Remaining Useful Life Prediction Based on Multi-scale Residual Convolutional Network for Aero-engine

Jinhui Yu, Yuhuai Peng, Qingxu Deng

Abstract: Accurate and reliable remaining useful life (RUL) prediction is crucial to the use and maintenance of mechanical manufactures, which can improve usage efficiency and boost economic benefits. The traditional RUL prediction methods are lack of consideration on various features and difference of aero-engine data, which causes low accuracy and stability. In response to these problems, a multi-scale residual temporal convolutional network (MSR-TCN) is proposed. With a multi-scale convolution structure, the complex features contained inside the data can be extract comprehensively, which contributes to the distribution fitting. Through the attention mechanism fused in the structure, the sensor data can be refactored based on relevance with RUL, the influence of low relevance data can be avoided in the prediction process. To optimize the algorithm's feature extraction process and alleviate the problem of overfitting, the residual module is added in the network. Finally, experiments are implemented on the turbofan engine monitoring data provided by NASA, the results are compared with state-of-art methods, which proves the accuracy and effectiveness of the proposed method.