Remaining Useful Life Prediction via Bayesian Temporal Convolutional Networks

Abstract

In modern industry, the accurate prediction of remaining useful life(RUL) contributes to the equipment safety and economic effectiveness. Aiming at the RUL prediction, this paper proposed a Bayesian temporal convolutional network (BayesianTCN) under the Bayesian deep learning framework. BayesianTCN outputs not only the RUL prediction, but also the associated confidence interval by Monte-Carlo simulation. This quantifies the RUL prediction uncertainty. Experimental results on CMAPSS datasets show that our model has higher fitting degree and lower uncertainty than BayesianLSTM, and performs well whether in simple or complex conditions.