Experimental investigation and physics-informed neural network based modelling of heat transfer coefficient and pressure drop in R513A flow boiling through micro-fin tube

Abstract

This research investigates the heat transfer coefficient (HTC) and frictional pressure drop (FPD) during flow boiling of refrigerant R513A in a micro-fin tube under varying operational conditions. The primary goal is to enhance understanding of flow boiling characteristics to improve heat exchanger designs for environmentally friendly refrigerants. Experimental analysis was conducted using a 1000 mm-long tube with an outer diameter of 9.52 mm, considering heat flux (12–36 kW·m⁻²), vapor quality (0.02–0.96), mass flux (50–300 kg·m⁻²·s⁻¹), and saturation temperature (290.15–300.15 K). A Physics-Informed Neural Network (PINN) model was developed to predict HTC and FPD using these operational parameters as inputs, integrating experimental data with physical constraints for enhanced prediction accuracy. The PINN architecture, comprising 4-64-128-64-2 neurons, was optimized through systematic hyperparameter tuning, achieving well-converged training and validation losses. Key findings reveal the significant impact of heat flux, mass flux, and vapor quality on HTC and FPD, with micro-fin geometries enhancing heat transfer efficiency. This study bridges experimental data with advanced predictive modeling, offering a novel framework for optimizing thermal systems using sustainable refrigerants. The results provide valuable insights for designing high-efficiency, environmentally sustainable cooling systems.