基于残差神经网络的滚动轴承故障诊断方法

针对轴承故障诊断方法在变工况条件下诊断效果不佳的问题,提出了一种基于残差神经网络的滚动轴承故障诊断方法.该方法首先以滚动轴承时域信号数据作为输入,针对信号的时变性改进了数据池化层,改进的数据池化层利用三个连续的卷积层串联构建而成,目的在于能够有效地提取振动信号中的故障特征信息,并减少残差神经网络中参数的计算量;然后设计了一种空洞卷积和残差块相结合的空洞残差块,用于特征信息的学习;最后通过在全连接层后加入Dropout层丢弃一定比例的神经元,能有效避免过拟合的负面影响.使用凯斯西储大学轴承数据集进行仿真实验,与SVM+EMD+Hilbert包络谱、BPNN+EMD+Hilbert包络谱和Resnet三种方法作对比分析,结果表明该方法在变工况下的滚动轴承故障诊断中具有更高的诊断准确率、更强的抗噪性和泛化能力.

Rolling bearing fault diagnosis method based on residual neural network

Because the rolling bearing has been working for a long time in an environment with complex and changeable working conditions and large noise interference, the bearing fault diagnosis method has a poor diagnostic effect under variable working conditions, aiming at this problem, a method of rolling bearing fault diagnosis based on residual neural network is proposed. First, the time domain signal data of rolling bearing is used as the input. Because the signal has a strong time-varying nature, the data pooling layer is improved for this characteristic. The improved data pooling layer is constructed by using three consecutive convolution layers in series. The purpose is to effectively extract feature information of the vibration signal and reduce the calculation amount of residual neural network parameters. Then a kind of dilated residual block combined with dilated convolution and residual block is designed for feature information learning. Finally, the dropout layer is added after the full connection layer to discard a certain proportion of neurons, which can effectively avoid the negative effects of over fitting. The bearing data sets of Case Western Reserve University is used for simulation experiment, and compared with SVM+EMD+Hilbert envelope spectrum, BPNN+EMD+Hilbert envelope spectrum and Resnet, the results show that the proposed method has higher diagnosis accuracy, stronger noise resistance and generalization ability in the fault diagnosis of rolling bearing under variable working conditions.