基于FSST-ResNet的滚动轴承变工况数据故障诊断研究

Research on fault diagnosis of rolling bearing variable condition data based on FSST-ResNet

针对滚动轴承运行工况复杂多变,难以诊断的问题,提出一种基于短时傅里叶的同步压缩变换(FSST)与残差神经网络(ResNet)相结合的故障诊断方法。该方法对轴承振动信号做短时傅里叶变换后,将时频系数压缩重排,生成基于FSST时频图在3种变工况下的训练集和测试集。考虑到深度模型的网络退化问题,采用ResNet模型对数据集进行时频特征提取和故障诊断,进一步提升轴承故障诊断的精度。通过3种变工况实验证明了该方法的有效性和可行性,平均诊断准确率高达98.9%,与其他方法相比,诊断精度有较大提高。

Aiming at the problem that rolling bearing operating conditions are complex and changeable and difficult to diagnose,a fault diagnosis method based on the combination of short-time Fourier Synchrosqueezing Transform (FSST) and Residual Neural Network (ResNet) was proposed.This method performs short-time Fourier transform on the bearing vibration signal,compresses and rearranges the time-frequency coefficients,and generates a training set and a test set based on the FSST time-frequency diagram under three variable operating conditions.Taking into account the network degradation problem of the deep model,the ResNet model was used to extract time-frequency features and fault diagnosis of the data set to further improve the accuracy of bearing fault diagnosis.The effectiveness and feasibility of this method were proved through three variable-condition experiments.The average diagnostic accuracy rate is as high as 98.9 %.Compared with other methods,the diagnostic accuracy is greatly improved.