No abstract

The number of unlabeled data is much more than labeled data. There are some challenges in using existing complex data for fault diagnosis. Therefore, we proposed a semi-supervised fault diagnosis algorithm based on improved fuzzy C-means algorithm (IFCM) and Convolutional Neural Network (CNN). Firstly, the fault original signal is extracted by variational mode decomposition (VMD) and the singular value decomposition (SVD) to extract the fault feature vector. Secondly, IFCM is used to obtain the membership degree of the data to all categories, which integrates the labeled data into the cluster center of fuzzy C-means (FCM) to obtain better clustering results. The label of labeled data will spread to unlabeled data in the same category. Thirdly, data with a membership degree greater than the value of 0.9 will be selected as the core data set. Finally, the original signal and labels of core data set will be input into CNN for fault diagnosis. The result shows that the fault recognition rate of IFCM-CNN reaches 95%. The fault diagnosis effect is significantly better than other algorithms.

Equipment usually breaks down suddenly and irregularly, so most of the data sets obtained for fault diagnosis have unbalanced characteristics, and the amount of data varies greatly from different fault types. In this paper, three problems in the application of synthetic minority oversampling technique (SMOTE) are studied, and the improved SMOTE algorithm combined with support vector machine (SVM) is proposed. The validity of the model is verified by CWRU bearing data compared with SVM and SMOTE+SVM methods, and the result of fault diagnosis is satisfactory.

No abstract