基于机器学习的轴承故障分类方法研究

Research on Bearing Fault Classification Based on Machine Learning

针对传统轴承故障诊断方法需要人为提取故障特征及依赖专家经验的问题,提出一种端到端的轴承故障诊断方法。将归一化的振动信号转换为二维图片,对图片分别进行均值滤波、高斯滤波、中值滤波和双边滤波的降噪预处理对比,最后选定中值滤波降噪方法。在改进LeNet-5模型的基础上,通过超参数设置,对全寿命轴承数据进行分类试验,平均精度达到97%,交叉熵损失达到0.06,分类效果明显。

In order to solve the problem that traditional bearing fault diagnosis method needs man-made fault feature extraction and expert experience, an end-to-end bearing fault diagnosis method is proposed in this paper. In this paper, the normalized vibration signals are converted into two-dimensional images, the experiments of mean filter, gauss filter, median filter and bilateral filter were carried out, and the median filter was selected. On the basis of the improved LeNet-5 model, the classification test of the data faults is carried out by setting the over-parameter, the average accuracy is 97%, and the cross entropy loss is 0.06, which shows an obvious classification effect.