An enhanced hybrid model for batch sugar crystallization based on the pattern recognition for overall heat transfer coefficient using a machine learning approach

Abstract

A hybrid model is developed and evaluated to simulate the heat and mass transfer in the crystallization unit of a sugar factory. While the mass transfer is modeled by using the kinetic growth rate model, the heat transfer is simulated by applying the energy balance to the model. Here, the overall convection heat transfer coefficient of the crystallizer's heat exchanger is considered as a temperature‐dependent function. As this makes the governing equations more realistic, it can help to increase the model accuracy. Additionally, a thorough examination of key practical equations and principles governing the sugar crystallization process is presented. A regression learner method is applied to extract the pattern of the overall heat transfer coefficient. According to our results, the regression learning model successfully predicts the heat transfer coefficient with an average 7% deviation from experimental results. For the hybrid model, an average deviation of about 10% is observed. The crystallizer's behavior is somehow linear, indicating a constant growth rate of sugar crystals. Furthermore, the heat transfer in the crystallizer is improved by increasing the working temperatures.Practical applicationsThe method and obtained results of this work could be used in the following practical purposes: to find the optimum working temperatures of crystallizers used in sugar industry and to predict total working time of batch crystallizers versus working parameters.