9. enGineerinG Science
9.2 combustion
Combustion has played an important role in the history of human evolu-tion and today it is still very important for our society, as it is the major approach of converting chemical energy to heat and power, making up of more than 80% of the total energy utilization worldwide. In Sweden, about 60% of the total energy supply in 2010 was from combustion of various fu-els, including traditional fossil fuels typically used in the transport sector, and biofuels, peat and waste typically used for heat and electricity
produc-tion. Combustion of hydrocarbon fuels contributes to air pollution and greenhouse emission which affects the earth climate and our daily life. Fire accident and explosion are other undesirable combustion processes. The ad-vancement of combustion science is essential for the control of combustion accident and for the design of efficient combustion devices that have low pollutant emissions.
Numerical simulation of combustion processes is, today, one of the three major approaches, together with theory and experimentation, used by scien-tists and engineers to develop quantitative description of the various details involved in combustion, such as flame structures and propagation, intrinsic flame instability, ignition and quenching, pollutant emissions, and turbu-lence/chemical reaction interaction. In Sweden there is a strong industry in combustion engines through companies such as Volvo and Scania, who are world leaders in combustion engine development and manufacturing. High fidelity numerical simulations are used by engineers in these companies for the design and optimization of their combustion devices. It is expected that high performance numerical simulations will play a more important role in the future R&D work in the Swedish industry involved in energy and power systems since it will lead to reduced cost by avoiding extensive and expansive experimental campaigns. This will improve the competitiveness of their products on the worldwide market.
Combustion is a multiple disciplinary science involving physics (fluid flow and heat transfer), chemistry (chemical kinetics) and mathematics (partial differential equations). The flow velocity involved in combustion of engineering interest is typically high enough that the flow is turbulent.
Turbulence eddies with a wide spectrum of scales interact with chemical reactions in a highly nonlinear way, leading to various combustion phenom-ena such as flame stabilization and flashback, thermo-acoustics induced oscillations in gas turbines, and lift-off/self-ignition, knocking, and cyclic variations in piston engines. For gas turbines and piston engines, liquid fuel injection, spray evolution, vaporization, and mixing of vapour fuel with air need to be considered. Combustion of biofuel that makes up of one third of the Swedish energy supply today involves pyrolysis and inter-particle transport processes. Numerical simulation of these combustion processes is a challenging research area of crucial importance.
Currently there are three different approaches used in combustion sim-ulations. The first is the model-free direct numerical simulation (DNS), which provides detailed spatial and temporal distribution of all important flow and thermo-chemical variables. It is typically used to gain insightful information into a specific combustion process such as the onset of igni-tion in homogeneous charge compression igniigni-tion condiigni-tions in a 53mm3
domain. DNS of this type for simple fuels, such as methane, requires sev-eral millions of CPU core-hours. It will therefore remain not feasible for combustion simulations in practical combustion devices in the near future.
The most widely used simulation approach in industry R&D groups is based on the Reynolds averaged Navier-Stokes (RANS) equations and transport equations for species and energy. The smallscales of turbulence and reaction layers are not resolved in this approach, which significantly reduces the res-olution of grid and the number of time steps required to achieve a statisti-cally averaged field. The third approach is the large eddy simulation (LES) approach, which resolves the energy containing eddies of turbulence, and hence it is more suitable to simulate processes involving large-scale coherent structure such as vortex-shedding in jet burners and bluff-body stabilized flames, precessing vortex core (PVC) in swirl-stabilized flames in modern gas turbines and cyclic variations in piston engines. LES requires a much finer grid and hence it is more computationally expensive than RANS; how-ever, due to the advantage in capturing unsteady phenomena, LES is ex-pected to be more frequently used in future industrial R&D work.
As an example we consider here a combustion LES project carried out at Lund University. The problems examines the use of LES to study liquid n-heptane combustion in a diesel engine with two different injector geom-etries. LES was shown to have a superior capability in capturing the process of spray evaporation, mixing of the vapour fuel with the ambient hot air and ignition of the fuel/air mixture, as well as the stabilization of the diesel flames. The influence of flow swirl is clearly identified in the LES, with the upwind side of the flame being stabilized further downstream of the jet than the downwind side of the flame. The inter-jet angle is shown to affect the flame structures significantly. For this simulation 3.5 million mesh cells were used; with 512 cores the run took 3 weeks for simulation of the fuel injection and combustion process.
Since both LES and RANS do not resolve the reaction zone structures, a model is needed to take into account the interaction between turbulence eddies and the chemical reactions. For simple problems, such as low Karlo-vitz number flamelet combustion, there are different models developed and tested for engineering combustion simulations. For more complex prob-lems, there is a need to do further model development and validation. This is currently being done in university research groups where DNS is used to assist the development of combustion models. This trend is expected to continue in the future.
A typical three-dimensional DNS combustion simulation of academic interest will require at least several millions of CPU core hours and a few billions of grid cells. This requires computers with 104 cores in order to
fin-ish within a few weeks. For RANS and LES, however, the requirement on computers are quite different. Instead of large number of cores needed for one DNS job run, there will be more jobs run for RANS and LES in order to study a wide range of combustor geometry and/or operating conditions for the design and optimization of the combustors. Each job employs far fewer grid cells, typically 105 for RANS to 106 for LES. The number of cores required for each job in RANS and LES then ranges from 10–102. Currently the SNIC Tier-1 computers are suitable only for RANS and LES. To perform combustion DNS the SNIC computers should be upgraded to Tier-0 ma-chine level.
9.2.1 Potential breakthroughs
• Direct numerical simulation of full combustion processes, both in free space and close to solid walls, for realistic Reynolds numbers and temper-atures.
• With the help of DNS data and experimental data, development of high fidelity models for the interaction of turbulence and chemical reactions, and application of LES and RANS simulations to study large-scale engi-neering combustion processes.
• Improvement of engineering design processes in the area of automotive engines, gas turbine engines, and industrial furnaces.