Presentation
Data Fusion for Nuclear Fusion – Using HPC To Put a Star in a Bottle
Presenter
Event Type
HPC Impact Showcase
W
TUT
TP
EX
EXH
Industry
TimeWednesday, November 14th1:30pm - 2:15pm
LocationD171
DescriptionFusion energy offers the prospect of a carbon-neutral, environmentally responsible, and inexhaustible energy source. TAE Technologies Inc. is trying to greatly accelerate fusion research to develop the world’s first commercially viable fusion-powered generator for electricity production.
To this end, TAE has invested $100 million of capital expenditure in the construction of its latest magnetically confined fusion experiment, “Norman,” an advanced beam-driven field-reversed configuration (FRC) plasma device. A central challenge of understanding the physics in a fusion plasma experiment is that most of the experimental diagnostics are indirect in nature and require inverse problems, such as tomographic inversion, and that there are many interacting degrees of freedom each of which requires its own diagnostic and inversion process. To understand the “what” of the complete plasma state measured in this way requires data science, in particular, sensor fusion and Bayesian Inference. To understand the “why” of the plasma state requires theory and advanced computation, which is made challenging by multiple time and space scales and multi-physics interactions. The central problem of keeping the plasma hot enough for long enough to achieve fusion cannot be addressed without using HPC to understand non-linear wave-particle interactions, which can bring both great benefit in terms of kinetic stabilization of modes at the macroscale, and detriments in the form of heat losses due to kinetic microturbulence.
Analysis of existing experiments, and prediction of performance in new parameter regimes, requires a fusion of the “what” and the “why” with a combination of data science and HPC modeling. TAE is partnering with data science heavyweights – Google and others – and big iron heavyweights such as the Department of Energy Leadership Computing Facilities – to bring the most advanced data science algorithms and the fastest computers to bear on these problems.
To this end, TAE has invested $100 million of capital expenditure in the construction of its latest magnetically confined fusion experiment, “Norman,” an advanced beam-driven field-reversed configuration (FRC) plasma device. A central challenge of understanding the physics in a fusion plasma experiment is that most of the experimental diagnostics are indirect in nature and require inverse problems, such as tomographic inversion, and that there are many interacting degrees of freedom each of which requires its own diagnostic and inversion process. To understand the “what” of the complete plasma state measured in this way requires data science, in particular, sensor fusion and Bayesian Inference. To understand the “why” of the plasma state requires theory and advanced computation, which is made challenging by multiple time and space scales and multi-physics interactions. The central problem of keeping the plasma hot enough for long enough to achieve fusion cannot be addressed without using HPC to understand non-linear wave-particle interactions, which can bring both great benefit in terms of kinetic stabilization of modes at the macroscale, and detriments in the form of heat losses due to kinetic microturbulence.
Analysis of existing experiments, and prediction of performance in new parameter regimes, requires a fusion of the “what” and the “why” with a combination of data science and HPC modeling. TAE is partnering with data science heavyweights – Google and others – and big iron heavyweights such as the Department of Energy Leadership Computing Facilities – to bring the most advanced data science algorithms and the fastest computers to bear on these problems.
Presenter