Distributional Difference-Based Dynamic Adaptation for Bearing Performance Degradation Assessment

Bearings are critical components in rotating machinery, where real-time monitoring of its health is essential to improve machine safety and productivity. However, the degradation trend of vibration signals collected during equipment operation is often unconspicuous, poor monotonicity, and large wave nature, making it difficult to accurately assess bearing performance. Considering the above issues, this study proposes a distribution difference-based dynamic adaptation method (DD-DAM), which uses the constructed distribution structure to evaluate the bearing state information. It includes three simple steps: first, the sample set was established based on a small number of healthy samples, from which a certain percentage of samples were randomly selected and their mean distribution was calculated as the baseline distribution. Then, Kullback-Leibler (KL) divergence was used to obtain the maximum distribution difference distance D max in the healthy sample set. Finally, the evaluation results are given by comparing the difference between D max and the monitoring status. The validity of the method was verified on the bearing experimental data set. In addition, compared with Gaussian mixture model (GMM) and principal component analysis (PCA) through quantitative assessment, DD-DAM showed significant advantages in evaluating the bearing health status.