STDM: A new two-stage degradation model for Remaining useful life prediction

The two-stage degradation model is frequently used to describe the degradation path and is typically formulated with two mathematical functions as the drift terms. However, at the stage division point (SDP), an abrupt change in drift term rate affects the accuracy of model in describing the degradation path. Additionally, existing change-point-based SDP identification method fail to account for the impact of local random fluctuations, which reduces the model performance and prediction accuracy. Therefore, this paper proposes the steady two-stage degradation model (STDM), which addresses the abrupt change by retaining partial parameter information from the slow degradation stage, thereby achieving a smooth transition between models. Besides, the anomaly SDP identification (ASI) method is proposed. It constructs a health indicator (HI) called the anomaly of model-HI (AMH). By analyzing the increments and the increment threshold of AMH, SDP identification is achieved to exclude the influence of local random fluctuations. Furthermore, a metric based on AMH, referred to as AMH for metrics (AMHM), is developed to evaluate model performance. Finally, the proposed model and metric are validated using the XJTU-SY dataset and laboratory dataset, showing an improvement in cumulative relative accuracy by 4.4% and 8.3%.