Rolling Bearing Fault Diagnosis Under Data Imbalance and Variable Speed Based on Adaptive Clustering Weighted Oversampling

Rolling bearings are critical for maintaining the stability, reliability, and safety of mechanical systems. However, diagnosing faults in rolling bearings objectively can be challenging due to the lack of fault data and the difficulty of feature extraction at variable speeds. To solve the variable speed problem, the segmented variable speed data is processed using nuisance attribute projection (NAP) to remove the condition information in the feature domain. Meanwhile, considering the imbalanced data, the adaptive clustering weighted oversampling (ACWOS) method is proposed to process the imbalanced data. The method, firstly, to solve the problem that density peak clustering (DPC) requires human intervention, proposes a strategy based on the γ-parameter jump phenomenon and soft thresholding to determine the number of clusters and cluster centers adaptively. Then, the proposed ACWOS also assigns different oversampling weights and variable K-nearest neighbors (VKNNs) to different samples based on the sample density and relative distances to increase some minority samples, which solves the problem of imbalanced and uneven distribution of failure data. Finally, the effectiveness and superiority of the method are demonstrated by comparing five weights, three classifiers, and seven imbalanced data processing methods on the Ottawa and measured datasets, respectively.