By toluene swell index for cross link density level of prevulcanized (PV) rubber latex knowledge, toluene swell of PV latex is measured for trading and production management. Therefore, aim of this research is to use the Fourier transform near infrared (FT-NIR) spectroscopy with machine learning to classify different cross link density levels by toluene swell index including, Unvulcanized (U) (> 160%swell), Lightly vulcanized (L) (100-160%swell), Moderately vulcanized (M) (80-100%swell), Fully vulcanized (F) (< 80%swell) of prevulcanized (PV) natural rubber latex of raw PV latex and 50% solids content PV latex (PV50). The result shows that toluene swell index of rubber prevulcanized latex could be 91.8% correct classified into L group and M group using PV50 MSC pretreated spectra with PLS-DA classifier. Unfortunately, sample obtained for this experiment were loss of U and F groups. In future, to develop the robust model, the sample of all crosslink density levels should be collected.

Rapid method in measurement of crosslink density is required in factory. The objective of this study was to develop the prediction model of crosslink densities based on near infrared (NIR) spectroscopy method. The prediction models were developed using partial least squares regression (PLSR) with spectral pre-treatment of fractional order derivatives (FOD) and variable selection methods including successive project algorithm (SPA) and genetic algorithm (GA). The result demonstrated that prevulcanised (PV) latex model had higher accuracy than that of PV50 latex model. Effective model in predicting crosslink densities of PV and PV50 latices could be pre-processed with FOD=1 and 0.75, respectively. The prediction model generated with full wavelength had the standard error of cross validation (SECV) of 3.21% and 3.52%, respectively. The model performance of PV latex could improve with variable selection method of GA which reduced the SECV from 3.21% to 3.17% and number of wavelengths reduced from 1059 to 937. The model performance of PV50 could not reduce by using the variable selection method. However, the GA could reduce the number of wavelengths from 1059 to 216.