A tutorial on inversion-based shape control with design application to NSTX-U

This paper presents a comprehensive tutorial on inversion-based shape control (IBSC) for tokamaks, detailing design variations and a systematic procedure that, when applied to NSTX-U, successfully eliminates vertical control oscillations and improves phase margin by decoupling shape and vertical control interactions.

The Gist

Imagine you are trying to balance a broom on the palm of your hand. Now, imagine that broom is a giant, super-hot ball of gas (plasma) floating inside a donut-shaped machine called a tokamak. This plasma is incredibly unstable; if you don't control it perfectly, it will crash into the walls of the machine and shut down.

This paper is essentially a masterclass on how to build the "hand" that keeps the plasma balanced and shaped, specifically for a machine called NSTX-U.

Here is the breakdown of their method, using simple analogies:

1. The Core Problem: The "Broom" is Wobbly

In a tokamak, you have a set of giant electromagnets (coils) surrounding the plasma. By turning the power up or down in these coils, you can push and pull the plasma to keep it in a specific shape (like a D-shape) and in the center of the machine.

The authors call their method Inversion-Based Shape Control (IBSC).

• The Analogy: Imagine you are a chef trying to bake a cake in a specific shape. You have a map that tells you: "If I add 1 cup of flour, the cake rises 1 inch. If I add 2 eggs, it gets 1 inch wider."
• The Inversion: Instead of guessing, you work backward. You say, "I need the cake to be 2 inches wider and 1 inch taller. Based on my map, I need to add 2 eggs and 1 cup of flour."
• The Paper's Job: They figured out the perfect "map" (mathematical model) to tell the magnets exactly what to do to fix the plasma shape instantly.

2. The Three Big Choices (The Recipe)

The paper explains that there isn't just one way to make this map. You have to make three big decisions, like choosing ingredients for a recipe:

• Current vs. Voltage: Do you tell the magnets how much electricity to push (Current), or how hard to push the electricity (Voltage)?
  • Analogy: Do you tell the car engine how many RPMs to spin, or how hard to press the gas pedal? The paper shows that sometimes pressing the pedal (Voltage) is a more direct way to control the car's speed.
• Static vs. Dynamic: Do you use a map based on a still photo (Static), or a video of how the plasma moves over time (Dynamic)?
  • Analogy: A static map is like a snapshot of a runner. A dynamic map is a video of them running. The paper found that for some machines, the "snapshot" (Static) actually works better than the "video" because the plasma behaves in a tricky way that the video model gets wrong.
• The Plasma Model: Do you treat the plasma like a solid rock that just moves up and down (Rigid), or like a jelly that wobbles and changes shape as it moves (Non-rigid)?
  • Analogy: The paper discovered a surprise: treating the plasma like a solid rock (Vacuum model) actually worked better for their specific machine than treating it like a wobbly jelly, because the metal walls of the machine dampen the wobbling.

3. The "Ghost" Problem: The Right-Half Plane Zero

This is the most technical part, but here is the simple version.
Sometimes, when you try to push the plasma up to fix a wobble, the physics of the machine makes it dip down for a split second before going up. It's like trying to steer a car that has a broken steering wheel: you turn left, but the car jerks right first.

• The Paper's Solution: They realized that if you try to fix the shape and the vertical wobble at the same time using the same "map," the controller gets confused and makes the wobble worse.
• The Fix: They built a Decoupling Filter.
  • Analogy: Imagine you have a car with two drivers. One driver is in charge of steering (Shape), and the other is in charge of the gas pedal (Vertical Stability). If they both try to steer at the same time, the car spins out. The paper's solution is to tell the "Shape Driver": "Hey, don't touch the gas pedal or the vertical steering! Leave that to the other driver. You just focus on the shape." This stops them from fighting each other.

4. The "Bobble" Fix

The NSTX-U machine had a problem called a "vertical bobble." It was a tiny, annoying up-and-down shaking of the plasma that happened even when everything seemed fine.

• The Cause: The shape controller was accidentally pushing the plasma up and down while trying to change its shape, fighting against the vertical stabilizer.
• The Result: By applying their new "Decoupling" rules (telling the shape controller to stop messing with the vertical position), they eliminated the bobble completely. They also made the system more stable by adding a few extra magnets to the vertical control team, which was like adding a second set of stabilizing fins to a rocket.

5. The "Traffic Cop" (Quadratic Programming)

Finally, the paper talks about handling limits. The magnets can only push so hard before they break or overheat.

• The Analogy: Imagine a traffic cop directing cars. If a car tries to go too fast, the cop doesn't just let it crash; they gently slow it down while still keeping traffic moving.
• The Tech: They used a mathematical tool called a "Quadratic Program" (QP). This is like a smart traffic cop that calculates the perfect way to adjust all the magnets at once to get the shape right, without ever asking any single magnet to do more than it can handle.

Summary

This paper is a tutorial for engineers on how to stop a super-hot ball of gas from crashing into the walls of a fusion reactor. They did this by:

1. Creating a better "instruction manual" (the map) for the magnets.
2. Realizing that the "Shape" and "Vertical" controls were fighting each other, and separating them so they work as a team.
3. Using smart math to make sure the magnets don't get overworked.

The result? A much smoother, more stable plasma that stays in the center of the machine, bringing us one step closer to clean, limitless fusion energy.

Technical Summary

1. Problem Statement

Magnetic shape control in tokamaks is a complex Multi-Input Multi-Output (MIMO) control problem involving 10–20 coil actuators and numerous shaping parameters (e.g., plasma position, elongation, triangularity). While Inversion-Based Shape Control (IBSC) is the industry standard—linearizing an equilibrium to create a sensitivity map and inverting it to determine control currents—several critical challenges remain:

• Design Ambiguity: There is a lack of systematic guidance on choosing between static vs. dynamic maps, current vs. voltage mappings, and specific plasma response models (vacuum, rigid, or non-rigid).
• Vertical Instability Coupling: A major issue on the National Spherical Torus Experiment-Upgrade (NSTX-U) was the "vertical bobble" (undesired vertical oscillations). This was caused by a lack of decoupling between shape control and vertical stability control.
• Inverse Response (RHP Zeros): For elongated plasmas, static linearization models often predict control actions that are initially in the wrong direction (inverse response) due to Right-Half Plane (RHP) zeros introduced by the interaction between the vertical control loop and the shape controller. This leads to instability if not properly managed.

2. Methodology

The authors propose a comprehensive IBSC framework and a systematic design procedure to address these issues.

A. The IBSC Framework

The paper classifies IBSC designs based on four key dimensions (summarized in Table 1 of the paper):

1. Input Type: Current Map (mapping coil currents to shape) vs. Voltage Map (mapping power supply voltages to shape).
2. Dynamics: Static Map (derived from equilibrium linearization $C$ matrix) vs. Dynamic Map (derived from the stabilized plant transfer function $P(s)$).
3. Plasma Model: Vacuum (ignores plasma response), Rigid (plasma shifts as a rigid body), or Non-rigid (full Grad-Shafranov redistribution).
   • Key Finding: Surprisingly, for static maps on NSTX-U, the Vacuum model often aligns better with the true dynamic response than the high-fidelity Non-rigid model, likely because vessel currents dampen plasma motion on control timescales.
4. Inversion Method:
   • Pseudoinverse PID: Direct inversion with PID on individual errors.
   • SVD Control: Projects errors onto singular modes to reduce dimensionality (used in JET's XSC).
   • Quadratic Program (QP): Solves for optimal currents subject to hardware constraints (voltage/current limits).

B. Systematic Design Procedure

The paper outlines a step-by-step procedure (Fig. 7) for integrated vertical and shape control:

1. Linearization: Generate state-space models ($\dot{x}=Ax+Bu, \delta y=C\delta x$).
2. Vertical Stabilization: Design a dedicated PD controller for vertical stability.
3. Decoupling:
   • Vertical Decoupling: Ensure shape control commands are orthogonal to vertical control.
   • RHP Zero Mitigation: Apply penalties (weights) to "up-down antisymmetric" coil currents in the shape controller to slow down responses that excite vertical instability.
4. Inversion & Tuning: Use ALIGN algorithms (for dynamic maps) or QP optimization (for static maps) to compute control gains.

C. Application to NSTX-U

The authors applied this framework to NSTX-U, specifically addressing the vertical bobble and optimizing the use of Poloidal Field (PF) coils.

3. Key Contributions

• Unified Framework: The paper provides the first comprehensive taxonomy of IBSC design choices, clarifying the trade-offs between static/dynamic maps and different plasma models.
• Vertical Decoupling Strategy: It identifies the root cause of NSTX-U's vertical bobble as a coupling between shape and vertical control. The solution involves:
  • Adding vertical actuation penalties to the shape controller's cost function (penalizing antisymmetric coil currents).
  • Explicitly modeling the RHP zero behavior and ensuring the shape controller is slower than the vertical instability growth rate in the vertical direction.
• Optimization of Actuator Usage: Demonstrated that including PF1A and PF2 coils in the vertical control vector (in addition to the standard PF3U/PF3L) significantly improves performance.
• Tutorial & Reproducibility: The appendices provide extensive tutorials, MATLAB code snippets, and design procedures for NSTX-U, making the methodology accessible for other tokamak operators.

4. Results

Applying the systematic design procedure to NSTX-U yielded the following results:

• Elimination of Vertical Bobble: By applying vertical decoupling (penalizing antisymmetric currents) and using a QP-constrained controller, the unwanted vertical oscillations (bobble) observed in shots like 204660 were removed in nonlinear simulations (GSevolve).
• Improved Phase Margin: Including PF1A and PF2 in the vertical control vector increased the phase margin from 32° to 38° and reduced overshoot by 15%. This was achieved with no new hardware, only software changes.
• Robust High-Elongation Control: The new controller successfully simulated tracking a highly elongated plasma ($\kappa = 2.3$) and performing emergency ramp-downs (de-elongation) while maintaining vertical stability and keeping shaping errors below 2cm.
• Model Validation: Simulations confirmed that while the Non-rigid dynamic model is the most accurate for simulation, the Vacuum static map is often a superior choice for the controller's internal model when used with static inversion, due to vessel shielding effects.

5. Significance

This work is significant for the fusion community for several reasons:

• Operational Safety: It provides a proven method to eliminate vertical instabilities (bobble) that can lead to Vertical Displacement Events (VDEs) and plasma disruptions.
• Cost-Effective Upgrades: It demonstrates that significant performance gains (better phase margin, higher elongation limits) can be achieved through control algorithm optimization (e.g., adding PF1A/PF2 to the control vector) rather than expensive hardware modifications.
• Standardization: By formalizing the IBSC design process, the paper offers a blueprint for designing controllers for future devices (like SPARC or ITER) and improving existing ones.
• Educational Value: The inclusion of detailed appendices, code, and tutorials makes complex control theory (RHP zeros, SVD, QP) accessible to plasma physicists and control engineers.

In conclusion, the paper successfully bridges the gap between theoretical control design and practical tokamak operation, showing that a systematic, decoupled IBSC approach can resolve long-standing stability issues on NSTX-U and enable more aggressive plasma shaping.