32.5M hours of supercomputer time to aid GM, Ford engine projects with Oak Ridge Lab
6 August 2014
|Simulation of injector. Graphic from GM, Edwards AMR 2014 presentation. Click to enlarge.|
As part of its 2014 ASCR Leadership Computing Challenge (ALCC) awards of processor time (totaling more than 3 billion processor hours), the US Department of Energy’s (DOE) Office of Science has awarded 15 million hours on Oak Ridge National Laboratory’s (ORNL) Titan supercomputer to a project led by General Motors, and 17.5 million hours on Titan to a project led by ORNL with Ford and Convergent Science as co-investigators. Titan is current Nº2 on the Top 500 Supercomputer list, and offers 27.1 petaflop (PF) peak processing capacity, with about 300,000 compute cores.
The two projects are part of a larger multi-year DOE-funded project to develop and to apply innovative simulation strategies and tools to maximize benefits of predictive information from high performance computing (HPC) for internal combustion engines. The Principal Investigator on that DOE project is Dean Edwards of ORNL.
The umbrella projects addresses specific technology barriers identified by DOE and industry stakeholders; Ford and GM had expressed interest in working with Oak Ridge in these areas, and GE came later and was worked in under the same project, notes Edwards. CFD software developer Convergent Science (“Never make a mesh again”) is a partner on all three of these efforts. The projects include both open and proprietary aspects under the Oak Ridge Leadership Computing Facility (OLCF) User Facility Agreement.
The projects started in May 2012 and are ongoing; the new ALCC awards are the latest awards of processor time supporting the project. (Titan’s CPU time is valued at $0.03/h; hence the two awards combined are worth about $975,000.)
The DOE’s ALCC’s mission is to allocate supercomputer time for projects of interest to the Department of Energy (DOE) with an emphasis on high-risk, high-payoff simulations in areas directly related to the DOE mission and for broadening the community of researchers capable of using leadership computing resources.
GM/ORNL. The project led by Tang-Wei Kuo from General Motors is focused on optimizing multi-hole injectors for spark-ignited direct-injection (SIDI) gasoline engines. Co-investigators are Ronald Grover (GM), Sreekanth Pannala (ORNL), and Wael Elwasif (ORNL).
Recently, the most popular SIDI gasoline injectors have been multi-hole type. Multi-hole sprays offer the flexibility of manufacturing the nozzle holes at various orientations to engineer a variety of spray patterns. This flexibility offers a high degree of freedom for design and engine manufacturers need to balance the cost of hardware testing with turn-around time and practicality for mass production. However, optimization is challenging.
The current design optimization process is very time- and labor-intensive. As a result, it is not possible to fully investigate the numerous geometry and operating parameters and truly optimize the injector design for best fuel efficiency.—Sreekanth Pannala, ORNL Computer Science and Mathematics Division
Detailed analytical tools, such as computational fluid dynamics, become attractive to reduce the number of probable injector concepts for a given combustion system. The Titan supercomputer project supports the simulations of multi-hole injector designs for SIDI engines. HPC enables a thorough and rapid investigation of operational and geometric design spaces, reducing the time required from months to weeks and providing more thorough coverage of design space.
ORNL is developing a computational framework to automate model generation and design optimization. In this, it is leveraging its experience with automotive batteries (CAEBAT, earlier post) and fusion (IPS for ITER).
|Optimizing injector design. Source: Edwards, AMR 2014 presentation. Click to enlarge.|
ORNL/Ford/Convergent. This project, led by Charles Finney from ORNL, is simulating cyclic variability in dilute internal combustion engine operation to develop a method that identifies combustion instability.
Some strategies to reduce emissions include running the engine highly dilute, with excess air or with recirculated exhaust gases as part of the combustion charge. However, increased dilution leads to increased combustion instability resulting in reduced fuel-economy benefits. This limits the practical range of operation in real-world applications.
This project supports computational high-dimensional fluid dynamics simulations to explore the nature and causes of cyclic variability under highly dilute engine cycles. Through methods development in parallelization and intelligent parameter space sampling, the methodology aims to enhance the utility of complex CFD simulations for studying the dynamics of cycle-to-cycle variations in unstable dilute combustion operation.
In the past, researchers had to rely on highly simplified engine models to simulate the thousands of consecutive engine cycles needed to study the cycle-to-cycle variability with statistical accuracy.—Dean Edwards, ORNL
The researchers are creating low-order metamodels of deterministic response using multi-dimensional mapping of a CFD model’s response at sample points, and leveraging ORNL’s TASMANIAN (Toolkit for Adaptive Stochastic Modeling And Non-Intrusive ApproximatioN) algorithm. They have already successfully demonstrated the approach using a simple spark ignition model with cycle-to-cycle feedback.
Co-investigators include Sreekanth Pannala (ORNL); Miroslav K. Stoyanov (ORNL); Brad VanDerWege (Ford Motor Company); Daniel Lee (Convergent Science, Inc); Eric Pomraning (Convergent Science, Inc); Keith Richards (Convergent Science, Inc); and P. Kelly Senecal (Convergent Science, Inc).
ALCC. For 2014, 42 awards were made, supporting scientific and technological research in fields spanning Bioenergy, Engine Efficiency, Turbomachinery Design, Fusion Energy Sciences, High Energy Physics, Materials Science, Nuclear Physics, Nuclear Reactor Safety, Climate Modeling, and Seismology. In addition to the two noted above, other awards included:
|Select 2014 ALCC awards|
|J. C. Oefelein, Sandia National Laboratories
G. Lacaze (SNL), R. Dahms (SNL), Christopher Stone (Computational Science and Engineering), R. L. Davis (UC Davis), R.Sankaran (ORNL)
|Development of High-Fidelity Multiphase Combustion Models for Large Eddy Simulation of Advanced Engine Systems
The importance of understanding liquid-fuel injection and multiphase combustion processes in state-of-the-art transportation, propulsion, and power systems (e.g., reciprocating and gas- turbine internal-combustion engines) are widely recognized. Injection of liquid fuels largely determines fuel-air mixture formation, which governs the detailed evolution of chemical kinetics, combustion, and emissions. A lack of accurate models is a major barrier toward the design of advanced engine systems that are clean and highly efficient, and there is a critical need for advanced development in this area.
The objective of this project is to perform fundamental inquiries into the structure and dynamics of turbulent combustion processes that are dominated by high-pressure, high-Reynolds-number, multiphase flows at device relevant conditions. The simulations will be directly coupled to a set of companion experiments being performed at Sandia National Laboratories, Combustion Research Facility. The results will provide scientific advances required for improved predictive models for combustion design.
|K. Schulten, University of Illinois at Urbana-Champaign
R. C. Bernardi (University of Illinois)
|Molecular Dynamics Studies of Biomass Degradation in Biofuel Production
A critical bottleneck for the production of second-generation biofuel is the depolymerization of plant biomass. The breakdown of plant biomass is not straightforward because plants have evolved complex structural and chemical linkages that are highly resistant to degradation by the microbial enzymes used to release the sugar units for further ethanol production. Advanced computational tools make it feasible to address biomolecular-related problems associated with biofuel production through simulations of the enzymatic systems that act on plant fibers.
To address the challenge of producing second-generation biofuels, this project supports a joint computational and experimental approach to study major problems of the second-generation biofuel industry. The outcomes will be a deeper molecular understanding of challenges and solutions for development of second-generation biofuels and advancement along the path towards clean energy.
|G. Voth, U. Chicago / Argonne National Laboratory
C. Knight (ANL)
|Influence of Morphology on Proton Transport in Proton Exchange Membranes
The successful development of low-cost, high-performance electrochemical devices would have an important impact in the clean energy market. To date, the rate-limiting step in the optimization and design of new materials for such systems has been the use of intuition based “guess and check” strategies. The efficiency of this strategy can be significantly improved with the incorporation of a detailed fundamental understanding of the associated transport processes based on the results of combining experimental characterization and molecular modeling.
The project utilizes a combination of novel molecular simulation algorithms developed by this team to improve the understanding of fundamental processes that govern charge transport in polymer electrolyte membranes commonly used in fuel cells. The ability to satisfy all criteria for realistic calculations on these important systems will be accomplished by the coupling of accurate reactive dynamics models, a highly efficient simulation software package and parallelization strategies specifically tailored to multistate simulations. The outcome of the project will be a significant step forward on answering fundamental questions regarding proton transport in fuel cell membranes and stimulating the synergy of theoretical and experimental characterizations of these important systems.
ALCC grants one year awards. In 2013, 39 ALCC awards were granted, amounting to 1.8 billion processor hours with project sizes ranging from 3 to 250 million process hours.
K. Dean Edwards (2014) “Accelerating predictive simulation of IC engines with high performance computing (ACE017)” DOE AMR 2014 presentation
TrackBack URL for this entry:
Listed below are links to weblogs that reference 32.5M hours of supercomputer time to aid GM, Ford engine projects with Oak Ridge Lab: