Validation and Uncertainty Quantification of a digital model for an oxy-coal combustion power unit using Bayesian-based analysis

The transition to low-carbon energy systems has increased interest in oxyfuel combustion due to its potential for high CO2 capture efficiency. Currently, employing oxyfuel combustion in industrial applications presents a considerable obstacle to widespread use due to the substantial effort and financial investment required. High-fidelity modeling offers a cost-efficient method to investigate oxy-coal combustion in large-scale industrial boilers. This modeling demands an additional evaluation step to assure the accuracy and adequacy of these multiphysics models. This paper introduces a machine-learning strategy utilizing Bayesian Inference coupled with bound-to-bound data collaboration for Uncertainty Quantification, aimed at evaluating the uncertainty in both simulation and experimental data. This novel Bayesian Inference UQ technique is intended to identify the hyper-parameter space where simulations closely match experimental outcomes. The core achievement of this research is improving the predictive accuracy of key quantities of interest related to oxy-fuel combustion. Consequently, the average uncertainty in predicting these key quantities for the systems studied is reduced to within a 5% range. This findings provide a high-fidelity modeling framework that can accelerate the deployment of oxyfuel combustion by improving predictive accuracy and reducing design uncertainties. The insights gained contribute to the development of cost-effective, optimized oxyfuel combustion strategies, facilitating its integration into large-scale carbon capture initiatives.