A Combined Temporal Convolutional Network and Gated Recurrent Unit for the Remaining Useful Life Prediction of Rolling Element Bearings

An accurate prediction of bearing remaining useful life (RUL) can largely reduce the operation and maintenance cost of a machine. This then motivates the work presented in this study where a bearing RUL prediction technique based on a combined temporal convolutional network (TCN) and gated recurrent unit (GRU) is developed. The TCN is used to extract the detailed spatial features from the raw data whilst the GRU is employed to learn long-term dependencies between features simultaneously to enhance the RUL prediction accuracy of rolling element bearings. The validity of the proposed technique is examined using the published data from the IEEE-PHM2012 data challenge. It is found that the proposed technique can yield a satisfying lifecycle degradation trend using the end-to-end bearing degradation data. The efficacy of the proposed technique is also examined in the study by comparing the prediction result with those obtained using existing RUL prediction techniques. It is shown that the proposed technique has a better performance in bearing RUL prediction than other techniques.