Rolling Bearing Degradation Evaluation Method Based on mRMR-SVDD

Aiming at the problem of continuous degradation of rolling bearing components and working conditions over time, a rolling bearing degradation evaluation method based on the Max-Relevance and Min-Redundancy (mRMR) combined with support vector data description (SVDD) is proposed. According to the pauta criterion, the outliers and influential noises in the original data are eliminated, and the time domain characteristic indexes of the data are extracted. The sensitivity analysis and hierarchical clustering method between time-domain features are adopted, and the dimensionality reduction of data features is realized based on the principle of mRMR. And the time-domain features during normal operation are sent as training samples to the optimized SVDD evaluation model to realize performance degradation assessment. Finally, smooth optimization is performed according to the SVDD distance change curve, and a formula for mapping distance to health is proposed. The obtained result curve can be used as a fitting of the remaining useful life prediction function. Using the full life data sample of the bearing to verify the proposed method, accurate evaluation and life prediction results were obtained.