基于局部主成分保持投影的旋转机械故障数据降维方法  

Dimension reduction method for rotating machinery fault data based on local principal component preserving projection

作　　者：原健辉 赵荣珍[1] 马驰 YUAN Jianhui;ZHAO Rongzhen;MA Chi(School of Mechanical&Electronic 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：《振动与冲击》2023年第6期24-30,共7页Journal of Vibration and Shock

基　　金：国家自然科学基金(51675253)。

摘　　要：针对旋转机械高维故障特征集存在的特征冗余导致的分类困难问题,提出一种基于局部主成分保持投影(locality principal component preserving projection,LPCPP)的故障数据集降维算法。该算法将类间可分性判据、主成分计算两种思想与局部保持投影(locality preserving projection,LPP)相融合,使得算法具有剔除冗余特征、减小降维盲目性的能力,从而可以更好地保留能够反映机械运行状态的高价值密度的故障信息以及特征的主要成分。通过两个不同型号的双跨度转子系统的振动信号对所提算法进行验证,并分别以可分性指标和3种不同分类器的识别准确率对本算法的降维性能进行评价。结果表明,本算法能够达到降低故障分类难度与提高故障分类准确率的功能,其可为积累高价值密度的数据资源和基于“工业大数据”的旋转机械智能决策技术工程实现,提供一种数据运算的理论依据。Aiming at the classification difficulty caused by feature redundancy in high-dimensional fault feature sets of rotating machinery,a dimension reduction algorithm of fault data sets based on local principal component preserving projection(LPCPP)was proposed.In the algorithm,the two ideas of inter-class separability criterion and principal component calculation were integrated with locality preserving projection(LPP),making the algorithm have the ability to eliminate redundant features and reduce dimensionality reduction blindness,so,the fault information with high value density and the main components of features being able to reflect the operation status of machinery could be better retained.The proposed algorithm was verified by using the vibration signals of two different types of two span rotor systems,and the dimensionality reduction performance of the algorithm was evaluated by the separability index and the recognition accuracy of three different classifiers.The results show that the algorithm can achieve the functions of reducing the difficulty of fault classification and improving the accuracy of fault classification,which can provide a theoretical basis for data operation to accumulate the data resources with high value density and for the engineering implementation of intelligent decision-making technology for rotating machinery based on“industrial big data”.

关 键 词：故障诊断 局部保持投影(LPP) 可分性 主成分计算 旋转机械 

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]