Extracting a smart model for determining defrost time in household refrigerators using a combination of CFD, experimental data, and GMDH type artificial neural network

The present paper introduces a model that integrates Computational Fluid Dynamics (CFD), experimental data, GMDH-type Artificial Neural Networks (ANNs), and image processing to smartly forecast the optimal time for defrosting household refrigerators. Initially, five parameters that have a direct impact on defrost time are identified: compressor runtime, frequency of door openings, a door-open state duration, ambient temperature, and humidity. Subsequently, experiments are planned using the Response Surface Methodology (RSM) approach, with all tests conducted in compliance with ISO 15502 standards. To reduce the time and cost associated with experimental procedures, CFD simulations are used in parallel with the experimental data to help validate the model and to support the derivation of future refrigerators models. Extracted images from the experiments are analyzed to determine the optimal defrost time using image processing, and these data are used to model the objective function (optimal defrost time) via GMDH-type neural networks. The developed model predicts defrost time with high precision across various environmental and operational conditions. Traditional defrost schedules often rely on critical timing, which is rarely necessary in practice. The model proposed in this study dynamically adjusts to real-world conditions, thereby reducing energy consumption and enhancing the energy efficiency of household refrigerators.