Machine Learning and Computational Fluid Dynamics Applications for the Modeling of the Decompression-Induced Innovative Steaming Treatment of Paddy at the Bulk Level

Abstract

Instant controlled pressure drop (ICPD) treatment is a novel approach for the quality improvement of milled rice. In this investigation, temperature and moisture profiling for the bulk-level ICPD treatment of paddy and its effect on the gelatinization kinetics of milled rice were studied using machine learning approaches, namely, artificial neural network (ANN), adaptive neuro-fuzzy interface system (ANFIS), and computational fluid dynamics (CFD). For CFD-based temperature profiles, the coefficient of determination (R²) values ranged from 0.91 to 0.95, while for moisture profiles, the values were between 0.97 and 0.99. In case of ANN predictions, the 2-5-3 architecture showed adequate performance based on an R² value of 0.99 and a mean squared error (MSE) of 0.514. For ANFIS-based predictions, the R² values exceeded 0.99, while the MSE values ranged from 0.0034 to 0.0084. The 2-3-3-1 ANFIS architecture with gaussmf membership function and nine fuzzy rules showed the best results. These results indicated that the ANFIS-based model exhibited more accuracy as compared to ANN- and CFD-based predictions. In addition, the broken rice percentage (BRP) was determined to assess the impact of temperature and moisture profiles on the quality of milled rice, showing a decrease in BRP with the increase of treatment time. The parboiled rice treated for 40 s had the highest quality by considering BRP obtained from all the nodes.