Estimation of heat transfer parameters of shell and helically coiled tube heat exchangers by machine learning

Abstract

Shell and helically coiled tube heat exchangers (SHCTHEXs) are heat exchangers that only take up a small space and enable greater heat transfer area compared to traditional models. Information on 21 different SHCTHEXs obtained from catalog was considered for the modeling. Two other artificial neural network structures have been created to forecast the heat transfer coefficient, pressure drop, Nusselt number, and performance evaluation criteria values as outputs. In contrast, tubing and coil diameters, Reynolds and Dean numbers, curvature ratio, and mass flow rate are designed as inputs. In the network structures with 105 data points, 70% of the data was used for training, 15% for validation, and 15% for the testing stages. The Levenberg-Marquardt procedure was evaluated as the training algorithm in multi-layer perceptron network models. The coefficient of determination was as higher than 0.99. The mean deviation was less than 0.01%. The results show that the created artificial neural network structures can acqurately estimate the outputs.