Uncertainty quantification (UQ) plays a crucial role in predictive modeling in combustion chemistry, as it improves the accuracy of predictions and the reliability. To accurately predict ignition delay times and nitrogen oxides (NOx) emissions of ammonia (NH3) and hydrogen (H2) fuel blends, minimizing uncertainty in combustion kinetic models is critical. This study introduces a novel approach that integrates Bayesian analysis with Artificial Neural Networks (ANNs) to perform inverse UQ and update combustion kinetic models based on experimentally measured nitric oxide (NO) speciation time history. Traditional Markov Chain Monte Carlo (MCMC) methods are effective, but they are computationally intensive and require large datasets, which limit their practical applicability. ANNs are applied as surrogate models to replace traditional kinetic modeling, optimizing the combustion kinetic model of NH3/H2 fuel blends. By integrating Bayesian analysis with ANNs, the computational cost was significantly reduced compared to conventional MCMC methods, while maintaining high accuracy in uncertainty quantification and parameter optimization. This approach facilitates efficient exploration of parameter space and ensures reliable predictions, making it a valuable tool for complex combustion modeling.
Wang et al. (Mon,) studied this question.