Remaining useful life estimation of rolling element bearings based on an interpretable feature extraction method

Dealing with the remaining useful life (RUL) estimation issue of rolling element bearings (REBs), feature extraction is a fundamental step and is of great importance in the estimation of the RUL of REBs. However, few studies have considered the interpretability of the feature extraction. A new interpretable feature extraction method and an exponential model-based RUL estimation method are proposed in this study. The feature selection strategy and partial least squares (PLS) method were used to ensure the interpretability of the extracted features. First, the extracted time-domain features were screened by calculating the monotonicity, robustness, and predictability of the time-domain feature vectors. Meanwhile, time-domain feature vectors with calculation results within [0,1] are preserved, weighted, and reused to achieve good predictive performance. Subsequently, the PLS method was used for feature extraction, and a new feature vector describing the bearing degradation process was constructed. Finally, an exponential model was proposed for RUL estimation. A strong tracking filter is employed to fuse the residuals of the model parameters and improve the accuracy of the estimation. Under the three experimental working conditions, the root mean square error (RMSE) values of this study’s method were reduced by 15.87%–82.40%, and the scoring function values were reduced by 1.13%–34.70% compared with other methods. The obtained RMSE and scoring function values prove the rationality and superiority of the proposed method and indicate that the method is effective in reducing the safety risks associated with bearing failures.