| The production of tires requires materials such as rubber,fillers,curing accelerators,and antioxidants.Some materials are toxic and polluting.The use of toxic and polluting materials will pollute the environment and endanger human health.Therefore,many countries have formulated relevant laws and regulations to prohibit or restrict the use of toxic and polluting materials,and vigorously develop "green tires".Because of the high cost of environmentally friendly materials,some companies still use toxic and polluting materials to produce tires in order to save costs.The current testing methods of rubber and additives are difficult to meet the needs of testing due to their complex testing process,long cycle,and low efficiency.Therefore,the research of an accurate,rapid,and environmentally friendly testing method has important significance and application value.In this paper,the testing methods of rubber and additives are studied based on terahertz time-domain spectroscopy(THz-TDS)technology and machine learning algorithms.The main research contents are as follows:(1)Research on the classification method of rubber and additives based on THz-TDS technology and machine learning algorithm.Principal component analysis(PCA)was used to extract the features of the spectral data.Classification models were developed by using BP neural networks(BPNN)and support vector machines(SVM).Eight rubbers,five vulcanization accelerators and five antioxidants were classified.The experimental results show that the classification accuracy of SVM model is higher than the BPNN model.The classification accuracy of SVM model for rubber,vulcanization accelerator and antioxidant were 83.33%,100% and 91.67%.To address the problem that SVM parameters need to be optimized,the honey badger algorithm(HBA)was used to optimize SVM parameters.However,the HBA algorithm converged prematurely leading to a tendency to fall into local optimum.Therefore,this paper uses Bernoulli chaotic mapping,cosine density factor and Cauchy mutation to improve the HBA algorithm.Compared with the SVM models optimized by genetic algorithm(GA)and HBA algorithm,the classification accuracy of the SVM model optimized by improved honey badger algorithm(IHBA)was higher.The classification accuracy of the IHBA-SVM model for rubber,vulcanization accelerator,and antioxidant were 98.96%,100%,and 96.67%.(2)Research on the quantitative analysis of five-component mixtures of rubber and additives based on THz-TDS technology and machine learning algorithm.The PCA method was used to extract features of the spectral data.Quantitative analysis models were developed by using BPNN and support vector regression(SVR)methods.The carbon black content in the five-component mixture of rubber and additives and the antioxidant H content in the five-component mixture of rubber and additives were quantitatively analyzed.The experimental results show that the prediction accuracy of the SVR model is higher than the BPNN model.The quantitative analysis models based on PCA method has a good prediction effect in the detection of carbon black content,but has a poor prediction effect in the quantitative analysis of antioxidant H.Therefore,another variable selection method,iteratively retaining informative variables(IRIV)method,was adopted in this paper to select spectral data variables,and then established quantitative analysis models.The results show that PCA method and IRIV method have their own advantages.In the detection of carbon black content,the PCA-IHBA-SVR model gave the best prediction accuracy with correlation coefficients of 0.99785 and 0.98647 for the calibration set and test set,and root mean square errors of 0.46323% and 1.1594% for the calibration set and test set.In the detection of antioxidant H content,the IRIV-IHBA-SVR model gave the best prediction accuracy with correlation coefficients of 0.94279 and 0.93265 for the calibration set and test set,and root mean square errors of 2.5825% and 2.7947% for the calibration set and test set.(3)Establishment of terahertz spectral database and development of detection software for tire rubber and additives.The database contains spectra of eight tire rubbers,five curing accelerators,five antioxidants,five five-component mixtures with different carbon black contents,nine five-component mixtures with different antioxidant H contents,five twocomponent mixtures with different curing accelerator MBT contents,and five threecomponent mixtures with different silica contents,with 36 spectra each,for a total of 1512 spectral data.The terahertz spectral detection software was developed,which integrated the functions of spectral data import,data pre-processing,spectral visualization,classification,and quantitative analysis.The terahertz spectral detection software is capable of rapid and accurate detection of tire rubber and additives,and improves detection efficiency. |
