基于ResNet18和迁移学习的滚动轴承故障诊断
Rolling Bearing Fault Diagnosis Based on ResNet18 and Transfer Learning
张炎亮 1张伊童 1齐聪1
作者信息
摘要
针对基于深度学习的滚动轴承故障诊断网络层数过深模型易退化、难以收集大量故障样本的问题,提出一种基于ResNet18 和迁移学习的滚动轴承故障诊断方法.首先,采用小波变换将原始振动信号转换为二维时频图像,并通过图像增强方法凸显图像包含的时频信息;其次,通过迁移学习,将在ImageNet数据集上预训练的ResNet18 作为初始故障诊断模型;最后,利用轴承数据集微调网络所有参数,生成最终故障诊断模型.利用轴承数据集验证该方法,并与其他方法进行比较.结果表明,在故障标签样本较少的情况下,采用该方法对滚动轴承进行诊断的平均准确率高达98.85%;图像增强和权值微调的方法能有效提高模型的训练速度,提升模型的分类精度.
Abstract
To address the problems of deep learning-based rolling bearing fault diagnosis network with too deep layers and easy degradation of the model and difficulty in collecting a large number of fault samples,a rolling bearing fault diagnosis method based on ResNet18 and transfer learning is proposed.Firstly,the wave-let transform is used to convert the original vibration signal into a two-dimensional time-frequency image,and the time-frequency information contained in the image is highlighted by image enhancement methods;second-ly,the ResNet18 pre-trained on the ImageNet dataset is used as the initial fault diagnosis model through trans-fer learning;finally,the bearing dataset is used to fine-tune all parameters of the network to generate the final fault diagnosis model.The method is validated using the bearing dataset and compared with other methods.The results show that the average accuracy of the method for diagnosing rolling bearings is as high as 98.85%with a small number of fault label samples;the image enhancement and weight fine-tuning methods can ef-fectively improve the training speed of the model and enhance the classification accuracy of the model.
关键词
滚动轴承故障诊断/迁移学习/ResNet18/小波变换Key words
rolling bearing fault diagnosis/transfer learning/ResNet18/wavelet transform引用本文复制引用
基金项目
NSFC联合基金重大项目(U1904210)
河南省高等学校重点科研项目(23A630006)
出版年
2024
NETL
NSTL
万方数据