基于改进AlexNet模型的滚动轴承故障诊断方法

Fault diagnosis method for rolling bearings based on improved AlexNet model

针对传统方法难以提取滚动轴承故障特征、诊断精度不佳且模型结构复杂等问题,提出一种改进Alexnet模型的滚动轴承故障诊断方法。首先,将滚动轴承各故障振动信号转化为富含时频信息的二维特征图样本集,按一定比例划分为训练集与验证集;然后针对AlexNet模型中存在的模型训练速度慢、准确性不高等缺点进行改进,使用ImageNet图像数据集对改进模型进行训练,并保存训练过程获取的知识;最后将保存的训练信息迁移应用于改进模型对轴承故障数据集的诊断。通过改进前后模型对部分cifar-10图像数据集的训练与验证情况证明了改进模型的优化效果,对比常见网络模型对轴承10类别故障诊断情况,所提方法具有更好的诊断效率和诊断精度。

A rolling bearing fault diagnosis method based on an improved Alexnet model was proposed to address the issues of difficult feature extraction,poor diagnostic accuracy,and complex model structure in traditional methods.Firstly,the vibration signals of each fault of the rolling bearing were converted into a two-dimensional feature map sample set rich in time-frequency information,and divided into a training set and a validation set in a certain proportion.Then,improvements were made to the slow training speed and low accuracy of the Alexnet model.The ImageNet image dataset was used to train the improved model and save the knowledge obtained during the training process.Finally,the saved training information was transferred and applied to the improved model for diagnosing bearing faults in the dataset.The optimization effect of the improved model was demonstrated through training and validation on some cifar-10 image datasets before and after improvement.Compared with common network models for bearing 10 category fault diagnosis,the proposed method had better diagnostic efficiency and accuracy.