Study of the Soaking Process of a ready-to-eat rice of Assam (Komal Chaul): A Mechanistic and a Machine Learning Based Approach for spectra-based Estimation of Endpoint

Abstract

This research article focuses on two approaches to study the hydration behavior of a low amylose rice of Assam for the manufacture of a no-cooking rice known as Komal Chaul. Fick’s second law was used to study the diffusion of water during the soaking of brown Chokuwa rice. A machine learning (ML) approach to calibrate NIR spectral data with moisture values. ML models like PCR, and PLS were used for regression, and classification models like Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Classification and Regression Tree, Naïve Bayes, Support Vector Machines, and Random Forest Classifiers were used. The concentration-dependent diffusion coefficients as estimated by applying Fick’s model were found to lie within the range of 2.83 ×10-11 m2/s - 7.92 ×10-11 m2/s. The ML regression models didn’t work well however, the spectral data endpoint classification on a target moisture value of 30% during soaking showed that the Random Forest (RF) classifier predicted the best with classification accuracy close to 0.90. Mechanistic models help us understand the physical phenomenon and the advancement of numerical tools and concepts of digital twins for process operations have led to the use of a sensor-based approach.