Application of machine learning methods for investigating the heat transfer enhancement performance in a circular tube with artificial roughness

Abstract

This paper presents a hybrid approach for investigation of heat transfer enhancement performance using computational fluid dynamics and artificial neural network. More than 5,000 CFD simulations are carried out for turbulent flow in pipes provided with artificial roughness of transverse rectangular ribs to analyze heat transfer, pressure drop, and thermal hydraulic performance. The rib height and pitch are widely varied along with the flow Reynolds number, working fluid, and material of roughness elements. To accurately predict major parameters (Nusselt number, friction factor, and thermal hydraulic performance) a deep neural network is developed, trained, and tested by current CFD data. The ANN allowed finding optimal rib roughness parameters for the current problem and opened perspectives of industrial application due to low computational cost and prediction error of less than 1.5%.