In yesterday's research thread, I wrote about fusion plasmas, stellarators, and magnetic confinement fusion in general terms. Today, let's start with things I've been directly involved in. We'll begin with coil optimization. - ThreadSky

dbindel.bsky.social • 84 days ago

In yesterday's research thread, I wrote about fusion plasmas, stellarators, and magnetic confinement fusion in general terms. Today, let's start with things I've been directly involved in. We'll begin with coil optimization.

Reposted from David Bindel

For today's "advent of research" thread, let's introduce fusion plasma and stellarators.

Comments

dbindel.bsky.social•84 days ago

In stellarators, most of the magnetic field comes from an external set of electromagnetic coils. There may sometimes be a contribution from bootstrap current in the plasma, but it's usually small compared to what is seen in tokamaks.

dbindel.bsky.social•84 days ago

The standard approach to designing stellarators for many years has been in two-stages: first, one designs the plasma profile; then, one designs coils that produce the right magnetic field at the outermost surface of the plasma.

dbindel.bsky.social•84 days ago

Usually, we look for coils that minimize the field error normal to the outermost surface in a squared integral sense. This is arguably not the "right" metric, since small field errors at the surface may be more or less dangerous based on their shape. But it's the usual business.

dbindel.bsky.social•84 days ago

The optimized coil shapes are often weird and hard to manufacture. Worse, small manufacturing issues can cause unacceptable confinement losses.

The estimated coil tolerances for NCSX (last major US stellarator experiment) were 0.08%. The project went over time and budget, and was canceled.

dbindel.bsky.social•84 days ago

The idea of optimization under uncertainty (OUU) is to optimize a design in a way that it's robust to slight uncertainties, e.g. due to manufacturing error. We used so-called risk-neutral OUU, where the nominal objective f(x) is replaced by the smoother function E_U[f(x+U)] for random errors U.

dbindel.bsky.social•84 days ago

Prior work showed that OUU produced much better coil sets, but the computation in that work was done with a *lot* of Monte Carlo runs to evaluate the risk-neutral objective. We figured we would try the same thing, but faster.
https://doi.org/10.1017/S0022377820000227

Comments

Posting Rules

Reply