Robust predictive models for estimating frost deposition on horizontal and parallel surfaces

Abstract

The phenomenon of frost is affected by different parameters, and considerable complexity is involved in the process. A Multilayer Perceptron-Artificial Neural Network (MLP-ANN) is developed to eliminate the limitations by estimating the frost density and layer thickness over wide ranges in both horizontal and parallel plate configurations. A comparative study between the developed MLP-ANNs, the other most popular intelligent methods, and the well-known empirical and theoretical models highlights the overall better performance of the MLP-ANN models presented in the current study. The R2 for the MLP-ANN models were 0.9994, 0.9997, 0.9953, and 0.9965 for the frost thickness and density on horizontal and parallel surfaces, respectively. Additionally, the quality of the collected data samples and the applicability domain of the MLP-ANNs are assessed using the Leverage algorithm. The results demonstrate the predictability of the suggested scheme for precisely calculating frost deposition over wide ranges on both plate configurations under different conditions.