基于多尺度加权融合特征学习的转子故障诊断  

Rotor Fault Diagnosis Based on Multi-Scale Weighted Fusion Feature Learning

作　　者：王展 鲁晨琪 星施宇 王卿源 WANG Zhan;LU Chenqi;XING Shiyu;WANG Qingyuan(School of Mechanical Engineering,Shenyang Jianzhu University,Shenyang 110168,China;Shenyang Peltec Technology Co.,Ltd.,Shenyang 110000,China)

机构地区：[1]沈阳建筑大学机械工程学院,沈阳110168 [2]沈阳派尔泰科科技有限公司,沈阳110000

出　　处：《组合机床与自动化加工技术》2023年第11期154-158,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金项目(52175107,52205117,51975388);东北大学航空动力装备振动及控制教育部重点实验室研究基金资助项目(VCAME202008)。

摘　　要：针对一维振动信号表达故障特征信息不全面及转子故障信噪比低的问题,提出一种基于多尺度加权融合特征学习的转子故障诊断方法。首先,对时域振动信号的幅值进行标准化处理,利用对称点模式(SDP)原理将多传感器振动信息融合为二维SDP图像,通过选取适当的时间滞后系数和角增益,突出不同故障下SDP图像的特征;其次,构建了一种多尺度加权卷积神经网络(MSW-CNN)模型,利用3个不同的感受野分别提取图像特征,通过对多通道图像特征进行加权融合提高了模型的分类精度;最后,利用MSW-CNN模型对6种故障状态的SDP图像进行特征提取并分类。实验结果表明,与其他3种方法相比,所提方法的转子故障诊断精度更高,达到99.31%,在噪声干扰下的诊断精度为96.23%,验证了所提方法的有效性。Aiming at the problems of incomplete fault feature information expressed by one-dimensional vibration signal and low signal-to-noise ratio of rotor fault,a rotor fault diagnosis method based on multi-scale weighted fusion feature learning is proposed.Firstly,the amplitude of time-domain vibration signal is standardized,and multi-sensor vibration information is fused into two-dimensional SDP image using the principle of symmetrical point pattern(SDP).By selecting appropriate time lag coefficient and angular gain,the characteristics of SDP image under different faults are highlighted.Secondly,construct a multi-scale weighted convolution neural network(MSW-CNN)model,which uses three different receptive fields to extract image features respectively,and improves the classification accuracy of the model by weighted fusion of multi-channel image features.Finally,use the MSW-CNN model to extract and classify the SDP images of six fault states.The experimental results show that compared with the other three methods,the rotor fault diagnosis accuracy of the proposed method is higher,reaching 99.31%,and the diagnosis accuracy under noise interference is 96.23%,which verifies the effectiveness of the proposed method.

关 键 词：SDP图像 多尺度加权融合 特征提取 转子系统 故障诊断 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]