吕欢1, 许涛1,2, 麻爱松1, 李建平1, 陈玉立1.基于改进DenseNet和迁移学习的变负载滚动轴承故障诊断[J].轻工机械,2023,41(1):53-58 |
基于改进DenseNet和迁移学习的变负载滚动轴承故障诊断 |
Rolling Bearing Fault Diagnosis Method Based on Improved DenseNet and Transfer Learning under Load Changes |
|
DOI:10.3969/j.issn.1005 2895.2023.01.009 |
中文关键词: 滚动轴承 故障诊断 DenseNet 迁移学习 LeakyReLU函数 |
英文关键词:rolling bearing fault diagnosis DenseNet transfer learning LeakyReLU |
基金项目:陕西省自然科学基金资助项目(2019JM 310));广东省精密齿轮柔性制造装备技术企业重点实验室开放基金(中山迈雷特数控技术有限公司&华南理工大学)(2021B1212050012 05);西安市现代智能纺织装备重点实验室资助项目(2019220614SYS021CG043)。 |
|
摘要点击次数: 124 |
全文下载次数: 101 |
中文摘要: |
由于在实际工作环境下滚动轴承故障样本不足,而且受到环境噪声以及负载变化的影响,故障样本分布存在差异性导致诊断泛化性差,对此课题组提出一种基于改进DenseNet与迁移学习结合的滚动轴承故障诊断方法。对原DenseNet中的ReLU激活函数,使用LeakyReLU函数替代,并在全连接层后添加Softmax层进行分类,使提取故障特征更为丰富;为了使轴承信号接近工厂采集的数据,对凯斯西储大学轴承数据集中添加了信噪比为 2 dB的高斯白噪声并进行模拟,经Z Score归一化处理后转化为二维灰度图作为样本数据。实验结果表明该方法在小样本变负载下的跨域诊断准确率都达到了90%以上,与其他模型对比具有更好的泛化性。 |
英文摘要: |
Due to the insufficient fault samples of rolling bearings in the actual working environment, and the influence of environmental noise and load changes, the difference in fault sample distribution leads to poor generalization of diagnosis. A rolling bearing fault diagnosis based on improved DenseNet and transfer learning method was proposed. The ReLU activation function in the original DenseNet was replaced by LeakyReLU, and a Softmax layer was added after the fully connected layer for classification to enrich the fault features extraction. In order to make the bearing signal close to the data collected by the actual factory, a Gaussian white noise with a signal to noise ratio of 2 dB was added to the Case Western Reserve University data set for simulation. After Z Score normalization, it was transformed into a two dimensional grayscale map as the sample data. The experimental results show that the cross domain diagnostic accuracy of the proposed method has reached more than 90% under small samples and load changes, and has better generalization compared with other models. |
查看全文 查看/发表评论 下载PDF阅读器 |
关闭 |