基于变分模态分解和深度多核极限学习机的轴承故障分类  

Bearing fault classification based on variational mode decomposition and deep multicore limit learning machine

作　　者：邵磊[1] 祝晓晨 李季[1] 刘宏利[1] 孙文涛[1] SHAO Lei;ZHU Xiaochen;LI Ji;LIU Hongli;SUN Wentao(Tianjin Key Laboratory for Advanced Mechatronic System Design and Intelligent Control,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学天津市先进机电系统设计与智能控制重点实验室,天津300384

出　　处：《天津理工大学学报》2024年第5期32-39,共8页Journal of Tianjin University of Technology

基　　金：国家自然科学基金(21106107)。

摘　　要：针对轴承故障分类任务中核极限学习机(kernel extreme learning machine,KELM)超参数选择困难、模型运算速度慢的问题,提出一种基于深度混合核极限学习机(deep hybrid kernel extreme learning machine,DHKELM)的轴承故障分类方法,利用天鹰优化算法(aquila optimization algorithm,AO)实现该模型超参数的优化选择。首先,以峰度指数作为鲸鱼优化算法(whale optimization algorithm,WOA)的适应度函数,对变分模态分解(variational mode decomposition,VMD)的相关参数寻优,利用最优参数组合进行VMD分解,得到k个模态分量并求其希尔伯特-黄变换(Hilbert-Huang Transform,HHT)边际谱作为特征数据,将其作为天鹰优化DHKELM分类器的输入,对不同状态的轴承故障进行识别。实验结果表明,KELM,DHKELM,天鹰优化DHKELM三种分类模型故障识别准确率分别为94%,96.67%,98.34%,运算时间分别为0.0631,0.0360,0.0175 s,证明AO-DHKELM识别准确率和运算速度均具有明显优势。Aiming at the problem that the kernel extreme learning machine(KELM)hyperparameter selection is difficult and the model operation speed is slow in the bearing fault classification task,a bearing fault classification method based on deep hybrid kernel extreme learning machine(DHKELM)is proposed,and the optimization selection of the hyperparameters of the model is realized by using the aquila optimization algorithm.Firstly,using the kurtosis index as the adaptability function of the whale optimization algorithm(WOA),the relevant parameters of variational mode decomposition(VMD)are optimized,and the optimal combination of parameters is used to decompose VMD,and k modal components are obtained and the Hilbert-Huang Transform(HHT)marginal spectrum is obtained as the characteristic data,which is used as the input of the aquila optimized DHKELM classifier to identify bearing faults in different states.Experimental results show that the fault recognition accuracy of KELM,DHKELM and aquila optimized DHKELM is 94%,96.67%and 98.34%,and the operation time is 0.0631 s,0.0360 s and 0.0175 s,respectively,which proves that the recognition accuracy and operation speed of AO-DHKELM have obvious advantages.

关 键 词：滚动轴承 深度混合核极限学习机 天鹰优化算法 变分模态分解 边际谱 

分 类 号：TH122[机械工程—机械设计及理论]