摘要
针对最小二乘支持向量机(LSSVM)实现过程中盲目选择核函数的现象,提出了一种基于核极化的多核LSSVM与EMD相结合的滚动轴承故障识别算法。首先,对滚动轴承振动信号进行EMD信号提取,进而提取故障特征向量;然后,根据多核构造原理,引入核极化确定基本核函数的组合权系数,构造多核函数;最后,结合多核函数与LSSVM,形成多核LSSVM学习器,进行故障识别。分析滚动轴承正常状态、内圈故障、外圈故障和滚动体故障的诊断实验结果,可知,EMD与多核LSSVM的故障识别算法可以准确地判断滚动轴承的工作状态和故障类型,并与SVM、LSSVM算法的诊断结果进行对照,表明所提算法的故障识别率更高。
For the phenomenon of selecting kernel function blindly for least squares support vector machine （LSSVM）, a fault diagnosis algorithm for rolling bearing based on empirical mode decomposition （EMD） and LSSVM with multiple kernels was propased. First, select the main fault feature vectors by using EMD method to decompose the vibrational signal of rolling bearing. Then, introduce kernel polarization to determine the weights of basic kernels, according to the principle of multiple kernel learning, and then construct the multiple kernel function, which was combined with LSSVM classifier to make a new classifier, namely LSSVM with multiple kernels （MK_LSSVM）. Last, apply the selected fault features served as input parameters of MK LSSVM to identify the fault diagnosis of rolling bearing. The experimental results consist of fault-free, inner-race fault, outer-race fault and rolling-ball fault of rolling bearing. We analyze the experimental results, compare with SVM and LSSVM, and we can know that the working status and fault types of roUing bears can be identified accurately and effectively by the proposed algorithm based on EMD and MK_LSSVM.
出处
《机械设计与制造》
北大核心
2018年第2期249-252,共4页
Machinery Design & Manufacture
基金
国家自然科学基金项目(21366017)
内蒙古科技大学科研启动项目(2014QDL024)
内蒙古自然科学基金(2016MS0543)
