Model-based Fault Diagnosis Of Electrified Driven Powertrains In Pure Electric Vehicles

In the severe energy and environmental crisis, the rapid development and popularization of green and efficient electric vehicles is considered as one of the important actions to reduce energy consumption and emissions. Permanent magnet synchronous motor(PMSM) and lithium-ion batteries are the key components of powertrains in pure electric vehicles, and their functionality and reliability are correlated to the vehicle’s dynamic performance, safety and stability. The components of PMSM and lithium-ion batteries will present various kinds of faults in the actual driving environment, resulting in the decrease of system performance and efficiency, even the safety risk. In order to ensure the vehicle working in a safe, efficient and reliable environment, it is critical to develop a fault diagnosis scheme for powertrains. Model based fault diagnosis for PMSM and lithium-ion batteries has attracted more and more attentions from academia and industry, which is of great values for the theoretical research and engineering application.This dissertation presents a generalized unscented Kalman filter(UKF) based fault detection and isolation(FDI) scheme for PMSM to detect and isolate current sensors and position sensor fault. The proposed methodology could properly handle the nonlinear properties with good robustness and high diagnosis accuracy. In addition, structural analysis based FDI scheme is presented for a lithium-ion battery to detect and isolate the current, voltage and temperature sensor fault. One advantage of this methodology is that it could pre-analyze the sensors fault detectability and isolability without the accurate knowledge of battery parameters, but just depending on the structural information of battery dynamics. This is useful in the early design stages of diagnostic system. A further advantage is that it helps to construct different diagnostic tests to achieve the online FDI. Finally, adaptive extended Kalman filter(AEKF) based sensor FDI scheme is proposed for a series lithium-ion battery pack with a low computational effort, which solves a challenging problem of sensor FDI from a large number of sensors in the battery pack. Also the proposed method is robust to system random noise. The detail work are as follows,Firstly, the working principle and components of PMSM, voltage based inverter and lithium-ion battery system are described. Through using the failure modes and effects analysis(FMEA) method, different types of failures, failure effects and causes are analyzed. Taking the lithium-ion battery system as an example, risk priority number(RPN) is used to evaluate the risk level of different system failures. By using the fault tree analysis(FTA) technique, the propagations from components faults to system failures are represented by a tree-structured graph.Secondly, model-based fault diagnosis methods applied in this work are discussed. Considering the system nonlinear properties, measurement noise and unknown disturbance, EKF, AEKF and UKF –based fault diagnosis method are used. In order to know the system fault detectability and isolability in the early design stage, structural analysis based fault diagnosis method is applied. To decrease the probability error when declaring the presence of faults, a statistical inference residual evaluation method named cumulative sum(CUSUM) test is used.Thirdly, a generalized UKF based FDI scheme is presented for PMSM to detect and isolate the current sensors and position sensor fault. The PMSM drive model is built, including PMSM model, field oriented control(FOC) technique used to control the PMSM, inverter model, space vector pulse width modulation(SVPWM) technique used to modulate the inverter to output the required voltage, driver model and other subsystem models. The propagation effects of PMSM sensors fault, demagnetization of permanent magnet, and switch-on failure of two transistors on the same bridge of inverter on the whole vehicle are analyzed. In final, three UKFs are used, and the inputs of each UKF include different sensor measurements, hence faults in any of these sensors can be detected by using this UKF. All faults can be efficiently isolated by using these UKFs as different faults affect each UKF differently. The proposed method could properly handle the nonlinear properties with good robustness and high diagnosis accuracy.Finally, observer based and structural analysis based sensor FDI scheme is presented for a lithium-ion battery and a battery pack. The battery testing system is introduced, a comprehensive battery testing program is proposed, and a database with three types of lithium-ion batteries is established. The effects of current or voltage sensor faults on the battery management system are analyzed. EKF based fault detection scheme is proposed to detect the current and voltage sensor fault, and the effectiveness of proposed approach is validated through experimental data. In order to know the fault detectability and isolability in the early design stage, structural analysis based fault FDI scheme is presented for a lithium-ion battery. Also, it could help to construct different diagnostic tests to achieve the online current, voltage and temperature sensor FDI, and the effectiveness of proposed scheme is validated through lots of experimental data. Finally, AEKF based sensor FDI scheme is proposed for a series battery pack with a low computation effort, which solves a challenging problem of sensor FDI from a large number of sensors in the battery pack. The main idea is that based on the differences of voltage responses under various sensors fault, the batteries in the battery pack are separated into two parts. In one part, the batteries are monitored online, and in the other part, the batteries are monitored offline using a long-time interval. With the help of AEKF, this method could efficiently achieve the sensor online FDI.