Membrane Solutions, Oakville, Ontario) placed in the aeration tank. ZeeWeed is a hollow fiber membrane assembled in a shell-less module that is immersed directly in the mixed liquor. The membrane module has a surface area of about 12.2 m 2 and a nominal pore size of 0.04 micrometers. Filtration is from the outside-in under a negative pressure applied to the permeate side. The hollow fibers are continuously scoured by coarse bubble aeration. The other train was a conventional enhanced biological phosphorus removal ( C E B P R ) system consisting of the biological treatment process and a secondary clarifier for the final solids-liquid separation. A l l settled solids were returned to the aerobic zone with the sludge recycle stream as shown in Figure 2.1. It was decided to return the settled sludge from the conventional train clarifier back to the aerobic zone, rather than to the anoxic zone to achieve identical operating conditions for both the M E B P R and the C E B P R treatment trains to accommodate the requirement for a comparative study for another research project. Both treatment processes were identical in process configuration and operation with the exception of the solids separation stage. Raw sewage from a nearby U B C main sewer is pumped, at different times of the day, into two storage tanks at the pilot plant. The sewage is kept well mixed in the tanks using mixers to avoid solids settling at the bottom of the tank. The raw sewage is then pumped to a primary clarifier which also acts as a common fermenter. Overflow from the primary clarifier, primary effluent, serves as the influent to both biological treatment systems. The 48 influent enters the anaerobic reactors and overflows to the anoxic and then the aerobic reactors. For the M E B P R process, the air needed in the aeration stage is introduced into the reactor at the membrane module to create the required agitation to reduce membrane fouling. Membrane filtered effluent exits the system after the aerobic stage. Mixed liquor is recycled from the aerobic reactor to the anoxic reactor. A recycle stream also leaves the anoxic reactor to the anaerobic reactor. In the C E B P R train, the stream leaving the aerobic zone flows to the clarifier where the final effluent overflows from the top to a holding tank and is discharged to the sewer. The settled solids are withdrawn from the bottom of the clarifier and recycled back to the aerobic zone. The sludge was recycled back to the aerobic zone, rather than the anoxic zone, in an attempt to achieve similar biomass distribution characteristics in both the M E B P R and C E B P R trains to accommodate a comparison study between the two systems for another research project, which required identical operating conditions for both treatment trains. Process reactor volumes are given in Table 2.1. More details about the U B C wastewater treatment pilot plant process description are given by Monti (2006a). Table 2.1. Design specifications of the biological zones of the U B C wastewater treatment pilot plant Biological Reactor MEBPR Process CEBPR Process Anaerobic zone 234 L 240 L Anoxic zone 585 L 618 L Aerobic zone 1311 L 1369 L Total Reactor volume 2130 L 2227 L 2.2 Process Operating Conditions The M E B P R process was started on December 19, 2002 with a sludge retention time (SRT) of 20 days and a hydraulic retention time (HRT) o f 10 hours. On February 13, 2003, the C E B P R process was started with the same operating conditions. 49 Problems were encountered due to sludge bulking in the secondary clarifier o f the C E B P R process at the beginning of March of 2003, so it was decided to reduce the process SRT to 12 days on March 17, 2003, to ease the load on the clarifier. The SRT of the M E B P R process was changed as well to keep similar operating conditions on both sides. The process H R T was maintained at 10 hours for both systems for the duration of the study. F i g u r e 2.1. U B C wastewater treatment pilot plant process The SRT was calculated based on the solids concentration in all zones. The total suspended solids (TSS) concentration was measured three times a week and the sludge wasting from the aerobic zone, was calculated based on these measurements to obtain the desired SRT. Part of the sludge wasting was done automatically using a controlled flow pump while the rest was wasted manually once a day. For the C E B P R process, the sludge blanket in the secondary clarifier was included in the calculation of the SRT since it acted as an extra denitrification compartment. The following expressions were used for the 50 calculation of the daily wasting volume (Qwaste) withdrawn from the aerobic zone of the two systems as given by Mont i (2006a): M E B P R : i=l XAER S R T (2.1) C E B P R : £ v , - x i + v C L - x C L i=l Q INF X EFF XAER ' S R T X (2.2) AER where V ; and X i are the volume and solids concentration in each reactor zone respectively, V C L and X C L are the volume and solids concentration of the clarifier, X A E R is the solids concentration in the aerobic zone, QINF is the process influent flowrate, and XEFF is the suspended solids concentration in the effluent. Figure 2.2 shows the TSS concentration in all biological zones, anaerobic ( A N A ) , anoxic (ANO) and aerobic ( A E R ) for both systems. The solids distribution in both processes was different due to the incorporation of the sludge blanket as part of the active biomass in calculating the process SRT. The M E B P R process was always operated with higher sludge concentrations in all biological zones. The sudden drop in the TSS concentration shown in Figure 2.2 for the C E B P R process on August 20 and 27 for example, are due to incidents of accidental sludge loss, but the system recovered quickly there after. 51 7 4-- « - M E B P R _ A N O -hr MEBPR_AER - £ - C E P B R _ A N A C E B P R A N O - 0 - CEBPR AER MEBPR ANA 6 + 5 + § A s 4 34 2 1 0 ^ - r - -nr- r - - r - - r - - I 29-Jan-03 20-Mar-03 9-May-03 28-Jun-03 17-Aug-03 6-Oct-03 Date Figure 2.2. Total suspended solids in all biological zones during the experimental phase in the M E B P R and C E B P R processes The membrane system was operated at a constant flux of 22.8 L/m 2-hr at the process H R T of 10 hours. A timer was set up so the membrane was back-flushed with effluent for 30 seconds after every nine and a half minutes of suction to reduce fouling. When the trans-membrane pressure approached the recommended limits o f operation o f 20 in H g (67.7 kPa), due to fouling, the membrane module was replaced by a clean one and the fouled membrane was cleaned and soaked in a bleach solution for the next use. The membrane cleaning method is described in more details by Monti (2006a). The process influent storage tanks were filled using an automatic sump pump twice a day, in the morning (around 9 am) and evening (around 9 pm). However, beginning on September 3, 2003, the storage tanks were filled four times a day to better capture the daily concentration variation. The filling of the tanks started at 7 am and filling events were scheduled 6 hours apart. The last filling was in the middle of the night when the concentration of sewage was the lowest for the day. A s a result, the change in the pumping regime resulted in lower overall organic loading rates. 52 Data used in this study were collected from the M E B P R and C E B P R processes for the period between February and October of 2003. During this period, the influent flowrate of the M E B P R system was set to 3.55 L/min, while the influent flowrate to the C E B P R process was set to 3.71 L/min. The slight difference in the flowrates was introduced to account for the slight difference in the biological reactor volumes to ensure similar H R T values in both systems. The anoxic recycle flow for both systems was equivalent to the influent flowrate. The aerobic recycle flow was changed during the course of the experiment since high concentrations o f nitrate were observed in the anoxic zone of both systems. Therefore, the aerobic recycle flow was reduced to reduce the anoxic nitrate concentration and avoid leakage of nitrate to the anaerobic zone for its negative impact on the phosphorus release mechanism. Table 2.2 lists all operating conditions for the experimental period. Table 2 . 2 . Experimental operating conditions for the C E B P R and M E B P R Processes HRT SRT Influent Influent Anoxic Aerobic RAS S Date (Days) (Days) Flow Flow Recycle Recycle Recycle (L/min) (L/min) Ratio Ratio Ratio Both Both MEBPR CEBPR Both Both CEBPR Start* - 03/04/03 10 20 3.55 3.71 1 1 I 03/04/03-03/30/03 10 12 3.55 3.71 1 1 1 03/30/03-09/11/03* 10 12 3.55 3.71 1 2 1 09/11/03-09/15/03 10 12 3.55 3.71 1 1.6 1 09/15/03-10/22/03 10 12 3.55 3.71 1 1 1 * The MEBPR process was started on 12/19/02 and the CEBPR process was started on 02/13/03 * Starting 09/03/03, the storage tank was filled 4 times a day. $ RAS is the return activated sludge The process temperature, p H and dissolved oxygen (DO) concentration were measured daily. Temperature and pH were measured using a portable V W R probe, while D O concentration was measured by a dedicated in situ Y S I probe. The temperature was in the range of 16 - 24 °C through the course of the experiment. Both systems experienced the same environmental conditions as shown in Figure 2.3. The p H in all zones was measured and approximately 1.5 kg of sodium bicarbonate was added once a day to each feed storage tank to prevent the aerobic p H from dropping as a result of the nitrification activity and related alkalinity consumption. Figure 2.4 presents the aerobic p H for both systems. The p H values in the aerobic zone were in the range o f 6.3 - 7.3, but both 53 systems were mostly operated at a pH value that was above 6.9. It should be noted that for the period between M a y 26 and August 18, it is believed that the pH probe was not calibrated and therefore pH values seem to be higher than usual and exhibit less variation. The dissolved oxygen (DO) concentration was measured daily as well and the air flowrate was adjusted to maintain a D O concentration of about 3 g 02/m3. The same D O was maintained for both systems as shown in Figure 2.5. 24 15 14 -I 1 , , , , • 19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 7-Aug-03 26-Sep-03 15-Nov-03 (Date) Figure 2.3. Temperatures in the M E B P R and C E B P R processes during the course of the study 54 7.5 7.3 7.1 6.9 6.7 5, 6.5 6.3 6.1 5.9 5.7 4 MEBPR_AER CEBPR AER 5.5 19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 7-Aug-03 26-Sep-03 15-Nov-03 (Date) Figure 2.4. p H in the M E B P R and C E B P R processes during the course of the study o 3 O Q 0.5 0 MEBPR_AER CEBPR AER 29-Jan-03 20-Mar-03 9-May-03 28-Jun-03 17-Aug-03 6-Oct-03 25-Nov-03 (Date) Figure 2.5. Dissolved oxygen concentrations in the M E B P R and C E B P R processes during the course of the study 55 2.3 S a m p l i n g a n d A n a l y s i s During the course of the experiment, the process was monitored closely for irregular behavior (pipe blockage, clarifier bulking or overflow, pump leakage, process failure, etc.) and steps were taken to resolve any problem as quickly as possible to minimize its effect on the process performance and stability. Process flowrates were checked and adjusted regularly to ensure steady hydraulic loads. Sampling was done regularly by taking grab samples of influent (primary effluent), effluent and the three biological zones. Samples were collected and analyzed for ammonium-nitrogen ( N H 4 - N ) , nitrate plus nitrite-nitrogen (NO x -N) , ortho-phosphate (PO4-P), total Kjeldahl nitrogen (TKN) which is the sum of organic and ammonium-nitrogen, total phosphorus (TP) including phosphate, V F A including acetic acid and propionic acid, total chemical oxygen demand ( C O D t o t) , filtered chemical oxygen demand ( C O D s o i ) , which is the C O D of the filtered fraction passing through a 0.04 pm pore-size membrane, and TSS. It was decided not to sample for B O D due to the large error associated with the laborious technique of its measurement and because it was decided to use the method presented by Meijer et al. (2001) for the estimation of the inert and slowly biodegradable particulate fractions of C O D in the influent, which does not require B O D data. Samples taken from the M E B P R and C E B P R biological systems showed no N O 2 - N present in both systems, therefore N O 3 - N was used in the rest o f this thesis to represent N O x - N . Ammonium and nitrate are used in this thesis to refer to ammonium-nitrogen and nitrate-nitrogen for simplicity. Samples were collected and analyzed according to Table 2.3. Influent and effluent samples were collected five times per week, while samples from the biological zones were collected three times per week. However, sampling was performed more frequently during some periods in order to monitor the process more closely when a change occurred in the process. Filtered samples were obtained using a 0.04 micrometer membrane filtration to correspond with the standard definition of the soluble fraction as indicated by Roeleveld and van Loosdrecht (2002). For the effluent of the M E B P R process, samples were 56 collected from the permeate tank directly. Samples collected from the biological zones for C O D t o t were diluted with distilled water using a 1:20 dilution ratio prior to analysis. Sample analyses were performed according to standard procedures ( A P H A et al., 1998). Table 2.3. Sample collection and analysis during the course of the experiment Influent CEBPR MEBPR Anaerobic Anoxic Aerobic Effluent Effluent (Both) (Both) (Both) N H 4 - N V •/ •/ N 0 3 - N •/ PO4-P V V F A T P / T K N •/ •/ C O D t 0 t C O D s o l •/ TSS * Sampling was based on grab samples. 2.4 Quality Assurance and Quality Control of Measurements Measurement techniques were checked for quality assurance and quality control to ensure reliability o f collected data and obtain the coefficient of variation ( C O V ) for measurements. Five samples were collected for the same measurement and the C O V was calculated at follows: C Q y = S.D. of Measurements x l Q ( ) % ( 2 3 ) Average of Measurements Table 2.4 shows the C O V values for the measurement techniques used in this study. Results show relatively low C O V for the soluble concentrations, PO4-P, NO3 -N and N H 4 -N . However V F A measurements were relatively less accurate since the stability of the samples was very low and its concentration varied. The C O D measurement technique was prone to error since it comprised several steps from vial cleaning, reagent preparation, sampling, digestion and spectrophotometer reading. For the total C O D measurements of sludge samples, a dilution ratio of 1:20 was used and as a result, a small error in the measurement could be magnified influencing data accuracy. 57 Table 2.4. Coefficient of variation of measurement data Sample # C O D T O T INF C O D S 0 , INF, C O D S O L M E B P R C O D T 0 , M E B P R C O D , O T C E B P R C O D , O T M E B P R P O 4 M E B P R N O A M B E P R N H 4 M E B P R V F A M E B P R E F F A E R A E R A E R A N A A E R A N A A N A 1 398 100 10.6 3778 2767 4311 14.7 8.9 28.9 9.6 2 417 103 8.0 4630 2767 4790 14.8 9.0 30.1 9.3 3 373 100 10.6 4257 2448 3991 14.8 9.0 28.9 8.8 4 387 122 10.6 4417 2501 3885 14.8 9.0 29.6 8.9 5 384 100 76.9* 4204 4524 16.4* 9.0 29.2 8.5 6 4364 Avg. 392 105 9.9 4257 2621 4311 14.8 9.0 29.3 9.0 S.D. 16.4 9.5 1.3 315 170 • 335 0.0 0.0 0.5 0.4 C O V % 4.2 9.0 13.3 7.4 6.5 7.8 0.3 0.2 1.7 4.8 * Ignored (outlier) The adequacy of the TP and T K N measurements was assessed using a different technique than the one used for the other analyses. In this approach, samples collected from the process, including influent, effluent, and sludge samples were analyzed to determine their T K N and TP concentrations. The samples were then spiked with a known amount of a standard concentration solution of 200 mg/L and re-analyzed. The results of the original samples were compared with the spiked samples to examine the reliability of the test method. Furthermore, a blank sample o f distilled water and another sample of the standard concentration solution of 200 mg/L were also analyzed for verification purposes. Table 2.5 shows the results obtained for these analyses of TP and T K N measurements. Results show satisfactory measurement errors for the samples ranging between 5 to 25%. Table 2.5. Results of Q A / Q C for TP and T K N measurements EFF AER Blank Std. Sol. Test Type CEBPR m * CEBPR 200 mg/L TP TKN TP TKN s TP TKN TP TKN TP TKN Original Trial 1 0.68 1.51 3.36 26.77 129.7 234.2 6 -0.83 184 224.3 Sample Trial 2 0.79 1.45 3.23 28.72 146.7 219.9 5.9 0.78 200 194.4 Cone. Avg. 0.73 1.48 3.30 27.74 138.2 227.1 (mg/L) COV 10.8 2.7 2.7 5 8.7 4.5 Volume of original 9.9 9.9 9.9 9.9 1 1 sample (mL) QA/QC Test with Standard Solution Added Volume of 200 mg/L Std. Sol. (mL) 0.1 0.1 0.1 0.1 1 1 Expected cone. (mg/L) 2.72 3.46 5.26 29.47 169.1 213.5 Measured cone. (mg/L) 2.04 3.62 4.50 31.77 159.7 197.8 Estimated % error -25 5 -14 8 -6 -7 58 2.5 TUDP Model The updated version o f the T U D P model as presented by Meijer (2004) was used for the modeling purpose of this project. The model was implemented in A Q U A S I M 2.0 simulation software. The T U D P model rate equations, components and parameters used in this project are given in Appendix I. 2.6 Influent Characterization Influent characterization was performed according to the standard procedures recommended by S T O W A , the Dutch Foundation for Applied Research on Water Management, and reported by Roeleveld and Loosdrecht (2002) for A S M 2 d with minor differences as described below. 2.6.1 Influent Characterization of COD Fractions The total influent C O D (COD t ot,iNF) was measured directly using the total sample and the soluble influent C O D fraction (COD S OI,INF) was measured after filtering the sample through a 0.04 pm membrane unit. The difference between the total and soluble fractions was considered to be the particulate C O D fraction. The soluble C O D fraction was considered to be the sum of V F A (S A ) , fermentable and readily biodegradable organic substrate (SF) and the inert soluble fraction of C O D (Si) as follows. where S A was directly measured in terms of acetic acid and propionic acid according to S T O W A ' s standards procedures as: The inert fraction, Si, was calculated using the measured soluble effluent C O D (COD S O I,EFF)- In the procedures outlined by S T O W A , the COD S O I ,EFF was corrected with a COD tot ,INF - C O D s o l , l N F + CODpart,rNF (2.4) COD s oi ;rNF — S A + SF + Si (2.5) S A = 1.07 • (Acetic Acid) + 1.51 • (Propionic Acid) (2.6) 59 factor 0.9 due to some production of S i during the treatment process. However in this case, it was assumed that the measured effluent soluble C O D was equal to the influent S i . S, = CODsCEFF (2.7) The fermentable and readily biodegradable organic substrate SF was then calculated as: S F = C O D s o i , r N F - ( S A + S , ) (2.8) The CODPart,[NF in the model was considered to consist of the following model components. CODpart,INF = X H + XAUT + XpAO + XpHA + XGLY + Xpp + X] + X s (2.9) where X H was the heterotrophic microorganism concentration, X A U T was the autotrophic microorganism concentration, X P A o was the phosphorus-accumulating microorganism concentration, X P H A was the concentration of poly-hydroxy-alkanoates, XGLY was the glycogen concentration, X P P was the concentration of poly-phosphate, X s was the slowly biodegradable particulate substrate and X i was the inert particulate organic material. The concentrations of the particulate components o f active biomass in the influent were assumed to be negligible in the simulation model and were set as follows. X H = 0.01, XAUT = 0.01, X P A O = 0.01, X P H A = 0, X G L Y = 0, and X P P = 0 (g C O D / m 3 ) B y assuming negligible active biomass concentration, the particulate fraction consisted of the inert and slowly biodegradable fractions as: C O D p a r U N F = X i + X s (2.10) X s and X i fractions were determined as suggested by Meijer et al. (2001) using fxsin defined as: f - = 3 ^ < Z 1 1 ) 60 fxsin was used as calibration parameter to fit C O D t o t data of the sludge. This method was used instead of using B O D measurements to estimate X s and X T fractions since B O D measurements are not considered reliable. The concentration of total suspended solids (TSS) in the T U D P model was defined as a fraction of the particulate C O D (COD p a r t ) as: TSS = 0.7 • C O D p a r t (2.12) The value 0.7 was determined by correlating the measured TSS values to the particulate C O D values determined as the difference between C O D t o t and C O D s o i . 2.6.2 Characterization of N and P Fractions in the Influent Soluble nitrogen compounds were directly measured as ammonium-nitrogen (NH4-N) and nitrate plus nitrite-nitrogen (NOx-N) for use in the T U D P model ( N O x - N is referred to as N O 3 - N for simplicity). The influent T K N was calculated in the model as: TKNINF = NH4-N + i N B M ' (XH,INF + XPAO,INF + XAUT.INF) + INSF ' SF,INF + + iNSI " S^iNF + iNXS ' Xs,rNF + iNXI' Xi.rNF (2-13) where INBM» JNSF, INSI, JNXS and iwxi were the nitrogen fractions of biomass, SF, SI XS and X i respectively. The total nitrogen (TN) was used in the simulation for all zones and was calculated as: T N = NO3 -N + T K N (2.14) The measured T K N and N O 3 - N concentrations were used to calculate the T N . The calculated T N values based on the measured concentrations are referred to as measured T N in this thesis. The measured influent and biological zone T N concentration values were used to calibrate the iNxs and iNxi parameters of the T U D P model while the default values for the 61 nitrogen fraction of the soluble C O D components, JNSI and JNSF and for the biomass, INBM, were used as given in Appendix I. The ortho-phosphate compounds were measured and presented as PO4-P. The phosphorus fractions for the biomass and the various C O D components were determined in the same manner as the nitrogen fractions. The TP concentration in the influent in the simulation was calculated using Equation 2.15. TP INF = PO4 -P + ipBM ' (XH,INF + XpAO,INF + XAUT,INF) + IPSF ' SF,INF + IPSI - Si,rNF + + 1PXS ' Xs,rNF + 1PXI " X^rNF + Xpp.rNF (2.15) ^ where ipBM? IPSF, ipsi, ipxs and ipxi were the nitrogen fraction of biomass, SF, SI X s and X i respectively. The measured influent and biological zone T P concentration values were used to calibrate the ipxs and i P x i parameters of the T U D P model while the default values for the phosphorus fractions of the soluble C O D components, ipsi and i PsF were used as given in Appendix I. The default value was used as well for i P B M . Table 2.6 lists the approximate typical ranges of concentrations for the influent components during the course of the experiment. It should be noted that there was a large variation in concentration ranges since the data were collected over a long period of time and in different seasons with various temperatures and weather conditions. The average and the median values are presented to give a better indication of the typical values. Table 2.6. Influent concentration ranges of different components Components Concentration Ranges Average Median Value C O D t o t (g C O D / m 3 ) 211 -580 357 346 C O D s o l ( g C O D / m 3 ) 5 2 - 207 113 104 V F A ( g C O D / m 3 ) 10 -39 22.8 21.7 TP (g P/m 3) 2.8--6 .5 4.26 4.2 T K N (gN/m 3 ) 2 5 - 41.7 33.9 34 PO4-P (g P/m 3 ) 1.45 - 4 . 0 2.61 2.61 N H 4 - N (g N /m 3 ) 16.7--34.8 26.7 26 N O 3 - N (g N /m 3 ) 0 - 0.5 0.057 0.02 62 2.7 Model Set-up in AQUASIM 2.0 A Q U A S I M 2.0 was used in this research as the simulation engine. A Q U A S I M does not include a visual presentation of the process diagram. Window-based menus are used instead for identifying variables, processes (rate equations), compartments and the links between compartments. The software includes a list o f pre-identified compartments along with their hydraulic equations. B y specifying the compartment properties and identifying the links between them, the system hydraulics can be identified. Prior to the start-up of the pilot M E B P R and C E B P R systems, tracer studies were carried out to confirm that the biological reactors are well mixed with no dead zones or short-circulating flow. Therefore, each reactor was modeled in A Q U A S I M using one mixed compartment. 2.7.1 MEBPR Process Model in AQUASIM The M E B P R process at the U B C pilot plant was identified as illustrated in Figure 2.6. The reactor volumes were set as given for the M E B P R in Table 2.1 and the process operating conditions are given in Table 2.2. A Q U A S I M does not include a membrane module; therefore the immersed membrane was modeled as an additional small mixed compartment following the aerobic reactor with a high recycle flow back to the aerobic zone. Perfect solids-liquid separation was assumed for the membrane unit so all suspended solids were recycled back to the aerobic zone. 0.234 m 3 Reactor ] (Anaerobic) 0.585 m 3 Reactor 2 (Anoxic) ANO Rec. 1.311 m 3 Reactor 3 (Aerobic) R 3 -Membrane AER Rec. Membrane 0.02 m 3 Figure 2.6. M E B P R process model in A Q U A S I M Membrane -* Permeate, PermeateN Qerr Tank J 0.02 m 3 1000 m3/d ^recycle ' fraciion 63 2.7.2 CEBPR Process Model in AQUASIM The C E B P R process was modeled in A Q U A S I M as illustrated in Figure 2.7 with all specified reactor volumes. The clarifier was modeled as a mixed compartment with a volume of 0.1 m 3 . A n additional mixed reactor was added as a sludge compartment to simulate the sludge blanket at the bottom of the clarifier. The amount of biomass retained in the clarifier was measured and used as the starting point in estimating the volume of the sludge compartment in the model. Then the compartment volume was adjusted to 0.2 m 3 to match the nitrate concentrations measured in the final effluent. The sludge recycle flow was set to the influent flowrate. A solids-liquid separation efficiency o f 99.21% was assumed, which resulted in a value of 0.0079 for fxEFF , which is the fraction of solids that leave the system with the effluent. 0.24 m 3 Reactor 1 (Anaerobic) 0.618 m 3 1.3686 m3 R,^R2 CReactorT^j l y R , f Reactor 3^| (Anoxic) ANO Rec. AER Rec. Calr. . (Aerobic) I \ Sec. Clar/ \0.1 ra3/ Sludge Compartment 0.2 m 3 QRAS = Qi„m3/d Q R A S • ^fraction ( I - W W R A S + Q E F F ) - X F R A C L I O Figure 2.7. C E B P R process model in A Q U A S I M 64 3 STEADY STATE MODELING OF THE MEBPR USING THE TUDP MODEL Steady state modeling of wastewater treatment processes is usually performed prior to the dynamic modeling of the process. This is done to assess the predictive power of the model and to check process operating conditions and measurements. The model is usually calibrated using averaged process data and the calibrated parameters are used as initial parameter values for the dynamic calibration. This chapter starts by describing the general process behavior of the M E B P R and the C E B P R processes during the course of the experiment. Following that, the selection of the data set used for the steady state modeling of the M E B P R process and the results o f the error diagnosis and data reconciliation analysis of the selected data set are discussed. Then the M E B P R steady state modeling results are described and discussed. Finally the steady state modeling results for the C E B P R process are discussed and compared to those of the M E B P R process. 3.1 MEBPR and CEBPR Process Data and Behavior This section describes the process behavior of the M E B P R and C E B P R processes during the experimental phase and the effect of different operating conditions on the process performance. Data were collected from both systems during the period between February 17 and October 22 of 2003 for the purpose of this project. This period exhibited variable process behavior and operating conditions as discussed in section 2.2. Data collected for the M E B P R process are presented in Figures 3.1, 3.2, 3.4 and 3.6 for C O D s o i , N H 4 - N , N 0 3 - N and PO4-P respectively. Data collected for the C E B P R process are presented in Figures 3.1, 3.3, 3.5 and 3.7 for C O D s o i , N H 4 - N , N O 3 - N and P 0 4 - P respectively. Figure 3.1 shows that the major change in C O D data was mainly observed in the influent CODtot resulting from the change in the filling schedule of the plant wastewater storage tanks on September 3, 2003. Prior to that, storage tanks were being filled twice a day at 9 am and 9 pm when the concentration of the wastewater was high. However, it was decided to fi l l the tanks four times a day to generate wastewater loads and concentrations 65 that would be more similar to those of a real plant subject to normal daily variations. Therefore, the storage tank filling schedule was changed to four times a day, starting at 7 am and repeating at 6-hour intervals. This resulted in a large decrease in the average daily influent C O D t o t concentration o f about 100 g C O D / m 3 as shown in Figure 3.1. The CODsoi concentration for the effluent of the M E B P R process is also shown along with the effluent C O D s o i concentration of the C E B P R process. Both systems exhibited efficient C O D removal over the course of the experiment, averaging a 91.5% C O D removal with a standard deviation of 7.2% for the M E B P R process and a 90.0% removal with a standard deviation of 9.0% for the C E B P R process. 700 29-Jan-03 20-Mar-03 9-May-03 28-Jun-03 17-Aug-03 6-Oct-03 25-Nov-03 (Date) Figure 3.1. Influent and effluent C O D concentrations of the M E B P R and C E B P R Processes during the course o f the experiment Figures 3.2 and 3.3 show the ammonium profiles for the M E B P R and C E B P R processes respectively. The ammonium profiles for both systems were more or less similar, as the concentrations in the anaerobic and anoxic zones varied according to the influent concentration variations while the effluent ammonium concentrations were essentially zero throughout the experimental period. This stability in the nitrification process is 66 believed to be due to the fact that the aerobic zones for both systems were over-designed. A study by Ramphao et al. (2005) recommended that the sludge mass fraction distribution in the anaerobic, anoxic and aerobic zone should be 0.15, 0.35 and 0.50 respectively. However, in the U B C M E B P R system, the corresponding mass fractions were 0.05, 0.20 and 0.75, indicating that the aerobic zone may have been over-sized. The high sludge concentration in the aerobic zone resulted in consistent complete nitrification and therefore, little variation was observed in the effluent ammonium concentrations. Figure 3.2 shows the ammonium profiles in all zones o f the M E B P R process. The difference in the aerobic and effluent concentrations (increase in the aerobic ammonium concentrations) noticed during the period M a y 2 - August 16 was due to an experimental error in the measurement of the aerobic concentration that was detected and corrected on August 16. Therefore, for the purpose of modeling in this study, it was decided to use the effluent ammonium concentration instead of the aerobic concentration for that period. 60 29-Jan-03 20-Mar-03 9-May-03 28-Jun-03 17-Aug-03 6-Oct-03 25-Nov-03 (Date) Figure 3.2. Ammonium profiles in the M E B P R process during the course of the experiment 67 Figure 3.3 shows the ammonium profiles in all zones of the C E B P R process. The behavior is similar to that of the M E B P R process. The high concentrations in the aerobic zone and effluent ammonium in February of 2003 were due to the process start-up on February 13 of 2003. 60 19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 7-Aug-03 26-Sep-03 15-Nov-03 (Date) Figure 3.3. Ammonium profiles in the C E B P R process during the course of the experiment Nitrate profiles for the M E B P R and the C E B P R processes are shown in Figure 3.4 and 3.5 respectively. Both systems experienced periods of high concentrations of nitrates in the anoxic zone. This phenomenon is not desired in E B P R systems since nitrates leaking to the anaerobic zone with the anoxic recycle can negatively impact anaerobic P-release. Therefore, the aerobic recycle ratio was reduced for both systems from 2 to 1, to ease the load on the denitrification zone, and was switched back to 2 when the anoxic nitrate concentrations in both systems dropped to nearly zero. Process operating conditions are described in more detail in section 2.2. 68 19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 7-Aug-03 26-Sep-03 15-Nov-03 (Date) Figure 3 . 4 . Nitrate profiles in the M E B P R process during the course o f the experiment C E B P R A N A C E B P R A N O C E B P R A E R CEBPR EFF 19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 7-Aug-03 26-Sep-03 15-Nov-03 Figure 3 .5. Nitrate profiles in the C E B P R process during the course o f the experiment 69 The magnitudes of the anoxic zone nitrate concentrations were almost the same in both systems, indicating that this phenomenon was independent of the solid-liquid separation step. Rather, it was related to factors affecting both biological systems to the same degree. It was believed that the high nitrate concentrations in the anoxic zone were related to the size of the anoxic zone, for which the zone H R T was lower than that required to achieve complete denitrification. Furthermore, the sludge mass fraction of 0.20 in the M E B P R anoxic zone was much lower than the recommended mass fraction of 0.35 suggested by Ramphao et al. (2005). These design limitations resulted in periodic incomplete denitrification as reflected in the anoxic zone nitrate concentrations. The phosphorus removal process showed interesting results as well for both systems. The ortho-phosphate profiles in all zones of the M E B P R and the C E B P R processes are shown in Figures 3.6 and 3.7 respectively. Both systems exhibited two major failures in phosphorus removal during the experimental period. The first failure started at the beginning of M a y of 2003 and lasted for about 2 months, until the end of June. The system was able to recover on its own and ortho-phosphate concentrations in the effluent eventually returned to essentially zero. A second, more significant failure started around August 22 and lasted until the end of the experimental phase. Both the M E B P R and the C E B P R processes experienced the same failures in phosphorus removal. However higher concentrations, reaching up to 3.6 g P/m in the second failure, of effluent PO4-P were observed in the M E B P R system on both occasions while the effluent PO4-P in the C E B P R system did not exceed 1.8 g P /m 3 . Furthermore, it took a longer time for the M E B P R process to recover from the first failure than the C E B P R process, as illustrated in Figures 3.6 and 3.7. 70 = _ O O a, c -19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 (Date) I t* P-Failure 7-Aug-03 26-Sep-03 15-Nov-03 2" P-Failure Figure 3.6. Ortho-phosphate profiles in the M E B P R process during the course of the experiment INF CEBPR ANA CEBPR_AMO CEBPRAER CEBPR EFF 19-Jan-03 10-Mar-03 (Date) 5-Nov-03 Figure 3.7. Ortho-phosphate profiles in the C E B P R process during the course of the experiment 71 Both the M E B P R and the C E B P R processes were operated without a fermenter. The raw sewage from the storage tanks was settled in a primary clarifier and then was pumped to both biological treatment systems. Therefore, the only source of V F A in the process was that coming in with the primary-treated influent stream. Figure 3.8 presents the influent V F A / T P ratio along with the effluent PO4 -P concentrations o f both systems. The ratio o f the influent V F A / T P was plotted since it is reported in the literature that 7-9 mg of V F A are needed to remove 1 mg of phosphorus (Mulkerrins et al., 2004). Therefore, the influent V F A / T P ratio was examined. Examining Figure 3.8, it can be noticed that both system failures were accompanied by a decrease in the influent V F A / T P ratios. When the ratio stayed relatively constant for a period of time, the process was able to recover. The first failure in P-removal was preceded by a decrease in the ratio from about 7 to 5 in May. The influent V F A / T P remained at about 5 during the summer as the system recovered. Following that, a sharp increase in the influent V F A / T P ratio followed by a decrease starting around mid August seemed to be associated with the second failure at the end of August which continued until the end of the experimental period. 30-Mar-03 29-Apr-03 29-May-03 28-Jun-03 28-Jul-03 27-Aug-03 26-Sep-03 26-Oct-03 25-Nov-03 (Date) Figure 3.8. Influent V F A / T P ratio and effluent P O 4 - P concentrations for the M E B P R and C E B P R process during the course of the experiment 72 Monti et al. (2006b) used the process data from the same system to compare the process behavior o f the M E B P R and C E B P R systems. The authors reported that the decrease in the influent C O D t o t concentration observed in September, due to the changes in the filling schedule o f the raw wastewater storage tanks, further reduced the influent V F A concentration, since the residence time in the storage tanks was reduced, resulting in shorter fermentation times. It is believed that this decrease in the influent V F A concentration is also the reason for the low denitrification efficiency. Research has shown that the spillover of V F A to the anoxic zone in a process with a fermenter could increase the denitrification rate by as much as 30 or 40% above that measured in an anoxic zone that must rely solely upon the organic carbon resources of the raw sewage (Oldham and Rabinowitz, 2001). The high nitrate concentrations observed in the anoxic zone in September of 2003, could have also contributed to the second failure in P-removal, since nitrate leakage to the anaerobic zone with the anoxic recycle would compete for the acetate available for the bio-P process. A detailed discussion of the reasoning for the observed failure in P-removal in the M E B P R process and a comparison to the process behavior of the C E B P R process is given by Monti et al. (2006b). In summary, the measured data indicate that the M E B P R and the C E B P R processes exhibited similar behavior for the C O D removal, nitrification and denitrification processes. Some differences were observed for the phosphorus removal process behavior of the two systems. Furthermore, the influent V F A / T P ratio was thought to be a critical factor in the stable operation of the E B P R process of both systems. Results showed that a minimum influent V F A / T P ratio is required to achieve a stable E B P R process operation, however, a sudden decrease in the influent V F A / T P ratio, regardless of the value, also appeared to disturb the bio-P removal process. Therefore, both the value of the influent V F A / T P ratio and keeping the ratio constant appeared to be important. 73 The aerobic recycle flowrate was found to be a critical and challenging operating parameter to control the nitrate concentration in the anoxic zone. Results showed that controlling the nitrate concentration in the anoxic zone by varying the aerobic recycle flowrate was an important factor for process stability. Results also showed the importance o f proper system design to allocate the required reactor volume and to control the sludge mass distribution in each zone. It was found that the anaerobic and anoxic zones were under-designed while the aerobic zone was over-designed, which was believed to have contributed also to the encountered process failures. The results described above, created a motive for modeling both the M E B P R and C E B P R processes to better understand their behavior and the actual kinetic differences, i f any, between the two systems. Furthermore, the M E B P R process was modeled for subsequent utilization of the model in simulation studies aimed at developing guidelines for the design and operation of the M E B P R process for effective operation under high influent flowrates. The following sections describe the steady state modeling o f the M E B P R and the C E B P R processes. 3.2 D a t a Set Se l ec t ion fo r the S teady State M o d e l i n g o f the M E B P R Process Steady state modeling of W W T P s requires data collected from the process under steady state or pseudo-steady state conditions. A s discussed earlier, the M E B P R process experienced some periods of low P-removal efficiency and other periods of low denitrification rates where the concentration profiles showed dynamic variations. Furthermore, the particulate TP, T K N and C O D t o t mixed liquor concentrations were not measured frequently due to the laborious and time consuming analytical measurement techniques resulting in periods of incomplete data as shown in Figures 3.9, 3.10 and 3.11 respectively. Therefore, the data sets used for the steady state modeling and calibration of the M E B P R process were selected carefully to obtain a representative set of process data while avoiding periods of process upset. In general, process data were selectively averaged over the period between M a y 26 and September 3 of 2003 depending on data availability and process stability during that period. Table 3.1 identifies the date range 74 used for each variable (concentration profile) for the steady state modeling of the M E B P R and the reason for the choice. Table 3.1. Date range used for averaging the concentration data for the steady state Concentration Date Range Used Reason A l l influent, effluent, anaerobic, anoxic and aerobic concentrations (except for the ones identified below) M a y 26 - Sep 3 A l l data were available and were at pseudo-steady state conditions T K N for anaerobic, anoxic and aerobic Mar 24 - Sep 3 T K N concentrations were not measured regularly and this period covered all the data available at pseudo-steady state conditions TP for effluent, anaerobic, anoxic and aerobic Mar 3 - A u g 22 TP concentrations were not measured regularly and this period provided all the data available while avoiding the second failure in P-removal PO4-P for effluent, anaerobic, anoxic and aerobic M a y 26 - A u g 22 This period was chosen to avoid the second failure in P-removal MEBPRANA MEBPRANO MEBPRAER MEBPR EFF 19-Jan-03 10-Mar-03 29-Apr-03 18-Jun-03 7-Aiig-03 26-Sep-03 15-Nov-03 (Date) Figure 3.9. Mixed liquor TP concentrations for the M E B P R process during the course o f the experiment 75 700 600 500 400 300 o U H 200 100 0 • INF A MEBPR_EFF - • - MEBPR_ANA A MEBPRANO • MEBPR AER 29-Jan-03 20-Mar-03 9-May-03 28-Jun-03 17-Aug-03 6-Oct-03 (Date) Figure 3.10. Mixed liquor T K N concentrations for the M E B P R process during the course of the experiment 6000 5000 Q O 4000 M 1 3000 a o o s 2000 1000 0 • MEBPR_AJNA A MEBPRANO -•-MEBPR AER 19-Apr-03 8-Jun-03 28-Jul-03 (Date) 16-Sep-03 5-Nov-03 Figure 3.11. Mixed liquor C O D t o t concentrations for the M E B P R process during the course of the experiment 76 3.3 Error Diagnosis and Data Reconciliation of the MEBPR Process Data Experimental data often contain measurement errors resulting from various sources of errors. The data selected for steady state simulation of the M E B P R process were carefully analyzed for the detection and removal o f outliers. Outliers were detected by visually comparing concentration profiles of different zones i f no reaction occurred for that component in the two zones. A n example of such an approach is shown in Figure 3.12 in which the ortho-phosphate concentrations in the effluent and the aerobic zone are compared for possible outliers since the effluent concentrations were expected to be essentially the same as the concentrations of the soluble fraction of the aerobic sample. The spike in the aerobic PO4-P concentration on March 12 was not accompanied by a spike in the effluent, so this spike was not justified and was therefore neglected. However, the spike in the effluent PO4-P concentration around August 29 was not considered an outlier since it had a corresponding spike in the PO4-P concentration in the aerobic zone on the same day. In general, soluble concentration measurements contained very few outliers; however, measurements for mixed liquor concentrations (mainly C O D t o t ) contained outliers at a frequency o f up to 5% o f the total number o f measurements. 77 29-Jan-03 20-Mar-03 9-May-03 28-Jun-03 17-Aug-03 6-Oct-03 25-Nov-03 14-Jan-04 (Date) Figure 3.12. Outlier analysis of the M E B P R PO4 -P concentrations in the effluent stream and the aerobic zone during the course o f the experiment After the detection and removal of outliers, TP and flow measurements were used in mass balance equations, since phosphorus is conserved in the process, to verify measured TP data prior to simulation. Mass balance equations were implemented in the Macrobal 2.02 software (Hellinga, 1992). Macrobal is a program that applies conservation principles for the estimation and balancing of conversion rates in biotechnological processes. The software can be downloaded for free from the website o f the Department o f Biotechnology at the Technical University of Delft. Macrobal was used by Meijer et al. (2002a) for error detection and data reconciliation of concentration and flow measurements of a full-scale W W T P by checking the balance residuals identified by: Z Q i X C , n = i n (3.1) i=l 78 where Q ( is the average flow to be balanced and Cin is the average concentration of the n t h component in that flow. The program uses a statistical test to examine the ^-square distribution of the residuals to detect gross errors. When the system of equations can be balanced within the specified standard deviations for the flows, the program indicates that no proof of gross error in the measurements is detected. In the work of Meijer et al. (2002a), the program was mainly used to check the operational data such as SRT and flows, since these were not directly measured, while assuming concentration measurements were correct. In the present work, the SRT was closely monitored and controlled using TSS measurements and waste flow. Furthermore, suspended solids did not leave the system with the effluent due to the efficient solids-liquid separation of the membrane filtration unit. Moreover, flows were checked and adjusted regularly to ensure constant flowrates. Therefore, in the present research, Macrobal was used solely to check TP concentration measurements, assuming correct operational data. Figure 3.13 presents the abbreviations used in the Macrobal program for each flow and biological zone. QINF / A n a e r o b i c X Q R ^ / H Anoxic v QR2 / Aerobic \ QEFF v R 1 y ^ R2 1 •( R3 1 • t Q R C 2 ~ 1 Q R C 3 QEX . Figure 3.13. The M E B P R process flow diagram as identified in Macrobal The following mass balance equations were implemented in Macrobal. Flow balance for R I : QFNF - Q R 1 + Q R C 2 = 0 (3.2) Flow balance for R 2 : Q R 1 - Q R 2 + Q R C 3 - Q R C 2 = 0 (3.3) Flow balance for R 3 : Q R 2 - Q E F F - Q R C 3 - Q E X = 0 (3.4) Flow balance for W W T P : QINF - QEFF - QEX = 0 (3.5) TP balance for R I : QINF • TPFNF - Q R 1 • TP ANA - Q R C 2 • T P A N 0 = 0 (3.6) TP balance for R2: Q R 1 • TP A NA - Q R 2 • T P A N o + Q R C 3 • T P A E R -- Q R C 2 • T P A N O = 0 (3.7) 79 TP balance for R3 : QR2 • T P A N o - QEFF • T P E F F - Q R C 3 • TPAER -- Q E X - T P a e r = 0 (3.8) The TP averaged concentration values were entered into the program as constant values and the flowrate values were entered along with their allowed variance. Then the program performed a statistical test to check the ^-distribution of the residuals of the equations listed above for gross errors. If the program detected an error, it indicated that a gross error was detected. That meant that the program could not balance the given set of equations with the given constant values (TP concentrations in this case) and the given flowrate values with the allowed variances for the flowrates. In that case, the TP values were adjusted manually, within the measurement variance of each reading, until the program indicated that no proof for error measurement was detected. The program provides a list o f corrected (Balanced) flowrate values that are within the variance specified by the user, to balance the set o f given equations and given values (TP concentrations). A l l equations are balanced simultaneously. It was not possible to balance the system using the averaged TP measurements obtained. However, when the anoxic TP concentration was manually changed from the averaged value of 81.5 g P /m 3 to 96.2 g P /m 3 , the program indicated that no proof for error measurement was detected and was able to adjust the flowrates within the provided variance. The corrected averaged TP value o f 96.2 fell within the variance for the TP measurements for the anoxic zone o f ± 40 g P /m 3 for that period. The difference could have been the result of grab sampling. The results of the program are presented in Tables 3.2 and 3.3. Table 3.2 displays the mass balance equations listed above. Each vertical column represents a closed balance. For example the third column titled " R l ( A N A ) flow" forms the flow balance over the first reactor, which is the anaerobic zone with the equation: QINF - QR1 - Q R C 2 = 0. In Table 3.2, the averaged TP concentration measurements obtained based on the collected data are shown in bold. The standard deviations of the averaged TP concentration values are displayed in italics for reference only, as they were not entered in the Macrobal program. The adjusted averaged TP concentration measurements, to fit the balance, are shown in the parentheses below the actual value. Table 3.3 shows the results of the balance for each flow. The third column 80 indicates i f this flow could be balanced in this program using " * B " . It also shows the measured flowrate and the user specified standard deviation of the measurement in the following column. The last column shows the value predicted by the program to close the balance and the standard deviation associated with the prediction. If the predicted flowrate value fell within the measured standard deviation o f each flow, then the measurement was considered correct. The results o f the last column show that the balanced values are close to the measured values indicating that the entered values are correct and the adjusted averaged TP concentration values resulted in a balanced system. T a b l e 3.2. Error diagnosis and data reconciliation of averaged flow and total phosphorus measurements. ( A l l empty spaces in the matrix represent zeros) Flow, Load U n i t R 1 ( A N A ) R 2 ( A N O ) R3 (AER) W W T P RI (ANA) R2 (ANO) f l O W f l n w f l n u / l i , . . . T D I-11 flo  flow flow T P T P R3 (AER) T P Q l N F QR1 QR2 Q E X Q R C 3 Q R C 2 m7d m3/d m3/d m7d m3/d 1 -1 4.6 ± 7.8 -49.9 ±3 7.6 -1 -1 -1 -1 -1 81.5 ±40. J (96.: 49.9 ±37.6 -81.5 ±40.1 (96.2) 142.6 ±40 .7 -81.5 ±40.1 81.5 ±40.1 (96.2) -0.9 ±0.5 (-0.72) -142.6 ±40.1 -142.6 ±40.1 Value in parentheses () is the manually adjusted averaged TP values to fit the balance Italics is the TP standard deviation T a b l e 3.3. Results of the error diagnosis and data reconciliation of averaged flow and total phosphorus measurements. (The balanced system is presented in the last column) Degree of redundancy = 6 No proof for measurement error based on the X-square distribution * = measured Flow, Load Unit C = calculated B = balanced by 'ram Measured ± Standard Deviation Balanced ± Standard Deviation Qin m3/d *B 5.112 ±0.1 5.191 ±0.0190 QR1 m3/d *B 10.22 ±0.1 10.27 ± 0.0376 QR2 m3/d *B 15.34 ±0.1 15.44 ±0.0565 Qeff m3/d *B 4.962 ±0.1 5.043 ±0.0185 Qex m3/d *B 0.15 ±0.001 0.148 ± 0.000542 QRC3 m3/d *B 10.22 ±0.1 10.25 ±0.0375 QRC2 m3/d *B 5.112 ± 0.1 5.079 ±0.0186 81 3 . 4 Steady State Model Calibration Results for the MEBPR Process Once the flowrate data and the averaged TP measurements were checked and adjusted for use in the steady state simulation of the M E B P R process, the influent characterization was performed according the S T O W A standards as described in section 2.6 for C O D , phosphorus and nitrogen fractions. The averaged influent data were then used as inputs to the simulation and the model parameters were calibrated to fit the averaged concentrations measured and adjusted by Macrobal for the biological zones and the effluent. Default parameter values as reported by Meijer (2004), and given in Appendix I, were used initially and parameter values were then adjusted manually to obtain a reasonable fit o f the averaged measured data values. The calibration was performed in the following sequence. 1. It was not possible to predict the correct bioreactor particulate concentrations for C O D , TP and T N for the measured influent CODpart. concentration by adjusting the calibration parameters suggested by Meijer et al. (2001), fxsin, INXS, INXI, ipxs and ipxi. It was observed that the predicted sludge concentrations were always higher than the measured ones. The default sludge yield appeared to be too high for this system. Work done by Zhang and Hal l (2006) on estimating the heterotrophic yield of sludge taken from the U B C M E B P R pilot plant process operated under different SRT values, resulted in Y H values ranging from 0.37 to 0.62 (g CODxH/g C O D ) with an average of 0.50 ± 0.08, which is lower than the default reported value of 0.63 for the A S M 2 d model. Upon calibration of this parameter, it appeared that 0.58 was the most suitable in this case. Using the value of 0.58 for Y H resulted in a better fit o f the mixed liquor data for C O D , TP and T N . Then the averaged measured values were further fitted by adjusting the parameters suggested by Meijer et al. (2001). 2. The COD tot concentrations in the biological zones were fitted using the fraction, fxsin, defined as Xs/(Xs+Xi). This parameter represented the fraction of slowly biodegradable C O D in the total influent particulate C O D . In order to fit the mixed liquor C O D t o t concentrations, fxsin was changed from 0.439 (Meijer, 2004) to 0.85 reflecting less inert component and more substrate in the influent C O D t o t . The low 82 inert solids content was thought to be reasonable because the sewerage treated in this process was of a domestic origin and most o f the inert solids were removed in the primary clarifier. 3. T N concentrations of the influent stream and the biological zones were fitted by changing the coefficient for nitrogen content of particulate substrate, i N x s , from 0.03 to 0.065 and the coefficient for nitrogen content of inert particulate C O D , iNxi, from 0.03 to 0.04. These coefficients are dependant on the wastewater characteristics and are different from one region to another (Roeleveld and van Loosdrecht, 2002). 4. TP concentrations of the influent stream and the biological zones were fitted by changing the coefficient for phosphorus content of particulate substrate, i P xs, from 0.01 to 0.005 and the coefficient for phosphorus content of inert particulate C O D , ipxi from 0.01 to 0.0045. These coefficients are also dependant on the wastewater and are different in each case (Roeleveld and van Loosdrecht, 2002). 5. The predicted effluent N H 4 - N concentration was found to be higher than the average measured value and so it was fitted by changing the half-saturation parameter for soluble N H 4 - N for growth of A U T biomass, K N H , from 1 to 0.1. A s indicated by the I A W Q Task Group for the A S M No. 2 model, Henze et al. (1999), the half saturation parameter is related to the diffusion limitation in the floes and in cases with high turbulence and small floes (as often found in pilot-scale experiments), this value tends to be lower than that of a full-scale installation, which supports larger floes and less turbulence. Therefore, the lower value estimated for K N H was in accordance with this theory. The calibrated parameters are listed in Table 3.4. B y adjusting these values, the rest of the averaged measurements were predicted reasonably well within the standard deviation of each measurement. Results for the steady state simulation are given in Table 3.5. It should be noted that the large standard deviations for some of the averaged measurements are because of the long experimental period of about three months during which the plant experienced some dynamic behavior due to influent variations. The long time period was chosen to ensure obtaining process 83 data that were representative of the normal behavior of the plant. Furthermore, measurements were based on grab sampling which could have also contributed to the variations. Table 3.4. T U D P model parameters used for the M E B P R process for fitting the averaged measurements Parameter Units Default Value* Manually calibrated value for steady state data fxsin g C O D / g C O D 0 . 4 3 9 0 . 8 5 iNXI g N / g C O D 0 . 0 3 0 . 0 4 iNXS g N / g C O D 0 . 0 3 0 . 0 6 5 ipxi g P / g C O D 0.01 0 . 0 0 4 5 ipxs g P / g C O D 0 .01 0 . 0 0 5 KNH g N / m 3 1 0.1 Y H g C O D / g C O D 0 .63 0 . 5 8 * As presented by Meijer (2004) Table 3.5. Averaged measurements and simulation results o f the calibrated T U D P model using pseudo-steady state data collected from the M E B P R process Measured/ Component Influent Anaerobic Anoxic Aerobic Effluent Simulated (g/m3) (g/m3) (g/m3) (g/m3) (g/m3) Meas. COD t o t 417±80 1903 ±151 3354± 529 4638 ±1069 26 ±23 Sim. x + s 417 1976 3534 5117 27 Meas. COD s o l 117 ± 37 49 ± 14 37 ±44 26± 18 26 ±23 Sim. SA+SF+S, 117 54 30 27 27 Meas. T P 4.6 ±0.78 50 ±38 96* 143 ± 63 0.55* ±0.5 Sim. P04-P+XPp+iP 4.47 55 105 155 0.1 Meas. PO4-P 2.5 ±0.39 10.6 ±2.6 4.9 ± 1 0.7 ±0.7 0.5 ±0.6 Sim. PO4-P 2.5 12.8 5 0.1 0.1 Meas. TKN 38.7 ±2.7 120 ±27 184 ±33 274 ± 48 1.1 ±0.4 Sim. NH4-N+iN 45 121 196 272 0.4 Meas. TN 38.8 120 184 283 9.6 Sim. NH4-N+iN+N03-N 45 121 197 280 8.4 Meas. NH 4-N 24.4 ±3.3 17.4 ±2.4 10.4 ± 6 0.1 ±0.1 0.05 ± 0.05 Sim. NH 4-N 24.4 17.8 9.6 0.11 0.11 Meas. NO3-N 0.05 ± 0.07 0.02 ± 0.03 0.8 ± 1.16 8.4 ± 1.95 8.5 ± 1.5 Sim. NO3-N 0.05 0 0.12 8.0 8.0 * Adjusted to fit mass balance equations in Macrobal In general, the mixed liquor measured concentrations, C O D t o t , TP and T N , were fitted reasonably well by the simulation after adjusting the heterotrophic yield and the particulate fractions. The ammonium concentrations were also predicted well after adjusting the KNH parameter. The nitrate concentrations were predicted reasonably well in the aerobic zone and in the effluent; however, the anoxic nitrate was under-predicted when compared to the averaged measured values, even though it was still within the 8 4 measured standard deviation. The ortho-phosphate concentrations in the anaerobic, aerobic and effluent were also slightly different but they were still within the standard deviation of the measurements. Considering that the averaged measured data were predicted reasonably well within the measured standard deviations for all concentrations and since this calibration was intended as a starting point to assess the predictive power o f the model and the accuracy of the measured operating conditions and concentrations, it was decided that this calibration was satisfactory. In the work by Mejier et al. (2002a), the net calculated oxygen consumption based on the balanced data was used as a final check on the calibrated model. They used Macrobal to calculate the nitrified mass flow, denitrified mass flow, oxidized C O D and the net oxygen consumption using open mass balance equations for T K N , nitrate (NO3-N), C O D and D O respectively as given by Equations 3.9, 3.10, 3.11 and 3.12. - nitrified = T K N I N F x Q I N F - Q E F F x T K N E F F — QWASTE X T K N ^ (3.9) Qdenitrified = Q.NP X (NO3 - N ) ^ - Q E F F X (NO3 - N ) E F F - Q WASTE X (NO3 - N ) A £ R + Q R A T R I F I E D (3.10) Oxygen Consumpt ion C 0 D = C O D t o t I N F x Q I N F - C O D E F F x Q E F F - C O D t o u A E R x QWASTE - 2.87 x Q D E M T N F I E D (3.11) Oxygen Consumption N e t = Oxygen Consumpt ion C 0 D + 4.57 x (3.12) The program reported the calculated loads along and their calculated standard deviations. The standard deviations were estimated considering the errors in the flow measurements while neglecting the errors in the concentration measurements. That was reasonable in their case since they were dealing with data collected from a full-scale W W T P with large flowrate values and measurement errors compared to errors found in concentration measurements. However, in the present work, data were collected from a pilot plant system where the flowrates were monitored and corrected regularly. Therefore, errors 85 encountered in averaged concentration measurements were significant and had to be considered in the calculations. In the present work, the net oxygen consumption was used as the final check on the calibrated model to ensure a reasonable fit o f the averaged measured data. The net oxygen consumption was calculated manually using Equations 3.9 to 3.12 and the averaged measured concentrations, as reported in Table 3.5. The flows used in the calculation of the net oxygen consumption for the simulated results were based on the simulated flows of 5.1 m 3 /d for QIN F , 4.96 m 3 /d for Q E F F and 0.148 m 3 /d for QWASTE-However, for the calculations of the net oxygen consumption for the measured data, the balanced flow values produced by Macrobal and presented in the last column o f Table 3.3 were used. The aerobic concentrations were used for the waste sludge stream since sludge was wasted from the aerobic zone. The standard deviations were calculated manually using Equation 3.13. The standard deviations of the averaged measurement, as listed in Table 3.5, were used. The flowrate values were assumed constant because they were monitored and controlled closely in the pilot plant. x = f(p,q,r,...) 2 0~„ 2 2 (dx) 2 2 (dx) ° n + o' + P q ydv) • 2 (3.13) a, +••• The calculated net oxygen consumption based on measured data was (1687 ± 462 g CVd) . This value closely matched the calculated value based on simulated results (1675 g CVd) indicating a good fit of the M E B P R steady state measured data. 3.5 Comparison to the Steady State Modeling Results of the CEBPR Process It was important to compare the results of the modeling of the steady state behavior of the M E B P R process to that of the C E B P R process, to gain a better understanding of the effect of the membrane system, i f any, on the biological treatment process. Furthermore, it would help identifying whether the T U D P model was more suitable for the conventional system or its prediction power was equally applicable to the membrane system. 86 Data collected from the C E B P R process were averaged over the same periods as for the M E B P R process, to ensure consistency of the results. The C E B P R process was modeled in A Q U A S I M as described in section 2.7.2. The model parameter values used for the M E B P R process were sufficient to predict the averaged measured concentrations o f the C E B P R process without further calibration. However, for the C E B P R process, the default value o f 0.63 was used for Y H since it resulted in a good agreement between the simulated and measured values. Table 3.6 lists the values of the calibrated parameters for the C E B P R T U D P model, while the rest of the model parameter values were kept the same as those suggested by Meijer (2004). The results of the model prediction and the averaged measurements are presented in Table 3.7. Table 3.6. T U D P model parameters used for the C E B P R process for fitting the averaged measurements _ , T T ., _ ., . Manually calibrated value Parameter Units Default Value - / . for steady state data W S COD/g COD • N x r gWgCOD n n r 0.85 »*S gWgCOD °™ 0.04 ' P X I gP/gCOD ™? 0.065 lXS gP/gCOD ° -°045 _ g W _ ° f 0.005 0.1 The mixed liquor concentrations ( C O D t o t , TP and TN) and the ammonium concentration were predicted reasonably well as shown in Table 3.7. The ortho-phosphate concentration was also predicted relatively well , except that the predicted aerobic and effluent ortho-phosphate concentrations were slightly lower than the averaged measured values. Since there was a large variation in the measured values, the fit was within the standard deviation of the measurement. However the difference between the predicted and measured values was smaller for the C E B P R process than for the M E B P R process. The aerobic and effluent nitrate concentrations were reasonably well predicted, but the anoxic nitrate was slightly under-predicted. 87 In general, initial modeling results based on the pseudo-steady state data collected from the M E B P R and the C E B P R showed that the T U D P model is capable of predicting the process behavior of both systems quite well . However, it was noticed that the C E B P R data were fitted somewhat better than those of the M E B P R process, especially for ortho-phosphate concentrations, as indicated by comparing values in Tables 3.7 and 3.5 respectively. Nonetheless, it was not possible to draw definite conclusions about the ability of the T U D P model to predict the dynamic behavior of the M E B P R process and the differences in the model parameters, i f any, between the M E B P R and the C E B P R systems at this stage based on the pseudo-steady state modeling results. Therefore, further investigation was required using dynamic modeling of the two processes, which is the focus of the next chapter. Table 3.7. Averaged measurements and simulation results of the calibrated T U D P model for the C E B P R process Measured/ Component Influent Anaerobic Anoxic Aerobic Effluent Simulated (g/m3) (g/m3) (g/m3) (g/m3) (g/m3) Meas. COD t o t 417±80 1663 ±280 2334± 376 3451 ±691 82 ± 100 Sim. X+S 417 1568 2719 3890 101 Meas. COD s o l 117 ± 37 46 ± 15 37 ±54 26 ±22 38 ±36 Sim. SA+SF+S, 117 64 42.2 39 39 Meas. TP 4.6 ±0.78 36 ±16 70 ±25 109 ±34 5.8 ± 21 Sim. P04-P+XPp+iP 4.47 39 . 74 108 1.8 Meas. PO4-P 2.5 ±0.39 9.2 ± 1.87 3.5 ±0.8 0.2 ±0.3 0.3 ± 0.3 Sim. PO4-P 2.5 10.4 3.4 0.05 0.05 Meas. TKN 38.7 ±2.7 103 ±27 164 ±43 210 ± 48 3.6 ±3.4 Sim. NH4-N+iN 45 99 154 209 4.4 Meas. TN . 38.8 103 165 219 9.6 Sim. NH4-N-HN+NO3-N 45 99 154 215 10.4 Meas. NH 4-N 24.4 ±3.3 18± 1.8 9± 1.2 x 0.2 ±0.1 0.1 ±0.1 Sim. NH 4-N 24.4 17.6 9.6 0.4 0.4 Meas. NO3-N 0.05 ± 0.07 0.03 ± 0.03 0.6 ± 1 7.1 ± 1.9 6.6 ± 1.9 Sim. NO3-N 0.05 0 0.12 6.2 6.2 88 4 DYNAMIC MODELING OF THE MEBPR PILOT PLANT PROCESS AT UBC USING THE TUDP MODEL The pseudo-steady state modeling of the M E B P R using the T U D P model showed promising results for the model prediction of averaged measured concentration values, as discussed in the last chapter. Building on these results, dynamic modeling was carried out using data collected from the M E B P R process. Then the model parameters of the T U D P model used in the dynamic modeling of the M E B P R process were used in simulating the C E B P R process during the same period of operation to compare the kinetic parameter values and model predictions for the two systems. The objectives of these studies were the following. 1. Perform dynamic modeling of the M E B P R process using the T U D P model. 2. Compare the results for pseudo-steady state modeling to dynamic modeling of activated sludge systems. 3. Examine the prediction power of the T U D P Model in fitting the M E B P R system. 4. Compare the kinetic parameter values and modeling results of the M E B P R process to those of the C E B P R process to investigate the effect of the membrane separation module on the biological kinetics, i f any. The focus of this chapter is directed towards presenting and discussing the dynamic modeling results o f the M E B P R process using the T U D P model to address these objectives. 4.1 Application of a New Practical Protocol for the Dynamic Modeling of Activated Sludge Systems for the Purpose of Process Control and Optimization Studies Prior to performing dynamic calibration of activated sludge models, it is important to identify the objective of the modeling exercise since it determines the required level of specification o f the model (Hulsbeek et al., 2002; Vanrolleghem et al., 2003; Gernaey et al., 2004). General modeling objectives include the following. 1. Design and upgrade of new or existing W W T P s ; 2. Optimization studies for existing W W T P s ; and 3. Development of control strategies for existing and new W W T P s . 89 Modeling for the purpose of developing control strategies requires the highest level of specification (Hulsbeek et al., 2002). However, most modeling studies reported in the literature for W W T P s were mainly performed to address the first objective and some tackled the second objective while less attention has been paid to modeling for the purpose of process control design. In the present study, modeling of the M E B P R was intended to provide a tool to achieve objectives 2 and 3. Therefore, it was critical for the resulting model to accurately reflect process behavior under various dynamic conditions and allow for precise identification of variable interactions. Upon review of practical protocols for dynamic modeling of activated sludge systems available in the literature, there was no clear indication of the steps required to ensure achieving these modeling objectives. Therefore, it was decided to include the development of a practical protocol for dynamic modeling of activated sludge systems for the purpose of process control and optimization studies in the list o f tasks required to achieve the objectives of this project. This section describes the protocol utilized in the current study to model the M E B P R pilot plant process at U B C . Most calibration protocols presented in the literature are mainly based on dedicated lab-scale experimental methods for wastewater characterization and estimation of the kinetic and stoichiometric parameters. Sin et al. (2005) provides a detailed discussion on the limitation in transferring lab-scale data to full-scale models. Parameters estimated via batch tests vary depending on experimental conditions and may not reflect the actual process behavior at full-scale. Therefore, the protocol utilized in the present study was aimed at obtaining a reliable and validated dynamic model while minimizing the reliance on batch test parameter estimation techniques. The protocol used in the present work was based on utilizing system identification tools to carefully design dynamic experiments and data collection procedures to obtain the data required for proper calibration of activated sludge models under dynamic conditions. Then data screening tools were used to further enhance the quality of the collected data. Finally, a set of most identifiable parameters was selected based on process knowledge and sensitivity analysis and these were then calibrated using a combination of parameter 90 estimation techniques and manual calibration to result in a set o f parameters reflecting the best fit o f experimental data. The following sections describe, in detail, the steps of the practical protocol used in the present work for the dynamic modeling of the M E B P R process using the T U D P model. 4.1.1 D y n a m i c E x p e r i m e n t a l D e s i g n A s indicated previously, it is important to identify the objective of the modeling study prior to performing the modeling task, to ensure that the resulting model is capable of fulfilling the desired objective. In this case, the goal of the modeling exercise was to facilitate process control and optimization studies. It was desired to study the M E B P R process design and the effect of different bioreactor zone volume allocations on the sludge distribution and activity of the process. Furthermore, it was desired to examine process variable interactions to determine the set of operating conditions, such as process SRT, influent V F A / T P ratio and recycle flow rates of the M E B P R process, that result in high quality permeate with low concentrations o f nitrogen and phosphorus while operating at a high flowrate making full use of the membrane capacity. A s described earlier, during the course of the experiment, the process experienced periods of failure in P-removal, during which the effluent ortho-phosphate concentration exceeded 3.5 g P /m 3 , between August 22 and October 20, 2003 (Figure 3.6). During the same period, high concentrations of nitrate in the anoxic zone, exceeding 4.5 g N / m 3 at some times, were also observed (Figure 3.4). In order to reduce the high concentrations of nitrate in the anoxic zone, the aerobic recycle flow was reduced from an influent ratio of2,to 1. Since the M E B P R process exhibited poor phosphorus removal performance during this period, it was decided to model the M E B P R process under these conditions to examine the predictive power of the T U D P model under such conditions, which is a critical requirement for using the model to study the effect of different process design and operating conditions on the process performance. 91 Sampling Durat ion The process failure in P-removal started on August 22. It was important that the process be sampled for a period of time that was long enough to ensure fully capturing the sludge microbial dynamics during that period. Since the process SRT during this period was 12 days, as described earlier, October 13 was chosen as the end date for sampling, resulting in a sampling period of over three times the value of the process SRT. The chosen sampling duration was higher than twice the dominant time constant o f the process which is the recommended duration in system identification practice (Soderstrom and Stoica, 1989). The sampling duration is a very important factor in obtaining experimental data suitable for a modeling exercise. Data collected prior to the process failures, during operation at pseudo-steady state, were used to predict the conditions leading to the failure (obtain the initial concentrations for the dynamic simulation). Therefore, experimental data collected during the entire period between July 10 and October 13, 2003 were used for the dynamic calibration of the T U D P model parameters. Sin et al. (2005) indicated that one of the deficiencies of the S T O W A model is that it does not identify a method for determining the initial heterotrophic and autotrophic biomass concentrations for the dynamic modeling. This deficiency was addressed in the current study by using the averaged data collected during pseudo-steady state operation to obtain the initial biomass concentration for the dynamic simulation. A s for model validation, it was decided to use the data collected during an intensive sampling campaign in the period between M a y 2 and M a y 15 of 2003 to validate the calibrated model. This decision was based on the following reasoning. 1. The operating conditions o f the M E B P R process, such as process H R T , SRT, recycle flows and influent flowrate, were the same in M a y as that of July -October, period used for the calibration. 2. During M a y of 2003, sampling was done twice a day for the bioreactor zones and effluent concentrations and four times a day for the influent concentrations. So the 92 process was monitored closely and a complete set of data was available for this period. 3. The model calibration was performed using data collected during the summer, when the process temperature was relatively higher, so it was interesting to examine the model prediction using data collected during the spring season, at a slightly lower temperature. 4. This validation period exhibited failure in P-removal as well , so it provided a good opportunity to validate the model capabilities in predicting other process failures. Sampl ing Frequency The overall experimental phase in this project started on February 17 o f 2003. It was decided to carry out an intensive sampling campaign during the period between M a y 2 and M a y 15, in which samples were collected twice a day, corresponding to the number of times the influent storage tanks were filled daily during that period. Grab samples were collected manually from the influent, effluent, anaerobic, anoxic and aerobic zones and the samples were analyzed for N H 4 - N , N 0 3 - N , P 0 4 - P , T K N , TP, V F A , C O D t o t , C O D s o i and TSS. This sampling period produced a large number of samples (about 1120 for all tests). Analyzing all these samples was both costly and time consuming. Therefore, once the steady state simulation was completed, the model was used to choose a revised sampling frequency which would result in a reasonable number of samples, while still adequately capturing the important process dynamics. It was important to decide on the proper sampling frequency for each variable (concentration measurement). A recommended sampling frequency should be between (1/10 - 1/4) o f the dominant time constant of the response o f that variable to a step change in the process input, to avoid loosing important information about the process, aliasing (Soderstrom and Stoica, 1989). In order to determine a proper sampling frequency for each variable (concentration measurement), a simulation of the M E B P R process using the T U D P model calibrated using steady state data was used to examine model responses simulated for all process 93 variables to a step change in the influent PO4 -P concentration and the process H R T . Averaged measured influent concentrations were used in this simulation. The step change in the influent PO4-P was mainly used to examine the dominant time constant o f the process dynamics in response to disturbances to the process resulting from a change in the influent phosphorus concentration. The PO4-P and TP concentration profiles o f the biological zones showed significant variations when the influent PO4-P concentration was increased from 2.5 to 8 g P /m 3 while other variables were not affected significantly. The TP and PO4-P concentration profiles showed a first-order behavior and so the dominant time constants were estimated as the time at which the variable reached 63.2% of its ultimate value at steady state. The P 0 4 - P concentration in the anaerobic zone increased almost instantly. The increase in the anaerobic PO4-P concentration was proportional to the influent concentration increase indicating that the increase was not due to kinetic reactions. Therefore, the change in the anaerobic PO4-P concentration was not considered for selecting the sampling frequency, since the focus o f the modeling exercise was to calibrate the T U D P model to predict the M E B P R process kinetics. On the other hand, the anoxic and aerobic PO4-P concentrations increased as result of kinetic reactions with time constants of 6 and 19 days respectively. The time constants for the TP concentration profiles in the anaerobic, anoxic and aerobic zones were about 8 days for all zones. The change in the TP concentration was a reflection of changes in the biomass mass concentration for the bio-P removing organisms as a result of changes in the bio-P kinetics. Figure 4.1 shows the method for obtaining the dominant time constant of the TP concentration in the aerobic zone in response to a simulated increase in the influent ortho-phosphate concentration from 2.5 to 8 g P /m 3 . 94 220 Time (Days) Figure 4.1. M E B P R process dynamic simulation results for the TP profile in response to a step increase in the influent PO4-P concentration (PO4-P concentration increased from 2.5 to 8 g P /m 3 on day 201 at process SRT = 12 days and H R T = 10 hours) The same steps were repeated to examine model responses to a step change in the process H R T from 10 hours to 7 hours. The change in the H R T was introduced by increasing the influent flowrate. It was found that only mixed liquor concentrations (COD t o t , TP and TN) were significantly affected by the increased flowrate, while the soluble concentrations showed insignificant changes, which remained within the experimental error of each measurement. The time constants o f the mixed liquor concentrations varied between 6 to 12 days. A s discussed earlier, a suitable sampling frequency is suggested to be between '/w and '/4 of the dominant time constant o f the response o f each variable. Therefore, in light o f the results obtained in these simulations, it was decided to measure the soluble concentrations daily Monday to Friday and the mixed liquor concentrations in the bioreactor zones three times per week (Monday, Wednesday and Friday). These chosen 95 sampling frequencies resulted in a reasonable sampling effort and they fell within the recommended sampling frequency range. 4.1.2 Dynamic Data Collection and Screening Samples were collected from different locations in the M E B P R process to ensure capturing process dynamics along all intermediate stages. The data were then screened for outliers and missing data were estimated. Following that, the data were checked for proper excitation to ensure their suitability for dynamic calibration of the complex T U D P model. Finally, particulate phase concentrations, including TP, T N and C O D t o t , were filtered to reduce measurement noise resulting from grab sampling. M A T L A B 6.5 was used for the steps o f data screening and analysis. These steps are discussed in more detail in the following sections. Outlier Detection The data selected for the modeling were carefully analyzed for detection and removal of outliers using the outlier detection technique described in Section 3.3. Soluble concentration measurements contained very few outliers, however, mixed liquor concentrations ( C O D t o t , TP and TN) showed a higher percentage of outliers, accounting for up to 5% of the total number o f measurements. Visual inspection was used for the detection of outliers since the data set analyzed was not large (about 3 months of daily sampled data). However, for larger data sets or for more accurate and reliable detection o f outliers, the data set can be fitted to a model (e.g. A R times series model) or lightly filtered (data filtering is described below), and then the residuals, C(t) = ymeasured(t) - yPredicted(t) (4.1) can be plotted to detect possible spikes in the sequence. I f a spike in e(t) is abnormally large, then the corresponding output ymeasured(t) can be replaced by an estimated value. A simple estimation method is to take the average of proceeding and following measurements: 96 ymeasured(t): - 0.5 [VmeasuredO 1) + YmeasuredCt+l)] or to simply replace the measured value by the predicted, modeled, value. (4.2) Miss ing Data Influent data were used as inputs in the simulation model and so missing data (due to missing samples or outliers) in all influent concentration profiles needed to be estimated. A state-space model was used to estimate the missing data. Data estimation was done iteratively. The reconstructed data series (including the estimated data for the missing point) were modeled using a state-space model. The state-space model was then used to predict the missing data again. The iterations were terminated when the difference between two series of missing data estimates differed by less than the specified tolerance o f 1%. To begin iterating, the first model was built using linearly interpolated values for the missing data points. Figure 4.2 below shows the original data set along with the reconstructed values for the influent C O D t o t concentrations. 600 28-Jul-03 12-Aug-03 27-Aug-03 ll-Sep-03 26-Sep-03 ll-Oct-03 26-Oct-03 (Date) Figure 4.2. Estimated missing data for influent C O D t o t concentration used for the dynamic simulation of the M E B P R process 97 Proper Excitation The collected data were tested for sufficient excitation to ensure that the collected data set was suitable for dynamic modeling of the system. The degree o f excitation was checked by examining the power spectral density plot, a graph of signal intensity against frequency (Olsson and Newell , 1999). The cross-spectrum between the output and input was calculated as: M <M W) = zZK^WM(r)e-'WT (4.3) r = - A / where k > ) = ~ity{t)u(t-T) (4.4) and, ^ yu ( w ) ^ s m e cross-spectrum between the output and input. R (r) is the cross-covariance function between the output and input. WM (T) is the so-called lag window and M is the width of the lag window x lag value y(t) the output signal u(t) the input signal N the number of data points t time w frequency [0 to 2rc] The spectra were computed using a Hamming window at 128 equally spaced frequency values between 0 (excluded) and n. The power spectrum of each output variable (concentrations of N H 4 - N , PO4 -P , C O D t o t , TP and T K N in all zones) was obtained with respect to its corresponding influent variable to examine the range of excited frequencies 98 for each plot. It was important in this exercise that the input-output time series were in sequence. Therefore missing data were predicted. Results showed that all variables were mainly excited (showed high amplitude values) over the lower range of frequencies while the higher frequency range showed little excitation (low amplitude values). It should be noted that biological systems exhibit slow dynamics and, therefore the power spectrum of the biological processes are mainly excited over the lower frequency range. Figure 4.3 shows the power spectrum of the anaerobic zone ortho-phosphate signal with respect to the influent ortho-phosphate, while Figure 4.4 shows the power spectrum of the anaerobic total phosphorus concentration signal with respect to the influent total phosphorus. The figures show that the amplitude starts dropping as the frequency increases, which indicates that the data are excited at the lower frequency range. The other concentration profiles showed similar behavior for the power spectrum. Since the aim of the study was to model the biological process behavior, and not the fast hydraulic dynamics of the system, the degree of excitation exhibited by the collected data was deemed sufficient for modeling the dynamic behavior o f the M E B P R process. E < 1Q 1 I i i i i i ' • • 1 i i i i • * * • ' i i i i i • • » I 10'2 10'1 10° 101 Frequency (rad/day) Figure 4.3. Power spectrum o f the anaerobic zone PO4 -P concentrations with respect to the influent PO4 -P concentrations for the data used in the dynamic modeling of the M E B P R process 99 104 103 -' ' 1—1 • • . 1 1 1 1 — 1 — 1 1 1 > 1 — Amplitude 102 101 1 fv^ 7 : 10° 10 -2 10"1 10° 10' Frequency (rad/day) Figure 4 .4 . Power spectrum of the anaerobic zone TP concentrations with respect to the influent TP concentrations for the data used in the dynamic modeling of the M E B P R process Data Fi l ter ing Filtering of measured data was intended to remove measurement errors and large fluctuations in the signal, which could result in biased estimated model parameters. A filter was used to extract as much information as possible from the data. Filtering was carried out for the mixed liquor concentration data (TP, T K N and C O D t o t ) for the M E B P R process since these data exhibited large variations. These fluctuations were believed to be due to grab sampling of the mixed liquor. It was necessary to filter these data to remove these fluctuations prior to model calibration so that when the model predictions were fitted to the measured data, a distinction could be made between real dynamic changes in the process and unrealistic changes. The common digital first order low-pass filter given in Equation 4.5 (Ogunnaike and Ray, 1994) was used to remove high frequency noise. The filtered value, y ; , was computed by adding a weighted version of the previous filtered point, y w , to the previous measured 1 0 0 value, y j . The coefficient a usually has a value between 0 and 1. If a is near 1, the noise is greatly reduced at a cost of poor agreement with real changes in the measurement signal (may loose real process dynamics). If a is small, near 0, there is a poorer reduction in measurement noise level, but the filter w i l l track real signal changes more easily (similar to the original signal). For a = 0, the filter output is identical with the measurement value. y ; = o r y i _ 1 + ( l -a )y , . (4.5) The power spectrum was used to choose the proper value of a using Equation 4.6: _T S a = e T r (4.6) where if is the filter's time constant and T s is the sampling frequency. The filter's time constant was estimated as suggested by Ogunnaike and Ray (1994) using Equation 4.7: — > w c (4.7) where w c is the corner frequency identified from the power spectrum as the frequency at which the signal intensity (amplitude values) started dropping. The term — is referred to as the cut-off frequency. Selecting the filter time constant based on the relationship in Equation 4.7 removed unwanted high frequency noise while capturing important process dynamics. Figure 4.5 shows the method used for determining the corner frequency and the cut-off frequency from the power spectrum for the anaerobic zone TP data for the M E B P R process. The corner frequency was found to be 0.2 and the selected cut-off frequency was 0.4. These values resulted in a filter time constant of 2.5 days. Analyzing the power spectrum for the mixed liquor concentrations for TP, T N and C O D t o t in all zones resulted in a similar filter time constant of 2.5 days. Using Equation 4.6 with a sampling 101 frequency, T s , o f 1 day, a was found to be 0.67. Therefore the following digital low pass filter was used for all mixed liquor concentrations: • y, =0.67 y M + 0 . 3 3 y,. (4.8) Missing data points were estimated prior to filtering the signals. The function used for the filter was a zero-phase forward and reverse digital filtering. After filtering in the forward direction, the filtered sequence was then reversed and run back through the filter given in Equation 4.8. The filtered data, y i ? were the time reverse o f the outputs of the second filtering operation. M A T L A B 6.5 was used for data filtering. Figure 4.6 shows a plot of the filtered and unfiltered data for the anaerobic zone TP concentration profile for the M E B P R process. It is clear that large fluctuations were removed from the signal resulting in a smoother response. Figure 4.7 shows the frequency content of the filtered and unfiltered signals. The frequency response of the two signals was computed and plotted to ensure that the signal still retained the important process dynamics at the low frequency range as shown in Figure 4.7, where the two signals are identical at the low frequency range (< 1 rad/day). Therefore, the filter only removed high frequency noise from the signal. 102 10 i t r 10' E < 10 10 Coraer freq. 10" 10" 10" Frequency (rad/day) 10 Figure 4.5. Choosing the proper cut-off frequency for the filter design to filter the anaerobic zone TP concentrations of the M E B P R process 110 -e Unfiltered data —*- Filtered data 60 Time (Days) 140 Figure 4.6. Filtered and unfiltered measured dynamic data for the TP concentration in the anaerobic zone of the M E B P R process Frequency (rad/day) Figure 4.7. Frequency content of the filtered and unfiltered measured dynamic data for the TP concentration in the anaerobic zone of the M E B P R process 4.1.3 Dynamic Simulation Using the TUDP Model Parameters Determined During the Steady State Calibration of the MEBPR Process Once the dynamic data collected from the M E B P R were analyzed and processed, the next step was to use the measured dynamic data for modeling the process. The dynamic simulation was first carried out using the T U D P model with the previously determined steady state calibrated parameters, presented in Table 3.3, and the default model parameters reported by Meijer (2004). This step was undertaken to examine the adequacy of the T U D P model parameters determined during the steady state model calibration for predicting the dynamic behavior of the process. The M E B P R process configuration was implemented in A Q U A S I M 2.0 as shown in Figure 2.6. The reactor zone volumes were set as given for the M E B P R in Table 2.1 and the process operating conditions given in Table 2.2. The data collected from the M E B P R pilot plant during the period between August 22 and October 13 of 2003, when the process experienced failure in P-removal and high anoxic 104 nitrate concentrations in the period, were used for the dynamic modeling for the M E B P R process. However data collected from the process between July 10 and October 13 were used for the dynamic modeling to be able to model the process behavior prior to the failure in case certain conditions lead to such behavior. Moreover, the averaged measured influent concentrations used for the steady state modeling o f the M E B P R process, described in Chapter 3, were used as inputs to the simulation for the first 100 days prior to the dynamic simulation, to allow the process to reach steady state and to obtain the correct initial biomass concentrations required for the dynamic modeling. The daily measured influent concentrations, along with the recorded operating conditions including flowrates, p H , temperature and D O concentrations were used as inputs to the dynamic simulation model. Concentration profiles predicted by the model for the M E B P R anaerobic, anoxic, and aerobic zones and the effluent (permeate) were compared to the measured data obtained from the pilot plant during the period between July 10 and October 13. The predicted concentrations for the TP, T N and C O D t o t were compared to the unfiltered measured mixed liquor concentrations. The resulting model predictions showed reasonable agreement with measured data for C O D and ammonia concentrations (results not shown). However, the concentration profile of nitrate in the anoxic zone showed large discrepancies between the predicted and measured values as shown in Figure 4.8. These discrepancies were mainly present during the periods in which the process experienced elevated nitrate concentrations in the anoxic zone. The simulation model predicted low nitrate concentrations for the entire experimental period. This indicated that further calibration was required to better predict the anoxic nitrate concentrations. 105 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) Simulated_ANO - ± - MeasuredANO Simulated_AER -•-Measured_AER Figure 4.8. T U D P model predictions for N O 3 - N concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined in the steady state calibration stage Furthermore, the model was not capable of predicting the P-removal failure using the T U D P model parameter values determined in the steady state calibration stage. The TP concentrations in all zones were overestimated by the model as shown in Figure 4.9. The PO4-P concentration profiles in all biological zones were not predicted accurately as well , as presented in Figure 4.10. The model predicted higher concentrations o f PO4-P in the anaerobic zone, indicating higher simulated phosphate release rates than the observed rates and the anoxic PO4-P concentrations were over-predicted as well . The model also predicted low PO4-P concentrations, essentially zero, when the measured data showed elevated concentrations of aerobic and effluent PO4-P. 106 250 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 • Simulated_ANA —•— Measured ANA - Simulated_ANO -A— Measured ANO •Simulated_AER Measured AER Figure 4.9. T U D P model prediction of the TP concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined in the steady state calibration stage "I 30 25 20 S 15 o U a. p e. 10 28-Jun-03 II 3 1 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 — • Simulated_ANA -•— Measured ANA - Simulated_ANO -A— Measured ANO —— Simulated_AER -•—Measured AER Figure 4.10. T U D P model prediction of the P0 4-P concentration profiles in all biological zones of the U B C M E B P R process using the T U D P Model parameters determined in the steady state calibration stage 107 In light of these results, it was clear that further calibration was required to better fit the dynamic M E B P R process data under these operating conditions. Specifically, further calibration of the denitrification and P-removal components of the model was required. Calibration techniques presented in the literature have certain limitations, as discussed in Chapter 1, especially in the area of parameter estimation. Model parameter calibration was mainly performed manually in the studies presented in the literature. Dochain and Vanrolleghem (2001) indicated that no perfect minimization algorithm for nonlinear objective functions exists so far. Sin et al. (2005) indicated that using a combination o f process expert knowledge and mathematical/statistical tools for model parameter calibration would be a better alternative. Therefore, in the present work it was decided to use a combination of process knowledge and mathematical tools in the selection of the T U D P model parameters to be fitted for the dynamic calibration and in estimating the values o f the selected parameters. The utilized approach is described in detail in the following sections. 4.1.4 Selection of the TUDP Model Parameters to be used for Calibration in the Dynamic Modeling of the MEBPR Process In a correct calibration, the parameters have a unique value with a physical interpretation. This is only possible i f the model is identifiable, resulting in a unique set of parameters when tuning the model to a given set of (output) data. If parameters are not identifiable, then a change in one parameter can be compensated almost completely by a proportional shift in another, still producing a satisfying fit between the measured data and the model predictions (Dochain and Vanrolleghem, 2001). A s a result, usually a set of parameters are selected for calibration while keeping the rest at their reported default values. The first study on identifiability analysis o f activated sludge model parameters was presented by Wejiers and Vanrolleghem (1997). Drawing on control theory, they focused on the problem of the selection of best identifiable parameter subsets for parameter estimation. The authors applied their technique in calibration of the A S M 1 model. 108 Another approach was suggested by Brun et al. (2001) based on linear regression diagnostics. Based on these two studies, Brun et al. (2002) presented a paper on identifiability of A S M 2 d parameters. However their approach to determining the most identifiable parameters relied heavily on expert knowledge and experience with activated sludge systems and since they only had data available for soluble ammonia and ortho-phosphate, they decided to fix many parameters, as these would not be reasonably estimated from the available data. Therefore, only a very small set of parameters was tested using their proposed technique. The disadvantage of these techniques lies in their difficulty when applied to the complex E B P R models such as the A S M 2 d or T U D P models, as these always require and rely on extensive system knowledge and assumptions. For these reasons, other groups such as Meijer et al. (2002b) solely based their selection of the subset o f parameters fitted for model calibration on process knowledge and uncertainty in parameters. Basically parameters that were known to be dependent on the characteristics of the system and those that were not very well known or defined were selected for calibration, while other parameters based on reaction stoichiometry for example, were kept constant. However, Brun et al. (2002) and Weijers (2000) argued that a major drawback of this method is that parameter values obtained may be biased by the values o f the fixed parameters and their uncertainty is usually underestimated. In the present work, the selection of the calibration parameters was a critical and challenging step in the overall calibration procedure. Since the T U D P model had over 60 parameters and, due to model complexity, a strong correlation existed between many model parameters, which reduced parameter identifiability. Furthermore, many of the state variables (e.g. X G L Y , X P P , X P A O , etc.) were not measured and so a limitation existed on the information that could be utilized in the calibration process. Due to these limitations, it was difficult to apply any o f the mathematical methods described above to determine the set o f model parameters to be used for the dynamic model calibration step. Furthermore, using the set o f parameters identified by the S T O W A calibration procedure (Hulsbeek et al., 2002) or Meijer et al. (2002b) for calibrating the T U D P was not sufficient to fit the measured nitrate and phosphorus concentrations obtained in the 109 present work. Therefore, a combination of different methods was used for selecting the parameters of the T U D P model to be used for the dynamic calibration. First, the parameters estimated for the steady state calibration were selected for further calibration. Then the other parameters were selected based on: 1. literature review of modeling studies in which the A S M 2 d and the T U D P models were used, 2. results o f sensitivity analysis using dynamic influent data, and 3. trial and error. A n extensive review of studies published in the literature, in which the A S M 2 d or the T U D P models were used, was carried out to identify all parameters that were estimated in these studies. This was done to determine the least certain parameters that are system specific and therefore need to be calibrated for each system. In the present research, emphasis was placed on parameters related to the denitrification and P-removal processes since dynamic nitrate and ortho-phosphate profiles were not predicted well using the T U D P model parameter values estimated in the steady state modeling of the process. Table 4.1 lists the studies reviewed for this purpose and the calibration parameters chosen in each o f these studies. Table 4.1 shows that some parameters required calibration for each system such as: K N H IT (saturation coefficient for ammonium) for nitrification, r\ N o 3 (heterotrophic reduction factor) and K02 (saturation/inhibition coefficient for oxygen) for denitrification, and r |f e (anaerobic hydrolysis reduction factor), q f e (maximum fermentation rate), r ] P N 03 (reduction factor for denitrifying P removal) and gpp (saturation reduction factor for poly-P formation) for the E B P R process. Therefore, it was decided to estimate these parameters in the present calibration process. The suggested values reported in literature for these parameters were used as constraints for each parameter during parameter estimation. Vanrolleghem et al. (2003) emphasized the importance of using dynamic sensitivity analysis to examine the sensitivity of the model output variables (concentrations) to 110 specific model parameters. Therefore, prior to deciding on the final set of the T U D P model parameters to be used in the dynamic model calibration, it was decided to examine the dynamic sensitivity of certain concentrations to the T U D P model parameters. Table 4.1. Summary o f calibration parameters reported in the literature for A S M 2 d and T U D P models Study Model Parameters Original Value Estimated Values Comments Carrette et al. (2001) ASM2d Tlfe "1 N03 b A U T N / A N / A N / A N / A N / A N / A 41% of total COD originating from a textile industry Penya-Roja et al. (2002) ASM2d KNH "1 N03 1 0.8 0.25 0.58 Referenced Gernaey et al. (1998) 0.06 ±0.23 Referenced Henze (1987) 0.56-0.58 Satoh et al. (2000) ASM2 PAUT KNH qfe 1 1 3 0.9 8 0.5 r)fek h 0.3 0.6 (at k h of 3.0 and 20 °C, n f e changes from 0.1 to 0.2) Van Veldhuizen et al. (1999) TUDP qfe kpp n P M N03 1.71 0.05 0.5 1.0 0.07 0.7 Brdjanovic et al. (2000) TUDP qfe "1 N03 ^ G L Y 3 0.5 1.09 1 0.8 0.45/0.15 Meijer et al. (2001) TUDP f x S i n K(32 N / A 0.2 0.493 0.7 Meijer et al. (2002a) TUDP f x S i n KQ-2 N / A 0.2 0.2024 0.3 Should not calibrate i N B M Meijer et al. (2002b) TUDP f x S i n K(32 k G L Y Ypo4 N / A 0.2 1.09 0.48 0.493 0.7 0.93 0.35 Dynamic Modeling of Start-up q s m a x 9.67 8 Meijer (2004) TUDP kpp kpHA gpp 0.45 7.55 0.25 0.1 5.51 0.22 Ph.D. Thesis Dynamic sensitivity analysis was carried out in A Q U A S I M 2.0 using the measured dynamic influent concentrations, actual process conditions (temperature, D O concentrations, H R T , SRT, etc.) and the steady state-calibrated T U D P model. The 111 sensitivity function given by Equation 4.9, was used to obtain a relative change in a model output variable to the relative change in a model parameter value. where x is the model parameter, y is the variable (concentration), and S is the sensitivity function value. This function was chosen since it results in a non-dimensional measure allowing for comparison of the effect o f different parameters on a common variable as well as the effect o f a specific parameter on different variables. Dynamic sensitivity analysis was performed for the variables with the concentration profiles that were not predicted well by the steady state-calibrated T U D P model such as PO4-P in the anaerobic, anoxic and aerobic zones and N O 3 - N in the anoxic zone. The dynamic sensitivity analysis for the V F A component in the anaerobic zone was also analyzed because it is a critical factor in the P-removal process. Using the dynamic influent data resulted in a dynamic sensitivity function that changed with time. Most parameters demonstrated a relatively constant sensitivity profile over the experimental time period. However for some parameters, the sensitivity function profile fluctuated, showing high sensitivity o f the variable (concentration) to the parameter value in some instances, while it remained insensitive to the parameter for the rest of the period. Examples of the sensitivity function profiles for the sensitivity of the PO4-P concentration in the anaerobic zone to three parameters, Y p P (aerobic yield for formation o f Xpp), Ypo4 (anaerobic yield for ortho-phosphate release) and fxsin (fraction of slowly biodegradable C O D in the influent particulate C O D ) are shown in Figure 4.11. The plot presents the daily relative change in the anaerobic PO4-P concentration, predicted by the steady state calibrated T U D P model, for a 100% change in the parameter. The change in the sensitivity function value from day to day is due to changes in the dynamic influent concentrations, which affect the state variables of the system and in this way the sensitivity to the parameters. S = (4.9) 112 3 Y P P O Y P 0 4 * * * f xsin (Date) Figure 4.11. Dynamic sensitivity function values for PO4-P in the anaerobic zone of the M E B P R process for Y p P , Ypcw and fxsin parameters of the T U D P model ( H R T = 10 days, S R T = 12 days and QINF = 5.1 m 3/day) The plot of the sensitivity function for Y ° p , for example, shows large fluctuations reaching a maximum sensitivity value o f about 7 on July 24 while it drops to a minimum of about -2 at other times and remains essentially zero at most times. However, the sensitivity functions for both Ypo4 and fxsin, remained essentially constant at about 1 and 0.4 respectively for the whole time period. Since the model contained a large number of parameters and it was almost impossible to compare the dynamic sensitivity profiles of all parameters to analyze them, the maximum and minimum values were determined for each profile and plotted on a bar graph along with the median to give a better indication of the sensitivity of each variable (concentration) to different parameters o f the T U D P model. The bar graphs presenting the sensitivity analysis results for these variables are shown in Figures 4.12 to 4.16. Only parameters with maximum, minimum or median sensitivity values above 0.1 or below -0.1 are presented in these figures. 113 Sensitivity analysis alone can not differentiate between more defined (e.g. stoichiometric model parameters) or less defined model parameter values (Gernaey et al., 2004). Therefore, the results of the dynamic sensitivity analysis were combined with results of calibration studies in the literature. Parameters that have been used in literature as calibration parameters and resulted in sensitivity values above 0.1 were considered as potential calibration parameters. Figure 4.12 presents the sensitivity analysis results for the PO4-P concentrations in the anaerobic zone. The median value was mainly used to determine the sensitivity o f the variable to a parameter value, while the maximum and minimum sensitivity values were presented to illustrate the range o f variation for the parameter o f interest. Figure 4.12 shows high sensitivity for the parameters used in the literature for the calibration of the E B P R process, as listed in Table 4.1, including r\fe (anaerobic hydrolysis reduction factor), f X s i n (fraction o f slowly biodegradable C O D in influent particulate C O D ) and Ypo4 (anaerobic yield for phosphate release) for example. A s expected, the concentration of PO4-P in the anaerobic zone, is most sensitive to the parameter Ypo4, since it directly affects this concentration. The plot also shows high sensitivity towards the process SRT, but since the SRT in this study was tightly controlled, it was assumed to be essentially constant in the simulation. Examining the results of Figure 4.12 and comparing them to results reported in the literature, the following were considered as potential calibration parameters for anaerobic zone P O 4 - P : r)fe, f X s i n , * N X S (nitrogen fraction o f particulate substrate), k p p (poly-phosphate formation rate), q f e (maximum fermentation rate), q s m a x (maximum anaerobic acetate uptake rate), Y H (heterotrophic yield for growth on substrate) and Y P o4 (anaerobic yield for phosphate release). 114 s a u a s to • Maximum Sensitivity Value • Minimum Sensitivity Value • Median Sensitivity Value II II ~T3~ S o-ft, J j L B . C M S - * * J 5 ei as > . J J — i 5 n 2 a o z i : -o u Parameters Figure 4.12. Averaged sensitivity analysis values of the P 0 4 - P concentration profile in the anaerobic zone to T U D P model parameters for The M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, SRT = 12 days and QIN F = 5.1 m 3/day) Figure 4.13 presents the sensitivity of the PO4-P concentration profile in the anoxic zone to the T U D P model parameters. The figure shows high sensitivity for yield coefficients. However, since the yield coefficients are well established and defined (van Veldhuizen et al., 1999) based on reaction stoichiometry, it was decided not to include the yield parameters in the calibration process except for Y H and Ypo4, since they have been calibrated in other studies and are known to depend on the system. Based on the sensitivity analysis results and the review o f the literature studies, the following parameters were to be selected as calibration parameters for fitting the anoxic P 0 4 - P : r|fe, n H N o 3 (reduction factor for denitrification), n p N o 3 (reduction factor for 115 denitrifying P removal), f X S i n , gpp (saturation reduction factor for poly-P formation), i N X S , ipxs (suggested by Meijer, 2004), k p p , and Y H and Y P 0 4 . » 2 m > 0 1 -1 • Maxium Sensitivity Value • Minimum Sensitivity Value • Median Sensitivity Value -2 -3 -5 •=> S Z - - 3 > - S • R • R R ( • 1 Q ' n CL a-5 J ! 5 2 to cn 1 Jd cn x I Q n Q B g n E L s y y S . O, a. a O a. pi JJ Jd * >. a. 5 i e i a _ 3 - — I a - — i -»' a L J 3 s Parameters Figure 4.13. Averaged sensitivity analysis values of the PO4-P concentration profile in the anoxic zone to T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, S R T = 12 days and QINF - 5.1 m 3/day) Furthermore, Figure 4.14 presents the sensitivity values for the PO4-P concentration in the aerobic zone. Again, the yield parameters show the highest sensitivity, followed by the kinetic rate parameters such as k p p (the poly-phosphate formation rate) and the sludge fractions such as ipxs- Based on these results and the literature studies, the following parameters were selected as candidates for calibration for fitting the aerobic PO4-P: r)fe, H P jp r| NO3, r| NO3, fxsin, gpp, INXS, ipxs, k G L Y (glycogen formation rate), k P H A ( P H A degradation rate), K N H , k p p , q f e , q s m a x , Y H and Y P 0 4 . 116 B Maximum Sensitivity Value • Minimum Sensitivity Value • Median Sensitivity Value Figure 4.14. Averaged sensitivity analysis values of the PO4-P concentration profile in the aerobic zone to the T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13 (HRT = 10 hours, SRT - 12 days and QINF = 5.1 m 3/day) Results for sensitivity analysis for the N O 3 -N concentrations in the anoxic zone are presented in Figure 4.15. It is clearly shown that the anoxic nitrate concentration is most sensitive towards the yield parameters, followed by the parameter for the nitrogen content of particulate substrate, iNxs- When comparing the results for sensitivity analysis with the results o f the reported calibration studies in the literature, the following parameters can serve as potential calibration parameters when compared to calibration studies reported in the literature: r|f e, n H N 0 3 , r)?N03, fxsin, INXS, K02 (saturation/inhibition coefficient for oxygen), k P H A , k p p , q f e , and q s m a x . 117 • Maximum Sensitivity Value j • Minimum Sensitivity Value • Median Sensitivity Data^  2 + ss > 1 e e •p § 3 0 £> '> I e so -2 I n ' -3 + P J. VI z , z * « - A 2 8 B x a e j B O TJ1 TT [I ill a °-, 8! H 3 B j i . i - i - 5 - u Parameters g z z •Jo o ^ Ba — i r > > Figure 4.15. Averaged sensitivity analysis values of the NO3 -N concentration profile in the anoxic zone to the T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, S R T = 12 days and QmF = 5.1 m 3/day) Finally, sensitivity analysis results for the V F A concentration in the anaerobic zone, presented in Figure 4.16, indicate the following as potential calibration parameters: rjfe, fxSin, iNXS, K-02, KpHA, q f e , q s ™ " , koLY , Y H > and Ypo4-Furthermore, work carried out by Zhang and Hal l (2006) for the estimation of the heterotrophic yield, Y H , and decay coefficient, b H , for sludge samples taken from the M E B P R and C E B P R processes at the U B C pilot plant showed similar values for both systems. However, both values determined for Y H by Zhang and Hal l (2006) for the C E P B R and the M E B P R processes of 0.59 ± 0.14 and 0.5 ± 0.08 respectively were lower than the default value o f 0.63 for the A S M 2 d model. Therefore, it was decided to include Y H in the set of parameters used for the T U D P model calibration. 118 10 - 7 . 5 + - 1 0 ^ - N Parameter Figure 4.16. Averaged sensitivity analysis values of the V F A concentration profile in the anaerobic zone to the T U D P model parameters for the U B C M E B P R process using dynamic influent data for the period July 10 - October 13. (HRT = 10 hours, SRT = 12 days and QFNF = 5.1 m 3/day) It should also be noted that other parameters could also have been selected for calibration based on process knowledge and trial and error depending on the calibration step. Every system is unique and parameters calibrated in one system may not be adequate for another one, so the process could be iterative. During the calibration step in the present study, it was found that in order to predict the elevated N O 3 - N concentrations observed in the anoxic zone, the anoxic hydrolysis reduction factor, r | L N 0 3 , needed to be adjusted, so it was added to the parameter calibration list as well . The calibration of the anoxic hydrolysis reduction factor is discussed in more detail in the parameter estimation section. In light of these findings, the concern raised by Sin et al. (2005) regarding the standard set of calibration parameters suggested by the S T O W A calibration procedure is valid. In the S T O W A calibration procedure for activated sludge systems (Hulsbeek et al., 2002), the authors suggested a fixed list o f calibration parameters to be used for all 119 calibration studies, however, in the present work, it was found that every system is unique and parameters other than the ones suggested in the S T O W A protocol required adjustment to fit the measured data. Based on the findings of these analyses, the final list o f calibration parameters was prepared. Table 4.2 lists the parameters selected for the dynamic calibration of the T U D P model for the M E B P R process. The table also lists the default values o f these parameters as reported by Meijer (2004) and the selected higher and lower limits for calibration. The ranges for the parameters were mainly based on the reported values in the literature and the results of manual calibration as discussed in the next section. T a b l e 4.2. Selected parameters for calibration of the T U D P model for the dynamic M E B P R process data Parameter Units Meijer (2004) Minimum Maximum r|fe 0.2 0.01 0.3 r| N03 0.8 0.5 0.8 R\ N03 0.8 0.2 0.8 U N03 0.8 0.5 0.9 fxSin g C O D / g C O D 0.439 0.5 1 gPP 0.22 0.1 0.5 JNXI g N/g C O D 0.03 0.03 0.07 INXS g N/g C O D 0.03 0.03 0.07 !PXI g P/g C O D 0.01 0.0005 0.01 !PXS g P/g C O D 0.01 0.0005 0.01 k G L Y T g COD/(g C O D • d ) 0.93 0.15 1.09 gN/m 3 1 0.06 1 K 0 2 g 0 2 / m 3 0.2 0.2 2 k p H A g COD/(g C O D • d ) 5.51 5.51. 7.55 k T K PP T g P/(g C O D • d ) 0.1 0.1 0.45 qfe max T I s g COD/(g C O D • d ) 3 0.5 3 g COD/(g C O D • d ) 8 8 9.67 Y H g C O D / g C O D 0.63 0.37 0.63 YpQ4 g C O D / g C O D 0.35 0.35 0.77 All parameters are reported at 20 °C. T These parameters are temperature dependent. * Default values as reported by Meijer (2004) 4.1.5 P a r a m e t e r C a l i b r a t i o n After identifying the model parameters to be used in the calibration process, the next step was to estimate the model parameters that would result in a good model fit o f the measured concentrations. In the present work, it was decided to use manual model 120 parameter estimation based on the method described by Meijer et al. (2002b) and to compare the results to those o f a mathematical parameter estimation method. Both methods were used since it has been reported in the literature that the use o f advanced mathematical approaches increases the accuracy o f the calibration process (Sin et al., 2005). However, other studies indicated that relying solely on mathematical algorithms does not differentiate between the more certain parameters (stoichiometric) and the less accurate ones (Meijer et al., 2002b). Therefore, it was decided to compare the results of both methods. The daily measured influent concentration data, along with the temperature and p H profiles, were used as model inputs. First, manual calibration of the T U D P model parameters was completed using the stepwise calibration approach of Meijer et al. (2002b). Then the mathematical parameter estimation module in A Q U A S I M 2.0 was used for calibrating the T U D P model parameters using the simplex algorithm. The parameters determined during the manual calibration step were used as the initial parameter values in the simplex algorithm. Then some of the parameters determined by the simplex parameter estimation were used to better fit the measured data in the manual calibration. The process was iterative between using manual calibration and information of the mathematical parameter estimation to obtain the set of parameters that resulted in a reasonable fit o f all concentration profiles simultaneously. This calibration approach and the results of both calibration methods are presented in the next section. In order to obtain a measure of the fitness of the model predictions to the measured data, the Pearson correlation coefficient (sample correlation coefficient) between the measured and predicted data was computed for all data sets according to Equation 4.10. r = (4.10) 121 where r is the correlation coefficient, x and y are the predicted and measured data sets and n is the number of data points. The value of r is between -1 and 1. If the value of r is closer to 1 or -1, this indicates a strong correlation between the measured and predicted data, while a value of 0 indicates no correlation between the two. It should be noted that for the manual calibration and the automatic parameter estimation methods, the filtered measured mixed liquor concentrations for TP, T N and COD t o t were used for comparing model predictions to the measured concentrations. The filtered mixed liquor concentrations were used to avoid biased parameter estimation due to measurement error and noisy data. 4.1.5.1 M a n u a l M o d e l P a r a m e t e r E s t i m a t i o n M e t h o d The T U D P model parameters determined in the steady state calibration step were used as the starting point for the stepwise calibration o f the T U D P model. The default values as given by Meijer (2004) were used for the rest of the T U D P model parameters. The manual calibration was mainly based on the stepwise approach o f Meijer et al. (2002b) in addition to process knowledge and trial and error. The calibration process is described in the following section. I. Calibrating C O D Removal Fitting CODtot. The first step in the stepwise calibration approach of Meijer et al. (2002b) was to fit the COD t o t concentrations in the biological zones using the fraction, fxsin, Xs/(Xs+Xi). This parameter represents the fraction o f slowly biodegradable C O D in the total influent particulate C O D . During steady state calibration, this parameter was found to be 0.85, which resulted in a reasonable fit o f the averaged measured C O D t o t concentrations in the biological zones. The same value o f 0.85 was sufficient to predict the trend of the dynamic C O D t o t data in the biological zones as shown in Figure 4.17. The filtered measured C O D t o t data was used for anaerobic and anoxic zone data, but the actual measured COD t o t profile was used for the aerobic zone. The aerobic C O D t o t measured data set had a larger number of missing data points (due to measurement error) and therefore estimation of the missing points, which is required prior to filtering, was not 122 reliable. A s a result, it was decided to use the actual measured data set especially since it did not exhibit large fluctuations. The predictions of the C O D t o t concentrations in all zones were considered reasonable since the correlation coefficient values of the predicted and filtered measured data were found to be 0.87, 0.91 and 0.56 for the anaerobic, anoxic and aerobic concentration profiles respectively. Furthermore, the TSS profiles in all biological zones were very well predicted by the simulation as shown in Figure 4.18. The correlation coefficient values obtained were 0.94, 0.97 and 0.91 for the anaerobic, anoxic and aerobic profiles respectively. The change in the C O D t o t and TSS concentration profiles observed early September is believed to be due to the change in the filling times of the feed storage tanks, which resulted in an overall drop in the influent organic loading. 6000 7-Aug-03 17-Aug-03 27-Aug-03 6-Sep-03 16-Scp-03 26-Sep-03 6-Oct-03 16-Oct-03 (Pate) — • Simulated A N A Simulated A N O Simulated A E R • Filtered Measured A N A A Filtered Meausred A N O — • — Measured A E R Figure 4.17. T U D P model prediction for the C O D t o t concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 123 4500 4000 3500 3000 H 2500 (SB —^' ag 2000 H 1500 1000 500 0 28-Jun-03 — • Simulated_ANA Simulated_ANO SimulatedAER —•— Measured_ANA A Measured_ANO > Measnred_AER Figure 4.18. T U D P model prediction for the TSS concentration profiles in all biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data Fitting Soluble COD in the Effluent. The effluent COD S O ] concentration profile was well predicted and no further calibration was required. Simulation results are shown in Figure 4.19. The prediction closely represents the measured data profile with a correlation coefficient o f 0.93, indicating that the model well describes the mechanism of C O D removal. II Calibrating Nitrification and Denitrification Calibration of nitrogen components was performed according to the procedure outlined by Meijer et al. (2002b) starting with fitting the T N concentrations in the influent and the biological zones, followed by the N H 4 - N and NO3 -N concentration profiles in all biological zones. Fitting TN. The T N data were fitted by adjusting the activated sludge N-fractions, INXI to 0.035 and iNxs to 0.04. The parameters were adjusted to fit the influent T N data. Measured and simulated concentration profiles of T N in the M E B P R are shown in Figure 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) 124 4.20. After adjusting the parameters iwxs and itoa, the influent T N concentrations were reasonably predicted. The correlation coefficient for the measured and predicted T N concentrations was 0.46, which was deemed reasonable considering the measurement error. The filtered anaerobic and anoxic T N concentrations were well fitted by the model with correlation coefficient values of 0.96 and 0.8 respectively. The measured T N concentrations for the aerobic zone showed an increase in value between mid September to mid October, which was not predicted by the model resulting in a low correlation coefficient of 0.12 between the filtered measured and predicted data. However, beside that, the general trend was reasonably predicted. 250 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) Figure 4.19. T U D P model prediction for C O D s o i concentration profile in the effluent of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 125 350 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 15-Nov-03 (Date) Simulated INF " - — Simulated ANA Simulated ANO Simulated AER Simulated E F F —©-• Measured INF — • — Filtered Measured ANA — i — Filtered Measured ANO — • — Filtered Measured AER — * — Measured E F F Figure 4.20. T U D P model prediction of the filtered measured T N concentration profiles in the biological zones, influent and the effluent of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data Fitting Nitrification. The N H 4 - N concentration profiles were fitted well using the value of 0.1 for the half-saturation coefficient parameter o f the T U D P model, KNH, determined during the steady state calibration stage. Results o f the T U D P model prediction of N H 4 - N concentration profiles are plotted in Figure 4.21. The correlation coefficient values for the anaerobic and anoxic concentration data were 0.79 and 0.89 respectively. Both the predicted and measured effluent N H 4 - N concentrations were essentially zero and therefore, the correlation coefficient was not computed. In the present study, the sludge mass fraction in the aerobic zone was determined to be 75% of the total sludge mass in the system, while Ramphao et al. (2005) suggested a target sludge mass fraction o f 50% for the aerobic zone. In light of this finding, it is believed that the present M E B P R aerobic zone was over-designed and so there was no limitation on the nitrification process and therefore, low effluent N H 4 - N concentrations were achieved throughout the experimental period. 126 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) — • Simulated ANA SimulatedANO SimulatedEFF —•— Measured_ANA —ir— Measured_ANO —•— Measured_EFF Figure 4.21. T U D P model prediction for the N H 4 - N concentration profiles in the biological zones and the effluent of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data Fitting Denitrification. After calibrating the coefficient for nitrogen content o f particulate substrate, iNxs, and the coefficient for nitrogen content of inert particulate C O D , i N x i , to fit the influent T N concentrations, the effluent nitrate concentration was found to be under-predicted. In the work of Meijer et al. (2002b), the effluent nitrate concentration was over-predicted and therefore, the saturation / inhibition coefficient for oxygen, K02, was increased to fit the effluent nitrate concentration. Changing K02 affects the anoxic gradient inside the floe in the aerobic zone. In full-scale plants, where the aerobic zone may not be fully mixed due to the large zone volume, anoxic conditions may be present inside the floe. Therefore, increasing the value of K02 basically increases the anoxic volume fraction in the system. However, in the present study, the situation was the opposite, as the effluent nitrate concentration was under-predicted and adjusting K o 2 did not result in a good fit o f the measured effluent nitrate concentrations. Therefore, it was decided not to adjust this parameter in the calibration stage and it was kept at its default value of 0.2. 127 Instead of calibrating K02, the anoxic heterotrophic reduction factor, r » H N 0 3 , was adjusted. This parameter was introduced in the model to reduce the heterotrophic growth rate under anoxic conditions as compared to the aerobic heterotrophic growth. The effluent N O 3 - N concentration profile was better predicted by the T U D P model when r i H N03 was reduced from the default value of 0.8 to 0.65. Reducing this parameter results in a lower anoxic heterotrophic biomass growth and so less biomass available for denitrification (i.e., higher effluent nitrate concentration). After fitting the effluent nitrate concentration profile, further calibration was required to better predict the high anoxic nitrate concentrations experienced during that period. The sensitivity analysis results for the anoxic nitrate, Figure 4.20, showed that the iwxs parameter was the most critical calibration parameter (after the yield parameters). However, this parameter had been adjusted previously to fit the T N concentrations in the biological zones. Further calibration efforts showed that it was only possible to simulate the high nitrate concentrations in the anoxic zone by reducing the anoxic hydrolysis reduction factor, r | L N 03- This parameter was introduced in the model to reduce the hydrolysis rate under anoxic conditions when compared to aerobic conditions, since research results showed that hydrolysis rates were higher under aerobic conditions compared to anoxic conditions (Henze et al., 2000). The fact that it was only possible to simulate the high anoxic nitrate concentrations by reducing r) LN03 was in agreement with the results of Siegrist et al. (1999). In their work, the authors reported that the reduced denitrification activity could only be modeled in the I A W Q A M S Models No. 1 and 2 by reducing the anoxic hydrolysis rate. Therefore, in the present work, r | L No3 was adjusted to fit the anoxic nitrate concentration profile by reducing it from the default value of 0.8, to 0.2. The value o f 0.2 for r | L N03 was outside the range reported in literature for this parameter. However, in light of the model limitation described by Siegrist et al. (1999) and the fact that neither the A S M 2 d nor the T U D P model were tested under conditions in which high nitrate concentrations were observed in the anoxic zone, this value was deemed acceptable. 128 Results of the model predictions of the nitrate profiles in all zones are shown in Figure 4.22. The aerobic nitrate concentration was well predicted with a correlation coefficient o f 0.74. However, the model was not capable o f predicting the exact values o f the anoxic nitrate concentrations, especially the peak observed between September 3 and 12 o f 2003, and that resulted in a low value for the correlation coefficient for the anoxic nitrate. Nonetheless, the general trend, indicated elevated nitrate concentrations, was reasonably predicted. 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) Simulated_ANO Simulated_AER — A — Measured_ANO —•— Measured AER Figure 4.22. T U D P model prediction for the NO3 -N concentration profiles in the biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data In general, the simulation gave reasonable predictions of the nitrate concentration profiles in both the anoxic and aerobic zones. Since the model was capable of predicting the general trend o f the nitrate concentration profiles, it was believed to be sufficient for the purpose of this study with no further modification. However, i f more accurate and precise predictions are needed, then further investigation of the heterotrophic kinetics of the model may be required. 129 I l l Calibrating E B P R The calibration o f the E B P R process created a challenge, due to the observed failure in P-removal which was believed to be due to carbon limitation, an application for which the T U D P model has not been tested. Hence, this was a test of the model's capability to give reasonable predictions under these conditions. Fitting TP. The filtered TP concentration profiles in the biological zones were fitted using 0 . 0 0 5 for both the coefficient for phosphorus content of particulate substrate, ipxs, and the coefficient for phosphorus content of inert particulate C O D , ipxi-Fitting phosphate Release and Uptake. Examining the ortho-phosphate profile in the anaerobic zone of the M E B P R during the failure periods, Figure 3 .6 , it is clear that there was a decrease in the anaerobic ortho-phosphate concentration in the period between August 1 5 and September 8, reflecting a reduced anaerobic phosphorus release rate. Moreover, the observed decrease in the influent V F A concentration accompanying the failure in P-removal discussed earlier, in section 3 . 1 , may be the explanation. In addition, the V F A concentration in the anaerobic zone, shown later in Figure 4 . 2 5 , was essentially zero at some times during that period, which may indicate a reduced fermentation rate. Therefore, the parameters of the fermentation process were adjusted. The anaerobic hydrolysis reduction factor, r|fe, was decreased from the default value of 0 . 2 to 0.1 to reduce the hydrolysis rate and therefore, the amount of substrate available for fermentation. The fermentation rate, qfe, was also calibrated by decreasing it from 3 to 1. This would reduce the amount of fermented V F A available for the P-release process. The half-saturation coefficient for X P A o on acetate, K A P was decreased from 4 to 1 in order to increase the amount of V F A uptake by the P A O s to simulate the low V F A concentrations observed in the anaerobic zone. This resulted in a better prediction o f the anaerobic V F A concentration and in turn, the PO4-P and TP concentration profiles in all zones. This parameter was not part of the selected parameter set to be calibrated, but analysis of the model showed that the uptake of V F A might have been limited by the value of 4 for K A P , so it was decreased to better fit the data. 1 3 0 The predicted and simulated TP and PO4 -P concentration profiles are presented in Figures 4.23 and 4.24 respectively. The calibration clearly resulted in a much better fit o f the measured data, relative to the simulation with the T U D P model parameters determined during steady sate calibration stage, Figures 4.9 and 4.10. Figure 4.23 shows that the decrease in the concentrations of TP (mainly in the aerobic zone), experienced between the end of August and mid September, accompanied by the high concentration of ortho-phosphate in the effluent (Figure 4.24), were not predicted well by the simulation. However, the data for the second half o f the experimental period in Figures 4.23 and 4.24 were simulated well . This is mainly due to the fact that the elevated anoxic nitrate concentration profile was not predicted well for that period so the negative effect of the nitrate leaking to the anaerobic zone was not modeled. On the other hand, for the following period in which high concentrations of anoxic nitrate were simulated reasonably well , the effluent PO4 -P concentrations were predicted relatively well . This explanation is further substantiated by the prediction of the V F A concentrations shown in Figure 4.25, where for the period between September 5 and 15, when the process experienced the highest anoxic nitrate concentration, the simulation predicted a V F A concentration in the range of 1 - 2 g C O D / m 3 when the measured data showed zero concentration values. However, following that period, the N O 3 - N concentration was predicted relatively well and so were the PO4-P and the V F A concentration profiles. Figure 4.26 shows the concentration profiles of the effluent P 0 4 - P , anoxic N O 3 -N , aerobic TP and anaerobic V F A for the period between August 18 and September 28 to allow for a better comparison of the results. The calculated correlation coefficient values for the TP concentration profiles were 0.70, 0.79 and 0.67 for the anaerobic, anoxic and aerobic zones respectively, which indicate a reasonable fit. However, the correlation coefficient values for the PO4-P profiles were 0.66, 0.11 and 0.26 for the anaerobic, anoxic and aerobic zones respectively. The low anoxic and aerobic coefficient values were due to the fact that the exact measurements were not predicted well as described earlier. However, the general trend was satisfactorily 131 predicted in both cases. The correlation coefficient for the fitted V F A concentrations was found to be 0.45, which is reasonable considering the results described above. Table 4.4 lists the set of calibrated parameters for the manual calibration of the T U D P model that resulted in a reasonable prediction of the process behavior. A l l parameters were found to be within the reported ranges in literature except for r | L N 0 3 as discussed earlier. In general, despite the fact that the exact profiles for the bio-P process variables were not simulated at some points, the calibrated model predictions were considered acceptable and sufficient for the sake of this study, since the model was capable o f predicting an episode of transient behavior resulting in elevated effluent ortho-phosphate concentrations for the given process configuration, operating conditions and influent characterization. This model's capability in identifying this irregular or undesired behavior makes it a useful tool to be utilized in studying the effect of different process design and operating conditions on the process performance o f the M E B P R system. 200 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) • Simulated_ANA Simulated ANO — Simulated_AER —•— Filtered Measured ANA — F i l t e r e d Measured ANO —•— Filtered Measured AER | Figure 4.23. T U D P model prediction for the filtered measured TP concentration profiles in the biological zones of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 132 s a; 0 a. c -• i \ f. f * 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 • SimulatedANA - Measured ANA Simulated_ANO Measured ANO -SimulatedEFF Measured EFF Figure 4.24. T U D P model prediction for the P 0 4 - P concentration profiles in the anaerobic and anoxic zones and the effluent of the U B C M E B P R process using the T U D P model parameters determined during model calibration using dynamic data E o o u o •v < > 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oet-03 26-Oct-03 Figure 4.25. T U D P model prediction for the V F A concentration profile in the anaerobic zone of the U B C M E B P R process using the T U D P model parameters determined during the model calibration using dynamic data 133 Figure 4.26. T U D P model prediction results for the U B C M E B P R process for (a) effluent P 0 4 - P (b) anoxic NO3 -N (c) aerobic TP and (d) anaerobic V F A for the period between August 22 - September 26 4.1.5.2 Parameter Est imation in A Q U A S I M 2.0 In addition to the manual calibration of the T U D P model parameters, numerical parameter estimation techniques were also explored to compare the results of the manual calibration to the results obtained using automatic calibration. The results of both methods are discussed in this section. The simplex parameter estimation algorithm o f A Q U A S I M 2.0, explained in Reichert (1998), was used in the present study. The simplex algorithm was preferred over the 134 secant method, which is another parameter estimation method in A Q U A S I M 2.0, since it looks for a downward direction in a very robust way and it can also be applied to a poorly defined parameter estimation process with starting values o f the parameters far from those leading to the minimum. The A Q U A S I M function estimates model parameters by minimizing the sum of the squares of the weighted deviations between measurements and calculated model results using: x 2 ( p ) = E ymeasuj Yi (l*) o meas.i (4 .H) where y m e a s , i is the i-th measurement, amea$j is the i-th measurement's standard deviation, yi(P) is the calculated value of the model variable corresponding to the i-th measurement and evaluated at the time and location of this measurement, P = (pi, . . . , p m ) are the model parameters, and n is the number of data points. The standard deviations ameas,i c a n be defined individually for each data point or globally for all data points. In the present work, the standard deviation was defined globally for all data points of one variable (e.g. effluent C O D s o i ) and was estimated by taking the standard deviation of a pseudo-steady state portion o f each dataset. The parameter estimation function performs minimization of the sum of squares with the constraints: Pmi„,i ^ P i ^Pmax,i (4-12) where pmi n ,i and pm ax,i are the allowed minimum and maximum values for pi as specified by the user. The maximum and minimum values that were specified for each parameter, in the present work, were based on the reported values in literature as shown in Table 4.2. A s described earlier, sampling of the M E B P R was not carried out daily; therefore the concentration profiles were characterized by missing data. The missing data for all the concentration profiles were estimated as described in section 4.1.2. The filtered measured 135 C O D t o t , TP and T N profiles were used for the calibration process to smoothen the concentration profiles, thus avoiding biased parameter estimation. Each concentration profile required a standard deviation for use in computing the sum of the squares given by Equation 4 . 1 1 . Table 4 .3 lists the measured concentration profiles that were used for fitting the T U D P model predictions in the numerical parameter estimation and their estimated standard deviations (o~ m e a s i ) . Empty cells in the table indicate that the variable was not used in the parameter estimation. Fitting these 3 1 variables for the period between August 18 and October 2 1 of 2 0 0 3 resulted in a large set of 1 9 2 5 data points. The numerical parameter estimation run took over 2 days to converge, estimating 1 9 parameters using a Pentium 4 (2 .4 GHz) and 5 1 2 M B R A M computer. The initial T U D P parameter values used for the simulation were based on a set of manually calibrated parameters that resulted in a reasonable fit o f the data (prior to the parameter estimation run and not the final set of manually calibrated T U D P model parameters). The resulting set of estimated parameter values is given in Table 4 . 4 along with the initial values used and the default values as given by Meijer ( 2 0 0 4 ) . Table 4 . 5 lists the results for the sum of the errors for all fitted concentration profiles as determined by the numerical parameter estimation method of A Q U A S I M . It should be noted that the half-saturation coefficient for X P A O on acetate, K A P , was set to the default value of 4 and was not adjusted in the parameter estimation. The results of Table 4 . 5 show that most of the soluble concentration variables (e.g. NO3 -N and N H 4 - N ) have a higher weight in contributing to the total sum of errors when compared to the mixed liquor concentration data (e.g. T N and TP). This is due to the standard deviations of each measured variable, Table 4 . 3 , since the mixed liquor data carry larger measurement error while the soluble measurements exhibit less measurement error. Even though the mixed liquor data were filtered to smooth out large errors, they still exhibited larger measurement errors than those of the soluble components. 1 3 6 Table 4.3. List of variables and standard deviations used in parameter estimation algorithm in A Q U A S I M Influent Anaerobic Anoxic Aerobic Effluent TP (g P/m 3 ) 0.25 5.0 8.5 6.4 0.18 T N (gN/m 3 ) 1.3 2 6.2 5.4 1 TSS (g/m 3) 14 41 74 CODtot (g C O D / m 3 ) 29 135 239 C O D s o l ( g C O D / m 3 ) 1.8 1 1.9 N H 4 - N (g N /m 3 ) 0.55 0.32 0.007 N 0 3 - N ( g N / m 3 ) 0.001 0.11 0.91 0.22 PO4-P (g P/m 3) 0.25 0.12 0.14 0.12 VFA (g C O D / m 3 ) 0.5 Further examination of the results in Table 4.5 shows that many of the variables have a lower sum of errors using the estimated parameters over the initial guess, except for the nitrate concentrations in all zones. The estimated parameters resulted in a worse fit o f the experimental nitrate concentrations. The nitrate concentrations in the anoxic and aerobic zones reflect the behavior of the denitrification process, which is an important factor in the calibration process. Since the calibration algorithm moved in the direction o f minimizing the sum of the errors without differentiating between. data, it resulted in a better fit o f the variables at the expense of fitting the anoxic NO3 -N concentration profiles. B y examining the results for the estimated parameter values, given in Table 4.4, some useful information was obtained to better fit the data manually. For example, the estimated values of N-fractions of the activated sludge, iNxs and i N x i , were 0.031 and 0.041, which were different from the initial values and resulted in a much better fit o f the T N concentrations (Table 4.5). These values were then used in the manual calibration process to better fit the T N concentration profiles. The final manually calibrated values used were 0.04 and 0.035 respectively. Furthermore, the higher value for the maximum anaerobic acetate uptake rate, q s m a x , indicated that this rate was limited and needed to be increased to better fit the data. B y further examining the anaerobic uptake rate equation, it was found that the default value of 4 for the saturation coefficient for growth of P A O s pn acetate, K P A , was limiting the 137 acetate uptake in the anaerobic zone and so the low V F A concentrations in that zone were overestimated (results not shown). Once this value was reduced to 1, the lower V F A values were predicted better. In addition, the parameter estimation results confirmed the values of some parameters (such as k P H A and kp P), indicating that they did not require further calibration. These findings improved the calibration o f the model and produced a relatively good fit for all concentration profiles. The sum o f the error values for the model prediction using the manually calibrated parameters are given in Table 4.4. Even though the simplex parameter estimation method resulted in a better fit for some of the variables and a lower total %2 value, the manual calibration resulted in an overall better fit for most variables (as shown earlier) and a slightly higher value (only 8%) for j^. Results show that due to model complexity for activated sludge systems, calibration o f such models can be very tedious and tricky for the general user. Numerical parameter estimation results can provide valuable information about the model behavior which can help in better calibrating the model manually. However, it is recommended to calibrate these models manually with a good understanding o f the model structure. Therefore, a combination o f manual calibration and parameter estimation can go a long way in providing good information about the model behavior and resulting in a good set o f calibrated model parameter values. It was decided to use the set of the T U D P model parameters determined using manual calibration, which was also based on information obtained from the numerical parameter estimation method. The set of parameters used in the present study are listed under manual calibration parameters in Table 4.4. 138 Table 4.4. T U D P model parameters estimated using the simplex parameter estimation method and manual calibration for fitting the measured data collected from the U B C M E B P R process Parameter Units Meijer (2004) Initial Values for Parameter Estimation Parameter Estimation for Dynamic Calibration Manual Calibration Parameters rife 0.2 0.1 0 . 0 3 1 0.1 r| N03 0.8 0.65 0 . 6 5 7 0 . 6 5 r| N03 0.8 0.3 0 . 3 0 3 0 .2 r| N03 0.8 0.8 0 . 7 3 1 0.8 f x S i n g COD/ g COD 0 .439 0.8 0 . 8 1 1 0 . 8 5 g P P 0.22 0 .22 0 . 2 8 1 0 . 2 2 INXI g N/g COD 0.03 0 .04 0 . 0 4 1 3 0 . 0 3 5 INXS g N/g COD 0.03 0.065 0 . 0 3 1 1 0 . 0 4 ipxi g P/g COD 0.01 0 .0045 0 . 0 0 5 5 0 . 0 0 5 ipxs g P/g COD 0.01 0 .005 0 . 0 0 4 3 0 . 0 0 5 k G L Y T g COD/(g COD • d ) 0.93 0.93 1 .064 0 . 9 3 KNH gN/m 3 1 0.1 0 . 1 2 2 0.1 K<)2 g0 2/m 3 0.2 0.7 0 . 7 3 7 0 .2 k p H A g COD/(g COD • d) 5.51 5.51 5 . 5 1 5 5 .51 k T Kpp g P/(g COD • d) 0.1 0.1 0 . 1 0 2 0.1 T Qfe g COD/(g COD • d) 3 1 0 . 7 2 5 1 „ max T q s g COD/(g COD • d ) 8 8 9 . 6 6 5 8 Y H g COD/g COD 0.63 0.63 0 . 6 2 7 0 .63 Yp04 g COD/g COD 0.35 0.35 0 . 3 5 1 0 . 3 5 K P A * g C O D S A / m 3 4 4 4 1 Resulting %2 573061 481603 519600 A l l parameters are reported at 20 °C. T These parameters are temperature dependent. *Not used in parameter estimation Moreover, to further verify that the model captured the process dynamics correctly; residuals analysis of model predictions was performed. The results o f the analysis are discussed in the following section. 4.1.6 Residual Analysis Residual analysis can be a useful tool in assessing model prediction capabilities. Residuals, (measured variable - predicted variable) simply represent the plant-model mismatch. If the model fully captures process dynamics, then the residuals would basically consist of measurement noise and experimental error (Ljung, 1999). 139 Model residuals can be cross-correlated with model inputs to ensure that the model captures all process dynamics. In cases where the model fully captures the process dynamics, the residuals would have no correlations with the inputs. However since no model is perfect, a correlation usually exists between model residuals and input variables. Using these results, areas of model deficiency can be identified. The cross-covariance function given by Equation 4.4 can be used for this purpose. In the present study, it was decided to use residual analysis to evaluate the fit o f the calibrated T U D P model to the M E B P R process behavior. A cross-correlation between the model residuals (measured concentrations - model predicted concentrations) and the influent concentration profile was used as a measure of the model fit. If the cross-correlation function gave a value outside the 95%-confidence lines, that meant that some dynamics in the influent stream for that component were not captured by the model, and this would represent a deficiency in the model. However, i f all values fell within the 95%-confidence lines, then the model was considered sufficient. The cross-correlation function o f the residuals for effluent C O D S O | and the influent C O D S O | profile was computed after removing the trend in both signals (by removing the best straight-line fit linear trend from the data). The correlation function was plotted and examined to ensure a good fit of the data. Figure 4.27 presents the plot of the cross-correlation function. A l l the values for the function are within the 95% confidence interval lines (shown above and below the curve) indicating that.all process dynamics were adequately captured by the calibrated T U D P model. The same results were also obtained for the cross-correlation function for the residuals of the effluent C O D s o i and the influent C O D t o t as shown in Figure 4.28 140 Table 4.5. The sum of squares results for the model fit of the dynamic measured data of the M E B P R process for different calibration methods Concentration Profile Biological Reactor Initial Parameter Values for Simplex Method Estimated Parameters using Simplex Method Estimated Parameter using Manual Calibration COD s o , A N A 42945 42118 43011 C O D s o l A N O 91580 92015 91337 COD s o , EFF 6084 6084 6082 C O D t o t A N A 20355 20449 21115 C O D t o t A N O 995 972 987 C O D t o t A E R 116 102 82 N H 4 - N A N A 9278 6328 7390 N H 4 - N A N O 79922 47357 58516 N H 4 - N EFF 61074 34187 42560 N03-N ANA 76123 76617 76306 N03-N ANO 8883 9978 9140 NOrN AER 1556 1867 1596 N03-N EFF 27664 30015 26910 T N INF 8708 3617 4580 T N A N A 10974 8241 5736 T N A N O 1935 1600 1341 T N A E R 5767 5097 4415 T N EFF 9029 4400 5629 PO4-N A N A 8831 6906 9348 PO4-N A N O 53755 37915 56988 P 0 4 - N A E R 2771 2559 2836 P 0 4 - N EFF 4392 4106 4509 TP INF 1049 1035 1055 TP A N A 507 492 519 TP A N O 298 268 303 TP A E R 1631 1541 1620 TP EFF 3821 3933 3848 TSS A N A 27681 27118 26003 TSS A N O 1883 1793 1870 TSS A E R 799 831 1075 V F A A N A 2657 2063 2891 Resulting % 573061 481603 519600 141 1 0.8 0.6 o 1 0.2 a o To 0 CD 3 -0.2 to CO 5 -0.4 -0.6 -0.8 -1 -2 i 1 1 i 3 -15 -10 -5 0 5 10 15 20 Figure 4.27. Cross-correlation function of the effluent C O D S O i residuals for the T U D P model predictions and the measured influent C O D s o i concentrations of the M E B P R process 1 0.8 0.6 | 0.4 1 0.2 c o as 0 i i "ai 3 -0.2 cn CA B -04 -0.6 -0.8 -1 -2 3 5 -II ) -£ C 1 £ 1 3 1' 5 2( Figure 4.28. Cross-correlation function of the effluent C O D s o i residuals for the T U D P model predictions and the measured influent C O D t o t concentrations of the M E B P R process 142 The cross-correlation function of the residuals o f the fitted effluent N H 4 - N data and the influent N H 4 - N profile was examined as well (Figure 4.29). A peak was found at lag zero. This peak is believed to be due to the influent sampling protocol used. When the plant was sampled, a sample was taken first from the influent stream, followed by the anaerobic, anoxic and aerobic zones and the effluent stream. The time difference between the influent and effluent sampling was about 1 Vi - 2 hours, which is less than the process H R T of 10 hours. Therefore, the influent sample contained some dynamics that are part of the following effluent sample and since the N H 4 - N dynamics were mainly due to the influent load variations instead of the nitrification kinetics since the aerobic zone was over-designed, a correlation existed between the effluent N H 4 - N residuals and the influent N H 4 - N concentrations. The cross-correlation functions for the fitted T N , effluent PO4-P, anaerobic PO4-P, anoxic PO4-P and effluent TP concentrations are presented in Figures 4.30 to 4.34 respectively. Some functions showed low correlation values (around 0.4), but these values were considered insignificant since they do not reflect strong correlation. The effluent PO4-P cross-correlation function showed a correlation value of about -0.6 but it was identified at lag -20, which is also considered insignificant. The results of the residual analysis indicate that the model structure was correct in capturing the process dynamics. 143 Figure 4.29 Cross-correla t ion function o f the effluent N H 4 - N residuals for the T U D P mode l predictions and the measured influent N H 4 - N concentrations for the M E B P R process 1 0 . 8 0 . 6 .1 o I 0 -2 E o oS 0 01 <S - 0 . 2 CO CO S "0-4 - 0 . 6 - 0 . 8 -1 -2( i i i i s/ i ] - 1 5 - 1 0 - 5 0 5 1 0 1 5 2 0 Figure 4.30. Cross-corre la t ion function o f the effluent T N residuals for the T U D P mode l predictions and the measured influent T N concentrations for the M E B P R process 144 Figure 4.31. Cross-correlation function o f the effluent PO4 -P residuals for the T U D P model predictions and the measured influent P0 4-P concentrations for the M E B P R process t 0.8 0.6 .2 0 4 0 1 0.2 c 0 ts 0 5 -0.2 C O to 8 -0.4 -0.6 -0.8 -1 -2 ] -15 -10 -5 0 5 10 15 20 Figure 4.32. Cross-correlation function o f the anaerobic PO4 -P residuals for the T U D P model predictions and the measured influent PO4 -P concentrations for the M E B P R process 145 Figure 4.33. Cross-correlation function of the anoxic P 0 4 - P residuals for the T U D P model predictions and the measured influent PO4-P concentrations for the M E B P R process 1 0 . 8 0 . 6 .1 0 4 o c o O to CO p 0 . 2 0 - 0 . 2 H - 0 . 4 o - 0 . 6 - 0 . 8 h -1 I I I 1 1 1 1 - 2 0 - 1 5 - 1 0 - 5 1 0 1 5 2 0 Figure 4.34. Cross-correlation function of the effluent TP residuals for the T U D P model predictions and the measured influent TP concentrations for the M E B P R process 146 4.1.7 Model Validation Upon calibration of the T U D P model parameters, it was important to test the predictive power of the calibrated model using a data set that was different from that used in model calibration. This step was performed as a final check of the modeling exercise. It was decided to validate the model by predicting the process behavior for data collected during an intensive sampling campaign carried out between M a y 2 and M a y 15 of 2003. During this period, samples were collected twice a day from the M E B P R process, since the raw sewage tanks were filled twice a day around 9 am and 9 pm. The two samples should have captured the dynamic variations in the influent stream resulting from each tank filling. Process samples were collected daily around 7 am and 7 pm (just before the next fi l l ing o f the tank) to ensure that the process had reached pseudo-steady state conditions. Influent samples however, were collected four times a day to capture the concentration variation in the raw sewage storage tank over each filling cycle. The same process configuration as that described in section 2.7.1 was used. The calibrated model parameters as determined by the calibration process described above were used to predict the process behavior. A l l parameters used in the model validation were the same as those estimated by the manual calibration, except for the half-saturation coefficient for X P O A on acetate, K A P , which was kept at the default value of 4 since the influent V F A concentration was higher during M a y of 2003, than that experienced in July - October. Parameters used i i i model validation are listed in Table 4.6. The data used as the influent concentration in the simulation model were based on the influent samples collected from the M E B P R process 4 times per day, but the measured data for the biological zones and the effluent were based on samples collected from the process twice a day. Using the dynamic influent concentrations sampled 4 times a day in the simulation model, resulted in more dynamic predicted profiles for the biological zones and effluent concentrations than the measured concentration profiles. The measured and simulated anaerobic and aerobic C O D s o i concentration profiles are presented in Figure 4.35. A s expected, the C O D s o ] concentration trends were well 147 predicted with a correlation coefficient o f 0.74. The C O D t o t measured concentrations exhibited dynamics that were not predicted by the model as shown in Figure 4.36. The fit resulted in low correlation coefficient values between the measured and predicted concentration profiles. This could be due to the grab sampling o f the mixed liquor and since sludge wasting, of about 150 L/day, was performed manually once per day during that period, it is believed that the sludge wasting introduced dynamics in the process which affected the suspended solids concentrations in the mixed liquor. On the other hand, the wasting of the sludge in the simulation was modeled as a stream withdrawn from the process continuously over the day, resulting in smoother predicted concentration profiles. Considering this fact, the prediction of the C O D t o t profiles was deemed acceptable. Table 4.6. T U D P model parameters used for model validation for the M E B P R process Parameter Units Meijer Manual Calibration Validation MEBPR (2004) Parameters (May, 03 Data) 0.2 0.1 0.1 TT, N03 0.8 0.65 0.65 1<\ N03 0.8 0.2 0.2 TI N03 0.8 0.8 0.8 fxSin g C O D / g C O D 0.439 0.85 0.85 gPP 0.22 0.22 0.22 iNXI g N / g C O D 0.03 0.035 0.035 iNXS g N / g C O D 0.03 0.04 0.04 ipxi g P / g C O D 0.01 0.005 0.005 ipxs g P/g C O D 0.01 0.005 0.005 k G LY T g COD/ (g C O D • d ) 0.93 0.93 0.93 K N H g N / m 3 1 0.1 0.1 K02 g 0 2 / m 3 0.2 0.2 0.2 kpHA k T K P P T g COD/(g C O D • d) 5.51 5.51 5.51 gP/ (g C O D - d ) 0.1 0.1 0.1 qfe g COD/(g C O D • d) 3 1 1 max T qs g COD/ (g C O D • d ) 8 8 8 . Y H g C O D / g C O D 0.63 0.63 0.63 Yp04 K P A g C O D / g C O D 0.35 0.35 0.35 g C O D S A / m 3 4 1 4 148 160 140 120 = e c u 3 loo 80 60 40 20 I / \ j \t 1 _____ _ _ _ _ _ _ / ' % r- 11 -A' *G__P^ P^ —r~ A v 7 0 1-May 3-May 5-May 7-May 9-May (Date) 11-May 13-May 15-May 17-May • Simulated ANA —•— Measured ANA • Simulated A E R —•— Measured AER Figure 4.35. Model prediction of the C O D s o , concentration profiles in the anaerobic and aerobic zones for model validation o f the calibrated T U D P model using data collected from the M E B P R process in May o f 2003 • Simulated_ANA -•—Measured ANA Simulated_ANO Measured ANO Figure 4.36. Model prediction o f the C O D t o t concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 149 A l l nitrogen components were reasonably well simulated by the model. The total nitrogen concentrations in the mixed liquor were slightly under-predicted, but the influent profile (calculated by the model using Equations 2.11 and 2.12) was relatively well predicted with a correlation coefficient of 0.67, Figure 4.37. A s for the nitrate concentrations, shown in Figure 4.38, the aerobic nitrate profile was reasonably predicted as well as the anoxic nitrate profile including the high concentrations experienced by the process at some times. The correlation coefficient values for nitrate concentrations were 0.54 for the anoxic zone and 0.8 for the aerobic zone. The ammonium profiles were also well predicted as shown in Figure 4.39. The correlation coefficient values were found to be 0.83 for an anaerobic zone and 0.86 for the anoxic zone. The aerobic zone concentrations were essentially zero for both the measured and predicted profiles. The influent concentration profile was also plotted to show the dynamics of the influent as experienced by the model, while the measured data were sampled less frequently. It can be noticed that the simulated anaerobic zone profile nicely follows the measured influent concentration profile. 300 l-May 3-May 5-May 7-May 9-May 11-May 13-May 15-May 17-May (Date) —•—MeasuredINF —•— Measured_ANA —A—Measured_ANO —•— Measured AER Simulated INF • Simulated ANA Simulated_ANO — — Simulated_AER Figure 4 .37. Model prediction of the T N concentration profiles in the influent and biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 150 1-May 3-May 5-May 7-May 9-May 11-May 13-May 15-May 17-May (Date) -A—Measured_ANO Simulated_ANO —•—Measured AER •Simulated AER Figure 4.38. Model prediction of the N 0 3 - N concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 1-May 3-May 5-May 7-May 9-May 11-May 13-May 15-May 17-May (Date) — • Simulated_ANA —•— Measured_ANA Measured INF Simulated_ANO -A—Measured ANO •SimuIatedAER Measured AER Figure 4.39. Model prediction of the N H 4 - N concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 151 The phosphorus removal process was also predicted relatively well . The TP concentration profiles are plotted in Figure 4.40. The experimental data were simulated reasonably within the measurement error for the first half of the sampling campaign; however some dynamics were experienced by the process in the second half o f the period that were not predicted by the simulation. Interestingly though, these dynamics were not observed in the PO4-P and V F A profiles, Figures 4.41 and 4.42 respectively, as these were relatively consistent with the preceding period. Considering that, the TP predictions were deemed acceptable. Furthermore, the PO4-P and V F A profiles were found to be acceptable as well . The dynamics observed in the simulated data were the result of the dynamics encountered in the influent concentration profiles which fluctuated during the day especially for the influent V F A , since it depended on the residence time in the raw sewage storage tanks. These findings resulted in low correlation coefficient values for the phosphorus concentrations, however, the general trend for the concentration profiles were reasonably predicted. 0 4- • •^•^••^•^••• • • • • • • • • • • • •^•• • • • • • • • • • • • • • • • • • • • • •^ l-May 3-May 5-May 7-May 9-May 11-May 13-May 15-May 17-May (Date) Simulated_INF Measured INF • SimuIated_ANA Measured ANA - SimulatedANO - A — Measured ANO •SimuIated_AER - Measured AER Figure 4.40. Model prediction of the TP concentration profiles in the influent and all biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 152 16 1-May 3-May 5-May 7-May 9-May 11-May 13-May 15-May 17-May (Date) •—- • Simulated_ANA - " Simulated_ANO Simulated AER —•—Measured ANA —A—Measured ANO —•—Measured AER Figure 4.41. Model prediction of the P 0 4 - P concentration profiles in the biological zones for model validation of the calibrated T U D P model using data collected from the M E B P R process in M a y of 2003 16 1-May 3-May 5-May 7-May 9-May 11-May 13-May 15-May 17-May (Date) — • SimulatedANA • Measured_ANA Simulated_ANO — * r — Measured_ANO Figure 4.42. Model prediction of the V F A concentration profiles in the anaerobic and anoxic zones for model validation o f the calibrated T U D P model using data collected from the M E B P R process in M a y o f 2003 153 Model validation results showed that the calibrated dynamic model can reliably predict the general trend in the process behavior even at different operating conditions, considering that the data used for model validation were collected at an average process temperature of 17.5 °C, while the data used for the model calibration were collected during the summer season, when the process temperature was in range of 19 - 23 °C. Furthermore, the same calibrated parameters of the T U D P model were used for model validation except for the half-saturation coefficient of X P A O on acetate, K P A . Considering these findings, it is believed that this model can be a useful tool in studying the M E B P R process performance under different design and operating conditions. The model validation was the last step in the dynamic modeling calibration process. The effectiveness of the utilized dynamic modeling technique in the present study is discussed in the next section. 4.2 Effectiveness of the Utilized Dynamic Modeling Protocol The utilized dynamic calibration protocol is based on the S T O W A protocol (Hulsbeek et al., 2002) which was considered the most practical and easy to implement (Sin et al., 2005). However, the present protocol addressed the different areas of deficiency identified by Sin et al. (2005) in their analysis of the different calibration protocols (Chapter 1). 4.2.1 Steady State versus Dynamic Modeling The distinction between steady state and dynamic data is not made clear in most of the dynamic modeling protocols. The utilized protocol draws a clear distinction between steady state modeling using averaged data collected from a plant running at pseudo-steady state and dynamic modeling using data that exhibit signal excitation to the main process dynamics to be modeled and collected at a proper sampling frequency. When the protocol was applied to data collected from the U B C pilot plant, it was shown that the results of the model calibration using averaged data can be misleading when the same parameters are applied in modeling data collected under dynamic process behavior. 154 This result is believed to be valuable in clearly identifying these differences since a number of model calibration studies using the T U D P model have been performed using averaged full-scale plant data and referred to as dynamic modeling, including the work of Meijer et al. (2001) and Meijer et al. (2002a) among others. In most of these studies samples were collected over a short period of time, for example a 2-day period, resulting in data that was excited at the high frequency range, due to daily flow and load variations, and not the low frequency range required for dynamic modeling of activated sludge systems. 4.2.2 Steady State Modeling The S T O W A protocol did not acknowledge steady state modeling in the main structure o f the model calibration, while in the present protocol it is specified as an essential step to obtain a feeling for the model fit to process data. Performing the steady state calibration also provides an initial set o f parameters for the dynamic modeling. Furthermore, performing steady state modeling prior to dynamic modeling provides the initial biomass concentrations required for the dynamic simulation. Results o f the steady state modeling in this study gave insightful information about the model structure and the process configuration set-up in A Q U A S I M . It would have been very difficult to check for these using the dynamic data, since these exhibited large variations in concentrations over the course of the experiment. The steady state modeling also provided a good initial estimate for model parameter values for the half-saturation coefficient for N H 4 - N , K N H , and the biodegradable fraction of the influent C O D using f xS in - , 4.2.3 Experimental Design and Dynamic Data Calibration protocols available in the literature rarely address the issue o f experimental design. In most studies reported in the literature, experimental design was usually based on plant experience and data availability rather than system identification theory, except for some methods such as the optimal experimental design (OED) technique used in the B I O M A T H protocol by Vanrolleghem et al. (2003). 155 The S T O W A protocol puts little emphasis on mathematical or statistical methods to be used for better design of sampling campaigns. The protocol used in the present work addressed this deficiency by suggesting the use of the model to obtain information about the process dynamics and the proper sampling frequency. Furthermore, after collecting the data, the frequency content of the collected data was checked to ensure that the data covered a wide range o f frequencies reflecting the process dynamics. The calibrated T U D P model using steady state data was used to study the process dynamics to decide on a proper process sampling frequency. Prior to that, an intensive sampling campaign was carried out which resulted in a large number of samples. However, after examining the model dynamics, less frequent sampling was found sufficient to obtain excited process data used for model calibration. Furthermore, the resulting model, which was calibrated using less frequently sampled data, was found to reasonably predict the set of data collected during the intensive sampling campaign, which was collected at a higher sampling frequency. This indicated that the sampling frequency used for the data collected for model calibration was sufficient to model the process behavior satisfactorily and to predict the concentration profiles of data collected at a higher frequency. Equally important was the sampling duration. Often in the literature, data used for model calibration were collected from the process using a sampling frequency o f two hours and a sampling duration of 2 days when the process SRT was between 10 - 20 or more days. The data collected this way could be sufficient for steady state calibration, but not for dynamic calibration since the data set would not be excited to reflect the process dynamics to be modeled. The data used in the present study for the dynamic calibration of the T U D P model for the M E B P R process were collected over a period of about 103 days, during which the process experienced various dynamic changes such as the change in the filling times of the storage tank which resulted in a drop in the influent V F A concentration. A 156 examination of the power spectra of the collected data showed some excitation of the data over a reasonable range of frequencies. These results emphasize the importance o f obtaining properly excited dynamic data for the purpose o f calibrating a dynamic model. Data should be collected using a suitable sampling frequency and for a sufficient period of time to capture process dynamics. Using dynamic data in model calibration would ensure that the calibrated model can properly predict the process behavior under different dynamic conditions. 4.2.4 In f luen t C h a r a c t e r i z a t i o n Another deficiency of the S T O W A calibration protocol was that it relied on the B O D method for characterization of the influent inert particulate fraction. In the current study, the method described by Meijer et al. (2001) of introducing the calibration parameter, fxsin. for the inert particulate fraction, was used. Using fxsin as a calibration parameter to estimate the inert particulate fraction eliminated the error associated with the B O D measurement, which was one of the concerns raised by Sin et al. (2005) about the S T O W A protocol. Furthermore, in the present study, the influent and the mixed liquor T N and TP concentration profiles were used to calibrate the factors for the N fractions, INXS> INXI, and the P fractions, ipxs, ipxi of the particulate C O D components. The factors for the N and P fractions of the soluble C O D components were kept at their default values. This method was believed to be more accurate than the one suggested by the S T O W A protocol (Roeleveld and van Loosdrecht, 2002) in which the T K N concentration measurements of the total sample and the soluble fraction of the influent and effluent (which usually contain high measurement errors) were used to estimate the factors for the N and P fractions of the C O D components according to Equations 4 . 13 -4 . 17 . These factors were usually in the range of 0.005 to 0.06 and therefore, small measurement errors in the concentration measurements used for estimating these factors, could introduce significant errors in their estimated values. Moreover, incorrect estimation o f these factors might have affected the model predictions since sensitivity analysis results presented in Figures 4 . 1 2 - 4 . 1 6 showed high sensitivity o f the predicted model concentrations to changes in 157 these factors. Therefore the method used in the present work can result in a better model calibration and prediction of the measured concentrations. SNH4 - N H 4 - N r N F (4.13) T K N t o t , r N F - SNH4 + (SF • INSF) + (Si • INSI) + ( X i • iNxi) + (Xs • iNxs) (4.14) T K N s o i ; INF - S N H4 + (SF • INSF) + (Si • iNsi) (4.15) T K N s o i > E F F = Si • i N s i + N H 4 - N E F F (4.16) T K N part ,INF ~ T K N tot.INF - T K N S 0 1 ,INF - (Xi • iNXl) + (Xs • iNxs) (4.17) 4.2.5 Initial Biomass characterization Another deficiency encountered in the S T O W A protocol, as indicated by Sin et al. (2005), was the determination of the initial autotrophic and heterotrophic biomass concentrations in a W W T P for a dynamic calibration. The S T O W A protocol did not specify a method for determining the initial biomass concentrations for a set of dynamic data, even though these concentrations had a significant effect on the model prediction. To address this deficiency in the present work, the averaged influent data used for the steady state simulation was used for a simulation period of 100 days prior to the dynamic modeling period to generate the initial biomass concentrations for the dynamic modeling part. This method was found to give a good estimate of the initial biomass concentration that resulted in a reasonable fit o f the dynamic data. 4.2.6 Model Calibration The calibration of the dynamic model was the most challenging part of the modeling exercise. Selecting the sub-set of parameters to calibrate was one part of the challenge and the other part was determining the values for the parameters that would result in a good fit o f the measured data. In addressing the first challenge, dynamic sensitivity analysis results were used in addition to model knowledge and literature review results for selecting the sub-set of parameters. The dynamic sensitivity analysis provided 158 valuable information about the model behavior which helped in selecting the sub-set o f the T U D P model parameters used for the model calibration. Performing sensitivity analysis using dynamic data is far more informative than using averaged steady state data, since it provides information about the change in model outputs in response to a change in model parameters for the entire data series. Therefore, the user can focus on the results of the sensitivity analysis for the period when the model predictions do not match the measured data, to better understand the effect of different parameters on the model outputs. The results of the dynamic sensitivity analysis were combined with the findings of a comprehensive review of the parameters used for model calibration o f the A S M 2 d and T U D P models reported in the literature to identify the acceptable ranges for the parameters used for the calibration. Furthermore, it was found that experience and knowledge of the model are essential to better understand the effect o f each parameter on the model outputs. This was experienced while fitting the anoxic nitrate concentration, as it was found that the anoxic hydrolysis reduction factor, which is not typically used for calibration, had to be adjusted to predict the high nitrate concentration in the anoxic zone. The combination of these methods was found to be effective in selecting the sub-set of parameters to use for model calibration. After deciding on the sub-set o f parameters to use for model calibration, the second challenge in model calibration was parameter estimation. In the current work, batch tests were not used for estimating some kinetic parameters to avoid the known limitation of transferring lab-scale results to full-scale models. The limitation lies in the difficulty in estimating the correct initial biomass concentration (Vanrolleghem et al., 2004) in the batch test in addition to possible differences in the floe size in both systems which was shown to affect the mass transfer rates (Gapes et al., 2004). Moreover, differences in reactor sizes, configuration and hydrodynamics may also affect the results (Sin et al, 2005). It is well known that the mass transfer rates observed in well mixed batch tests are different than those in full-scale or pilot-scale plants and therefore, may also affect the measured values of the kinetic parameters. 159 Using the simplex parameter estimation method of A Q U A S I M in the current study provided a better estimate of some model parameters than the ones obtained in the initial manual calibration exercise, which helped in better calibrating the model parameters manually (e.g. i N x i and iNxs)- However, since mathematical techniques aim at reducing the sum of the errors, this can result in a better fit o f some variables at the expense of others and result in a biased parameter estimation. Nonetheless, combining the findings of both techniques resulted in a set of parameter values that was within the ranges reported in the literature for all the parameters, except for anoxic hydrolysis reduction factor (t]L N03), and gave a reasonable fit o f the data. This showed that relying solely on manual calibration can be misleading, especially in the case of calibrating complex models with many interacting parameters. 4.2.7 Summary The protocol utilized in the present work was found to provide an efficient dynamic calibration technique for modeling the behavior of activated sludge systems. The protocol was based on the S T O W A calibration protocol but addressed most of its deficiencies as described by Sin et al. (2005) and had the following advantages. 1. Cost effective calibration technique avoiding extensive sampling and batch tests. 2. Provided extensive data analysis to ensure obtaining a dynamic set o f data suitable for the dynamic model calibration. 3. Easy to follow steps and practical to implement for modeling the most complex system. 4. Combined wastewater treatment expert knowledge with system identification theory to utilize the better o f the two worlds in data analysis and processing. 5. Relied on mathematical methods and process knowledge in selecting and estimating the values for the calibration parameters. 6. Tested using the complex T U D P model for modeling data collected from an M E B P R process under challenging operating conditions imposed by the carbon-limited influent resulting in process failure in P-removal. 160 4.3 Comparison of Dynamic Modeling Results of MEBPR to CEBPR Process Samples were collected from both the M E B P R and the C E B P R during the course of the experiment as described earlier. The data collected from the C E B P R process for the same period, July 1 0 - October 21 o f 2 0 0 3 , used for calibrating the M E B P R model, were used to model the C E B P R using the T U D P model to compare the model predictions for both systems. The C E B P R process was modeled in A Q U A S I M 2 . 0 as described in section 2 . 7 . 2 . The T U D P model and the parameter values, as determined by the calibration process for the M E B P R system, were used to model the C E B P R process to compare both fits. The same T U D P model parameter values used for the M E B P R process were found to give a reasonable fit o f all C E B P R measured concentrations except the anoxic PO4 -P concentration profile which was over-estimated (results not shown). In order to better predict this profile, the polyphosphate formation rate, kpp, was increased from 0.1 to 0 .2 g P/(g COD-d). This resulted in a better fit o f the anoxic P 0 4 - P data. Table 4 . 7 lists the parameters used for the modeling the C E B P R process along with those used in the calibration and validation of the M E B P R process. It should be noted that the half-P 3 saturation coefficient o f XPAO on acetate, K A , was lowered from 4 to 1 g C O D S A / m in both the M E B P R and the C E B P R to predict the low V F A concentrations observed in the anaerobic zone, since both systems experienced the reduced influent V F A concentrations during that period. Simulation results are presented below. The effluent C O D s o i concentration profile was fitted reasonably well , as shown in Figure 4 . 4 3 , with a correlation coefficient of 0 . 5 5 . The predicted effluent C O D s o i concentration of the M E B P R process, Figure 4 . 1 9 , shows a relatively better fit o f the concentration profile, which explains the higher correlation coefficient of 0 . 9 3 obtained for the M E B P R data. This difference is believed to be due to the nature of the conventional process and the sampling approach used. For the M E B P R process, the effluent sample was taken from the permeate tank and preserved. However for the conventional side, the sample was taken from the effluent tank after the secondary clarifier, and was then filtered using the ZeeWeed membrane filter and preserved. These extra steps of filtering may have 161 produced some measurement errors resulting in fluctuations in the readings. Taking that into consideration, both fits were considered to be satisfactory. Table 4.7 T U D P model parameters used in the modeling of the M E B P R and C E B P R processes Parameter Units Meijer Calibration Validation CEBPR (2004) MEBPR MEBPR rife 0 .2 0.1 0.1 0.1 r i N03 0.8 0 . 6 5 0 . 6 5 0 . 6 5 r) N03 0.8 0 .2 0 .2 0 .2 Tl N03 0.8 0.8 0.8 0.8 fxSin g C O D / g C O D 0 . 4 3 9 0 . 8 5 0 . 8 5 0 . 8 5 gPP 0 . 2 2 0 . 2 2 0 . 2 2 0 . 2 2 INXI g N / g C O D 0 . 0 3 0 . 0 3 5 0 . 0 3 5 0 . 0 3 5 INXS g N / g C O D 0 . 0 3 0 . 0 4 0 . 0 4 0 . 0 4 ipxi g P/g C O D 0 .01 0 . 0 0 5 0 . 0 0 5 0 . 0 0 5 ipxs g P/g C O D 0.01 0 . 0 0 5 0 . 0 0 5 0 . 0 0 5 k G L Y T g COD/(g C O D • d ) 0 . 9 3 0 . 9 3 0 . 9 3 0 . 9 3 KNH g N / m 3 1 0.1 0.1 0.1 K<D2 g 0 2 / m 3 0.2 0 .2 0 .2 0 .2 kpHA^ g COD/(g C O D • d) 5.51 5 .51 5 .51 5 .51 k p p T gP/ (g C O D - d ) 0.1 0.1 0.1 0.2 qfe „ max T q s g COD/(g C O D • d) 3 1 1 1 g COD/(g C O D • d ) 8 8 8 8 Y H g C O D / g C O D 0 . 6 3 0 .63 0 .63 0 .63 Ypo4 g C O D / g C O D 0 . 3 5 0 . 3 5 0 . 3 5 0 . 3 5 K P A g C O D S A / m 3 4 1 4 1 The CODtot profiles were also simulated reasonably well using the calibrated parameters for the C E B P R process. The correlation coefficient for the anaerobic measured and predicted concentrations was 0 . 5 . However, due to the large fluctuations in the anoxic and aerobic measured C O D t o t data, resulting from the dynamic nature of the sludge in the conventional system, low correlation coefficient values were obtained for both concentration profiles. In general, the fit was deemed acceptable. Results of the simulated and measured data are shown in Figure 4 . 4 4 . 1 6 2 250 ^200 J Q o u 55150 B e •a = 100 u s a o u 50 0 MeasuredEFF •Simulated EFF 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 Figure 4.43 T U D P model prediction of the C O D s o | concentration profile o f the effluent o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 6000 5000 — o o 4000 3000 2000 — o u 1000 7-Aug-03 17-Aug-03 27-Aug-03 6-Sep-03 16-Sep-03 (Date) 26-Sep-03 6-Oct-03 16-Oct-03 — • Simulated_ANA • Measured ANA Simulated_ANO Measured ANO "Simulated_AER Measured AER Figure 4.44. T U D P model prediction of the C O D t o t concentration profiles in the biological zones of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 163 The T N concentration profiles in all zones of the C E B P R , Figure 4.45, exhibited the same patterns as those observed in the M E B P R , Figure 4.20, except for the sharp increase in the aerobic T N concentration experienced after October 6, which was not observed on the M E B P R side. It is clear that both systems experienced an increase in the T N content of the aerobic mixed liquor during the period between September 3 and the 29, which was not correctly predicted by the simulation in both cases. This indicates that this disturbance in the concentration was due to variations in influent concentrations that were not captured by the model. This resulted in a poor correlation coefficient for the T N concentrations in the aerobic zone. However, in general, the predictions in both systems were satisfactory using the same model parameter values. The calculated correlation coefficient values of the measured and fitted T N concentrations for the anaerobic and anoxic zones were 0.46 and 0.62 respectively. These values were considered reasonable especially that the measured data exhibited drops in the concentrations due to accidents encountered during process operation, such as the one experienced on August 20, when the underflow for the secondary clarifier stopped working so solids accumulated in the clarifier and were not returned to the aerobic zone, resulting in a drop in the bioreactor concentrations. These events were not reflected by the model. The N H 4 - N concentration profiles, presented in Figure 4.46 for the C E B P R process show a good fit o f the measured data. The calculated correlation coefficients for the anaerobic and anoxic zones were 0.86 and 0.93 respectively, which reflect a high correlation between the measured and predicted data. Also the aerobic measured and predicted concentrations were essentially the same. The same behavior was observed for the M E B P R process, Figure 4.21, as both pilot systems showed complete nitrification. The same result of complete nitrification was also achieved in other studies involving M E B P R systems through the monitoring of the effluent (Fleischer et al., 2005; Lesjean et al., 2003; Adam et al., 2002). 164 On the other hand, results of the batch tests performed by Mont i et al. (2006b) using sludge samples collected from the U B C M E B P R and C E B P R pilot plants showed that the C E B P R process exhibited 15 to 75% greater nitrification rates at different times. Monti et al. (2006b) modeled the results of the batch test for N H 4 - N and NO3 -N data using the T U D P model and found that the maximum specific growth rate of autotrophs, PAUT, in the M E B P R process, was 30% lower than that of the conventional counterpart. However, research carried out by Manser et al. (2004) for comparing the maximum nitrication activity for a membrane and a conventional nitrogen removal system showed that the same values were obtained for both systems. These findings indicate the need for more research to further investigate the differences in nitrification rates between the M E B P R and the C E B P R process. 450 400 350 300 J 250 S 200 3 B <3 150 Z 100 50 28-Jun-03 < r 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 — • Simulated_ANA -—•— Measured ANA — Simulated_ANO Measured ANO — SimuIated_AER • — Measured AER Simulated_EFF Measured EFF Figure 4.45. T U D P model prediction of the T N concentration profiles in the biological zones and the effluent C E B P R process using the T U D P model parameters calibrated for the M E B P R process 165 30 28-Jun-03 18-JuI-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) Simulated - • - Measured _ANA ANA Simulated_ANO -A- Measured ANO ~ ~ Simulated - • - Measured AER AER Figure 4.46. T U D P model prediction of the N H 4 - N concentration profiles in the biological zones o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process The results of the denitrification process, presented by the NO3 -N concentration profiles, are shown in Figure 4.47 for the C E B P R process. The process experienced the same high anoxic nitrate concentrations during the experimental period as described earlier for the M E B P R process, Figure 4.22. The effluent profile was reasonably well predicted in both systems using the same heterotrophic kinetic parameters, Table 4.7. The calculated correlation coefficient for the measured and predicted effluent nitrate concentration in the C E B P R process was 0.8, which indicate a good fit o f the measured data. It should be noted that for the conventional system, a sludge blanket compartment o f 0.2 m 3 was modeled as an extra anoxic zone to account for the extra denitrification activity observed in the sludge blanket in the clarifier. Adding this sludge blanket resulted in a good model prediction of the lower nitrate effluent concentrations observed in the conventional side when compared to the M E P B R system. The anoxic NO3 -N concentration profile in the C E B P R was simulated in a similar trend as that of the M E B P R , but with a slightly higher correlation coefficient value of 0.41 compared to the M E B P R value of 0.19. However, the exact concentrations were not 166 predicted in both system, but the general trend was reasonably simulated and therefore, both fits were deemed satisfactory. The similar heterotrophic kinetic activity for the M E B P R and the C E B P R processes found in this study is in accordance with the results of the study by Zhang and Hall (2006), in which similar values, within the 95% confidence interval, for the heterotrophic yield, Y H , heterotrophic decay, bn, and the maximum heterotrophic growth rate, pn, parameters were measured for sludge samples taken from the U B C M E B P R and C E B P R pilot plant processes. 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 Simulated ANO —A—Measured ANO •Simulated_AER —•—Measured_AER Figure 4.47. T U D P model prediction of the N O 3 - N concentration profiles in the biological zones of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process The biological phosphorus removal process was modeled using the same calibrated T U D P parameters used for the M E B P R process, except for the rate of polyphosphate formation, k P P , which was increased from 0.1 to 0.2 g P/(g C O D P A O ' d) to better predict the anoxic ortho-phosphate concentration profile. The value of 0.2 for k P P falls within the acceptable range of values reported in literature for this parameter. 167 Results of fitting the TP concentration profiles for the C E B P R process are presented in Figure 4.48. The predicted profiles closely matched the measured data, after the sudden decrease observed on August 20. The calculated correlation coefficients for the anaerobic, anoxic and aerobic TP concentrations were 0.85, 0.92 and 0.55 respectively. The fit o f the measured TP data is definitely better for the C E B P R process when compared to that of the M E B P R , Figure 4.23. 250 -r 200 — 28-Jun-03 I8-Jul-03 7-Aug-03 27-Aug-03 16-Sep-03 6-Oct-03 26-Oct-03 (Date) — 1 Simulated_ANA Simulated_ANO SimulatedAER —•— Measured_ANA — A — Measured_ANO —•— Measured_AER Figure 4.48. T U D P model prediction of the TP concentration profiles in the biological zones and the effluent of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process The PO4-P concentration profiles also showed slightly different model fitting characteristics for both the M E B P R and the C E B P R . Results for the predicted PO4-P concentrations for the C E B P R using the T U D P model parameters for the M E B P R process are shown in Figure 4.49. The concentration profiles in the anaerobic zone were comparably well predicted in both processes. The calculated correlation coefficient for the anaerobic PO4-P concentrations in the C E B P R process was 0.62 which is close to the M E B P R value of 0.66. However, the anoxic PO4-P concentration was relatively better predicted in the M E B P R process, Figure 4.24, than that for the C E B P R process, even 168 after increasing the value for the rate of polyphosphate formation, kp P, in the C E B P R case. However, both systems resulted in low correlation coefficient values for the anoxic PO4-P concentrations. The PO4-P profile in the aerobic zone was, on the other hand, much better predicted in the conventional side, with a correlation coefficient of 0.73, when compared to the membrane side, which resulted in a low correlation coefficient o f 0.26. The peak in the aerobic PO4-P profile was predicted relatively better on the C E B P R side. The results clearly show that the mechanism of P-removal of the C E B P R process is better presented by the T U D P model when compared to that of the M E B P R process. A number of studies in the literature reported modeling results of the conventional E B P R process as described earlier, while other research showed successful applications for the M E B P R process, resulting in effluent with low phosphorus content (Fleischer et al., 2005; Patel et a l , 2005; Adam et al., 2002). More recently, Mont i et al. (2006b) presented a comprehensive study on the comparison of the performance of the M E B P R and the C E B P R processes. However a comparison of the modeling results for these systems is clearly missing. Therefore, it was not possible to compare the results found in this study with other results in the literature. However, results of the phosphorus removal batch tests performed by Monti (2006a) using sludge from both the M E B P R and the C E B P R of the U B C pilot plant during the experimental period of this study, showed that the conventional process consistently exhibited higher specific maximum anoxic and aerobic phosphorus uptake rates measured in units of mg P/(g V S S • h) when compared to its counterpart on the membrane side. This could explain the increased rate of polyphosphate formation obtained for the conventional process since the formation of polyphosphate results in the net uptake of ortho-phosphate from the solution (Meijer, 2004). The predicted V F A concentration profile in the anaerobic zone of the C E B P R process is presented in Figure 4.50. The prediction resulted in a correlation coefficient of 0.47 which is similar to that its counterpart on the membrane side. The model prediction was 169 deemed acceptable since the general trend was reasonably presented and comparable to that of the M E B P R process shown in Figure 4.25. 20 18 — on e e U 0. i 6 — 28-Jun-03 18-Jul-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 ~~ • Simulated_ANA — Measured A N A Simulated_ANO Measured A N O •S imula t edAER Measured A E R Figure 4.49 T U D P model prediction of the PO4-P concentration profiles in the biological zones of the C E B P R process using the T U D P model parameters calibrated for the M E B P R process 4.4 T U D P M o d e l Prediction Capabilit ies The T U D P model was used to predict the process behavior of the M E B P R and C E B P R processes of the U B C pilot plant running under carbon-limited conditions that resulted in periods of elevated phosphorus in the effluent in both systems. Furthermore, both processes experienced periods of significant concentrations of nitrate in the anoxic zone. These conditions imposed a great modeling challenge. 170 28-Jun-03 18-JuI-03 7-Aug-03 27-Aug-03 (Date) 16-Sep-03 6-Oct-03 26-Oct-03 Figure 4.50. T U D P model prediction of the V F A concentration profile in the anaerobic zone o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process In spite of these challenges, the T U D P model was found to describe the process behavior of both the M E B P R and the C E B P R systems reasonably well after changing only 5 to 6 kinetic parameter values from their default values (sections 4.1.5.2 and 4.3 respectively). The model predictions were deemed satisfactory for the purpose of the present .study since they predicted the general trend of the data. Although the model is believed to be sufficient for use in this work, some areas were identified that require further attention to enhance the prediction capabilities of the T U D P model. Areas entailing further investigation based on the results o f this work are described below. The V F A concentration in the anaerobic zone was not predicted correctly, as periods of zero V F A , experienced more in the M E B P R side, were always over-predicted. Reducing the half-saturation coefficient of X P A O on S A , K A P , in the model resulted in an overall lower V F A concentration profile, affecting the prediction of periods with higher V F A 171 Figure 4.50. T U D P model prediction of the V F A concentration profile in the anaerobic zone o f the C E B P R process using the T U D P model parameters calibrated for the M E B P R process In spite of these challenges, the T U D P model was found to describe the process behavior of both the M E B P R and the C E B P R systems reasonably well after changing only 5 to 6 kinetic parameter values from their default values (sections 4.1.5.2 and 4.3 respectively). The model predictions were deemed satisfactory for the purpose of the present study since they predicted the general trend of the data. Although the model is believed to be sufficient for use in this work, some areas were identified that require further attention to enhance the prediction capabilities of the T U D P model. Areas entailing further investigation based on the results o f this work are described below. The V F A concentration in the anaerobic zone was not predicted correctly, as periods of zero V F A , experienced more in the M E B P R side, were always over-predicted. Reducing the half-saturation coefficient of X P A o on S A , K A P , in the model resulted in an overall lower V F A concentration profile, affecting the prediction of periods with higher V F A 171 5 DEVELOPMENT OF GUIDELINES FOR MEBPR PROCESS DESIGN AND OPERATION USING THE SIMULATION MODEL The B N R process is a complex and highly integrated and coupled system. The competing nature of the bio-P removal and nitrogen removal processes creates a challenge in designing and operating a B N R process. Furthermore, the addition of a membrane system for the final liquid-solid separation stage adds to this complexity, since little is known about the effect o f the different biomass distribution imposed by the membrane system, when compared to a system with a conventional clarifier, on the B N R process stability and performance. A number o f key parameters play an important role in the design and operation of B N R processes. The main design parameter for efficient nutrient removal is the zone volume fraction or more importantly, the sludge mass fraction in each biological zone, anaerobic, anoxic and aerobic (Ramphao et al., 2005). Important operating conditions include the process H R T , SRT, temperature and influent V F A concentration. Other important parameters also include the p H , alkalinity, availability of certain cations such as potassium, magnesium and calcium and the ratio of organic loading to mass of activated sludge (F/M). A detailed discussion on the effect of different parameters on bio-P removal is given by Mulkerrins et al. (2004). The M E B P R process adds another dimension to the process design challenge since it removes the limitation imposed by a clarifier for the minimum residence time required for the sludge to settle. Operation at higher flowrates is possible; however more information is required about the set of operating conditions that are suitable at such high rates. Few studies have suggested design and operating conditions for B N R processes including the work by Oldham and Rabinowitz (2001), in which they reviewed the development of B N R technology in western Canada and the exportation of the technology to the United States, Europe, Australia and Asia . Based on modeling and demonstration testing, they suggested a design criterion of winter SRT of 10 days for 12 °C mixed liquor temperature and a bioreactor H R T of 7.9 hours at average daily flow to achieve effluent 173 concentrations of 0.5 mg/L o f TP and 6 mg/L of T N on monthly average basis after effluent filtration. Another research study, by Ramphao et al. (2005) focused on the impact of membrane solids-liquid separation on the design of B N R systems. In their study, they developed equations for determining the volume of each biological zone required to achieve a specific sludge mass distribution for specified anoxic and aerobic recycle ratios. They also compared the difference between the sludge mass distribution in the membrane enhanced system to that of the conventional clarifier system for different process configurations. They emphasized the advantage of the membrane B N R system as it allows changing the mass fractions, by varying the recycle flows, to optimize biological N and P removal in conformity with influent wastewater characteristics and the effluent requirements. While these studies provide useful insight on some design aspects of the B N R process, detailed information on the combined effects o f several variables and guidelines for the design and operation of B N R systems under various process conditions are clearly missing. Also missing is information on the conditions suitable for operating an M E B P R process to allow efficient wastewater treatment at high flowrates. A t the start of this project, very little was known about the required sludge mass distribution, volume fractions, recycle flows, influent V F A / T P ratios and process SRT for stable process performance in N and P removal for the M E B P R system. Studies reported in the literature focusing on the design and operation of M E B P R systems are few in number due to many factors including: 1. process complexity and the resulting importance o f many parameters, which makes studying and discussing the effect of the different parameters very challenging, 2. the membrane separation system is a new technology and little is known about the effect it has on the process design and operation, and 174 3. design studies are usually carried out by private organizations and the results are not published. A s a result, the UCT-type M E B P R pilot plant process at U B C was designed based on recommendations from industry practitioners with expertise in designing conventional B N R systems using clarifiers for the final solids-liquid separation step, rather than membranes. Accordingly, the U B C M E B P R process was used in the work carried out by Mont i (2006a) to study the process performance when operated under different process conditions including high flowrates. The operation of the process was challenging, since some periods of failure in P removal were observed. The challenge faced in designing and operating the U B C M E B P R facility and the findings by Mont i (2006a) emphasized the need for the development of design criteria and procedures that can serve as guidelines for the design and operation of a UCT-type M E B P R process. In the previous chapters, the T U D P model was calibrated to predict a set of data collected from the U B C pilot plant of an M E B P R process with a UCT-type B N R configuration accompanied by a hollow-fiber membrane technology replacing the conventional clarifier. Results of the model calibration and validation showed a satisfactory model prediction of the nutrient concentrations in the intermediate biological zones and the final effluent permeate stream. The resulting A Q U A S I M simulation model was used to study the effect of different design and operating parameters on the process stability and performance o f a UCT-type M E B P R process under steady state conditions. The simulation results were analyzed to determine the design and operating conditions that would result in a stable process performance for N and P removal. 175 Based on the results obtained for operating the U B C M E B P R pilot plant during the experimental phase and the results of the steady state simulation studies, guidelines for the design and operation of a UCT-type M E B P R process were developed. This chapter presents the experimental results for operating the U B C M E B P R pilot plant process as presented by Monti (2006a), the simulation results for the effect of different sludge mass distributions and operating conditions on the effluent concentrations of an M E B P R process operating at high flowrate and the developed guidelines based on these results for the design and operation of a UCT-type M E B P R process. Finally some recommendations for process control strategies suitable for the M E B P R process are presented. 5.1 Summary of Experimental Results fori the Effect of Different Operating Conditions on Process Performance for the UBC MEBPR Pilot Plant Process The U B C M E B P R pilot plant process was used during the experimental phase o f this project. The process design and operating conditions are given in Chapter 3. The process was designed with bioreactor zone volume fractions of 11% anaerobic, 28% anoxic and 61% aerobic and an aerobic recycle ratio o f 2 and anoxic recycle ratio of 1. When the process was operated at an SRT of 12 days and an H R T of 10 hours, it was found that this configuration resulted in a sludge mass distribution of 4% anaerobic, 21% anoxic and 75% aerobic. This design configuration and the variations in the influent V F A / T P ratio over the course of the experiment resulted in periods with stable process performance, while other periods exhibited failures in P-removal, as discussed in Chapter 3. Furthermore, Mont i (2006a) carried out a series of experiments in which the U B C M E B P R pilot plant process was operated at different process conditions with varying process H R T , SRT, influent V F A / T P and aerobic recycle ratios. A summary of the results of the present study and those of Mont i (2006a) for the U B C M E B P R process is presented in Table 5.1. The results shown for the concentrations are for the general process behavior after it reached pseudo-steady state. 176 Table 5.1. Summary of the U B C M E B P R process performance under various operating conditions as reported by Monti (2006a) HRT SRT V F A / T P Temp Zone Volume Zone Mass Ano Aer Eff Eff Ano (INF) (°C) Fraction (%) Distribution (%) Rec Rec P0 4 -P N H 4 - N N O 3 - N Ana Ano Aer Ana Ano Aer 10 12 6 21.9 11 28 61 4 21 75 1 2 > 1 <0.2 >0.1 10 12 8 21.9 11 28 61 4 21 75 1 2 <0.5 <0.2 >0.1 10 12 6 21.1 11 28 61 4 19 77 1 1 > 1 <0.2 <0.1 10 12 10 18.1 11 28 61 4 21 75 1 2 <0.5 <0.2 <0.1 7 12 10 16.3 11 28 61 4 21 75 1 2 > 1 <0.2 >0.1 7 12 12 18.5 11 28 61 4 19 77 1 1 <0.5 <0.2 <0.1 7 20 12 11 28 61 4 19 77 1 1 <0.5 <0.2 <0.1 5 20 12 - 11 28 61 4 19 77 1 1 <0.5 <0.2 <0.1 Monti (2006a) gives a detailed explanation for the results presented in Table 5.1 for the behavior of the M E B P R process at the different operating conditions. A summary of the results presented by Monti (2006a) for the operation of the M E B P R process include the following. • A minimum influent V F A / T P ratio was required for effective B P R performance in the M E B P R system. A t the experimental conditions described in the work of Monti (2006a), the required influent V F A / T P ratio was found to be 10 - 12. • A s the process H R T increased, the required influent V F A / T P ratio for stable B P R process performance increased. • A s the process SRT increased, the required influent V F A / T P ratio for stable B P R process performance increased. • A t an aerobic biomass fraction o f 77%, the process exhibited stable low effluent NH4-N concentrations throughout the experimental period despite the changes in the operating conditions experienced by the process. • Stable process operation for N and P removal at an H R T as low as 5 hours was possible when sufficient influent V F A concentration was supplied. • High concentrations of NO3-N in the anoxic zone were controlled by manually adjusting the aerobic recycle flow. These results reflect the challenge associated with designing and operating a B P R process. There are many factors to be considered and certain conditions can result in process failures and high effluent nutrient concentrations. 177 The results presented in Table 5.1 form a good starting point to obtain a general overview of the M E B P R process performance under certain operating conditions. However, little variation in the process biomass distribution and process SRT were experienced. Further investigation of the M E B P R process performance under different operating conditions, especially at high flowrates, is required to better understand the effect o f these variables on the process performance. To that end, it was decided to use the T U D P model with the calibrated model parameters for the M E B P R process presented in Chapter 5, to perform steady state simulations to further investigate the M E B P R process performance under various operating conditions at high flowrate operation. 5.2 TUDP Model Predictions for the Effect of Different Design and Operating Conditions on MEBPR Process Behavior The calibrated T U D P model for the M E B P R process, as presented in the previous chapter, was used in the simulation environment to simulate a UCT-type M E B P R process under different design and operation conditions. Steady state runs were simulated with different sludge mass distributions, internal recycle ratios, process SRT, temperature and influent V F A / T P ratios to compare the predicted results. The objective of the simulation studies was to find the set of design and operating conditions that result in low effluent concentrations while operating at high flowrates. The average influent concentrations observed at the pilot plant during the experimental period were used as the influent stream to the simulation. The influent concentrations used are given in Table 5.2. The mass fraction distribution was varied in the system by changing the zone volume fractions and the internal recycle flows. The design equations presented by Ramphao et al. (2005) for the UCT-type process with membrane solids-liquid separation technology were used to determine the required volume zone fractions to achieve the desired sludge mass distribution in the system. Prior to using these equations, they were tested using the sludge mass fraction distribution and the recycle 178 flows of the U B C M E B P R process. The equations were found to result in the exact numbers for the volume fractions of the process. Table 5.2. Influent concentrations used for process design and operation studies Component Units Value C O D p a r t g C O D / m 3 291 C O D t o t g C O D / m 3 414 TSS g T S S / m 3 111 T N . g N / m 3 40.97 TP g P / m 3 4.53 SArNF g C O D / m 3 28 SFrNF g C O D / m 3 75 SIlNF g C O D / m 3 20 NH4-N g N / m 3 27 PO4-P g P /m 3 2.7 NO3-N g N / m 3 0.1 D O g 0 2 / m 3 3 Temperature °C 21.9 Initially, several simulations were carried out to identify the biomass fraction distribution ranges of interest and the model predictions at these conditions. Table 5.3 lists the conditions for the initial simulation runs carried out. Figure 5.1 shows the results of these runs in terms of effluent N H 4 - N , PO4-P and NO3-N concentrations with respect to the run number. These simulations were run at an influent V F A / T P ratio of 6.18 (computed as SAINF/TP), temperature of 21.9 °C, SRT of 10 days and H R T o f 4 hours. Figure 5.1 was divided into four sections. Section I, runs number 1-9, shows the effect o f recycle flow on the effluent concentrations at a specified sludge mass distribution. Section II, runs number 10 - 18, also shows the effect of recycle flow but at a different sludge mass distribution. Section III, runs number 1 9 - 2 9 , shows the effect of different biomass distributions on the effluent concentrations, while keeping the recycle flows the same. Section IV, runs number 30 - 34, examines the effect of different recycle flows on the sludge mass distribution observed in section III, run 29 which resulted in low effluent N H 4 - N and PO4-P concentrations. 179 Table 5.3. Simulation test run conditions for the initial process design and operation studies for the M E B P R process . Zone Volume Zone Mass Run HRT SRT VFA/TP Temp Fraction (%) ANO AER Distribution (%) No (hr) (Days) (INF) (°Q ANA ANO AER Rec Rec ANA ANO AEI 1 4 10 6.2 21.9 10 30 60 1.0 1.0 3 20 77 2 4 10 6.2 21.9 10 30 60 1.0 2.0 4 24 72 3 4 10 6.2 21.9 10 30 60 1.0 3.0 5 26 69 4 4 10 6.2 21.9 10 30 60 2.0 1.0 4 20 76 5 4 10 6.2 21.9 10 30 60 2.0 2.0 5 24 71 6 4 10 6.2 21.9 10 30 60 2.0 3.0 6 26 68 7 4 10 6.2 21.9 10 30 60 3.0 1.0 5 19 76 8 4 10 6.2 21.9 10 30 60 3.0 2.0 6 24 70 9 4 10 6.2 21.9 10 30 60 3.0 3.0 7 26 68 10 4 10 6.2 21.9 25 40 35 1.0 1.0 11 33 56 11 4 10 6.2 21.9 25 40 35 1.0 2.0 12 38 50 12 4 10 6.2 21.9 25 40 35 1.0 3.0 13 40 47 13 4 10 6.2 21.9 25 40 35 2.0 1.0 14 32 55 14 4 10 6.2 21.9 25 40 35 2.0 2.0 16 37 48 15 4 10 6.2 21.9 25 40 35 2.0 3.0 16 39 45 16 4 10 6.2 21.9 25 40 35 3.0 1.0 15 31 54 17 4 10 6.2 21.9 25 40 35 3.0 2.0 17 36 47 18 4 10 6.2 21.9 25 40 35 3.0 3.0 18 38 44 19 4 10 6.2 21.9 10 30 60 2.0 2.0 5 24 71 20 4 10 6.2 21.9 11 28 61 2.0 2.0 6 22 72 21 4 10 6.2 21.9 15 31 54 2.0 2.0 8 25 66 22 4 10 6.2 21.9 15 35 50 2.0 2.0 9 29 62 23 4 10 6.2 21.9 16 34 50 2.0 2.0 9 28 63 24 4 10 6.2 21.9 20 35 45 2.0 2.0 12 30 58 25 4 10 6.2 21.9 20 50 30 2.0 2.0 13 46 41 26 4 10 6.2 21.9 25 39 37 2.0 2.0 15 35 50 27 4 10 6.2 21.9 25 45 30 2.0 2.0 16 42 42 28 4 10 6.2 21.9 25 34 41 2.0 2.0 15 30 55 29 4 10 6.2 21.9 26 29 46 2.0 2.0 15 25 60 30 4 10 6.2 21.9 26 29 46 2.0 2.5 16 26 58 31 4 10 6.2 21.9 26 29 46 2.0 3.0 16 27 57 32 4 10 6.2 21.9 26 29 46 1.0 3.0 13 28 59 33 4 10 6.2 21.9 26 29 46 1.0 2.5 13 27 60 34 4 10 6.2 21.9 30 30 40 2.0 2.0 19 27 54 Examining the results in Figure 5.1 shows the competing nature of the bio-P and nitrogen removal processes. In section I, where the sludge mass distribution varied between 3 -7% for anaerobic, 19 - 26% for anoxic and 68 - 77% aerobic, which provides a larger 180 aerobic zone mass distribution than in section II, the nitrification process seems stable and unaffected by the varying aerobic recycle flows. The ortho-phosphate concentration, on the other hand, appears more sensitive to changes in the aerobic recycle flows since higher recycle flows results in nitrate leaking to the anaerobic zone which negatively affects the ortho-phosphate release (Shehab et al., 1996). A s for denitrification, increasing the aerobic recycle flow ratio from 1 to 3, reduced the effluent nitrate concentration from about 12 to 7 g N / m 3 . While the denitrification was found to be greatly influenced by the aerobic recycle flow, it was less sensitive to the changes in the biomass distribution simulated in these runs. 14.0 Figure 5.1. Effluent concentrations for the steady state simulation results of the different run conditions (simulated process T= 21.9 °C, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) In section II, where the sludge mass distribution varied between 11 - 18% for anaerobic, 31 - 40% for anoxic and 44 - 56% aerobic, which results in smaller aerobic zones when compared to section I, the effluent ammonium concentration clearly increased and the 181 ortho-phosphate concentration dropped to low levels. In this case, the nitrification process was more affected by the changes in the aerobic recycle flow since increasing the flow results in a lower sludge mass fraction in the aerobic zone and therefore, a lower zone S R T and vice versa. However, bio-P removal appeared to be less sensitive to changes in the recycle flows. The effluent nitrate concentration showed the same pattern as in section one, however, since the sludge mass fraction in the anoxic zone was increased, the effluent nitrate concentrations were lower than those in section one. In section III, the effect of different sludge mass distributions is examined and it is clear that when the effluent ortho-phosphate concentration is low, the ammonium concentration is higher and vice versa, which emphasizes the competing nature of the two processes. The conditions of run number 29 were found to result in a reasonable effluent concentration providing a balance between the two processes. Therefore, these conditions were further studied in section IV to examine the effect of varying the recycle flows at this condition. It was found that anoxic and aerobic recycle ratios of 1 and 3 respectively, resulted in the lowest effluent concentrations. It should be noted that the objective here was mainly focused on comparing the change in the predicted effluent nutrient concentrations rather than concentrating on the absolute value of the predicted concentrations, since it was shown in Chapter 4 that the model was capable o f reasonably predicting the process trend but not the absolute concentrations in some cases. However, it was important to ensure that the results obtained from the steady state simulations for different operating conditions were valid under dynamic conditions as well . Therefore, the conditions tested in Table 5.3 for reactor volume fractions and recycle flows were used to model the M E B P R process with the dynamic influent data collected from the U B C M E B P R pilot plant during the period between July and October of 2003. The measured temperatures and influent concentrations collected from the pilot plant were used, while the bioreactor volumes and recycle flows were varied according to the conditions listed in Table 5.3. Results showed that runs in Table 5.3 resulting in 0.15 g/m , or less, for effluent concentrations for N H 4 - N and PO4 -P in the steady state simulations were found to result in a stable process performance when the same reactor volume fractions and recycle ratios were used to model the M E B P R process using the 182 influent dynamic data collected from the M E B P R process. Therefore the simulation studies carried out in this chapter were aimed at achieving effluent concentrations below 0.15 g/m 3 for P 0 4 - P and N H 4 - N . The model predictions obtained for running the simulation using the influent dynamic measured data for the M E B P R process and the bioreactor volume fractions and recycle flow ratio conditions o f run number 32 were examined to carefully evaluate these design conditions under dynamic operations. Simulation results showed lower predicted effluent P 0 4 - P concentrations than the ones predicted using the actual U B C M E B P R bioreactor volume fractions and recycle flow ratios. Complete nitrification was achieved; however, higher anoxic nitrate concentrations were predicted when compared to the ones predicted using the original design caused by the high aerobic recycle flow ratio o f 3. Therefore, it was concluded that process control of the aerobic recycle flow is critical for stable M E B P R process performance to ensure essentially zero nitrate in the anoxic zone while maximizing the denitrification load. These initial results gave a good starting point for selecting the conditions of the next set of simulation studies. Based on these results, it was decided to simulate the process performance under different operating conditions in an attempt to determine the design parameters that result in stable and efficient process performance. The process parameters varied were the process SRT, temperature, influent V F A / T P ratio and the sludge mass distribution in the system. A l l runs were performed at an H R T of 4 hours in an attempt to evaluate the process performance at high flowrates. The anoxic and aerobic recycle flow ratios were kept constant, at 2 for each, since it was found that on-line process control of these flows is required to ensure stable process operation provided that an adequate anoxic zone sludge mass fraction is achieved, as described earlier. The conditions used in the simulation studies for the M E B P R process are listed in Table 5.4. The influent V F A / T P ratio was varied in the model by changing the influent SA concentration resembling the addition of acetate. Testing the combination of these conditions resulted in 575 steady state simulation runs. The results from these runs and the initial runs were used in the following sections to discuss the effect of different 183 parameters on the performance of the M E B P R process as.predicted by the model. The results obtained from the initial runs were analyzed in more detail below to further elucidate the effect of the recycle flows on process performance. Table 5 .4 . Process conditions used in the M E B P R process design and operation studies Parameter Units Ranges H R T SRT Temperature Influent V F A / T P Ratio A N A / A N O / A E R mass fraction Hours Days °C g C O D / g P % 8 ,10 ,15 ,20 ,25 ,30 10, 15,20 5 ,10 ,15 ,20 4/21/75, 6/24/70, 8/27/65, 10/30/60, 15/30/55, 18/32/50, 20/35/45, 30/30/40 5.2.1 Effect of Biological Mass Distribution The biomass distribution in a B N R process is governed by the selection of the volume fractions o f each biological zone. In a UCT-type M E B P R system with the membrane in the aerobic reactor, the biomass distribution can be varied, within a range, by varying the interreactor recycle ratios (Ramphao et al, 2005). However, as discussed earlier, changes to the aerobic recycle may increase the nitrate load to the anoxic zone and i f the nitrate is not completely denitrified, it may leak to the anaerobic zone and negatively affect the P-release. Therefore there is a limitation on the window of variation for interreactor recycle ratios. This emphasizes the need to properly design the M E B P R process to ensure adequate biomass distribution in each biological zone to result a in stable and efficient process operation. In this section, the results of the initial runs number 1 - 29 were used to study the effect of the different biomass distributions on the effluent N H 4 - N , N O 3 - N and PO4-P. These simulation runs were performed at a constant S R T of 10 days, a process temperature of 21.9 °C, an influent V F A / T P ratio of 6.2 and an H R T of 4 hours. The biomass distribution was varied in the biological reactors by changing the volume fractions or the interreactor recycle flows. Therefore, it was decided to use the simulation results of these runs, as opposed to the results of the 575 runs, to discuss the effect of varying the 184 biomass distribution by changing the recycle flow ratios as well as the bioreactor volume fractions while keeping other parameters constant. Figure 5.2 presents the effect of varying the aerobic biomass fraction on the effluent N H 4 - N concentration and the results show that the effluent N H 4 - N decreases with increasing aerobic mass fraction until it reaches the value of about 65%, beyond which a further increase in the mass fraction has minimal effect on the N H 4 - N concentration. This is because as the mass fraction increased from 40 to 65%, the mass of nitrifiers also increased, resulting in higher nitrification rates until it reached a point where the F / M ratio became too low and the process became substrate limited. The results show that under these conditions, an aerobic mass fraction of 65% was sufficient to achieve stable and efficient process operation even at the low H R T of 4 hours. However, in the initial design of the U B C M E B P R pilot plant, the aerobic mass fraction was 75%, which explains the reason for the stable nitrification throughout the experimental study. Nonetheless, other factors need to be taken into consideration such as variations in the influent substrate concentrations and seasonal temperature variations. The effect of different temperatures is examined in section 5.1.4. 185 Figure 5.2. Effect of the aerobic biomass fraction on the nitrification process (simulated process temperature = 21.9 °C, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) Figure 5.3 presents the effect of the anaerobic biomass fraction on the effluent PO4-P concentration using the results of runs number 1 - 34 of Table 5.3. The results show that as the anaerobic biomass fraction increases, the effluent PO4-P concentration decreases. Increasing the anaerobic biomass fraction increases the rate of fermentation and therefore the V F A available for P-release. The rate increases with increasing anaerobic biomass fractions until it becomes limited by the fermentable C O D available. Therefore, a further increase in the mass fraction has little effect on the effluent P 0 4 - P concentration. It is clear that the data are more scattered in Figure 5.3 when compared to that of Figure 5.4, especially at lower biomass fractions. This is mainly due to the effect of varying the aerobic and anoxic recycle flows, since increasing these flows results in nitrate leaking from the anoxic zone to the anaerobic zone. For example, the maximum effluent PO4-P concentration was observed with an anaerobic biomass fraction of 7%, which corresponds to run number 9 in which both recycle flow ratios were set to 3. This effect is more severe at lower mass fractions where the P-release is less stable. 186 Figure 5.3. Effect of the anaerobic biomass fraction on the Bio-P removal process (simulated process temperature = 21.9 C, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) Figure 5.4 presents the effect of the anoxic mass fraction on the anoxic N O 3 - N concentration using the results of runs number 1 9 - 2 9 . The results show that an increase in the anoxic mass fraction up to 40% results in better denitrification (lower effluent N O 3 - N concentrations) beyond which the effect is minimal, which could be due to substrate limitation. Complete denitrification of anoxic zone N O 3 - N is important to decrease the total effluent nitrogen concentration, but more importantly, it is essential to avoid nitrate leakage to the anaerobic zone. However, other factors such as the aerobic recycle flow, which controls the nitrate load to the anoxic zone (discussed below), have a bigger effect on the effluent N O 3 - N concentration than the anoxic sludge biomass fraction. 187 Figure 5.4. Effect of the anoxic biomass fraction on the anoxic NO3 -N concentration (simulated process temperature = 21.9, H R T = 4 hours, SRT = 10 days and influent V F A / T P = 6.2) 5.2.2 Effect of SRT Different microbial populations are involved in B N R processes and they have different requirements in relation to SRT. Nitrifying bacteria are slow growing organisms that require relatively long sludge ages to achieve complete nitrification (Mulkerrins et al., 2004). A number of studies reported that an SRT of 10 days would result in good P-removal rate (Change et al., 1996; Choi et al., 1996). However, Furumai et al. (1999) suggested that for combined biological nitrogen and phosphorus removal systems, longer SRTs are beneficial for P-removal. Oldham and Rabinowitz (2001) indicated that modeling and demonstration for combined nitrogen and phosphorus removal have led to the use of an S R T of 10 days for 12 °C mixed liquor temperature and bioreactor of H R T of 7.9 hours at average daily flow. The authors added that these conditions are applied provided that a fermenter is used and good control of D O in the aerobic zone and plant flow is achieved. Moreover, Monti 188 (2006a) reported that stable and efficient operation of an M E B P R process can be achieved at an SRT of 20 days and an H R T of 5 hours with an external supply of acetate. While these studies suggest general guidelines for process design of B N R systems, specifications about system configuration, volume fractions, sludge mass distribution, and operating temperatures and the effect of these variables on the required SRT are clearly missing. In this section, the effect of process SRT and aerobic mass fraction on the effluent N H 4 - N is discussed using results from the simulation carried out at an H R T of 4 hours, 20 °C temperature and influent V F A / T P ratio of 10. These conditions were used to eliminate the effect of low temperature or insufficient influent V F A on process performance and focus mainly on the parameters of interest. The results are plotted in Figure 5.5 and show that at a constant aerobic biomass fraction, the effluent N H 4 - N decreased as the process SRT was increased. This is due to the slow growing nitrifying organisms that are generally lost at low SRTs (Mulkerins et al., 2004). The results also show that as the SRT was increased, the aerobic mass fraction required to achieve a low N H 4 - N concentration decreased. This was due to fact that increasing the process SRT allowed for the slow growing nitrifying organisms to grow and their concentration in the sludge increased and therefore, a lower aerobic mass fraction was needed to achieve the same nitrification rates. The U B C M E B P R process was operated at an aerobic mass fraction of about 75% and Figure 5.5 shows that this condition results in a low N H 4 - N concentration at all tested process SRTs, which explains the stable nitrification observed though out the course of the pilot plant experiment. 189 30 35 40 45 50 55 60 65 Aerobic Biomass Fraction (%) • SRT = 8 • SRT =10 • SRT = 15 -A SRT = 20 ^ SRT = 25 - • - SRT = 30 70 75 80 Figure 5.5. Effect of process SRT and aerobic biomass fraction on the effluent N H 4 - N concentration (simulated Process temperature = 20 °C, H R T = 4 hours and influent V F A / T P = 10) Figure 5.6 presents the effect o f SRT and anaerobic biomass fraction on the effluent PO4-P concentration for the simulation runs carried out at an H R T o f 4 hours, temperature of 20 °C, influent V F A / T P ratio of 5 and aerobic and anoxic recycle ratios of 2. The influent V F A / T P ratio o f 5 was used in this case to examine the effect of fermentation on the bio-P removal process. Results of Figure 5.6 show that at anaerobic biomass fractions lower than 15%, increasing the process SRT results in an increasing effluent PO4 -P concentration. This is the case because a higher process SRT results in a lower F / M ratio and a lower sludge yield in addition to extended biomass decay, which negatively affect the concentration of P A O s in the mixed liquor and the bio-P removal process. Results of Figure 5.6 also show that as the SRT increases, the anaerobic sludge mass fraction required to achieve stable and efficient bio-P removal increases. This can be explained by the fact that a higher SRT results in higher nitrification rates and in turn more nitrate being recycled to the anoxic zone, which leaks to the anaerobic zone. The 190 presence of nitrates in the anaerobic zone results in consumption of organic compounds by denitrifiers, thus decreasing the availability of organic matter for P A O s and reducing the efficiency of P-release. In that case, a larger sludge mass fraction is required to enhance the fermentation and result in the required V F A concentration for the P-release. 1.6 0 4 1 1 1 1 , , , , , , , , , , , 1 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 Anaerobic Biomass Fraction (%) Figure 5.6. Effect of process SRT and anaerobic biomass fraction on the effluent PO4-P concentration (simulated process temperature = 20 °C, H R T = 4 hours and influent V F A / T P = 5) 5.2.3 Effect of Recycle Flows The internal recycle flows in a UCT-type membrane enhanced B N R process have two major effects on the process. First, varying the recycle flows affects the sludge mass distribution in the system. Increasing the aerobic recycle flows, for example, decreases the aerobic mass fraction and increases the anoxic mass fraction and vice versa. This is presented by the results for the mass distribution for runs number 1 - 18 in Table 5.2. Secondly, increasing the aerobic recycle may increase the nitrate load to the anoxic zone beyond the denitrification capacity, resulting in nitrate leakage to the anaerobic zone. Therefore the second effect may limit the flexibility described by the first effect. 191 Rhamphao et al. (2005) indicated that the ability to vary the mass fraction distribution in an M E B P R process without the limitation of the clarifier is a significant advantage of the M E B P R option since it allows changing the mass fractions to optimize biological nitrogen and phosphorus removal in conformity with influent wastewater characteristics and the effluent nitrogen and phosphorus concentration requirements. However, this flexibility is limited, as described above, and therefore care should be taken when designing M E B P R processes to select suitable bioreactor volume fractions that result in a reasonable biomass distribution in the system. The results presented in this section are based on the initial simulation runs number 1 -34 of Table 5.3 in which the recycle flowrates were varied along with the biomass distribution in the bioreactors. Figure 5.7 shows the effect of the aerobic recycle ratio on the effluent NO3-N concentration at different anoxic mass fractions. The results show that an increase in the aerobic recycle ratio from 1 to 2 can result in a decrease in the effluent N O 3 - N concentration from approximately 11.5 g N / m 3 to about 8 g N / m 3 . On the other hand, increasing the anoxic biomass fraction from 22% to 46%, at the same aerobic recycle flow, resulted in a decrease o f about 1 g N / m in the effluent N O 3 - N concentration. This shows that the nitrate load to the anoxic zone has a larger effect on the denitrified load than the anoxic biomass fraction under these conditions. 192 a 3 B c « -0 = E u 6 5 I • AER Recycle Ratio = 1 • AER Recycle Ratio = 2 A AER Recycle Ratio = 3 • • • o.o 10.0 20.0 30.0 40.0 Anoxic Biomass Fraction (%) 50.0 60.0 Figure 5 . 7 . Effect of the aerobic recycle ratio on the effluent N O 3 - N concentration (simulated process temperature = 21.9 °C, H R T = 4 hours, and influent V F A / T P = 6.2) It was also important to examine the effect o f the aerobic recycle ratio on the anoxic nitrate concentration, for its critical effect on the P-release. The simulation results of runs 1 - 34 o f Table 5.3 were used to study this effect and these are presented in Figure 5.8. Results show that at an aerobic recycle ratio o f 3 and with an anoxic mass fraction o f less than 30%, the nitrate concentration can exceed 1 g/m 3 . However, when the anoxic mass fraction was increased to above 30%, the nitrate concentration decreased to 0.13 g/m 3 . This shows that a minimum anoxic mass fraction is required to achieve complete denitrification. Figure 5.8 shows that under the conditions of these simulation runs, a minimum anoxic mass fraction o f about 30% is needed to avoid elevated anoxic nitrate concentrations. It should be noted that these simulations were carried out at an influent V F A / T P ratio o f 6.2. Different results may be obtained i f a higher influent V F A / T P ratio was used since it is known that higher V F A concentration can increase denitrification rates (Oldham and 193 Rabinowitz, 2001). The effect o f the influent V F A / T P ratio on the M E B P R process performance is described in more detail in Section 5.1.5. 20.0 30.0 40.0 Anoxic Mass Fraction (%) 60.0 Figure 5.8. Effect of the aerobic recycle ratio on the anoxic NO3-N concentration (simulated process temperature = 21.9 °C, H R T = 4 hours, and influent V F A / T P = 6.2) These results explain the complexity of controlling the recycle flows and the competing nature of the nitrogen and phosphorus removal processes. Optimizing the recycle flows requires on-line control to maximize the denitrification load without affecting the bio-P removal process. 5.2.4 Effect of Temperature The influence of temperature on nitrification is well described and reveals a high correlation between temperature and nitrification (Metcalf and Eddy, 2003). The effect o f temperature on the bio-P process however is more complex because of the different factors involved. Baetens et al. (1999) indicated that the temperature effect on a full-scale biological nutrient removal plant is not straightforward because of the different 194 influences of temperature on the subprocesses. Therefore, the effects o f temperature on the nitrification and bio-P processes were examined using the results of the steady state simulations. The results are discussed in this section. The results of the simulations carried out at an SRT of 10 days, an H R T of 4 days and an influent V F A / T P ratio of 5, were used to investigate the simulated effect of temperature and the aerobic biomass fraction on the nitrification process. These conditions were chosen since they are close to the conditions observed at the U B C M E B P R process. The analysis results are presented in Figure 5.9. Results show that increasing the process temperature generally decreases the effluent N H 4 - N concentration at otherwise constant process conditions. For example, the results show that at an aerobic biomass fraction of 65%, the effluent N H 4 - N concentration goes from 1 to 0.11 as the temperature increases from 10 to 20 °C. This prediction was expected as discussed earlier. Results presented in Figure 5.9 also show that a decrease in temperature requires a higher aerobic mass fraction to achieve the same effluent N H 4 - N concentration. These findings show that different bioreactor biomass distributions may be necessary to accommodate for seasonal temperature variations. 195 z S = Temp = 10 deg C • Temp = 15 deg C • Temp = 20 deg C 0.01 1 Aerobic Biomass Fraction (%) Figure 5.9. Effect of the temperature and aerobic biomass fraction on the effluent N H 4 - N concentration (simulated process SRT = 10 days, H R T = 4 hours and influent V F A / T P = 5) Investigating the model's prediction for the effect of temperature on the effluent PO4-P concentration was done by plotting the effluent PO4-P concentration for the results for the simulation performed at a biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic, process H R T of 4 hours, process SRT of 20 days and an influent V F A / T P ratio of 10. This mass distribution was chosen since it is close to the conditions of run 29 of initial simulation runs of Table 5.3 which resulted in the lowest effluent concentrations. Furthermore, the influent V F A / T P ratio of 10 was used to ensure that the process was not carbon-limited. The results are presented in Figure 5.10 and indicate that as the process temperature increases, the effluent PO4-P concentration increases for the specified conditions. This phenomenon is explained by Baetens et al. (1999) who indicated that with increasing temperature, the A T P (energy) requirement for maintenance increases as 196 shown by Bradjanovic et al. (1997), which causes a decrease in the substrate available for net biomass growth. Furthermore, biomass decay increases with increasing temperature and both of these effects result in less biomass production for the same amount of substrate used, causing a decrease in the net observed yield and thus a decrease in P-removal capacity. This was verified by checking the simulated heterotrophic and P A O s biomass concentration at different temperatures. These were found to be higher at lower temperatures and they decreased as the temperature increased (results not shown). Earlier results showed that the performance of the bio-P removal process is better at lower process SRTs. Therefore, it was decided to investigate the effect o f varying the process temperature on the effluent PO4-P concentration under different process SRTs. The investigation was carried out by examining the steady state simulation results of the same runs used for Figure 5.9. The results of the simulations are presented in Figure 5.11. Figure 5.10. Effect of the temperature on the effluent PO4-P concentration (simulated process SRT = 20 days, H R T = 4 hours, influent V F A / T P = 10 and biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic) 197 A t these conditions, it was found that at a process temperature of 10 °C, the effluent PO4-P concentration was high at an SRT of 8 and 10 days and then decreased with increasing SRT, but stayed relatively constant as the SRT increased from 20 to 30 days. However the process was more stable at higher temperatures for the lower SRTs. This behaviour was also reported by Mamais and Jenkins (1992) who indicated that washout o f P-removing organisms was observed at an aerobic SRT of 2.1 days at 13.5 °C while washout only occurred at an aerobic SRT of 1.5 days for a temperature of 20 °C, indicating that the minimum aerobic SRT required to avoid bio-P organisms washout increases with decreasing temperatures. 0.01 Temp = 20 oC • Temp = 15 oC A Temp = 10oC SRT (Days) Figure 5.11. Effect of the temperature and SRT on the effluent PO4-P concentration (simulated process H R T = 4 hours, influent V F A / T P = 10 and biomass distribution o f 10% anaerobic, 30% anoxic and 60% aerobic) 5.2.5 Effect of Influent V F A / T P Ratio It is well accepted in the literature and in practice that V F A are essential for an effective B P R process. Mulkerrins et al. (2004) reported that 7 - 9 mg of V F A are needed to remove 1 mg o f phosphorus. Furthermore, Mont i (2006a) operated the U B C M E B P R process at influent V F A / T P ratios of 10 - 12 to achieve a stable E B P R process. 198 While these studies reported the required influent V F A / T P ratios for effective B P R process operations under specific process conditions, the effect of various operating conditions on the required ratio needed for effective phosphorus removal has not been addressed in the literature. Therefore, it was decided to use the simulation model to examine the effect of various process temperatures, SRTs, biomass distributions and influent V F A / T P ratios on the B P R process. The influent V F A / T P ratio was varied by increasing the SA concentration in the model resembling an external addition of acetate. Figure 5.12 presents the results for simulation runs carried out at an H R T of 4 hours, temperature of 20 °C and a biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic while varying the process SRT and influent V F A / T P ratio. The influent TP concentration was 4.53 g/m 3 as shown in Table 5.1. This biomass distribution was chosen since it is similar to the conditions of run 29 of Table 5.3. Furthermore, the results of Figure 5.11 show that the effect of the process S R T on the effluent PO4-P concentration at 20 °C is consistent. Therefore the simulations were carried out at this temperature. Figure 5.12 shows that in order to achieve the same effluent PO4-P concentration while increasing the SRT, the influent V F A / T P needs to increase as well . The same finding was reported by Mont i (2006a) who noted that as the process SRT was increased from 12 days to 20 days, the required influent V F A / T P ratio increased from 10 to 12 to achieve a low effluent PO4-P concentration. It should be noted that some process conditions such as the process biomass distribution, H R T , temperature and influent characterization were different between the present study and that of Mont i (2006a), so the comparison is not intended to be a quantitative but qualitative. 199 10 • SRT = 10 - • - SRT = 30 SRT = 20 25 0.01 V F A / T P Ratio Figure 5.12. Effect of the SRT and influent V F A / T P ratio on the effluent P 0 4 - P concentration (simulated process H R T = 4 hours, temperature = 20 °C and biomass distribution of 10% anaerobic, 30% anoxic and 60% aerobic) Another set of simulations was carried out at the same process conditions while decreasing the anaerobic biomass fraction from 10 to 4% and the anoxic biomass fraction from 30 to 21% and increasing the aerobic biomass fraction from 60 to 75%. This was done to examine the effect of changing the anaerobic mass fraction on the influent V F A / T P ratio required to achieve the same effluent PO4-P concentration. The results of these simulations are presented in Figure 5.13. Comparing the results o f Figure 5.12 and Figure 5.13 shows that to achieve an effluent PO4-P concentration of 0.07 g/m 3 at an SRT of 10 days and an anaerobic mass fraction of 10%, an influent V F A / T P ratio of 10 is required. However, to achieve the same effluent P 0 4 - P concentration at an anaerobic mass fraction o f 4%, results of Figure 5.13, the influent V F A / T P ratio needs to be increased to about 15. 200 10 E w = O 33 -e u = e U O cu c o = I N 0.1 0.01 SRT = 10 SRT = 20 SRT = 30 VFA/TP Ratio Figure 5.13. Effect of SRT and influent V F A / T P ratio on the effluent PO4-P concentration (simulated process H R T = 4 hours, temperature = 20 °C and biomass distribution of 4% anaerobic, 21% anoxic and 75% aerobic) In order to further examine the effect of different anaerobic mass fractions and influent V F A / T P ratios on the process performance, a number of simulations were performed at an H R T of 4 hours, an SRT o f 15 days and a temperature of 20 °C, while varying the sludge mass distribution and the influent V F A / T P ratios. A n SRT of 15 days was chosen to be the closest to the SRT of 12 days used for the measured concentrations used in the present study at the U B C M E B P R pilot plant. The results of the simulations are presented in Figure 5.14. The results show that at lower anaerobic biomass fractions the effect of the influent V F A / T P ratio is the greatest, but as the anaerobic biomass fraction increases to 30%, the effect o f different influent V F A / T P ratios becomes minimal. This is due to the fact that at higher anaerobic biomass fractions, the rate of fermentation is higher and therefore, more carbon becomes available. Furthermore, the results of Figure 5.14 show that as the influent V F A / T P ratio increases, the effect o f varying the anaerobic mass fraction on the effluent PO4-P concentration is decreased. For example at the influent 201 V F A / T P ratio of 15, the effluent P 0 4 - P concentration remained less than 0.1 g/m 3 as the anaerobic mass fraction varied between 4 and 30%. E 6X a H-• -= o a e U 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 • VFA/TP Ratio = 5 - • - V F A / T P Ratio = 10 -A- VFA/TP Ratio = 15 10 15 20 25 Anaerobic Biomass Fractions (%) 30 35 Figure 5.14. Effect of the anaerobic biomass fraction and influent V F A / T P ratio on the effluent PO4-P concentration (simulated process H R T = 4 hours, SRT = 15 days and temperature = 20 °C) A s for the effect of influent V F A / T P on the effluent NH4-N concentration, surprisingly, it was found that increasing the influent V F A / T P ratio negatively affected the nitrification process, resulting in an increase in the effluent NH4-N concentration as shown in Figure 5.15. Further examination of the model results revealed that increasing the influent carbon concentration resulted in an increased growth of the heterotrophic biomass, XH, and the phosphorus-accumulating organisms, X P A o , which also increased the biomass concentration in the system. Since the system was simulated under constant SRT, i.e., the waste depended on the total biomass concentration in the system, the nitrifiers concentrations were decreased over time, which in turn decreased the nitrification efficiency. 202 M e o e o u e o U = z a 3 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 10 15 V F A / T P Ratio 20 25 Figure 5.15. Effect of the influent V F A / T P ratio on the effluent NH4-N concentration (simulated process H R T = 4 hours, S R T = 10 days, temperature = 20 °C and sludge mass distribution of 4% anaerobic, 21% anoxic and 75% aerobic) The results presented in this section are intended to give a general overview of the effect of the different operating conditions and biomass distribution on the process performance as predicted by the simulation model. It should be noted that all these simulations were performed at steady state conditions and therefore, they indicate the long term effect of these variables. Different results and even process failure may occur due to sudden changes in influent composition or process temperature (Mulkerrins et al., 2004). While this section was focused on discussing the effects of varying one or two variables at a time on the process performance for nitrogen and phosphorus removal, the combined effect of varying all discussed variables on process performance and the recommended design conditions based on simulation results are discussed in the following section. 203 5.3 Recommended Design Conditions Based on Simulation Results In the last section, the effects of different operating conditions and design parameters on the process performance for P and N removal were discussed in terms of simulation results. The clear competition between the N and P removal processes makes it difficult to decide on the design and operating conditions for the M E B P R to result in efficient process performance. Therefore, the simulation environment was used to test different process conditions to choose the conditions that result in a stable simulated process performance. To that end, the simulation results for the 575 steady state runs were analyzed to determine the conditions resulting in effluent PO4-P and NH4-N concentrations that were less than 0.15 g/m 3 . This low concentration o f 0.15 g/m 3 was chosen to ensure stable process performance. A l l simulations were carried out at an H R T of 4 hours to take full advantage of the high rate capacity made available by membrane separation. In the previous sections it was found that changing the aerobic recycle flow has the most effect on the effluent NO3-N concentration and can therefore be used as a manipulated variable to optimize the denitrification process via the application of process control. In this control loop, the aerobic recycle flow is increased (to increase the denitrification load) as long as the anoxic NO3-N concentration is essentially zero and when the NO3-N increases above zero, the recycle flow is decreased (to avoid leakage of NO3-N to the anaerobic zone). More details about this process control loop are given in Section 5.4. The nitrification and B P R processes, on the other hand, were found to be more influenced by the biomass distribution and therefore, a constant aerobic and anoxic recycle ratio of 2 was used in all runs while other parameters were varied. The operating conditions for the simulated runs are presented in Table 5.4. The simulation results were analyzed and the selected recommended biomass distributions for the aerobic and anaerobic zones at different influent V F A / T P ratios are plotted in Figures 5.16, 5.17 and 5.18 for temperatures of 20, 15 and 10 °C respectively. The recommended values selected were found to result in low (less than 0.15 g/m 3) effluent PO4-P and NH4-N concentrations. In cases for which it was not possible to 204 achieve the low effluent concentrations at any biomass distribution, these conditions were not included in the plots o f Figures 5 .16-5 .18 . Examining the results in Figure 5.16 for example shows that at 20 °C, influent V F A / T P ratio o f 5 and SRT o f 10 days, the recommended aerobic biomass fraction is 60% as shown in Figure (a) while the recommended anaerobic biomass fraction is 15% as presented in Figure (b). This leaves an anoxic mass fraction of 25%. According to the design equations given by Rhamphao et al. (2005), this results in an aerobic volume fraction of 31.6%, anoxic fraction of 26.3% and aerobic fraction of 42.1% using an aerobic recycle ratio of 2 and an anoxic recycle ratio of 1. The high anaerobic volume fraction required is due to the low influent V F A / T P ratio assumed for the simulation. The anaerobic volume allows for the production of more V F A via the fermentation process. The general trend of the results in Figures 5.16 to 5.18 show that as the process temperature is decreased to 15 °C, in Figure 5.17, and 10 °C, in Figure 5.18, the recommended aerobic biomass fraction increases and the recommended anaerobic biomass fraction decreases (for the different influent V F A / T P ratios and process SRTs). The results presented in Figures 5.16 to 5.18 give a general idea of the recommended design values for an initial process design or upgrade. It should be noted that at low. process temperatures, the aerobic volume fraction may limit the nitrification process, resulting in an increased effluent NH4-N concentration. Therefore, the seasonal temperature variation needs to be taken into consideration during the initial process design. 205 10 1 5 2 0 S R T ( D a y s ) 2 5 3 0 3 5 (a) c U c o ^ a! w a = o s 2 c — • s < 2 0 18 16 14 12 1 0 8 6 4 2 0 • — • • • • • (X) at O r — u at w 10 1 5 2 0 S R T ( D a y s ) 2 5 3 0 3 5 V F A / T P = 5 • V F A / T P = 1 0 • V F A / T P = 1 5 O V F A / T P = 2 0 (b) Figure 5.16. Recommended biomass distributions for a UCT-type M E B P R process for (a) aerobic zone and (b) anaerobic zone for process H R T = 4 , temperature = 20 °C and different process SRTs and influent V F A / T P ratios to ensure effective P and N removal 206 80 cu s — CM a. ea u 2 75 70 65 b '— Vi CS E 60 55 CU 50 16 10 .2 ^ S=>°8 CS w VF AT/TP = 5 14 • VFA/TP = 10 A VFA/TP = 15 12 -- O VFA/TP = 20 • • • V FA/TP = 5 • VFA/TP = 10 VFA/TP = 15 -e- VFA/TP = 20 10 15 20 SRT (Days) 25 30 35 (a) • • • M • • / X i • • 1 1 1 1 1 1 10 15 20 SRT (Days) 25 30 35 (b) Figure 5.17. Recommended biomass fraction distributions for a UCT-type M E B P R process for (a) aerobic zone and (b) anaerobic zone for process H R T = 4, temperature = 15 °C and different process SRTs and influent V F A / T P ratios to ensure effective P and N removal 207 80 75 a. « a. OS u 70 65 60 a £ 55 « E o 5 .a o 50 45 40 • VFA/TP = 5 • VFA/TP = 10 A VFA/TP =15 10 15 20 SRT (Days) 25 30 35 (a) a. =£ a. aa H s — o E o S o •-a B S? 6 10 15 20 SRT (Days) 25 30 35 (b) Figure 5 .18. Recommended biomass fraction distributions for a UCT-type M E B P R process for (a) aerobic zone and (b) anaerobic zone for process H R T = 4, temperature = 10 °C and different process SRTs and influent V F A / T P ratios to ensure effective P and N removal 208 The following section presents a proposed design and operation guidelines for UCT-type M E B P R processes based on the experimental and simulated results presented in this chapter. 5.4 Proposed Guidelines for the Design and Operation of a UCT-type MEBPR Process Based on the process performance results obtained from the U B C M E B P R pilot plant system and the analysis of the simulation results, a design procedure for a UCT-type M E B P R process that is aimed at resulting in high effluent quality while operating at a high flowrate is proposed. The proposed procedure ensures low effluent N H 4 - N and PO4-P concentrations via using a recommended biomass distribution in the aerobic and anaerobic zones respectively, while low effluent NO3-N concentrations are achieved by manipulating the aerobic recycle flow. Recommendations for process control applications in M E B P R process are discussed in the following section. The proposed guidelines consist of the following steps. 1. Determine the average process temperature and influent V F A / T P ratio that the M E B P R system is to be designed for. Historical sewer data can be used for this purpose. 2. Choose a desired process SRT. The plots in Figure 5.16, 5.17 and 5.18 can be used to obtain an indication of a suitable SRT. 3. Using Figures 5.16, 5.17 and 5.18 for the chosen process temperature, influent V F A / T P and SRT, determine the recommended aerobic and anaerobic sludge mass distributions. 4. Using the design equations presented by Rhamphao et al. (2005), calculate each zone's volume fraction as follows. 209 f vana D r f (l - f - f V mana i maer ) vano D (5.1) f. vaer (a + l ) D mana r where f v a n a and f m a n a are the anaerobic volume and mass fractions respectively, f v ano and fmano are the anoxic volume and mass fractions respectively, f v a e r and f m a e r are the aerobic volume and mass fractions respectively, r is the anoxic recycle ratio and a is the aerobic recycle ratio. reactor. These values can provide a general guideline for the required zone biomass fractions; however it should be noted that variations in the process conditions under dynamic conditions can affect the process performance and may even result in process failures. A s indicated by Shehab et al. (1996), the E B P R process is sensitive to sudden In general, it is recommended to design the system to allow for flexibility in changing the zone volumes depending on the process requirement. For example, multiple smaller tanks can be used in series instead of one big tank to allow for varying the zone volumes by the addition of aeration or introducing a recycle flow. B y introducing this flexibility, the process design can be changed to vary the sludge mass distribution to accommodate seasonal temperature variations for instance. The following section presents the recommendations for process control application in a UCT-type M E B P R process. Then each zone's volume can be determined based on the total volume of the biological disturbances in the system. 210 5.5 Proposed Guidelines for the Application of Process Control for the Operation of a UCT-type MEBPR Process Once the design of the process is completed, the following process control loops can be utilized for the process operation of the UCT-type M E B P R process. These control loops were found to be important based on the experience gained during the operation of the U B C M E B P R pilot plant process. a. It is critical to control the aerobic recycle flow to ensure that the maximum denitrification capacity is utilized. Manual control of the flow was used in operating the U B C M E B P R process, but was found inefficient since the response time o f the control loops is in the order o f hours (Vrecko et al., 2002) and the lack of on-line sensors made it difficult to determine the anoxic NO3-N concentration in a timely manner. Meijer (2004) used the change in the oxidation reduction potential as an indication of increased or decreased NO3-N concentration in the anoxic zone and changed the recycle flow accordingly to optimize denitrification. A simple PI (Proportional and Integral) controller is sufficient for this control loop. b. The data collected from the U B C M E B P R pilot plant process during the period between February and December of 2003 showed that each case of failure in P removal in the process was preceded by a decrease in the influent V F A / T P ratio (results shown in Figure 3.8). It was noticed that the decrease in the influent V F A / T P ratio had a negative effect on the B P R process regardless of the V F A concentration. The process was found to be more stable during periods of relatively constant influent V F A / T P ratio than during periods with varying ratios. Therefore, it is believed that the control of the influent V F A / T P ratio is crucial for a stable B P R process performance. Due to the lack of reliable instrumentation for on-line measurement of V F A and TP concentrations, manual control of this parameter is suggested. A daily composite sample can be collected from the process and used to determine the average daily influent V F A / T P ratio and an additional supply of acetate can be used to adjust the influent V F A concentration accordingly. 211 c. In addition to the aerobic recycle control and the control of the influent V F A / T P ratio, the standard D O control and waste sludge stream are required. A D O set-point of 3 g/m was found to be suitable for the operation of the M E B P R process while the waste stream can be controlled using the total suspended solids (TSS) concentration in the aerobic zone. The set-point for the TSS in the aerobic zone is set to achieve the desired process SRT. 5.6 Summary In general, experimental results and simulation studies have shown that stable process performance in N and P removal can be achieved in a high rate UCT-type M E B P R process under proper process design and operation. The biomass distributions in the aerobic and anaerobic zones were found to be an important design parameter for the nitrification and B P R processes. On the other hand, the aerobic recycle flow can be used to utilize the maximum anoxic zone capacity for denitrification, while avoiding leakage of nitrate to the anaerobic zone. This can be used to reduce the effluent nitrate concentration provided that a reasonable anoxic biomass fraction, that is higher than 20%, is achieved. Maintaining a constant influent V F A / T P ratio was also found to be critical for achieving a stable B P R process. 212 6 CONCLUSIONS AND DIRECTIONS FOR FUTURE RESEARCH The membrane enhanced biological phosphorus removal ( M E B P R ) process for the treatment of municipal wastewater offers a number of advantages over the conventional biological nutrient removal (BNR) process, including a superior effluent quality and a small footprint. A number of studies reported in the literature showed promising results for stable process performance of M E B P R s under different operating conditions resulting in high quality effluent. However, studies focusing on determining the set of design and operating conditions required for effective process operation were scarce. Therefore, the main objective of the current study was to develop guidelines for the design and operation of a UCT-type M E B P R process for stable and efficient operation under high flowrates. The M E B P R process design and operation studies were performed in the simulation environment using a calibrated dynamic T U D P model. The model was calibrated using data collected over six months from the U B C M E B P R pilot plant process. The calibration technique used was based mainly on the S T O W A calibration protocol, however additional steps were carried out to address some deficiencies in the S T O W A protocol. Furthermore, the calibrated model was used to predict the concentration profiles of data collected from the conventional biological phosphorus removal process at U B C where gravity settling was used for the final solids-liquid separation step. This was carried out to identify the differences in the model kinetic parameters of the two systems. Finally the dynamic model was used to perform simulation studies aimed at determining the set of process design and operating conditions that result in a stable process performance for the M E B P R process operating at high flowrates. 6.1 Conclusions The following sections present the conclusions drawn from the different parts of the current study. 213 6.1.1 D y n a m i c C a l i b r a t i o n o f A c t i v a t e d S l u d g e M o d e l s From the dynamic calibration of the T U D P model to predict the measured data collected from the M E B P R process, the following conclusions were drawn. • A dynamic calibration protocol for activated sludge systems, based on the S T O W A protocol, that addresses some of the deficiencies reported for the S T O W A protocol was developed and used to calibrate the T U D P model to fit real process data collected from the M E B P R process at U B C . The developed protocol uses a combination of mathematical methods and expert knowledge in the model calibration process and was found effective in ensuring proper fit o f the measured data. • Using a fixed set of calibration parameters for the activated sludge system as proposed by the S T O W A protocol was found ineffective in predicting the measured data in this study. Results of the current study showed that parameters other than the ones proposed in the S T O W A protocol had to be adjusted to obtain a satisfactory fit of the measured data. • Data collected at steady state or pseudo-steady state are not suitable for the calibration of a dynamic model. A different set of parameters was obtained when calibrating the T U D P model using data collected from the U B C M E B P R process at pseudo-steady state conditions than the set of parameters obtained using dynamic data. The calibrated T U D P model using pseudo-steady state data was not capable of predicting the measured data collected under dynamic conditions. • Using a combination of mathematical methods and process knowledge is essential in the data analysis and parameter selection and estimation stages of the dynamic model calibration of activated sludge models. Relying solely on either one may result in an incorrect set of calibrated model parameters, especially for the inexperienced user. 214 6.1.2 Modeling of MEBPR and CEBPR Processes • The T U D P model was capable of predicting the trend of the measured data collected from the M E B P R process reasonably well after changing only 5 to 6 kinetic parameter values, indicating that activated sludge models developed for the conventional biological activated sludge systems are capable of predicting the process behavior of their membrane counterparts. • Although the general trend of the measured data was reasonably well predicted, the exact concentrations for the anaerobic volatile fatty acids ( V F A ) and the elevated anoxic nitrate and effluent ortho-phosphate were not predicted accurately by the model. Processes of the T U D P model that may require further investigation are anaerobic hydrolysis and fermentation, denitrification and phosphorus uptake. • The T U D P model calibrated using data collected from the M E B P R process was capable o f predicting the concentration profiles o f data collected from the C E B P R process with essentially no change to model parameters, except for the rate of polyphosphate formation, k P P , which was found to be higher in the conventional process. 6.1.3 Guidelines for MEBPR Process Design and Operation • A set o f guidelines was proposed for the design and operation of a UCT-type M E B P R process based on simulation studies and experimental results to achieve low effluent concentration while operating at high flowrates. • Experimental results and simulation studies confirmed that stable process performance in nitrogen and phosphorus removal can be achieved in a UCT-type M E B P R process operated at 4 hours hydraulic retention time, under proper process design and operation. • The biomass distributions in the aerobic and anaerobic zones were found to be an important design parameter for both the nitrification and biological phosphorus 215 removal processes. On the other hand, controlling the aerobic recycle can reduce the effluent nitrate concentration, provided that a reasonable anoxic biomass fraction, that is higher than 20%, is achieved. The aerobic recycle flow can be adjusted to utilize the maximum anoxic zone capacity for denitrification, while avoiding leakage of nitrate to the anaerobic zone. • Maintaining a constant influent V F A / T P ratio was found to be critical for achieving a stable biological phosphorus removal process. Experimental data showed that variations in the influent V F A / T P ratio can disturb the B P R process significantly. Control of the influent V F A / T P ratio is recommended via the addition of external V F A , relative to the incoming TP concentration, to achieve a constant influent V F A / T P ratio in the process. • Guidelines for the application o f practical process control strategies for the stable operation of a UCT-type M E B P R process were proposed based on the experience gained during the experimental phase of the project. 6.2 Research Significance The membrane enhanced biological phosphorus removal process offers many advantages over the conventional E B P R alternative and it has been shown in the present study and in other studies in the literature that superior effluent quality can be achieved using the membrane module for the final solids-liquid separation step. However, information about the effect of differing bioreactor. mass distribution on the process performance under various conditions was missing. This research study was conducted to model real plant data collected from the U B C M E B P R pilot plant data to develop a dynamic model that can be used to explore the design parameters and configuration of the M E B P R process required to achieve stable process performance under various operating conditions. The current research has resulted in a number of contributions to various areas that are of interest to researchers and industrial practitioners. These contributions include the following. 216 • A modified S T O W A calibration protocol was developed and utilized in the modeling stage in the current study. The developed protocol combines expert knowledge with system identification theory to ensure proper model calibration and addresses the various deficiencies of the S T O W A protocol described by Sin et al. (2005). • The calibrated T U D P model for the M E B P R process can serve as a valuable tool in further research into M E B P R systems, such as process control applications. For example, the calibrated T U D P model of the present study has been successfully used to model an E B P R process coupled to a phosphorus recovery system (Srinivas, 2007). • Modeling results of the M E B P R and the C E B P R processes presented in the present study showed that no significant difference exists in the process kinetics of the two systems which indicates that the membrane unit does not impact the process performance as was first anticipated. Knowing that, research efforts can now be focused on ways to further develop the M E B P R process for high flowrate operations utilizing the full capacity of the membrane system. • The proposed guidelines for the design and operation of a UCT-type M E B P R and for the application of process control provide useful information about the M E B P R process operation under high flowrates for new researchers and industrial practitioners. These guidelines provide a platform that can be built on for future process development. 6.3 Research Needs The results of the current study have generated new ideas for further research work that was considered out o f the scope o f the thesis. Below is a list o f research needs that can further improve the knowledge about the M E B P R process and ways o f enhancing its design. • The guidelines proposed in the current study for the design and operation of a U C T -type M E B P R form a good starting point for new system design or an upgrade to an 217 existing wastewater treatment process. It would be interesting to apply the proposed guidelines to modify the design of the U B C M E B P R pilot plant process to verify the proposed conditions in an experimental environment. • In the current study, the UCT-type configuration of the M E B P R was tested. However, the developed simulation model can be used to examine different process configurations for the M E B P R process. Using the UCT-type configuration for the M E B P R process in the current study resulted in operation challenges due to the high nitrate concentrations in the anoxic zone and its negative effect on the phosphorus release process in the anaerobic zone. In an attempt to decouple the nitrate concentration in the anoxic zone and its affect on the anaerobic zone through the recycle flow, a two-stage anoxic bioreactor as presented in Figure 6.1 can be tested in a simulation environment to further develop the M E B P R process design and operation. Anoxic Recycle Influent Anaerobic l i l l lllfll^  Aerobic ^ \ w • • Aerobic (Nitrate) Recycle ^ Sludge Waste Figure 6.1. Different Process Configuration for an M E B P R Process • The simulation studies carried out in the current studies were performed by simulating the M E B P R process under steady state conditions. Steady state simulations were used to study the long term effect of a certain process design under specific operating conditions. However the calibrated model for the M E B P R process is a dynamic model and can therefore be used to perform dynamic simulation studies 218 by introducing daily load and flow variations to further develop the process performance under dynamic conditions. 219 R E F E R E N C E S Adam C , Gnirss R., Lesjean B . , Buisson H . and Kraume M . (2002) Enhanced biological phosphorus removal in membrane bioreactors. Water Science and Technology 46 (4-5), 281-286. A h n K . H . , Song K . G. , Cho E. , Cho J., Y u n H . , Lee S. and K i m J. (2003). Enhanced biological phosphorus and nitrogen removal using a sequencing anoxic/anaerobic membrane bioreactor ( S A M ) process. Desalination 157, 345-352. A h n Y . T., Choi Y . K . , Jeong H . S., Chae S. R. and Shin H . S. (2006). Modeling o f extracellular polymeric substances and soluble microbial products production in submerged membrane bioreactor at various STRs. Water Science and Technology 53(7), 209-216. Alex J., Beteau J. F., Copp J. B . , Hellinga C , Jeppsson U . , Mars i l i -Libel l i S., Pons M . N . , Spanjers H . and Vanhooren H . (1999). Benchmark for evaluation control strategies in wastewater treatment plants. Proc. European Control Conference (ECC'99), Karlsruhe, Germany. August 31 - September 3. Al -Malack M . H . (2006). Determination of biokinetic coefficients of an immersed membrane bioreactor. Journal of Membrane Science 271(1-2), 47-58. A P H A - A W W A - W E F (1998). Standard methods for the examination o f water and wastewater. 20 t h Ed. , Washington, D . C . A Q U A S I M (2005). E A W A G , Swiss Federal Institute for Environmental Science and Technology, http://www.aquasim.eawag.ch. Baetens D. , Vanrolleghem P. A . , van Loosdrecht M . C. M . and Hosten L . H . (1999). Temperature effects in bio-P removal. Water Science and Technology 39(1), 215-225. Barker P. S. and Dold P. L . (1997). General model for biological nutrient removal activated-sludge systems: model presentation. Water environment research 69(5), 969-984. Brdjanovic D . (1998). Modeling biological phosphorus removal in activated sludge systems. PhD Thesis, Delft University of Technology, The Netherlands. Brdjanovic D. , van Loosdrecht M . C. M . , Versteeg P., Hooijmans C. M . , Alaerts G . J. and Heijen J. J. (2000). Modeling C O D , N and P removal in a full-scale W W T P Haarlem Waarderpolder. Water Research 34(3), 846-858. Brun R. and Reichert P. (2001). Practical identifiability analysis of large environmental simulation models. Water Resources Research 37(4), 1015-1030. 220 Brun R., Kuhni M . , Siegrist H . , Gujer W . and Reichert P. (2002). Practical identifiability o f A S M 2 d parameters - systematic selection and tuning o f parameters subsets. Water Research 36,4113-4127. Carrette R., Bix io D . , Thoeye C. and Ockier P. (2001). Full-scale application of the I A W Q A S M No. 2d Model . Water Science and Technology 44(2-3), 17-24. Canales A . , Pareilleux A . , Rols J. L . , Goma G . and Huyard A . (1994). Decreased sludge production strategy for domestic wastewater treatment. Water Science and Technology 30(8), 97-106. Chang W . C , Choiu R. J. and Ouyang C. F. (1996). The effect o f residuals substrate utilization on sludge settling in an enhanced biological phosphorus removal process. Water Science and Technology 34(1-2), 425-430. Choi Y . , Shin E. B . and Lee Y . D . (1996). Biological phosphorus removal from wastewater in a single reactor combining anaerobic and aerobic conditions. Water Science and Technology 34(1-2), 179-186. Cicek N . (2003). A review of membrane bioreactors and their potential application in the treatment o f agricultural wastewater. Canadian Biosystems Engineering 45, 6.37-6.49 Cicek N . , Dionysiou D. , Suidan M . T., Ginestet P., and Audic J. M . (1999). Performance deterioration and structural changes of a ceramic membrane bioreactor due to inorganic abrasion. Journal of Membrane Science 163(1), 19-28. Cote P. and Thompson D. (2000). Wastewater treatment using membranes: the North American experience. Water Science and Technology 41(10-11), 209-215. Cote P., Buisson H . and Praderie M . (1998). Immersed membranes activated sludge process applied to the treatment of municipal wastewater. Water Science and Technology 38(4-5), 437-442. Davies W . J., Le M . S. and Heath C. R. (1998). Intensified activated sludge process with submerged membrane microfiltration. Water Science and Technology 38 (4-5), 421-428. De Silva D . G . V . , Urbain V . , Abeysinghe D . H . and Rittmann B . E . (1998). Advanced analysis of membrane bioreactor performance with aerobic-anoxic cycling. Water Science and Technology 38(4-5), 505-512. Dochain D . and Vanrolleghem P. (2001). Dynamical modeling and estimation in wastewater treatment processes. I W A Publishing, London, U K . Engelhardt N . , Firk W . and Warnken W . (1998). Integration of membrane filtration into the activated sludge processing municipal wastewater treatment. Water Science and Technology 38(4-5), 429-436. 221 Fan X . J., Urbain V . , Qian Y . and Manem J. (1996). Nitrification and mass balance with a membrane bioreactor for municipal wastewater treatment. Water Science and Technology 34(1-2), 129-136. Fleischer E . J., Broderick T. A . , Daigger G . T., Fonseca A . D . Ffolbrook R. D . and Murthy S. N . (2005). Evaluation of membrane bioreactor process capabilities to meet stringent effluent nutrient discharge requirements. Water Environment Research 77(2), 162-178. Furumai H . and Rittmann B . E . (1992). Advanced modeling of mixed populations of heterotrophs and nitrifiers considering the formation and exchange of soluble microbial products. Water Science and Technology 26(3-4), 493-502. Furumai H . , Kazmi A . A . , Furuya Y . and Sasaki K . (1999). Effect of sludge retention time (SRT) on nutrient removal in sequencing batch reactors. Journal of Environmental Science and Health 34(2), 317-328. Galluzzo M . , Ducato R., Bartolozzi V . and Picciotto A . (2001). Expert control of D O in the aerobic reactor of an activated sludge process. Computers and Chemical Engineering 25 ,619-625. Gehlert G . and Hapke J. (2003). Rigorous approach to modeling of a continuous aerobic membrane bioreactor. Engineering in Life Sciences 3(7), 292-298. Gernaey D . V . , van Loosdrecht M . C. M . , Henze M . , Morten L . , and Jorgensen S. B . (2004). Activated sludge wastewater treatment plant modeling and simulation: state of the art. Environmental Modeling & Software 19, 763-783. Gernaey K . , Vanrolleghem P. A . and Verstraete W . (1998) On-line estimation of Nitrosomonas kinetic parameters in activated sludge samples using titration in-sensors-experiments. Water Research 32, 71-80. Grelier P., Rsenberger S. and Tazi-Pain A . (2006) Influence of sludge retention time on membrane bioreactor hydraulic performance. Destination 192(1 -3_, 10-17. Gujer W. , Henze M . , Mino T. and van Loosdrecht M . (1999). Activated sludge model No . 3. Water Science and Technology 39(1), 183-193. Gujer W. , Henze M . , Mino T., Matsuo T., Wentzel M . C. and Marais G . v. R. (1995). The activated sludge model No . 2: biological phosphorus removal. Water Science and Technology 31(2), 1-11. Hellinga C. (1992). Macrobal 2.02. Delft University of Technology. The Netherlands. Download: http://www.bt.tudelft.nl/content/download/macrobal.html. 222 Henze M , Grady C. P. L . , Gujer W. , Marais G . v. R. and Matsuo T. (1987). Activated sludge model N o . 1. I A W Q Scientific and Technical Report No . 1. I A W Q , London Henze M . , Gujer W. , Mino T. and Loosdrecht M . (2000). Scientific and technical report No. 9: Activated sludge models A S M 1 , A S M 2 , A S M 2 d and A S M 3 . IWA Task Group on Mathematical Modeling for Design and Operation of Biological Wastewater Treatment. I W A Publishing, London, U K . Henze M . , Gujer W. , Mino T., Matsuo T., Wentzel M . C. and Marais G . v. R. (1994). Activated sludge model No . 2. I A W Q Scientific and Technical Report No . 3. I A W Q , London. Henze M . , Gujer W. , Mino T., Matsuo T., Wentzel M . C. , Marais G . v. R. and van Loosdrecht M . C. M . (1999). Activated sludge model No 2d, A S M 2 d . Water Science and Technology 39(1), 165-182. H u Z. , Wentzel M . C . and Ekama G A . (2002). Anoxic growth of phosphate-accumulating organisms (PAOs) in biological nutrient removal activated sludge systems. Water Research 36(19) 4927-4937. H u Z. , Wentzel M . C . and Ekama G . A . (2003). Modell ing biological nutrient removal activated sludge systems - a review. Water Research 37(14), 3430-3444. Hulsbeek J. J. W. , Kruit J., Roeleveld P. J. and Loosdrecht M . C. M . (2002). A practical protocol for dynamic modeling of activated sludge systems. Water Science and Technology 45(6), 127-136. Ingildsen P. (2002a). Realising full-scale control in wastewater treatment systems using in situ nutrients sensors. PhD Thesis, Lund University, Sweden. Ingildsen P., Jeppsson U . and Olsson G . (2002b). Dissolved oxygen controller based on on-line measurements of ammonium combining feed-forward and feedback. Water Science and Technology 45(4-5), 453-460. Jiang T. (2007) Characterization and Modell ing of Soluble Microbial Products in Membrane Bioreactors. PhD thesis, Ghent University, Belgium. Jiang T., L i u X . , Kenndy M . D . , Schippers, J. C . and Vanrolleghem P. A . (2005) Calibrating a side-stream membrane bioreactor using Activated Sludge Model No. 1. Water Science and Technology 52(10-11), 359-367. Koch F. (2002). Personal communications Kuba T., Murnleitner E. , van Loosdrecht M . C. M . and Heijnen J. J. (1996). A metabolic model for the biological phosphorus removal by denitrifying organisms. Biotechnology and Bioengineering 52, 685-695. 223 Kuba T., Wachtmeister A . , van Loosdrecht M . C. M . and Heijnen J. J. (1994). Effect of nitrate on phosphorus release in biological phosphorus removal systems. Water Science and Technology 30(6), 263-269. Lee T. T, Wang F. Y . and R. B . Newell (2004). Robust multivariable control of complex biological processes. Journal of Process Control 14, 193-209. Lee, T. T., Wang, F. Y . , Islam, A . , and Newell , R. B . (1999). Generic distributed parameter model control of a biological nutrient removal (BNR) activated sludge process. Journal of Process Control 9, 505-525. Lesjean B . , Gnirss R. and Adam C. (2002). Process configuration adapted to membrane bioreactors for enhanced biological phosphorus and nitrogen removal. Desalination 149, 217-224. Lesjean B . , Gnirss R., Adam C , Kraume M . and Luck F. (2003). Enhanced biological phosphorus removal process implemented in membrane bioreactors to improve phosphorus recovery and recycling. Water Science and Technology 48(1), 87-94. Lindberg C. F. and Carlsson, B . (1996). Nonlinear and set-point control of the dissolved oxygen concentration in an activated sludge process. In: Water Quality International '96, Singapore, Conf. Pre-print Bk. 3 Wastewater Treatment Plants and Sewerage Systems, 203-210. L i u W . T., Marsh T. L . and Froney L . J. (1998). Determination of the microbial diversity of anaerobic-aerobic activated sludge by a novel molecular biological technique, Water Science and Technology 37(4-5), 417-422. Ljung L . (1999). System identification - theory for the user, Prentice-Hall, N J . L u S. G , Imai T., Ukita M . , Sekine M . , Higuchi T. and Fukagawa M . (2001). A model for membrane bioreactor process based on the concept of formation and degradation of soluble microbial products. Water Research 35, 2038-2048. Mamais D . and Jenkins D . (1992). The effects o f M C R T and temperature on enhanced biological phosphorus removal. Water Science and Technology 26(5-6), 955-965. Manem J. and Sanderson R. (1996). Membrane bioreactors. In: Water Treatment Membrane Processes (Mallevialle, J. Odendaal, P., and Wiesner, M . R . , Eds.) McGraw-H i l l , New York, 17.1-17.31. Manoharan R. (1988). Readily Biodegradable C O D as an indicative parameter in estimating the efficacy of a sewerage for biological excess phosphorus removal. PhD Thesis, Univ. British Columbia, Vancouver, Canada. 224 Manser R., Gujer W . and Siegrist H . (2004). Membrane bioreactor vs. conventional activated sludge system - population dynamics of nitrifiers. Proc. 4th IWA World Water Conference. Marrakech, Marocco, Sept. 19-24. Manser R., Gujer W . and Siegrist H . (2005a). Membrane bioreactor versus conventional activated sludge system: population dynamics of nitrifiers. Water Science and Technology 52(10-11), 417-425. Manser R., Gujer W . and Siegrist H . (2005b). Consequences of mass transfer effects on the kinetics of nitrifiers. Water Research 39(19), 4633-4642. Manser R., Gujer W . and Siegrist H . (2006). Decay processes of nitrifying bacteria in biological wastewater treatment systems. Water Research 40(12), 2416-2426. Masse A . , Sperandio M . and Cabassud C. (2006) Comparison of sludge characteristics and performance of a submerged membrane bioreactor and an activated sludge process at high solids retention time. Water Research 40(12), 2405-2415. McCue T., Naik R., Zepeda M . , L i u Y . - H . , Vassiliev I. and Randall A . A . (2004). Changes in anoxic denitrification rate resulting from prefermentation of a septic, phosphorus-limited wastewater. Water Environment Research 76(1) 23-28. McCue T., Shah R., Vassiliev I., L i u Y . - H . , Eremektar F. G . , Chen Y . and Randall A . A . (2003). Evaluation of influent prefermentation as a unit process upon biological nutrient removal. Water Science and Technology 47(1), 9-15. Meijer S. C. F. (2004). Theoretical and practical aspects o f modeling activated sludge processes. PhD Thesis, Technical University of Delft, Delft, The Netherlands. Meijer S. C. F., van Loosdrecht M . C. M . , and Heijnen J. J. (2001). Metabolic modeling of full-scale biological nitrogen and phosphorus removing W W T P ' s . Water Research 35(11), 2711-2723. Meijer S. C. F., van Loosdrecht M . C. M . and Heijnen J. J. (2002a). Error diagnostics and data reconciliation for activated sludge modeling using mass balances. Water Science and Technology 45(6), 145-156. Meijer S. C. F., van Loosdrecht, and Heijnen J. J. (2002b). Modell ing the start-up of a full-scale biological phosphorus and nitrogen removing W W T P . Water Research 36(19), 4667-4682. Meinhold J., Arnold E . and Issacs S. (1999). Effect of nitrite on anoxic phosphate uptake in biological phosphorus removal activated sludge. Water Research 33(8), 1871-1883. 225 Melcer H . , Dold P. L . , Jones R. M . , Bye C. M . , Takacs I., Stensel H . D . , Wilson A . W. , Sun P. and Bury S. (2003). Methods for wastewater characterization in activated sludge modeling. Water Environment Research Foundation (WERF) , Alexandria, V A . U S A . Metcalf and Eddy (2003). Wastewater Engineering: Treatment and Reuse. 4 t h Ed. , M c G r a w - H i l l , New York. Monti A . (2006a) A comparative study of biological nutrient removal processes with gravity and membrane solids-liquid separation. Ph.D. Diss. Environmental Engineering Program, Department of C i v i l Engineering, The University of British Columbia, Vancouver, Canada. Monti A . , Hal l E . , Dawson R. N . , Hussain H . and Kel ly H . G . (2006b). Comparative study o f biological nutrient removal (BNR) processes with sedimentation and membrane-based separation. Biotechnology and Bioengineering 94(4), 740-752. Mulkerrins D . , Dobson A . D . W . and Colleran E . (2004). Parameters affecting biological phosphate removal from wastewaters. Environment international 30(22), 249-259. Murnleitner E . , Kuba T., van Loosdrecht M . C. M . and Heijnen J. J. (1997). A n integrated metabolic model for the aerobic and denitrifying biological phosphorous removal. Biotechnology and Bioengineering 54, 434—450. O 'Re i l ly J. and Leithead W . E . (1991). Multivariable control by "individual control design". International Journal of Control 54(1), \—46. Ogunnaike B . A . and Ray W . H . (1994). Process dynamics, modeling, and control. Oxford Press. New York, N Y . Okada M . , L i n C . K . , Katayama Y . and Murakami A . (1992). Stability of phosphorus removal and population of bio-P-bacteria under short term disturbances in sequencing batch reactor activated sludge process. Water Science and Technology 26(3-4), 483-491. Oldham W . K . and Rabinowitz B . (2001). Development of biological nutrient removal technology in western Canada. Canadian Journal of Civil Engineering 28(1), 92-101. Olsson G . and Newell , B . (1999). Wastewater Treatment Systems - Modeling, Dagnosis and Control, I W A Publishing, London, U K . Patel J., Nakhla G . and Margaritis A . (2005). Optimization of biological nutrient removal in a membrane bioreactor system. Journal of Environmental Engineering 131(7) 1021-1029. Pattarkine V . M . and Randall C. W . (1999). The requirement of metal cations for enhanced biological phosphorus removal by activated sludge. Water Science and Technology 40(2), 159-65. 226 Penya-Roja J. M . , Seco A . , Ferrer J. and Serralta J. (2002). Calibration and validation of activated sludge model No.2d for Spanish municipal wastewater. Environmental Technology 23, 849-862. Pochana K . and Keller J. (1999). Study of factors affecting simultaneous nitrification and denitrification (SND). Water Science and Technology 39(6), 61-68. Ramphao M . , Wentzel M . C , Merritt R., Ekama G . A . , Young T. and Buckley C A . (2005). Impact of membrane solid-liquid separation on design of biological nutrient removal activated sludge systems. Biotechnology and Bioengineering 89(6), 630-646. Reichert P. (1998). A Q U A S I M 2.0 - User Manual. Swiss Federal Institute for Environmental Science and Technology, Dubendorf, Switzerland. Rieger L . ; K o c h G. , Kuhni M . , Gujer W . and Siegrist H . (2001). The E A W A G Bio-P module for activated sludge model No . 3. Water Research 35(16), 3887-3903. Roeleveld P. J. and van Loosdrecht M . C. M . (2002). Experience with guidelines for wastewater characterization in the Netherlands. Water Science and Technology 45(6), 77-87. Saito T., Brdjanvoic D . and van Loosdrecht M . C. M . (2004). Effect of nitrite on phosphate uptake by phosphate accumulating organisms. Water Research 38, 3760-3768. Samuelsson P. and Carlsson B . (2002). Control of the aeration volume in an activated sludge process for nitrogen removal. Water Science and Technology 45(4-5), 45-52. Shehab O., Deininger R., Porta F., and Wojewski T. (1996). Optimizing phosphorus removal at the A n n Arbor wastewater activated sludge system. Water Science and Technology 37(4-5), 263-270. Siegrist H . , Brunner I., Koch G. , Phan L . C. and Le V . C. (1999). Reduction of biomass decay rate under anoxic and anaerobic conditions. Water Science and Technology 39(1), 129-137. Sin G . , Van Hulle S. W . H . , De Pauw D . J. W. , Van Griensven A . and Vanrolleghem P. A . (2005). A critical comparison of systematic calibration protocols for activated sludge models: A S W O T analysis. Water Research 39(12) 2459-2474. Smolders G . L . F., van der Meij J., van Loosdrecht M . C. M . and Heijnen J. J. (1994a). Model of the anaerobic metabolism of the biological phosphorus removal process: stoichiometry and p H influence. Biotechnology and Bioengineering 43(3), 461-470. 227 Smolders G . L . F., van der Meij J., van Loosdrecht M . C M . and Heijnen J. J. (1994b). Stoichiometric model of the aerobic metabolism of the biological phosphorus process. Biotechnology and Bioengineering 44(7), 837-848. Soderstrom, T. and Stoica, P. (1989). System identification, Prentice-Hall, Cambridge, U K . Sperandio M , Masse A , Espinosa-Bouchot M . C. and Cabassud C. (2005) Characterization of sludge structure and activity in submerged membrane bioreactor. Water Science and Technology 52(10-11), 401-408. Srinivas H . (2007). Coupling of phosphorus recovery to an enhanced biological phosphorus removal process through a sidestream: a pilot scale study. M . A . S c . Thesis, Environmental Engineering Program, Department of C i v i l Engineering, The University of British Columbia, Vancouver, Canada. Tay J. H . , Zeng J. L . and Sun D . D . (2003). Effects of hydraulic retention time on system performance of a submerged membrane bioreactor. Separation Science and Technology 38(4), 851-868. Urbain V . , Mobarry B . , de Silva V . , Stahl D. A . , Rittmann B . E . and Manem J. (1998). Integration of performance modular biology and modeling to describe the activated sludge process. Water Science and Technology 37, 223-229. V a n Veldhuizen H . M . , van Loosdrecht M . C. M . and Heijnen J. J. (1999). Modeling biological phosphorus and nitrogen removal in a full-scale activated sludge process. Water Research 33(16), 3459-3468. Vanrolleghem P. A . , Insel G , Petersen B . , Sin G. , De Pauw D. , Nopens I., Dvermann H . , Weijers S. and Gernaey K . (2003). A comprehensive model calibration procedure for activated sludge models: Proc. 76th Annual Technical Exhibition and Conference, Water Environment Federation (WEFTEC), Los Angeles, U S A , October 11-15. Vanrolleghem P. A . , Sin G . and Gernaey K . (2004). Transient response of aerobic and anoxic activated sludge activities to sudden concentration changes. Biotechnology and Bioengineering 86, 277-290. Vrecko D. , Hvala N . and Kocijan J. (2002). Wastewater treatment benchmark: what can be achieved with simple control?. Water Science and Technology 45(4-5), 127-134. Vrecko D. , Hvala N . , Kocijan J. and Zee M . (2001) System analysis for optimal control of a wastewater treatment benchmark. Water Science and Technology 43(7), 199-206. Weijers S. R. and Vanrolleghem P. A . (1997). A procedure for selecting best identifiable parameters in calibrating activated sludge model No. 1 to full-scale plant data. Water Science and Technology 36(5), 69-79. 228 Weijers S. R., (2000). Modeling, identification and control of activated sludge plants for nitrogen removal, PhD thesis, Eindhoven University of Technology, The Netherlands. Wen X . H . , X i n g C. H . and Qian Y . (1999). A kinetic model for the prediction of sludge formation in a membrane bioreactor. Process Biochemistry 35, 249-254. X i n g C. H . , W u W . Z. , Qian Y . and Tardieu E . (2003). Excess sludge production in membrane bioreactors: a theoretical investigation. Journal of Environmental Engineering 291-297. Zhang Z . and Hal l E . R. (2006). Heterotrophic kinetic parameter estimation for enhanced biological phosphorus removal processes operated in conventional and membrane -assisted modes. Water Quality Research Journal of Canada 41(1), 72-83. 229 A p p e n d i x I - T h e T U D P M o d e l R a t e E q u a t i o n s a n d P a r a m e t e r s Table 1.1. Stoichiometric matrix of the T U D P model (Meijer, 2004) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Component -> So S F S A S N U S N O S N 2 Spo Si S H C O X I Xs X H XpAO Xpp XpHA X G L Y X A X T S S <— Process HE o JS-CS O U -oo "E Q O CJ 00 "E z 00 "E z 00 "E z 00 "E oo "E 3 o u 00 "E "E "a D o o c 00 "E Q O U 00 "E 3 o o oo gCOD/m' "E a; oo "E 3 O 00 "E "E -3 a o o u o 00 oo "E "ob Aerobic gCODxs/d 1-fsi fs. 1 r° Hydrolysis CN.l Cp.l Ce,l -1 CTSS.l 2 r h» Anoxic Hydrolysis gCODxs/d 1-fsi C N . I Cp.l fsi ce,i -1 C T S S . I 3 rh*> Anaerobic Hydrolysis gCODxs/d 1-fsi C N . I Cp.l fs. cc.i -1 CTSS.l 6 r" Regular Heterotrophic Organisms Xu 4 & Growth on SF ^ C O D ™ / d " " Z ^ " ' ^ Aerobic 5 rs° Growth on gC0DX H/d -(1/YH-1) S A NO Anoxic S F Growth on Sp Anoxic 7 Growth on gC0DX H/d S A Fermentation gCODsp/d -1 Heterotrophic Lysis CN.4 -1/Y„ C N , 5 gCODxn/d -1/Y„ CN.6 -A N ' H L gC0D x„/d " ! / Y H C N , 7 1 C N , 8 CN.9 C/Y H -1) (1/YH -1) 2.86 2.86 (1/YH-1) (1/Y„ -1) 2.86 2.86 Cp,4 C e,4 Cp.5 C e,5 Cp.6 C e , 6 Cp,7 C e.7 Cp.8 Cp,9 Ce.9 fxi .H p t> 1-Xl ,H CTSS.5 CTSS.6 CTSS.7 CTSS.8 CTSS.9 Phosphorus Accumulating Organisms Xp A O L N A N Anaerobic , , 1 0 r - Storage of SA ZCOD^/d 11 R A N Anaerobic M Maintenance & NO Anoxic „ „ ,, 1 2 Storage of SA g C ° D s A / d 13 r, Anoxic PHA I - H A Consumption g C O D ™ ,/d Anoxic 14 rpN° Storage of gP/d poly-P Anoxic 15 rGN°Y Glycogen gCODG L Y/d Formation 1 6 C S n a n c e S C 0 D ^ d Aerobic PHA Consumption Aerobic 18 r° Storage of gP/d poly-P Aerobic 19 r°L Y Glycogen gC0DG L Y/d l - l /Y 0 ° L T Formation Aerobic -1/Y° C N . 1 4 C N . 1 5 CN.16 CN.17 C N . I 8 C N . 1 9 (1-YST) (1-YSNA0) 2.86 2.86 U - 1 / Y „ N ° ) (1-1/Y P N„°) 2.86 2.86 (1/YpT) '(l/YPNp°) 2.86 2.86 O / Y Z - l ) (l/Y 0 N L 0v-D 2.86 2.86 -1/2.86 1/2.86 Y P 0 c e. lO 1 C e , l l Y P 0 C=.12 Cp,13 Ce.13 Cp,14 C e , H Cp,15 C„,15 / A N V A N -Y„ Cp.16 Cp,17 C e , l6 C e , l 7 2 0 $ Maintenance g C 0 I W d -1 Cp.18 C e,]8 Cp,19 C e,]9 Cp,20 C e,20 1/YP^A -l/YpN P° - l / Y N 0 V G L Y -1 1 /Y P °„A -l/Yp°P - l /Y° V G L Y -1 CTSS.18 Autotrophic Nitrifying Organisms X A 2 1 G r o l ° P h i C g C 0 D - / d ' - 4 - 5 7 / Y * 22 Autotrophic Lysis gCODxA/d CN.21 CN.22 1/YA Cp,21 Cp.22 Cc.21 Ce.22 fx] 1 CTSS.21 -1 CTSS.22 231 CN.I = 'N.XS 'N.SI ' *si 'N_S i • • • ( " - » ; , ) CP.I = 'l-.xs ~ 'p.SI ' 'si.ll ~ 'p.SI (1- »;,.„ > c«.i = c N , / I 4 -cN.4 = i / Y - i CP4 = 'p.SI.-/Y„ - i | . , | M Co.4 = c N . 4 / l 4 CNo = ~ ' N . I I M C P . S = ~ ' P . B M CN.„ = 'N.SI ^ Y n ~ 'N.IIM cp.r. = W / Y,, - i | , „ M c . . (. = cN.f,/14 < V T = ~ ' N . H M CP.7 = - , P , B M Cc.7 =cN.7/14 ^N.S = 'N.SI- C P . X = ' p . S P CC . « = < V K / I 4 -= ' N . U M ~ 'N.XI ' X^Ul ~ 'N.XS ' (' — f xu i ) CP.!> = 'p.liM ~ 'P.XI ' fxi.ll ~ 'P. . cc..> =<=„..,/'4" CN.I» = 0 c - Y A N P^.Hi — 1 PO W'.lll - 'po I C N . I I ~ 0 CP.I1 = 1 c c j l =-1.5/31 CN..: = 0 c - Y ™ *"P.I2 "PO ^ : = -Y,r,'-( CNJ? = - i / YN O c - - i / Y N U UP.I.I ' P I I M ' "PICA C«.I3=CN.I.1/I4 CN.I4 = i / YN n ' N . U M ' 1 t'P S'.I4 - 'p .BM ' 'pp 1 C».i4 =cN J J/14 C N . I 5 = : / V N ( ) *N.HM ' *(il.V C _ : / V N ( ) Co.l5=CN., i/14 N^.lfi = 'N.IIM Cp.ir> - 'p .BM Cc.l« =CN.,f,/14 = N.P7 = - i / Y ° 'N.11M ' 1 PIJ.-\ C P . | i = 'p.DM ' Y,"LA cc.i7 =cN 1 7/14 C N . I K = */Y° N.UM ' M'P C P . U = 'p .BM ' Y,", - 1 C ^.IH ~ C N . I X ^4 C N . W = N.IIM ^Ci.V CP.1V = 'p .BM ' Y G I y • . C . . W = CN.,,/14 CN.2() = N.QM CP.2<) — 'p .BM Cc.2i) = CN.20 C N . : I = ~ ' N . U M ~ CP.2I = ~ ' p . B M C,.2, =cN.,,/14-C N . : : = N.UM ~ ' N . , \ I ' f.\I.A ~ N.XS ' 0 ~ fxi.A ) C P . ; : = 'P .BM _ 'p.xi ' fxi.A ~ 'p.x s-d- - f . X , . A ) c c „ = c N „ / 1 4 C I S.S.I = ~' ISS.XS crss .4 = •ISS.BM CISS.5 = 1 ISS.UM C ISS.fi ~~ 1 ISS.ItM CISS.7 =1 11'SS.UM c iss.y _ 1 issjti "fxi.it + 'TSS.XS 0 *" fxui ) ' c iss.w - ' I S S . P H ' ( ~ Y m ) + i |ss.nu '^SA ' CiSS.II ~ ~LTSS,t'P C I S S . I : - ' i s s . i ' i - ' ( " Y m ^ + 'iss.i'UA ' Y s A ° c = i / v N n _ i ISS.n 1 ISS.UM / 1 PHA 'ISS.PIIA ' ISS.UM ' , S M . , V ( l - Y ^ ) 4 'ISS.BM/ 1 PP T11SS.PP >— 'rss.HM/ 1 ni.v 'iss.r.i.Y ^ = —'t.SS.UM = i / Y " - i ! 'tSS.HM/ . PHA 'iSS.PUA = - i / Y " +i ! MSS.BM/ 'PP '^fSS.PP - - i / Y ( > +i Table 1.3. Component composition factors (Meijer, 2004) Component composition factors i N , i P and i T S S / Nitrogen content of inert soluble COD, Si iN.SI 0.01 gN-g'CODsi Henze et dl., 1999; ASM2d 2 Nitrogen content of soluble substrate, SF lN,SF 0.03 gN-g 'CODSF " 3 Nitrogen content of inert particulate COD, X, tN.XI 0.03 gN-g 'COD X I Meijer, 2004 4 Nitrogen content of particulate substrate, X s 1N.XS 0.03 gN-g" 'CODxs Henze et al., 1999; ASM2d 5 Nitrogen content of biomass, X H , X P A O , X A U T 1N.BM . 0.07 gN-g" C O D B M 6 Phosphorus content of inert soluble COD, Si iP,si 0 gPg" 'CODs, " 7 Phosphorus content of soluble substrate, SF !P,SF 0.01 gP-g CODSF " 8 Phosphorus content of inert particulate COD, Xi ip.xi 0.01 gP-g CODx, " 9 Phosphorus content of particulate substrate, X s 1P,XS 0.01 gP-g' CODxs " • 10 Phosphorus content of biomass, X H , X P Ao, X A U T . 1P,BM 0.02 gP-g- C O D B M 11 Ratio Total Suspended Solids to X| 1TSS.XI 0.75 gTSS- r 'COD X I 12 Ratio Total Suspended Solids to X s 'TSS.XS 0.75 gTSS-j 5'CODxs " 13 Ratio Total Suspended Solids to biomass X H , X P A 0 , X A U T (CH 209O 0 5 4 N 0 2 P 0 0 ] 5 ) N 1TSS.BM 0.90 gTSS-g ' C O D B M Smolders etal., 1994b 14 Ratio Total Suspended Solids to X P P ( M G L 2 ; 3 K ^ P O , ) „ iTSS.PP 3.23 gTSS-g'PPP Smolders el al., 1994a 15 Ratio Total Suspended Solids to X P HA (C<H 60 2) 1 M iTSS,PHA 0.6 gTSS-g - 'COD P H A Doi, 1990 16 Ratio Total Suspended Solids to XOLY ( C 6 H 1 0 O 5 ) „ 6 1TSS.GLY 0.84 gTSS-g ' C O D G L Y Stryer, 1975 17 Ratio COD to Oxygen (S0) 'COD.O -1 g C O D g ' 0 2 18 Ratio COD to Nitrate (SN 0) ICOD.NO -2.86 gCOD •g_,NsNO 233 Table 1.4. Stoichiometric parameters of the T U D P model (Meijer, 2004) Hydrolysis of Particulate Substrate / Fraction of inert COD generated in hydrolysis fs. 0 gCOD s,-g"1COD(xH +xpAO) Henze el al., 1999; ASM2d Regular Heterotrophic Organisms X H / Heterotrophic yield for growth on substrate 2 Fraction of. inert COD generated in biomass lysis Y H fxi.H 0.63 0.10 gCODxHg'COD gCODxi-g'CODxH Henze eta!., 1999; ASM2d Autotrophic Nitrifying Organisms X A / Autotrophic yield for growth 2 Fraction of inert COD generated in biomass lysis Y A fxi.A 0.24 0.10 gCODxA-g'NsNH gCODxi-g'CODxA Henze et al., 1999; ASM2d Phosphorus Accumulating Organisms X P A O / ATP produced per NADH or P/O ratio 2 Observed biomass ratio TOC over COD 5 a 1.85 0.334 moleATP-mole'NADH gCpAO-g'CODpAO Smolders et al, 1994b 4 Anaerobic yield for phosphate release 1 PO 0.184x pH- 0.94 « 0.35 gPsro-g-'CODsA Smolders et al., 1994a 5 Yield for anaerobic formation of PHA from SA ! S A 1.50 gCODpuB-g'CODsA 6 Observed yield for anoxic phosphate release Y N O 1 PO 0.23 gPsrog'CODsA Kuba etal, 1994 7 Yield for anoxic formation of PHA from SA vN ° ' S A ( 8 . 3 - Y P N 0 ° - 4 . 9 + 8-5) ' 9-5 . g COD P HBg 'CODsA Meijer, 2004 8 Anoxic yield for degradation of X P H B 1 PHA 19 + 3 % — = 1.72 4 + 9-5 g COD P HB-g''COD P Ao + * Murnleitner et ai, 1997 9 • Anoxic yield for formation of X G L Y GLY 12.6 + 2-% — *1.18 9 + 6-5 g CODoLY-g-'CODpAo 10 Anoxic yield for formation of X P P v N O I pp 57 + 9-% — « 3.02 28 + 4-5 g Ppp-g-'CODpAo / / Aerobic yield for degradation of X P H B Y° 1 PHA 9.4 + 3 % — «1 .39 2 + 9-<5 g CODpHB-g'CODpAo + Murnleitner etal., 1997 12 Aerobic yield for formation of X G L Y Y° G L Y 12.6 + 4-% —K1 . 1 1 9 + 12-5 g CODcLvg'CODpAo 13 Aerobic yield for formation of X P P Y° i pP 57 + 18-% — * 4.42 28 + 4-5 g Ppp-g'CODpAo + * Murnleitner etal, 1997 •Parameters that contain a typesetting error in Henze et al, 1999 (ASM2d). Parameters that contain rounding errors in van Veldhuizen et al. (1999). 234 Table I.5a. Kinetic parameters for Hydrolysis, X H and X A (Meijer, 2004) Hydrolysis of Particulate Substrate / Hydrolysis rate' 1 1 3 ( ) x e ( 0 . 0 4 0 6 , , -20)) gCODxs-g 'CODp^xpAO) d"1 Henze etal, 1999; ASM2d < T ) Henze et al, 1999; ASM2d 2 Anoxic hydrolysis reduction factor I N O 0.8 - Meijer, 2004 3 Anaerobic hydrolysis reduction factor 0.2 - Henze et al., 1999; ASM2d 4 Saturation / inhibition coefficient for oxygen Ko 0.2 g0 2-nV 3 " 5 Saturation / inhibition coefficient for nitrate K N O 0.5 gNsNo-m-3 6 Saturation coefficient for particulate C O D ( T ) K x 0 A x e m T -20)) gCODxs-g'COD(xH +xPAO) Henze et al., 1999; A S M 2 d m Henze et al., 1999; A S M 2 d Regular Heterotrophic Organisms X H / Maximum heterotrophic growth rate , T ) HH 6.0xei0MHT-20)) gCODxng-'CODxH-d" 1 Henze etal., 1999; ASM2d ( T ) Henze et al., 1999; A S M 2 d 2 Maximum fermentation rate ( T ) (0.069-(7-~ 3.0 x e 20)) gCODsFg-'CODxH-d' 1 3 Heterotrophic decay rate ( T ) b H - (0.069(r-0.4 x e 20» gCODxHg'CODxH-d- 1 4 Reduction factor for denitrification f|NO 0.8 - Henze etal., 1999; A S M 2 d 5 Saturation / inhibition coefficient for oxygen Ko 0.2 g 0 2 m - 3 " 6 Saturation coefficient for growth on SF K F 4.0 gCODsF-m"3 7 Saturation coefficient for fermentation of SF K f , 20.0 gCODsr-m-3 * Gujer et ai, 1995; A S M 2 . 8 Saturation coefficient for growth on Acetate K A 4.0 g C O D S A m - 3 Henze et al., 1999; ASM2d 9 Saturation / inhibition coefficient for nitrate K N O 0.5 gNsNo-m"3 10 Saturation coef. for Ammonium (nutrient) K N 0.05 gNsNH -m 3 11 Saturation coefficient for Phosphate (nutrient) K P 0.01 .gPspo;m'3 12 Saturation coefficient for alkalinity (HCO3) KHCO 0.1 m o l e H C 0 3 - m - 3 Autotrophic Nitrifying Organisms X A 1 Autotrophic growth r a t e m HA l W ' ' 0 5 ^ 20)) gCODxAg'CODxA-d- ' Henze et ai, 1999; A S M 2 d ( T ) Henze et al, 1999; ASM2d 2 Autotrophic decay rate ( T ) b A 0 . 1 5 x e ( 0 " ° ' ( r -20)) gCODxAg'CODxA-d- ' -3 Saturation coefficient for oxygen Ko 0.5 g 0 2 m ' 3 Henze etal, 1999; ASM2d 4 Saturation coefficient for.Ammonium K N H 1.0 gNsNH-m-3 " 5 Saturation coefficient for Phosphate (nutrient) K P 0.01 gPspo-m-3 6 Saturation coefficient for alkalinity (HC03 -) KHCO 0.5 moleHC03-m" 3 Temperature dependant parameters and their reference. * Parameters that contain a typesetting error in Henze et al., 1999 (ASM2d). 235 Table I.5b. Kinetic parameters for X P A Q (Meijer, 2004) Phosphorus Accumulating Organisms X P A O Maximum anaerobic acetate uptake rate<r) Anaerobic maintenance rate Maximum anoxic acetate uptake rate(T) PHA degradation rate (T) Glycogen formation rate ,T) Poly-phosphate formation rate<T) Observed oxygen consumption for maintenance 8.0 xe (0.090-(7-20)) mAN „ max.NO kpHA kcLY kpp m a e r 0 (0.069-(7'-20)) 0.05 x e , , rmx.AN (O.09O-(7'-2O)) 5.51 x e 0.93 x e 0.\0xel0mH 0.096 (0.118-IT-20)) gCODsA-g'CODpAo-d: 1 gPpp-g-'CODpAo-d"1 gCODsAg'CODpAod' 1 gCODpHAg'CODpAod' 1 gCODcLY'g 'CODpAO'd 1 gPppg'CODpAod' 1 gOi-g-'CODpAo-d'1 Meijer, 2004 m Brdjanovic et al., 1998 ( T ) Murnleitner et al:, 1997 Kuba etal., 1994 < T ) Brdjanovic et al., 1998 Meijer, 2004 , T ) Brdjanovic etal., 1998 Meijer, 2004 ( T ) Meijer, 2004 Murnleitner et al., 1997 0 1 Brdjanovic etal., 1998 Smolders et al, 1994b 8 Aerobic maintenance rate (T) m 0 3 • 3 • maero 3.2 + % « 0 . 0 6 x e ( 0 0 6 9 ( r - 2 0 » gCODpAog'CODpAod;1 + ° Murnleitner etal., 1997 0 1 Murnleitner etal., 1997 9 Anoxic maintenance rate (T) rnuo 6 • 5 • maero 63 + % « 0 . 0 9 x e < 0 0 6 9 ( r - 2 0 ) ) gCODpAog'CODpAod-1 10 Saturation reduction factor for poly-P formation gpp 0.22 - Murnleitner etal, 1997 II Reduction factor for denitrifying P removal r |No 0.8 - Meijer (2004) 12 Saturation coefficient for poly-P formation . Kpo 1.0 gPspo-m-3 " . .13 Saturation coefficient for growth on acetate K A 4.0 gCODS A-m-3 Henze et al., 1999; ASM2d 14 Saturation / inhibition coefficient for nitrate K N O 0.5 gNsNo-m'3 " 15 Saturation / inhibition coefficient for oxygen Ko 0.2 g0 2m- 3 " 16 Saturation coefficient for fpHA KfpHA 0.2 gCODpHAg'CODpAo Meijer, 2004 17 Saturation coeff. for Phosphate (nutrient) K P 0.02 gPspo-m'3 18 Saturation coefficient for N H 4 (nutrient) K N 0.05 gNsNH-m"3 " 19 Maximum poly-phosphate fraction of PAO's fm™ iPP 0.35 gCODppg'CODpAo Wentzel etal, 1989 20 Maximum glycogen fraction of PAO's f™ GLY .0.5 gCODoLY-g'CODpAo Brdjanovic etal, 1998 21 Saturation coefficient for poly-P Kpp 0.01 gPpp-m-3 Switch 22 Saturation coefficient for glycogen K G L Y 0.01 gCODcLY-m-3 " 23 Saturation coefficient for PHA KpHA 0.01 gCODpHA-m"3 " 24 Saturation coefficient for fcLY KfGLY 0.01 gCODcLYg'CODpAo " 25 Saturation coefficient for fPP Kfpp 0.01 gCODppg'CODpAo " 26 Saturation coefficient for alkalinity (HCO3) KHCO 0.01 moleHCOj-m3 + Parameters that contain rounding errors in van Veldhuizen et al. (1999). 0 Parameters that contain typesetting errors in van Veldhuizen et al. (1999). ( T ) Temperature dependant parameters and their reference. 236 Table 1.6. Kinetic rate equations of the T U D P model (Meijer, 2004) Process Kinetic Rate Equation (r?) Switch function (on / off) Hydrolysis of Particulate Substrate X s Aerobic Hydrolysis (gCODxsd"') rh° = K h X S / ( X H + X P A 0 ) ^ x + x ^ K x + x s P B + X P A O ) Anoxic Hydrolysis (gCODxsd1) . Anaerobic Hydrolysis (gCODxsd"1) r h N ° - ' / N O ' K h X S ^ ( X H + XpAo) Sjjp ' V H + A P A O - > K X + X s /(XH + X p A Q ) K N O + S N R A N _ „ X S /(XH + X p A O ) F h - Ik — Y , , r i A H + A P A o ) K X + A S ' v A H + A P A o J Regular Heterotrophic Organisms X H Aerobic Growth on Sp (gCODxH-d"') . Aerobic Growth oh SA (gCODxH-d1) . Anoxic Growth on S F (gCODxH-d1) Anoxic Growth on S A (gCODxH-d') „ Fermentation o f S F (gCODspd1) Q Heterotrophic Lysis (gCODxH-d1) rs A = A H -SA + SF Kp + SF K 0 + S0 R S F ~ ' / N O ' ' r s A ' / N O ' Mu ' S A + S F K A + S A K 0 + S 0 S F S , S M N "V. = q f c -S F K f e + SF ^ - N + S . N H K p + S P 0 K H C 0 + S H C O K 0 + S Q K N + S ^ , K p + S P Q K H C O + S H C O ^o S N H S P O S H C 0 K D + S 0 K N + S N H K p + S P O K H C O + S H C O Phosphorus Accumulating Organisms X P A O 10 Anaerobic storage of S A (gCODsA-d1) Anaerobic Maintenance (gPd-1) A N max.O % A — 4 s A K„ +S a KKI X„, K 0 + SQ K N 0 + SN 0 K G L Y + X G L Y Kpp + XPp K n K N n X„D 12 13 14 Anoxic storage of S A (gCODsA-d1) Anoxic PHA consumption (gCODxpHA-d"1) Anoxic storage of poly-P (gP-d-1) Anoxic Glycogen 15 formation (gCODxcLY-d1) 16 17 18 Anoxic Maintenance (gCODxPAO-d"1) Aerobic PHA consumption (gCODxpHA-d-1) Aerobic storage of poly-P (gPd 1) N O _ m a x , N O F S A — 4 s A rNO - n . k P H A ' / N O PHA - •x t ^ f P H A + ^ P H A ^ P A O ^ - N O + ^ N O WNO ^ P P Aerobic Glycogen 19 formation (gCODxcLY-d"1) X p p K p 0 + S p 0 gpp-KfjQ + Sj, X ' G L Y ' / N O ^ G L Y P H A ^ N O ' M " * N O r ° = k P H A P H A ipP i^ Pp L P A O J P O G L Y " - G L Y L- ^ P H A ^ O V ^o So + S^ K p + SpQ K H C 0 + SHCO K n X B U . K 0+S Q K P H A +X P H A K,pp + (f™ - X p p /Xp A 0 ) K n K 0 +S 0 KpH A+X P H A K,j L V + (fGLY - X 0 L Y / X P A O ) K n XDI fpP - Xpp / X P A 0 KpHA-*-XpH A. Kjpp + (fPp - Xpp/XPA0) K p H A + X p H A K ^ J L Y + (fGLy - X G L Y / X p A 0 ) .20 Aerobic Maintenance (gCODxPAod"') M " ' O Autotrophic Nitrifying Organisms X A 21 Autotrophic growth (gCODxA'd-1) ~y Autotrophic Lysis (gCODxA-d1) r A L = b A - X A 237 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0228857/manifest

Comment

Related Items