GOVERNMENT

Leveraging the power of grid computing, simulated scenarios to analyze concepts, tactics and technologies can now begin to address the key variable of human response: Large-scale simulations to model and analyze concepts, tactics and technology that involve people historically lack trade studies on the key variable of human decision making. Dr. Dimitri Mavris, a Professor, and Dr. Patrick T. Biltgen, a System-of-Systems Researcher in the Aerospace Systems Design Laboratory (ASDL) at the Georgia Institute of Technology, are pioneering solutions to the problem using software from Ternion Corporation and Phoenix Integration, Inc. Ternion Corporation's FLAMES software provides a "Constructive Simulation Framework" to allow creation of virtually any type of simulated environment for performing Monte Carlo analysis and parametric trade studies. The software is commonly used for testing, training, and analysis of battlefield tactics in the Aerospace & Defense industry, but has capabilities for other government agencies and commercial organizations to conduct simulations in any setting. Phoenix Integration's PHX ModelCenter and PHX CenterLink software enable conceptual modeling for engineering design and conducting trade studies on design models involving huge data volume and numerous variables. Earlier this year, the collaborating companies demonstrated how running FLAMES simulations in ModelCenter provided a platform for performing trade studies in simulated environments that were previously never thought possible. ASDL's research at Georgia Tech is the latest example of new methods development using these software tools. According to Biltgen, "FLAMES allows human cognition modeling in a simulated environment in order to analyze human-like decisions in that environment. At times, this requires running very large numbers of data points through thousands of trade studies." He also notes that these studies can be very time consuming. "Running the FLAMES scenarios in ModelCenter, CenterLink, or through CenterLink as a broker for a High-Performance Computing (HPC) cluster speeds the process to make this type of analysis possible," he says. Crawl, Walk, Run, Sprint To illustrate his point, Biltgen describes his creation of an "intelligent battle manager" to simulate human decisions on a battlefield. This technique uses machine learning and agent-based modeling approaches to identify high-payoff combinations of systems, technologies, and tactics by running many FLAMES cases and building a surrogate model of the results. To manually set up and run a single case in FLAMES typically takes 45 minutes. "This was a show-stopper. The technique we proposed can't work in this environment, so we needed a way of automating and integrating the analysis process which is why we introduced ModelCenter," says Biltgen. After a one-time integration of FLAMES with ModelCenter, the one case and any additional cases can be run very quickly in what Biltgen calls the "crawl" mode-faster but still insufficient for the large-scale studies needed to create an "intelligent battle manager." In what he calls a "walk" mode, Biltgen uses a Design Of Experiments (DOE) tool in ModelCenter to generate multiple cases through FLAMES, then run them in ModelCenter on, for example, 10 computers running 1,000 cases each. It takes approximately 8 hours to run the 10,000 cases compared to 7,500 hours if attempted in "crawl" mode. "The only drawback to this approach is the labor required to manually distribute and reassemble the cases," says Biltgen. "To address this technical challenge, we introduced CenterLink." In the "run" mode, Biltgen takes advantage of Phoenix Integration's PHX CenterLink grid computing solution to manage and run even more cases on more computers, without having to update simulations and data files by visiting each computer, which CenterLink does automatically. His next step to a "sprint" mode will have CenterLink managing case loads across a 128-processor HPC cluster. "I will be able to get all the data I need in less than one hour," says Biltgen. Integrating FLAMES with ModelCenter not only enables testing different systems in different scenarios, but can also begin to account for the human factor. The capability will add tremendous value whether analyzing a battle space, simultaneously trading tactics and technologies, simulating responses to natural disasters or terrorist attacks, or other similar projects in business and industry. Academia-Friendly Software Packages More than 50 leading universities worldwide use Phoenix Integration's software in classroom and industry-related research environments, including the Air Force Institute of Technology, Beijing Institute of Technology, Berlin University of Technology, Carnegie Mellon University, University of Manitoba, Michigan, MIT, Penn State, Purdue, Vanderbilt and Seoul National University. All take advantage of a special Phoenix Integration program that recognizes university budget constraints and offers educational institutions the same integration software used throughout industry with discounted "academic software bundles."