A Hybrid 1DCNN-SVM Model for Bearing Fault Recognition and Classification
A Hybrid 1DCNN-SVM Model for Bearing Fault Recognition and Classification
复制标题
DOI：
10.1109/icac55051.2022.9911115
复制
发表时间：
2022-09
期刊：
影响因子：
--
通讯作者：
Guoliang Wang;Hua‐Liang Wei
中科院分区：
文献类型：
--
作者：
Guoliang Wang;Hua‐Liang Wei
滚动轴承在旋转机械的运行中起着重要的作用。早期发现轴承故障，不仅可以保证设备的安全运行，而且可以避免经济损失。传统的机器学习（ML）和深度学习（DL）是两类重要的故障诊断方法。特征提取过程对于基于最大似然的方法来说是非常重要的，许多信号处理技术都被应用到了这个过程中。然而，在处理海量复杂信号时，这些技术提取的特征可能不能完全描述故障的特征。卷积神经网络（CNN）模型在图像识别和分类、模式识别、语音处理等领域有着广泛的应用。本文提出了一种将一维CNN（1DCNN）和支持向量机（SVM）相结合的混合模型用于轴承故障诊断和分类，称为1DCNN-SVM模型或方法。模型的第一部分，1DCNN，用于提取原始数据的特征，而第二部分，SVM，用于执行故障识别和分类任务。所提出的方法的评估是基于一个真实的数据集。实验结果表明，与传统的基于SVM的方法相比，基于混合模型的方法的故障诊断精度有了很大的提高。
Rolling bearings play a significant role in the operation of rotating machinery. Finding bearing faults in the early stage can not only keep machinery running safely but also avoid economic loss. Traditional machine learning (ML) and deep learning (DL) are two important classes of fault diagnosis methods. The process of feature extraction is significant for ML based methods and many signal processing techniques have been used in this process. However, when dealing with massive and complex signals, features extracted by these techniques may not fully describe the characteristics of faults. Convolutional neural network (CNN) models are applied in many areas, including image identification and classification, pattern recognition, and speech processing. In this paper, a novel hybrid model by combining one dimensional CNN (1DCNN) and support vector machine (SVM) is proposed for bearing fault diagnosis and classification, which is called 1DCNN-SVM model or method. The first part of the model, 1DCNN, is used to extract features of raw data, whereas the second part, SVM, is employed to carry out the fault recognition and classification tasks. The evaluation of the proposed method is operated based on a real dataset. The experiment results show that the fault diagnosis accuracy of proposed method based on the hybrid model is highly improved compared to the traditional SVM based methods.