A pre-partitioned adaptive chemistry methodology for the efficient implementation of combustion chemistry in particle PDF methods

Large Eddy Simulation/particle Probability Density Function (LES/PDF) approaches are now well developed, and can be applied to turbulent combustion problems involving complex flows with strong turbulence-chemistry interactions. However, these methods are computationally expensive, restricting their use to simple fuels with relatively small detailed chemical mechanisms. To mitigate the cost in both CPU time and storage requirements, an adaptive strategy tailored for particle PDF methods is presented here, which provides for each particle a specialized reduced representation and kinetic model adjusted to its changing composition. Rather than performing chemical reduction at runtime to determine the optimal set of equations to use for a given particle, an analysis of the composition space likely to be accessed during the combustion simulation is performed in a pre-processing stage using simple Partially Stirred Reactor (PaSR) computations. In the pre-processing stage, the composition space is partitioned into a user-specified number of regions, over which suitable reduced chemical representations and kinetic models are generated automatically using the Directed Relation Graph with Error Propagation (DRGEP) reduction technique. A computational particle in the combustion simulation then carries only the variables present in the reduced representation and evolves according to the reduced kinetic model corresponding to the composition space region the particle belongs to. This region is identified efficiently using a low-dimensional binary-tree search algorithm, thereby keeping the run-time overhead associated with the adaptive strategy to a minimum. The performance of the algorithm is characterized for propane/air combustion in a PaSR with pairwise mixing. The results show that the reduction errors are well controlled by the specified error tolerance, and that the adaptive framework provides significant gains in cost and storage compared to traditional non-adaptive reduction approaches.