基于MGCD的转子故障数据集降维方法  

Dimension Reduction Method of Rotor Fault Dataset Based on MGCD

作　　者：张勇飞 赵荣珍[1] 邓林峰[1] ZHANG Yongfei;ZHAO Rongzhen;DENG Linfeng(School of Mechanical and Electrical Engineering,Lanzhou University of Technology Lanzhou,730050,China)

机构地区：[1]兰州理工大学机电工程学院,兰州730050

出　　处：《振动．测试与诊断》2024年第2期266-273,408,共9页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金资助项目(62241308,51675253)。

摘　　要：针对由于特征维数过高导致故障数据集分类困难及故障模式辨识精度偏低的问题,提出一种基于多图协同决策(multi graph collaborative decision-making,简称MGCD)的转子故障数据集降维算法。首先,在边缘Fisher分析(marginal Fisher analysis,简称MFA)算法框架基础上,通过建立近邻图和远邻图解决因单一图结构导致的故障类别局部不可分问题;其次,采用最大化散度加权差分方式去削弱小样本问题造成的影响;最后,利用两个不同结构型式的转子系统故障模拟数据集对算法性能进行了验证。结果表明,使用本算法对故障数据集进行降维得到的敏感故障数据集使故障类别之间的差异性更加突出,能够提高故障模式识别准确率,为提高旋转机械智能故障诊断技术水平提供一定的研究参考依据。A rotor fault dataset dimensionality reduction algorithm based on multi graph collaborative decision making(MGCD)is proposed,in order to address the issues of difficulty in classifying fault datasets and low accuracy in fault pattern recognition due to high feature dimensions.This algorithm first builds on the framework of marginal Fisher analysis(MFA)algorithm to solve the problem of local inseparability of fault categories caused by a single graph structure,through establishing nearest neighbor graphs and far neighbor graphs.Secondly,it uses the maximum divergence weighted difference method to try to weaken the impact of small sample problems.The performance of the algorithm is verified using two different structural types of rotor system fault simulation datasets.The results show that the sensitive fault dataset obtained by using this algorithm to reduce the dimensionality of the fault dataset,can make the differences between fault categories more prominent,thereby improving the accuracy of fault pattern recognition.This study can provide a certain research reference for improving the level of intelligent fault diagnosis technology in rotating machinery.

关 键 词：故障诊断 降维 远邻图 小样本 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TH165[自动化与计算机技术—控制科学与工程]