Fault diagnosis of rolling bearing based on improved stacking ensemble learning

Abstract In response to the challenges posed by the complex operating conditions of electromechanical systems, non-stationary noise interference, difficulties in sensor installation on bearings, and the scarcity of real fault sample data, this paper proposes an advanced fault diagnosis method for rolling bearings. This method employs an enhanced stacking ensemble learning model, improved with wavelet soft and hard threshold algorithms, and utilizes both motor current and vibration signals. Initially, preprocessing of the raw motor current and vibration signals is performed using these wavelet algorithms to mitigate noise and highlight subtle fault features. Data fusion of the motor current and vibration signals is then conducted, from which time-domain and frequency-domain features are extracted to serve as feature vectors for the diagnostic model. Feature selection is optimized through filtering and embedded methods, and principal component analysis is applied to reduce dimensionality. The base learners—GBDT, RandomForest, AdaBoost, and SVM—are employed for polynomial feature processing and weight adjustment. A logistic regression (LR) meta-learner is then trained on newly constructed meta-training and meta-test sets to form the improved stacking ensemble learning model. Comparative evaluations with single models and stacking ensemble models demonstrate that the proposed method significantly enhances fault diagnosis accuracy, achieving an impressive accuracy rate of 99.36%.