基于高斯滤波与多尺度CNN的抗噪轴承故障诊断方法  

Fault Diagnosis Method for Anti-noise Bearings Based on Gaussian Filter and Multi-scale CNN

作　　者：路新华 韩风超[2] 马丽 张凌晓 孙鹏 LU Xinhua;HAN Fengchao;MA Li;ZHANG Lingxiao;SUN Peng(School of Information Engineering,Nanyang Institute of Technology,Nanyang 473004,Henan,China;School of Information Engineering,Zhengzhou University,Zhengzhou 450001,China;School of Computer and Software,Nanyang Institute of Technology,Nanyang 473004,Henan,China)

机构地区：[1]南阳理工学院信息工程学院,河南南阳473004 [2]郑州大学信息工程学院,郑州450001 [3]南阳理工学院计算机与软件学院,河南南阳473004

出　　处：《噪声与振动控制》2024年第4期132-137,共6页Noise and Vibration Control

基　　金：国家自然科学基金资助项目(61901417);河南省自然科学基金资助项目(222300420504);河南省高等教育学改革研究与实践资助项目(学位与研究生教育)成果(2021SJGLX262Y);河南省科技攻关资助项目(212102210173)。

摘　　要：针对滚动轴承在强背景噪声环境下故障诊断性能不佳,已有基于深度学习和降噪处理的故障诊断模型规模较大,复杂度较高导致难以实际部署的问题,提出一种基于高斯滤波(Gaussian-filter)和多尺度卷积神经网络(MultiScale Convolution Neural Network,MSCNN)的滚动轴承故障诊断方法,可在噪声环境中实现低复杂度、高精度的故障诊断,并对不同负载情况具有高鲁棒性。该方法首先构造适应不同信噪比的最佳滤波核,然后对有噪信号进行降噪处理,最后使用MSCNN自适应提取信号多尺度特征,实现多类别故障诊断。实验结果表明,与当下最先进的各种故障诊断方法相比,该方法在各种强度噪声下均具有较高故障诊断精度,且在时间和空间维度上具有较低的复杂度,有望在工业生产中得到实际应用。The problems that the fault diagnosis performance of rolling bearings under strong noise background of noisy environment is poor and the size of the existing fault diagnosis model based on deep learning and noise reduction has large scale and high complexity which leads to practical difficulty to deploy,are studied.A fault diagnosis method for rolling bearings based on Gaussian-filter and Multi-scale Convolution Neural Network(MSCNN)is proposed,which can realize the fault diagnosis for the bearings effectively and accurately in noisy environments,and has high robustness in different load conditions.In this method,firstly the optimal filter kernel that adapts to different signal-to-noise ratios is constructed;noise reduction processing is then performed on the noisy signal,and finally the MSCNN is used to adaptively extract the rich multi-scale features of the signal to realize multi-class fault diagnosis.The experimental results show that compared with the current state-of-the-art various fault diagnosis methods,this method has higher fault diagnosis accuracy under various intensity noises,and has lower complexity in time and space dimensions,which can be practically deployed in industrial production.

关 键 词：故障诊断 降噪 深度学习 高斯滤波 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TH133.3[自动化与计算机技术—控制科学与工程]