Fault Diagnosis With a Model-Based Recurrent Neural Network

The utility of a model-based recurrent neural network (MBRNN) is demonstrated in fault diagnosis. The MBRNN can be formatted according to a state-space model. Therefore, it can use model-based fault detection and isolation (FDI) solutions as a starting point, and improve them via training by adapting them to plant nonlinearities. In this paper, the application of MBRNN to the IFAC Benchmark Problem is explored and its performance is compared with ‘black box’ neural network solutions. For this problem, the MBRNN is formulated according to the Eigen-Structure Assignment (ESA) residual generator developed by Jorgensen et al. [1]. The results indicate that the MBRNN provides better results than ‘black box’ neural networks, and that with training it can perform better than the ESA residual generator.