Fault Diagnosis and Fault-Tolerant Control for Multi-Sensor of Fuel Cell System Using Two-level Neural Networks

Sensors are critical for the monitoring and real-time control of fuel cell system, according to the reliability requirements of multi-sensor of 60kW automotive fuel cell system designed by our group, a two-level neural networks based fault diagnosis method is put forward in this paper. The two-level neural networks include a main net and five sub nets which are corresponding to inlet hydrogen pressure sensor, inlet air pressure sensor, outlet temperature sensor, output voltage sensor and output current sensor, and they are trained with 2100 different groups of normal data of the five sensors stated above from time t-1 to t-3 with LM algorithm. In the faults detection of these sensors, we set the threshold errors of the main net and the five sub nets at 0.005 and 0.05 respectively, when the errors of both main net and a certain sub net exceed the threshold values set, the fault of sensor corresponding to the certain sub net is detected, then its output signal is replaced by the sampled value of the former time and updated into the two-level neural networks accordingly. Finally, taking output current sensor for instance, the simulation results are presented, which validates that the approach adopted can facilitate the real-time fault detection and active fault-tolerant control for multi-sensor of automotive fuel cell system.