基于压缩采集特征提取与CNN_SVM的滚动轴承的故障诊断

Rolling Bearing Fault Diagnosis Method Based on the Compressed Sampling Feature Extraction and CNN_SVM

目的 针对传统卷积神经网络训练时间长、易过拟合、故障诊断精度低、抗噪能力差等问题,提出一种基于压缩采集特征提取与CNN_SVM的滚动轴承的故障诊断模型,降低滚动轴承故障数据的冗余度。方法 首先,使用压缩采集技术去除实验样本中的冗余信息;然后,使用三层卷积神经网络(CNN)对采集数据进行故障特征提取,在网络中加入Dropout层、Batch Normalization层、全局平均池化层来防止网络的过拟合,加强网络提取特征的能力;最后,用多分类支持向量机(SVM)对提取特征进行分类。结果 研究表明:模型对故障诊断精度达到了99.4%,比CNN_SVM,PCA_SVM,1D_CNN等模型故障诊断效果突出,对含噪的实验数据具有去噪功能。结论 笔者所提出的模型诊断精度高,且具有很强的学习能力和降噪能力。

In order to reduce the redundancy of the fault data of rolling bearings, and to address the problems of long training time, easy overfitting, low fault diagnosis accuracy and poor noise immunity of traditional convolutional neural networks, a fault diagnosis model of rolling bearings based on the compression, acquisition and feature extraction of signal transform domain and CNN_SVM is proposed.First, the redundant information in the experimental samples is removed using compression and acquisition technique;then, a three-layer convolutional neural network(CNN) is used to extract fault features from the acquired data, and dropout layer, batch normalization layer, and global average pooling layer are added to the network to prevent overfitting and enhance the ability of the network to extract features;finally, a vector machine(SVM)which support multi-classification is used to classify the extracted features.The results of the study show that the model achieves 99.4% of fault diagnosis accuracy, which is outstanding than CNN_SVM,PC_SVM,1 D_CNN and other models for fault diagnosis.The model also has denoising function for experimental data containing noise.The author′s proposed model has high diagnostic accuracy and strong learning ability and noise reduction ability.