black and white photo of Mike Churchill

Michael Churchill

Combining simulation and machine learning with experiment to drive the future of fusion energy

Michael Churchill

Princeton Plasma Physics Laboratory

Tuesday, March 15, 2022


PSFC Seminars

Abstract: With fusion devices such as SPARC and ITER nearing the burning plasma regime, fusion energy researchers will become increasingly reliant on analysis, modeling, and simulation for various engineering and physics
tasks, including prediction and knowledge discovery. These tasks are performed over a range of timescales, from quick between-shot analysis to inform next-shot adjustments, to more detailed simulations of plasma dynamics. In this seminar I will present a view for how simulation and machine learning can enhance the work of fusion scientists and engineers, and the various research projects we have pursued to realize this view. First, I show how fast, deep convolutional neural networks can be applied to multi-scale plasma diagnostics to automate prediction and event identification. Next, I show how normalizing flows can be used as neural density estimators to both speed up Bayesian inference of physical quantities from experimental diagnostic data, and in uncertainty quantification of ad-hoc inputs to scrape-off layer fluid codes, consistent with experimental data (simulation-based inference). Finally, I discuss the role of large-scale, high-fidelity first-principles simulation, such as the edge gyrokinetic code XGC, and how these can be accelerated with machine learning for more routine use in plasma modeling.

Bio: Michael Churchill is a research scientist at the Princeton Plasma Physics Laboratory, in the Computational Sciences Department. His research focuses on computational methods such as machine learning applied to fusion energy simulation and experiment, in particular to understand tokamak physics in the challenging edge region. He is a member of the US SciDAC Center for High-Fidelity Boundary Plasma Simulations (HBPS), working on the edge gyrokinetic code XGC. Michael earned his Ph.D. degree at the Massachusetts Institute of Technology (MIT), where he performed experimental research on the Alcator C-Mod tokamak. He received his Bachelor's degree in Electrical and Computer Engineering from Brigham Young University.