Performance degradation and maintenance optimization strategy of rolling bearings based on data fusion and adaptive partition

The traditional method of performance degradation prediction and maintenance of rolling bearings only considers a single sensor signal, which makes it difficult to automatically partition degradation stages and prone to over-detection. A new method of performance degradation evaluation and maintenance of rolling bearings based on data-level fusion, adaptive health state partitioning, and state maintenance is proposed. Firstly, considering the degradation and impact in the process of bearing deterioration, the multi-sensor signals are dynamically weighted to achieve data-level fusion. Secondly, a bearing health index was established based on fast spectral correlation, Wasserstein distance, and linear rectification techniques. On this basis, by combining the Bayesian information criterion and the elbow rule, the precise division of rolling bearing health state is realized through hidden Markov model regression. Then, random forest was used to classify and predict the data to verify the validity of the proposed data fusion method and health indicator. Finally, condition-based maintenance strategy based on the fourth moment, stress-strength interference model, and Gamma process is proposed to avoid excessive detection and reduce maintenance costs. Through accelerated degradation experiments and field validation tests on the rolling bearing test data set of Xi’an Jiaotong University and FEMTO (PRONOSTIA), the accuracy and superiority of the proposed method in the prediction and maintenance of bearing health state are verified.