Fault Diagnosis of PEMFC Systems In The Model Space Using Reservoir Computing

Artificial neuron network provides a promising solution for fault diagnosis of fuel cell systems. A recently proposed novel framework of recurrent neuron network named Reservoir Computing is focused with only its output weights to be trained, which is rather advantageous for online adaption in real-time applications. In a previous work, its simplicity and efficiency has been demonstrated. This paper focus on a novel attempt of performing fault diagnosis directly in the reservoir computing based model space (current-voltage model) instead of the original data space (voltage signal). No additional feature extraction procedure is needed and abnormal health states could be detected directly in the model space (in the form of evolution of output weights).