The challenges involved in employing projection-based methods to develop accurate and robust Reduced-Order Models (ROMs) for reacting flows are investigated. The evaluations are based upon a representative benchmark problem that provides tractable CFD datasets containing the essential physics encountered in typical combustion dynamic problems involving both premixed and non-premixed reactants. Datasets from non-reacting solutions of the same problem are also considered providing a systematic four-way comparison of the effects of gradients of temperature only, species only, species and temperature combined, and species, temperature and reactions. Evaluations are based on both dataset reconstruction and future state predictions. ROM robustness is shown to be sensitive to snapshot selection and sampling and to local, steep gradients in temperature and species mass fractions. Comparisons of Galerkin and Petrov-Galerkin projections indicate that global stabilization in itself is insufficient for problems of this complexity; local stabilization techniques are necessary as well. Overall, the investigations narrow the challenges in reacting flow ROM development to small-scale physics arising from local sharp gradients in temperature and species mass fractions.
展开▼