Bayesian model selection for graphite oxidation by molecular oxygen informed by new data

Flow tube experiments and Bayesian model selection and calibration methods are used to assess kinetics models for graphite oxidation at pressures of 33 and 330 Pa and temperatures from 1100 to 2284 K. Bayesian methods, which are well established but not commonly used for carbon oxidation modeling, are employed to select and calibrate models based on the novel experiments. The experiments isolate high-temperature gas-surface interactions and mass loss, enabling low computational cost model evaluations, extensive statistical sampling, and inference methods. Influential reaction parameters are selected by global sensitivity analysis and parameter influence criteria. Bayesian inference is used for both parameter calibration and model class selection among three candidate model classes proposed for air--carbon ablation. The two classes with dual adsorption mechanisms, in which oxygen can adsorb strongly or weakly, are shown to be more plausible than the model with a single adsorption mechanism. Posterior distributions of calibration parameters are constructed to quantify parameter uncertainties and model errors for uncertainty quantification when using these models.