Rolling bearing fault diagnosis with combined convolutional neural networks and support vector machine

[Display omitted] •A fault diagnosis method with three stopping conditions is proposed for small sample.•The convolutional neural networks is applied for feature extraction.•Fault classification is carried out by the support vector machines.•The addition of stop conditions can improve efficiency and accuracy of the method. For small sample data, it is difficult to complete the requirements of training complex models in the field of fault diagnosis. To solve the problem, this paper combines convolutional neural network's excellent feature processing ability with the excellent generalization ability of Support Vector Machine (SVM). The proposed CNN-SVM system is applied in bearing fault diagnosis, which takes the time domain diagram of bearing vibration data as the system input. The features are extracted by CNN, and realizes the final bearing state recognition by SVM. The contribution of the paper is to add three conditions for automatically switch CNN to SVM. The results show that the system has the advantages of less time-consuming, high precision and strong generalization ability. Experimental results show that the time consumption of this model is 1/3 of CNN, and the accuracy of the training set and the testing set are 100% and 99.44%.