基于改进OTSU-CNN的轴承智能故障诊断  

Intelligent Fault Diagnosis of Bearing Based on Improved OTSU-CNN

作　　者：张伟[1] 鲍泽富[1] 李寿香 徐浩 张迪 Zhang Wei;Bao Zefu;Li Shouxiang;Xu Hao;Zhang Di(School of Mechanical Engineering,Xi'an Shiyou University,Xi'an 710065,China)

机构地区：[1]西安石油大学机械工程学院,西安710065

出　　处：《机电工程技术》2023年第3期222-227,共6页Mechanical & Electrical Engineering Technology

基　　金：中石油科技中青年创新基金项目(编号:05E7040)。

摘　　要：针对传统故障诊断方法在小样本数据集下诊断准确率低且故障特征提取难的问题,提出了一种改进大津阈值分割算法(OTSU)和卷积神经网络(CNN)相结合的智能故障诊断方法。首先,对采集到的振动信号进行希尔伯特变换(Hillbert)得到信号的包络谱,同时使用小波变换对包络谱信号处理,获取二维特征时频图;其次,建立最大类间方差目标函数模型,通过算术优化算法(AOA)得到时频图的最佳分割阈值,再将变换后的阈值分割图像作为CNN的输入得到最优训练模型,最后得到分类结果。试验结果表明:所提方法相比于传统OTSU方法,所提取的故障特征更为突出,为模型提供了优秀的训练样本;在较小数据样本条件下,所提方法的准确率达99.01%,远高于传统故障诊断方法,且模型有着良好的泛化能力。Aiming at the problem that the traditional fault diagnosis method has low diagnosis accuracy and difficult fault feature extraction under small sample datasets,an intelligent fault diagnosis method combining the Otsu threshold segmentation algorithm(OTSU)and the Convolutional Neural Network(CNN)was proposed.Firstly,the Hillbert transform was performed on the collected vibration signal to obtain the envelope spectrum of the signal,and the wavelet transform was used to process the envelope spectrum signal to obtain a two-dimensional feature time-frequency map.Secondly,the maximum between-class variance objective function model was established,the optimal segmentation threshold of the time-frequency graph was obtained by the arithmetic optimization algorithm(AOA),and then the transformed threshold segmentation image was used as the input of CNN to obtain the optimal training model,and finally the classification result was obtained.The experimental results show that compared with the traditional OTSU method,the extracted fault features of the proposed method are more prominent,which provides excellent training samples for the model.Under the condition of small data sample,the accuracy of the proposed method is 99.01%,which is much higher than that of traditional fault diagnosis methods,and the model has good generalization ability.

关 键 词：OTSU 故障特征提取 卷积神经网络 时频图 

分 类 号：TH133.33[机械工程—机械制造及自动化]