Multi‐objective optimization of cane sugar continuous crystallization system design based on computational fluid dynamics

Abstract

This work focused on improving circulation and mixing of the massecuite and reducing the energy loss in the cane sugar continuous crystallization system. The developed Computational Fluid Dynamics (CFD) model is based on the continuous crystallization system of a sugarcane mill in Guangxi, China and verified with the actual operation data. The calculated entropy production and pressure drop of the system are used as indices for assessing the performance of the system. Next, 350 CFD simulations are conducted in the parametric experiment platform with 9 parameters and produces a collection of data. Then based on the CFD data set, the data‐driven model is employed for regression of the relations between key parameters and indices. The implemented data‐driven model is used for Non‐dominated Sorting Genetic Algorithm (NSGA‐II) to obtain the optimal result of Pareto frontier. CFD simulation with the optimized parameters reduces entropy by 9.76% and reduces pressure drop by 11.52%. Novelty impact statement Multi‐objective optimization of cane sugar continuous crystallization system is performed. The entropy is reduced by 9.76% and the pressure drop is reduced up to 11.52%. The proposed method show the good applicability.

Application of computational fluid dynamics simulations in food industry

“…Then the datadriven model was employed for regression of the relations between key parameters and indices. CFD simulation with the optimized parameters reduced entropy and pressure drop [125].…”

Applications of Artificial Intelligence and Machine Learning Algorithms to Crystallization

“…Meng et al proposed a generalizable workflow in organic molecules crystallization by generating data from 350 CFD. The authors validated the CFD model comparing its output with in situ measurements, selected the most important parameters via PCA, and optimized them by combining particle swarm optimization-support vector machine (PSO-SVR) and the non-dominated sorting generic algorithm (NSGA-II).…”

“…166 Overall, most ML applications in the field attempt to circumvent prohibitively expensive optimization problems by training ML algorithms on CFD data for very fast evaluation of the effects of input conditions on the equipment operation and resulting outputs within a given design space. Meng et al 167 proposed a generalizable workflow in organic molecules crystallization by generating data from 350 CFD. The authors validated the CFD model comparing its output with in situ measurements, selected the most important parameters via PCA, and optimized them by combining particle swarm optimization-support vector machine (PSO-SVR) and the non-dominated sorting generic algorithm (NSGA-II).…”