融合注意力机制的二维CNN变工况轴承故障诊断

Improved CNN Bearing Fault Diagnosis under Variable Working Conditions with Attention Mechanism

针对重载滚动轴承在交变工况条件下运行时导致的故障诊断效果差、分类精度低的问题,提出一种融合注意力机制的二维CNN滚动轴承诊断方法。对滚动轴承在正常运行时和发生不同故障时的原始信号采用短时傅里叶变换处理,并为其设定标签作为训练样本。为提高特征识别率,设计了融合注意力机制的多尺度卷积模块,通过对滚动轴承有效的特征给予更多“注意”,从而提高滚动轴承的故障诊断精度,并借助江南大学轴承数据集加以实验验证。研究表明:以准确率和混淆矩阵为评价指标,在单工况条件下,融合注意力机制二维卷积神经网络平均准确率为99.60%,在变工况条件下的平均准确率为98.41%,均优于BP网络、标准CNN和支持向量机(SVM)模型,实现了轴承故障的高效诊断。

Aiming at the problems of poor fault diagnosis effect and low classification accuracy caused by the operation of heavy rolling bearings under alternating working conditions,a two-dimensional CNN rolling bearing diagnosis based on the attention mechanism was proposed.The original signals of rolling bearings in normal operation and different faults were processed by short-time Fourier transform,and labels were set for them as training samples.In order to raise the feature recognition rate,a multi-scale convolution module integrating attention mechanism was designed to improve the fault diagnosis accuracy of rolling bearings by giving more“attention”to the valid features of rolling bearings.Experimental verification was carried out with the help of the bearing data set of Jiangnan University.The research showed that with the accuracy and confusion matrix as the evaluation index,the average accuracy of the convolution neural network fused with the attention mechanism reached to 99.60%under the condition of single mode,and 98.41%under the condition of the variable modes,superior to BP network,standard CNN model and support vector machine(SVM),realizing the effective diagnosis of bearing faults.