注意力机制结合CNN卷积网络的滚动轴承复合故障诊断

Composite Fault Diagnosis of Rolling Bearing Based on Attention Mechanism and CNN Convolution Network

针对滚动轴承复合故障诊断精度低的问题,运用注意力机制对滚动轴承振动信号不同特征进行自动权重分配,以强化重要特征,弱化冗余特征;采用CNN卷积神经网络对滚动轴承故障进行诊断;在公开的滚动轴承数据集上对ATT-CNN模型进行验证。结果表明:ATT-CNN模型可准确诊断0 kW~3 kW负载和单一工况、多工况下的滚动轴承故障,对单一工况滚动轴承故障诊断的平均准确率可达97.23%,对两种或三种多工况滚动轴承故障诊断的平均准确率可分别达96.48%和83.40%。

In view of the low accuracy of composite fault diagnosis of rolling bearing faults,attention mechanism is applied to perform automatical weight allocation on different characteristics of rolling bearing vibration signals for important feature enhancement and and redundant feature weakening.CNN convolution network is adopted to diagnose rolling bearing faults.Based on the rolling bearing data set publishehed by Western Reserve University,ATT-CNN model is tested and verified.The results show that the model can accurately diagnose the single working condition and multi working condition rolling bearing faults under 0 kW~3 kW load,among which average accuracy of single working condition rolling bearing fault diagnosis can reach 97.23%,while the average accuracy of two or three multi working condition rolling bearing fault diagnosis is 96.48%and 83.40%respectively.