A study on prediction performance of the mechanical properties of rubber using Fourier-transform near infrared spectroscopy. Issue 6 (December 2018)

• Record Type: Journal Article
• Authors: Pornprasit, Rattapol; Pornprasit, Philaiwan; Boonma, Pruet; Natwichai, Juggapong
• Is Part Of: Journal of near infrared spectroscopy. Volume 26:Issue 6(2018)
• Page Start: 351
• Page End: 358
• Publication Date: 2018-12
• Subjects: Fourier-transform near infrared -- rubber -- mechanical strength -- hardness -- tensile strength -- partial least squares regression; Near infrared spectroscopy -- Periodicals; 543.5
• DOI: 10.1177/0967033518805277
• Languages: English
• ISSNs: 0967-0335

Abstract

Near infrared spectroscopy is a spectroscopic method used for quality and quantity analysis of agriculture products and industry materials. Rubber is a mostly raw material of any products. NIR spectroscopy had been using to analyze the mechanical properties of rubber and polymer materials. Prediction models were built from the correlation between the NIR spectra and mechanical strength values (hardness and tensile strength). Raw data were pretreated to improve the prediction models, where the prediction models were based on partial least squares regression and support vector regression. In the case of hardness prediction, the raw dataset was pretreated with standard normal variate transformation or a combination of Savitzky–Golay smoothing and multiplicative scatter correction, following which orthogonal signal correction and uninformative variable elimination were used for feature selection, and partial least squares regression and support vector regression were applied for the prediction model. For tensile strength prediction, the pretreatments were multiplicative scatter correction or combination of Savitzky–Golay smoothing and multiplicative scatter correction, following which orthogonal signal correction and uninformative variable elimination were used for feature selection, and partial least squares regression and support vector regression were applied for the prediction model. From these processes, the r 2 values were greater than 0.9, the bias values were among ±0.5, and the RMSEP values were lower than 5.