Study On Some Key Problems Of Fault Dignosis And Maintenance For Automotive Fuel Cell System

As the dominating part of Fuel Cell Electric Vehicle(FCEV), atuomotive fuel cell system can hardly avoid malfunction, faults or even serious accidents for its complicated structure, atrocious operational circumstance and poor dynamic permformance, so it is important to study on its fault diagonosis to enhance the safety, relability, maintainability and market acceptance of both itself and FCEV. Based on the 60kW automotive fuel cell system designed by our group, in this paper, some key problems related to the safety and maintainability of the whole system, and the relability of its accessorial subsystems and fuel cell stack are studied. The main research works are as follows:The common problems of automotive fuel cell system's fault characteristic and fault mechanism are put forward systemically, the fault characteristic is qualitatively analyzed from the points of coupling, gradual change, randomicity and uncertainty, and the fault mechanism is qualitatively analyzed from the aspects of mechanism factor, electric fator, device factors, human factors and environmental factors, based on the automotive fuel cell system designed, its faults are classified and ranked.According to the hydrogen safety of the whole system, the fault tree analysis model of hydrogen leaking is established and qualitatively analyzed, its weakness is figured out, to evaluate the hydrogen safety, considering the uncertain fault probability of each elementary event, the fuzzy maths theory is combined into the traditional fault tree analysis method to quantitatively analyze the probability of top event and the importance degree of each elementary event, then some improvement measures and advice on hydrogen leaking are brought forward accordingly.As to the fault dianosis and signals fault tolerant of multisensors in the accessorial systems, the united fault diagnosis and fault tolerant model based on two layer main and sub neural networks is set up, the difference between the predictive output of the main neural network and subnetworks with the sensors's practical output is constructed, these sensors are diagnosed to be in fault when the difference of the main network exceeds the threshold setted and lasts for the threshold steps time, and the output of certain sub network is used to replace the practical output of its corresponding sensor. Finally, the simulation is presented, it proves that the model set up can be effectively used to the fault diagosis and signals restoration of those sensors which have redundancey relationship, and it can meet the needs of the both temperature and air flow fault tolerant control of automotive fuel cell system.Considering flooding and drying of fuel cell stack are difficult to be inerrably distinguished, an information fusion fault dignosis method combining binary tree support vector machine(SVM) and D-S evident theory is put forward, the fault characteristic is extracted from the real time process variables and their variational parameters feature which affect and reflect the water transmission in fuel cell stack, the output of binary tree SVM is utilized as the initial healthy state diagosis results, then they are fused with D-S evident theory to get the final fault dignosis conclusion. The simulation results reveal that, compared with neural networks it has preponderant precision and speed with binary tree SVM, and the belief degree is high and the uncertainty of the conclusion is largely lowered down when the conflicting results of SVMs are fused with D-S evident theory.Finally, according to the requirement of off-line and online fault diagosis and maintenance, an embedded handheld fault diagosis instrument is designed, the diagnostic trouble code(DTC) is designed and the fualt diagnosis strategy and flow are provided. Besides, the visual state monitoring and fault diagnosis expert ssystem is designed with VC++ languange, the specific interfaces are realized, and the fault diagnosis example and its results prove their practicability.Above all, hydrogen leaking, multi sensor fault diagnosis and tolerant, fuel cell stack healthy state diagnosis, the design of monitoring and fault diagnosis system are studied in this paper, the calculation and simulation results show the methods and ideas adopted are effective and practical, which can provide some reference to some other problems on its fault dignosis and maintenance in future.