Investigation of jet-fuel flames using an implicit parallel solver, on-the-fly chemical reduction, and experimental validation

The numerical simulation of the combustion in a jet engine is an important milestone towards the production of more efficient combustors. However, obtaining accurate theoretical predictions is challenging. The first main difficulty is generating an accurate model for the oxidation of the fuel molecules, which must include hundreds (if not thousands) of intermediate chemical species. Even when a model is available, the solution of the governing equations requires extremely large computational resources. As of today, the detailed simulation of an engine is well beyond technological capabilities, and even simple problems, such as homogeneous isotropic turbulent flows, can strain very large supercomputers.;Three major issues render this problem difficult to treat: (i) combustion is a multiscale problem in which large integral lengthscales are coupled with small submillimeter scales (e.g., a flame front at high pressure); (ii) the problem is stiff, because it couples slowly evolving scales with fast-evolving turbulent eddies and unstable chemical compounds; and (iii) combustion simulations tend to require very large amounts of memory to store all the variables associated with the problem. Furthermore, unlike cold-flow simulations, in which only variables at neighboring gridpoints are coupled together, chemically reacting flows form a tightly coupled system of equations, in which, at any gridpoint, all the chemical compounds are coupled together by chemical reactions.;One possible approach to multidimensional reactive flow simulation is to rely on implicit solvers, which are unconditionally numerically stable and hence robust to the stiffness of the problem. However, for these solution methods, memory requirements and computational time scale nonlinearly with the number of chemical species, making them potentially inefficient when complex fuel molecules are considered.;Two novel solution algorithms have been developed to mitigate the issue. The first one relies on parallel solution on distributed memory machines, while the second one operates a chemical mechanism reduction that eliminates unimportant chemical species from the problem.;First, in order to devise an efficient and robust parallel solver, different implementations of Newton's method have been studied in detail. Specifically, three different communication layouts were considered with particular attention to the overhead generated by large chemical mechanisms. The final parallel implementation is scalable even for very small subdomains (10 × 10 gridpoints), and it enabled the efficient solution of kerosene laminar flames.;Second, concurrent with the parallel calculations, a mechanism reduction technique has been developed, which removes unnecessary unknowns from the problem. The simplification mechanism is based a novel directed relation graph method which operates "on the fly," and, in contrast to many equivalent techniques, explicitly considers the effect of transport fluxes. The resulting numerical scheme acts in a spatially inhomogeneous fashion and leads to very large reduction ratios, in which most of the flame structure is resolved using about 30% of the chemical species employed in the full mechanism.;The calculations have been verified against both previous solution algorithms and experimental results. More specifically, gas sampling and chromatographic analysis on counterflow flames were initially considered to validate the chemical mechanism. Finally, laser diagnostic measurements of temperature and major species in coflow flames were used as an overall verification of the simulation code in a more complex fluid-mechanical environment.