Study On Health Evaluation And Fault Prognosis Methods For Vehicle Systems

Vehicle health management is the management activity that directly relates to the health state of avehicle system. It can monitor the real-time state of a system and its components, adjust the systemback to normal state when unexpected abnormalities occur, and minimize the damage to system safetyand the undertaking task. Therefore, fault diagnosis, prognosis and restoration are the key tasks in ahealth management system.This thesis focuses on health evaluation and fault prognosis; two hot issues in the field of vehiclehealth management.Two novel methods are developed and tested on a vehicle simulation platform.The key contributions are summarized as bellows.1. A fault propagation model is developed by integrating a qualitative modeling method intosigned directed graph (SDG) and Bayesian network (BN).2. A system health evaulation scheme is proposed by combining the lifetime prediction methodfor electronic components and the performance evaluation method for large-scale complex systems.The proposed health evaluation model takes into account four major factors: component’s importance,reliability, historical break-down frequency and current fault degree. The model takes form of linearor nonlinear weighted model, where Analytic Hierarchy Process method (AHP) is adopted todetermine the model’s parameters.3. A fault prognosis method is proposed via Bayesian network. A simple multi-layer Bayesiannetwork is developed which can handle time-dependent information by incorporating the qualitativesignal trend information into the nodes of the Bayesian network. The failure probabilities of childnodes are calculated based on the integrated health index of parent nodes, which can improve theprediction performance of the presented fault prognosis method. Different parameter learningalgorithms are adopted for complete data and incomplete data, respectively. The Pearl’s poly treepropagation algorithm is used for joint probability reasoning.4. The application results on a Quanser3-DOF Hover simulation system can verify theeffectiveness and feasibility of the proposed methods.