A new dynamic neural modelling for mechatronic system prognostics

This paper proposes a new dynamic neural modelling method for mechatronic system prognostics. A simulated turbofan engine system is firstly introduced. Then, so-called health index (HI) for the system is evaluated by an artificial neural network model called extreme learning machine (ELM) to recognize the health conditions. In order to track and predict the propagation process of the system based on the health index, a new dynamic neural network (DNN) with feedback from outputs to both output layer and hidden layer is developed. With the batch learning type of least square, both the feedback weights and output weights are optimized. With the turbofan engine datasets from NASA, simulations are carried out, using both the ELM model and the proposed DNN. The excellent performance of the proposed DNN shows that the DNN with hidden layer feedback has a superior capability of capturing the dynamic features of the mechatronic system.