针对强噪声环境下传统轴承故障诊断方法对故障识别率低,深度残差收缩网络在降噪时对频域信号丢失的问题,提出了一种基于连续小波变换(continuous wavelet transform,CWT)和改进的深度残差收缩网络(improved deep residual shrinkage network,IDRSN)的故障诊断模型.首先,利用CWT将轴承振动信号转换为二维时频图,作为输入样本,用于解决深度残差收缩网络在直接处理振动信号时引起的频域失真问题;其次,设计了一种改进的软阈值函数(improved soft threshold func-tion,ISTF),解决了因软阈值化引起的信号失真,设计了改进的软阈值模块(improved soft threshold block,ISTB)和自适应斜率模块(adaptive slope block,ASB),构建了改进的残差收缩单元(improved residual shrinkage building unit,IRSBU)以实现自适应地确定最佳阈值并进一步调整输出;最后,利用凯斯西储大学滚动轴承数据集与风机轴承振动数据采集实验台收集的滚动轴承数据集对所提方法进行实验验证.结果证明相较于其他方法,所提的故障诊断方法有更好的泛化性和通用性,分类准确率分别达到了99.75％和99.69％.
Fault Diagnosis of Fan Rolling Bearing Based on CWT-IDRSN
Aiming at the problems of low fault recognition rate of traditional bearing fault diagnosis method in strong noise environment and loss of signal in frequency domain when deep residual shrinkage network is used for noise reduction,a novel method based on continuous wavelet transform ( continuous wavelet trans-form,Fault diagnosis model of CWT and improved deep residual shrinkage network ( IDRSN ) . Firstly,CWT is used to convert the bearing vibration signal into a two-dimensional time-frequency graph,which is used as the input sample to solve the frequency domain distortion caused by the depth residual contraction network when directly processing the vibration signal. Secondly,an improved soft threshold function is de-signed to solve the signal distortion caused by soft threshold,an improved soft threshold module and an a-daptive slope module are designed,and an improved residual contraction unit is constructed to determine the optimal threshold and further adjust the output. Finally,using the rolling bearing data set collected by Case Western Reserve University and the wind turbine bearing vibration data collection experimental platform,the proposed method is verified experimentally. The results show that compared with other methods,the proposed fault diagnosis method has better generalization and universality,and the classification accuracy reaches 99.75％ and 99.69％ respectively.
万方数据
维普
