Enhancing bearing life prediction: Sparse Gaussian process regression approach based on sequential ensemble and residual reduction for degradation prediction

Abstract

Bearings are critical components of wind turbines, and predicting their remaining useful life is essential for ensuring safe and reliable operation of wind power generators. However, the degradation of wind turbine bearings exhibits distinct multi-stage characteristics, and full-lifecycle degradation samples are rarely available. This lack of samples makes it challenging to accurately predict the service life of bearings. Therefore, we propose a residual-reduction sequential ensemble sparse Gaussian process regression model to enhance bearing life prediction. The proposed model introduces a primary learner based on a Gaussian process regression serial ensemble strategy, effectively simulating the multi-stage dynamic bearing degradation process. Building on this learner, the model constructs a secondary learner within the gradient-boosting framework by applying residual-reduction techniques to further increase predictive accuracy. The proposed method is applied to a public dataset and a real wind turbine bearing degradation dataset, and its superiority is validated through comparisons with existing methods.