Remaining Useful Life Prediction of Mechanical Equipment Based on Temporal Convolutional Network and Asymmetric Loss Function

Abstract

Remaining Useful life (RUL) prediction plays a very significant role in the health management of machinery and equipment. Accurate life prediction can maximize the working capacity of the equipment and reduce costs. This paper proposes a method of mechanical equipment lifetime prediction based on a Temporal Convolutional Network (TCN) and an asymmetric loss function. Time-series convolutional networks are accurate, simple, and clear in mining time-series features for sequence modeling. The asymmetric loss function enables the remaining life prediction to be more tend to over-predict, avoiding enormous economic loss due to late prediction. The effectiveness of the proposed method is tested on the public dataset C-MAPSS. Comparison with other deep learning methods such as gated recurrent unit network (GRU), Bi-directional Long and Short Term Memory network (BiLSTM), and two-dimensional convolutional neural network (2D-CNN) shows the superiority of TCN. Finally, the loss function is adjusted to improve the overall prediction accuracy.