基于FB-LSTM ResNet的滚动轴承故障诊断方法

Fault Diagnosis Method for Rolling Bearings Based on FB-LSTM ResNet

提出一种基于FB-LSTM ResNet的滚动轴承故障诊断方法,并将故障诊断过程划分为3个部分。原始振动信号获取与处理模块利用振动信号窗平移方法完成重叠信号的分割,并利用Inception网络高效完成一维信号预处理;振动信号特征提取模块采用FB-LSTM ResNet网络,可以有效处理层深所引起的退化情况;故障诊断分类模块选用全局池化层代替全连接层,能够削减网络参量,从而有效规避过拟合情况。采用CWRU与QPZZ-II故障轴承样本集的试验结果表明,FB-LSTM ResNet方法在原始样本和加噪样本中均获得了最高的故障诊断准确率,并可在较少的迭代过程中达到较优的准确率与损失值,其效果优于单独的FB-LSTM,ResNet方法以及其他融合方法。

A fault diagnosis method for rolling bearings is proposed based on FB-LSTM ResNet,and the fault diagnosis process is divided into three parts.The original vibration signal acquisition and processing module mainly uses the vibration signal window translation method to segment the overlapping signals,and uses Inception network to efficiently preprocess the one-dimensional signals.The vibration signal feature extraction module adopts FB-LSTM ResNet network,which is able to deal with degradation caused by layer depth effectively.The fault diagnosis classification module selects global pooling layer instead of full connection layer,which is able to reduce the network parameters and thus effectively avoid the over-fitting situation.The test results using CWRU and QPZZ-II fault bearing sample sets show that the FB-LSTM ResNet method has the highest fault diagnosis accuracy in both original and noisy samples,and achieves better accuracy and loss value in less iterative process than that of FB-LSTM,ResNet method and other fusion methods.