Unbalanced Data Classification Model in PHM Field Based on Oversampling Algorithm

Fault detection based on data-driven artificial intelligence has always been a research hotspot. Due to the long- term operation of rotating machinery in a healthy state, the lack of historical data on faults leads to data unbalance problems, which hinder data-driven fault diagnosis and have become one of the stubborn problems in the field of PHM. From the perspective of data preprocessing, this paper explores the effects of SMOTE and LR-SMOTE oversampling algorithms on unbalanced data of rotating machinery. This paper uses the public data of gears and bearings to artificially establish various types of unbalanced data and combines the SMOTE and LR-SMOTE oversampling algorithms with SVM, RF, and GBDT three classifiers into multiple models for experiments. The experimental results show that the algorithm combining LR-SMOTE and SVM can achieve better classification results and has better stability. And compared with the SMOTE algorithm, LR-SMOTE can effectively avoid the overfitting problem of the GBDT classifier.