Predicting Lean Blowout in Combustion Systems Using Multi-Fidelity Modeling

The accurate prediction of lean blowout (LBO) in combustion systems is essential for designing safe and efficient energy solutions, especially in aircraft. However, the complexity and cost associated with high-fidelity experimental data acquisition often limits the quantity of data available for model training. To address this challenge, we present a multi-fidelity modeling approach that leverages both high-fidelity data from limited experiments and low-fidelity data from cost-effective reactor network simulations. The high-fidelity dataset consists of LBO measurements obtained under specific experimental conditions, providing a trustworthy but sparse foundation for model development. In contrast, the low-fidelity dataset is derived from a reactor network designed to use liquid fuel, enabling the generation of additional data points under a wider range of conditions not covered by the experiments. Using these two sources of data, we train a machine learning model that effectively predicts LBO across various conditions. Our multi-fidelity model not only enhances prediction accuracy by capturing the underlying physical processes more effectively but also demonstrates the potential of multi-fidelity modeling in overcoming the limitations posed by data scarcity from experimental measurements. The resulting model showcases significant promise for predicting LBO in combustor systems, providing a powerful tool for optimizing combustion processes and enhancing operational safety