Integrated multi-objective optimization of a horizontal evaporator structure in domestic refrigerators: Comparison between the semi-empirical model and GMDH neural networks for enhanced pareto frontiers

Abstract

In this paper, an Integrated Multi-Objective Optimization for optimising the horizontal evaporator structure of a domestic refrigerator is proposed by comparing the semi-empirical formulation model with the Group Method of Data Handling (GMDH) neural network model. Using STAR-CCM+ for 3D simulation and Amesim for a 1D system model, it examines the pressure drop, flow rate, and heat transfer of a horizontally finned tubes evaporator with trapezoidal fins. It compares the adaptability of a semi-empirical formula model and the GMDH model. The semi-empirical formulas uses row distance and fin pitch to model air-side flow rate and pressure drop, while the GMDH models uses six design parameters (tube distance, row distance, tube outer diameter, fin pitch, number of units, and fan speed) for its calculations. The semi-empirical models and GMDH models employ multi-objective optimization algorithms with friction factor (f) and heat transfer coefficient (j) as objectives. The design Pareto fronts of both methods do not overlap, creating a comprehensive Pareto front. The unification of these Pareto fronts provides a comprehensive design space to analyze. The evaluation using LINMAP, TOPSIS, and Shannon’s entropy methods showed that LINMAP and TOPSIS provided superior solutions for the more linear semi-empirical model. In contrast, the Shannon entropy method offered a more robust solution for the highly nonlinear and complex GMDH model, making it more suitable for conditions with high uncertainty. The optimal design point in the GMDH model was selected using the Shannon entropy method, with f and j values of 0.174 and 19.430, respectively.