Developing And Validating Models With Multi-parameters For Estimating The Top Electrode Voltages In Radio Frequency Heating Systems

The top electrode voltage is the energy source of the parallel plate-configured radio frequency(RF)heating systems and a critical boundary condition of relevant computational simulation of the RF heating processing,which dominates the heating rate and pattern.Systematic analysis of the factors affecting the voltage of charged electrode as related to fundamental properties of food is crucial for understanding the working principle of radio frequency(RF)heating systems and accurate-efficient computational modelling.Precise and fast estimations of the voltage of free-running oscillator(FRO)RF system are a difficult issue since the traditional methods including the analytical and trial and error methods have the drawbacks of poor accuracy and time-consuming,respectively.The published model for estimating the voltage is established based on the parameters that can be easily acquired,such as RF power.However,the model was specialized for the soybeans with specific range of moisture content,volumes and RF heating systems with fixed electrode configurations.Therefore,the aim of this study was to systematically investigate the effects of RF generator types,electrode configurations,sample categories and volumes on the voltage and relevant model parameters.Hence,more generalized models with improved accuracy could be developed as compared to the previous one.The major contents and conclusions of this research are as follows:(1)The voltage of the FRO systems was influenced by the electrode configuration as the installed copper strip enhanced the power coupling process and thus raised the output RF voltage and power.The increase rate of voltage versus RF power decreased with the increasing moisture content of soybeans within the range of4.7%-11.7% on wet basis(w.b.).The multiple regression analysis showed that the moisture contents and anode currents had extremely significant influence(p<0.0001)on the developed equations.The previously published one-parameter model was thus extended to the two-parameter models and more accurate in predicting voltages for the computer simulations.The optimal regression models were both quadratic polynomials and the coefficient values of the equations were totally different under different electrode configurations;(2)A coupling factor was proposed as related to the dielectric properties,thicknesses of sample and electrode gaps of RF systems.Theoretically,the voltage was influenced by the coupling factor,RF power,sample volume and energy efficiency based on the analytical solution.The FRO system had a relatively higher energy efficiency due to larger voltage as compared to the 50-Ω system;(3)The FRO system maintained a stable voltage by self-regulation when loaded with samples of different moistures and volumes.A modified power model with exponent value of 0.5 was better used in fitting the relationship between voltage and RF power than the linear model.The slope parameter of the voltage-power model increased with the increasing coupling factor and decreasing sample volume.The relevant slopes for soybeans and wheat flour were not significantly different(p>0.05)since their coupling factors and volumes were comparable.The sensitivity analysis showed that the coupling factor was the most important parameter and the effect of sample volume on the slope was negligible;(4)The electrode gaps,coupling factors and RF power were employed as the input variables,and the measured voltages as output to train five machine learning models.Their regression performances and application ability were compared.The regression trees and ensemble trees showed worse regression than the Gaussian process regression(GPR),support vector machine regression(SVR)and interaction linear regression.Moreover,SVR showed the best application ability among the models when applied in the RF simulation and processing of rice,wheat flour,wheat kernels,soybeans,and peanut butter.In conclusion,fast estimation of the top electrode voltage is a non-linear issue involving multi-parameters.The improved accuracy and generalization ability of voltage for computational simulation could be achieved by SVR methodology incorporating the electrode gaps and output powers of the RF heating system,dielectric properties and thicknesses of the treated sample.