基于改进CNN的轴承声学故障诊断

针对轴承振动信号在复杂机械中难采集和跨转速域工况下传统故障诊断方法精度低的问题，文中提出了一种基于Teager能量算子和卷积神经网络的滚动轴承声学故障诊断方法，即TEO-CNN。将轴承声学信号的Teager能量算子作为模型的输入，使用卷积神经网络学习输入的抽象特征，并结合全局平均池化层和全连接层实现轴承健康状态识别。模型验证基于轴承声学实验数据，并通过构建不同的轴承声学数据集模拟跨转速域工况。试验结果表明，与传统卷积神经网络和机器学习模型相比，TEO-CNN表现出明显的优势，并且在跨转速域工况下的预测精度始终高于95%。

An Improved CNN Method for Bearing Acoustic Fault Diagnosis

In view of the difficulty to collect bearing vibration signals in complex machinery and the poor accuracy of traditional fault diagnosis methods under cross working speed conditions, a rolling bearing acoustic fault diagnosis method is proposed based on TEO-CNN. Teager energy operator of raw acoustic signals is taken as the input of TEO-CNN model, the CNN is employed to extract the abstract features from inputs, and the global average pooling layer and the fully connected layer are combined to recognize the bearing health status. TEO-CNN is verified on bearing acoustic experimental data sets, and cross working speed conditions are simulated by constructing different bearing acoustic data sets. Experimental results show that compared with traditional convolutional neural networks and machine learning models, the proposed TEO-CNN shows obvious superiority, and the prediction accuracy is always higher than 95% under cross working speed conditions.