Empirical and computational modelling of glutinous rice drying

Drying of glutinous rice is a time and energy-intensive operation with a significant contribution to postharvest loss. Incomplete, improper and over-drying are highly detriment to the grain quantity and quality. Therefore, effective modelling and optimization of glutinous rice quality during and after the drying process is crucial for ensuring the efficient production of high-quality dried grains. Thus, in this study, freshly harvested glutinous rice was dried under different drying conditions i.e. temperature (50 °C, 60 °C and 70 °C) and layer thickness (15 mm, 25 mm, 35 mm, and 45 mm) in a hot air box dryer. The progression in the grain qualities (moisture content, golden index and colour change) during the drying process and macronutrient retained in the milled rice were monitored using hyperspectral imaging (HSI). The grain qualities were modelled using empirical model and computational intelligence. The intelligence model indudes decision trees (DT), ensembles of trees (EoT). artificial neural network (ANN), support vector machines (SVM), and Gaussian process regression (GPR). To obtain the best conditions for the glutinous rice drying, response surface method (RSM) and genetic algorithm (GA) were used to optimize the drying process in terms of drying kinetics, energy consumption, milling quality and nutrients retained in the grains. The result showed that the moisture diffusivity of the grain ranged from 1.62 x 10-12 m/s to 1.68 × 10-11 m/s while the activation energy was found to range between 40.6 kJ/mol and 42.1 kJ/mol. The grain moisture and golden index increased with drying temperature and reduced with layer thickness. Monitoring the drying process with HSI showed that the Savitzky-Golay first derivative (SG1D) preprocessing technique coupled with competitive adaptive reweighted sampling (SG1D-CARS-PLSR) and variable iterative space shrinkage (SG1D-VISSA-PLSR) gave the best performance (0.9091 ≤ R$ s 0.9501) in predicting the quality of glutinous rice during drying. For the macronutrients retained in the milled glutinous rice, the SG1D-PLSR showed the highest preprocessing accuracy (>98%). The addition of the VISSA algorithm for effective variable selection increased the prediction accuracy to -100%. For the empirical modelling of the grain quality as a function of the drying conditions, the Page, first order, and second order kinetic model had the best performance with 99% accuracy, respectively, but limited to the experimental condition. Using computational intelligence algorithm overcame the limitation of the empirical. The GPR had the highest performance with prediction accuracy ranging between 82.54% and 99.67%. The optimization of the drying process using RSM and GA showed that the drying of glutinous rice should be optimally carried under the temperature of 63.26 + 2.29 °C and layer thickness of 25.19 + 1.24mm for optimal performance. Therefore, the developed optical processing sequence and computational intelligence models are recommended for the effective development of a smart and reliable spectral system. This system enables rapid monitoring, inspection, control, and automation of the glutinous rice drying process, ensuring greater efficiency and consistent quality during production.

Text FK 2024 28 - Declaration Form.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB) Text FK 2024 28 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (15MB) Text FK 2024 28.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (15MB) Official URL or Download Paper: http://ethesis.upm.edu.my/id/eprint/18993

Actions (login required)

View Item