基于优化残差卷积网络的滚动轴承变工况故障诊断

Variable Condition Fault Diagnosis of Rolling Bearings Based on Optimized Residual Convolution Network

针对滚动轴承在实际运行环境中同时存在变负荷和变噪声的复合工况干扰而产生的故障诊断效果不理想的问题,提出了一种用于滚动轴承变工况故障诊断的一维残差卷积神经网络方法。将归一化后整理完的原始轴承振动信号输入到网络模型中,利用具有残差连接的多个一维卷积层提取特征,再经过多个卷积池化,最后输入到Softmax层进行分类,输出轴承振动信号的故障类型。将所提方法与一维卷积神经网络(CNN)、LeNet-5和AlexNet几个经典模型进行对比分析,结果表明,本文方法在变噪声实验和变负荷实验中的平均准确率分别为94.16%和95.31%,均高于其他经典神经网路,具有较强的抗噪性和泛化性能力。

In order to solve the problem that the fault diagnosis effect of rolling bearings is not ideal due to the interference of variable load and variable noise in the actual operating environment,a one-dimensional residual convolutional neural network method was proposed for the fault diagnosis of rolling bearings under variable working conditions.The normalized original bearing vibration signals were input into the network model,and multiple one-dimensional convolution layers with residual connections were used to extract features.After multiple convolution pooling,they were finally input into the Softmax layer for classification,and the fault types of bearing vibration signals were output.The proposed method was compared with the classical models of one-dimensional convolutional neural network(CNN),Lenet-5 and AlexNet.The results show that the average accuracy of the proposed method is 94.16%and 95.31%in the variable noise experiment and the variable load experiment,respectively,which are higher than other classical neural networks.The proposed method has strong noise resistance and generalization ability.