Fault Identification and Remaining Useful Life Prediction for Complex Equipment under Multiple Fault Modes

Abstract

To ensure the safe operation of complex equipment and specify maintenance strategies, the prediction of the remaining useful life (RUL) and the identification of failure modes are of great significance. Generally, it is difficult to directly differentiate the fault modes from the raw data. Thus, a novel fault clustering and RUL prediction framework under multiple failure modes is proposed in this paper. First, dynamic time warping (DTW) distance is combined with k-medoids to identify multiple failure modes. Following that, a convolutional neural network connecting with a long short-term memory (CNN-LSTM) is constructed to extract hidden features and predict RUL. In the end, the experiment is conducted on an engine dataset provided by NASA. The results show that the proposed framework has a better performance compared with the competitive prediction methods.