UBC Undergraduate Research

Noise and PID control for low consistency pulp refiner Reynolds, Byron; Pang, Haotian; Zhang, Lionel Jan 17, 2011

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

Item Metadata

Download

Media
52966-Reynolds_Byron_et_al_APSC_479_2011.pdf [ 852.33kB ]
Metadata
JSON: 52966-1.0074452.json
JSON-LD: 52966-1.0074452-ld.json
RDF/XML (Pretty): 52966-1.0074452-rdf.xml
RDF/JSON: 52966-1.0074452-rdf.json
Turtle: 52966-1.0074452-turtle.txt
N-Triples: 52966-1.0074452-rdf-ntriples.txt
Original Record: 52966-1.0074452-source.json
Full Text
52966-1.0074452-fulltext.txt
Citation
52966-1.0074452.ris

Full Text

    Noise and PID Control for Low Consistency Pulp Refiner     Byron Reynolds Haotian Pang Lionel Zhang    Project Sponsors: Dr. Jens Heymer Dr. James Olsen    APSC 479 Engineering Physics The University of British Columbia January 17, 2011  Project Number: 1059 ii    Executive Summary The main objective of this project was to remove the noise on various measurement signals for the low consistency pulp refiner, and implement a controller for the gap distance in two different modes of operation. The project is sponsored by Dr. Jens Heymer and Dr. James Olson from UBC Mechanical Engineering.  A design of the noise reduction algorithm and gap size control algorithm was implemented and tested by Byron Reynolds, Haotian Pang and Lionel Zhang.  Spectrum analysis showed that the majority of the noise on the gap signal is above 20Hz. As a result, a 12Hz Butterworth low pass filter of order 10 was chosen to reduce the noise on raw gap signal in volts. As the filtered signal is converted to size in mm, noise is also amplified. Therefore, an averaging of every 15 samples was employed to keep the signal clean enough so that it does not compromise the performance of the gap size controller. It was found after filtering and averaging of every 15 samples; the gap size uncertainty can be kept under 0.02mm, which meets the specification.  Before the gap size controller was implemented, a safety algorithm was developed and fully tested to ensure that no plate clashing or equipment damage occurs as it constantly checks the current gap size and keeps the gap size in a safe range. Based on the results from the actual effects of the P, D and I term. It was determined that a traditional PID controller is not suitable for the system due to the fact that the system is never stable with any value of P, D and I gain. The instability of the system is likely caused by the non-linearity and slow response of the actuator. Consequently, a combination of proportional control, pre-control and lock algorithm was used as the final gap size controller. The testing results showed that it is capable of moving the plate to any targeted position with a precision and maximum overshoot of ±0.05mm. Also, the gap size can be changed by 1mm within 10 seconds.  The energy based mode of the position controller cannot be fully implemented and tested prior to the completion of this report due to some issue with the refiner that needs to be fixed before it can be run properly with load.  All software codes including the filters and the gap size based position controller have been provided to our sponsor Dr. Jens Heymer. A list of recommendations is also included in this report, aiming to further reduce noise, help compensate for signal delay caused by averaging, and some suggestions on safety.  iii    Table of Contents Executive Summary          ii List of Figures           iv List of Tables           v 1. Introduction           1 1.1 Background and Significance of the Project      1 1.2 Project Objective          3 1.3 Scope and Limitation         4 1.4 Organization          4 2. Discussion           5 2.1 Theory           5 2.1.1 Electrical Noise and Shielding       5 2.1.2 Butterworth Filters         5 2.1.3 PID Control and Tuning Basics       6 2.2 Methods/Testing Protocol         8 2.2.1 Noise Reduction         8 2.2.2 Position Controller         8 2.3 Experimental Equipment, Flow Diagram and Algorithms     10 2.3.1 Experimental Equipment        10 2.3.2 Flow Diagram         11 2.3.3 General Description of Algorithms       11 2.4 Results           12 2.5 Discussion of Results         16 3. Conclusions           17 4. Project Deliverables          18 5. Recommendations          29 Appendix A. LabVIEW Front Panel and Codes       21 Appendix B. Hardware Equipment and Setups       23 Appendix C. Result of Noise Reduction for Other Signals     25 Appendix D. Description and Analysis of the Incident      28 Appendix E. Matlab Codes and Functions       29 References           33  iv    List of Figures Figure 1: Bars and Grooves on the Rotor Plate of a Pulp Refiner     1 Figure 2: Side View of the LC Refiner        2 Figure 3: Gain Response from Different Filters       6 Figure 4: Effects of the P, I and D Terms on a Typical Closed-loop System   7 Figure 5: Side View of the Open-hooded Refiner       10 Figure 6: Signal Flow Diagram         11 Figure 7: Effect of Filtering on the Gap Signal of a Pulp Refining Trial with Load  12 Figure 8: Power Vs. Gap Before and After Filtering with a 12Hz Lowpass Filter  13 Figure 9: Effect of P Gain on Gap Size Control       14 Figure 10: Effect of D and I Gain on Gap Size Control      14 Figure 11: Result of the Final Position Control Algorithm     14 Figure 12: Front Panel While Running the Gap       21 Figure 13: Filters in LabVIEW         22 Figure 14: PID Controller in LabVIEW        22 Figure 15: Long Distance Cables as Shown in Red Lines      23 Figure 16: Variable Frequency Drive (VFD)       23 Figure 17: Front View of the Actuator and Gap Motor      24 Figure 18: Top View of the Actuator and Gap Motor      24 Figure 19: Pressure In Signal with Optimal Cutoff Frequency 12Hz    25 Figure 20: Pressure Out Signal with Optimal Cutoff Frequency 11Hz    25 Figure 21: Power Signal with Optimal Cutoff Frequency 12Hz     26 Figure 22: Speed Signal with Optimal Cutoff Frequency 12Hz     26 Figure 23: Flow Signal with Optimal Cutoff Frequency 11Hz     27 Figure 24: Valve Signal with Optimal Cutoff Frequency 12Hz     27    v    List of Tables Table 1: PID Gains and Their Effects        7 Table 2: List of Deliverables         12 1    1. Introduction 1.1 Background and Significance of the Project The pulp and paper industry is one of Canada’s most significant and profitable industries, and it is especially concentrated in British Columbia (Wikipedia). According to the BC Pulp and Paper Industry Task Force, the BC pulp and paper industry contributes over $4 billion annually to the economy of British Columbia in 2007.  Pulp refining is a crucial step in papermaking process. It is intended to improve paper quality by modifying the fibre morphology or structure. The technology can be traced back to the early 19th century from simple batch beaters to the flow-through disc, conical and cylindrical refiners in use today. In pulp refiners, as pulp in suspension flows through an annulus between a rotor and stator, which have bars and grooves on both surfaces (see Figure 1), fibres are trapped in the gaps and subject to cyclic compression and shear forces during bar crossings;  as a result, the fibre properties are strengthened and paper quality is improved.  Figure 1: Bars and Grooves on the Rotor Plate of a Pulp Refiner  Compared to the traditional High Consistency refining technology, which typically has a water consistency above 15%, Low Consistency (LC) refining with a water consistency less than 5% is the primary industrial technique and provides significant energy savings in mechanical pulp production, 2    considering that an average pulp mill in BC consumes 300,000 MWH of power per year (PricewaterhouseCooper). A new state-of-the-art experimental LC pulp refining facility was created at UBC’s Pulp and Paper Centre. This facility is specifically designed to promote the use of the more energy efficient LC refining to replace less efficient, high consistency refining. It employs an existing experimental pulp pump loop consisting of two tanks (4 m3), one 40 kW pump, two 10 kW agitators, pipes and etc. In addition, a new 16 inch single-disc LC refiner was donated by AIKAWA/Advanced Fibre Technologies Inc. The side view of this LC refiner is presented in Figure 2.   Figure 2: Side View of the LC Refiner  This project is sponsored by Dr. Jens Heymer and Dr. James A. Olson from UBC Mechanical Engineering. Dr. Heymer and his group are mostly concerned with the correlation between the gap distance and power consumption of the refiner. The gap distance between the rotor and the stator is typically between 0-10 mm. The previous manual control of the gap motor was inconvenient and incompetent so that an advanced controller is required. The main objective of the project is to reduce the electrical noise on the measurement signals and develop a PID controller for the gap distance.  3    1.2 Project Objectives The first priority of this project is to remove the noise on the gap signal, as the noise was making an accurate position control rather difficult. Therefore, efforts were made to ensure that the signals that the PID controller is based on are clean and smooth. The main objectives and requirements of this project are listed below.  1. Noise control - Primarily focus on the gap distance signal since this is the main parameter for the PID controller. The signal should have an uncertainty within 0.02mm. - Secondary focus is on power, motor speed and flow signals - Pressure signal noise should also be reduced given sufficient time  2. Design and implement a PID controller for the gap motor - Capability of two modes of operation: gap size based and specific energy based - Allows high speed when gap is wide open (5-10 mm). - Overshoot should be avoided as it could cause the plates to clash resulting in damage. - Emergency option should be included; the gap should be withdrawn to a safe distance if the gap is outside of the safe range or plate clashing is detected by the user.  3. Optionally, design and implement a controller for the ball valve using LabVIEW   4    1.3 Scope and Limitations This report will address the method of reducing noise on the measurement signals and the design of a PID controller for the gap distance. The reasons why these methods were chosen to be implemented will be discussed. Moreover, it will present the results of the noise reduced signals and the PID controller. Our sponsor is mostly interested in the correlation between the gap distance and the energy consumption; however, the exact relationship between the gap distance and the energy efficiency of LC refining facility is beyond the scope of this recommendation report.  1.4 Organization This report will first introduce the theory part of this project. The source of the noise, Butterworth low pass filter and PID controller will be described in the theory section. Next, the methods of how to remove the noise and design the PID controller will be provided. The equipment of the experiment will be listed as pictures and a flow diagram will be shown as well. Our final results of both noise reduction part and PID controller part will be stated and discussed. The percentage of error in terms of noise reduction, the steady state error, the settling time and the overshoot with respect of the PID controller will be included. Furthermore, The LC Refining Facility incident happened in November will also be discussed in appendix D. The conclusion of this project will be drawn from the results. Finally, we would like to give recommendation in terms of both convenience and safety.  5    2. Discussions 2.1 Theory The source of the electrical noise, the Butterworth filter and the PID controller will be discussed in this section.  2.1.1 Electrical Noise and Shielding Electrical noise caused by the data communication cables could be very frustrating to correct as it can cause corruption of the signals being sent across the cable and render any actuating action useless. Ideally, the best strategy to eliminate noise is to prevent it from escaping the source. As a more common technique, shielding can be used to prevent noise signals from being transmitted from the source to the receiver. Shields can be placed at the source, the receiver or somewhere in between. However, a shield itself must be correctly designed. Most high-frequency shielding problems are resulted from openings in the shield. The effectiveness of a shield can be seen as a water bucket that holds electrical signals, because even a small hole in the bucket could cause a significant shielding problem; therefore, it is crucial to be sure of the integrity of the shielding surface. (SVC, 2010)  However, finding the source of noise could be difficult when the noise is intermittent and the source might be external to the system and inaccessible. Generally, high frequency noise is easier to filter than low frequency noise. Intermittent noise, such as that generated by motors, can sometimes be minimized by clamping power lines with diodes or transient absorbers. (Ample Power) After we investigated the gap distance noise, we decided to implement a low frequency filter to remove most of the noise.  2.1.2 Butterworth Filters In signal processing, filters refer to devices used to remove undesired components of the signals. After analyzing the spectrum of the gap distance signal, we realized a low pass filter has to be implemented so that high frequency noise components are not transmitted. The ideal filter should have an abrupt transition from passband to stopband; however no practical circuit can achieve such response. The Butterworth filter has a transfer function |Hሺjωሻ|ଶ ൌ Gబ మ ଵାሺ ω ωౙ ሻమ౤  , where n is the order of the filter and ωୡ is the cutoff frequency. N here can be used to adjust the sharpness of the transition from passband to stopband. The Butterworth filter has a feature such that it is extremely flat in the pass band so that no ripples occur in the passband. Although Butterworth filter has a slower roll-off, it has more linear slope in the phase response. In addition, both Matlab and LabVIEW have build in functions of Butterworth filter. These are the 6    reasons why we chose it against other filters. Figure 3 shows the gain responses of different filters. It is clear that Butterworth has no ripples and a good sharpness of transition from passband to stopband.  Figure 3: Gain Response from Different Filters (http://www.circuitstoday.com/active-filter-types)  2.1.3 PID Control and Tuning Basics PID stands for proportional, integral, and derivative. It is the most common type of feedback controller. In process control today, more than 95% of the control loops are of PID type (Karl Astrom, 2002). Figure 4 shows a block diagram of a typical PID controller.  A PID controller needs to be tuned by adjusting the values of the gains Kp, Ki and Kd to get the system to behave as desired. Table 1 summarizes the effects of each term on a control system, and Figure 5 shows the behavior of an example system with P, I and D term added separately.             7    Term Math Function Effect on Control System P - Proportional Kp * Verr Typically the main drive in a control loop. It reduces most of the overall error. However, if a steady-state error exists, even though it can be decreased by increasing Kp, the tendency to oscillate also increases. I - Integral Ki *  ∫ Verr dt Reduces the steady-state error in a system by summing small errors over time to produce a corrective signal large enough to drive the system toward the target, eliminating the offset D - Derivative Kd * dVerr / dt Counteracts the Kp and Ki terms when the closed-loop system is oscillatory. It helps reduce overshoot and ringing. Note that damping increases with Kd, but decreases again when Kp becomes too large. Table 1: PID Gains and Their Effects   Figure 4: Effects of the P, I and D Terms on a Typical Closed-loop System (Courtesy of ExperTune Loop Simulator)  In order to design a PID controller that meets the requirements, it is also necessary to consider issues such as noise filtering and high frequency roll-off, set point weighting, windup and etc. The details of these design considerations can be refereed to any typical control system design textbook. It should be kept in mind that the tuning process of a PID controller can be very time consuming and is largely based on trial and errors as real-life systems are usually too complicated to model with mathematical equations.   8    2.2 Methods and Testing Protocols The method of reducing high frequency noises and designing a position controller will be presented in this section.  2.2.1 Noise Reduction Raw data from the refining facility was collected and the Fast Fourier Transform (FFT) was used in Matlab to analyze the spectrum of the gap distance and other measurement signals. It was found that the majority of the noise is above a certain frequency while the signals are very close to DC. Therefore, Butterworth low pass filters, with different cutoff frequency for different signals, were chosen to reduce the noise, and the optimal cutoff frequencies were determined by trial and error such that the filtered signals have minimum fluctuation. In addition, as the noise appears to be periodic, averaging was employed to further reduce the noise and smooth the curves.  It is worth mentioning that even though averaging does help reduce the periodic noise in the signals it might also cause significant delay to the real time signals. The delay is undesired especially when the gap size is very small as it could easily cause plate clashing. Also, it can cause the gap signal to be more oscillatory, resulting in more overshoot and making PID control less effective. Consequently, an investigation was conducted to determine the tradeoff between the noise level and amount of delay, and the optimal averaging size was determined by trial and error until the noise was reduced without an unreasonable small delay.  The steps mentioned above were performed first on the data with the refiner off, and then repeated when the refiner was running without load, and finally for data collected from real pulp refining process, progressing through increasing amounts of interference and noise.   2.2.2 Position Controller It was tricky to test and debug the LabVIEW code at the very beginning. It was found that the actuator control the plate in the wrong direction by using the default output of the build-in LabVIEW pid.vi. This cannot be known in advance. As a result, an incident occurred because of the LabVIEW Abort problem and the fact that no particular safety precaution was taken. A record of this incident and a brief analysis is included in Appendix D.  9    Later on, some safety precaution was taken prior to a working Safety Algorithm. To be more specific, the shaft between the actuator and the plate was removed during the testing of the Safety Algorithm. As the actuator receives signals from the LabVIEW, the user can follow the rotation of the actuator by wheeling the plate manually. Meanwhile, the LED panel that shows a rough current gap size was watched to ensure that the plate position is a safe range. This method can prevent any clash caused to the motor. A picture of this setup is included in section 2.3.1.The shaft was not put back until the safety algorithm was fully tested and proven to be working under any circumstances.  The full range of the gap distance is 0-10mm. The valid input gap distance was chosen to be 0.1 to 9.9mm to avoid plate clashing. The input value from the user is automatically checked and values less than 0.1 would be converted to 0.1, while values larger than 9.9mm would be converted to 9.9mm. During testing, a safer range is set to be 2mm-8mm; the gap would withdraw automatically if the gap hits the limits. If the safety range is 2mm-8mm, the user could set the hysteresis range to be 2.5mm-7.5mm, so the gap would not stop the withdrawing until it hits the hysteresis range. For example, if the gap hits 8mm, it would stop and starts to close, only when the gap distance is less than 7.5mm, would it stop.  The actual tuning starts with the proportional gain P value, since we want to find the balance between the responsive time and the stability of the systems. Derivative gain D was introduced to reduce the oscillation and the integral gain is supposed to help eliminate any steady-state error. The actual effects of the gains were studied by adding the P, D and I term one at a time. Once the behavior of the system under P, PD, PID was determined. The final design of the controller can be decided and tuned with the assistance of the Pre-control Algorithm and Lock Algorithm.  The performance of the controller was evaluated in terms of amount of overshoot, transient time, and stability and steady-state error.    10    2.3 Experimental Equipment/Flow Diagram/Algorithms This section will present a list of equipment used in the project, the data flow of the code and a brief description of the algorithms used in the controller.  2.3.1 Experimental Equipment Our project is mostly written in software (Matlab and LabVIEW). We tested our filters in MATLAB and then implement the filters and PID controllers in LabVIEW. Facilities in UBC Pulp and Paper Centre are the experimental equipment we used: -Gap Motor and the LC Refiner -DAQ system and a controlling computer -Variable Frequency-Drive (VFD) system  Figure 5 is the side view of the LC refining facility. There is a LED panel which shows the current gap distance at the front. The near side is the manual control of the gap distance and the far side is the actuator so that the gap distance can be adjusted when signal coming from LabVIEW program. Figure 5 is the top view of the control part of the LC refining facility. The LabVIEW program can control the actuator so that the distance between the rotor and stator can be adjusted. The shaft between the actuator and the motor is shown removed for safety purpose during tuning.  Figure 5: Side View of the Open-hooded Refiner Actuator  LED Panel  Position Wheel  11    2.3.2 Flow Diagram  Figure 6: Flow Diagram 2.3.3 General Description of Algorithms Safety Algorithm This algorithm continuously checks if the current gap size is between 0.1mm and 9.9mm, called the safety boundary. If this condition is not met, all other algorithms are stopped automatically and the plate is withdrawn to a predefined safe range (within 1mm and 9mm) at maximum speed (±10V). Note that the safety boundary that triggers the automatic withdrawal is different from the safe range that the plate is withdrawn to. This hysteresis settings act as an anti-glitch and anti-noise buffer band analogous to these used in digital logic chips.  Pre-Control Algorithm Move with full speed (±10V) when the current gap size is more than 0.5mm away from the target.  Proportional Control Algorithm Whenever the difference between the current gap size and target ≤5mm and >0.05mm, output the actuation voltage = KP * |difference|. KP is chosen to be 30 as when the difference is slightly greater than 0.05mm, it would give an actuation voltage > ±3V, which is the minimum voltage required for the actuator to move. Note that the maximum speed the plates can move now is only half of the full speed to minimize overshoot  Lock Algorithm Whenever the difference between the current gap size and target <0.05mm, output 0V to actuator to avoid oscillation caused by over-actuation. Raw Data in  Voltage Low Pass Filter Gain&Offset Signals in proper  unitsAveraging  Gap Controller  and Output to  Acuator 12    2.4 Results Spectrum analysis showed that the majority of the noise on the gap signal is above 20Hz, and the optimal cutoff frequency was found to be 12 Hz. As shown in figure 7, the filtered signal (black curve) with a 12Hz low pass Butterworth filter of order 10 has an uncertainty of 0.46%, while the original signal (yellow curve) has an uncertainty of 9.03%. Note that this graph is based on the raw gap signal (in volts) with the worst noise level observed during wet pulp refining trials, which have more interference and vibration, and thus have more noise than those taken while the refiner is running without load.  Figure 7: Effect of Filtering on the Gap Signal of a Pulp Refining Trial with Load  Since the sponsor is particularly interested in determining the correlation between power consumption and the gap size, the gap signal must be related to the power signal. Figure 8 shows the plot of raw power vs. gap signals (both in Volts) before and after filtering. Note that since the power signal has more low frequency noise (<12Hz), which cannot be removed by the lowpass filter, and the fluctuation on the power signal (horizontal) is larger than on the gap signal (vertical).  The power, speed, flow and pressure signals all show significant improvement after the filter is applied. The optimal cutoff frequencies were determined individually and all very close to 12Hz. The plots of the results are included in Appendix C. 13     Figure 8: Power Vs. Gap Before and After Filtering with a 12Hz Lowpass Filter  Although the gap signal (in volts) shows significant improvement after the 12 Hz low pass filter is applied, the absolute fluctuation of the gap size (in mm) is still too large for an accurate position control, because the noise was also amplified during the conversion of gap signal in volts to the actual gap size in mm. As a result, an averaging algorithm is used to further reduce noise. As shown in Figure 9, the filtered gap size (mm) has a much smaller fluctuation with an averaging of every 25 samples. However, it should be noticed that the delay between the averaged curve (blue) and the filtered signal without averaging (red) might to large enough to cause clashing when the gap size is decreasing to a small target. The optimal averaging number that can reduce noise to ±0.02mm effectively with negligible delay was determined to be 15, which is used in the final design.  Figure 9 is also the result of a controller with proportional gain only. The rotating plate was moved from 5.5mm to the targeted 4mm, and it can be observed that the gap size does not converge to 4mm over time. Instead, it is sinusoidal and only marginally stable. It should be noted that a lower value for the P gain results in a smaller oscillation with a longer period, but does not help the gap size to converge. 14     Figure 9: Effect of P Gain on Gap Size Control  Figure 10: Effect of D and I Gain on Gap Size Control  15    Since the gap size oscillates about the target, a derivative gain was added to help dampen the oscillations. As shown in Figure 10 (the yellow curve & the averaged green curve), the derivative gain indeed helps reduce oscillation. However, it also introduces an offset, which can only be corrected by a proper integral gain. Unfortunately, the system becomes unstable for any value of KI. The red curve with the averaged blue curve is the result from a small KI (only 1% of KP) is added to the PD controller.  Consequently, a proportional controller with the pre-control and the lock algorithm, as described in section 2.3.3, was implemented and chosen to be the final design because it has capability of moving the plate to any desired position with a precision and maximum overshoot of ±0.05mm. Figure 11 show the process of moving the plate without load from a large gap distance to 6mm, 5mm, 4mm and 3mm, respectively, and held for about 10 seconds at each target. It can also be seen that the gap size can be changed by 1mm within 10 seconds.  Figure 11: Result of the Final Position Control Algorithm   16    2.5 Discussion of Results The 12Hz Butterworth lowpass filter of order 10 shows best results in noise reduction. An additional experiment was performed to find the effect of the filter order on noise reduction. It shows no improvement of noise reduction with a higher order, but instead causing more delay to the signals. Therefore, order 10 was chosen to the filter setting in LabVIEW for all the signals.  It was mentioned that the noise on the raw gap signal in volts gets amplified when gap is converted to the actual size in mm. The amplified noise was reduced by an averaging algorithm, which introduces delays to the signal depending on the number of averaging used. An investigation of the effect of averaging number on noise reduction and delay was conducted. Just as shown in Figure 9, an averaging of 25 causes a larger delay. Therefore, the final averaging number was chosen to be 15, and the final testing of the control algorithm confirms that corresponding delay does not cause excessive overshoot.  Based on the results from the actual effects of the P, D and I term. It was determined that a traditional PID controller is not suitable for the system due to the fact that the system is never stable with any value of P, D and I gain. The instability of the system is likely caused by the non-linearity and slow response of the actuator. To be more specific, the actuator is not powerful enough to control the plate when the actuation signal is between -3V and 3V; and it takes a while to switch the moving direction of the plate due to its relatively large momentum.  The results of the final position control algorithm were obtained from trials without load. Fortunately, the overshoot is expected to be smaller or even completely eliminated after real pulp (load) is running between the plates thanks to the increased mechanical resistance. The result from a real pulp refining trial could not been obtained prior to the completion of this report due to some vibration issue with the refiner that needs to be fixed before it can be run properly with load. It should be mentioned that the process of increasing the gap to larger target positions were also conducted and satisfactory results were obtained with the curve to be just the reverse of the one in Figure 11.  Additional tests which might help evaluate the quality of the final design is to set the target to be the minimum size allowed and to see if the delay and noise would cause plate clashing or premature withdrawal  due to the safety algorithm.  Moreover, it might also be desirable to verify whether the plates would be just about to move with an actuation voltage of exact ±3V, which in turn gives the minimum resolution that the target can step accurately. 17    3. Conclusion This report presents the design and results of noise reduction on measurement signals and an accurate gap position controller. The primary objectives of this project were achieved successfully. Namely, a 12Hz Butterworth lowpass filter of order 10 with an averaging algorithm of 15 samples can effectively reduces the noise on gap size to be less than 0.02mm. The gap size controller can achieve a precision and maximum overshoot within 0.05mm from the target. Also, the performance of the safety algorithm was particularly ensured so that the system would not be accidentally damaged.  A traditional PID controller was not used since the gap system is not stable over time due to the nonlinear property of the actuator. Any fine actuation that requires the gap to move at very small voltage (<3V) is ignored. As errors would add up when the actuation voltage is too small to move the plate, the integral gain would push the system from marginally stable to completely unstable.  As a solution to this complication, an accurate control of the plate position was achieved by a combination of the pre-control algorithm, proportional controller and lock algorithm; that is, the gap size moves at full speed when the it is more than 0.5mm away from the desired position, and a proportional controller would be in effect only when the gap plates are within 0.5mm but at least 0.05mm away from target. The controller will be stopped when the position is 0.05mm from target, and the inertia would bring the plate to the target. A careful choice of the proportional gain was calculated and tested to minimize or even overshoot. Accordingly, the gap size can be changed by 1mm within 10 seconds.  The energy based mode of the position controller is not able to be fully implemented. The system has a small vibration at this moment, and no pulp could be applied for testing and tuning. Hence, we could not determine the exact relationship between the energy consumption and the gap distance, which is necessary for the implementation of the energy based mode. Nevertheless, a preliminary code of the controller will be provided to the sponsor in January 2011. The exact testing and tuning of this code will be conducted either by the team members of this project or the sponsor after the vibration issue is fixed.  All software codes including the filters and the gap size based position controller have been provided to our sponsor Dr. Jens Heymer. Although we had initially hoped to be able to finish the ball valve controller, this proved to be an unattainable goal due to the failure of the machine in November. However, we are still pleased with the completion of the noise reduction and position controller part of the project, and believe that this represents a major step towards the completion of the system. 18    4. Project Deliverables 1. List of Deliverables File Name Environment Description my_fft.m Matlab Spectrum analysis of signals lowpass_to_signal.m Matlab Butterworth lowpass filter of order 10 averaging.m Matlab Averaging algorithm plot_gap.m Matlab Comparison of Gap vs. Time plots before and after filtering power_gap.m Matlab Comparison of Power Vs. Gap plots before and after filtering compare_gap.m Matlab Comparison of filtered only Vs. filtered & averaged Gap Size AIKAWA REFINER.vi LabVIEW Modified main program based on the sponsor's original version; filters, position controller and other features added. SafetyCheck.vi LabVIEW Sub VI for checking whether the current gap size is in range PositionControl.vi LabVIEW Sub VI for the position controller based on the desired gap size PositionControl2.vi LabVIEW Sub VI for the position controller based on the desired power Table 2: List of Deliverables  2. Financial Summaries LabVIEW software copies provided by our sponsor Our project is mostly based on software, so there is no cost associated with this project.  3. Ongoing commitments by team members A preliminary code of the controller will be provided to the sponsor in January 2011. The exact testing and tuning of this code will be conducted later this term either by the team members or the sponsor after the vibration issue is fixed. 19    5. Recommendation 5.1 LabVIEW Abort Problems We have determined that the gap motor DAQ maintains the last signal it was provided even after losing contact with the computer. This leaves the refiner in danger of the computer crashing or freezing. LabVIEW programs are interpreted by the LabVIEW virtual machine, which can abruptly stop the program's execution, such as through the task manager. In this situation, the LabVIEW virtual machine would not gracefully exit by running code to clean up the actuator signals upon exit leaving the gap motor to continue moving the plate until the plates clash, or the actuator breaks. This would require extending the range limiter to include mechanical/electrical components to control the motion, such as sensing the position with micro-switches which could be precision-adjusted using a micrometer handle, and use power relay to control the motor. As an interim solution, the abort button in LabVIEW has been deactivated to prevent user mistakes.  5.2 Compensating for Averaging Delay By looking at the graphs of the averaged and unaveraged signals, it can be seen that while averaging smoothes the signal out, it introduces a time delay. This delay appears constant and determined by the number of samples used. It would be useful for research purposes to subtract this delay from the calculated value before printing it to the file. This delay can be calculated using the sampling frequency from the DAQ and multiplying it by the number of samples used to average the signal. Averaging the signal does distort the signal when it is rapidly changing, but at steady state the signal is sufficiently accurate for our sponsor's purposes.  5.3 Further Reduction of Noises It was proposed that the cable is acting as a capacitor, with the shield and signal wire acting as the electric plates, and the insulator acting as a dielectric. This effect has been documented in other systems (SVC, 2010); however, we have concluded that this is unlikely because the noise is distributed evenly along the high frequency portion, such as the case in white noise. It is likely that the system will have to be debugged in order to isolate the issue. We recommend the following procedure:  Firstly, disconnect the sensors and cables, and connect the DAQ terminals together and measure the signal’s stability. This would tell if there is an issue with the DAQ or the power supply. If the signal is noisy, then an electrician’s experience and equipment would be needed to perform more detailed analysis 20    and possibly provide a dedicated power supply.  Secondly, connect the cable to the DAQ, and connect the other ends together, effectively short circuiting the sensors. The signal should then be measured. If the signal is unstable then there is interference caused by other systems. In this case, a new DAQ would have to be built close to the sensors to digitalize the data before sending it to the computer, or the signal would have to be amplified before being sent through the cable, and attenuated after leaving, resulting in the noise introduced being attenuated while the signal is reduced to the original value.  Finally, if noise is not present in either of the prior tests, then the sensors must be tested individually. They cannot be tested together because while the DAQ is shielded well from external noise, internal noise from one input can contaminate the others. It is expected that one of the sensors will be sensitive to the noise, and introduce it to the DAQ where it will interfere with other measurements.  21    Appendix A. LabVIEW Front Panel and Codes   Figure 12: Front Panel While Running the Gap 22     Figure 13: Filters in LabVIEW  Figure 14: PID Controller in LabVIEW 23    Appendix B. Hardware Equipment and Setups  Figure 15: Long Distance Cables as Shown in Red Lines  Figure 16: Variable Frequency Drive (VFD) 24     Figure 17: Front View of the Actuator and Gap Motor  Figure 18: Top View of the Actuator and Gap Motor  25    Appendix C. Result of Noise Reduction for Other Signals  Figure 19: Pressure In Signal with Optimal Cutoff Frequency 12Hz  Figure 20: Pressure Out Signal with Optimal Cutoff Frequency 11Hz 26     Figure 21: Power Signal with Optimal Cutoff Frequency 12Hz  Figure 22: Speed Signal with Optimal Cutoff Frequency 12Hz 27     Figure 23: Flow Signal with Optimal Cutoff Frequency 11Hz  Figure 24: Valve Signal with Optimal Cutoff Frequency 12Hz      28    Appendix D. Description and Analysis of the Incident On November 23, 2010, Thursday, Lionel and Haotian were testing the PID program on the computer. We ran our LabVIEW program shortly and probed the output voltage signal to the motor. We realized that the direction of the actuation signal was wrong, so we turned off the program to fix the code. Unfortunately, due to our inexperience with LabVIEW and the system, we did not realize that the motor would still be receiving the actuation signal (-10V) from the DAQ even when the LabVIEW program and Variable- Frequency Drive (VFD) were both off. The VFD only controls the spinning of the plates instead of the gap motor. As we were still editing the code, we heard a very loud cracking noise and realized that something was broken. This occurred approximately 3-5 minutes after we stopped the LabVIEW program. The same problem occurred on Nov 19, Friday, when we tried a small actuation signal (-1V) and we believed that it caused a very tiny crash which we did not realize. After investigated by Physics department machine shop, the cast-iron bracket and the worm gearbox from the refining machine were found to be broken. Both components have been replaced by newly fabricated ones.  The following reasons caused this failure:  1. The “Abort” button in LabVIEW does not shut down the actuator voltage signal. The actuator would still be receiving the actuation signal (-10V) from the DAQ after the program has been aborted. An additional stop button has been implemented so that we can stop the gap motor when needed. The user of this facility must be very cautious at this part, because after click the “abort” button, there is no way to stop the motor unless turn off the switch of the motor. This is a build in nature of the LabVIEW.  2. The actuator opens the gap when a negative voltage applies, while it closes the gap when a positive voltage is applied. The build in nature of the actuator could only be found by testing. We should have removed the shaft connecting the actuator and the motor before we have done any testing.  3. The main power switch only controls the VFD, not the gap motor. The switch of the motor locates at the other side of the computer. It is very difficult to control the power of the motor while testing on the computer.  29    Appendix E. Matlab Codes and Functions function [] = my_fft(signal)  fs = 1000;               % Sample freq = 1000 Hz max_signal_f = fs/2;     % Nyquist sampling theorem num_data = length(signal);  y = abs(fft(signal)); f = (max_signal_f)*linspace(0,1,num_data);  % Plot single-sided amplitude over the full spectrum. subplot(2,1,1); plot(f,y); %plot(f(100:num_data),y(100:num_data)) axis ([0 500 0 5000]) title('Single-Sided Amplitude over the Full Spectrum') xlabel('Frequency (Hz)') ylabel('|Y(f)|')  % Only plot the lower frequency part to visulize the noise freq values more easily subplot(2,1,2); plot(f,y); %plot(f(100:num_data),y(100:num_data)) axis ([0 50 0 10]) title('Lower freq portion only') xlabel('Frequency (Hz)') ylabel('|Y(f)|')   function [] = lowpass_to_signal(f_cutoff, data, parameter)  t = data(:,1); Pin = data(:,2); Pout = data(:,3); gap = data(:,4); power = data(:,5); speed = data(:,6); flow = data(:,7); valve = data(:,8);  signal = parameter;  fNorm = f_cutoff/500; [b, a] = butter(10, fNorm, 'low');       % Construct a butterworth low pass filter of order 10 Hd = dfilt.df1(b,a); % Now use nonzero initial conditions by setting ICs after before you filter. Hd.persistentmemory = true; Hd.states = signal(1);      % Uses scalar expansion.  y = filter(Hd,signal); 30     %Calculate the absolute error before filter applied high1 = max(signal); low1 = min(signal); err_before = (high1 - low1)/2;  %Calculate the absolute error after filter applied high2 = max(y); low2 = min(y); err_after = (high2 - low2)/2;  plot(t,signal,'yellow') hold on plot(t, y, 'black') title(['Uncertainty after filtering = +/- ', num2str(err_after,'%10.3f'), '%; c.f. original +/- ', num2str(err_before,'%10.3f'), '%.']); xlabel('Time(Seconds)') ylabel('Signal (Volts)')   end   function[]= gap_plot(data)  t=data(:,1); gap=data(:,2);  %Calculate the +/-error high = max(gap); low = min(gap); avg = (high + low)/2; error = (high - avg);  plot(t,gap) hold on plot(t,6, 'black') plot(t,5, 'black') plot(t,4, 'black') plot(t,3, 'black') %axis([1 10 5.91 5.914]); xlabel('Time (sec)') ylabel('Gap Size (mm)') title(['Filtered and Averaged Gap Size Vs. Time']);  end  function [] = power_gap(input)  t = input(:,1); gap = input(:,4); power = input(:,5); 31     fNorm = 12/500;                         % 12 Hz lowpass filter with max signal frequency 500Hz [b,a] = butter(10, fNorm, 'low');       % Construct a butterworth low pass filter  Hd = dfilt.df1(b,a); % Now use nonzero initial conditions by setting ICs after before you filter. Hd.persistentmemory = true;     % Enable initial condition Hd.states = power(1);           % Set initial condition power_new = filter(Hd, power);  % Apply the low-pass filter to the input signal  Hd.states = gap(1);             % Set initial condition gap_new = filter(Hd, gap);      % Apply the low-pass filter to the input signal   plot(gap,power,'yellow'); hold on plot(gap_new,power_new,'black');  title('Power vs Gap: yellow - original; black - filtered'); %axis([3 4 1 7]); xlabel('Gap Signal (Volts)'); ylabel('Power Signal (Volts)');  end   function [] = averaging(num, data)  t = data(:,1); gap = data(:,4);  fNorm = 12/500; [b, a] = butter(10, fNorm, 'low');       % Construct a butterworth low pass filter of order 10 Hd = dfilt.df1(b,a); % Now use nonzero initial conditions by setting ICs after before you filter. Hd.persistentmemory = true; Hd.states = gap(1);      % Uses scalar expansion.  gap_new = filter(Hd,gap);  % Averaging every n samples n=num; len=floor(length(t)/(n)); for i=1:len gap_avg(i)=mean(gap_new((i-1)*n+1:i*n)); t_avg(i)=t(1+(i-1)*n); end  32     %Calculate the absolute error before averaging applied high1 = max(gap_new); low1 = min(gap_new); err_before = (high1 - low1)/2;  %Calculate the absolute error after averaging applied high2 = max(gap_avg); low2 = min(gap_avg); err_after = (high2 - low2)/2;  plot(t,gap_new,'blue') hold on plot(t_avg, gap_avg, 'red') axis([t(1) t(end) 3.75 3.85]) title(['Gap signal swing after averaging = +/- ', num2str(err_after,'%10.3f'), 'mm; c.f. filtered only +/- ', num2str(err_before,'%10.3f'), 'mm.']); xlabel('Time(sec)') ylabel('Gap Size(mm)')  end  33    References “Control System Design”. Chapter 6.Astrom, Karl Johan. 2002.  “Report on the Economic Impact of the BC Pulp and Paper Industry, Prepared for BC Pulp and Paper Industry Task Force”. PricewaterhouseCooper. 2007. http://www.llbc.leg.bc.ca/public/pubdocs/bcdocs/431477/final_pwc_report_to_task_force_nov_07.pdf  “Pulp and paper industry in Canada”. http://en.wikipedia.org/wiki/Pulp_and_paper_industry_in_Canada. Wikipedia. 2010 “Electrical Noise Sources”. http://www.amplepower.com/apps/noise/index.html. “Shielding”. http://svconline.com/news/avinstall_shielding/#container. SVC. 2010   “PID controller”.  http://en.wikipedia.org/wiki/PID_controller. Wikipedia. 2010 “PID Controller”. http://www.ecircuitcenter.com/circuits/pid1/pid1.htm. eCircuit Center. 2002. “What Is PID—Tutorial Overview”. http://www.expertune.com/tutor.html. ExperTune Inc. 2010   “Circuits Today-Active Filter Types”. http://www.circuitstoday.com/active-filter-types. 2010 

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}]}"
                            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.52966.1-0074452/manifest

Comment

Related Items