The FreeGSNKE Pulse Design Tool (FPDT): a computational framework for evolutive plasma scenario and control design

The paper introduces the FreeGSNKE Pulse Design Tool (FPDT), an open-source Python framework that couples an evolutive equilibrium solver with a virtual control system to simulate, design, and validate tokamak plasma scenarios and control strategies, demonstrating high accuracy against MAST Upgrade experimental data to reduce the need for costly physical testing.

The Gist

Imagine you are trying to teach a very delicate, super-hot balloon (a plasma) to dance inside a giant, invisible magnetic cage (a tokamak). This balloon is so hot it would melt any physical container, so we have to hold it in place using only magnetic fields.

The problem? This balloon is wobbly, unpredictable, and if it touches the walls of the cage, the whole experiment fails. To keep it dancing perfectly, you need a "dance instructor" (the control system) that adjusts the magnetic fields thousands of times per second.

The Problem:
Before, if scientists wanted to test a new dance routine, they had to go to the actual machine, turn it on, and try it. If the routine was bad, the balloon might crash, wasting millions of dollars and days of work. It's like trying to learn to fly a plane by crashing it every time you tried a new maneuver.

The Solution: The "Flight Simulator" for Plasma
This paper introduces a new tool called the FreeGSNKE Pulse Design Tool (FPDT). Think of it as a high-tech, hyper-realistic flight simulator for nuclear fusion.

Instead of crashing a real machine, scientists can now run the experiment inside a computer. They can tell the computer, "Okay, let's try this new shape," and the software simulates exactly how the plasma would react, how the magnets would heat up, and whether the "balloon" would stay stable.

How It Works (The Analogy)

The tool has two main parts working together like a team:

1. The Physics Engine (The "Balloon"):
   This part of the software is a super-smart calculator that knows the laws of physics. It predicts how the hot plasma balloon will wiggle, stretch, and move based on the magnetic fields. It's like a physics teacher who can predict exactly how a soap bubble will pop if you poke it.

2. The Virtual Instructor (The "PCS"):
   This is the "brain" of the operation. In the real world, this is a computer system that reads sensors and flips switches on the magnets. In the simulation, this is a Virtual Control System.

   • Feedback (The Reflex): If the balloon starts drifting left, the Virtual Instructor instantly yells, "Push the magnets right!" to correct it.
   • Feedforward (The Plan): The Instructor also has a script. It knows, "At 3 seconds, we need to stretch the balloon," so it prepares the magnets in advance.

What Makes This Tool Special?

• It's a "Digital Twin": The authors tested this on the MAST-U tokamak (a real machine in the UK). They took a real experiment that already happened, fed the exact same instructions into the computer, and watched the simulation run.
• The Results: The computer simulation matched the real-world experiment almost perfectly. The "virtual balloon" danced exactly the same way the "real balloon" did.
• Speed vs. Accuracy: The tool offers three different "modes":
  • The "Slow & Perfect" Mode: Takes hours to run but is incredibly accurate (like a high-end flight simulator).
  • The "Fast & Good Enough" Modes: Takes only minutes to run. These are so fast that scientists can use them between real experiments to quickly test new ideas without waiting for the machine to cool down.

Why Does This Matter?

Think of it like a video game where you can practice a difficult level over and over again before you try it for real.

• Safety: Scientists can test dangerous scenarios on the computer. If the simulation shows the plasma will crash, they know not to try it on the real machine.
• Money & Time: Real fusion experiments are expensive. This tool saves money by letting researchers "fail" in the computer so they can "succeed" in the lab.
• Collaboration: Because the tool is open-source (free for anyone to use), scientists all over the world can build on each other's work, making the path to clean, limitless energy (fusion power) faster.

The Bottom Line

This paper presents a virtual playground where scientists can design, test, and perfect the control systems for fusion reactors. It turns the risky, expensive process of "trial and error" on a real machine into a safe, fast, and cheap process of "trial and error" on a computer screen. It's a massive step forward in teaching humanity how to harness the power of the stars.

Technical Summary

1. Problem Statement

In magnetic confinement fusion, operating tokamaks requires robust Plasma Control Systems (PCS) to maintain plasma position, shape, and current while adhering to strict machine safety limits (coil currents and voltages). Developing and testing control strategies for new, high-performance scenarios typically requires expensive and time-consuming physical experiments on the tokamak.

• The Gap: While existing Pulse Design Tools (PDTs) exist, many are proprietary, lack open-source accessibility, or are not easily customizable for different machine configurations. There is a need for a flexible, open-source, machine-agnostic framework that allows for in silico testing of control algorithms, uncertainty quantification, and scenario design without risking physical hardware.
• Specific Challenge: The control of plasma involves complex, non-linear interactions between the plasma equilibrium, active Poloidal Field (PF) coils, and passive conducting structures. Simulating this requires solving coupled Grad–Shafranov (GS) equilibrium equations and time-dependent circuit equations in real-time or near real-time.

2. Methodology

The authors present the FreeGSNKE Pulse Design Tool (FPDT), a Python-based computational framework that couples the FreeGSNKE evolutive equilibrium solver with a new, modular Virtual Plasma Control System (PCS).

Core Architecture

The FPDT operates in a closed-loop simulation cycle:

1. Simulator (FreeGSNKE): Solves the coupled evolution of plasma equilibrium and metal currents.
2. Virtual PCS: Takes simulated measurements (plasma current, shape parameters, position) and calculates the required voltage actuation for PF coils using a combination of Feedforward (FF) and Feedback (FB) control.
3. Feedback Loop: The calculated voltages are fed back into the simulator to evolve the plasma state to the next time step.

Governing Equations

The simulator solves three coupled systems:

1. Circuit Equations: Describes current evolution in active PF coils and passive structures ($M \dot{I} + RI = V$).
2. Plasma Current Equation: A lumped 1D equation governing the total plasma current evolution.
3. Grad–Shafranov Equation: The non-linear elliptic equation describing the magnetohydrodynamic (MHD) equilibrium of the axisymmetric plasma.

Control System Design

The Virtual PCS is modular and includes:

• Plasma Controller: Regulates total plasma current using a blended FB/FF PID strategy.
• Shape Controller: Regulates local and global shape parameters (e.g., elongation, triangularity, X-point positions) using PID controllers.
• Virtual Circuits (VC) Controller: Transforms shape/current requests into specific PF coil current requests using sensitivity matrices (Jacobians) to decouple control actions.
• Systems & Safety Controllers: Enforce machine limits on coil currents, voltage ramp rates, and absolute voltage/current limits.
• Coil Activation Controller: Dynamically switches coils on/off by altering resistance matrices, simulating pre-plasma and ramp-up phases.
• Vertical Controller: A dedicated FB controller to stabilize vertically unstable elongated plasmas.

Solution Modes

To balance accuracy and computational speed, FPDT offers three modes:

1. NL (Non-Linear): Fully non-linear evolution of all equations. Most accurate but computationally expensive.
2. PwLD (Piecewise Linear Dynamics): Linearizes the circuit/plasma dynamics but solves the full non-linear GS equation at every step.
3. PwL (Piecewise Linear): Linearizes both the dynamics and the GS equation. Uses reduced-order descriptors for control. Fastest but relies on frequent re-linearization triggers.

3. Key Contributions

• Open-Source Framework: FPDT is released as part of the open-source FreeGSNKE suite, promoting reproducibility and collaborative research.
• Machine Agnosticism: While the controllers are inspired by the MAST-U PCS, the framework is designed to be easily adapted to any tokamak geometry and control logic.
• Hybrid Control Implementation: It successfully implements a realistic blend of Feedforward (pre-programmed waveforms) and Feedback (PID-based) control, including Virtual Circuit matrices for decoupling shape parameters.
• Safety Limit Enforcement: The framework explicitly models machine protection systems, clipping voltages and currents to prevent simulated damage.
• Efficiency Optimization: The introduction of PwLD and PwL modes allows for simulation speeds an order of magnitude faster than the full non-linear mode, making them viable for inter-shot scenario planning.

4. Results

The FPDT was validated against experimental data from two challenging MAST Upgrade (MAST-U) discharges (Shot 52570 and Shot 50366) during the "flat-top" phase.

• Quantitative Agreement:
  • Plasma Current & Vertical Position: All simulation modes (NL, PwLD, PwL) achieved excellent agreement with experimental data. Relative Mean Absolute Differences (RMAD) were < 0.5% for plasma current and < 1% for vertical position.
  • Shape Parameters: Controlled shape parameters (e.g., midplane radii, X-point positions) tracked reference waveforms with high fidelity. Uncontrolled parameters (e.g., strike point location) evolved consistently with experimental reconstructions.
  • Divertor Configurations: The tool successfully simulated complex divertor scenarios, including the Super-X divertor (Shot 52570) and the X-point target divertor (Shot 50366), accurately capturing the dynamics of strike point sweeping and secondary X-point formation.
• Control Signals: The simulated PF coil voltages and currents matched experimental trends, including the application of safety limits and the response to feedforward drives.
• Computational Performance:
  • NL Mode: ~80 minutes per shot (too slow for real-time planning).
  • PwLD Mode: ~4.5 minutes per shot.
  • PwL Mode: ~2 minutes per shot.
  • The linearized modes (PwLD/PwL) reduced computational cost by an order of magnitude while maintaining accuracy comparable to the NL mode for control-relevant quantities.

5. Significance

• Reduced Experimental Risk: FPDT allows researchers to test new control algorithms and challenging plasma scenarios in silico before running them on the physical tokamak, significantly reducing the risk of machine damage or failed discharges.
• Training and Calibration: The tool serves as a "flight simulator" for control room operators and can be used to calibrate control gains and optimize parameters for specific operating regimes.
• Future-Proofing: The framework is designed to support the development of AI-driven control schemes (e.g., Reinforcement Learning) and can be extended to include transport and current diffusion models in the future.
• Community Impact: By providing a transparent, open-source tool, the authors aim to standardize plasma control research and accelerate the development of robust control strategies for future fusion reactors like ITER and DEMO.

In conclusion, the FPDT represents a significant advancement in plasma control simulation, offering a high-fidelity, flexible, and computationally efficient tool for the fusion community to design and validate the next generation of tokamak scenarios.