Development of a machine-learning-based performance prediction model for indirect regenerative evaporative cooling applications supported by experimental and numerical techniques

Abstract

Advanced prediction tools are essential for assessing suitability of regenerative evaporative cooling systems, significantly reducing the time and effort required for extensive testing. Smart algorithms enable optimizing operating conditions and system performance, making the implementation of artificial intelligence tools crucial. This work aims to create first open-source artificial neural network model for performance prediction of a novel a multi-pass crossflow indirect regenerative evaporative cooler configuration. With this purpose, an artificial neural network structure was established for estimating the product air temperature, relative humidity, cooling capacity and the effectiveness of the proposed cooling system. The model was developed using 50 data points from experiments and validated numerical models, with inlet temperature, humidity, and working air ratio as the input parameters. The cooling capacity ranged between 0.27 and 1.33 kW, while wet bulb and dew point effectiveness were 0.49-0.95 and 0.37-0.67, respectively. The developed model achieved a coefficient of determination value of 0.997 and mean deviation less than 0.08%. The study results demonstrated that neural networks are promising engineering tools for regenerative evaporative cooling systems, reducing the effort and time required for complex numerical modeling or experimental testing.