Machine learning for efficient heat transfer coefficient prediction in complex helical plate heat exchanger geometries
摘要
Accurate prediction of heat transfer coefficients (HTCs) is essential for optimizing the performance of helical plate heat exchangers (HPHEs), especially given their complex flow structures. This study develops a machine-learning-based framework to predict HTCs and improve HPHE thermal performance without relying on computationally expensive turbulence modelling. Using experimental data, geometric factors (pitch ratio, helix diameter, and plate spacing), and thermal parameters, the proposed models effectively capture the nonlinear behaviour of heat transfer. The results demonstrate that increasing flow rates enhances HTC from 450 to 680 W m−2 K−1, while surface modifications such as graphene oxide and nanofluid coatings improve the thermal enhancement factor (TEF) to 1.52 and 1.58, respectively. A CNN-based Bayesian optimization algorithm (BOA) further identified optimal operating conditions, including a pitch ratio of 0.67 and fluid velocities of 0.93 m s−1 (hot) and 0.19 m s−1 (cold), achieving an optimized HTC of 580 W m−2 K−1. The machine-learning framework produced accurate HTC predictions within 2.03 s, compared to 45 min required for high-fidelity simulations, demonstrating a substantial reduction in computational cost. This confirms the potential of ML models as efficient surrogates for complex numerical simulations. The study provides a practical pathway for designing next-generation heat exchangers with enhanced thermal performance. Future scope includes integrating advanced nanomaterials, expanding the ML framework to multi-objective optimization, incorporating real-time adaptive learning for dynamic systems, and validating the approach at industrial scale to further strengthen the deployment of ML-driven thermal system design.