ENHANCING COMBUSTION EFFICIENCY THROUGH ADVANCED RADIACTIVE HEAT TRANSFER ANALYSIS AND ML PREDICTION OF GAS EMISSIVITY AND FLAME TEMPERATURE

Abstract

This study presents a hybrid computational framework integrating radiative heat transfer (RHT) physics, Design-Expert-based response surface methodology (RSM), and Machine Learning (ML) for predicting and optimizing combustion characteristics in gaseous flame systems. RHT is modeled using the Stefan–Boltzmann law under gray-gas assumptions, while gas emissivity is characterized using temperature-, composition-, and soot-dependent correlations. A Design-Expert central composite design (CCD) systematically investigates nonlinear relationships among key combustion variables and develops response surface models for combustion performance. In parallel, a Random Forest ML model is developed using combustion-related features, including flame temperature, spectral radiation characteristics, soot concentration, gas composition, and operating conditions. The model achieves high predictive performance, with R² values of 0.998 for flame temperature and 0.978 for effective emissivity. However, these results require cautious interpretation because the dataset contains approximately 800 samples and covers a relatively narrow temperature range of 1150–1400 K, potentially limiting model generalizability under substantially different operating conditions. Comparison with physics-based evaluations further indicates that the physics-based models tend to overestimate radiative heat losses; therefore, strong agreement between the two approaches cannot be claimed. Instead, ML provides an accurate data-driven representation within the investigated range, while physics-based models provide the theoretical basis for interpreting RHT behaviour. Design-Expert desirability optimization identifies operating conditions that balance flame temperature, emissivity, RHT, and combustion performance. The framework provides a useful approach for combustion analysis and optimization, while emphasizing the need for larger, more diverse datasets and broader operating conditions to improve generalization. Potential applications include high-temperature furnace design, thermal-energy management, and data-driven combustion optimization.