Researchers from the Federal University of Campina Grande in Brazil have developed a new approach to optimize the design of industrial burners, which could help improve their efficiency and reduce emissions. The team, led by Patrick Souza Lima and Paulo Roberto Santana dos Reis, has proposed a multi-fidelity Bayesian optimization (MF-BO) framework that integrates computational fluid dynamics (CFD) evaluations with Gaussian-process surrogates to balance the trade-off between accuracy and computational cost.
The researchers focused on a non-premixed burner configuration, aiming to enhance thermal efficiency when using hydrogen-enriched fuels. They defined a design vector that includes the burner’s height, length, and mesh element size, which together form a continuous fidelity index. This index allows the optimizer to adaptively combine low- and high-resolution simulations, enabling more efficient use of computational resources.
The MF-BO framework employs a calibrated runtime model to penalize computationally expensive queries and a constrained noisy expected improvement (qNEI) to guide sampling under an emissions cap for nitrogen oxides (NOx). The researchers found that the surrogates trained on CFD data exhibited stable hyperparameters and physically consistent sensitivities. For instance, the mean temperature increased with reactor length and fidelity but decreased slightly with height, while NOx emissions grew with temperature but tended to decrease with length.
The best design achieved an average temperature of approximately 2,000 Kelvin while satisfying the NOx limit. Compared to a hypothetical single-fidelity campaign, the MF-BO framework achieved comparable convergence with about 57 percent lower total wall time. This efficiency is due to the framework’s ability to learn the design landscape through fast low-fidelity evaluations and reserve high-fidelity CFD simulations for the most promising candidates.
The methodology offers a generalizable and computationally affordable path for optimizing reacting-flow systems, where mesh-driven fidelity inherently couples accuracy, cost, and emissions. This approach could accelerate design cycles and reduce resource requirements in industrial burner development and other high-cost CFD-driven applications. The research was published in the Journal of Computational Physics.
In practical terms, this research could help energy companies design more efficient burners that produce less pollution. By reducing the computational resources required for burner design, the MF-BO framework could also lower the costs associated with research and development in this area. Additionally, the methodology could be applied to other reacting-flow systems, making it a versatile tool for the energy sector.
This article is based on research available at arXiv.