Physics-informed neural networks as industrial surrogate models for heat exchanger performance prediction: experimental validation and CFD benchmarking

Artificial Intelligence has increasingly been adopted as a practical tool to support engineering analysis, particularly in scenarios where traditional numerical methods are computationally expensive or unsuitable for rapid decision-making. In this context, this study presents a Physics-Informed Neural Network framework applied as a surrogate thermal model for a concentric tube heat exchanger operating with water in both hot and cold streams.The proposed approach embeds the energy conservation equation directly into the neural network loss function, enabling the PINN to learn thermally consistent solutions with minimal reliance on labeled data. Unlike conventional data-driven models, the network is trained using physics-based constraints and boundary-consistent synthetic targets, allowing generalization without direct exposure to experimental measurements. The PINN is formulated to predict outlet temperatures for both parallel and counterflow configurations, ensuring robustness across varying operating conditions while significantly reducing computational cost.The methodology is applied to a real concentric tube heat exchanger installed on a laboratory test bench. Experimental measurements of inlet and outlet temperatures and mass flow rates are used exclusively for validation, while the PINN predictions are benchmarked against both experimental data and high-fidelity Computational Fluid Dynamics (CFD) under steady-state conditions.Results show that the PINN predicts outlet temperatures with relative errors below 5% compared to experimental data and within 7% relative to CFD, while achieving approximately 88% reduction in computational time. These findings demonstrate that PINNs can serve as fast, physically consistent surrogate models for thermal analysis of heat exchangers, supporting parametric studies, design evaluation, and engineering decision-making in industrial thermal systems.