摘要
针对传统轴承故障诊断模型学习关键故障特征能力不足,且在噪声干扰下诊断精度受限的问题,提出了一种多分支多尺度卷积神经网络结合通道注意力(MBSACNN)的故障诊断方法。该方法采用多通道多输入的方式弥补传统模型只能分析单一维度故障特征信息的不足;进行连续小波变换将样本转化为时频信号,增强样本信息的多样性;利用多尺度并行卷积获取关键特征,增强特征学习能力;结合通道注意力机制有效融合多分支故障特征,提升故障诊断的准确性。与传统故障诊断模型相比,MBSACNN模型在特征学习和抗噪性能方面都表现出一定的优势。在凯斯西储大学(CWRU)实验数据集零噪声与强噪声情况下,故障分类准确率分别为99.99%和96.97%;工程应用中,在噪声干扰强烈的3类水泥生产设备上故障分类准确率均优于97.25%,具有较高的诊断精度与噪声鲁棒性。
A multi-branch multi-scale convolutional neural network with channel attention(MBSACNN)method is proposed to enhance feature extraction and improve accuracy with noises in bearing fault diagnosis.Different from the traditional methods,where only one dimension fault feature was considered,multi-channel multi-input is constructed to extract multi-dimension abundant features and enhance the diversity of sample information from the wavelet transform timefrequency signal,which are combined by the channel attention mechanism.Higher diagnosis accuracy and better noiserobustness are obtained by the MBSACNN compared with the traditional methods,which is verified by Case Western Reserve University(CWRU)bearing dataset and real-world application in cement industry.In the case of a bearing dataset,the accuracy of zero noise and strong noise is 99.99%and 96.97%respectively.Under strong noise,the accuracy of three kinds of cement equipment are all above 97.25%.
作者
郭海宇
于威
张晓光
陆凡凡
陈洋
赵学义
GUO Haiyu;YU Wei;ZHANG Xiaoguang;LU Fanfan;CHEN Yang;ZHAO Xueyi(School of Electrical Engineering,Shenyang University of Technology,Shenyang,Liaoning 110870,China;Shanghai Intelligent Quality Technology Co.Ltd,Shanghai 201801,China;School of Computer Science and Technology,University of Science and Technology of China,Hefei,Anhui 230026,China;Yangtze Delta Information Intelligence Innovation Research Institute,Wuhu,Anhui 241000,China;Anhui Conch Cement Co.Ltd,Wuhu,Anhui 241200,China)
出处
《计量学报》
北大核心
2025年第2期222-232,共11页
Acta Metrologica Sinica
基金
国网辽宁省电力有限公司科技项目(2023YF-21)
中国博士后科学基金(2024M753116)。
关键词
振动计量
轴承故障诊断
卷积神经网络
多分支
通道注意力机制
水泥生产设备
vibration measurement
bearing fault diagnosis
convolutional neural networks
multi-branch
channel attention mechanism
cement equipment
