Prediction Method of Bearing RUL Based on DNN and GBDT Algorithm

Abstract

In view of the lack of effective model and the large prediction error in the traditional prediction methods, a collaborative prediction method for remaining useful life of bearing based on DNN and GBDT is proposed. Firstly, the degradation characteristics are constructed through normalization processing of parameters in time domain and frequency domain that can clearly represent the healthy running state of bearings, in order to improve the correlation of degradation characteristics, the prior model features are generated based on DNN. Secondly, a regression model of GBDT based on the prior model features is presented. Finally, the experimental results show that compared with other algorithms such as DNN, GBDT, SVR, RF, DT, the proposed method has better prediction performance evaluation results, higher prediction accuracy and efficiency compared with other algorithms.