One- and two-dimensional simulations of diffusion flames using complex fuels and distributed-memory computing clusters

One-dimensional counterflow and two-dimensional coflow diffusion flame models have been applied to problems that provide significant insight into more complicated, practical combustion systems. Laminar, oxygen-enriched, counterflow methane diffusion flames have been studied under conditions approximating those in methane-fired industrial furnaces. The present study demonstrates that the inclusion of a radiation submodel that accounts for contributions from both gaseous species and soot particles is essential to model NO concentration in industrially important oxyfuel flames. Further, this study demonstrates that oxyfuel combustion is an effective strategy to mitigate NO formation only if the oxidizer and the fuel streams can be guaranteed to be greater than 99% pure.;As part of a broader effort to validate the chemical kinetics of practical, multicomponent aviation fuels, counterflow diffusion flames have been examined computationally and experimentally using both a six-component JP-8 surrogate and individual surrogate components as fuels. Reasonably good agreement exists between predicted and measured temperature profiles and extinction limits in flames using both individual surrogate components and the six-component JP-8 surrogate as fuels. A significant reduction in the computational time required to obtain a solution has been realized by implementing the counterflow flame model in parallel. Applying a sensitivity analysis to the problem has generated preliminary skeletal mechanisms that retain the comprehensiveness of the full semi-detailed kinetic mechanism.;Memory constraints prevent the simulation of sooting, axisymmetric, coflow JP-8 surrogate diffusion flames on a single processor, thus necessitating a parallel implementation of the two-dimensional coflow flame model. The first step towards this goal is the solution of a sooting, axisymmetric ethylene-air diffusion flame across sixteen processors of a distributed-memory computing cluster using a damped modified Newton's method coupled to a multi-domain preconditioned Bi-CGSTAB solver. Both the parallelized nonlinear solver and the multi-domain preconditioned Bi-CGSTAB solver scale very efficiently across sixteen processors, and this parallel implementation reduces the per-processor memory consumption by roughly a factor of six. These results bode well for the examination of axisymmetric JP-8 surrogate diffusion flames that require significantly more memory and that require additional scaling beyond sixteen processors.