Abstract

View factors (VF) are fundamental for accurately modeling radiative heat transfer between surfaces, as they govern the distribution of radiative fluxes and play a key role in thermal management and fire-safety analysis. This work presents a numerical methodology for estimating view factors and solving associated inverse problems using machine learning (ML) and deep learning (DL) algorithms. The proposed data-driven framework is benchmarked against high-fidelity view factor values computed through contour-integral methods derived from Stokes' theorem. Three representative geometric configurations are analyzed to characterize the spatial and geometric behavior of radiative exchange and to evaluate the predictive capabilities of the surrogate models. Among the tested approaches, the Multilayer Perceptron (MLP) consistently achieved the highest accuracy, successfully capturing nonlinear geometric dependencies and reproducing the spatial VF distributions with near-perfect agreement relative to analytical references. The trained models offer several orders of magnitude improvement in computational efficiency over classical numerical integration, enabling real-time evaluation suitable for parametric studies, optimization tasks, and inverse design. The results demonstrate that ML-and DL-based surrogates constitute a robust and scalable alternative to traditional integral-based formulations, providing accurate, fast, and physically consistent estimations of view factors across a wide range of radiative transfer scenarios.