Seminar featuring Dr. Qian: “Scalable computational methods for engineering decision-making”

ABSTRACT: Engineering design and decision-making rely on many-query computations, where a system must be simulated not once, but many times, for example within an optimization, control loop, or Monte Carlo simulation. Traditional simulation methods that solve partial differential equations (PDEs) on finely-resolved computational meshes, such as finite element or finite volume methods, can be prohibitively expensive for this many-query setting. To enable engineering decision-making for large-scale systems, efficient reduced models are needed. This talk will first present Lift & Learn, a scientific machine learning method that uses data to learn scalable reduced models for nonlinear PDEs. The talk will then introduce multifidelity methods for many-query computations, which can combine reduced models with traditional simulations in a principled way for faster results with accuracy guarantees. A multifidelity method for global sensitivity analysis is presented, with application to a thermal distortion model for the James Webb Space Telescope.

BIOSKETCH: Elizabeth is a von Karman Instructor in the Department of Computing + Mathematical Sciences at Caltech. Her research is motivated by the need for computational methods used in engineering decision-making to be efficient and scalable, with particular focus on model reduction, scientific machine learning, and multifidelity methods. Elizabeth received SB and SM degrees in aerospace engineering and a PhD in computational science and engineering from MIT. She has also been the recipient of a Fulbright student grant, the NSF Graduate Research Fellowship, and a Fannie and John Hertz Foundation Fellowship.

Date/Time:
Date(s) - Feb 18, 2022
11:00 am - 12:00 pm

Location:
47-124 Engineering IV
420 Westwood Plaza Los Angeles CA