多核的LSSVM的轴承故障诊断算法研究

A Research on the Bearing Fault Diagnosis Algorithm of Least Squares Support Vector Machine with Multiple Kernels

针对滚动轴承故障诊断中出现的多故障类型识别与诊断问题,在LSSVM和多核学习的基础上,提出了多核的LSSVM的轴承故障诊断算法。多核的LSSVM实现的关键是如何确定多核函数的每个基本核函数的权系数,采用核度量标准——核极化来解决此难题。首先,选择基本核函数及其核参数值;然后,用核极化求解基本核函数的权系数,组合多核函数;最后,创建多核的LSSVM算法模型,进行轴承故障诊断。美国西储大学的滚动轴承的实验结果表明,与5-fold SVM和LSSVM相比,多核的LSSVM算法具有更优的故障识别率,验证了所提算法的有效性。

For the problems of multiple fault types identification and diagnosis of rolling bearing,on the basis of LSSVM and multiple kernel learning,the paper proposed a bearing fault diagnosis algorithm of LSSVM with multiple kernels. The key to realize the proposed algorithm is how to construct the kernel function with multiple basic kernels,and the solution is that weights of basic kernels is determined by kernel polarization.Firstly,select the basic kernels and parameter values. Then,determine the kernel weight by using kernel polarization,and construct the multiple kernel. Last,create the LSSVM model with the multiple kernel to carry out fault diagnosis of bearing. The experimental results of rolling bearing from Case Western Reserve University show that,compared with 5-fold SVM and LSSVM,the proposed algorithm can obtain a better fault recognition rate,verifying the effectiveness of the proposed algorithm.