基于量纲一指标和极限学习机的滚动轴承故障诊断方法

Rolling Bearing Fault Diagnosis Method Based on Dimensionless Parameters and Extreme Learning Machine

时域中的量纲一指标因对故障敏感,被广泛运用于机械故障诊断中,但是目前量纲一指标在诊断过程中存在严重交叉问题,即量纲一指标对于不同故障状态在特征空间中存在混叠现象。为了解决这个问题,提出基于量纲一指标和极限学习机的滚动轴承故障识别方法,采用美国西储大学轴承数据中心网站公开发布的轴承探伤数据集,验证算法诊断效果。为了进一步验证算法的优越性,将该算法与BP神经网络、支持向量机(SVM)和GripsearchSVM3种算法进行比较,结果表明:基于量纲一指标和极限学习机的故障诊断方法能够提高滚动轴承故障诊断效率和分类准确率。

The time-domain dimensionless parameters that are sensitive to the fault models are used widely in mechanical fault diagnosis,but it also has serious crossover problems,that is,the dimensionless parameters have aliasing in the feature space for different fault states.In order to solve this problem,a method based on time-domain dimensionless parameters and Extreme Learning Machine(ELM)is presented,which is used for fault diagnosis of rolling bearings.The accuracy and effectiveness of the method was validated by the actual Rolling experimental data from a public site on the net.The experimental results show that further compared with the Back Propagation(BP),State Vector Machine(SVM)and Grip search SVM for advantages,rolling bearing fault diagnosis method based on time-domain dimensionless parameters and ELM has faster speed and higher classification accuracy.