Algorithm 3 Bearing fault diagnosis method based on Feature Fusion

Input:  bearing signal sequence $f\left(t\right),t=1,2,\cdots ,m\lambda$, window width $\lambda$, wavelet packet scale l, wavelet packet function type, Number of runs $\tau$. Output:  classification accuracy $\eta$, calculation time, Variance of classification accuracy $\delta$. 1:The feature matrix of bearing fault is obtained $T_{m\times r}$ by Algorithm 2; 2:The feature matrix $T_{m\times r}$ is randomly divided into the training set and test set, and different state types are labeled; 3:The SVM classifier is trained by the training set to obtain the SVM-based classification model; 4:The test set is input to the SVM-based classification model to obtain the predicted label of the test set. The actual label and predicted label of the test set are calculated according to Equation (26) to calculate the classification accuracy $\eta$ of the diagnostic model. The total running time of bearing signal from MVSVD feature extraction to training SVM classification model to test result is calculated; 5:The model is run $\tau$ times in sequence to get the classification accuracy $\eta$ each time, and the variance $\delta$ of classification accuracy is obtained according to Equation (27).