Bearing fault diagnosis method based on CNN-LightGBM

Abstract

Abstract In recent years, deep neural networks have been widely used in bearing fault diagnosis. Aiming at the problem that the traditional convolutional neural network (CNN) has insufficient generalization ability and low accuracy in rolling bearing fault diagnosis, this paper presents a new fault diagnosis model based on CNN and light gradient boosting machine (LightGBM). By using the original vibration signal directly, the new model firstly uses the deep convolutional neural network with small kernel to extract features and introduces the batch normalization and the Adam algorithm to improve the rolling bearing fault diagnosis model convergence speed and generalization ability. Then, combined with the efficient and accurate characters of LightGBM in classification prediction, the extracted features are imported into LightGBM for training to complete bearing fault diagnosis of different fault types. This new model can realize the fault diagnosis of rolling bearing directly from end to end without extracting the fault feature of rolling bearing vibration signal by hand. The results of the comparative experiments on the Case Western Reserve University (CWRU) public bearing dataset show that the new model improves the accuracy compared to just using CNN. The experimental results also confirm that the proposed method has feature learning ability and has good ability for fault diagnosis.