Predictive models of the top electrode voltage of radio frequency heating systems for low moisture foods

Abstract

Precise and rapid estimations of the top electrode voltage in free-running oscillator (FRO) radio frequency (RF) heating systems are important for effective computer simulation to improve the heating uniformity. In this study, factors affecting the voltage were comprehensively analyzed in the fixed frequency heating system (50-Ω), which was experimentally and subsequently utilized in the model development of the FRO system using the machine learning (ML) methodology. The ML models were validated by the combined heating experiments involving in typical low-moisture foods (LMFs) (including soybeans, rice, wheat kernels, wheat flour, and peanut butter) with different volumes and moistures under various electrode gaps. An assemble factor describing the power coupling state of samples along with the RF powers and electrode gaps was employed as input parameters, and published experimental voltages of soybeans as output parameters to train five types of ML models. Results showed that the support vector machine model had the optimal regression performance of the soybeans data and the best application capability over the boarder parameter range. The proposed approach may help to transfer the previously published model from soybeans into other LMFs and the developed model could serve as an effective and accurate tool for relevant computer simulation.