Optimization of combustion kinetic models on a feasible set

We introduce a new approach to optimizing combustion models by constraining the optimization not only to parameter uncertainties but also to the uncertainties in experimental data. Doing so assures that the best-fit solution produces a model whose predictions deviate the least from the experimental targets while remaining within their experimental uncertainties. The new method is built using the uncertainty-quantification framework of Data Collaboration, and hence allows one to examine the influence of parameter and experiment uncertainties on the optimal solution, and to identify experimental targets that are most difficult to match as well as model parameter values that are likely to be questionable. The proposed methodology is flexible to formulate a variety of objectives and thus to answer seemingly heuristic questions. The technique is demonstrated employing the GRI-Mech 3.0 kinetic model and 76 experimental targets.