Digital Twins and ML
Fast, data-informed models of a device that mirror its behavior for study, control design, and what-if analysis.
What a digital twin is
A digital twin is a computational model of a physical or designed system that runs alongside it, kept consistent with its behavior, and used to predict, analyze, and test scenarios. For fusion, a twin couples physics models of the plasma and machine with data-informed components to stay accurate and fast enough to be useful.
Where ML fits
- Surrogates that make the twin fast enough for interactive use
- State estimation that keeps the twin aligned with measurements
- Learned corrections closing the gap between model and reality
- Anomaly detection comparing observed behavior to the twin's prediction
Uses
A twin lets engineers test control strategies, explore operating scenarios, and study fault responses without touching hardware. It supports training operators, planning campaigns, and diagnosing discrepancies when the real system departs from expectation. The ML components are what make it fast and self-correcting.
Fidelity and honesty
A twin is only as good as its models and the data that inform them. A twin extrapolated beyond its validated range gives confident but unreliable predictions, the same trap as any surrogate. A trustworthy twin states where it is validated and flags when the real system enters territory it has not been checked against.
Design context
For a design-stage device such as the Kronos breeder, a digital twin is a simulation of the design, informed by physics and by data from other devices where relevant. It is a tool for engineering and study, and its outputs are model predictions for a design, not measurements of built hardware. No performance claim rests on the twin beyond what its underlying physics supports.