AbstractIn heat exchange applications, frost formation on the cold surface causes a decrease in the rate of heat transfer and growth in the pressure drop. Thus, the study on the frost thermal conductivity has a significant and vital place for the engineers and researchers dealing with the heat exchangers. In the literature, there is a lack of accurate and applicable methods for determination of frost thermal conductivity. Additionally, the high cost and difficulties of experimental works clarify the importance of computational and mathematical methods. The errors in the determination of frost thermal conductivity on parallel surface channels can cause inaccuracy in estimations of frost density and thickness. The main aim of present work is suggesting Gaussian Process Regression (GPR) models based on four different kernel functions for the estimation of frost thermal conductivity in terms of time, air velocity, relative humidity, air temperature, wall temperature, and frost porosity. To achieve this purpose, a total number of 57 frost thermal conductivity values has been collected. Comparing the suggested GPR models and other available computational methods express the quality of the developed models. The best predictive tool has been selected as a GPR model, including Matern kernel function with R2 values of 0.997 and 0.994 in training and testing phases, respectively. In addition, the effectiveness of discussing variables on frost thermal conductivity has been investigated by sensitivity analysis and showed that air temperature is the most effective parameter. The present work gives engineers an insight into frost thermal conductivity and the effective parameters in its determination.The significant advantage of present work is the accurate prediction of thermal conductivity by a brief knownledge in artificial intelligence.
An insight into the estimation of frost thermal conductivity on parallel surface channels using kernel based GPR strategy
