Geoffrey Oxberry, Massachusetts Institute of Technology
Large combustion chemistry models, consisting of hundreds of species and thousands of chemical reactions, are used to study the mechanisms behind coking and pollutant formation in engines. However, incorporating these models into CFD simulation of engines is a computationally expensive endeavor. One strategy to reduce the computational requirement of chemistry models is to replace the full model with smaller, reduced models that approximate the full model over specified ranges of temperatures, pressures, and chemical compositions. Here, model reduction is carried out by solving an integer linear program (ILP). The solution of the ILP yields a reduced model obtained by eliminating reactions and species from the full model. At a finite number of control points (temperatures, pressures, and compositions), constraints ensure that the solution of the reduced model differs from the solution of the full model by no more than a user-specified error tolerance. Case studies were carried out on an n-heptane model consisting of 561 species and 2539 reactions. Profiles of temperature and species mass fractions generated from solving the reduced models accurately predict the solution of the full model within conditions of interest. Reduced models also accurately predict ignition delay times, and their computational requirements are reduced by roughly an order of magnitude, compared to the full model. In addition, the tradeoff between error tolerance and reduced model size is quantified and shows that the number of species included in a reduced model scales logarithmically with error tolerance. These results demonstrate that species and reaction elimination can be used to approximate large chemistry models in an error-controlled fashion for use in CFD simulations.
Abstract Author(s): G. M. Oxberry, P. I. Barton, W. H. Green, and A. Mitsos