发动机油中水分的近红外光谱分析  

Near Infrared Spectroscopy Analysis of Moisture in Engine Oil

作　　者：刘阁 陈彬 商芷萱 全宇轩 LIU Ge;CHEN Bin;SHANG Zhi-xuan;QUAN Yu-xuan(Hebei Key Laboratory of Hazardous Chemicals Safety and Control Technology,School of Chemical and Environmental Engineering,North China Institute of Science and Technology,Yanjiao 065201,China;School of Environmental Engineering,Hebei Key Laboratory of Safety Monitoring of Mining Equipment,North China Institute of Science and Technology,Yanjiao 065201,China)

机构地区：[1]河北省危险化学品安全与控制技术重点实验室,华北科技学院化学与环境工程学院,河北燕郊065201 [2]华北科技学院机电工程学院,河北省矿山设备安全监测重点实验室,河北燕郊065201

出　　处：《光谱学与光谱分析》2023年第2期449-454,共6页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(51375516);中央高校基本科研业务费项目(3142019001,3142021009);河北省物联网监控技术创新中心绩效后补助经费项目(21567693H);中央引导地方科技发展资金项目(206Z5201G)资助。

摘　　要：发动机油是发动机的核心部件,发动机油中极易混入水分,水分容易加速发动机油的劣化和变质,进而危害发动机的安全运行。对发动机油中水分进行检测是保障发动机油质量的重要指标。因而采用近红外光谱结合偏最小二乘法(PLS)回归方法对不同含水量的发动机油进行了检测。首先根据含水发动机油的近红外光谱的特征,分析了931,1195~1212和1391~1430 nm波长的较强吸收峰的机制;采用正交信号校正(OSC)和几种其他的光谱预处理方法构建了PLS回归模型,根据回归系数进行了特征波长的选择。结果表明,OSC预处理后的PLS模型具有较好的预测能力,而多元散射校正(MSC)和标准正态变量变换(SNV)预处理降低了模型的校正性能。选择了166个特征波长,占全谱的32.42%。采用所建的近红外全谱PLS模型和特征波长选择的PLS模型分别对预测集14个油样进行预测,两个模型都能实现较好地预测,预测标准差分别为0.0007和0.0006;而特征波长选择对含水发动机油的预测最稳健,性能指标最好(R^(2)_(P)为0.9930,R^(2)_(CV)为0.9887,且RMSE CV和RMSE P值分别为3.1401×10^(-4)和2.4190×10^(-4),RPD值为11.9884),特征波长选择的PLS模型与全光谱模型相比,经过特征波长选择消除了全光谱中大量无用信息,对发动机油中含水量预测最稳健,性能指标最好,使模型的性能得到了显著提高。根据所建的OSC预处理后的全谱PLS模型以及特征波长选择的PLS模型,对油样的预测集进行验证,特征波长选择后的PLS模型对预测集的预测效果较优,每个油样的预测值更接近实测值。说明经过特征波长选择后建立的PLS模型不仅没有降低模型的精度和预测能力,反而由于消除了不相关变量的信息,使所建模型更具有泛化性能。因而近红外光谱技术对发动机油中水分的检测具有较好的精确性、可靠性,为发动机的状态监测提供一种可行的解决方案。Engine oil is the core component of the engine.It is easy to mix with water in the engine oil,which can easily accelerate the deterioration and deterioration of the engine oil,and then harms the safe operation of the engine.Detecting water in the engine oil is an important indicator to ensure the quality of the engine oil.Moisture is easy to accelerate the deterioration and degradation of engine oil,and it is harmful to the safe operation of the engine,and its detection is an important index to ensure the quality of engine oil.Therefore,near-infrared(NIR)spectroscopy combined with partial least squares(PLS)regression method was used to detect engine oil with different water content.Firstly,the mechanism of 931,1195~1212 and 1391~1430 nm wavelengths with strong absorption peaks were analyzed according to the NIR characteristics of water-containing engine oil.Orthogonal signal correction(OSC)and several other spectral pretreatment methods were used to construct the PLS regression model,and the characteristic wavelength was selected according to the regression coefficient.The results showed that the PLS model pretreated by OSC had the better predictive ability,while the pretreated by MSC and SNV reduced the correction ability of the model.The 166 feature wavelengths were selected,accounting for 32.42%of the spectrum.The fourteen oil samples in the prediction set were predicted using the established near-infrared full spectrum PLS model and the characteristic wavelength selected PLS model.Both models can achieve good prediction,and the standard deviation of prediction is 0.0007 and 0.0006,respectively.The PLS model selected by characteristic wavelength had the most robust prediction and the best performance index(R^(2)_(P)was 0.9930,R^(2)_(CV)was 0.9887,RMSE CV and RMSE P were 3.1401×10^(-4)and 2.4190×10^(-4),RPD was 11.9884).Compared with the full-spectrum model,the PLS model with characteristic wavelength selection can eliminate much useless information in the full spectrum,predict the water content of engine oil

关 键 词：发动机油 水分 近红外光谱 特征波长 

分 类 号：O657.3[理学—分析化学]