NRL NREIP student helps generate cost-saving physics modeling algorithm

WASHINGTON  –   U.S. Naval Research Laboratory (NRL) Naval Research Enterprise Intern Program (NREIP) intern, Liam Magargal, and a team of NRL scientists and collaborators from the University of Washington (UW) develop novel algorithms to significantly reduce time and costs of simulating the behavior of complex multiphysics systems with machine learning.
 
These efforts are aimed at enabling efficient, accurate, and low-cost designs of defense systems and applications.
 
NRL offers undergraduate and graduate students with a strong interest in scientific research an opportunity to learn under the tutelage of professionals through the NREIP.
 
During the ten-week internship program, students work with mentors at participating Navy laboratories who help hone or further develop their skillsets. Magargal, a Lehigh University doctoral student, helped design and build computational multiphysics models and subsequently used them to generate synthetic data to train algorithms developed by the NRL and UW team.
 
“Computational multiphysics is a field of computational mathematics and physics that enables scientists and engineers to model complex phenomena, such as modeling airflow over an airplane,” said Magargal.  “While these tools have become indispensable in engineering design, they are often too computationally expensive to be used in many time-critical analyses.”
 
Dr. Steven Rodriguez, an NRL research scientist from the Computational Multiphysics Systems Laboratory who is heading the NRL and UW team in this effort, guided Magargal through the development of an in-house code based on numerically modeling the physics of multiphase flow with smoothed particle hydrodynamics. The tandem composed code that can be customized and user-defined to allow for physical inputs such as conductivity, density, viscosity and other physical and computational parameters.
 
Magargal and Rodriguez first focused on generating training data for  NRL’s algorithms with simple fluid flow often seen in natural convection, such as Rayleigh-Bénard instabilities – a phenomenon which can be seen when you boil water. 
 
“Liam focused on helping me code up a mathematical technique used to model fluids called the ‘Smooth Particle Hydrodynamics Method,’ or SPH for short, which was originally developed to model astrophysics,” said Rodriguez. “SPH, is recognized among the scientific computing community as an effective modeling tool, and has shown to be useful for problems involving different types of fluids with different densities – for example, how oil and water interact at room temperature.”
 
This past summer, Magargal learned the mathematical framework of SPH and how to communicate these ideas to a computer to run fluid simulations and study the behavior of intermixing fluids. After modeling the Rayleigh-Bénard Convection, Magargal leveraged the code to systematically generate training data for the Projection-Tree Reduced-Order Model (PTROM) – the algorithm developed by NRL and UW team. 
 
“The PTROM is a class of reduced-order modeling, which is a discipline in applied and computational mathematics that aims to reduce the costs of simulating complex multiphysics systems,” Rodriguez said. “It is an approach akin to machine learning, where you feed an algorithm data over a couple of different user inputs runs and the algorithm is able to predict output data of many other desired inputs it was not trained on.”
 
Magargal’s code and data will enable the deployment of the PTROM for many query applications such as design optimization, uncertainty quantification, and control. Rodriguez went on to say “Using Liam’s SPH code, we can train the PTROM to learn the behavior of intermixing fluids over a few physical properties, such as different densities and viscosities. So that if we train our PTROM over the interactions of air and water, it can guess how honey and milk will interact – as a fun and extreme example.”
 
“I was drawn to NRL because of Dr. Rodriguez and his organization’s research, which involves applied mathematics and machine learning methods and how they relate to computational physics models,” Magargal said. “I had a healthy amount of freedom to explore new interests while working toward an end goal, and I was excited to build skills in new areas that will be beneficial to me throughout my career.”
 
The resulting code Magargal and Rodriguez developed is now being used to for new developments that will further extend the capabilities of PTROM algorithm.
 
“NRL has always supported mentorship and encouraged mentoring students,” Rodriguez said.  “On a personal level, I had many great mentors over my career and NREIP is an opportunity to provide other students the help I received when I was first starting in research.”
 
Whether it be receiving access to advanced software and hardware to working alongside Nobel Prize caliber scientists, Rodriguez encourages post-doc students to participate in NREIP internships across the Department of the Navy.
 
Magargal recounted that the NREIP internship provided him real-world experience in developing physics models and associated computing.
 
“I plan to continue collaborating with Dr. Rodriguez throughout graduate school, as our research areas are closely aligned,” Magargal said.
 
The Office of Naval Research is offering summer appointments at a Navy lab to current sophomores, juniors, seniors and graduate students from participating schools.
 
For more information about NREIP opportunities, please contact NRL’s NREIP coordinator at: NREIP@nrl.navy.mil
 

About the U.S. Naval Research Laboratory

NRL is a scientific and engineering command dedicated to research that drives innovative advances for the U.S. Navy and Marine Corps from the seafloor to space and in the information domain. NRL is located in Washington, D.C. with major field sites in Stennis Space Center, Mississippi; Key West, Florida; Monterey, California, and employs approximately 3,000 civilian scientists, engineers and support personnel.
 
For more information, contact NRL Corporate Communications at (202) 480-3746 or nrlpao@nrl.navy.mil.