978-1-4577-0557-1/12/$26.00 ©2012 IEEE 1 Data-driven Fault Diagnosis in a Hybrid Electric Vehicle Regenerative Braking System Chaitanya Sankavaram 1 , Bharath Pattipati 1 , Krishna Pattipati 1 , Yilu Zhang 2 , Mark Howell 2 , and Mutasim Salman 2 1 Department of Electrical and Computer Engineering, University of Connecticut, 371 Fairfield Road, U-2157, Storrs, CT 06269, USA 2 GM R&D, General Motors Company, 30500 Mound Rd, Warren, MI 48090, USA {chaitanya, krishna}@engr.uconn.edu Abstract— Regenerative braking is one of the most promising and environmentally friendly technologies used in electric and hybrid electric vehicles to improve energy efficiency and vehicle stability. In this paper, we discuss a systematic data-driven process for detecting and diagnosing faults in the regenerative braking system of hybrid electric vehicles. The process involves data reduction techniques, exemplified by multi-way partial least squares, multi-way principal component analysis, for implementation in memory- constrained electronic control units and well-known fault classification techniques based on reduced data, such as support vector machines, k-nearest neighbor, partial least squares, principal component analysis and probabilistic neural network, to isolate faults in the braking system. The results demonstrate that highly accurate fault diagnosis is possible with the pattern recognition-based techniques. The process can be employed for fault analysis in a wide variety of systems, ranging from automobiles to buildings to aerospace systems. TABLE OF CONTENTS 1. INTRODUCTION.................................................................1 2. MODELING OF REGENERATIVE BRAKING SYSTEM ........2 3. FAULT UNIVERSE AND MONITORED SIGNALS ................3 4. FAULT DETECTION AND DIAGNOSIS PROCESS................4 5. EXPERIMENTAL RESULTS ................................................8 6. CONCLUSIONS ..................................................................9 ACKNOWLEDGMENTS ..........................................................9 REFERENCES ........................................................................9 BIOGRAPHIES .....................................................................10 1. INTRODUCTION Hybrid electric vehicles (HEVs) employ regenerative braking to improve fuel economy, enable energy regeneration and provide environmental protection [1]. The primary function of a regenerative braking system (RBS) is to convert kinetic energy into electrical energy and store it in batteries during braking mode for later use in propelling the vehicle (see Fig 1) [2]. Failures in a regenerative braking system may significantly degrade the performance and efficiency of vehicles. Hence, an intelligent diagnostic process is crucial to quickly detect and isolate faults in order to aid in vehicle health monitoring and, consequently, enhance the reliability of vehicular systems. Diagnostic methods have mainly evolved upon three major paradigms, viz., physics-based modeling, data-driven and knowledge-based approaches. The physics-based modeling approach employs consistency checks between the sensed measurements and the outputs of a mathematical model. The expectation is that inconsistencies are large in the presence of malfunctions and small in the presence of normal disturbances, noise and modeling errors. Two main methods of generating the consistency checks are based on observers (e.g., Kalman filters, reduced-order unknown input observers, interacting multiple models, particle filters) and parity relations (dynamic consistency checks among measured variables stemming from hardware or information redundancy relations). A data-driven approach is preferred when the system monitoring data for nominal and degraded conditions is available. Neural network and statistical classification methods are illustrative of data-driven techniques. The knowledge-based approach uses graphical models such as dependency graphs (digraphs), Petri nets, multi-signal (multi-functional) flow graphs, and Bayesian networks for diagnostic knowledge representation and inference [3][4][5]. In this paper, we develop a systematic data-driven fault detection and diagnosis (FDD) process for diagnosing faults in a regenerative braking system. For FDD analysis, Powertrain System Analysis Toolkit (PSAT) [6], a vehicle simulation software tool, was used to create a Matlab/Simulink ® model of RBS with series-parallel drivetrain configuration. The fault diagnosis process involves data reduction techniques, such as multi-way partial least squares, multi-way principal component analysis [7], Fig. 1: Regenerative Braking System - Energy Flow Diagram in HEVs