Design Of A Fault Diagnosis System For Fuel Cell Engine Based On Neural Network

As an important part of the fuel cell vehicle, performance and reliability of fuel cell engine have close relationship with safety and cost of fuel cell vehicle. The engine would easily break down during the operation time because it works in a closed space and the work environment is more severe. It is great significant to study fuel cell engine operating characteristics and establish the fault diagnosis model for improving the reliability of fuel cell engine. This thesis has completed the fault diagnosis system of fuel cell engine based on neural network, the main contents are as the following:The architecture of fuel cell engine and the main factors which impact the output performance of fuel cell engine were analyzed. Based on working principle of fuel cell engine and real data, performance-based fuel cell engine failure and subsystem-based type failure were analyzed. Engine faults were graded to four levels according to severity, a list for typical faults at all levels and the corresponding measures to deal with was made.As the fuel cell engine with a strong non-linearity, unstablity and uncertainty, etc. It can not get the precise diagnosis results by using the traditional method of analytical model. This thesis presented a fault diagnosis method based on neural network, discussed the pros and cons of various improved BP algorithm, and BP networks fault diagnosis model were established for different subsystem, access to initial results used in fault diagnosis process of the engine.According to the fuel cell engine system which was developed independently, fuel cell engine fault diagnosis expert system software was designed. Software was divided to several modules based on requirement analysis, including the communications interface module, real-time monitoring module, fault diagnosis module, neural network self-learning modules and database management module. The system can monitor the completion of the engine operating state, inquiry real-time or historical working parameters, display alarm, diagnose and mapping the common fault, output the proposed decision-making advices, while the system can diagnosis off-line. The accuracy of the fault diagnosis model and software diagnostic capabilities was verified by on-site commissioning. The correct rate of diagnostic results were rarely high, obviously the software can diagnosis the failure of the engine maintenance correctly and give professional opinions on the repair. Using modular design, the system can be highly reused. If demands change, it can reduce the difficulty in changing procedures.