Scaling and efficiency of prism in adaptive simulations of turbulent premixed flames

Abstract

The dominant computational cost in modeling turbulent combustion phenomena numerically with high-fidelity chemical mechanisms is the time required to solve the ordinary differential equations associated with chemical kinetics. One technique applied toward reducing computational cost develops an inexpensive surrogate model that accurately represents the evolution of chemical kinetics. One such approach, PRISM, constructs a polynomial representation of the chemistry evolution in a local region of chemical composition space, to be stored for later use. Within this region the representation provides a good approximation to the kinetics. As the computation proceeds, the chemistry evolution for other points within the same region are computed by evaluating the polynomials instead of calling an ordinary differential equation solver. If chemistry evolution is required for a region for which a polynomial has not been defined, the methodology dynamically samples that region and constructs a new representation for it. In this paper we assess the PRISM methodology in the context of a turbulent premixed flame in two diniensions. The practicality is influenced by the size and number of regions necessary to model the subset of composition space that is active during a simulation. The number required grows with decreasing size in a manner that scales exponentially with the dimensionality of active composition space. We considered a range of turbulent intensities, ranging from weak turbulence that had little effect on the flame to strong turbulence that tore pockets of burning fluid from the main flame: for each case we explored the scaling behavior as a function of region size and turbulent intensity.