采用近红外透射光谱研究了汽车制动液品牌及新旧的鉴别。 采集宝马(BMW), 丰田(Toyota), 沃尔沃(Volvo)以及嘉实多(Castrol)四种品牌的汽车制动液全新样本以及用过的样本的透射光谱。 分别对每一种品牌下全新与用过汽车制动液样本的光谱数据进行主成分分析(PCA), 主成分得分图表明不同品牌制动液以及该品牌下全新样本以及用过的样本能够被较好的区分, 其光谱特性存在差异。 基于主成分载荷(Loadings)进行特征波数选择, 偏最小二乘判别分析(PLS-DA), 线性判别分析(LDA), 簇类独立软模式法(SIMCA), k最邻近分类算法(KNN), 随机森林(RF), 误差反向传播人工神经网络(BPNN), 径向基神经网络(RBFNN), 极限学习机(ELM), 支持向量机(SVM), 最小二乘支持向量机(LS-SVM)等判别分析方法用于建立基于特征波数的判别分析模型, 判别模型的建模集和预测集判别正确率均略低于或达到了100%。 与其他三种品牌汽车制动液相比, 嘉实多全新样本与用过样本的差异较小, KNN与LS-SVM模型的建模集正确率均低于100%。 结果表明, 近红外透射光谱结合特征波长选择以及判别分析模型对不同品牌制动液以及同一品牌下全新样本以及用过的样本进行识别是可行的, 为开发在线或便携式仪器提供理论支持。

Near-infrared transmittance spectroscopy was used to identify brake fluid brands, new and used brake fluid of each brand. The transmittance spectra of the new and used samples of 4 different brands of brake fluid, including BMW, Toyota, Volvo and Castrol were collected. PCA was conducted to the spectral data of the new samples of the four brake fluid and the spectral data of the new and used samples of each brand. The PCA scores scatter plot indicated that there were differences among the four brands of brake fluid, and there were also differences between new and used samples of each brand. Optimal wavelengths were selected for identifying different brands and new and used samples of each brand by loadings of PCA. Classification models were built using the optimal wavelength, including Partial least squares-discriminant analysis (PLS-DA), Linear discriminant analysis (LDA), Soft independent modeling of class analogy (SIMCA), k-nearest neighbor algorithm (KNN), Random forest (RF), Back propagation neural network (BPNN), Radial basis function neural network (RBFNN), Extreme learning machine (ELM), Support vector machine (SVM), Least-squares support vector machine (LS-SVM). All classification models obtained good performances, the classification accuracy of the calibration set and the prediction set are 100% for most models. Compared with other three brands, new and used samples of Castrol showed slighter difference, and KNN and LS-SVM models performed worse with classification accuracy under 100% in the calibration set. The overall results indicated that near-infrared transmittance combined with optimal wavenumber selection and classification methods could be used to identify brake fluid brands, new and used brake fluids, the results of this study could provide theoretical support for developing online and portable devices.