Remaining Useful Life Prediction Based on Improved Temporal Convolutional Network for Nuclear Power Plant Valves

In order to ensure safe operation, the risks associated with Nuclear Power Plants (NPP) needs to be precisely assessed. For that very reason, Prognostics and Health Management (PHM) is so crucial as it determines the best time for each equipment maintenance. At present, the bottleneck of predictive maintenance is the prediction of equipment’s remaining useful life (RUL). With the development of artificial intelligent techniques, deep learning algorithms are becoming more and more popular for RUL prediction. Upon this, this paper studied RUL prediction techniques for nuclear electric gate valves with temporal convolutional network (TCN). The main advantage of using TCN is its ability to avoid loss of information due short-term changes. More importantly, we enhanced the effectiveness of traditional TCN by incorporating an additional auto-encoder in its structure along with improving its residual convolution mode. The technique was successfully applied and then verified on simulated crack data obtained from electric gate valve platform (estabilshed by authors). Comparison of results with typical methods shows that the proposed method can predict electric valves’ RUL with higher accuracy. Furthermore, this paper tested the generalization ability and universality of the proposed method through aero-engine standard datasets. Aforesaid in view, it is envisaged that PHM systems for NPPs can be further developed through application of proposed methodology.