Abstract: Turbulence is a common phenomenon in environmental and engineering flows, and it involves chaotic and 3-dimensional motions in a wide range of scales. In practical applications, resolving all the scales is prohibitively expensive, so numerical simulations such as Reynolds-averaged Navier Stokes (RANS) solve the transport equations by resolving only the large scales and modeling the small scales. Traditional RANS turbulence models have been around since 1970’s and have had limited success in simple turbulent flows, but are still notoriously inaccurate in most complex flows. In the past decade, machine learning and big data have made significant inroads in seemingly intractable problems in computer vision and natural language processing, so these tools attracted significant interest in the turbulence modeling community. In this talk, we discuss data driven approaches to generate alternative turbulence models, in particular leveraging random forests and deep neural networks trained with high fidelity turbulent simulations. We argue that such methods should be developed in a way that is informed by underlying physical principles and numerical considerations. We describe how this framework led to the development of the tensor basis neural network (TBNN), a deep learning model for Reynolds stress that embeds Galilean invariance. Then, we delve deep into a case study in turbulent scalar mixing for aerospace applications, which culminates with the development of a scalar mixing version of the TBNN, called TBNN-s, that shows significant promise compared to existing RANS mixing models.
 
Bio: Dr. Julia Ling is a tech lead on the Tidal project at X, the Moonshot Factory.  She leads the software and machine learning team at Tidal, a project whose mission is to protect the ocean while feeding humanity sustainably.  Prior to joining Tidal, Julia was the CTO at Citrine Informatics, a 70+ person start up in the Bay Area building an AI platform to accelerate new materials development.  She was a Harry S. Truman fellow at Sandia National Labs where she helped pioneer the application of deep learning to turbulence modeling.  She holds a PhD in Mechanical Engineering from Stanford University and a Bachelors in Physics from Princeton University.

Bio: Dr. Pedro M. Milani is a research scientist on the Tidal project at X, the Moonshot Factory in Mountain View, CA. He is responsible for developing and improving machine learning models for scientific applications that further Tidal's mission of ocean sustainability. Prior to Tidal, Pedro worked as an associate in the Thermal Sciences division of Exponent, a scientific consultant in Menlo Park, CA where he focused on wildfire modeling. Pedro holds a PhD in Mechanical Engineering from Stanford University, where he worked with Dr. Ling in machine learning models for turbulent mixing. He also holds a Bachelors in Mechanical Engineering from Stanford University.

Location: https://argonne.zoomgov.com/j/1619902756?pwd=cEpWUU1DbUU1OGhKbmdObzZIdE…