Not Just Data: HPC for Design Decisions

Not Just Data: HPC for Design Decisions

Currently, DoD assets operate in a largely uncontested air, space, and electromagnetic (EM) environment. Air, ground, and navy vehicle systems with current capabilities (e.g., stealth) operate without opposition, and GPS and other C4I systems provide reliable and accurate positioning, navigation, and timing data without contention. This situation is changing rapidly as adversaries develop capabilities. To counter these emerging threats, DoD must rapidly develop and deploy advanced capabilities such as high-speed vehicles, selectively automated decision-making systems, rapid-fire low-consumable effects systems to deal with large numbers of drones (swarms), and maintain C4I in a contested EM environment. Even Michael Griffin, Undersecretary of Defense for Research and Engineering, has declared the development of hypersonic capabilities as the Pentagon’s “highest technical priority.”1

The challenges are daunting and must be met with new approaches that exploit emerging HPC, artificial intelligence (AI), and other technologies in novel ways. HPC is needed to apply data analytics, machine learning (ML), and deep learning (DL) to decision making.

I apply digital modeling techniques with the aim of reducing the amount of time and money needed to design, test, and build systems. Shorter design times mean systems could be operational sooner, which is important to maintaining U.S. defense positions. We can digitally test designs and performance before actual construction begins, and supercomputers are needed to churn through the massive amounts of data needed for modeling.

HPC for Design Decisions Blog Engility is working with the DoD on advanced computational fluid dynamics (CFD) code for vehicle design

HPC as an Enabler

Some of the CFD datasets we are working with are petabytes in size, which is non-trivial to process, and therefore require HPC systems. We are developing artificial intelligence systems to be able to detect when to output certain subsets of data of interest, and also use AI-based auto-encoders to compress the data for storage. The system we are currently developing is able to detect certain small-scale physics in a CFD system, and then extract specific information only when such a condition is detected. We are also using AI systems to try to understand more of the underlying physics driving vehicle performance.

Analyzing the Data

CFD scientists can choose from a number of tools to visualize and post-process CFD results. Many of these tools have been designed to run on distributed cluster systems to process and visualize large amounts of data. We also use HPC to build deep neural networks that take CFD solutions as their input and predict information that can be used to make informed decisions.

Achieving Exascale Computing

HPC processing capabilities are expanding as the state of technology advances towards the exascale era, in which computers will be able to execute a billion billion calculations per second. Exascale computing will allow us to more effectively and rapidly perform modeling and simulation for national security missions—allowing us to make technological advances before our adversaries. National security requirements for both peer-to-peer conflict and economic competitiveness necessitate increasingly rapid advancements to solve some of our most critical challenges in hypersonic flight.

The Big Picture

As we approach new capabilities, like traveling from New York to Los Angeles in less than 39 minutes, I’m passionate about the mission to maintain U.S. national security and proud to contribute my digital modeling and data analysis expertise to develop tomorrow’s vehicles.


Share this Post:

Posted by Wes Brewer

I am a DoD HPCMP PETTT data scientist. I work with DoD scientists to develop deep learning solutions on supercomputers to solve complex problems in the following domains: condition-based maintenance, hyperspectral imagery, cyber data science, real-time object detection, and hypersonics. I have an M.S. in Ocean Engineering from MIT and a Ph.D. in Computational Engineering from Mississippi State University.