基于随机森林的高强化柴油机关键摩擦副磨损诊断方法

刘杰¹，冯海波¹，刘峰春¹，谢俊¹，陈创¹，董红霞²，李闯¹，毛玉欣¹

1.中国北方发动机研究所，天津  300406；2.山西柴油机工业有限责任公司，山西 大同  037000

摘要：为明确柴油机润滑油中包含的摩擦副磨损信息，对某强化柴油机耐久试验过程中的170组润滑油液进行光谱分析，运用非线性拟合方法分析油液中Fe、Cu、Al元素的质量分数与关键摩擦副磨损的相关性；采用Python编程软件构建基于Fe元素的随机森林预测模型，并结合受试者工作特征（receiver operating characteristic，ROC）曲线评价预测模型的准确性。结果表明：与关键摩擦副磨损关联的主要元素Fe、Cu、Al的质量分数显著集中于特定区间，Fe-Al元素的质量分数可用来关联活塞-活塞环-缸套摩擦副磨损，Fe-Cu元素的质量分数可用来关联曲轴-曲轴瓦摩擦副磨损；根据Fe、Cu、Al元素的质量分数设置磨损正常区间和磨损警觉区间，构建的Fe元素随机森林预测模型的准确率为88.24%；与关键摩擦副磨损关联的3种主要金属元素Fe、Cu、Al的质量分数存在强非线性相关关系，磨损警觉区间可作为诊断高强化柴油机活塞-活塞环-缸套和曲轴-曲轴瓦这两类关键摩擦副异常磨损的补充依据。

关键词：摩擦副；光谱分析；磨损区间；磨损警觉；非线性相关：ROC

Wear diagnosis method for key friction pairs ofhigh-strengthened diesel engines based on random forest method

LIU Jie¹, FENG Haibo¹, LIU Fengchun¹, XIE Jun¹, CHEN Chuang¹, DONG Hongxia², LI Chuang¹, MAO Yuxin¹

1.China North Engine Research Institute, Tianjin 300406, China;

2.Shanxi Diesel Engine Industry Co., Ltd., Datong 037000, China

Abstract: To clarify the friction pair wear information contained in diesel engine lubricating oil, spectral analysis is performed on 170 groups of lubricating oil samples collected during the durability test of a high-strengthened diesel engine. The nonlinear fitting method is adopted to analyze the correlation between the mass fractions of Fe, Cu, and Al elements in the oil and the wear of key friction pairs. A Python programming software is used to construct a random forest prediction model based on Fe element, and the receiver operating characteristic (ROC) curve is combined to evaluate the accuracy of the model. The results show that the mass fractions of Fe, Cu, and Al elements, which are mainly correlated with the wear of key friction pairs, are significantly concentrated in specific intervals. The mass fractions of Fe-Al elements can be used to correlate the wear of piston-piston ring-cylinder liner friction pairs, while the mass fractions of Fe-Cu elements can be used to correlate the wear of crankshaft-crankshaft bearing friction pairs. The normal wear interval and wear alert interval are set according to the mass fractions of Fe, Cu, and Al elements, and the accuracy of the constructed Fe-element random forest prediction model reaches 88.24%. There is a strong nonlinear correlation among the three main metal elements of Fe, Cu, and Al, and the wear alert interval can be used as a supplementary basis for diagnosing abnormal wear of piston-piston ring-cylinder liner and crankshaft-crankshaft bearing friction pairs in high-strengthened diesel engines.

Keywords: friction pairs; spectral analysis; wear interval; wear alert; nonlinear correlation; ROC