Abstract:As one of the important functions to realize industrial intelligence in Industrial Internet of Things (IIoT), remaining useful lifetime (RUL) prediction can predict the future degradation states of industrial equipment based on monitoring data and then reckon its remaining service time. Furthermore, the corresponding predictive maintenance strategies can be formulated in advance, and the reliability, availability, and safety of the equipment can be enhanced. In this paper, an attention-based remaining useful lifetime prediction method (ARULP) for IIoT was proposed. Firstly, in the model training stage, a local attention computation algorithm was designed by constructing a data-driven local attention computation model and utilizing the training data to compute the local attention measurements. Thus, the prediction model can pay more attention to the key information within a large amount of data. Then, a relevance vector machine based on local attention was designed by introducing the local attention mechanism into the learning process of implicit variables in the prediction model. The attention weights were updated to adjust the state prediction model adaptively and improve the prediction accuracy of the RUL for IIoT device. Finally, the ARULP method was applied in the model prediction stage to predict the degradation status and reckon the RUL for IIoT device. Extensive experiments based on the dataset of life test for accelerated rolling element bearings released by Xi’an Jiaotong University were carried out to predict the RUL for bearings with outer ring failure, inner ring failure, and holder failure under different working conditions. The experiment results showed that the prediction data of the ARULP method are the closest to the actual bearing degradation data and more accurate RUL prediction results can be obtained compared with RVM, AR, ARIMA, and LSTM benchmarks.