Identification of adulteration in GTL synthetic lubricant via DD-SIMCA and C-H stretching Raman spectra

作     者：Yu, Yingtao Li, Jinlin Wang, Yuxuan Wang, Zhongqi Fu, Mengyu Zhou, Ziru Han, Haoxuan Yu, Yingxia Yang, Jiawei 

作者机构：Dalian Maritime Univ Coll Environm Sci & Engn Dalian 116026 Peoples R China Tsinghua Univ Future Lab Beijing 100084 Peoples R China Fushun Vocat Tech Inst Dept Finance & Econ Fushun 113122 Liaoning Peoples R China 

出 版 物：《MICROCHEMICAL JOURNAL》 (Microchem. J.)

年 卷 期：2025年第208卷

核心收录：

学科分类：081704[工学-应用化学] 07[理学] 08[工学] 0817[工学-化学工程与技术] 070302[理学-分析化学] 0703[理学-化学] 

基　　金：Fundamental Research Funds for the Central Universities 

主　　题：DD-SIMCA Model optimization C-H stretching vibration Principal component Preprocessing method Normalization 

摘      要：DD-SIMCA was combined with Raman spectra of C-H stretching vibrations to accurately distinguish between authentic gas-to-liquid (GTL) synthetic lubricating oil and samples adulterated with mineral oil at various levels. The effects of principal components, normalization and preprocessing methods on the model s performance were investigated, by which the model optimization was comprehensively considered for authentic samples in both the training and test sets, as well as for the adulterated samples in test set. Heatmap analysis on a parameter epsilon, defined as 1 - (sensitivity of training set x sensitivity of test set x type II error), was conducted to optimize the model for authentic samples in the training and test sets. The model s performance for adulterated samples was evaluated using a parameter delta, which measured the distance in the ODlog dimension between the centroid of samples in the test set and the acceptance boundary for authentic samples. Generally, increasing the number of principal components had a significant effect on enlarging the ODlog of the authentic sample centroid in the test set while it slightly affected the training set. When the training and test sets both achieved 100 % sensitivity, area normalization was observed to be more effective than maximum normalization in optimizing the model for authentic GTL samples. In addition to accurately classifying authentic samples, multiplicative scatter correction (MSC) resulted in a model with excellent identification ability for adulterated samples when employing either maximum intensity normalization with six principal components or area normalization with four principal components.