A Hybrid Approach for Rolling Element Bearing Health Condition Identification

The health condition of rolling element bearings profoundly affects the longevity, operational performance, and efficiency of the entire machinery system. This makes accurate division of health stages and condition identification essential for machine monitoring. To meet this need, this paper introduces a hybrid method combining three key steps. Initially, time‐domain analysis based on Pearson correlation coefficients (PTDA) selects initial features that are most sensitive to progressive degradation. Subsequently, a combined technique using fuzzy C‐means clustering with validity function, simulated annealing and genetic algorithm (called VSAGAFCM) achieves optimal segregation of the data into distinct health states, thereby eliminating the subjectivity associated with manual stage division. Finally, a probabilistic neural network (PNN) leverages these results to accurately identify each health state, capitalizing on its rapid training speed and probabilistic output capabilities for effective pattern recognition. The proposed method's effectiveness is assessed on a publicly accessible dataset.