Applied Fluid Dynamics
NASA's Numerical Propulsion Simulation System (NPSS) focuses on reducing both the cost and the time to develop aeropropulsion systems. NPSS will build a simulation environment that allows for the construction of arbitrary engine configurations for design and analysis. The environment will permit a choice of analysis techniques, analysis complexity, languages with which to describe the algorithms, and the ability to access and manage data from various sources.
NPSS MOD1 was released in 1994. It employs an object-based model for engine simulations: engine components, such as the compressor, combustor, turbine, and shaft, can be modeled as independent entities that can be replaced with models of greater fidelity that execute on different computing platforms in a dynamic environment. An engineer can configure engines with ease using this environment and a graphical user interface.
This work is being conducted at NASA Lewis Research Center in Cleveland, OH. Researchers at Lewis are working with the U.S. aeropropulsion industry to define a U.S. standard for select design codes.
One component of NPSS is the MSTAGE multistage compressor analysis software that has been implemented on the Cray Research C90, IBM SP-1, and IBM workstation cluster at NASA Lewis. This software has been used to analyze the flow physics involved in compressor stall that has suggested several approaches to improving compression system performance and increasing stall margins. This work is part of a NASA/MIT/Pratt and Whitney effort. A 1.5 percent reduction in Specific Fuel Consumption (SFC) for a large commercial aircraft engine was demonstrated at Pratt and Whitney; this result was achieved in half the historical design time.
Other NASA accomplishments include the installation of a 160-node IBM SP-2, located at NASA Ames Research Center, which performs at a sustained rate of 11.64 Gflops in NASA Parallel Benchmark codes; this was accomplished through a Cooperative Research Agreement, a new method of doing business for NASA. Additionally, the Framework for Interdisciplinary Design and Optimization (FIDO) project, located at NASA Langley Research Center, is developing a general computational environment for performing multidisciplinary design using networked heterogeneous computing systems. Such a graphical interface provides easier understanding and access to data than previous text-based methods, and requires less training for users.
http://cesdis.gsfc.nasa.gov/hpccm/accomp/94accomp/cas94.accomps/cas4.html
Analysis to define the flow physics involved in compressor stall. It suggested a variety of approaches to improve the performance of compression systems, while providing increased stall margins. A Cray Research C-90, IBM SP-1, and IBM workstation cluster were used to formulate and develop this model.
Coupled Field Problems and GAFD (Geophysical and Astrophysical Fluid Dynamics) Turbulence
This project represents engineering and scientific Grand Challenges linked through common enabling computer science advancements: High Performance Computational Methods for Coupled Field Problems and Coherent Structures and Dynamics in Geophysical and Astrophysical Turbulent Flows. The former addresses three engineering problems that exhibit strong interaction between components modeled by different disciplines: (1) aeroelasticity of a complete aircraft, (2) distributed control of flexible structures, and (3) coupling of electromagnetic, thermal, and superconducting phase-change behavior. The latter addresses four problems in oceans, atmospheres, and stars in which turbulence coexists with large-scale coherent structures and mean flows: (1) geostrophic turbulence, (2) ocean convection, (3) deep convection in planetary atmospheres, and (4) compressible convection in stars like the sun, all constrained by the effects of rotation and stratification. Both efforts are enabled by advances in networking, algorithm development, programming environments, and performance prediction and analysis.
Aeroelasticity studies the mutual interaction between aerodynamic and elastic forces for an aerospace vehicle. A flexible aircraft structure immersed in a flow is subjected to surface pressures induced by that flow. Moreover, structural dynamic motions induced by these pressures in turn change the boundary conditions of the flow. The accurate prediction of aeroelastic phenomena requires extensive computation to solve simultaneously the coupled fluid and structural equations of motion. For example, the aeroelastic response of a detailed wing-body configuration using potential flow theory requires about five CPU hours on a Cray Computer Cray-2. To establish the transonic flutter boundary for a given set of aeroelastic parameters, about 30 aeroelastic response analyses are typically required, which brings the total CPU time to six days. The aeroelastic simulation of a complete aircraft (shown above) is made possible by the new generation of massively parallel computing systems as well as by methods developed under this Grand Challenge. This simulation required the simultaneous solution of 463,674 nonlinear fluid equations and 45,108 linear structural equations, thousands of times. These computations were carried out in a heterogeneous mode on a 128-processor Cray Research T3D, a 128- processor IBM SP-2, and a 128-processor Intel Paragon XP/S.
In another effort, a billion-zone (1,024-by-1,024 by 1,024) turbulence problem was solved on an array of 16 Silicon Graphics Challenge XL workstations with 28 GB total system memory. The computation required a week and achieved a sustained computational rate of 4.9 Gflops. The workstations were configured as a 2-by-2- by-4 toroidal array using FDDI networking. This computation demonstrated the high performance of a networked cluster of shared memory multiple-processor machines.
This team is also installing their software on systems from different vendors, including the Cray Research T3D. In these efforts they are early users of implementations of High Performance Fortran (HPF) compilers, participate in the Message Passing Interface Forum, and develop performance analysis tools. Participants in this NSF- funded project include the University of Colorado, NCAR, and the University of Minnesota.
http://lcd-www.colorado.edu/Blue_book/bb_joint.html
An image from a video illustrating the flutter analysis of a FALCON jet under a sequence of transonic speed maneuvers. Areas of high stress are red; areas of low stress are blue.
Combustion Modeling: Adaptive Grid Methods
Combustion is a major source of energy, plays a dominant role in transportation, and is an important factor in many industrial processes. Requirements for energy efficiency and for emission reduction have led industry to increased use of computer simulations to design combustion devices. One such device is the pulse combustor, which is characterized by a periodic combustion process. Available simulation software has been limited in its ability to represent the detailed physical processes and the complex geometries of such practical engineering devices. This Grand Challenge team is developing parallel software that models fully three-dimensional fluid dynamics in the combustion chamber and incorporates lower-dimensional approximations for the inlet valve that controls the injection of fuel and the tail pipe whose acoustic properties control the cyclic behavior of the device.
Participants in this project include researchers at DOE's Lawrence Livermore and Los Alamos National Laboratories, the Courant Institute for the Mathematical Sciences, and the University of California at Berkeley. They are collaborating with Coen Co. of Burlingame, CA, and Babcock and Wilcox of Alliance, OH, in extending and validating their numerical methods to simulate low-NOx natural gas burners. They also participate in benchmarking software for modeling natural gas burners sponsored by the Gas Research Institute that is funded by the gas production industry.
http://www.nersc.gov/doc/Comp_Research/CCSE/bb96.html
Fuel flow around the stagnation plate in a pulse combustor. A burning cycle drives a resonant pressure wave, which in turn enhances the rate of combustion, resulting in a self-sustaining, large-scale oscillation. The figure shows the injection phase when the pressure in the combustion chamber is low. Fuel enters the chamber, hits the stagnation plate and becomes entrained by a vortex ring formed by flow separation at the edge of the splash plate. Researchers are developing computational models to study the interplay of vortex dynamics and chemical kinetics and will use their results to improve pulse combustor design.
Oil Reservoir Modeling: Parallel Algorithms for Modeling Flow in Permeable Media
This DOE-funded Grand Challenge has the goal of using high performance parallel processing to simulate (1) the behavior of petroleum reservoirs to enhance oil recovery, and (2) groundwater aquifers to aid removal of contaminants from fresh water aquifers. Improved understanding of petroleum reservoirs leads to better reservoir management and more efficient U.S. oil and gas production. Numerical modeling of fluid flow in permeable media is critical to the management and protection of groundwater supplies. These two problems have common scientific and engineering fundamentals involving multiphase transport in permeable media.
Using current scalar and vector computing systems, most industrial reservoir engineering codes use fewer than 100,000 gridblocks and model fewer than 1,000 wells. Key geological, physical, and chemical features are only crudely approximated. Preliminary studies suggest that a million gridblocks are feasible with parallel computing technologies, enabling more accurate and more useful results.
This work is being conducted by interdisciplinary teams in applied mathematics, computational science, and chemical, petroleum, and environmental engineering at the University of Texas at Austin, Rice University, and industrial affiliates. Participants have developed accurate and efficient serial and parallel numerical algorithms for solving linear and nonlinear, coupled partial differential equations, including advective dominated transport equations and elliptic/parabolic flow equations; have developed efficient parallel domain decomposition solvers for the large sparse linear systems that result from temporal and spatial discretizations; and work with computational science applications in modeling subsurface multiphase flow and multi-component reactive transport, surface water, and root-soil systems, and the interpretation of microscopic data on macroscopic scales. The reservoir simulation study was used to evaluate the use of horizontal wells with vertical drainholes for tertiary oil recovery using carbon dioxide.
http://www.pe.utexas.edu/HPCC/hpcc.html
Numerical Tokamak Project (NTP)
The goal of DOE's NTP Grand Challenge is to develop and integrate particle and fluid plasma models on large-scale parallel computing systems as part of a multidisciplinary study of Tokamak fusion reactors. The kinetic particle simulations and the fluid simulations have different physics and computational attributes and advantages, and comparisons of their results enable calibration and lead to improvements in each and to the development of hybrid models that embody aspects of both.
Accomplishments include (1) the development and optimization of a suite of fluid and kinetic three-dimensional simulation codes for massively parallel computing systems, (2) testing and comparison of various MPP (massively parallel processing) systems and programming paradigms and studies of MPP performance scaling with the number of processors, (3) improvement of code performance by a factor of 10 to 100 with careful MPP optimization and by an additional factor of 10 to 100 with optimized spatial grids, (4) development of advanced perturbative, implicit, and hybrid algorithms to improve simulation efficiency further, (5) development of new tools to handle storage and retrieval of large data sets and to post-process and visualize the data interactively on distributed communications networks, (6) building portable code and code modules for MPPs and clusters, class libraries, parallel fast Fourier transforms and elliptic matrix solvers, convolution routines, and parallel I/O using NetCDF software.
These accomplishments have enabled or accelerated (1) larger, more efficient three-dimensional toroidal gyrofluid and gyrokinetic simulations relevant to Tokamak experiments using a Thinking Machines CM-5 and a Cray Research C90, (2) code comparisons leading to closer agreement between gyrokinetic and gyrofluid simulations of a Princeton Tokamak Fusion Test Reactor (TFTR) test case and studies of the scaling of the turbulent transport with respect to the physical parameters of interest and the isotopic composition of the plasma, (3) advances in simulated thermal diffusivities, fluctuation spectra, and parameter scaling agreeing more closely with experiments (for example TFTR discharges), (4) quantitative determination of the importance of nonlinear self- generated sheared flows in influencing turbulence, (5) progress in simulations of ion-temperature-gradient, trapped electron, and alpha-particle-driven toroidal Alfven eigenmode instabilities, and (6) inclusion of more physics in the simulation (impurities, collisions, kinetic electrons, velocity shear, toroidicity, curvature drives and resonances, and magnetically trapped particles).
Algorithmic improvements and use of large-scale parallel computing systems result in 100- to million-fold performance improvements. Information processing and visualization tools accelerate comparison of computational models to each other, to experimental data, and to analytic theory, enabling better understanding of the target physics.
This work is being conducted at NERSC, LLNL, LANL, and PSC.
http://www.acl.lanl.gov/GrandChal/Tok/BlueBook.html
Particle trajectories and electrostatic potentials from a three- dimensional implicit tokamak plasma simulation employing adaptive mesh techniques. The boundary is aligned with the magnetic field that shears around the torus. The strip in the torus is aligned with the local magnetic field and is color mapped with the local electrostatic potential. The yellow trajectory is the gyrating orbit of a single ion.