宽转速范围下的航发主轴轴承故障诊断方法  

Fault diagnosis method of aero engine main shaft rolling bearings in wide rotating speed range

作　　者：张伟涛[1] 崔丹 刘璐[1] 黄菊 ZHANG Weitao;CUI Dan;LIU Lu;HUANG Ju(School of Electronic Engineering,Xidian University,Xi’an 710071,China;Guiyang Engine Design and Research Institute,China Aero Engine Corporation,Guiyang 550081,China)

机构地区：[1]西安电子科技大学电子工程学院,西安710071 [2]中国航发贵阳发动机设计研究所,贵阳550081

出　　处：《振动与冲击》2023年第5期253-262,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金资助项目(62071350)。

摘　　要：针对航空发动机主轴转速范围大而导致现有卷积神经网络(convolution neural network, CNN)故障诊断性能急剧下降的问题,提出了一种基于小波包重构成像与深浅层特征融合分类网络的故障诊断方法。利用小波包分解(wavelet packet transform, WPT)提取滚动轴承振动信号中的有效成分,消除与故障特征无关的干扰分量。后采用短时傅里叶变换对重构后的振动信号进行成像,得到时频谱样本。针对转速时变下的轴承故障分类问题,通过跳跃连接方式建立具有深浅层特征融合特性的卷积神经网络,实现故障分类预测。利用航发轴承试验机采集得到的多路轴承振动信号对提出的方法进行有效性验证,结果表明,在训练集和测试集样本具有不同转速的情况下,使用提出方法对不同类型故障仍具有很高的识别精度。Here, aiming at the problem of fault diagnosis performance of existing convolutional neural network(CNN) declining sharply due to wide rotating speed range of main shaft of aeroengine, a fault diagnosis method based on fusion classification network of wavelet packet reconstruction image and deep-shallow layer features was proposed. Firstly, the wavelet packet transform(WPT) was used to extract effective components of rolling bearing vibration signal, and eliminate interference components irrelevant to fault characteristics. Then, the reconstructed vibration signal was imaged with short-time Fourier transform to obtain time-frequency spectrum samples. Finally, aiming at the problem of bearing fault classification under time-varying rotating speed, the CNN with deep-shallow layer features fusion characteristics was established by means of skip connection to realize fault classification prediction. The effectiveness of the proposed method was verified by using multi-channel bearing vibration signals collected with the aeroengine bearing test machine. The results showed that the proposed method has high recognition accuracy for different types faults although training set and test set samples have different rotating speeds.

关 键 词：滚动轴承 故障诊断 小波包分解(WPT) 卷积神经网络(CNN) 

分 类 号：TP206[自动化与计算机技术—检测技术与自动化装置] TH133.3[自动化与计算机技术—控制科学与工程]