A rolling bearing fault diagnosis method based on GADF-CWT-GCNN

基于GADF-CWT-GCNN的滚动轴承故障诊断方法研究

Key Laboratory of Road Construction Technology and Equipment for Ministry of Education, Chang’an University, Xi’an 710064, China

Received: 5 September 2023

Abstract

Because of poor model generalization ability and low diagnostic accuracy caused by rolling bearing fault diagnosis in a small sample environment, a novel method based on the Gram angle division field (GADF), the continuous wavelet transform (CWT) and the parallel two-dimensional group normalizatio convolutional neural network (P2D-GCNN) for the fault diagnosis of rolling bearings is proposed. Firstly, collected data are preprocessed and one-dimensional vibration signals are converted into two-dimensional images by using the Gram angle division field and the continuous wavelet transform as the input of the model. Then the data enhancement technique is used to expand the sample sub-graph to meet the input requirements of the network. The sample sub-graph is imported into the convolutional neural network with the group normalization algorithm for diagnostic detection. The results show that the generalization ability of the data processing method and the model built in this paper in the small-sample environment is much higher than that of other network models such as the small vector machine and the 1D-CNN. In order to further verify the recognition ability of the model in the small sample environment, the sample sizes of 70%, 40% and 20% of the dataset are used to do experiments many times. Their corresponding training accuracy and test accuracy were 99.38%, 99.02%, 99.47%, 98.29%, 99.05% and 97.08% respectively, indicating that the model is highly accurate for the fault diagnosis of rolling bearings in the small sample environment.

摘要

针对滚动轴承故障诊断在小样本环境下引起的模型泛化能力差、诊断精度低的问题, 提出一种基于格拉姆角分场(GADF)和连续小波变化(continuous wavelet transform, CWT)与并行二维组归一化卷积神经网络(parallel convolutional neural network, P2D-GCNN)的滚动轴承故障诊断方法。对采集的数据进行预处理, 采用格拉姆角场和连续小波变换将一维振动信号转换成二维图像作为模型输入, 再选用数据增强技术扩充样本子图, 满足网络输入要求, 并将其导入搭建的组归一化卷积神经网络中进行诊断检测。结果表明: 文中数据处理方法与搭建模型在小样本环境下泛化能力远高于SVM和1D-CNN等其他网络模型。为进一步验证模型在小样本数据下的识别能力, 取数据集的70%, 40%和20%样本量进行多次实验, 所对应的训练准确率及测试准确率分为99.38%, 99.02%, 99.47%, 98.29%, 99.05%, 97.08%。结果证明, 文中模型在小样本环境下对轴承故障诊断具有很高的准确率。