Development of Reduced-Order Models for Engine Applications

Detailed chemical kinetics is critical for accurate prediction of complex flame behaviors, such as ignition and extinction in engine applications, but difficult to be applied in multi-dimensional flame simulations due to their large sizes. Reduced-order models are needed in such cases to enable high fidelity combustion simulations. This dissertation is focused on developing new model reduction strategies and reduced-order models for engine combustion applications. First, a linearized error propagation (LEP) method for skeletal mechanism reduction is proposed. LEP is based on Jacobian analysis of perfectly stirred reactors (PSR) and can more accurately predict the propagation of small reduction errors compared with the previous methods of directed relation graph (DRG) and DRG with error propagation (DRGEP). Skeletal models generated by using LEP are further validated for auto-ignition and 1-D laminar premixed flames to demonstrate the feasibility of reaction state sampling using only PSR for mechanism reduction. Second, a direct method is developed to accurately and efficiently compute the ignition and extinction turning points of PSR by solving a local optimization problem formulated based on analytic Jacobian. It is shown that the direct method features significantly better accuracy and efficiency compared with the continuation methods that march along the S-curves. Third, reduced and skeletal mechanisms for gasoline surrogates with and without ethanol are developed based on a 1389-species detailed mechanism developed by the Lawrence Livermore National Laboratory (LLNL). The skeletal reduction was performed with DRG, sensitivity analysis, isomer lumping, and the time-scale based reduction is based on linearized quasi-steady-state approximations. The skeletal and reduced mechanisms are extensively validated against the detailed mechanism and available experimental data for ignition delay time and flame speed. The skeletal mechanism is employed in cooperative fuel research engine simulations and the results agree well with experimental data. Lastly, skeletal mechanisms are generated for three gasoline/bio-blend-stock surrogates respectively based on a 2878-species detailed LLNL mechanism for engine simulations. An upgraded solver combining analytical Jacobian and sparse matrix techniques is employed to accelerate the reduction process, such that the reduction time becomes linearly proportional to the mechanism size and a speedup factor of approximately 100 is achieved