Gray relation weighted wavelet neural network integrated model and its application in rotating machinery fault diagnosis

Considering the variability and complexity of rotating machinery fault diagnosis, the fault diagnosis information carried out by a single model is not comprehensive and the fault identification rate is not high, so it cannot fully reflect the mechanical operation status. In response to this situation, this paper proposes an integrated learning method based on wavelet neural networks as base classifiers for fault diagnosis of rotating machinery, which largely meets the requirements of strong interpretability, high accuracy, and good stability. To start with, the method utilizes empirical mode decomposition (EMD) processing and decision layer fusion on multiple independent sensor signals separately, resulting in interpretative and discriminative extraction of fusion features. Next, a rose diagram classifier based on gray correlation weighting, which integrates the outputs of multiple wavelet neural networks (WNNs), is constructed, which is a more intuitive and reliable method for fusing multiple decision results using graphics. In addition, decision fusion based on rose diagram circularity is performed to realize rotating machinery fault diagnosis. Finally, the method is validated using the gear dataset and the bearing dataset, and the results of the analysis showed that the fault diagnosis accuracies for gears and bearings under two operating conditions are 97.72%, 99.41%, and 99.32%. The new method has a better fault diagnosis stability and higher fault diagnosis accuracy compared to other comparative methods.