Dennis Assanis
Vehicle manufacturers are finding themselves in a quandary: the 1990 amendments to the Clean Air Act require considerable reductions of pollutant-forming emissions by the year 2000, and there is ever-increasing pressure to improve fuel efficiency. At the same time, competitive market pressures demand a faster turnaround time for new designs than ever before. Fortunately, vehicle manufacturers have some new tools to help them keep pace with the demands of the environment and the market. One of them is engine modeling, or the creation of a "numerical engine"--allowing manufacturers to test new engine components on computer before creating expensive prototypes.
Dennis Assanis, associate professor in the UIUC Department of Mechanical and Industrial Engineering, has been performing pioneering work modeling turbulent flows in engines on NCSA's high-performance computers. He spent part of his 1991-92 sabbatical at NCSA, working to improve engine modeling techniques, and part of it traveling to major vehicle manufacturers worldwide. Assanis wanted to tailor his research efforts to the problems and challenges faced by the automotive industry.
Dennis Assanis
One of the most powerful multi-dimensional engine simulation codes is KIVA and its offshoot, KIVA-II, developed by the Los Alamos National Laboratory (LANL). Although KIVA-II, a vectorized code for Cray supercomputer systems, can deal with many of the phenomena occurring in internal combustion engines, Assanis and his students at UIUC found that it does not adequately predict some aspects of fuel spray dynamics and pollutant formation. Their improvements are being incorporated into a new version of KIVA called KIVA-3. Unlike KIVA-II, KIVA-3 can handle both the intake and exhaust flows, allowing for the simulation of an entire cycle.
Assanis is currently working on improving KIVA-3 to model direct injection of natural gas into the combustion chamber. In the near future, he hopes to collaborate with experimentalists to validate code predictions. "Lately, people have started to realize that natural gas is a fuel with great promise in terms of reducing pollution and improving fuel econ-omy," says Assanis. "This presents a new challenge. We would like to be able to provide some guidance to industry on how to improve the design of gaseous fuel injectors, how to position and target them, where to place the glow plug (an ignition assist device), and how to improve the mixing and penetration of the gaseous fuel plume with the rest of the air in the combustion chambers."
"I don't see KIVA being routinely used for component and system design for at least another ten years," he says. Modeling all the complex processes inside the combustion chamber means little if engine designers are unable to interpret and couple those processes with other engine components. Yet the results of these computational fluid dynamics (CFD) computations are not readily interpretable by most designers.
Illustration of transient heat flow
Assanis is taking the results from KIVA and incorporating them into thermal, structural, and other systems level codes, in an attempt to bridge the gap between modeling and actual design and manufacturing. Working with NCSA's Software Development Group, Assanis hopes to create visualizations and animations of fluid and spray dynamics and their interactions, an approach that he feels will allow designers to better understand the results of the KIVA code.
Another problem is that a typical simulation of just one cycle takes about 30 hours on the CRAY Y-MP system, a time that Assanis says is simply unacceptable to industry designers. Advances in rapid proto-typing have allowed designers to cheaply test some¥but not all--components. He says many medium-sized companies simply do not own or have access to Cray computers--they typically do all of their engine development using simplified, zero-dimensional tools on PCs or workstations. The answer, according to Assanis, is parallelization of KIVA--a task that KIVA's developers at LANL are currently pursuing. NCSA is working with LANL and Thinking Machines Corp. to port KIVA to NCSA's CM-5. Assanis will then incorporate his latest models of engine flow and spray processes into the parallelized version of KIVA.
Not only would parallelization allow the modelers to introduce even more variables into their calculations and achieve finer grid resolution, but a parallelized code would be faster and could be ported to clusters of networked workstations.
Two-stroke engines may be the automobile engine design of the future, according to Assanis, if the scavenging process can be improved so that the engines can meet emission standards.
"If a code is ever to be successful," says Assanis, "you need the participation and close collaboration of the modeler; the experimentalists who validate the models; the specialists in visualization, parallelization, and algorithm development; and from designers in industry. We will see a lot more multimachine, multiperson collaborations in the future."
