摘要
为解决匀速工况下印刷机轴承振动信号噪声大、故障特征难以捕捉的问题,提出了一种基于CEEMDAN-MCNN-SVM的诊断方法。首先,利用CEEMDAN对原始振动信号进行分解与重构,有效抑制噪声并凸显故障特征;随后,对重构信号将小波变换为时频图像,并结合时域特征构建多维特征集;然后,将图像与时域特征输入多通道卷积神经网络(MCNN)进行深层特征提取,采用交叉熵损失函数训练网络;最后,将提取的特征送入SVM分类器,并通过t-SNE可视化验证特征可分性。实验在CWRU数据集及实验台采集数据集进行测试,分类准确率分别达到99.44%和99.80%,结果表明该方法在匀速条件下实现了高精度、强鲁棒性的轴承故障诊断。
To address the challenges of high noise and difficult fault feature extraction in bearing vibration signals under uniform-speed conditions in printing machines,a CEEMDAN-MCNN-SVM diagnostic method is proposed.First,the original vibration signals are decomposed and reconstructed using CEEMDAN,effectively suppressing noise and highlighting fault features.Then,the reconstructed signals are transformed into time-frequency images via wavelet transform and combined with time-domain features to construct a multi-dimensional feature set.Next,the images and timedomain features are input into a multi-channel convolutional neural network(MCNN)for deep feature extraction,trained using a cross-entropy loss function.Finally,the extracted features are fed into an SVM classifier,and t-SNE visualization is employed to verify feature separability.Experiments conducted on both the CWRU dataset and the laboratory-acquired dataset achieved classification accuracies of 99.44%and 99.80%,respectively,demonstrating that the proposed method can achieve high-precision and robust bearing fault diagnosis under uniform-speed conditions.
作者
于明洋
李婷
马添翼
温杰
YU Mingyang;LI Ting;MA Tianyi;WEN Jie(School of Mechanical and Electrical Engineering,Beijing Institute of Graphic Communication,Beijing 102600,China)
出处
《北京印刷学院学报》
2026年第3期61-68,共8页
Journal of Beijing Institute of Graphic Communication
基金
中国国家自然科学基金(62403065)
北京市教委科研计划(KM202310015003)
北京印刷学院青年英才计划项目(Ea202405)
北京市属高等学校高水平科研创新团队建设支持计划项目(BPHR20220107)研究成果。
关键词
轴承故障诊断
模态分解
小波变换
卷积神经网络
支持向量机
bearing fault diagnosis
modal decomposition
wavelet transform
convolutional neural network
support vector machine
