摘要
采用可见，近红外光谱技术对不同品牌的汽车自动变速箱油进行了判别分析研究。采集壳牌、MOTUL（ATF一1A）、美孚ATF220、嘉实多、福斯泰坦等5种不同品牌的自动变速箱油在325--1075nm范围的透射光谱，发现5个不同品牌自动变速箱油的平均透射光谱存在一定的差异。采用变量标准化（Standard Normalized Vafiam，SNV）进行光谱数据预处理，并基于预处理后的光谱建立了偏最小二乘判别分析（Partial Least Squares Discfiminant Analysis，PLS—DA）模型。建模集和预测集判别正确率分别为94．87％和100％。基于预处理后的光谱采用连续投影算法（Successive Projections Algorithm，SPA）算法提取了13个特征波长，并基于特征波长分别建立PLS—DA、反向传播神经网络（Back—prop—agation NeuralNetwork，BPNN）和支持向量机（Suppoa Vector Machine，SVM）判别分析模型。其中BPNN和SVM模型取得了最优的判别分析效果，预测集和建模集的判别正确率均为100％，而基于特征波长的PLS—DA模型与基于全谱的PLS-DA模型的判别准确率相似。此外，SPA选择的特征波长与全谱数据相比，变量数减少了97．83％，并且基于特征波长的判别分析模型效果相当或更优，表明特征波长适用于自动变速箱油品牌的鉴别。本文研究结果表明，可见近红外光谱分析技术结合判别分析模型能够实现对汽车自动变速箱油品牌鉴别。
Visible and near-infrared spectroscopy was applied to identify different brands of automatic transmission fluid （ATF）. Transmittance spectra between 325--1075 nm of five different brands of ATF （namely Shell, MOTUL （ATF-1A） , Mobil ATF220, Castrol and FUCHS TITAN） were collected. Average spectrum of each brand of ATF showed some differences with each other. Standard normalized variate （SNV） was applied to preprocess the origin transmittance spectra, then partial least squares discriminant analysis （PLS-DA） model was built using the preprocessed spectra with classifi- cation accuracy of 94.87% in calibration Set and classification accuracy of 100% in prediction set. Successive projections algorithm （SPA） was used to select sensitive wavelengths based on.the preprocessed spectra, and 13 sensitive wavelengths were selected. PLS-DA, back-propagation neural network （BPNN） and support vector machine （SVM） models were built with the selected 13 sensitive wavelengths. BPNN model and SVM model obtained the best performances with the classification accuracy of 100% in both calibration set and prediction set. PLS-DA model using sensitive wavelengths ob- tained similar performance as PLS-DA model using full spectra. Compared with full spectra, the num- ber of variables of sensitive wavelengths reduced by 97.83%, and the performances of models using sensitive wavelengths were similar or better than the PLS-DA model using full spectra, which indicated the sensitive wavelengths could be used to identify brands of ATF. The overall results indicated that visible and near-infrared spectroscopy combined with classification modeling methods could be used to identify different brands of ArlT. Key words Visible and Near-infrared Spectroscopy; Automatic Transmission Fluid; Successive Projections Algorithm; Brand Identification; Pattern Recognition
出处
《光谱实验室》
CAS
2014年第2期254-259,共6页
Chinese Journal of Spectroscopy Laboratory
基金
国家自然科学基金资助项目（31072247）
关键词
可见近红外光谱
自动变速箱油
连续投影算法
品牌鉴别
模式识别
Visible and Near-infrared Spectroscopy
Automatic Transmission Fluid
Successive Projections Algorithm
Brand Identification
Pattern Recognition
