MACHINE LEARNING APPLICATIONS FOR PREDICTIVE MODELING AND OPTIMIZATION OF ROUGHNESS GEOMETRIES IN SOLAR AIR HEATERS: A COMPREHENSIVE REVIEW

Abstract

Solar air heaters (SAHs) represent a cost-effective renewable energy technology but suffer from inherently poor convective heat transfer coefficients, limiting thermal efficiencies to below 35%. Artificial roughness techniques enhance heat transfer but create complex design optimization challenges requiring extensive computational fluid dynamics (CFD) simulations. Machine learning (ML) has emerged as a transformative approach for rapid performance prediction and multi-objective optimization of roughness geometries. This comprehensive review analyze and emphases on artificial neural networks (ANN), deep learning architectures, and hybrid CFD-ML methodologies applied to SAH roughness design. Key findings reveal that ANN models achieve prediction accuracies exceeding R²=0.95 for Nusselt number and friction factor, reducing computational time by 200× compared to CFD. NSGA-II coupled with ANN surrogates identifies Pareto-optimal roughness configurations achieving thermo-hydraulic performance parameters (THPP) exceeding 2.4 with 98% time reduction. Critical research gaps include absence of ML studies on non-uniform honeycomb geometries (0 identified studies), limited validation under Indian climatic conditions (31% geographical coverage), and lack of physics-informed neural network implementations. Future directions encompass reinforcement learning for real-time control, transfer learning across roughness types, and edge-deployed ML for adaptive SAH operation. This review provides a comprehensive roadmap for researchers and practitioners implementing ML-driven SAH design optimization.