Unbalanced Data Classification Model in PHM Field Based on Stacking Integration Strategy

Data-driven artificial intelligence fault detection has always been a hot topic of research. The lack of historical data on faults due to rotating machinery and equipment operating in a healthy state for a long time leads to the problem of data unbalance. This unbalance hinders data-driven fault diagnosis and has become a persistent problem in the field of prognostics and health management (PHM). In this paper, we explored the effectiveness of a mechanical fault diagnosis framework for unbalanced data. This framework combines oversampling algorithms and integrated learning strategies to address the challenge of unbalanced data. We conducted extensive experiments to evaluate the proposed framework using actual unbalanced data. The results showed that the fault diagnosis framework proposed in this paper can significantly improve the prediction accuracy, particularly for a few classes of samples. By applying the GR-SMOTE oversampling algorithm, the proposed model improves the Pre_Small index by up to 93.41% and effectively addresses the data unbalance problem.