Identification of Engine Oil Types Using IR Based on LDA-SVM Model

Abstract

In order to realize the rapid identification of engine oil types by infrared spectroscopy (IR), linear discriminant analysis (LDA) was used to reduce the dimension of spectral data, and the characteristic differences of three engine oil types were obtained. The dimensionality reduction data was used as input of Support vector machine(SVM), the identification model of engine oil types is established. A total of 86 samples of three engine oil types were collected. 69 samples were selected as calibration set by Kennard-Stone (K/S) method, and the remaining 17 samples were used as validation set. By setting the LDA threshold parameter, different dimensionality reduction results are obtained. After SVM cross-validation training, the optimal parameters of Linear, Poly, Rbf, and Sigmoid kernel functions are selected. By comparing the prediction results of the calibration set and the validation set, When the kernel function of SVM is sigmoid, the model has good robustness and strong generalization ability. Through LDA-SVM modeling, when the LDA threshold is $10^{-2}$ or $10^{-3}$, the kernel function of SVM is Sigmiod, and the parameter coef equal 1, the accuracy of the calibration set and validation set predicted by the model are both 100%. The LDA-SVM method has a good classification and identification function, and provides a new method for the identification of engine oil types.