原子发射光谱是分析油液中微小磨损颗粒元素浓度的重要方法.以综合传动全寿命磨损试验不同阶段采集的多个油液样本为研究对象,分别运用基于模糊隶属度的稳健核主成分分析(RKPCA)与传统主成分分析(PCA)对光谱数据进行主成分提取与对比.在剔除光谱数据中的干扰元素后,计算与比较两种方法的主成分数量与贡献率,并利用RKPCA主成分进行综合传动多摩擦副的分类识别;对光谱数据和RKPCA特征值分别进行模糊C均值聚类,对比两种聚类结果应用在磨损状态评价中的效果.研究表明,由于光谱数据离群值与非线性影响,RKPCA较PCA的主成分数量稍小且累积贡献率高,说明前者能更有效地降低变量维数;通过RKPCA主成分与摩擦副组件的相关性分析可以看出,该方法可以精确的实现综合传动多摩擦副、多磨损部位的分类与识别,进而分类评价不同摩擦副的磨损状态;RKPCA特征值的模糊C均值聚类结果与光谱数据直接聚类结果相比,前者能更精确的定位磨损状态转化的临界点,从而准确评价综合传动整体磨损状态.油液光谱RKPCA分析方法的创新在于将特征值变化规律引入整体磨损状态评价,实现整体评价与关键摩擦副的分类评价相结合.这样不仅有助于综合传动大修期的准确判断,还能给出需维修部件建议.该方法也适用于其他复杂机械系统的磨损监测与评价等相关领域.

The most common methodology used in element concentration measurement and analyzing of wear particles is Atomic emission(AE) spectroscopy.The present paper presents an evaluation method on wear in power-shift steering transmission(PSST).By removing the problematic components which were highly correlated with oil additives,the robust kernel principal component analysis(RKPCA) method and the principal component analysis(PCA) method were accessed to extract the principal components of spectral data for oil samples collected from the life-cycle test of PSST in different stage and to calculate the amount of each principal component and its contribution rate respectively.A comparison between the above mentioned two methods was made to show that RKPCA method has fewer amounts of principal components and higher cumulative contribution rate indicating that RKPCA method acts more effectively in variable dimension reduction due to the outliers and nonlinearity of spectral data.Therefore,the effectiveness of RKPCA method in classification and identification of the wear in friction pairs was demonstrated subsequently through the correlation analysis between the variable coefficients of RKPCA and metal elements of friction pairs.The demonstration showed that RKPCA functioned precisely in the classification and identification of the wear in friction pairs,and in the evaluation on the wear in PSST.Thereafter,to detect the threshold point where the wear took place,the fuzzy C-means clustering algorithm was introduced to classify the RKPCA eigenvalues,and the results were compared with that of the spectral clustering algorithm.The fuzzy C-means clustering algorithm showed higher sensitivity in detecting the threshold point indicting a more precise evaluation on the wear in PSST.It is clear that the introduction of RKPCA method in wear evaluation,which takes the eigenvalues of spectral data as a critical variable to classify and identify the wear in different friction pairs as well as in the integral PSST configuration,shows better accuracy in wear prediction and will contribute to the reliable determination of life between overhauls and the accurate positioning of worn-out parts.As might be expected,the proposed method can be extended to other cases of wear detection and evaluation in complex mechanical system.