3
Combustion and Cyberinfrastructure

OVERVIEW

Combustion has been a scientific and engineering discipline for hundreds of years. The chemical and physical aspects of the field are well known. Today’s central challenges for combustion research are, in principle, easily summarized. Revolutionary changes are needed in the way that fuels are converted into work and heat to significantly reduce carbon dioxide (CO2) and other emissions into the atmosphere while at the same time significantly increasing combustion efficiency so as to make fuels last longer.

Most combustion systems are based on technologies that are very old; gasoline and diesel engines were invented more than 100 years ago. Although they have been improved and refined continuously, they continue to emit toxic chemicals and greenhouse gases, especially CO2. Factory burners, boilers, and other industrial combustion systems are based on even older technology and still produce carbon black (soot) along with toxic chemicals and CO2. To the extent that these systems rely on petroleum—which is the case for most transportation, in particular—they are using up a limited resource that by some estimates will be largely depleted in 50 to 100 years (U.S. EIA, 2007). Thus, it is essential to improve the efficiency with which they use that resource. At the same time, indications are that the world cannot tolerate another hundred years of the toxic emissions and global warming currently produced through petroleum combustion (U.S. Global Change Research Program, 2009). Mere tinker-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 41
3 Combustion and Cyberinfrastructure OVERVIEW Combustion has been a scientific and engineering discipline for hun- dreds of years. The chemical and physical aspects of the field are well known. Today’s central challenges for combustion research are, in prin - ciple, easily summarized. Revolutionary changes are needed in the way that fuels are converted into work and heat to significantly reduce carbon dioxide (CO2) and other emissions into the atmosphere while at the same time significantly increasing combustion efficiency so as to make fuels last longer. Most combustion systems are based on technologies that are very old; gasoline and diesel engines were invented more than 100 years ago. Although they have been improved and refined continuously, they con- tinue to emit toxic chemicals and greenhouse gases, especially CO2. Fac- tory burners, boilers, and other industrial combustion systems are based on even older technology and still produce carbon black (soot) along with toxic chemicals and CO2. To the extent that these systems rely on petroleum—which is the case for most transportation, in particular—they are using up a limited resource that by some estimates will be largely depleted in 50 to 100 years (U.S. EIA, 2007). Thus, it is essential to improve the efficiency with which they use that resource. At the same time, indica- tions are that the world cannot tolerate another hundred years of the toxic emissions and global warming currently produced through petroleum combustion (U.S. Global Change Research Program, 2009). Mere tinker- 41

OCR for page 41
42 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE ing with extremely mature engine and burner technologies will not solve existing combustion problems. A fundamental change in the way that combustion is studied will be required to meet these societal goals. Unfor- tunately, combustion is a complex problem involving enormous ranges of temporal and spatial scales and a wide range of disparate disciplines. Historically, these different disciplines have studied different aspects of the combustion problem independently, with relatively loose coupling between disciplines. Traditionally this decoupling has been particularly acute between the communities dealing with the detailed characterization of molecular properties and the large-scale continuum modeling com- munity that is focused on the complexities of modeling turbulent flows. This decoupling has arisen primarily because the tools to bridge the gap between molecular properties and turbulent flows were not available. Two developments in recent years that have had major impacts on progress in science and engineering have been laser diagnostics and high- performance computing (HPC). Used together, these two technologies have accelerated the pace of research and development (R&D) in many important fields, including combustion research, where they offer the potential to bridge the gap between different disciplines that study dif- ferent facets of the combustion problem. Laser diagnostics have made it possible to examine combustion systems at the molecular level and in complex physical systems where in situ observations had not been pos- sible. For example, it is now possible to probe turbulent flames experi- mentally in ways that elucidate turbulent-flame structure in detail both spatially and temporally. In addition, time-resolved velocity fields and two-dimensional planar images of flame markers to capture the interac - tion of a flame with turbulent flows can now be measured. Computational combustion is quite well established (Westbrook et al., 2005), but it is only within the past decade that HPC has progressed to the point that detailed simulations of the physical and chemical structure of combustion have become possible. Modeling a wide range of two-dimen - sional combustion problems has now become relatively routine. It has even become possible recently to model turbulent flows at the laboratory scale with detailed chemical kinetics and transport (Puduppakkam et al., 2009). This knowledge is essential in exploiting the information obtained by sophisticated observations and optical diagnostics to bring together the different communities working on combustion and to help effect the necessary revolution in combustion science. However, to realize this potential and to achieve significant advances, a much more coherent approach to simulations and data management than exists today is needed. Currently, most combustion investigators work independently or as part of small research groups, each of which develops its own simulation tools and has access primarily to its own

OCR for page 41
43 COMBUSTION AND CYBERINFRASTRUCTURE data. Every group saves experimental and computational data in its own style and format, and many groups develop their own software tools in their own idiosyncratic manner. For example, almost every chemical kinetics group in the world has developed its own submechanisms for the oxidation of hydrogen, methane, propane, and other fuels, rather than each contributing to a collective kinetics model that can be revised by everyone and used by everyone. The only significant exception occurred when a common kinetic mechanism was developed: the GRIMech (Fren- klach, 2007) mechanism for methane and natural gas combustion (see Appendix A in this report), assembled by a team of kinetics experts from multiple research groups and optimized exhaustively using results from many experimental studies. That reaction mechanism was responsible for an enormous number of excellent subsequent studies and provided a thoroughly tested research tool that was used worldwide. GRIMech played a significant role in unifying the community before funding to support continued development of the mechanism was lost. Another example of a community-wide resource was the CHEMKIN family of combustion software tools developed at the Sandia National Laboratories in Livermore, California (see Appendix B in this report). For many years, CHEMKIN provided a valuable common resource world- wide that was used in countless computational combustion studies. The CHEMKIN software tools were particularly important because a combus- tion researcher could walk into any combustion facility or office in the world and immediately be capable of carrying out modeling studies of simple zero- and one-dimensional combustion problems, since the soft - ware tools were the same everywhere. HOW INDIVIDUAL RESEARCHERS WOULD BENEFIT FROM A COMBUSTION CYBERINFRASTRUCTURE As has been found in other scientific and engineering disciplines, the coordination of computer software tools, data storage, formats, and proto- cols can be extremely effective in increasing dramatically the rate of prog - ress in a given field. A well-designed cyberinfrastructure (CI) applied to combustion is a means of enabling this coordination. Among the features of a cyberinfrastructure that would immediately be very helpful would be common repositories and/or registries for experimental and computa- tional data, which would make data sets available to any researcher. The establishment of collaborative groups working to produce, improve, and maintain these common data sets would lead to more accurate and robust models for chemical kinetics, molecular transport, radiation parameters, and thermochemical databases. Similar collaborative groups working at the continuum level would lead to better models for turbulence and other

OCR for page 41
44 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE submodels needed for simulating realistic combustion systems. In addi - tion, a suitably designed CI would facilitate interchange among the dif - ferent disciplines that would serve as a catalyst to change fundamentally the way that combustion research is pursued. Communication of robust and well-validated software tools and databases from developers to users would also be greatly facilitated, and improvements made by one group or individual would immediately be available to many other researchers. As discussed in Chapter 2, the existence of a community CI facilitates such interaction. A HIERARCHICAL APPROACH TO COMBUSTION MODELING The starting point for computer simulations of gas-phase combustion is the compressible Navier-Stokes equations, which describe the conserva- tion of mass, momentum, and energy in a compressible fluid. For com- bustion applications, this system is augmented with additional chemical species equations that describe the transport and reactions of the chemical components in the gas. To examine even a simple laboratory problem such as a steady laminar Bunsen flame, a detailed description of the properties of the gas under consideration must be developed. The thermodynamic behavior of the gas and the chemical kinetics that describes the burning process must be characterized. Also needed are transport properties that specify the viscosity and thermal conductivity of the fluid as well as inter- species diffusive processes. In addition, radiative heat transfer properties are needed to capture accurately the thermal environment of the flame. For more realistic combustion systems, it is necessary to introduce additional complexity to the model. Most practical combustion devices operate in a turbulent environment; consequently, even for gaseous com- bustion, one must deal with the complexities of turbulent flows. Many systems, such as internal combustion engines or jet engines, burn liquid fuels instead of gaseous fuels. Such fuels are introduced into the system as a spray, requiring a treatment of multiphase effects. Other types of sys - tems use particulate fuels, such as pulverized coal in a coal-fired burner or solid fuels in a rocket. The formation of soot during the combustion process introduces additional multiphase effects. This complexity means that a wide range of disciplines are needed to study combustion, which in turn has led to significant specialization within the field. Researchers developing chemical kinetic mechanisms, for example, will have expertise in quantum mechanics, computational chemistry, spectroscopy, and a number of zero- and one-dimensional experimental techniques, but they are unlikely to be experts in turbu - lent fluid mechanics or multiphase flow problems. Similarly, researchers with expertise in turbulence modeling or multiphase systems may have

OCR for page 41
45 COMBUSTION AND CYBERINFRASTRUCTURE expertise in the development of sophisticated numerical algorithms but are unlikely to have any expertise in the development of chemical mecha- nisms or transport properties. Progress in each of these specialties would be more rapid and targeted if each of those communities had ready access to the best data, models, and software from the other specialties. Historically, the linkages between the different subfields of combus- tion have been fairly weak. In particular, the linkages have been minimal between researchers investigating fundamental molecular-level proper- ties such as kinetics and transport and researchers studying continuum behavior of flames. This historical structure of the field reflects the types of tools available to combustion scientists. Simulation capabilities for modeling reacting flows for most problems were limited to relatively simple (e.g., one- or two-step) chemical mechanisms that approximate (often poorly) the basic energetic and flame-propagation properties; com - putations of realistic flames with detailed kinetics were simply not fea - sible. Over the past decade, the advent of HPC and advances in numerical algorithms have changed this, and it is now feasible to simulate realistic turbulent flames with detailed kinetics for simple fuels (Chen, 2011; Lu and Law, 2009). Moreover, new quantitative diagnostic methods are being developed at all scales (Kohse-Hoinghaus et al., 2005), from individual reactive encounters, to controlled molecular ensembles, to in situ stud- ies inside combustion chambers. As these measurement techniques are improved, much better characterization of detailed flame and ignition behavior will provide the data needed for the accurate validation of new simulation capabilities. The changes in the tools available for studying combustion—com- bined with the need to address the challenges posed by developing cleaner, more efficient systems for new, alternative fuels—are driving a need to change the way that combustion research is being done. COMBUSTOR CONFIGURATIONS Practical combustion devices are usually complex systems in which the actual chemical reactions are only one part. To predict and evaluate combustor performance using a computer model, all of the processes that contribute to the overall system must be understood and included in the model, ideally with well-quantified fidelity. For safety concerns, fuel is almost always stored separately from the oxidizer, and they are not mixed at the molecular level until they are about to be burned. Therefore, the mixing of fuel and oxidizer either at the engine process level or within the flame structure is an essential process in the full combustion system, and in many cases the overall performance of the combustion system depends on the details of this mixing. Fuels may be initially in the gas, liquid, or

OCR for page 41
46 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE Hot-Flame Region: Hot-Flame Region: Low-Temperature Combustion: NOx NOx and Soot Ultra-Low Emissions (<1900 K) FIGURE 3.1.1 Internal combustion engine concepts. BOX 3.1 Figure 3.1.1 C urrent and Future Internal Combustion Engine Concepts In a typical gasoline engine, a premixed fuel-air mixture with just enough air to burn all of the fuel injected is compressed and spark-ignited (SI) at a specified, opti- mal time. The high temperature resulting from the stoichiometric combustion leads to significant NOx formation. However, the NOx, along with other pollutants formed, can largely be removed by the three-way catalyst aftertreatment system on all SI engines on the road today. The power output (load) of an SI engine is controlled by restricting (commonly called throttling) the amount of air drawn into the engine and injecting correspondingly less fuel. Overall, the throttling and low compression ratio of an SI engine result in a significantly less fuel-efficient engine than a diesel engine. In part, the lower efficiencies can be overcome with more recent developments that use direct injection of gasoline into the cylinder to eliminate the throttling losses. This option adds substantial flexibility toward ultralean combustion but at the expense of more challeng- ing exhaust gas aftertreatment systems. Direct injection can also be an enabler for homogeneous charge compression ignition (HCCI) engine technology, a new combustion strategy under investigation that approaches the high fuel efficiency of a diesel engine while producing very low NOx and soot emissions—so low that there is the potential to meet the 2010 emissions stan- dards without exhaust aftertreatment. As in an SI engine, fuel and air are premixed; however, combustion is started by a compression-ignition process similar to a diesel engine. Many challenges exist for this promising ultraclean combustion strategy before use in high-efficiency engines can be achieved. The challenges include robust methods of controlling the timing of ignition for optimal performance, expanding the usable load range, controlling the engine through transients, and determining the optimal fuel characteristics. Overcoming these challenges, especially in a diversified-fuel-source future, requires a vastly improved science base around the fundamentals of fuel igni- tion, combustion, and emission formulation chemistry, and fuel spray and turbulent fuel-air mixing processes, all at the high temperature and pressure conditions relevant to engines. These engine types are shown in Figure 3.1.1. SOURCE: Adapted from www.er.doe.gov/bes.reports/files/ReproCTF_rpt.pdf.

OCR for page 41
47 COMBUSTION AND CYBERINFRASTRUCTURE solid phase, so fuel and air mixing can be a multiphase process. If the fuel is introduced into the combustion system in the liquid phase, such as in a diesel engine, then the combustion simulation must include a treatment of a liquid fuel spray and its vaporization and subsequent mixing with air. If the fuel is a solid particulate, a multiphase entrainment model is needed to characterize the system. The combustible fuel and oxidizer mixture must then be ignited, and different practical devices accomplish this ignition in different ways. In some cases, such as diesel or homogeneous charge compression ignition (HCCI) engines, ignition is produced by the compressive heating of the reactants, whereas in other systems such as a gasoline engine, a spark or other ignition system is employed. These concepts are illustrated in Box 3.1. Once the fuel and oxidizer have been ignited, chemical reactions con- vert them into chemical products, and many different types of combus- tion can be encountered. Overall combustion can take place by means of nearly simultaneous fuel oxidation as in HCCI engines, but in most cases, combustion occurs through the propagation of flames. Some engines, such as spark-ignited gasoline engines, feature premixed flames, and other systems, including diesel engines and aircraft gas turbine engines, com - bine features of premixed and non-premixed flames—the latter being the central feature of most industrial burners. Reciprocating engines, found in most internal combustion engines, operate as repetitive time-dependent systems in a cyclically changing environment, whereas turbines used in aircraft and for power generation and industrial burners are statistically stationary turbulent flames (see Box 3.2) in which oscillation instabilities related to acoustic flame interactions are important phenomena. Each type of combustion system or flame requires its own solution algorithms, models, and data analysis. In every important combustion system, the hot gas products transfer heat to do useful work, either by moving a piston or turbine blades or by heating a working fluid such as water in a boiler or furnace. In some large- scale systems, such as furnace combustion, building fires, and wildfires, radiation constitutes an essential long-range mode of heat transfer, and as such it must be included in the simulation. Furthermore, not all of the fuel in a combustion reaction is converted to chemical products such as CO2 and H2O, and incomplete fuel consumption is a primary source of carbon monoxide and other toxic emissions from combustion. In addi- tion, the high temperatures produced by combustion convert molecular nitrogen (N2) into oxides of nitrogen (NOx), which are main components of photochemical smog. In some combustors, the incomplete or inefficient mixing of fuel with oxidizer can lead to the production of soot, which is produced by means of a complex sequence of chemical reactions that starts with gaseous species and steadily build larger molecules until they

OCR for page 41
48 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE BOX 3.2 A ircraft Emissions Produced by Combustion o f Fuel and Air Proposed high-efficiency engines will have combustors that must operate at ex- treme pressures (~50 atmospheres) and inlet temperatures (~950 K). Few facilities with necessary flow rates are available for testing such hardware and are very expensive to operate. Combustor modeling has assisted significantly in the design of recent- generation engines and can be expected to be even more critical at the extreme conditions in cycles of high-efficiency engines. To enable such modeling, accurate simulations of phenomena such as fuel injec- tion, spray and vaporization, fuel and air mixing, mixing and combustion in swirling and jets in-cross flows, heat transfer, fuel chemistry, and heat release rates in flows with Reynolds numbers on the order of 106 are required. Such capabilities are needed to ensure proper combustor performance by enabling quantitative predictions of flow splits and pressure drops, liner temperatures, exit temperature profiles at high power, emissions (gaseous and particulates) at various power settings, ignition, pre-ignition and altitude relight, lean blow-off, and acoustics. Furthermore, fuel effects need to be simulated—or experimentally validated—to ensure safe operation with new synthetic fuels. Testing new combustor designs through the use of advanced and validated modeling methods leads to significant savings. The operation of such an engine is shown in Figure 3.2.1. FIGURE 3.2.1 Turbine engine combustion.

OCR for page 41
49 COMBUSTION AND CYBERINFRASTRUCTURE become so large that they become solid-phase particles. In some cases, such as in industrial furnaces, soot at an intermediate stage in the combus- tion process is a good thing, enhancing radiative heat transfer from the burning gases to the surfaces of the combustor where heat transfer to a working fluid is the goal of the combustion process prior to its subsequent oxidation. In other cases, such as in diesel engines, soot is harmful and contributes to emissions of air pollutants. MODELS, SUBMODELS, AND REDUCTIONS One goal of combustion simulation is to provide system designers with a suite of simulation tools that can solve the relevant equations for all of the physical and chemical processes in an integrated model with quantified uncertainty. This full-system model thus has to include sub - models for each of the processes that contribute to the performance of the combustor. Since the interactions between these subprocesses determine the overall performance of the combustor, each submodel must include the appropriate level of detail to be able to supply the overall model with enough information to simulate those interactions. The fidelity of the sub- models needs to be categorized, and the impact of uncertainty in submod- els on the overall simulation needs to be assessed. Some subprocesses do not contribute to every type of combustor; for example, in many gasoline- fueled, spark-ignition engines, the fuel is prevaporized prior to entering the combustion chamber, so a spray evolution submodel is not required. However, a spray submodel is essential for the simulation of combustion in a diesel engine, in which the liquid diesel fuel is injected directly into the combustion chamber and must vaporize before it can be burned. Most practical combustion systems operate in a turbulent environ - ment. Thus, a full-system model for a combustor is usually based on an underlying computational fluid dynamics (CFD) framework for turbulent flow that provides sufficient spatial resolution to describe the overall combustion process. In most cases, the physical combustor has a complex geometry, so this CFD framework must describe enough geometrical detail, usually in a fully three-dimensional setting. This geometric com - plexity can include the following: • Fluid flows through moving intake and exhaust valves or through an injector, • Swirling flows intended to enhance mixing between fuel and air, • Flames propagating from spark plugs across a combustion chamber, • Explosions occurring within some portions of a combustion cham- ber, and

OCR for page 41
50 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE • Radiation across a burner from a flame to tubes carrying secondary fluid to be heated. The combination of three spatial dimensions and sufficient spatial resolution for each important subprocess often requires a very large com- putational grid, which leads to high computational costs. Each model for the subprocesses involved also can require extensive amounts of detail and therefore computer capacity. Since full-system detailed combustion models are prohibitively expensive for general use, the most common approach for full-system models is to combine simplified submodels for each of the important processes involved in a given combustion appli - cation. A balance is usually necessary, trading computational detail in each submodel against the need for an efficient computer solution, often combining many simulations at lower fidelity with a few simulations at higher fidelity to help bound uncertainties. The starting point of a model for a combustion system is the descrip- tion of the molecular properties of the fluids in the system. The physi- cal and chemical parameters in a typical combustion model cover wide ranges, and the characteristics of these data are one of the primary reasons for the development of a CI for combustion. In particular, combustion sys- tems cover temperature ranges from room temperature to levels of 3000 K in the products of fuel and oxygen flames, and from very low pressures of a few torr in laboratory flame studies to hundreds of atmospheres of pres- sure in diesel engines and some types of detonations. There can be many thousands of different chemical species when a hydrocarbon fuel is being burned, and each species has physical and chemical properties that are often complex functions of temperature and pressure. These include phys- ical parameters such as equations of state, specific heat, heats of fusion and vaporization, ionization potentials and other spectroscopic quantities, frequency-dependent radiative opacities, molecular diffusivities, and oth- ers, as well as chemical parameters including elementary reaction rates with other species, with their activation energies and A-factors, heats of formation, and many other parameters. For most applications, particularly those involving the modeling of a three-dimensional turbulent-flow, high-fidelity characterization of the fluid properties is not computationally feasible (see Box 3.3). For example, rather than using a chemical kinetic reaction model with 1,000 chemical species, a reduced chemistry model with perhaps 10 to 50 species needs to be used. The challenge is that it can be difficult for such a reduced chem - istry model to reproduce the important features from a thousand-species chemistry model. Fortunately, extensive research in model reduction for chemical systems has shown that such reductions can be defined that are reasonably accurate and comprehensive within extensive parametric

OCR for page 41
51 COMBUSTION AND CYBERINFRASTRUCTURE FIGURE 3.3.1 Direct simulation of a turbulent jet flame. BOX 3.3 S tabilizing a Lifted Autoignitive Turbulent Jet Flame Figure 3-3-1 i n a Heated Co-flow as Revealed by Direct Numerical S imulation When a jet flame is lifted from the burner by increasing the fuel or surround- ing air co-flow velocity, the flame can stabilize without a physical element to use for stabilization, and a lifted jet flame is generated. The flame can exist over a range of jet velocities until a critical velocity is reached and the flame blows out globally. Figure 3.3.1 shows a direct numerical simulation of a typical laboratory combustion configuration corresponding to a (small) turbulent, premixed, methane-air, slot Bunsen flame. The chemical model includes 12 reactive species and the computational domain is decomposed into 200 million cells. Lifted flames are found in practical applications such as industrial burners for power generation, in which a lifted jet flame is utilized to reduce damaging thermal stresses to the nozzle material by minimizing contact between the flame and the nozzle. Lifted flames are also found in stratified combustion—for example, in direct injection gasoline engines, diesel engines, and gas turbines—where the fuel and oxidizer streams are partially premixed prior to combustion. The position downstream of a fuel injector where a diesel fuel jet establishes a flame, the so-called stabilization point, influences the degree of premixing between the cold fuel and the heated air prior to combustion, which, in turn, affects the com- bustion and soot formation processes downstream. For example, soot levels decrease as the fuel and air streams are better mixed upstream of the stabilization point and can even be eliminated if sufficient premixing has occurred upstream. In a heated (continued)

OCR for page 41
57 COMBUSTION AND CYBERINFRASTRUCTURE existing multiscale research. Two examples are multiscale simulations of nanoparticle formation from combustion sources (Viola and Voth, 2005) and flamelet modeling of turbulent combustion in which parameterized solutions from simple laminar flames are used to model the small-scale processes in turbulent flames (Peters, 1984). DATA AND DATA FLOW All of the data discussed above, spanning the range of conditions encountered in combustion systems, need to be made available to the combustion research community. However, it is frequently difficult to find all the data that one needs, and in some cases there can be quality-control problems. Although there have been data representation efforts—such as the representation of kinetics data at the Chemical Kinetics Database on the Web at kinetics.nist.gov, and the representation of molecular proper- ties at the National Institute of Standards and Technology Chemistry WebBook at webbook.nist.gov—some data are very sparse or unavailable, because experimental measurements can be difficult or unrewarding. In addition, some data, both experimental and computational, are not suf - ficiently well characterized to meet the needs of the larger combustion community. However, the CI alone will not be sufficient. The community must pay attention to data reuse, including that of metadata, documenta - tion, standards, and registries. Species‑Based Data A common core of physical and chemical data is required for most combustion applications. These include specific heats, entropies, and enthalpies of formation; molecular transport coefficients; heats of fusion and vaporization; and other quantities that are specific for each indi- vidual chemical species involved in kinetic mechanisms, which are used to evaluate macroscopic variables such as the equations of state. Other species-specific quantities of importance include ionization potentials, radiative opacities and cross sections, and other spectroscopic parameters. Some data needs are specific to particular submodels, whereas others are needed for almost every submodel. For example, radiative opacities affect only the propagation of radiation within a combustor, but specific heats and pressures are needed for all cases. These data are determined in a variety of ways, and some can be provided by more than one physical model (Ritter and Bozzelli, 1991; Muller et al., 1995). Most thermochemi- cal data are based on the “bond additivity” approach of Benson (Benson, 1976), which conceptually builds a molecule by combining all of the

OCR for page 41
58 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE contributions from the atoms in the molecule and the bonds between them. However, these “groups” are evaluated in different ways, which result in similar but not identical values. There is a common data format for the results, based on a convenient approach from the Department of Defense’s Joint Army-Navy-NASA-Air Force (JANNAF) tables (McBride et al., 1993). Almost every computational combustion researcher in the world uses thermochemical data tabulated in the JANNAF format, which is a good example of how standardization provides significant economies of time and effort. However, there is at present no ongoing effort to pro - vide convergence of the various chemical models that generate JANNAF data into a single, consensus set of thermochemical data. Chemical Databases A central component of a combustion simulation is the submodel for combustion chemistry. For coupled chemical kinetic models, one dif- ferential equation must be solved for each chemical species in the reac- tion mechanism, and these differential equations are coupled through their mutual chemical reactions. While a hydrogen oxidation mechanism may be as small as 8 to10 species and is relatively easy to solve numeri- cally, a mechanism for the important diesel reference fuel n-hexadecane (n-C16H34) can include 2,100 chemical species and 8,300 individual chemi- cal reactions (Westbrook et al., 2009). Since chemistry submodels, espe - cially those that compute overall rates of reaction and heat release, can include hundreds or thousands of different chemical concentrations, these databases can be extremely large. Each chemical species adds a new differential equation to the overall combustion model; consequently, as the complexity of the chemical model grows, it can easily dominate the computational cost for a simulation. One approach to dealing with this chemical complexity is to reduce the dimen- sionality of the system. For example, for large kinetics models, many simulations are made on simplified, one-dimensional spatial grids that have several hundred spatial zones where large chemical mechanisms can be treated numerically. The best example of such one-dimensional models is the premixed laminar flame, which is widely used to study combus- tion kinetics for many fuels. Many other numerical studies, including a large variety of ignition problems as well as studies of oxidation in jet-stirred reactors and flow reactors, can be carried out under spatially homogeneous (i.e., single zone) conditions with no significant sacrifice of chemical generality. Although these types of simulations can provide important insights into the behavior of the kinetic model, more realistic combustion simula- tions cannot treat a complex fuel with a comprehensive chemical mech -

OCR for page 41
59 COMBUSTION AND CYBERINFRASTRUCTURE anism. Thus, this chemical complexity motivates a need for reduced mechanisms, where the trade-off of using a simpler mechanism versus its reduced cost is well characterized. The CI must provide not only com- prehensive mechanisms for a given fuel but also a hierarchy of reduced mechanisms for which the trade-off of fidelity versus computational com- plexity is well characterized. The data needed to develop the high-fidelity chemical kinetics mod - els represent a synthesis of data from a broad range of disparate sources. For example, the molecular enthalpies needed to simulate the combus - tion of an alternative fuel are extracted by combining information from spectroscopy, calorimetry, synchrotron photoionization, vapor pressure, reversible reaction kinetics, and mass spectrometry experiments with the results of quantum mechanical calculations. Each of those types of data is generated by different research communities, whose members attend dif - ferent conferences and publish in different journals. When a new enthalpy value is reported, it is often discovered that there is a need to supplement previous measurements or calculations using new techniques in order to improve the precision or resolve technical issues with the older data. The CI needs to gather and disseminate all these disparate types of informa- tion, and therefore it must develop interfaces that are useful for each of the subcommunities in addition to the tools that facilitate the integration of these heterogeneous data (see Figure 3.2). The combustion community is broken up into disparate subcommuni- ties, each specializing in different types of data and methods. Combustion chemistry requires data from all of these communities. The CI will need to interface with many different subcommunities and handle very hetero- geneous data. Because of the quantity and variety of data, data-curation services will be an important component of the combustion CI. These services maintain data integrity and enable data discovery, retrieval, and use over time through a range of activities for identification, selection, authentication, representation, preservation, and other processes that span the entire life cycle of data from production to archiving and reuse. Continuum‑Based Data At its core, combustion is the synthesis of chemistry and fluid mechan- ics. The previous subsection discussed the role that the CI could play in enhancing progress at the molecular level. A CI could play a similar role at the continuum level, unifying the efforts of different research groups working on different aspects of the problem and providing a framework for dramatically improving the flow of information. Broadly speaking, there are essentially three subcommunities whose research focuses on combustion at the continuum level. One of those sub -

OCR for page 41
60 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE Kineticists measuring Quantum Calorimetry reversible chemists community reactions Vapor Spectroscopy pressure community experts Synchrotron Molecule- Kinetic photoionization oriented user modelers community interface Central Mass spectrometry Chemistry database community and shared community simulation tools FIGURE 3.2 Expanded view of the chemistry community block from Figure 3.1. Figure 3-2.eps communities consists of experimentalists whose research spans a range of problems, from simple laminar flames such as diffusion flames or vortex- flame interaction, to idealized turbulent flames in a laboratory setting, to in situ measurements in a realistic combustor. The second of these subcommunities consists of researchers who develop and apply simulation methodology to study flame behavior at the continuum scale using a “first principles” approach; namely, solving the reacting Navier-Stokes equations. At the forefront of this subcom - munity is a small number of research groups whose work focuses on harnessing the most powerful supercomputers to perform large-scale direct numerical simulations of turbulent flames. However, this subcom - munity also includes groups using DNS approaches for less demanding turbulent-flame simulations as well as groups studying laminar-flame phenomena computationally. The third subcommunity, whose efforts focus on continuum model- ing, includes the groups that focus on the development of approaches to turbulence and combustion modeling and the application of those mod - els to the simulation of realistic combustors. As with the other subcom - munities, this group also spans a broad range of activities, ranging from those researchers developing high-fidelity LES approaches that capture significant features of a turbulent flow in detail, to researchers focused on improving the fidelity of simplified RANS-type approaches to full-

OCR for page 41
61 COMBUSTION AND CYBERINFRASTRUCTURE system simulations, to engineers applying these tools as part of the design cycle to develop new combustion systems. A focal point of current LES efforts is the development of “multiregime” subgrid models for mixing and combustion that automatically treat the existence of mixed modes of combustion prevalent in advanced combustors operating at high pres- sure and low temperature, and in dilute conditions to maximize fuel effi- ciency and minimize emissions. For example, mixed combustion modes can include partially premixed flame propagation into low-temperature autoigniting mixtures for which traditional, purely premixed or purely non-premixed combustion models do not apply. Experimentation in these adverse environments is limited, and high-fidelity simulation is providing complementary data required to understand and model these complex combustion regimes. These regimes operate near combustion limits where intermittent phenomena—extinction, ignition, flammability limits—result in low-Damköhler-number flames and ignition phenomena character- ized by strong sensitivities to variations in fuel chemistry and properties. Hence, there is demand for simulations that can differentiate fuel effects in fundamental “turbulence-chemistry” interactions encountered in prac - tically relevant environments. The CI requirements for the continuum community share a number of the issues with the infrastructure for molecular properties discussed above. The system would need to support a diverse set of data from a broad range of sources and to provide tools to facilitate the comparison of data from different sources. One example in this context would be tools for the direct comparison of experimental and simulation data. These types of comparisons would range from detailed comparison of laser measurement of nitrogen oxides (NOx) in a diffusion flame, to a statistical comparison of local flame wrinkling in a turbulent premixed flame. Another type of comparison that would be required is the ability to perform comparisons between different types of models at different scales and fidelity. For example, the infrastructure should provide the ability to calibrate the fidelity of a coarse-grained turbulence model by comparing it with high-fidelity DNS data. For some of the data, quality issues are also important. For example, an experiment may not be adequately calibrated for it to be quantitatively compared to a simulation. Similarly, standards of validation and verification are needed to ensure the quality of simula - tion flame data. The committee believes that tools already developed by the combustion community, such as model versus data comparison tools developed by PrIMe,1 will be made available to the combustion CI. CI support for continuum data also presents some significant and 1 See www.primekinetics.org. Accessed December 3, 2010.

OCR for page 41
62 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE unique challenges. The basic issue is that of the sheer volume of data asso - ciated with high-fidelity simulations. Time-dependent direct numerical simulations of turbulent flames being performed today are able to model billions of spatial zones for relatively simple gaseous fuels using either detailed or reduced chemical mechanisms. Typically, detailed mecha- nisms for hydrogen, syngas, or methane can be incorporated in DNS, whereas reduced mechanisms for larger hydrocarbon and oxygenated fuels—for example, n-heptane, di-methyl ether, butanol—accurately rep - resenting low-, intermediate-, and high-temperature kinetics at pressure (i.e., transporting approximately 30 to 80 reactive species) are required for feasibility in DNS on petascale supercomputers. These simulations currently generate data sets measured in the hundreds of terabytes and are only expected to increase in size over time. The data are characterized by velocity, temperature, and dozens of reactive scalar composition fields and additional hundreds of terabytes of particle data used to aggregate Lagrangian statistics in a turbulent flow. The generation of these data sets represents a significant computational investment on the world’s largest supercomputers and, in many cases, is the only means of providing com - prehensive temporally and spatially resolved validation data for model development. Hence, it would be extremely valuable to make these data sets available to the larger modeling community. Providing community access to this type of data raises a number of questions. Should there be an effort to collect such data at a central site, or should data be distributed, with individual data sets remaining at the facilities where they are gener- ated? What types of tools are needed to catalog and access these data? If the data are not centralized, how does one deal with the security issues? What standardized file input/output formats and application program- ming interfaces are needed to facilitate extensible postprocessing and visualization software that operate on subsets of the data? Similar to the GRIMech effort in kinetics, there has been one notewor- thy example of an effort to organize a subset of the continuum combus- tion community, the Turbulent Nonpremixed Flame (TNF) Workshop. The TNF Workshop, organized in 1996 by Rob Barlow from the Sandia National Laboratories, is an open international collaboration of approxi - mately 100 experimentalists and computational researchers focused on non-premixed and stratified turbulent flames. The TNF Workshop has established an Internet library of well-documented flames that are appro- priate for model validation and for comparison of experimental data and modeling results. Although the TNF Workshop is limited to certain classes of flames and has not addressed the issues of providing community access to full-simulation data sets, it has served as a focal point for the commu- nity and is viewed as making a significant contribution to the study of

OCR for page 41
63 COMBUSTION AND CYBERINFRASTRUCTURE non-premixed and, more recently, stratified turbulent flames. Thus far the scale of the experimental data is sufficiently small that data access, move - ment, and processing do not significantly hinder progress. The addition of high-fidelity simulation benchmark data and their comparison with experimental and RANS and LES models will require additional CI to accommodate the sheer volume of the simulated data. Data Flow All of the subcommunities involved in developing and verifying the molecular properties needed by the combustion community are in the chemical sciences, so they all communicate in a common scientific “language,” and they all could be served by similar molecule-oriented user interfaces. Similarly, the different types of groups studying turbulent flames are from the fluid dynamics community and could be served by interfaces specific to their detailed research needs. However, the combus- tion CI must also facilitate information transfer between more dispa- rate communities, and in some cases it would be appropriate to develop different interfaces for different communities. In particular, researchers studying continuum-level phenomena will require a substantially differ- ent interface to the chemical databases than that required by researchers studying kinetics of a particular molecule. Another key ingredient is the need for the data flow in both direc - tions. For example, researchers studying a turbulent flame may notice that there is a large discrepancy in the simulation predictions of a particular pollutant under certain reaction conditions, suggesting that something may be wrong with the chemistry model. The CI should allow for the rapid communication of this information back to the chemistry commu - nity (see Figure 3.3). There is very strong data dependency in the combustion R&D system: a change in one number can require changes in many other numbers at all levels of the system. A dramatic illustration of this problem was the recent discovery that the published enthalpy of formation of OH was incorrect (Ruscic et al., 2001). This enthalpy number had been used to determine the enthalpies of many other species over the years, all of which need to change, and virtually every combustion simulation ever constructed is sensitive to this number, and so must be corrected (Ruscic et al., 2005). In the absence of a CI, all of these thousands of corrections must be per- formed by hand—a process expected to take many years. Clearly the new CI must correct this deficiency of the present information flow, making it practical to update the databases and the combustion simulators quickly whenever new information becomes available.

OCR for page 41
64 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE 1. Fuel chemistry data Chemistry needed to simulate community turbulent flame 5. Notification: data discrepancy 4. Data showing Experimentalists problem in chemistry data Central measuring database pollutants in a and shared 3. Pollutant turbulent flame simulation prediction 2. Chemistry tools Reacting-flow data computationalist Reacting-flow and predicting pollutant formation in a turbulent-flame turbulent flame community FIGURE 3.3 The cyberinfrastructure should facilitate forward and backward in- formation flow between different communities, driving improvements in critical Figure 3-3.eps data and submodels. The choice of enthalpy data in the examples above is just for illustra - tive purposes. There are many other types of data needed for combustion simulations, and most have similar issues (e.g., they are derived from heterogeneous sets of measurements and calculations; their numerical values are interdependent with other quantities; they necessitate informa- tion flow across disciplinary boundaries). The development of submodels for many other processes could also be accelerated if coordinated partnerships, common data formats, and common software tools were available and used widely. Today, prob- lems such as soot inception, growth, and oxidation are being pursued by many independent, individual groups, with minimal interconnection. Radiation transport in combustion devices, turbulence modeling, and many other data-rich submodels have the same characteristics as those outlined above. For example, the accessibility of detailed DNS data sets to the community developing turbulence models could greatly accelerate the development, validation, and verification of those models. The overall rate of progress toward solutions can be much faster, and the community can develop better capabilities with coordinated projects that are built

OCR for page 41
65 COMBUSTION AND CYBERINFRASTRUCTURE around a centralized data-management and software tools core. This approach has never been attempted at this scale in the field of combus- tion, although in the few cases where this type of activity was pursued at a much reduced scope, it was very successful. Examples are PrIMe and CHEMKIN (see Appendix B in this report). Cyberinfrastucture: A New Mode of Organization for a Community‑Level Vision in Combustion Research A CI-enabled combustion community will evolve in a two-tier frame- work organized into three quasi-homogeneous subcommunities (the fuel and chemistry research community, the reacting-flow and flame research community, and the industrial R&D community) under the umbrella of a superstructure for community-level coordination (see Figure 3.1). To facilitate this transformation and to promote the fast, wide, and deep adoption of a CI by combustion researchers, it is envisioned that some of the current stakeholders of the research community will play an increasingly important leadership role. One such stakeholder is the Combustion Institute (www.combustioninstitute.org). The Combustion Institute is the premier international scientific society in combustion. With approximately 2,000 full-time members, it promotes the field of combustion science by organizing meetings, both national (the biennial U.S. National Combustion Meeting) and international (the biennial Inter- national Combustion Symposium), and by helping manage the premier journal publications in combustion science. The Combustion Institute can play a pivotal role by providing the technical expertise, the integrity, and the legitimacy that an effort to build and sustain such a CI will require when interacting with funding agen- cies, individual researchers, and external communities. The Combustion Institute can work at the U.S. national level (through the joint U.S. sec - tions of the Combustion Institute) or at the international level and thereby provide international coordination, should wide-scale CI efforts also be developed in Europe and/or Asia (in fact, the Combustion Institute can presumably promote the emergence of additional CI efforts in Europe and Asia). Finally, and perhaps more importantly, the Combustion Institute can further develop, articulate, and help promote the vision of a new com- bustion research community empowered by a community-wide CI. Other institutions that could undertake this role include the National Center for Supercomputer Applications at the University of Illinois at Urbana- Champaign and the National Institute of Standards and Technology.

OCR for page 41
66 TRANSFORMING COMBUSTION RESEARCH THROUGH CYBERINFRASTRUCTURE REFERENCES Benson, S.W. 1976. Thermochemical Kinetics. 2d ed. New York: Wiley. Bykov, V., and U. Maas. 2009. “Problem Adapted Reduced Models Based on Reaction- Diffusion Manifolds (REDIMS).” Proceedings of the Combustion Institute, Vol. 32, pp. 561-568. Chen, J.H. 2011. “Petascale Direct Numerical Simulation of Turbulent Combustion—Fun - damental Insights Towards Predictive Models.” Proceedings of the Combustion Institute, Vol. 33, pp. 99-133. Frenklach, M. 2007. “Transforming Data into Knowledge—Process Informatics for Combus - tion Chemistry.” Proceedings of the Combustion Institute, Vol. 31, pp. 125-140. Hughes, K.J., M. Fairweather, J.F. Grifiths, R. Porter, and A.S. Tomlin. 2009. “The Application of QSSA via Reaction Lumping for the Reduction of Complex Hydrocarbon Oxidation Mechanisms.” Proceeding of the Combustion Institute, Vol. 32, pp. 543-552. Kohse-Hoinghaus, K., R.S. Barlow, M. Alden, and J. Wolfrum. 2005. “Combustion at the Focus: Laser Diagnostics and Control.” Proceeding of the Combustion Institute, Vol. 30, pp. 89-123. Liang, L., J.G. Stevens, and J.T. Farrell. 2009. “A Dynamic Adaptive Chemistry Scheme for Reactive-Flow Computations.” Proceedings of the Combustion Institute, Vol. 32, pp. 527- 534. Lu, T.F., and C.K. Law. 2009. “Toward Accommodating Realistic Chemistry in Large Scale Computations.” Progress in Energy and Combustion Science, Vol. 35, pp. 192-215. McBride, B.M., S. Gordon, and M.A. Reno. 1993. “NASA Thermodynamics Database.” NASA Technical Memorandum 4513. Cleveland, Ohio: Lewis Research Center, NASA. Muller, C., V. Michel, G. Scacchi, and G.M. Come. 1995. “THERGAS: A Computer Program for the Evaluation of Thermochemical Data of Molecules and Free Radicals in the Gas Phase.” Journal of Chemical Physics, Vol. 92, pp. 89, 1154-1178. Peters, N. 1984. “Laminar Diffusion Flamelets in Non-Premixed Turbulent Combustion.” Progress in Energy Combustion Science, pp. 319-339. Peters, N. 2000. Turbulent Combustion. Cambridge, United Kingdom: Cambridge University Press. Poinsot, T., and D. Veynante. 2005. Theoretical and Numerical Combustion. Philadelphia, Pa.: R.T. Edwards. Pope, S.B. 1997. “Computationally Efficient Implementation of Combustion Chemistry Us - ing in Situ Adaptive Tabulation.” Combustion Theory Modelling, Vol. 1, pp. 41-63. Prager, J., H.N. Najm, M. Valorani, and D.A. Goussis. 2009. “Skeletal Mechanism Generation with CSP and Validation for Premixed n-heptane Flames.” Proceedings of the Combustion Institute, Vol. 32, pp. 509-518. Puduppakkam, K.V., L. Liang, C.V. Naik, E. Meeks, and B. Bunting, 2009. Modeling of a Gasoline HCCI Engine Using Model Fuels. Society of Automotive Engineers Publication SAE-2009-01-0669. Warrendale, Pa.: Society of Automotive Engineers. Ritter, E.R., and J.W. Bozzelli. 1991. “THERM: Thermodynamic Property Estimation for Gas-Phase Radicals and Molecules.” International Journal of Chemical Kinetics, Vol. 23, pp. 767-778. Ruscic, B., D. Feller, D.A. Dixon, K.A. Peterson, L.B. Harding, R.L. Asher, and A.F. Wagner. 2001. “Evidence for a Lower Enthalpy of Formation of Hydroxyl Radical and a Lower Gas-Phase Bond Dissociation Energy of Water.” Journal of Physical Chemistry A. 105(1):1-4. Ruscic, B., J.E. Boggs, A. Burcat, A.G. Scaszar, J. Demaison, R. Janoschek, J.M.L. Martin, M.L. Morton, M.J. Rossi, J.F. Stanton, P.G. Szalay, P.R. Westmoreland, F. Zabel, and T. Berces. 2005. “IUPAC Critical Evaluation of Thermochemical Properties of Select - ed Radicals, Part I.” Journal of Physical Chemistry Reference Data, Vol. 34, pp. 573-656.

OCR for page 41
67 COMBUSTION AND CYBERINFRASTRUCTURE U.S. EIA (U.S. Energy Information Administration). 2007. U.S. Crude Oil, Natural Gas, and Natural Gas Liquids Reserves, 2006 Annual Report. DOE/EIA-0216. Washington, D.C. November. U.S. Global Change Research Program. 2009. Global Climate Change Impacts in the United States. New York, N.Y.: Cambridge University Press. Viola, A., and G.A. Voth. 2005. “A Multi-Scale Computational Approach for Nanoparticle Growth in Combustion Environments,” in High Performance Computing and Communica- tions, Vol. 3726, pp. 938-947, Book Series: Lecture Notes in Computer Science, L.T. Yang et al. (eds.). Berlin/Heidelberg: Springer. Westbrook, C.K., Y. Mizobuchi, T. Poinsot, P.A. Smith, and J. Warnatz. 2005. “Computational Combustion.” Proceedings of the Combustion Institute, Vol. 30, pp. 125-157. Westbrook, C.K.,W.J. Pitz, O. Herbinet, H.J. Curran, and E.J. Silke. 2009. “ A Comprehensive Detailed Chemical Kinetic Mechanism for Combustion of n-Alkane Hydrocarbons from n-Octane to n-Hexadecane.” Combustion and Flame, Vol. 156(1), pp. 181-199.