Online fault diagnosis of fuel cell systems using independent MLP neural network model

In this paper, an independent neural networks is constructed for modelling and to perform fault diagnosis of a proton exchange membrane fuel cell systems which has a nonlinear behaviour. The fault detection is investigated based on the residual generation. The difference between the model and the process plant gives the modelling prediction errors which later been used in detecting faults occurring in the systems. The RBF network acts as a classifier to perform fault isolation. The faults are introduced in a simulator model of fuel cell systems developed by University of Michigan where five faults are introduced in online simulation. The simulation results show that both neural network models able to detect and isolate five faults accordingly under open-loop scheme and the results are almost similar.