Machine learning–integrated artificial intelligence framework for novel design of thermophotovoltaic micro-combustors

This study presents an integrated, Pareto-driven design framework for micro planar combustors in thermophotovoltaic systems, leveraging machine learning–based surrogate modeling, multi-objective optimization, and decision-making. To overcome the computational cost of high-fidelity combustion simulations, five artificial-intelligence-enhanced surrogate models (GA-ANN, PSO-ANN, IGWO-ANN, MPA-ANN, and MPSO-ANN) were developed to accurately predict pressure drop, output power, and system efficiency. These models demonstrated excellent predictive capability (R > 0.999), with GA-ANN achieving the highest accuracy for pressure-drop prediction, MPA-ANN yielding the most reliable power estimates, and PSO-ANN providing superior efficiency prediction. The surrogate models were subsequently integrated with four evolutionary multi-objective optimizers (MOEA/D, NSGA-II, MOPSO, and SPEA-II) to generate comprehensive and consistent Pareto fronts. Among them, MOPSO produced the most diverse Pareto set, effectively capturing trade-offs among pressure drop, output power, and efficiency. Performance evaluations showed that output power increased to 6.88 W at higher inlet velocities (∼7.9 m/s), offering a 14 % enhancement but doubling pressure losses. Maximum efficiency (∼2.95 %) occurred at moderate velocities (∼6.0 m/s), indicating thermodynamic limitations as the governing constraint. Pareto-optimal designs consistently favored near-stoichiometric operation (ER ≈ 0.99), moderate tube lengths (5.0–6.0 mm), intermediate diameters (1.70–1.75 mm), and balanced spacing (3.8–3.9 mm). Multi-criteria decision analysis using VIKOR and TOPSIS revealed differing optimal configurations depending on objective emphasis, underscoring the framework's capability to deliver tailored thermophotovoltaic combustor designs for diverse application priorities.