4

Computational Capabilities and Requirements

BACKGROUND

As in most advanced technical facilities, the computer plays a multifaceted role in the operation of the Arnold Engineering Development Center (AEDC). The role of the computer is generally well understood in its use in controlling tests and test design, as well as in data reduction and data analysis. Understanding the internal flow in the facilities through computational fluid dynamics (CFD) and, consequently, the improvement of that flow and the design of new facilities are also well-accepted uses of computational analysis. As test objectives become more demanding, it is important to understand the predicted flow over the model, especially for the purpose of instrumentation design and location. Chemically reacting flows, due to combustion and high stagnation temperature dissociation, are very complex and require a sophisticated analysis capability to evaluate test data. The requirements for computing power and central memory are driven primarily by the need to provide state-of-the-art CFD capability at AEDC.

TEST SUPPORT AND MANAGEMENT

The computing facility existing at AEDC at the time of the committee 's discussions (1991) was based on a 1-MW (million words) CRAY 1S, a 2-MW CRAY XMP-12, and remote access to off-site supercomputers via T1 land links. The first function of supporting computing services at AEDC derives from the need of test support and management to operate a modern data-intensive ground test facility, to perform data reduction (including test corrections), facility design (facility flow analysis), instrumentation, and load analysis. Because these requirements have outgrown AEDC's current computing capabilities, a new plan--the Advanced Scientific Computing Enhancement Project--was initiated in 1991 to upgrade the computing facility. The proposed plan provides four computers, one for each of the major test areas. Each computer has 31 MW of central physical memory and a minimum single-user central processing unit (CPU) capability of 35 MFLOP (millions of floating point computer operations per second) on a 100 by 100 all-FORTRAN LINPACK Benchmark. These four computers should be adequate to provide test support and management and will be an adjunct to a large memory, high throughput



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 43
Future Aerospace Ground Test Facility Requirements for the Arnold Engineering Development Center 4 Computational Capabilities and Requirements BACKGROUND As in most advanced technical facilities, the computer plays a multifaceted role in the operation of the Arnold Engineering Development Center (AEDC). The role of the computer is generally well understood in its use in controlling tests and test design, as well as in data reduction and data analysis. Understanding the internal flow in the facilities through computational fluid dynamics (CFD) and, consequently, the improvement of that flow and the design of new facilities are also well-accepted uses of computational analysis. As test objectives become more demanding, it is important to understand the predicted flow over the model, especially for the purpose of instrumentation design and location. Chemically reacting flows, due to combustion and high stagnation temperature dissociation, are very complex and require a sophisticated analysis capability to evaluate test data. The requirements for computing power and central memory are driven primarily by the need to provide state-of-the-art CFD capability at AEDC. TEST SUPPORT AND MANAGEMENT The computing facility existing at AEDC at the time of the committee 's discussions (1991) was based on a 1-MW (million words) CRAY 1S, a 2-MW CRAY XMP-12, and remote access to off-site supercomputers via T1 land links. The first function of supporting computing services at AEDC derives from the need of test support and management to operate a modern data-intensive ground test facility, to perform data reduction (including test corrections), facility design (facility flow analysis), instrumentation, and load analysis. Because these requirements have outgrown AEDC's current computing capabilities, a new plan--the Advanced Scientific Computing Enhancement Project--was initiated in 1991 to upgrade the computing facility. The proposed plan provides four computers, one for each of the major test areas. Each computer has 31 MW of central physical memory and a minimum single-user central processing unit (CPU) capability of 35 MFLOP (millions of floating point computer operations per second) on a 100 by 100 all-FORTRAN LINPACK Benchmark. These four computers should be adequate to provide test support and management and will be an adjunct to a large memory, high throughput

OCR for page 43
Future Aerospace Ground Test Facility Requirements for the Arnold Engineering Development Center supercomputer needed for CFD that is discussed in the CFD Computing Facilities section. Graphic workstations should be acquired in sufficient numbers to prevent postprocessing of experimental and computational data from becoming a bottleneck and reducing productivity. Graphical methods have become the most effective way of extracting useful information from complex three-dimensional data sets from steady and unsteady flows. These computer infrastructure capabilities are also important for the support of CFD activities at AEDC. The committee recommends that AEDC continue the implementation of computer enhancements required to ensure state-of-the-art capabilities in tunnel control, data reduction and analysis, instrumentation, and facility flow analysis. CURRENT STATUS OF APPLIED COMPUTATIONAL FLUID DYNAMICS (CFD) Computational simulation has emerged as one of the major tools, along with theory and experiment, for the development of technology, the study of science, and the development of new products and materials. In the aerospace industry, computational fluid dynamics is widely recognized as a practical tool for development of aerospace systems, even though there are still many unsolved problems and shortcomings in CFD technology. As noted, CFD can contribute much to a ground test facility such as AEDC through pretest analysis; post-test data analysis and interpretation; and facility design, development, and calibration. The use of CFD and ground testing is highly synergistic in that the combination of CFD and wind tunnel testing together can generate more reliable information and can do it faster and cheaper than either alone. An example of CFD's application to ground testing is AEDC's innovative use of CFD analysis and wind tunnel testing to demonstrate safe release of stores from aircraft. For a comprehensive, if somewhat dated, overview of CFD's capabilities the reader is referred to a 1983 report by the National Research Council (NRC, 1983). More recent overviews are provided by Peterson, et al. (1989), Rubbert (1990), and Holst (1991). There are many flow models available for the physical description of an aerodynamic flow. These include the Euler equations for inviscid flow, the Reynolds-averaged Navier-Stokes (RANS) equations, large eddy simulations (LES), and direct numerical simulations (DNS) of turbulent flow. Other models include the Burnett and Boltzman equations for low-density flows. The choice of a particular flow model depends on the demands for accuracy and the completeness required for the proper physical description of the problem at hand, as well as on the computing power available for the numerical solution. The AEDC computing power available today is adequate for routine application of the exact inviscid Euler and laminar viscous Navier-Stokes model for steady aerodynamic flow over complete aircraft configurations. Unfortunately, these descriptions apply to only a very small set of flow problems arising in the aerospace industry and at AEDC. Most of the aerodynamic flows of interest to AEDC and its customers require consideration of one or more of the following phenomena: turbulence, flow separation, vortices, unsteadiness, and chemical reactions. One of the major goals of CFD at AEDC is numerical simulation of the aerodynamic flow over complete aircraft configurations and within rocket and turbine engines using the full RANS equations. The RANS equations provide the most complete and accurate description of fluid

OCR for page 43
Future Aerospace Ground Test Facility Requirements for the Arnold Engineering Development Center flows within the continuum region, subject to the limitations of available turbulence models. A RANS description is required to span the full range of flow conditions encountered in aircraft and engine applications. It is also required to adequately predict flows with separation, stall, buffet, and maximum aerodynamic loads, and to otherwise completely cover conditions over the entire performance envelope of aircraft and propulsive systems. Although much progress has been made toward meeting this goal over the last five years, it is important to recognize that RANS simulations remain in the research stage and have not yet emerged as a practical tool for aircraft and engine design (Peterson, 1989; Rubbert, 1990; Holst, 1991). The major problem areas limiting the usefulness of RANS simulations for practical application are: computing power, turbulence modeling, geometry and grid generation, validation and refinement, and cost. To get some feeling for computing requirements, we note that it currently takes about 5 to 10 hours and 50 to 100 MW of memory to do a RANS simulation of a complete aircraft configuration using 1 million grid points, on a single processor CRAY YMP which has a peak speed of about 300 MFLOP. To have practical utility, RANS methods must be augmented with turbulence models that are effective for the particular problem under study. Existing turbulence models are useful for two-dimensional and simple three-dimensional streamlined flows. They are inadequate for complex three-dimensional flows, separated flows, vortex flows and turbulent chemically reacting flows, to name but a few areas in which significant limitations exist. However, promising results have recently been presented for complex three-dimensional and separated flows using the simple Johnson-King one-half equation model and improved k-∊ turbulence models. The initial stage of a CFD computation involves the definition of a discretized geometry and the generation of an appropriate grid. These are very labor-intensive and time-consuming tasks that can take from weeks to months to complete for complex geometric configurations. In addition, the need to deal with multiple small-scale flow features in many applications is a difficult problem that limits the accuracy of existing CFD codes. At present, we can effectively deal with problems that have at most two length scales, for example, a boundary layer and an outer inviscid flow. A solution adaptive grid for complex three-dimensional flows would appear to be the most promising approach for treating these problems. The issue of code validation and refinement continues to be a vexing problem that limits the acceptance of CFD technology by the design community. For designers to commit to CFD they must be certain that the information it provides is accurate and reliable. Since all CFD

OCR for page 43
Future Aerospace Ground Test Facility Requirements for the Arnold Engineering Development Center capability is subject to some limitations in one form or another, the designer must know what the limits of the capability are--where CFD can be trusted and where it cannot. To be accepted for industrial applications, CFD must be affordable; it must be competitive with other means of generating the information, or it must provide a unique capability. Although RANS-based CFD has some way to go before it can be accepted as a practical engineering tool, it is clearly a very useful capability that can effectively support AEDC's charter in aerospace ground testing. Although aerodynamic CFD applications are very important as an adjunct to ground testing, CFD applications in support of aeropropulsion work are more difficult and of particularly high value. In general, AEDC acquires CFD technology from third parties and invests some resources to adapt the technology to its specific requirements. On occasion (e.g., plumes), AEDC will also develop a CFD capability if it is needed and not available from other sources. CFD SUPPORT AT AEDC Computational fluid dynamics is a key supporting technology at AEDC and is a service that is expected by its customers. As described above, it is important for facility design as well as for pretest planning and post-test data analysis. AEDC recognizes the important role of CFD in ground testing and is committed to maintaining a state-of-the-art CFD capability. The committee finds that CFD based on the RANS equations is an emerging technology that will have increasing importance in ground testing applications. However, to achieve this potential at AEDC, RANS-based CFD technology requires advances in chemically reacting flows, turbulence modeling, and in geometry and grid generation. Chemically reacting flows are of obvious importance to AEDC. They arise in many problem areas that are central to the AEDC mission, including but not limited to combustion, liquid and solid rocket stability, high-enthalpy facilities and diagnostics, exhaust plume signatures, hypersonics, and high-speed civil transport emissions. The emergence of teraFLOP computers expected in this decade will greatly expand CFD's ability to compute complex chemically reacting flows. Existing turbulence models are inadequate for complex three-dimensional flows, separated flows, or flows with turbulent reactive chemistry. Improvements are needed in all these areas if RANS methods are to become a general and useful tool for applications at AEDC. The task of defining a discretized geometry and constructing a computational grid is a labor-intensive, time-consuming, and costly endeavor. Automated methods are needed to reduce the time and cost of a CFD simulation. The problem of numerically resolving multiple small-scale flow features continues to be a factor limiting the utility of CFD in many three-dimensional applications. Solution-adaptive grid generation is a promising approach for addressing this class of problems, although much work remains to be done in extending the technology to complex three-dimensional flows. AEDC should consider expanding its on-site efforts in the application and understanding of turbulence modeling and in numerical simulations of chemically reacting flows. A small investment in these areas would serve to keep AEDC at the state of the art and could substantially add to its CFD capability. However, the development of improved geometry and grid generation methods is a major task requiring substantial investments. AEDC should not

OCR for page 43
Future Aerospace Ground Test Facility Requirements for the Arnold Engineering Development Center attempt to do this on-site but should look to acquire the technology from third parties. AEDC is encouraged to continue its ongoing cooperative efforts with industrial, academic, and NASA consortia. CFD COMPUTING FACILITIES Although the committee believes that the Advanced Scientific Computing Enhancement Program plan is adequate to meet the operational test support and management requirements, it is not, however, adequate to perform state-of-the-art CFD. Therefore, it would be desirable for AEDC to acquire a supercomputer with sufficient memory and throughput to perform this function. As discussed previously, about 50 to 100 MW are required to carry out a complete aircraft calculation on a medium grid of 1 million points. It is important to note that a solution cannot be contemplated if sufficient memory is not available. According to Dongarra (1991) the fastest single processor computer could do the 100 by 100 all-FORTRAN LINPACK Benchmark with about 160 MFLOP compared to the minimum requirement of 35 MFLOP called out in the Advanced Scientific Computing Enhancement Program plan. AEDC should plan an upgrade of its computational facility to allow for applications of state-of-the-art numerical RANS simulations and chemically reacting flows. The availability and cost of supercomputers suggests that AEDC should consider the acquisition of one large supercomputer that is sized for large CFD applications. The large machine should have a minimum central processing unit (CPU) performance of at least 160 MFLOP on the 100 by 100 all-FORTRAN LINPACK Benchmark and a minimum 128 MW of central memory. AEDC should start planning for the acquisition of massively parallel computers. Practical massively parallel computers, which will likely be available sometime in this decade, offer the prospect of teraFLOP computing performance. The availability of this level of performance will revolutionize RANS to full aircraft configuration and to full three-dimensional propulsive flows including nonequilibrium chemistry. The committee recommends that AEDC plan to acquire a supercomputer with at least 128 MW and a minimum CPU performance of 160 MFLOP on the 100 by 100 all-FORTRAN LINPACK Benchmark in order to support a full three-dimensional CFD applications capability. The committee recommends that key personnel be added to the AEDC staff to enhance the understanding and application of turbulence models and chemically reacting flow CFD models.