Remaining Useful Life Prediction on Transfer Learning for Bearing

Remaining useful life (RUL) prediction is vital to formulating a suitable maintenance strategy in machinery health management. Limited by the time-varying operational conditions, conventional RUL prediction models trained on some run-to-failure (RTF) datasets are unlikely to be generalized to a new degraded process. To increase the generalizability, recent studies have focused on the development of the deep domain adaptation methods for RUL prediction, which mainly align the global temporal features across the source and the target domains, leading to inaccurate prognostic results under time-varying operational conditions. In this study, an operational condition attention (OCA) subnetwork is constructed to eliminate the entanglement of the time-varying operational conditions and monitoring data. Adversarial-based domain adaptation (ABDA) and distance-based domain adaptation (DBDA) methods were applied separately to reduce the distribution discrepancy of the temporal features. In this way, two novel domain adaption methods, i.e., OCA-LSTM-ABDA and OCA-LSTM-DBDA, were proposed for RUL prediction with time-varying operational conditions. Comprehensive experiments on aircraft turbofan engines were conducted to validate the proposed methods. Owing to the explicit modeling of the influence mechanism between operational conditions and monitoring data, the proposed methods exhibit an improved performance with higher prediction accuracy than conventional deep domain adaption methods.