Adaptive ensemble gaussian process regression-driven degradation prognosis with applications to bearing degradation

•Constructed basic predictors with polynomial family mean functions.•Proposed an adaptive time-varying ensemble learning method.•Reduces training time and resources, with higher accuracy and adaptivity.•The effectiveness of the proposed method was verified through practical examples. Degradation modeling and remaining useful life prediction of bearings is crucial for predictive maintenance of rotating machinery. However, the contradiction between limited full-life cycle samples and dynamically diverse degradation trends has become the main obstacle for degradation modeling and prediction. To address these challenges, this paper proposes an adaptive time-varying ensemble Gaussian process regression-driven degradation prediction method. Firstly, four different base predictors (i.e., global predictor, healthy stage predictor, impending degradation stage predictor and degradation stage predictor) are constructed based on Gaussian regression process to reflect the characteristics of different degradation stages. On this basis, a time-varying ensemble learning method with adaptive weights is proposed, and a corresponding adaptive ensemble Gaussian regression process is constructed to model the full-life degradation process. The model can effectively enhance the flexibility and prediction accuracy of the single-time invariant Gaussian regression model. Some real bearing degradation cases are used to validate the proposed method.