Research On Terahertz Spectral Feature Extraction And Quantitative Detection Of Rubber And Additive Multicomponent Mixtures

In the context of the continuous development of low-carbon economy related industries,in order to accelerate the development of“green tires”in the tire industry,various countries have successively introduced relevant regulations and strictly regulated the content of toxic and polluting raw materials used.Therefore,it is necessary to accurately detect the content of additives used in tire rubber.However,traditional rubber additives detection methods have disadvantages such as time-consuming and inaccurate detection results.There are important practical significance and application prospects to research a fast,non-destructive and accurate rubber detection method.As an emerging spectrum detection technology,terahertz spectroscopy technology can accurately and quantitatively detect substances through the"fingerprint"characteristics of different substances in the terahertz frequency band without damaging the experimental samples.Based on THz spectrum detection technology,the spectral characteristics extraction and quantitative detection methods of the components to be measured in rubber and additive mixtures with different components are studied,which have certain practical guiding significance for improving the accuracy of quantitative analysis of mixtures.The main contents are as follows:(1)Quantitative analysis was carried out on the two-component mixture of nitrile rubber and nano calcium carbonate,and the quantitative models of Partial Least-Squares Regression(PLSR)and Support Vector Regression(SVR)were established respectively.The absorbance of the experimental samples are two model parameters,and the content of nano-calcium carbonate in the mixture is quantitatively analyzed,and the prediction results of the two models are obtained.The results show that the correlation coefficients of the correction set and prediction set obtained by the SVR model are 0.9995 and 0.9931,respectively,which are greater than the PLSR model,that is,the prediction results obtained by the SVR model are better.(2)Legendre Moment(Legendre Moment,LM)is introduced as a method of spectral feature extraction into data preprocessing to eliminate redundant information and interference noise in the spectrum of multi-component mixtures.The characteristic information of the absorbance grayscale images of the three-component and four-component mixtures are extracted,and then the LM-PLSR and LM-SVR quantitative analysis models are established respectively and the results are compared with the results of the PLS and SVR models.The results show that in the quantitative analysis of the three components,the spectral features extracted by LM significantly improve the prediction results of the LM-PLSR and LM-SVR models,and the prediction results of LM-SVR are the best,and the correlation coefficient of the calibration prediction set The SVR model's 0.9573 was increased to 0.9726;in the four-component quantitative analysis,the LM-SVR model's prediction results were the best compared with the results of other models,and the correlation coefficient of the prediction set was increased from 0.9393 of the SVR model to0.9440.(3)Krawtchouk moment(Krawtchouk Moment,KM)is used to extract the features of the region of interest of the absorbance spectrum gray image of multicomponent mixture.According to the setting of appropriate parameters ₁p andp₂,the region of KM is selected to extract the feature information which is more conducive to the establishment of quantitative analysis model.Finally,the KM-PLSR and KM-SVR quantitative analysis models were established with the obtained characteristic information to quantitatively detect the content of the target components in the multi-component mixture.The results show that the spectral characteristics of the multi-component mixture extracted by KM can significantly improve the accuracy of the quantitative detection of the model,and the prediction results obtained by the KM-SVR model are better than other models:The quantitative test results of the accelerator CZ showed that the correlation coefficient of the prediction set of the KM-SVR model increased from 0.9440 of the LM-SVR model to 0.9696;the quantitative test results of Mg O showed that the correlation coefficient of the prediction set of the KM-SVR model was changed from LM-SVR The model's 0.9561 increased to0.9786.