基于特征融合的注意力增强卷积神经网络的航空发动机滚动轴承故障诊断方法

针对现有基于深度卷积神经网络的故障诊断方法只考虑对信息局部特征的提取、忽视全局信息的不足，将可以把握全局信息的注意力机制融入卷积层，使得注意力机制参数和卷积层参数参与网络的训练，提出一种注意力增强卷积神经网络的机械故障诊断方法。通过经验模态分解、变分模态分解和小波包分解的方法提取滚动轴承振动信号的高维特征模量；将特征模量组成多通道样本输入到注意力增强卷积神经网络中进行训练，利用网络对特征模量自适应地融合和选择，从而挖掘特征模量的隐式特征；使用Softmax分类器进行分类识别；通过训练好的网络对高转速下的滚动轴承进行故障诊断；利用不同信噪比的信号对所提方法进行测试，以验证网络的泛化能力和故障诊断效果。实验结果表明：该方法能准确、有效地对航空发动机滚动轴承不同故障的损伤程度进行分类识别。

Fault Diagnosis for Aero-engine Rolling Bearings Based on an Attention Augmented Convolutional Neural Network with FeatureFusion

The existing fault diagnosis methods based on deep convolutional neural networks have only considered the extraction of local features while ignoring global information. To address this problem, an attention mechanism that can grasp global information is integrated into the convolution layer. The purpose is to allow both attention mechanism parameters and convolutional layer parameters to participate in network training. An attention augmented convolutional neural network for mechanical fault diagnosis is proposed. First, the high-dimensional feature modulus of the vibration signal of a rolling bearing fault is extracted using empirical mode decomposition, variational mode decomposition, and wavelet packet decomposition. Then, multi-channel samples composed of the feature modulus are input into the attention augmented convolutional neural network for training. The feature modulus are adaptively fused and selected by the network to mine the hidden information. Finally, the classifier Softmax is used to recognize and classify the results. The fault diagnosis of rolling bearings at high rolling speeds is carried out by the trained model. The proposed model is further tested by signals with different signal-to-noise ratios to verify the generalization ability of the network and the effectiveness of fault diagnosis. The experimental results show that the proposed method can accurately and effectively classify the damage degree of different bearing faults in aero-engines.