Fuzzy Collaborative Diagnostic Method for Motor Bearings under Strong Noise

Objectives Marine motor bearings are subjected to strong noise interference caused by wind and waves, hull vibration, and electromechanical coupling during navigation. Existing methods have difficulty in effectively characterizing feature fuzziness and class-boundary uncertainty in vibration signals. To address these issues, a rolling bearing fault diagnosis method based on TCN-SCKF-IT2FNN is proposed. Methods First, a Temporal Convolutional Network (TCN) is employed to extract deep temporal features from vibration signals. An Interval Type-2 Fuzzy Neural Network (IT2FNN) is then used to model the feature fuzziness and class-boundary uncertainty induced by noise. Finally, a Square-Root Cubature Kalman Filter (SCKF) is introduced to recursively optimize the antecedent and consequent parameters of the IT2FNN at the rule level, thereby improving the stability of parameter estimation. Results Experiments are conducted on the Case Western Reserve University (CWRU) bearing dataset and a self-built bearing test-rig dataset. The proposed method achieves average diagnostic accuracies of 99.80% and 99.69%, respectively, under four operating conditions. Under composite noise at an SNR of −4 dB, the Macro-F1 scores reach 94.37% and 90.58%, representing improvements of 7.53 and 8.86 percentage points over the TCN baseline, respectively. Conclusions Compared with the other methods, the proposed approach exhibits good diagnostic accuracy and robustness under complex operating conditions and effectively alleviates the feature instability and class-boundary ambiguity caused by noise interference.