摘要
采用传统矩阵分类器(即支持矩阵机SMM)进行滚动轴承故障诊断时存在一定的局限性,即在进行冗余特征分类时难以提取有效特征进行建模,为此,提出了一种基于自适应冗余矩阵分类器(ARMC)的滚动轴承故障诊断方法。首先,在构造ARMC模型的过程中,通过核函数创建了高维分布空间,解决了样本数据线性不可分的问题;然后,采用约束L 1范数的思想,使得样本到所有聚类凸包边界的距离最短,进而将其转化为求解线性规划的问题,降低了模型计算的复杂度;通过正则化约束来控制低秩项,进而弱化冗余特征和噪声成分对模型的影响,得到了更加准确的预测模型;最后,为了验证ARMC方法的有效性,采用美国凯斯西储大学(CWRU)的滚动轴承实验数据和自制滚动轴承故障模拟实验台数据,分别进行了实验;并且将采用该方法所获得的结果与其他方法获得的结果进行了对比。研究结果表明:ARMC利用L 1范数和核函数来构造和求解目标函数,不仅可以保护待诊断对象的结构化信息,而且可以弱化模型复杂度和增强模型的鲁棒性;与支持矩阵机(SMM)和鲁棒支持矩阵机等相比,ARMC能够充分考虑样本冗余信息弱化的问题,平均识别准确率提高3%~8%。
There are some limitations when using traditional matrix classifier(support matrix machine,SMM)to diagnose rolling bearing faults,that is,it is difficult to extract effective features for modeling when classifying redundant features.Therefore,a rolling bearing fault diagnosis method based on adaptive redundancy matrix classifier(ARMC)was proposed.Firstly,in the process of constructing ARMC model,the kernel function was used to create a high-dimensional distribution space to solve the problem of linear indivisibility of sample data.Then,the idea of constrained L 1 norm was used to minimize the distance from the sample to all clustering convex hull boundaries,and it was transformed into a problem of solving linear programming,which reduced the model computational complexity.The low rank term was controlled by regularization constraint,and the influence of redundant features and noise components on the model was weakened to obtain a more accurate prediction model.Finally,in order to verify the effectiveness of the ARMC method,the rolling bearing experiment data of Case Western Reserve University(CWRU)and the data of the self-made rolling bearing fault simulation experiment platform were used to conduct experiments respectively.The results obtained by the this method were compared with those obtained by the other methods.The results show that L 1 norm and kernel function were used to construct and solve the objective function in the ARMC,which can not only protect the structural information of the object to be diagnosed,but also weaken the complexity of the model and enhance the robustness of the model.Comparing with support matrix machine(SMM)and robust support matrix machine,ARMC can fully consider the problem of weakening the redundant information of samples,and the average recognition rate is improved by 3%~8%.
作者
刘振华
胡思远
赵捷
LIU Zhen-hua;HU Si-yuan;ZHAO Jie(Department of Mechanical and Electrical Engineering,Xinzhou Vocational and Technical College,Xinzhou 034000,China;School of Mechanical Engineering,Shandong University of Technology,Zibo 255000,China;Shandong Yaohua Tenai Technology Co.,Ltd.,Binzhou 256619,China)
出处
《机电工程》
CAS
北大核心
2023年第3期384-390,共7页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(52075306)。
关键词
自适应冗余矩阵分类器
矩阵分类器(支持矩阵机)
高维分布空间
冗余特征分类
模型鲁棒性
目标函数
adaptive redundant matrix classifier(ARMC)
support matrix machine(SMM)
high-dimensional distribution space
redundant feature classification
model robustness
kernel function
