基于近红外光谱技术的润滑油磨粒含量检测研究  

Study on the Detection of Wear Particle Content in Lubricating Oil Based on Near-Infrared Spectroscopy Technology

作　　者：殷雄 崔洪帅 刘雪婧 马世一 周延[1] 种道彤[1] 熊兵 李锟 YIN Xiong;CUI Hong-shuai;LIU Xue-jing;MA Shi-yi;ZHOU Yan;CHONG Dao-tong;XIONG Bing;LI Kun(School of Energy and Power Engineering,Xi'an Jiaotong University,Xi'an 710049,China;AECC Sichuan Gas Turbine Establishment,Chengdu 610500,China)

机构地区：[1]西安交通大学能源与动力工程学院,陕西西安710049 [2]中国航发四川燃气涡轮研究院,四川成都610500

出　　处：《光谱学与光谱分析》2025年第3期816-825,共10页Spectroscopy and Spectral Analysis

基　　金：国家重点研发计划项目(2021YFE0112800);陕西省企业院所联合重点专项(2023-LL-QY-29)资助。

摘　　要：发动机内润滑油中的金属磨粒含量检测对于预防发动机磨损至关重要,准确、快速地检测出滑油中磨粒含量能及时判断机械设备的磨损状况。为了快速高效地检测出润滑油磨粒形成的固液两相流中的磨粒含量,提出了一种近红外光谱技术结合数学建模算法预测磨粒含量的方法。通过搭建的近红外吸收光谱实验系统,在Fe和Cu两种金属磨粒、5种不同粒径大小下共10组工况下,利用波长检测范围在900~2500 nm的近红外光谱仪,采集磨粒浓度在6~15μg·mL^(-1)范围内的光谱数据。针对单波长上光谱信息无法良好解释滑油内磨粒浓度变化的问题,采用光谱-理化值共生距离(SPXY)算法将光谱数据集进行划分。建立润滑油磨粒含量预测偏最小二乘(PLS)模型,分析各工况下的模型预测结果,得到各工况下的磨粒均可被有效检测,模型决定系数(R^(2))最高为0.8318。针对仅采用PLS建模预测效果不完全理想的问题,采用多种数据预处理方法对原始光谱数据进行数据矫正后建模,结果表明除个别异常工况外,其他工况的模型决定系数R^(2)均大于0.8,优化了PLS模型预测效果。为进一步优化润滑油磨粒模型预测效果,建立了润滑油磨粒遗传规划(GP)模型和润滑油磨粒遗传规划-偏最小二乘(GP-PLS)模型,其中,GP模型相比PLS优化模型更加稳健且预测效果更好,R^(2)最高可达0.9562,平均引用误差(MFE)最大为14.73%;GP-PLS模型,相对于GP模型,R^(2)最高为0.9430,MFE最大为10.86%,MFE得到有效降低,使模型预测准确度更高。通过对磨粒含量预测模型的研究分析,得出几种模型均能有效预测滑油中的磨粒含量变化,其中,GP-PLS模型在预测磨粒含量变化方面整体表现的更好。研究结果表明,采用光谱分析法结合建模算法来预测滑油磨粒固液两相流中的磨粒含量是具备可行性的,为发动机内设备机械磨损故障检测提供了一种有效的检测The detection of wear particle content in engine lubricating oil is crucial for preventing engine wear.Accurately and rapidly detecting the wear particle content in lubricating oil can timely assess the wear condition of mechanical equipment.To rapidly and efficiently detect the wear particle content in the solid-liquid two-phase flow formed by lubricating oil wear particles,a method combining near-infrared spectroscopy with mathematical modeling algorithms for predicting the wear particle content is proposed.Through the near-infrared absorption spectroscopy experimental system built,the spectral data of wear particle concentration in the range of 6~15μg·mL^(-1) were collected by using a near-infrared spectrometer with a wavelength detection range of 900~2500 nm under a total of 10 groups of working conditions under two kinds of metal wear particles,Fe and Cu,and five different particle sizes.To address the issue that spectral information at single-wavelength points cannot adequately explain the changes in wear particle concentration within the lubricating oil,the sample set partitioning based on joint x-y distances(SPXY)algorithm was employed to segment the spectral dataset.The partial least squares(PLS)model for predicting the wear particle content of lubricating oil was established,and the model prediction results under each working condition were analyzed.The results showed that wear particles could be effectively detected under each working condition,the highest coefficient of determination(R^(2))for the model was 0.8318.Various data preprocessing methods were employed to correct the raw spectral data before modeling to address the issue of less-than-ideal prediction performance when using PLS modeling alone.The results showed that,except for a few abnormal conditions,the coefficient of determination R^(2) for the models under other conditions was greater than 0.8,optimizing the predictive performance of the PLS model.To further optimize the prediction effect of the lubricating oil wear particle model,the

关 键 词：近红外光谱 磨粒检测 偏最小二乘 遗传规划 

分 类 号：TK314[动力工程及工程热物理—热能工程]