基于深度小波去噪自动编码器的轴承智能故障诊断方法  

AN INTELLIGENT FAULT DIAGNOSIS METHOD FOR BEARING BASED ON DEEP WAVELET DENOISING AUTOMATIC ENCODER

作　　者：李晓花[1] 江星星[2] Li Xiaohua;Jiang Xingxing(Shangqiu Polytechnic,Shangqiu 476000,Henan,China;School of Rail Transportation,Soochow University,Suzhou 215137,Jiangsu,China)

机构地区：[1]商丘职业技术学院,河南商丘476000 [2]苏州大学城市轨道交通学院,江苏苏州215137

出　　处：《计算机应用与软件》2021年第2期145-151,190,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(51705349)。

摘　　要：针对原始振动数据无监督特征学习问题,提出一种深度小波去噪自动编码器与鲁棒极限学习机相结合的滚动轴承的智能故障诊断方法。利用小波函数作为非线性激活函数设计小波去噪自动编码器,从而有效地捕获信号特征;利用多个小波去噪自动编码器构造一个深度小波去噪自动编码器来增强无监督特征学习能力;采用鲁棒极限学习机作为分类器,对不同的轴承故障进行分类识别。对实验所得的轴承振动信号进行对比分析,结果验证了该方法在原始振动数据无监督特征学习的条件下优于传统方法和标准深度学习方法。Aiming at unsupervised feature learning of original vibration data,an intelligent fault diagnosis method for rolling bearings is proposed,which combines deep wavelet denoising automatic encoder with robust limit learning machine.The wavelet function was used as a non-linear activation function to design an automatic wavelet de-noising coder to capture the signal features effectively;a deep wavelet denoising automatic coder was constructed by using multiple wavelet denoising automatic coders to enhance unsupervised feature learning ability;robust limit learning machine was used as classifier to classify and identify different bearing faults.The experimental results show that the proposed method is superior to the traditional method and the standard deep learning method under the condition of unsupervised feature learning of the original vibration data.

关 键 词：智能故障诊断 滚动轴承 深度小波去噪自动编码器 极限学习机 无监督特征学习 

分 类 号：V263.6[航空宇航科学与技术—航空宇航制造工程] TP273[自动化与计算机技术—检测技术与自动化装置]