A Novel Collaborative Bearing Fault Diagnosis Method Based on Multisignal Decision-Level Dynamically Enhanced Fusion

The development of effective methods for diagnosing bearing faults has received significant research attention. As the engineering environment tends to be complex, mechanical equipment in operation affected by the interference is more intricate and variable, resulting in a single sensor being increasingly unable to meet the needs of high-precision fault diagnosis of rotating machinery. In contrast, multisensors contain more comprehensive and redundant information, which can improve the reliability of fault diagnosis. In this article, a new collaborative fault diagnosis method based on multisensors is proposed, which dynamically enhances and fuses the diagnostic information of multiple sensors at the decision level to achieve more accurate and reliable results. First, a series of base models driven by statistical features and deep features in parallel are constructed to gain a series of pre-diagnosis results. Second, a new dynamically enhanced weighted voting strategy (DEWVS) is designed. Through the dual consideration of diagnostic accuracy and misclassification of the base models, the strategy constructs the diagnostic performance indicator matrix and realizes the dynamic assignment of the voting weights of each base model to enhance the effective information of each pre-diagnosis result, obtaining more reliable collaborative diagnostic results. Finally, the proposed method is evaluated by an experimental dataset of the axle box bearing of a high-speed train. The experimental results validate the necessity of multisensor collaborative diagnosis and demonstrate the superiority of the proposed method.