Progress Toward a Robust and Computationally Efficient Burner-Stabilized Premixed Flame Model for Propellant Burning Rate Predictions

Author(s):

Christopher P. Stone

Abstract:

To improve the robustness, computational efficiency, and maintainability of propellant burning rate prediction models, the suitability of the burner-stabilized flame solver, part of the open-source Cantera modeling framework, was studied and compared with the legacy Chemkin-III PREMIX flame model currently used in ARL’s CYCLOPS burning rate model. Cantera’s internal flame solver was shown to have lower single-core run times (~20%–33%) compared with PREMIX when applied to predicting the burning rate of nitroglycerin, a common benchmark, using both a semi-empirical and detailed chemical kinetics model. Only with a specialized version of PREMIX that supports multicore parallelism is the legacy solver able to overcome Cantera’s improved algorithm efficiency. While not currently supported, it is expected that Cantera would be substantially faster with parallelism. To investigate the possibility of a Cantera-based parallel solver, a new flame solver was implemented using Cantera’s internal kinetics, transport, and thermodynamic modeling capabilities but with support for parallelism. Promising parallel efficiencies were achieved (i.e., >75%) with the residual function, which is the backbone of the nonlinear flame solver. A semi-analytical Jacobian assembly method was also demonstrated, which reduced the matrix assembly cost by nearly 10 times over Cantera’s finite-difference method. To address concerns with Cantera’s (and PREMIX’s) nonlinear convergence criteria, a new nonlinear flame solver was demonstrated in which the residual error was directly used as the convergence criteria. This reduced predicted residual errors through the peak heat release zones of the propellant flames.