Effects of detailed finite rate chemistry in turbulent combustion
MetadataShow full item record
The development of advanced combustion energy-conversion systems requires accurate simulation tools, such as Direct Numerical Simulation (DNS) and Large Eddy Simulation (LES), for capturing and understanding ignition, combustion instability, lean blowout, and emissions. However, the characteristic timescales in combustion systems can range from milliseconds to picoseconds or even lower. This renders the use of detailed finite rate chemistry prohibitive in DNS/LES of turbulent combustion, which requires the calculation of a large number of species and reactions on a large number of grid cells. Due to these high computational costs, DNS and LES typically employ either a flamelet model with detailed chemistry or a simplified/reduced finite rate chemistry with non-stiff reactions. Both approaches, however, are of limited accuracy and may reduce the overall prediction quality. To address this, a framework with high fidelity by incorporating finite rate chemistry, while mitigating additional computational cost, is necessary for the development of advanced combustion systems. In this dissertation, a new numerical framework for DNS and LES of turbulent combustion is established employing correlated dynamic adaptive chemistry (CoDAC), correlated evaluation of transport properties (CoTran), and a point-implicit stiff ODE solver (ODEPIM). CoDAC utilizes a path flux analysis (PFA) method to reduce the large chemical kinetics mechanism to a smaller size for each location and time step. Thermo-chemical correlation zones are introduced and only one PFA calculation is required for each zone, which diminishes the CPU overhead of CoDAC to negligible computation costs. CoTran uses a similar correlation method to accelerate the evaluation of mixture-averaged diffusion (MAD) coefficients. This framework is firstly applied to investigate the non-equilibrium plasma discharge of C2H4/O2/Ar mixtures in a low-temperature flow reactor. The accelerated case has been verified against the benchmark case by both temporal evolution and spatial distribution of several key species and gas temperature. Simulation results show that it accelerates the total computation time by a factor of 3.16, the calculation of chemical kinetics by a factor of 80, and the evaluation of MAD coefficients by a factor of 836. The high accuracy and efficiency of this proposed framework illustrate its promise in the simulation of diverse combustion problems. Secondly, this framework is evaluated for a canonical turbulent premixed flame employing a conventional jet fuel kinetics model. Again, the results show that the new framework provides a significant speed-up of chemical kinetics and transport computation, enabling DNS with large kinetics mechanisms while maintaining high accuracy and good parallel scalability. Detailed diagnostics show that, for this test case, calculation of the chemical source term with ODEPIM is 17 times faster than that of a pure implicit solver. CoDAC further speeds up the calculation of chemical source terms by 2.7 times. CoTran makes the evaluation of MAD coefficients 72 times faster. Comparing to the conventional DNS, the total computation time of this framework in this test is 20 times faster, with that of chemical kinetics 46 times faster, and that of the evaluation of transport properties 72 times faster. Based on the above DNS framework, an efficient finite-rate chemistry (FRC) - LES formulation is developed for numerical modeling of a turbulent jet flame. Comparing to the conventional FRC-LES, this framework provides a speed-up of 8.6 times for the chemistry calculation, and 6.4 times for the total computation, using a 20 species kinetics model. Both the new FRC-LES and flamelet/progress-variable (FPV)-LES are conducted for a piloted partially premixed methane/air flame (Sandia Flame D). The two approaches provide similar predictions in terms of time-averaged flame field and statistics, which agree well with the experimental data. For the instantaneous flame field, FPV-LES predicts significantly smaller regions with high temperature than the FRC-LES case, especially in the downstream region. Near the stoichiometric region, FPV-LES over-predicts the radical generation with respect to the experimental data, but under-predicts the CO generation and heat release, which explains its under-prediction of temperature. In contrast, on the fuel rich side, CO is no longer the bottleneck species, thus the FPV-LES predicts a higher temperature than FRC-LES. With respect to the experimental data, FRC-LES provides overall better predictions than FPV-LES for both temperature and species. Most existing chemical kinetics models offer similar predictions of ignition and extinction in 0D/1D finite-rate simulations of laminar combustion processes. Is it appropriate, therefore, to extend this observation to a 3D turbulent combustion environment? In order to investigate the sensitivity of predictions to chemical kinetics models, two different kinetics models, GRI-Mech 3.0 and an 11-species syngas model, are compared by performing 3D finite-rate kinetics-based DNS of a temporally evolving turbulent non-premixed syngas flame. The framework enables computationally efficient simulation incorporating the detailed GRI-Mech 3.0. Both chemical kinetics models provide comparable qualitative trends, and capture local extinction/re-ignition events. However, significant quantitative discrepancies (e.g. 86~100 K difference in the temperature field) indicate high sensitivity to the chemical kinetics model. The 11-species model predicts a lower radicals-to-products conversion rate, causing more local extinction and less re-ignition. This sensitivity to the chemical kinetics model is amplified relative to a 1D steady laminar simulation by the effects of unsteadiness and turbulence (up to 7 times for temperature, up to 12 times for CO, up to 13 times for H2, up to 7 times for O2, up to 5 times for CO2, and up to 13 times for H2O), with the deviations in species concentrations, temperature, and reaction rates forming a nonlinear positive feedback loop under reacting flow conditions. The differences between the results from the two models are primarily due to: (a) the larger number of species and related kinetic pathways in GRI-Mech 3.0, and (b) the differences in reaction rate coefficients for the same reactions in the two models. Both (a) and (b) are sensitive to unsteadiness and other turbulence effects, but (b) is more pronounced. During local extinction events, the major differences between the results from the two chemical kinetics models are in the peak values and the volume occupied by the peak values, which is dominated by unsteady effects. During re-ignition events, differences are mainly observed in the spatial distribution of the reacting flow field, which is primarily dominated by the complex turbulence-chemistry interaction. Further analysis shows that GRI-Mech 3.0 predicts more net radical production associated with the major global pathways, explaining the prediction of less local extinction and more re-ignition.