Parallelized Real-time Physics Codes for Plasma Control on DIII-D

TL;DR

This work addresses the need for deterministic, low-latency real-time physics computations in fusion plasma control by developing a real-time safe multi-threading library for the DIII-D PCS. Using a Manager–Worker paradigm with C11 atomics, the library parallelizes independent tasks to run rt-TORBEAM and rt-STRIDE in real time, achieving cycle times of roughly $20\,\mathrm{ms}$ and $100\,\mathrm{ms}$ respectively. The TORBEAM implementation demonstrates $\rho$-tracking accuracy with $|\rho_{target}-\rho_{torbeam}|\leq 0.05$, while STRIDE leverages a Riccati-based shooting approach to compute $\delta W$ across $\sim$200 intervals within the same time budget. Together, these results show robust, deterministic execution suitable for real-time fusion control, with potential to enable other physics-based modules and future surrogates while maintaining strict reliability guarantees.

Abstract

A real-time safe multi-threading library was developed on the DIII-D plasma control system to optimize the realtime TORBEAM and real-time STRIDE physics codes. These physics codes are crucial for future fusion power plant operation as they provide information about electron cyclotron wave propagation and heating as well as inform about ideal plasma stability limits. The real-time TORBEAM code executed consistently in under 20 ms while the real-time STRIDE code computes in 100 ms. The multi-threading library developed in this work can be applied to other real-time physics-based codes that will be crucial for the next generation of fusion devices.

Figures (5)

• Figure 1: UML diagram of the multi-threading algorithm. The Manager and Workers each follow the vertical black lines downward to progress through the shot. The horizontal bright green arrows represent communications between the Manager and Workers. An arched pink arrows denotes a thread waiting until some condition X is made false where the condition is given by Spin X. The orange arched arrows denote Workers doing computational work. Finally, the dark green arrows represent the Workers returning to the post-initialization state to repeat for the next CPU cycle during a shot.
• Figure 2: Example ECH ray-tracing where ECH beam begins from the gyrotron, makes its way to the ECH mirror to be launched into the plasma at some poloidal and toroidal angle. It then travels through the plasma at a calculated trajectory until being absorbed at the vertical red resonance location. The maximum absorption location (purple X), is given by coordinates $(R,Z)$. Refer to poli_torbeam_2018 for further explanation.
• Figure 3: Experimental results of rt-TORBEAM calculation and gyrotron mirror control. (a) DIII-D shot 199073 with constant $\rho_{target}$ values for four controlled gyrotron mirror where for each individual gyrotron, $\rho_{target}$ (dashed line) and $\rho_{torbeam}$ (solid line) have matching colors. (b) DIII-D shot 199071 with dynamic $\rho_{target}$ (black dashed line) that is the same across all gyrotrons (solid colored lines). (c) CPU cycle time to compute rt-TORBEAM for all four gyrotrons.
• Figure 4: Shooting method of the STM. Each vertical line represents a cut in the integration of the STM. The solid red lines are rational surfaces where the integration of the STM must be split apart and the dashed red lines being artificial interval breaks to create more intervals to divide the calculation further for the Workers. This figure has been adapted from a similar figure in glasser_robust_2018.
• Figure 5: Results from rt-STRIDE calculation in the DIII-D PCS in shot 187066. (a) gives the $\delta W$ stability value over time during the shot and (b) shows the CPU cycle time required for one full rt-STRIDE calculation.