【摘要】 针对涡旋压缩机振动信号不平稳和噪声情况下故障振动信号弱、需要人为提取故障特征以及准确率有待进一步提高的问题，提出基于多尺度注意力机制（convolutional block attention mechanism，简称CBAM）-卷积神经网络（convolutional neural network，简称CNN）涡旋压缩机故障诊断方法。首先，通过多个不同尺度的卷积核对振动信号转化为灰度图的故障特征进行全面提取，并引入注意力机制，通过调整权重值的方式提取重要的故障特征；其次，利用降维卷积模块、深度可分离卷积模块和残差模块提取更高维度的深层次故障特征，提升网络计算效率；最后，设置舍弃率为0.5的Dropout层防止过拟合，提升了网络的鲁棒性、抗干扰能力和泛化能力。实验结果证明，该方法在无噪声和添加不同信噪比噪声的情况下，均能有效地对涡旋压缩机故障进行分类，具有更高的识别准确性和更快的收敛能力。更多还原

【Abstract】 The vibration signal of scroll compressor is unstable and the fault vibration signal is weak under the condition of noise, which requires manual extraction of fault features and is difficult to be recognized. Therefore, a fault diagnosis method of scroll compressor combining convolutional neural network(CNN) with multi-scale convolutional block attention mechanism(CBAM) is proposed. First, the fault features of the gray-scale map transformed from vibration signal are comprehensively extracted by using multiple-scale convolution check. Then, CBAM is introduced to extract important fault features by adjusting the weight value. Third, the reduced dimension convolution module, the deep separable convolution module and the residual module are utilized to extract rich fault features with higher dimensions and deep levels, and to improve the network computing efficiency. Finally, a Dropout layer with a rejection rate of 0.5 is set to prevent over fitting, which improves the robustness, anti-interference ability and generalization ability of the network. Experimental results show that the proposed method can effectively classify faults both without noise and noise with different signal-to-noise ratio, exhibiting higher recognition accuracy and faster convergence ability in scroll compressor fault diagnosis.更多还原