Fault diagnosis method of rotating machinery based on multi-sensor information decision-level fusion

In order to improve the reliability and safety of rotating machinery systems, a fault diagnosis method based on consistency class weighted soft voting strategy is proposed in this paper, which consists of the basis classifier and decision fusion. The basis classifier includes a multi-scale feature extraction module and a classification module. In the first module, A multi-scale convolution kernel is employed to achieve the extraction and fusion of diverse features. To further improve the generalization performance, feature dimensionality reduction of the global average pooling layer is also introduced to effectively suppress noise interference and reduce overfitting. Then, the multi-scale features are input into the classification module to obtain the prediction label. Finally, a consistent category-weighted soft voting strategy is designed to perform decision fusion, which takes into account the difference in recognition ability of different basis classifiers for the same class while focusing on the difference in recognition ability of the base classifier for each fault class. By combining the probability distribution of prediction labels, the combined weights of different prediction labels are reasonably assigned. The fault dataset of a mechanical fault comprehensive simulation test-bed with Gaussian white noise is used to evaluate the method. The experimental results show that the proposed method is superior to other existing fault diagnosis methods for rotating machinery under noise interference, which proves that it has strong noise suppression ability.