A modified echo state network based remaining useful life estimation approach

Abstract

An approach to estimate the remaining useful life (RUL) by Echo State Network (ESN) is presented, which is a new paradigm in recurrent neural network (RNN). ESN randomly establishes a large sparse reservoir to replace the hidden layer of RNN, which overcomes the shortcomings of complicated computing, difficulties in determining the network topology of traditional RNN. An ESN sub-models strategy composed by classified ESN models matching to the varied training data set by retraining and classification is explored to estimate the RUL of turbofan engine system. The experimental results with the turbofan engine data of NASA Ames Prognostics Data Repository show that the proposed method can achieve better RUL estimation precision compared with the approaches of classical ESN and ESN trained by Kalman Filter and potential prospective in application.