International Construction Specialty Conference of the Canadian Society for Civil Engineering (ICSC) (5th : 2015)

Telematics data-driven prognostics system for construction heavy equipment health monitoring and assessment Said, Hisham M.; Nicoletti, Tony Jun 30, 2015

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

Item Metadata

Download

Media
52660-Said_H_et_al_ICSC15_149_Telematics_Prognostics_Construction.pdf [ 1.98MB ]
52660-Said_H_et_al_ICSC15_149_Telematics_Prognostics_Construction_slides.pdf [ 569.17kB ]
Metadata
JSON: 52660-1.0076365.json
JSON-LD: 52660-1.0076365-ld.json
RDF/XML (Pretty): 52660-1.0076365-rdf.xml
RDF/JSON: 52660-1.0076365-rdf.json
Turtle: 52660-1.0076365-turtle.txt
N-Triples: 52660-1.0076365-rdf-ntriples.txt
Original Record: 52660-1.0076365-source.json
Full Text
52660-1.0076365-fulltext.txt
Citation
52660-1.0076365.ris

Full Text

5th International/11th Construction Specialty Conference 5e International/11e Conférence spécialisée sur la construction    Vancouver, British Columbia June 8 to June 10, 2015 / 8 juin au 10 juin 2015   TELEMATICS DATA-DRIVEN PROGNOSTICS SYSTEM FOR CONSTRUCTION HEAVY EQUIPMENT HEALTH MONITORING AND ASSESSMENT Hisham M. Said1,3, 4, and Tony Nicoletti 2 1 Department of Civil Engineering, Santa Clara University, USA.  2 Director, Sales and Business Development, DPL America, USA 3 Adjunct lecturer, Structural Engineering Department, Cairo University, Egypt 4 hsaid@scu.edu  Abstract: Construction heavy equipment is a valuable asset for construction and equipment rental companies, which requires continuous monitoring and assessment for potential failures. Predictive maintenance has recently been proposed to as an alternative to preventive maintenance strategy by scheduling maintenance tasks just before a predicted failure of the equipment. Such predictive approach is dependent on the existence of a data collection and analysis system that monitors the equipment performance, compares it to the previous history, and predicts the failure events before their occurrence. This paper presents the development and validation efforts of a data-driven prognostics system that utilizes timely collected telematics data to monitor the equipment health condition and predict its failure hazard. The system is designed to utilize equipment telematics data to develop regression-based Cox’s proportional hazards functions. Regression analyses are performed for the historical telematics data to develop time-varying hazard functions for the successive life intervals of the equipment to generate dynamic predictions of its failure events. Accordingly, the outcome of the system would be the predicted probability of the equipment failure event considering the timely collected telematics data. The proposed prognostics system was validated by developing the hazard functions of two fleets of dozers and backhoes that provided high fit to the observed data and high prediction accuracy for the testing data. For both analyzed fleets, higher predictive and data fitting performance were achieved for later life intervals due the increased reliability of failure prediction for equipment with longer survival lives.   1 INTRODUCTION Construction heavy equipment is a valuable asset for construction and equipment rental companies, which requires continuous monitoring and assessment for potential failures. The absence of a properly implemented maintenance program leads to premature equipment failure and increased construction crew idle time. Predictive maintenance (Gransberg et al. 2006) has recently been proposed as an alternative to corrective and preventive maintenance strategies by scheduling maintenance tasks just before a predicted failure of the equipment. Such predictive approach is dependent on the existence of a data collection and analysis system that monitors the equipment performance, compares it to previous history, and predicts the failure events before their occurrence. Despite the great promise of predictive maintenance, its wide implementation was not realized yet due to its need for large data collection process and supporting technology. 149-1 Telematics is a data collection technology that integrates wireless communications, vehicle monitoring systems, and location devices to provide real-time spatial and performance tracking of the fleet machines (Lovett et al. 2003). As a witness to its great benefits, the telematics industry has significantly grown to be installed in 5.8 million equipment units with revenue volume of $2 billion in 2009 (Fletcher and Lauron 2009). Telematics can be installed by either the original equipment manufacturer (OEM) or a third-party service provider (TSP). Telematics has been utilized as mainly a real-time monitoring system of equipment fleet for the purposes of theft protection, fuel consumption, and prevention of undesired behaviors of operators/drivers. Fleet managers are challenged to expand the utilization of telematics technology due to the difficulties in inking telematics data to business functions and performance metrics (Monnot and Williams 2011, Trimble and Bowman 2012, Jackson 2012, Sutton 2013).       2 RESEARCH OBJECTIVE AND METHODOLOGY This paper presents the development and validation efforts of a data-driven prognostics system that utilizes timely collected telematics data to monitor the equipment health condition and predict its failure hazard. Prognostics is the field and methodologies of predicting the future health behavior, failure events, and remaining useful like (RUL) of equipment and machines by diagnosing the recorded temporal behavior (Mesgarpour et al. 2013). Current approaches of prognostics can be classified into three main classes (Lee et al. 2006): 1) model-based prognostics that depends on developing virtual models of the machine that mimic its behavior under healthy and faulty conditions; 2) data-driven prognostics that utilize collected sensor data of the machine’s previous behavior toward failure, and 3) hybrid prognostics that ingrates models formulation with sensor data calibration. This paper proposes a data-driven prognostics model that utilizes equipment telematics data to estimate its failure probability.    To accomplish this objective, the research methodology encompassed four main tasks. First, a thorough literature review was performed to study the previous research in the areas of equipment health prognostics and telematics application in construction. Second, a brief description of telematics system architecture and data was presented. Third, the formulation of the proposed telematics-based prognostics system was developed and illustrated. Fourth, the performance of the developed system was validated by analyzing the telematics data of two types of heavy equipment fleets. The paper is concluded by summarizing the contribution of the proposed system to heavy construction and equipment rental companies, as well as recommendations for future research.       3 PREVIOUS RESEARCH Relevant previous research related to this paper is summarized into two main categories: equipment health prognostics and application of equipment telematics in construction. First, previous research on equipment and machine prognostics focused on monitoring the condition of stationary mechanical machines or electrical micro-machines by majorly measuring their vibrations (Dutta and Giurgiutiu 2000; Yan R. and Gao 2007; Da et al. 2011, Thomson 2013). Equipment manufacturers have encouraged research and development efforts to develop health monitoring systems that integrate remote sensing and equipment oil sampling to diagnose its condition and estimate its life expectancy (Murakami et al. 2002). Little research was found to indicate the potential of utilizing telematics data in prognostics, with no development of proven models or systems (Mesgarpour et al. 2013).  Second, little research was performed to investigate the use of the telematics technology in construction and heavy equipment fleet management. Monnot and Williams (2011) highlighted the possible use of telematics in various equipment fleet management tasks, like reporting of machines hours, locations, fuel consumption, and health. Aslan and Koo (2012) proposed an implementation plan for the use of telematics technology in the improving the productivity of roadway maintenance operations. The plan was to me completed in a future study with testing a telematics data collection system and developing productivity measurement metrics.     The careful study of previous studies identified the critical research gap and need for new prognostics systems that would enable the use of telematics data to assess equipment failure potential, as an essential part of effective predictive maintenance (Gransberg et al. 2006).  149-2 study the telematics data and collect its samples for the development and validation of the proposed heavy equipment prognostics system.    TransponderUnitTelematics Data Sent through  Wireless AntennaLocation data Received by GPS ReceiverJ1939 Data (engine speed, check lamps, oil pressure, fluids temperature) received through  CAN-bus ConnectionBasic Data Received (local temperature, battery voltage, engine runtime) through Main Interface CablesCAN-busControl Units Figure 2: Telematics Data Types 5 EQUIPMENT HEALTH PROGNOSTICS USING TELEMATICS-BASED SURVIVAL ANALYSIS The development of the proposed equipment prognostics system is presented by listing the proposed telematics health data, providing an overview of survival analysis, and describing the modeling approach of the equipment failure hazard functions.    5.1 Proposed Telematics-Based Health Parameters Telematics provides a rich data source that is utilized in this research to derive diagnosis and prognosis parameters of the equipment health. The proposed prognostics system is generic and can be applied to any set of telematics health parameters available in the collected data. However, the inputted telematics health parameters may affect the quality of the generated equipment health hazard functions. Accordingly, the following eight health parameters available in every telematics entry at time t, were proposed in this research based on available literature review (Murakami et al. 2002, Dekate 2013) and consultation with equipment telematics professionals:   1. Maximum coolant temperature (MCTt) in degrees Fahrenheit, which is observed on the day when the telematics data entry is received. 2. Maximum engine oil pressure (MOPt) in pounds per square inch (psi). 3. Maximum engine oil temperature (MOTt) in degrees Fahrenheit. 4. Maximum engine speed (MESt), in rounds per minute (rpm). 5. Maximum engine percent torque (MPTt), which indicates the load on the engine as a percentage value. 6. Maximum fuel rate (MFRt) in gallons/hour. 7. Engine working hours (HWt), which reports the cumulative number of hours the engine run with a speed (rpm) above a specified threshold, set by the fleet manager. 8. Engine Idling hours (HIt), which reports the cumulative number of hours the engine ran with a speed (rpm) less than the specified threshold.  149-4 5.2 Survival Analysis and Failure Hazard Functions Survival analysis is a regression approach of reliability studies to assess the times and probabilities to failure events. Survival analysis has been applied before to analyze the lifetime of orgasms, survival times of cancer patients, occurrence of accidents, and failure times of machines (Gu et al. 2011). Survival analysis deals with the failure event time as a random variable T using different representations (Allison 1995), such as 1) the cumulative distribution function of variable T, P(t) = Pr(T≤t); 2) the survival function S(t) as the complement of the distribution function, S(t) = Pr(T>t) = 1 – P(t); and 3) the hazard function h(t) that assesses the instantaneous at time t. The goal of the survival analysis is develop these representations of the failure event as a function of its determinant, i.e. its health parameters.    Survival analysis models can be classified as non-parametric, parametric, and semi-parametric models (Ma and Krings 2008). Cox’s proportional hazards model (Cox 1972), one of the fundamental semi-parametric survival models, is utilized for this study due to its ability to capture the failure covariates more effectively than non-parametric models with less modeling restriction like the parametric models (Bailey et al. 2006). As shown in Equation 1, a dynamic time-varying hazard function h(t) of Cox’s model is utilized in the proposed equipment prognostics system to estimate the probability of equipment failure at time t, which is based on: 1) the telematics health parameters X(t) (covariates) and their coefficients β(t) that can be determined by performing a regression analysis to fit the observed health parameter values to the failure hazard probability h(t); and 2) a baseline failure rate h0(t) represents the decay of the equipment health regardless of the values of its health parameters. The next section will explain in more details the methodology of developing the hazard functions and the calculation of the observed equipment hazard probability based on the available telematics data.  [1] h(t) = h0(t) × exp [β(t).X(t)]        5.3 Modeling and Development of Equipment Failure Hazard Functions The proposed heavy equipment prognostics system follows a novel methodology to model and develop Cox’s proportional survival functions as a function of the available telematics data. As shown in Figure 3, the equipment hazard functions are developed in four main steps:  Step (a) – Telematics data are used to mark the failure events over the lifetime of the equipment and the engine hours. An equipment failure event is recognized if one of two engine lights is reported in the telematics data entry: 1) red stop light (RSL) that indicates a severe enough condition that it warrants stopping the vehicle; and 2) amber warning light (AWL) that reports a problem with the vehicle system but the vehicle does not need to be immediately stopped. Other engine lights, such as engine protection light (EPL) and malfunction indicator light (MIL), are not used to indicate a major equipment failure as they report less severe health conditions related to the vehicle electronic system and emissions-related issues. As shown in Figure 2, the existence of either RSL or AWL marks a failure event and the start/end of a survival life of the equipment. As shown in Figure 3, survival analysis permits the right and left censoring (Ma and Krings 2008) of the data to only consider the complete survival lives that start and end with a recorded failure/survival event.   Step (b) – The observed failure hazard value is quantified using another telematics data element, which is the engine total hours (HT) that reports the cumulative engine hours up to the time of the telematics data entry. As shown in Figure 2, the observed failure hazard is quantified over each of the identified survival lives as the ratio between: 1) the engine hours Et since the start of the survival life to time t; and 2) the total engine hours L occurred during corresponding survival life. Accordingly, the failure hazard observed values range from 0 (at the beginning of the survival life) to 1.0 (at the end of the survival life).     Step (c) – All telematics entries are combined to form the analysis data population and divided into survival intervals, which each will have its own hazard function to represent the time-varying nature of equipment health performance. As shown in Figure 3, the analysis sample combines all telematics data entries of every unit survival life from the same-type equipment fleet (i.e. dozers, loaders, excavators). Each data entry include the following variables: 1) the observed failure hazard value 149-5 1836 for the dozers and 3315 for the backhoes that cover a one year observation period. Removal of the outliers resulted in reducing the data sample to 1767 and 3016 data entries for dozers and backhoes respectively. Five arbitrary life intervals were considered to develop the time varying hazard functions: less than 50 days; between 50 and 100 days, between 100 and 150 days; between 150 and 300 days, and more than 300 days. The data entries were split between these time intervals based on their survival life times, accordingly divided into estimation and validation data groups. Tables 1 and 2 list the data entries distribution over the life intervals, the estimated coefficients of the survival functions, and their prediction validation metrics. For example, 676 data entries of the dozers were located in the first life interval (0 – 50 days) and were split equally between the estimation and validation groups.  Equation 2 depicts an example of how the estimated coefficients shown in Table 1 can be used to formulate the time-varying hazard function of the dozers fleet.         [2]     [ ][ ][ ][ ][ ]≤⋅−⋅+⋅−⋅−<≤⋅+⋅+⋅−<≤⋅−⋅+⋅−⋅+⋅−<≤⋅+⋅−⋅+<≤⋅+⋅+⋅−=ttttEXPtEXPth300HI0.00515HW0.0079MOT0.0009MOP0.0132EXP300150HI0.00123MPT0.00537MOP0.0144EXP150100HI0.0013HW0.00156MPT0.0091MES0.00047MOP0.0145EXP10050HI0.00093MFR0.181MPT0.01835MOP1650.0-500HW0.0015MOT0.0077MOP0.0396)(     Table 1: Final Regression Results for the Dozers Hazard Functions Parameters Life Intervals (days) 0 – 50 50 – 100 100 – 150 150 – 300 > 300 Constant (C)  1 1 1 1 1 X1 (MCT) 0 -0.00848 a 0 0 0 X2 (MOP) -0.04917 a -0.0206 a -0.01491 a -0.00728 a 0 X3 (MOT) 0.01101 a 0.0096 a 0 0 0 X4 (MES) 0 0 0 0 -0.0001 a X5 (MPT) 0.02792 a 0.0135 b 0.01336 a 0.00659 a 0 X6 (MFR) -0.35908 a -0.1408 a -0.13776 a -0.09602 a 0 X7 (HW) 0.00364 a 0.00195 a 0.00193 a 0.00129 a 0.00016 X8 (HI) 0 0 0 0 0.00101 Estimation Observations 338 185 133 198 24 Multiple R 0.800193 0.854431 0.857417 0.845453 0.9064934 R Square 0.640309 0.730053 0.735165 0.714791 0.8217304 Adj. R Square 0.633072 0.716926 0.721035 0.705226 0.7571333 Significance F 1.054E-72 3.71E-48 1.46E-35 1.3E-51 7.37E-08 Prediction Observations 338 185 133 198 27 RMSE 0.2967 0.2696 0.223 0.1691 0.0499 Rcorr 0.3896 0.3745 0.4375 0.4965 0.7845 Observed t-Test  7.707 5.5236 5.6319 8.0085 6.3251 Critical t-Test  1.64912 1.65304 1.6563 1.65221 1.70562   a p < 0.001  b p < 0.05  c p < 0.01  149-7 Table 2: Final Regression Results for the Backhoes Hazard Functions Parameters Survival Intervals (days) 0 – 50 50 – 100 100 – 150 150 – 300 > 300 Constant (C)  1 1 1 1 1 X1 (MCT) 0 0 0 0 0 X2 (MOP) -0.0396 a -0.0165 a -0.0145 a -0.0144 a -0.0132 a X3 (MOT) 0.0077 a 0 0 0 -0.0009 X4 (MES) 0 0 0.00047 b 0 0 X5 (MPT) 0 0.01835 b -0.0091 b 0.00537 0 X6 (MFR) 0 -0.181 b 0 0 0 X7 (HW) 0.0015 a 0 0.00156 a 0 0.0079 a X8 (HI) 0 0.00093 -0.0013 b 0.00123 a -0.00515 a Estimation Observations 664 305 236 241 62 Multiple R 0.7687809 0.8111306 0.7679307 0.7890777 0.9518655 R Square 0.5910241 0.6579329 0.5897176 0.6226436 0.9060479 Adj. R Square 0.5882738 0.6512013 0.5782841 0.6152708 0.8839470 Significance F 8.52E-128 9.4E-69 1.03E-42 5E-50 9.16E-29 Prediction Observations 665 305 236 241 61 RMSE 0.374626 0.23463 0.22687 0.26875 0.1 Rcorr 0.119655 0.3971456 0.32333 0.49067 0.878159 Observed t-Test  3.103263 7.5326 5.22682 8.7055 10.7 Critical t-Test  1.647 1.65 1.651 1.651 1.671 a p < 0.001  b p < 0.05  c p < 0.01  The examination of the developed hazard functions provides useful insights on the dependency of equipment health prognosis on the telematics parameters over the successive life intervals. First, some telematics parameters have shown consistent correlation with the failure hazard probability for both analyzed equipment. The maximum oil pressure (MOP) and working hours (HW) variables were found to be a very significant health prognosis variables (p < 0.001) in the developed hazard function in most life interval. Also, it was consistently shown that higher MOP values result in lower failure hazard (all of its coefficients were negative) as it indicate a lower possibility of oil leakage. On the other hand, the positive coefficients of the working hours (HW) variable reinforce the concept that higher working hours increases the equipment’s failure hazard. Second, the fitting and prediction accuracy of the developed hazard functions were found to increase with the survival time of the equipment. This observation is illustrated by the high multiple R value, the high Rcorr value, and low RMSE value of the last life interval for both equipment types compared to earlier life intervals. This complies with the de facto that the failure of assets with longer survival time can be reliably anticipated compared to younger assets with shorter survival time that just recovered from a failure event.          7 CONCLUSION AND FUTURE RESEARCH This paper presents the development of a telematics-based equipment prognostics system that can support heavy construction and rental companies to effectively manage their predictive maintenance programs. Telematics is an efficient data collection technology as it integrates equipment onboard 149-8 sensing, global positioning, and telecommunication to provide a timely reporting of different equipment performance data. The telematics CAN-bus data provide a rich source of metrics that can be used to diagnose the equipment health. A new survival analysis methodology is proposed to develop the equipment hazard functions utilizing the available telematics data to: 1) identify equipment failure events and survival lives using the red and amber engine lights; 2) quantify the observed failure hazard probability using the reported engine hours; and 3) estimate the regression coefficients of the failure hazard covariates that are evaluated from a set of proposed telematics health parameters.      The proposed prognostics system was successful in developing the hazard functions of the analyzed equipment fleets that provided high fit to the observed data and high prediction accuracy for the testing data. For both analyzed fleets, higher predictive and data fitting performance were achieved for later life intervals due to the high tear and wear levels that results in increased failure probability. The proposed telematics-based prognostics system should prove useful to equipment fleet managers to successfully implement predictive maintenance programs. The system would expand the current uses of telematics systems by transforming its timely data into useful decision making information. Future possible research venues of the developed system includes: 1) developing alternative telematics data-driven prognostics systems that utilize other statistical and probabilistic analysis approaches, such as logistic regression and fuzzy clustering; and 2) implementing the proposed prognostics system into an automated online or desktop prototype module that can be integrated into available telematics systems and fleet management operations.      References Aslan B. and Koo D. (2012). “Productivity Enhancement for Maintenance Equipment Operations using Telematics Technology.” Construction Research Congress, American Society of Civil Engineers (ASCE), pp. 971 – 980, May 21 - 23, West Lafayette, IN. Bailey W. J., Weir-Jones I., Couet B., and Hogan J. R. (2006). “Survival Analysis: The Statistically Rigorous Method for Analyzing Electrical Submersible Pump System Performance." SPE Annual Technical Conference and Exhibition, Society of Petroleum Engineers, Dallas, Texas October 9 – 12. Allison, P.D.(1995). Survival Analysis Using the SAS System: A Practical Guide. SAS Institute. Cox D. R. (1972). “Regression Models and Life-Tables” Journal of the Royal Statistical Society. Series B (Methodological), Vol. 34, No. 2. (1972), pp.187-220. Da Y., Shi X. and Krishnamurthy M. (2011). "Health monitoring, fault diagnosis and failure prognosis techniques for Brushless Permanent Magnet Machines," Proceedings of Vehicle Power and Propulsion Conference (VPPC), IEEE , Chicago, IL, pp. 1 – 7, 6-9 Sept. 2011. Dekate D. A. (2013). “Prognostics and Engine Health Management of Vehicle using Automotive Sensor Systems.” International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064, 2(2),  Dutta S. and Giurgiutiu V. (2000). “Health Monitoring and Quality Assurance for Rotary Micro-Machines and Active Sensors.” 8th International Symposium on Transport Phenomena and Dynamics of Rotating Machinery (ISROMAC-8), Pacific Center of Thermal-Fluids Engineering, Honolulu, Hawaii, March 26-30. Field, A. (2005). Discovering statistics using SPSS, 2nd Ed., Sage, London. Gransberg D., Popescu C., and Ryan R. (2006). Construction Equipment Management for Engineers, Estimators, and Owners. CRC Press, Taylor and Francis Group, Boca Raton, FL. Jackson T. (2012). “Off-Road Telematics: Why the Disconnect?” Equipment World, online magazine, Randall-Riley, August 2013 <http://www.equipmentworld.com/maintenance-23/> last accessed 11/27/2013.  Lee J., Ni J., Djurdjanovic D., Qiu H., and Liao H. (2006). “Intelligent prognostics tools and e-maintenance.” Computers in Industry, 57, 476 – 489.  Lovett H., Melby E., Myren N., and Nielsen J. S. (2003) “The Telematics Business Landscape.” Teletronikk, Telenor Group, Vol. 1, 2003, pp. 83 – 89. Lucko G., Anderson-Cook C. M., and Vorster M. C. (2006). “Statistical Considerations for Predicting Residual Value of Heavy Equipment.” Journal of Construction Engineering and Management, ASCE, 132(7), pp. 723 – 732. 149-9 Lucko G. and Rojas E. M. (2010). “Research Validation: Challenges and Opportunities in the Construction Domain.” Journal of Construction Engineering and Management, ASCE, 136(1), pp. 127 – 135.  Ma Z. and Krings A.W. (2008). "Survival Analysis Approach to Reliability, Survivability and Prognostics and Health Management (PHM)," Aerospace Conference, 2008 IEEE , pp. 1 - 20, 1-8 March 2008. doi: 10.1109/AERO.2008.4526634 Mesgarpour M., Landa-Silva D., and Dickinson I. (2013). “Overview of Telematics-Based Prognostics and Health Management Systems for Commercial Vehicles.” 13th International Conference on Transport Systems Telematics, TST 2013, Katowice-Ustroń, Poland, October 23–26, 2013. Monnot J. and Williams R. (2011). “Construction Equipment Telematics.” Journal of Construction Engineering and Management, ASCE, 137(10), pp. 793 – 796. Montgomery, D. C., Peck, E. A., and Vining, C. G. 2001. Introduction to linear regression analysis, 3rd Ed., Wiley, New York. Murakami T., Saigo T., Ohkura Y., Okawa Y. and Taninaga T. (2002). “Development of Vehicle Health Monitoring System (VHMS/WebCARE) for Large-Sized Construction Machine.” Komatsu’s Technical Report, Komatsu, 48(150), 15 – 21. Sutton R. (2013). “Turbo Telematics.” Construction Equipment, online magazine, <http://www.constructionequipment.com/blog/turbo-telematics> last accessed 11/27/2013.  Thomson M. (2013). “Monitoring Machine Health.” Midstream Business Magazine, HART ENERGY, January 2013, pp 87 – 89. Trimble T. and Bowman D. (2012). Market Guide to Fleet Telematics Services. Report 12-UT-028, National Surface Transportation Safety Center for Excellence, Virginia Tech Transportation Institute, Blacksburg, Virginia. Yan R. and Gao R. (2007). “Approximate Entropy as a Diagnostic Tool for Machine Health Monitoring.” Mechanical Systems and Signal Processing 149-10  5th International/11th Construction Specialty Conference 5e International/11e Conférence spécialisée sur la construction    Vancouver, British Columbia June 8 to June 10, 2015 / 8 juin au 10 juin 2015   TELEMATICS DATA-DRIVEN PROGNOSTICS SYSTEM FOR CONSTRUCTION HEAVY EQUIPMENT HEALTH MONITORING AND ASSESSMENT Hisham M. Said1,3, 4, and Tony Nicoletti 2 1 Department of Civil Engineering, Santa Clara University, USA.  2 Director, Sales and Business Development, DPL America, USA 3 Adjunct lecturer, Structural Engineering Department, Cairo University, Egypt 4 hsaid@scu.edu  Abstract: Construction heavy equipment is a valuable asset for construction and equipment rental companies, which requires continuous monitoring and assessment for potential failures. Predictive maintenance has recently been proposed to as an alternative to preventive maintenance strategy by scheduling maintenance tasks just before a predicted failure of the equipment. Such predictive approach is dependent on the existence of a data collection and analysis system that monitors the equipment performance, compares it to the previous history, and predicts the failure events before their occurrence. This paper presents the development and validation efforts of a data-driven prognostics system that utilizes timely collected telematics data to monitor the equipment health condition and predict its failure hazard. The system is designed to utilize equipment telematics data to develop regression-based Cox’s proportional hazards functions. Regression analyses are performed for the historical telematics data to develop time-varying hazard functions for the successive life intervals of the equipment to generate dynamic predictions of its failure events. Accordingly, the outcome of the system would be the predicted probability of the equipment failure event considering the timely collected telematics data. The proposed prognostics system was validated by developing the hazard functions of two fleets of dozers and backhoes that provided high fit to the observed data and high prediction accuracy for the testing data. For both analyzed fleets, higher predictive and data fitting performance were achieved for later life intervals due the increased reliability of failure prediction for equipment with longer survival lives.   1 INTRODUCTION Construction heavy equipment is a valuable asset for construction and equipment rental companies, which requires continuous monitoring and assessment for potential failures. The absence of a properly implemented maintenance program leads to premature equipment failure and increased construction crew idle time. Predictive maintenance (Gransberg et al. 2006) has recently been proposed as an alternative to corrective and preventive maintenance strategies by scheduling maintenance tasks just before a predicted failure of the equipment. Such predictive approach is dependent on the existence of a data collection and analysis system that monitors the equipment performance, compares it to previous history, and predicts the failure events before their occurrence. Despite the great promise of predictive maintenance, its wide implementation was not realized yet due to its need for large data collection process and supporting technology. 149-1 Telematics is a data collection technology that integrates wireless communications, vehicle monitoring systems, and location devices to provide real-time spatial and performance tracking of the fleet machines (Lovett et al. 2003). As a witness to its great benefits, the telematics industry has significantly grown to be installed in 5.8 million equipment units with revenue volume of $2 billion in 2009 (Fletcher and Lauron 2009). Telematics can be installed by either the original equipment manufacturer (OEM) or a third-party service provider (TSP). Telematics has been utilized as mainly a real-time monitoring system of equipment fleet for the purposes of theft protection, fuel consumption, and prevention of undesired behaviors of operators/drivers. Fleet managers are challenged to expand the utilization of telematics technology due to the difficulties in inking telematics data to business functions and performance metrics (Monnot and Williams 2011, Trimble and Bowman 2012, Jackson 2012, Sutton 2013).       2 RESEARCH OBJECTIVE AND METHODOLOGY This paper presents the development and validation efforts of a data-driven prognostics system that utilizes timely collected telematics data to monitor the equipment health condition and predict its failure hazard. Prognostics is the field and methodologies of predicting the future health behavior, failure events, and remaining useful like (RUL) of equipment and machines by diagnosing the recorded temporal behavior (Mesgarpour et al. 2013). Current approaches of prognostics can be classified into three main classes (Lee et al. 2006): 1) model-based prognostics that depends on developing virtual models of the machine that mimic its behavior under healthy and faulty conditions; 2) data-driven prognostics that utilize collected sensor data of the machine’s previous behavior toward failure, and 3) hybrid prognostics that ingrates models formulation with sensor data calibration. This paper proposes a data-driven prognostics model that utilizes equipment telematics data to estimate its failure probability.    To accomplish this objective, the research methodology encompassed four main tasks. First, a thorough literature review was performed to study the previous research in the areas of equipment health prognostics and telematics application in construction. Second, a brief description of telematics system architecture and data was presented. Third, the formulation of the proposed telematics-based prognostics system was developed and illustrated. Fourth, the performance of the developed system was validated by analyzing the telematics data of two types of heavy equipment fleets. The paper is concluded by summarizing the contribution of the proposed system to heavy construction and equipment rental companies, as well as recommendations for future research.       3 PREVIOUS RESEARCH Relevant previous research related to this paper is summarized into two main categories: equipment health prognostics and application of equipment telematics in construction. First, previous research on equipment and machine prognostics focused on monitoring the condition of stationary mechanical machines or electrical micro-machines by majorly measuring their vibrations (Dutta and Giurgiutiu 2000; Yan R. and Gao 2007; Da et al. 2011, Thomson 2013). Equipment manufacturers have encouraged research and development efforts to develop health monitoring systems that integrate remote sensing and equipment oil sampling to diagnose its condition and estimate its life expectancy (Murakami et al. 2002). Little research was found to indicate the potential of utilizing telematics data in prognostics, with no development of proven models or systems (Mesgarpour et al. 2013).  Second, little research was performed to investigate the use of the telematics technology in construction and heavy equipment fleet management. Monnot and Williams (2011) highlighted the possible use of telematics in various equipment fleet management tasks, like reporting of machines hours, locations, fuel consumption, and health. Aslan and Koo (2012) proposed an implementation plan for the use of telematics technology in the improving the productivity of roadway maintenance operations. The plan was to me completed in a future study with testing a telematics data collection system and developing productivity measurement metrics.     The careful study of previous studies identified the critical research gap and need for new prognostics systems that would enable the use of telematics data to assess equipment failure potential, as an essential part of effective predictive maintenance (Gransberg et al. 2006).  149-2 study the telematics data and collect its samples for the development and validation of the proposed heavy equipment prognostics system.    TransponderUnitTelematics Data Sent through  Wireless AntennaLocation data Received by GPS ReceiverJ1939 Data (engine speed, check lamps, oil pressure, fluids temperature) received through  CAN-bus ConnectionBasic Data Received (local temperature, battery voltage, engine runtime) through Main Interface CablesCAN-busControl Units Figure 2: Telematics Data Types 5 EQUIPMENT HEALTH PROGNOSTICS USING TELEMATICS-BASED SURVIVAL ANALYSIS The development of the proposed equipment prognostics system is presented by listing the proposed telematics health data, providing an overview of survival analysis, and describing the modeling approach of the equipment failure hazard functions.    5.1 Proposed Telematics-Based Health Parameters Telematics provides a rich data source that is utilized in this research to derive diagnosis and prognosis parameters of the equipment health. The proposed prognostics system is generic and can be applied to any set of telematics health parameters available in the collected data. However, the inputted telematics health parameters may affect the quality of the generated equipment health hazard functions. Accordingly, the following eight health parameters available in every telematics entry at time t, were proposed in this research based on available literature review (Murakami et al. 2002, Dekate 2013) and consultation with equipment telematics professionals:   1. Maximum coolant temperature (MCTt) in degrees Fahrenheit, which is observed on the day when the telematics data entry is received. 2. Maximum engine oil pressure (MOPt) in pounds per square inch (psi). 3. Maximum engine oil temperature (MOTt) in degrees Fahrenheit. 4. Maximum engine speed (MESt), in rounds per minute (rpm). 5. Maximum engine percent torque (MPTt), which indicates the load on the engine as a percentage value. 6. Maximum fuel rate (MFRt) in gallons/hour. 7. Engine working hours (HWt), which reports the cumulative number of hours the engine run with a speed (rpm) above a specified threshold, set by the fleet manager. 8. Engine Idling hours (HIt), which reports the cumulative number of hours the engine ran with a speed (rpm) less than the specified threshold.  149-4 5.2 Survival Analysis and Failure Hazard Functions Survival analysis is a regression approach of reliability studies to assess the times and probabilities to failure events. Survival analysis has been applied before to analyze the lifetime of orgasms, survival times of cancer patients, occurrence of accidents, and failure times of machines (Gu et al. 2011). Survival analysis deals with the failure event time as a random variable T using different representations (Allison 1995), such as 1) the cumulative distribution function of variable T, P(t) = Pr(T≤t); 2) the survival function S(t) as the complement of the distribution function, S(t) = Pr(T>t) = 1 – P(t); and 3) the hazard function h(t) that assesses the instantaneous at time t. The goal of the survival analysis is develop these representations of the failure event as a function of its determinant, i.e. its health parameters.    Survival analysis models can be classified as non-parametric, parametric, and semi-parametric models (Ma and Krings 2008). Cox’s proportional hazards model (Cox 1972), one of the fundamental semi-parametric survival models, is utilized for this study due to its ability to capture the failure covariates more effectively than non-parametric models with less modeling restriction like the parametric models (Bailey et al. 2006). As shown in Equation 1, a dynamic time-varying hazard function h(t) of Cox’s model is utilized in the proposed equipment prognostics system to estimate the probability of equipment failure at time t, which is based on: 1) the telematics health parameters X(t) (covariates) and their coefficients β(t) that can be determined by performing a regression analysis to fit the observed health parameter values to the failure hazard probability h(t); and 2) a baseline failure rate h0(t) represents the decay of the equipment health regardless of the values of its health parameters. The next section will explain in more details the methodology of developing the hazard functions and the calculation of the observed equipment hazard probability based on the available telematics data.  [1] h(t) = h0(t) × exp [β(t).X(t)]        5.3 Modeling and Development of Equipment Failure Hazard Functions The proposed heavy equipment prognostics system follows a novel methodology to model and develop Cox’s proportional survival functions as a function of the available telematics data. As shown in Figure 3, the equipment hazard functions are developed in four main steps:  Step (a) – Telematics data are used to mark the failure events over the lifetime of the equipment and the engine hours. An equipment failure event is recognized if one of two engine lights is reported in the telematics data entry: 1) red stop light (RSL) that indicates a severe enough condition that it warrants stopping the vehicle; and 2) amber warning light (AWL) that reports a problem with the vehicle system but the vehicle does not need to be immediately stopped. Other engine lights, such as engine protection light (EPL) and malfunction indicator light (MIL), are not used to indicate a major equipment failure as they report less severe health conditions related to the vehicle electronic system and emissions-related issues. As shown in Figure 2, the existence of either RSL or AWL marks a failure event and the start/end of a survival life of the equipment. As shown in Figure 3, survival analysis permits the right and left censoring (Ma and Krings 2008) of the data to only consider the complete survival lives that start and end with a recorded failure/survival event.   Step (b) – The observed failure hazard value is quantified using another telematics data element, which is the engine total hours (HT) that reports the cumulative engine hours up to the time of the telematics data entry. As shown in Figure 2, the observed failure hazard is quantified over each of the identified survival lives as the ratio between: 1) the engine hours Et since the start of the survival life to time t; and 2) the total engine hours L occurred during corresponding survival life. Accordingly, the failure hazard observed values range from 0 (at the beginning of the survival life) to 1.0 (at the end of the survival life).     Step (c) – All telematics entries are combined to form the analysis data population and divided into survival intervals, which each will have its own hazard function to represent the time-varying nature of equipment health performance. As shown in Figure 3, the analysis sample combines all telematics data entries of every unit survival life from the same-type equipment fleet (i.e. dozers, loaders, excavators). Each data entry include the following variables: 1) the observed failure hazard value 149-5 1836 for the dozers and 3315 for the backhoes that cover a one year observation period. Removal of the outliers resulted in reducing the data sample to 1767 and 3016 data entries for dozers and backhoes respectively. Five arbitrary life intervals were considered to develop the time varying hazard functions: less than 50 days; between 50 and 100 days, between 100 and 150 days; between 150 and 300 days, and more than 300 days. The data entries were split between these time intervals based on their survival life times, accordingly divided into estimation and validation data groups. Tables 1 and 2 list the data entries distribution over the life intervals, the estimated coefficients of the survival functions, and their prediction validation metrics. For example, 676 data entries of the dozers were located in the first life interval (0 – 50 days) and were split equally between the estimation and validation groups.  Equation 2 depicts an example of how the estimated coefficients shown in Table 1 can be used to formulate the time-varying hazard function of the dozers fleet.         [2]     [ ][ ][ ][ ][ ]≤⋅−⋅+⋅−⋅−<≤⋅+⋅+⋅−<≤⋅−⋅+⋅−⋅+⋅−<≤⋅+⋅−⋅+<≤⋅+⋅+⋅−=ttttEXPtEXPth300HI0.00515HW0.0079MOT0.0009MOP0.0132EXP300150HI0.00123MPT0.00537MOP0.0144EXP150100HI0.0013HW0.00156MPT0.0091MES0.00047MOP0.0145EXP10050HI0.00093MFR0.181MPT0.01835MOP1650.0-500HW0.0015MOT0.0077MOP0.0396)(     Table 1: Final Regression Results for the Dozers Hazard Functions Parameters Life Intervals (days) 0 – 50 50 – 100 100 – 150 150 – 300 > 300 Constant (C)  1 1 1 1 1 X1 (MCT) 0 -0.00848 a 0 0 0 X2 (MOP) -0.04917 a -0.0206 a -0.01491 a -0.00728 a 0 X3 (MOT) 0.01101 a 0.0096 a 0 0 0 X4 (MES) 0 0 0 0 -0.0001 a X5 (MPT) 0.02792 a 0.0135 b 0.01336 a 0.00659 a 0 X6 (MFR) -0.35908 a -0.1408 a -0.13776 a -0.09602 a 0 X7 (HW) 0.00364 a 0.00195 a 0.00193 a 0.00129 a 0.00016 X8 (HI) 0 0 0 0 0.00101 Estimation Observations 338 185 133 198 24 Multiple R 0.800193 0.854431 0.857417 0.845453 0.9064934 R Square 0.640309 0.730053 0.735165 0.714791 0.8217304 Adj. R Square 0.633072 0.716926 0.721035 0.705226 0.7571333 Significance F 1.054E-72 3.71E-48 1.46E-35 1.3E-51 7.37E-08 Prediction Observations 338 185 133 198 27 RMSE 0.2967 0.2696 0.223 0.1691 0.0499 Rcorr 0.3896 0.3745 0.4375 0.4965 0.7845 Observed t-Test  7.707 5.5236 5.6319 8.0085 6.3251 Critical t-Test  1.64912 1.65304 1.6563 1.65221 1.70562   a p < 0.001  b p < 0.05  c p < 0.01  149-7 Table 2: Final Regression Results for the Backhoes Hazard Functions Parameters Survival Intervals (days) 0 – 50 50 – 100 100 – 150 150 – 300 > 300 Constant (C)  1 1 1 1 1 X1 (MCT) 0 0 0 0 0 X2 (MOP) -0.0396 a -0.0165 a -0.0145 a -0.0144 a -0.0132 a X3 (MOT) 0.0077 a 0 0 0 -0.0009 X4 (MES) 0 0 0.00047 b 0 0 X5 (MPT) 0 0.01835 b -0.0091 b 0.00537 0 X6 (MFR) 0 -0.181 b 0 0 0 X7 (HW) 0.0015 a 0 0.00156 a 0 0.0079 a X8 (HI) 0 0.00093 -0.0013 b 0.00123 a -0.00515 a Estimation Observations 664 305 236 241 62 Multiple R 0.7687809 0.8111306 0.7679307 0.7890777 0.9518655 R Square 0.5910241 0.6579329 0.5897176 0.6226436 0.9060479 Adj. R Square 0.5882738 0.6512013 0.5782841 0.6152708 0.8839470 Significance F 8.52E-128 9.4E-69 1.03E-42 5E-50 9.16E-29 Prediction Observations 665 305 236 241 61 RMSE 0.374626 0.23463 0.22687 0.26875 0.1 Rcorr 0.119655 0.3971456 0.32333 0.49067 0.878159 Observed t-Test  3.103263 7.5326 5.22682 8.7055 10.7 Critical t-Test  1.647 1.65 1.651 1.651 1.671 a p < 0.001  b p < 0.05  c p < 0.01  The examination of the developed hazard functions provides useful insights on the dependency of equipment health prognosis on the telematics parameters over the successive life intervals. First, some telematics parameters have shown consistent correlation with the failure hazard probability for both analyzed equipment. The maximum oil pressure (MOP) and working hours (HW) variables were found to be a very significant health prognosis variables (p < 0.001) in the developed hazard function in most life interval. Also, it was consistently shown that higher MOP values result in lower failure hazard (all of its coefficients were negative) as it indicate a lower possibility of oil leakage. On the other hand, the positive coefficients of the working hours (HW) variable reinforce the concept that higher working hours increases the equipment’s failure hazard. Second, the fitting and prediction accuracy of the developed hazard functions were found to increase with the survival time of the equipment. This observation is illustrated by the high multiple R value, the high Rcorr value, and low RMSE value of the last life interval for both equipment types compared to earlier life intervals. This complies with the de facto that the failure of assets with longer survival time can be reliably anticipated compared to younger assets with shorter survival time that just recovered from a failure event.          7 CONCLUSION AND FUTURE RESEARCH This paper presents the development of a telematics-based equipment prognostics system that can support heavy construction and rental companies to effectively manage their predictive maintenance programs. Telematics is an efficient data collection technology as it integrates equipment onboard 149-8 sensing, global positioning, and telecommunication to provide a timely reporting of different equipment performance data. The telematics CAN-bus data provide a rich source of metrics that can be used to diagnose the equipment health. A new survival analysis methodology is proposed to develop the equipment hazard functions utilizing the available telematics data to: 1) identify equipment failure events and survival lives using the red and amber engine lights; 2) quantify the observed failure hazard probability using the reported engine hours; and 3) estimate the regression coefficients of the failure hazard covariates that are evaluated from a set of proposed telematics health parameters.      The proposed prognostics system was successful in developing the hazard functions of the analyzed equipment fleets that provided high fit to the observed data and high prediction accuracy for the testing data. For both analyzed fleets, higher predictive and data fitting performance were achieved for later life intervals due to the high tear and wear levels that results in increased failure probability. The proposed telematics-based prognostics system should prove useful to equipment fleet managers to successfully implement predictive maintenance programs. The system would expand the current uses of telematics systems by transforming its timely data into useful decision making information. Future possible research venues of the developed system includes: 1) developing alternative telematics data-driven prognostics systems that utilize other statistical and probabilistic analysis approaches, such as logistic regression and fuzzy clustering; and 2) implementing the proposed prognostics system into an automated online or desktop prototype module that can be integrated into available telematics systems and fleet management operations.      References Aslan B. and Koo D. (2012). “Productivity Enhancement for Maintenance Equipment Operations using Telematics Technology.” Construction Research Congress, American Society of Civil Engineers (ASCE), pp. 971 – 980, May 21 - 23, West Lafayette, IN. Bailey W. J., Weir-Jones I., Couet B., and Hogan J. R. (2006). “Survival Analysis: The Statistically Rigorous Method for Analyzing Electrical Submersible Pump System Performance." SPE Annual Technical Conference and Exhibition, Society of Petroleum Engineers, Dallas, Texas October 9 – 12. Allison, P.D.(1995). Survival Analysis Using the SAS System: A Practical Guide. SAS Institute. Cox D. R. (1972). “Regression Models and Life-Tables” Journal of the Royal Statistical Society. Series B (Methodological), Vol. 34, No. 2. (1972), pp.187-220. Da Y., Shi X. and Krishnamurthy M. (2011). "Health monitoring, fault diagnosis and failure prognosis techniques for Brushless Permanent Magnet Machines," Proceedings of Vehicle Power and Propulsion Conference (VPPC), IEEE , Chicago, IL, pp. 1 – 7, 6-9 Sept. 2011. Dekate D. A. (2013). “Prognostics and Engine Health Management of Vehicle using Automotive Sensor Systems.” International Journal of Science and Research (IJSR), India Online ISSN: 2319-7064, 2(2),  Dutta S. and Giurgiutiu V. (2000). “Health Monitoring and Quality Assurance for Rotary Micro-Machines and Active Sensors.” 8th International Symposium on Transport Phenomena and Dynamics of Rotating Machinery (ISROMAC-8), Pacific Center of Thermal-Fluids Engineering, Honolulu, Hawaii, March 26-30. Field, A. (2005). Discovering statistics using SPSS, 2nd Ed., Sage, London. Gransberg D., Popescu C., and Ryan R. (2006). Construction Equipment Management for Engineers, Estimators, and Owners. CRC Press, Taylor and Francis Group, Boca Raton, FL. Jackson T. (2012). “Off-Road Telematics: Why the Disconnect?” Equipment World, online magazine, Randall-Riley, August 2013 <http://www.equipmentworld.com/maintenance-23/> last accessed 11/27/2013.  Lee J., Ni J., Djurdjanovic D., Qiu H., and Liao H. (2006). “Intelligent prognostics tools and e-maintenance.” Computers in Industry, 57, 476 – 489.  Lovett H., Melby E., Myren N., and Nielsen J. S. (2003) “The Telematics Business Landscape.” Teletronikk, Telenor Group, Vol. 1, 2003, pp. 83 – 89. Lucko G., Anderson-Cook C. M., and Vorster M. C. (2006). “Statistical Considerations for Predicting Residual Value of Heavy Equipment.” Journal of Construction Engineering and Management, ASCE, 132(7), pp. 723 – 732. 149-9 Lucko G. and Rojas E. M. (2010). “Research Validation: Challenges and Opportunities in the Construction Domain.” Journal of Construction Engineering and Management, ASCE, 136(1), pp. 127 – 135.  Ma Z. and Krings A.W. (2008). "Survival Analysis Approach to Reliability, Survivability and Prognostics and Health Management (PHM)," Aerospace Conference, 2008 IEEE , pp. 1 - 20, 1-8 March 2008. doi: 10.1109/AERO.2008.4526634 Mesgarpour M., Landa-Silva D., and Dickinson I. (2013). “Overview of Telematics-Based Prognostics and Health Management Systems for Commercial Vehicles.” 13th International Conference on Transport Systems Telematics, TST 2013, Katowice-Ustroń, Poland, October 23–26, 2013. Monnot J. and Williams R. (2011). “Construction Equipment Telematics.” Journal of Construction Engineering and Management, ASCE, 137(10), pp. 793 – 796. Montgomery, D. C., Peck, E. A., and Vining, C. G. 2001. Introduction to linear regression analysis, 3rd Ed., Wiley, New York. Murakami T., Saigo T., Ohkura Y., Okawa Y. and Taninaga T. (2002). “Development of Vehicle Health Monitoring System (VHMS/WebCARE) for Large-Sized Construction Machine.” Komatsu’s Technical Report, Komatsu, 48(150), 15 – 21. Sutton R. (2013). “Turbo Telematics.” Construction Equipment, online magazine, <http://www.constructionequipment.com/blog/turbo-telematics> last accessed 11/27/2013.  Thomson M. (2013). “Monitoring Machine Health.” Midstream Business Magazine, HART ENERGY, January 2013, pp 87 – 89. Trimble T. and Bowman D. (2012). Market Guide to Fleet Telematics Services. Report 12-UT-028, National Surface Transportation Safety Center for Excellence, Virginia Tech Transportation Institute, Blacksburg, Virginia. Yan R. and Gao R. (2007). “Approximate Entropy as a Diagnostic Tool for Machine Health Monitoring.” Mechanical Systems and Signal Processing 149-10  Telematics Data-driven Prognostics System for Construction Heavy Equipment Health Monitoring And Assessment Hisham Said, Ph.D. Santa Clara University           5th International/11th Construction Specialty ConferenceJune 10, 2015Tony NicolettiDPL America           Outline Introduction Research Need and Objective Telematics Overview – System Architecture And Data Telematics-Based Prognostics System System Validation Conclusion & Future Research2Overview Heavy equipment is a vital and expensive asset. Effective maintenance program is critical in heavy construction companies.Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchTypes of Equipment Maintenance ProgramsReactive (breakdown) Maintenance:Get the broken machine up and running asquickly as possible. Don’t touch running equipment!Proactive (Preventive) Maintenance:Periodic inspection, lubrication, and replacement of worn parts, filters, and fluids.Predictive Maintenance:Schedule maintenance tasks based on the past performance of engine parts.Example: Engine Oil AnalysisGransberg et al. 2006 3Overview Preventive Maintenance analysis = Prognostics  Prognostics is the field of predicting the future health behavior, failure events, and remaining useful life (RUL). (Mesgarpour et al. 2013)Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchLee et al. 2006PrognosticsModel-Based Data-DrivenHybridVirtual models of the machine behavior under healthy and faulty conditionsUtilize collected sensor data of the machine’s previous behavior toward failureIntegrating model formulation with sensor data calibrationPrognostics Previous Research: Stationary machines health assessment by vibrations monitoring (Dutta and Giurgiutiu2000; Yan R. and Gao 2007; Da et al. 2011, Thomson 2013) Integration of sensors and oil samples for prognostics (Murakami et al. 2002)4Research Need and Objective Preventive maintenance (data-driven prognostics) is efficient, but requires significant data collection and analysis processes    Rich data source …….. TELEMATICS!Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future Researchhttp://www.eetasia.com/ART_8800591622_499495_NT_410c8916.HTM Telematics = GPS + machine sensing system + wireless communication. Fleet managers are challenged to link telematics data to fleet functions and performance metrics.  Little research on construction equipment telematics (Monnot and Williams 2011, Aslan and Koo 2012) Research Objective: Develop and validate a data-driven prognostics System that utilizes equipment telematics data to estimate its failure probability.5Remote shutoff & controlUtilization, Location, Alarms, CAN-BUSWirelessNetwork ServersCommunication MediumGPS SatellitesTelematics OverviewIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchWireless Devices forreal time notification• Web Software 24/7 multi-user access• Data Download to customer programsUser Interface Typical System Architecture Telematics installed in equipment by:o Original equipment manufacturer (OEM) o Third-party service provider (TSP)Transponder Units on Fleet Assets 6Telematics OverviewIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchCAN-busControl Units TransponderUnitTelematics Data Sent through  Wireless AntennaLocation data Received by GPS ReceiverJ1939 Data (engine speed, check lamps, oil pressure, fluids temperature) received through  CAN-bus ConnectionBasic Data Received (local temperature, battery voltage, engine runtime) through Main Interface CablesTelematics Data – Types and Collection7Prognostics System Survival Analysis: a regression approach to assess the times and probabilities to failure events, in terms of its independent variables. Cox’s proportional hazards function h(t) to estimate the failure rate at time t.Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future Researchh(t) = h0(t) × exp [β(t).X(t)]h0(t) : baseline failure rate (asset decay over time)X(t) : failure/health parametersβ(t) : regression coefficientsh(t)timeHazard Function(Cox 1972)8Prognostics SystemSuggested Health Parameters X(t) : 1) Maximum coolant temperature (MCTt) in degrees Fahrenheit2) Maximum engine oil pressure (MOPt) in pounds per square inch (psi).3) Maximum engine oil temperature (MOTt) in degrees Fahrenheit.4) Maximum engine speed (MESt), in rounds per minute (rpm).5) Maximum engine percent torque (MPTt).6) Maximum fuel rate (MFRt) in gallons/hour.7) Engine working hours (HWt), which reports the cumulative number of hours the engine run with a speed (rpm) above a specified threshold.8) Engine Idling hours (HIt), which reports the cumulative number of hours the engine ran with a speed (rpm) less than the specified threshold.Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future Research(Murakami et al. 2002, Dekate 2013)9tLife Intervals and their Hazard Functions Engine Total Hours (HT)AWLTime (days)Survival Life (1) Survival Life (2)Right-Censored DatatEtLFailure 1Failure 2Failure 3h(t)h(t)Left-Censored DataEquipment (2)h(t) = Et  / L Prognostics SystemIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchEngine Total Hours (HT)Time (days)Right-Censored DataLeft-Censored DataRSLAWLEquipment (1)Survival Life (1) Survival Life (2)Failure 1Failure 2Failure 3tEtLh(t) = Et  / L h(t)h(t)tTelematics Health Parameters1) Max Coolant Temp (MCTt)2) Max Oil Pressure (MOPt)3) Max Oil Temp (MOTt)4) Max Engine Speed (MESt)5) Max Percent Torque (MPTt)6) Max Fuel Rate (MFRt)7) Engine Work Hours (HWt)8) Engine Idle Hours (HIt)Apply Data Linearization Regression to Find the Coefficients of Hazard FunctionsCombine telematics entries of all equipment and assign to their life periodsh1(t) h2(t) h3(t)t1 t2 t3All Fleet Equipment of the Same Type10System Validation The methodology was applied to: 21 dozers (1836 telematics data entries) 29 backhoes (3315 telematics data entries) 17 trucks (3880 telematics data entries) The data is divided into 5 survival intervals: 1) less than 50 days2) between 50 and 100 days3) between 100 and 150 days4) between 150 and 300 days5) more than 300 daysIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchSurvival Interval0 < t < 5050 < t < 100100 < t < 150150 < t < 300300 < tEstimation Data Group 338 185 133 198 24Prediction Data Group 338 185 133 198 27The data of each survival interval is divided into 2 main groups:- Estimation Data Group (Coeff. Values)- Prediction Data Group (System Validation)Dozers Prognostics Data(1836 telematics data entries)11System ValidationDozers Hazard Functions (Estimation Data Group = 878 telematics data entries)Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchParametersSurvival Intervals (days)0 < t < 50 50 < t < 100 100 < t < 150 150 < t < 300 300 < tValue P-Value Value P-Value Value P-Value Value P-Value Value P-ValueConstant (C) 1 N/A 1 N/A 1 N/A 1 N/A 1 N/AX1 (MCT) 0 N/A -0.00848 0.00002 0 N/A 0 N/A 0 N/AX2 (MOP) -0.04917 0 -0.0206 0 -0.01491 0 -0.00728 0 0 N/AX3 (MOT) 0.01101 0.0003 0.0096 0.00016 0 N/A 0 N/A 0 N/AX4 (MES) 0 N/A 0 N/A 0 N/A 0 N/A -0.0001 0X5 (MPT) 0.02792 0.0007 0.0135 0.0061 0.01336 0.00035 0.00659 0.00093 0 N/AX6 (MFR) -0.35908 0 -0.1408 0.00001 -0.13776 0 -0.09602 0 0 N/AX7 (HW) 0.00364 0 0.00195 0 0.00193 0 0.00129 0 0.00016 0.09811X8 (HI) 0 N/A 0 N/A 0 N/A 0 N/A 0.00101 0.07346[ ][ ][ ][ ][ ]≤⋅+⋅+⋅<≤⋅+⋅⋅+⋅<≤⋅+⋅⋅+⋅<≤⋅+⋅⋅+⋅+⋅⋅<≤⋅+⋅⋅+⋅+⋅ttttEXPtEXP300HI0.00101WH0.00016MES0.0001-EXP300150HW0.00129MFR0.09602-MPT0.00659MOP0.00728-EXP150100HW0.00193MFR0.13776-MPT0.01336MOP0.01491-EXP10050HW0.00195MFR0.1408-MPT0.0135MOT0.0096MOP0.0206-MCT0.00848-500HW0.00364MFR0.35908-MPT0.02792MOT0.01101MOP0.04917-=)(th12System ValidationIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchParametersSurvival Intervals (days)0 < t < 50 50 < t < 100 100 < t < 150 150 < t < 300 300 < tValue P-Value Value P-Value Value P-Value Value P-Value Value P-ValueEstimation Observations 338 185 133 198 24Multiple R 0.80 0.854 0.857 0.845 0.9065R Square 0.64 0.730 0.735 0.715 0.82173Adj. R Square 0.633 0.716926 0.721 0.705 0.7571Significance F 1.054E-72 3.71E-48 1.46E-35 1.3E-51 7.37E-08Prediction Observations 338 185 133 198 27RMSE 0.2967 0.2696 0.223 0.1691 0.0499Rcorr 0.3896 0.3745 0.4375 0.4965 0.7845Observed t-Test 7.707 5.5236 5.6319 8.0085 6.3251Critical t-Test 1.64912 1.65304 1.6563 1.65221 1.70562Dozers Hazard Functions (Estimation Data Group = 878 telematics data entries)Increased Function Fit Quality and Prediction Accuracy!The failure of assets with longer survival time can be reliably anticipated compared to younger assets with shorter survival time.13System ValidationBackhoes Hazard Functions (3315 telematics data entries)Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchParametersSurvival Intervals (days)0 < t < 50 50 < t < 100 100 < t < 150 150 < t < 300 300 < tValue P-Value Value P-Value Value P-Value Value P-Value Value P-ValueConstant (C) 1 N/A 1 N/A 1 N/A 1 N/A 1 N/AX1 (MCT) 0 N/A 0 N/A 0 N/A 0 N/A 0 N/AX2 (MOP) -0.03958 0 -0.0165 0 -0.01449 0.00001 -0.01442 0 -0.01318 0X3 (MOT) 0.007714 0 0 N/A 0 N/A 0 N/A -0.00093 0.023X4 (MES) 0 N/A 0 N/A 0.000467 0.001295 0 N/A 0 N/AX5 (MPT) 0 N/A 0.01835 0.00196 -0.00913 0.03857 0.005372 0.0595 0 N/AX6 (MFR) 0 N/A -0.18099 0.00614 0 N/A 0 N/A 0 N/AX7 (HW) 0.001505 0 0 N/A 0.00156 0 0 N/A 0.007903 0X8 (HI) 0 N/A 0.000929 0 -0.0013 0.01273 0.001227 0 -0.00515 0Estimation Observations 664 305 236 241 62Multiple R 0.76878 0.81113 0.76793 0.7891 0.95187R Square 0.591 0.65793 0.58972 0.6226 0.9060Adj. R Square 0.5883 0.6512 0.57828 0.61527 0.884Significance F 8.52E-128 9.4E-69 1.03E-42 5E-50 9.16E-29Prediction Observations 665 305 236 241 61RMSE 0.3746 0.23463 0.22687 0.26875 0.1Rcorr 0.1197 0.3971 0.32333 0.49067 0.87816Observed t-Test 3.1033 7.5326 5.22682 8.7055 10.7Critical t-Test 1.647 1.65 1.651 1.651 1.671Increased Function Fit Quality and Prediction Accuracy!14System ValidationTrucks Hazard Functions (3880 telematics data entries)Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future ResearchParametersSurvival Intervals (days)0 < t < 50 50 < t < 100 100 < t < 150 150 < t < 300300 < tValue P-Value Value P-Value Value P-Value Value P-ValueConstant (C) 1 N/A 1 N/A 1 N/A 1 N/ANo Failure recorded for this periodX1 (MCT) 0 N/A -0.00326 0.0108 -0.00292 0 0 N/AX2 (MOP) -0.05807 0 0 N/A 0 N/A 0 N/AX3 (MOT) 0.00394 0.0139 0.00287 0.00353 0 N/A 0 N/AX4 (MES) 0 N/A -0.00024 0.00669 0 N/A -0.00004 0X5 (MPT) 0 N/A 0 N/A 0 N/A 0 N/AX6 (MFR) 0.0724 0.00006 0 N/A 0.023205 0.000078 0 N/AX7 (HW) 0.000463 0 0.000394 0 0.000066 0.065468 0 N/AX8 (HI) 0 N/A -0.0004 0.0012 0 N/A 0 N/AEstimation Observations 1064 411 232 37Multiple R 0.8084 0.891 0.824 0.8245R Square 0.6535 0.794 0.6788 0.68Adj. R Square 0.652 0.789 0.672 0.652Significance F 4.74E-242 1.366E-136 4.25498E-56 2.51E-10Prediction Observations 1063 410 232 36RMSE 0.3468 0.1744 0.1307 0.066Rcorr 0.225 0.3762 0.2435 0.286Observed t-Test 7.5267 8.2 3.8075 1.74Critical t-Test 1.6463 1.6486 1.6515 1.69Increased Function Fit Quality and Prediction Accuracy!15System ValidationIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future Research012345CoolantTempOilPressureOil Temp EngineSpeedPercentTorqueFuel Rate WorkHoursIdleHoursNumber of Survival Intervals Dependent on the ParameterHealth ParametersDozer Backhoes Trucks2.7E-036.7E-076.8E-031.5E-031.8E-022.1E-031.4E-022.6E-020.0E+005.0E-031.0E-021.5E-022.0E-022.5E-023.0E-02CoolantTempOilPressureOil Temp EngineSpeedPercentTorqueFuel Rate WorkHoursIdleHoursAverage P-ValueHealth ParameterThe Maximum Oil Pressure (MOP) was found to be the most controlling and significant telematics health parameter for dozers and backhoes.16System ValidationIntroduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation Future Research0.50.60.70.80.911 2 3 4 5Survival Function R-SquareSurvival IntervalDozers Backhoes Trucks00.050.10.150.20.250.30.350.41 2 3 4 5Prediction RMSE Survival IntervalDozers Backhoes TrucksThe survival function fitness to the data is the highest for the last survival interval.The survival function prediction error decreasesfor later survival intervals.17Conclusion and Future ResearchFuture Research Telematics provide a rich data source for equipment prognostics. New telematics-based survival analysis methodology. Proposed research is applicable for newer equipment tiers with CAN-Bus data.  Future Research:o Develop alternative telematics data-driven prognostics systems (logistic regression, fuzzy clustering).o Implementing the proposed methodology as an automated system.o Support other fleet management functions with telematics-based data analytics.Introduction Need and ObjectiveTelematics OverviewPrognostics SystemSystem Validation 18Thank you!Your Questions and Feedback are welcomed!Telematics Data-driven Prognostics System for Construction Heavy Equipment Health Monitoring And Assessment Hisham Said, Ph.D. Santa Clara University           5th International/11th Construction Specialty ConferenceJune 10, 2015Tony NicolettiDPL America           

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:
http://iiif.library.ubc.ca/presentation/dsp.52660.1-0076365/manifest

Comment

Related Items