1DAMCNN方法在滚动轴承故障诊断中的应用

Application of 1DAMCNN Method in Fault Diagnosis of Rolling Bearings

针对传统滚动轴承故障诊断方法中特征提取依赖人工经验、处理过程繁琐以及识别准确率较低的问题,提出一种端到端基于一维注意力混合卷积神经网络(One Dimensional Attention Mixed Convolution Neural Network,1DAMCNN)的轴承故障诊断方法。该方法首先引入空洞卷积,构造混合卷积用于增大特征提取的感受野范围,以获取更全面的特征信息。然后加入注意力机制,增强模型对关键特征信息的提取能力,实现对轴承故障的智能诊断。试验数据分析结果表明,相比其他故障诊断方法,固定负载工况下该方法自适应性强,准确率高达99%以上。在只有60个样本量的情况下,该故障诊断方法准确率超过88%,表明其具有出色的特征提取能力。最后通过对比实验和可视化技术,验证所提方法的有效性。

In order to solve the problems of the dependence of feature extraction on manual experience,tedious processing process and low recognition accuracy in traditional rolling bearing fault diagnosis methods,an end-to-end bearing fault diagnosis method based on one-dimensional attention mixed convolution neural network(1DAMCNN) is proposed.Firstly,the hole convolution is introduced and the mixed convolution is constructed to enhance the receptive field range for feature extraction,so as to obtain more comprehensive feature information.Then,the attention mechanism is added to enhance the ability of the model to extract the key feature information,and the intelligent diagnosis of bearing fault is realized.The analysis results of test data show that under fixed load conditions,compared with other fault diagnosis methods,this method has strong adaptability and high accuracy,and the accuracy is higher than 99 %.When the sample size is only 60,the fault recognition accuracy of this method is higher than 88 %,indicating that it has excellent feature extraction ability.Finally,the effectiveness of the proposed method is verified by comparing its results with those of experiments and visualization techniques.