基于递归图和增强残差网络的轴承故障诊断

Fault Diagnosis for Bearings Based on Recurrence Plot and Enhance ResNet

针对噪声干扰情况下轴承振动信号特征难以充分提取,故障识别精度低的问题,提出将递归图与增强深度残差网络相结合的RP-EResNet模型并应用于轴承故障诊断。将非线性的振动信号嵌入到具有可变时滞的延迟坐标空间中生成二维的递归图,并将压缩-激励模块、多尺度卷积、分组卷积网络模块融合到残差网络结构中得到增强的RP-EResNet模型,最终将递归图输入RP-EResNet模型中进行轴承故障诊断。使用不同的轴承数据集验证了RP-EResNet模型的性能,消融试验和对比试验的结果表明:与不同的深度学习方法相比,RP-EResNet模型能够在强噪声下增强特征提取能力,提升轴承故障的识别精度,具有良好的泛化性能和抗噪性能。

In order to solve the problems of insufficient feature extraction from bearing vibration signals under noise interference and low fault identification accuracy,the RP-EResNet model combining recurrence plot(RP)and enhance ResNet(EResNet)is proposed and applied to bearing fault diagnosis.The nonlinear vibration signals are embedded into delay coordinate space with variable time delay to generate a two-dimensional recurrence plot.The squeeze excitation block,multiscale convolution and grouped convolution network block are integrated into ResNet structure to obtain enhance RP-EResNet model.The recurrence plot is then fed into RP-EResNet model for bearing fault diagnosis.The performance of RP-EResNet model is validated using different bearing datasets,and the results of ablation and comparison tests indicate that compared with other deep learning methods,the RP-EResNet model can enhance the feature extraction ability under strong noise,improve the identification accuracy of bearing faults,and demonstrate good generalization and anti-noise performance.