Polymers are highly important organic materials in industry. Due to the excellent properties of polymers, they are applied in many products, e.g., construction and automotivematerials1,2. In3, it is reported that the consumption of polymers around the world was as high as 20.78 million tons in 2005, and this number increased to 26.73 million tons in2015.The excellent properties of polymers, specifically polyisoprene (a polymer of isoprene, (C5H8) ), which is one of the most important primary chemical constituents of natural rubber, include resilience, elasticity, abrasion resistance, efficient heat dispersion, and impact resistance4,5.These properties cannot be easily obtained from synthetic polymers. Moreover, the properties of rubber can be further enhanced with fillers, where carbon black is one of the most popular fillers due to its ability to enhance certain properties, especially mechanical properties such as elasticity and volume6–8. Such carbon black reinforcement of rubber has been extensively studied as follows. Carbon black physically adsorbs rubber molecules to its surface9,10 or occludes them in internal voids11 which results in partial immobilization of the rubber and an apparent increase in the filler volume. Carbon black also forms an agglomerated inter-particular structure9–12 , which may be associated with specific elastic properties and continuous breakup and rearrangement, finally leading to a strong nonlinear viscoelasticbehavior10–12 .One of the important issues in polymer production is quality measuring and monitoring. Such processes usually require a lot of resources both in terms of time and cost, e.g., chemical agent, labor, and sample cost. There are many attempts to address such issues in past decades13,14. Near infrared (NIR) spectroscopy is a non-destructive technique that can provide detailed analysis in terms of the quantity and the quality of agricultural products15. Typically, NIR reflectance information in spectra from an agricultural product sample is used to predict the chemical composition of such sample by extracting the relevant information from many overlapping peaks. Then, the predicted chemical composition can be interpreted as the quality. Before the quality measurement by NIR can be applied, the measurement system has to be calibrated for accurate results. In general, the calibration can be difficult and this is caused by the complex nature of the NIR spectra, where each of the interesting spectra are almost completely overlapped by the others. The calibrated models require routine checking to improve the accuracy and reduce the estimation error 16. In this paper, we propose extensive experiment results to show the performance of the prediction models built from NIR spectroscopy for the mechanical properties of vulcanized rubbers. Our main contribution is a guideline for creating prediction models of such material. The guideline is tailor made according to the prediction techniques to be applied, and the data pretreatment methods, which highly affects the quality measuring, in a particular scenario.
NIR spectroscopy is a non-destructive technique which provides detailed analysis in terms of the quantity and the quality of agricultural products. Specifically, NIR light covers the region from 4,000 to 12,500 cm-1 (700-2500nm). The C-H, O-H, C-O, and N-H bands in the subjects can be observed due to stimulations of such vibrations in this spectral range15. The NIR spectroscopy technique has several attractive features including a short analytical time, ease of operation, and having a diffused reflectance mode. Generally, multivariate calibration analysis, such as a partial least squares (PLS) regression model, is built to extract information from NIR spectra17. Specifically, the models are developed from the relationship between the spectral data and the interested constituents. Kwolek et al.18 were one of the first groups who evaluated the properties of the resin and the rubber concentration in guayule by NIR spectroscopy. In addition, NIR spectroscopy has been used to study the composition of synthetic polymers and rubbers19–21. Takeno et al.22proposed a Fourier transform NIR (FT-NIR) spectroscopy technique coupled with a PLS regression model to quantify natural polyisoprene in Eucommia ulmoides leaves. It was reported that the optimal models were obtained with second derivative NIR spectra in the region between 400-6000cm-1 (R2, 0.95). Marinho et al.21 studied the application of NIR spectroscopy to analyze natural trans- and cispolyisoprenes from Ficus elastica (cis-1,4-polyisoprene),gutta-percha (trans-1,4-polyisoprene), and mixtures of these polymers. Sirisomboon et al.23 used FT-NIR spectroscopy in the wavelength of 1100-2500 nm to evaluate the dry rubber content of rubber latex. Sirisomboon et al.24 also used short-wave NIR spectroscopy in the wavelength of 700-950nm to evaluate the dry rubber content and the total solids content. Their work can be applied in concentrated latex factory settings. In NIR analytical processes, the observed spectra are usually pretreated as the first process, in which this process is one of the most important steps for successful analysis25. The data pretreatment processes usually refer to transformation of the NIR spectra with the goal of reducing large baseline variations, dimensionality, collinearity, and/or the noise level of observed spectra. For removal of undesirable variations in the data, two types of pretreatment are commonly applied in the analytical chemistry literature, i.e. differentiation and signal correction. Generally, analysts combine more than a single pretreatment technique in order to create precise spectroscopy models. The common approaches include Savitzky Golay smoothing (SG) 26, multiplicative scatter correction (MSC)27, signalcorrection28, and variable selection29,30. Such approaches will be elaborated in the following. One of the most important basic data pretreatments is data smoothing and differential filters proposed in26, 31. SG, a well-known data smoothing and differential pretreatment, aims to optimally fit a set of data points to a polynomial in the least-squares of signal-to-noise26. Such technique first determines the rate of change of absorbance with respect to wavelength, or the slope of the curve at that point. For the derivation, a basic method is finite differences: the first order derivative is estimated as the difference between two subsequent spectral measurement points; the second-order derivative is then estimated by calculating the difference between two successive points of the first-order derivative spectra. Multiplicative scatter correction (MSC) 27,32 is another widely applied pre-processing technique for NIR spectroscopy. Applying MSC can reduce optical interference from the equipment, i.e. spectral noise and background noise in NIR data. The MSC main processes are composed of estimation of the correction coefficients and correcting the recorded spectrum. In the work by Martens et al.32, MSC was applied to overcome optical interference. The observed reflectance was corrected by the coefficients based on different linearization before the prediction models are built. Orthogonal signal correction (OSC) is another approach to reduce the variation in observed spectra correlated to the reference. The approach determines highly related spectra data and subsequently removes the non-related data. Such parts of data can be distinguished by orthogonal data considering the reference in high-dimensional space. Wold et al.33 proposed to apply OSC for the NIR spectroscopy pre-treatment process. Sjoblom et al. further applied the technique to reduce variation when the calibration models are transferred in real-life28. Marklund et al.34 applied OSC to improve the correlations and predictive quality of the PLS models. They obtained high-quality correlations between the NIR spectral properties of pulp and strength properties of paper derived from the pulps. By applying OSC, the correlations from the spectroscopy results can be traced back to the quality of the wood from which the pulps were produced from. Last, the wavelength selection from the whole spectra is an important issue, since it can complicate the NIR spectroscopy and reduce the prediction capability of the model. Uninformative variable elimination (UVE)35 is one of the most important methods for the selection. It is based on a PLS regression coefficient which has been widelyapplied36–38. The method first determines the root mean square error of the prediction, then the PLS models for an individual spectrum are built into a matrix. Subsequently, spectra which do not improve the prediction ability will be eliminated from the matrix. Last, the new prediction model can be built based on the remaining spectrum. The process will be repeated until no improvement can be achieved. Therefore, the UVE does not present any configuration problems from the variable selection issue.