基于1D CNN-XGBoost的滚动轴承故障诊断

Fault Diagnosis of Rolling Bearing Based on 1D CNN-XGBoost

在滚动轴承故障自动分类的研究中,使用传统的机器学习方法需要通过手动提取特征,因此特征的提取并不充分且自适应性不强。针对以上问题,提出一种一维卷积神经网络(1D CNN)结合XGBoost算法的单通道滚动轴承故障分类模型。该模型结合1D CNN和XGBoost的优势,对采集到的轴承振动信号进行数据集划分;使用训练集对1D CNN进行训练,把训练好的1D CNN模型进行保存并用来实现轴承数据特征的自动提取;将提取的特征数据集代入XGBoost算法中进行训练和分类。为验证所提模型的有效性,使用凯斯西储大学轴承数据中心提供的数据对1D CNN模型、XGBoost模型和1D CNN-XGBoost模型进行实验对比;为验证1D CNN-XGBoost的泛化性,使用一组新的滚动轴承数据集进行实验。结果表明:1D CNN-XGBoost模型的分类准确率更高,是一种有效的轴承故障分类模型,具有很好地分类性能和泛化性。

In the research of automatic classification of rolling bearing faults,the use of traditional machine learning methods requires manual extraction of features,so the extracted features are not sufficient and the adaptability is not strong.In view of the above problems,a single-channel rolling bearing fault classification model combined one dimensional convolutional neural network(1 D CNN) with the XGBoost algorithm was proposed.This model combined the advantages of 1 D CNN and XGBoost,the collected bearing vibration signals were divided into data sets;the training set was used to train the 1 D CNN,and the 1 D CNN model was saved and used to realize the automatic extraction of bearing data features;the extracted feature data set was substituted into the XGBoost algorithm for training and classification.In order to verify the effectiveness of this model,the data provided by Case Western Reserve University Bearing Data Center was used to compare the 1 D CNN model,XGBoost model and 1 D CNN-XGBoost model;to verify the generalization of 1 D CNN-XGBoost,an another rolling bearing data set was used in the experiment.The results show that the classification accuracy of the 1 D CNN-XGBoost model is higher and it is an effective bearing fault classification model with good classification performance and generalization.