Applications of a novel model-based real-time observer for electron density profile control experiments in TCV

This paper demonstrates the successful application of a novel RAPDENS-based real-time observer on the TCV tokamak to control electron density profiles across diverse plasma scenarios, including detachment studies, L-mode heating, and high-performance H-mode operations, thereby enhancing confinement, reproducibility, and robustness against diagnostic limitations.

The Gist

Imagine you are trying to bake the perfect cake, but instead of flour and sugar, you are cooking with super-hot plasma (a soup of charged particles) inside a giant, magnetic donut-shaped oven called a Tokamak.

The goal? To make this plasma hot and dense enough to fuse atoms together and create clean, limitless energy (like the sun does). But there's a catch: if you get the recipe wrong, the cake collapses, or worse, it explodes and ruins the oven.

This paper is about a new, super-smart "Smart Chef" (an observer) that helps the Tokamak kitchen run perfectly. Here is the breakdown of what they did, using simple analogies.

1. The Problem: The "Blind" Chef

In the past, the chefs (the control systems) had to guess how much "dough" (electron density) was in the oven. They had two main tools:

• The Interferometer: Like a laser pointer that shoots through the whole cake to measure the average thickness. It's fast, but it gets confused if there's too much "flour dust" (impurities) floating around the edges.
• Thomson Scattering: Like taking a high-resolution photo of a specific slice of the cake. It's very accurate, but it only happens 60 times a second (slow compared to the laser).

The Issue: When the cake gets too dense or the edges get messy (a "detached" state), the laser pointer gets confused by the dust at the edges. The chef thinks the cake is thicker than it really is and stops adding ingredients, causing the cake to collapse.

2. The Solution: The "Super-Observer"

The researchers built a Multi-Rate Observer. Think of this as a Smart Chef's Assistant that combines the best of both tools:

• It takes the fast, blurry average from the laser.
• It takes the slow, sharp photos from the camera.
• It uses a mathematical recipe book (a physics model called RAPDENS) to guess what's happening in between.

By fusing these together, the assistant creates a live, 3D map of the entire cake, knowing exactly how thick the dough is in the center and how thin it is at the edges, even if the laser is confused by dust.

3. What They Tested (The Experiments)

The team tested this Smart Assistant on the TCV Tokamak (a test kitchen in Switzerland) in three different scenarios:

A. The "Detachment" Test (Cleaning the Oven)

• The Goal: In fusion, you need to cool down the very edge of the plasma so it doesn't melt the oven walls. This is called "detachment."
• The Old Way: The laser would see the cooling dust at the edge and think the whole cake was getting too thick. It would stop feeding gas, and the cake would get too thin and unstable.
• The New Way: The Smart Assistant knows the difference between the "cake" (inside the magnetic ring) and the "dust" (outside the ring). It keeps the cake at the perfect thickness while letting the edge cool down safely. Result: They could control the edge cooling without messing up the core.

B. The "Microwave" Test (Cooking without Burning)

• The Goal: They used microwaves (Electron Cyclotron Heating) to heat the center. But if the plasma gets too dense, the microwaves bounce off like a mirror (the "cutoff" limit) and don't heat anything.
• The New Way: The assistant monitored the density exactly where the microwaves were hitting. It adjusted the gas valve in real-time to keep the density just below the "bounce-off" limit.
• Result: They could heat the plasma efficiently without the microwaves getting blocked. They even figured out how different heating methods (microwaves vs. particle beams) changed the shape of the cake, allowing them to tweak the recipe on the fly.

C. The "High-Performance" Test (The Perfect Cake)

• The Goal: Run the oven at maximum power (High-Performance H-mode) without it exploding.
• The Challenge: High power usually means high density, which risks hitting the "density limit" (the point where the plasma collapses).
• The New Way: The assistant controlled two things at once: the pressure (Beta) and the edge density. It acted like a tightrope walker, keeping the plasma right on the edge of stability without falling.
• Result: They achieved a very high-performance state that is stable and repeatable. Even if the plasma shape shifted or the sensors got "glitchy" (fringe jumps), the assistant corrected the data and kept the show going.

4. Why This Matters for the Future

This isn't just about one experiment; it's about building the next generation of fusion power plants (like ITER or DEMO).

• Robustness: Future power plants will be huge and harsh. Sensors might fail or get dirty. This "Smart Assistant" can guess what's happening even if a sensor breaks, making the plant safer.
• Automation: It allows the plant to run itself, adjusting the "recipe" automatically to stay in the "Goldilocks zone" (not too hot, not too cold, not too dense).
• Universal: Because it uses physics models rather than just raw sensor data, it works regardless of which specific machine you are on.

The Bottom Line

The researchers built a digital twin of the plasma's density. Instead of blindly guessing based on a single sensor, this system uses a smart combination of fast data, slow data, and physics math to see the whole picture. This allows fusion scientists to cook the perfect energy-generating plasma without burning the kitchen down.

Technical Summary

1. Problem Statement

Real-time control of tokamak plasmas requires precise estimation of kinetic quantities, particularly the electron density profile ($n_e$), to ensure stable operation, maximize fusion performance, and avoid disruptions (e.g., density limit disruptions). Current challenges include:

• Diagnostic Limitations: Traditional density control relies heavily on line-integrated interferometry (FIR). In high-density or detached regimes, these signals suffer from "pickup" of density in the Scrape-Off Layer (SOL) and divertor regions, leading to inaccurate estimates of the core/plasma density.
• Signal Corruption: High-density plasmas often cause Far-Infrared (FIR) interferometer fringe jumps and reduced signal-to-noise ratios, causing data loss or errors.
• Profile Control Complexity: Controlling specific local density points (e.g., to stay below ECH cutoff limits) or edge density (for proximity control) is difficult when the relationship between global fueling and local profile shape is non-linear and influenced by heating schemes (ECH, NBI) that alter particle transport (pinch vs. diffusion).
• Machine Agnosticism: Current control strategies often depend on specific diagnostic geometries, making them hard to transfer to future reactors like ITER.

2. Methodology

The authors present the integration and application of a novel multi-rate Extended Kalman Filter (EKF) observer based on the RAPDENS (RApid Plasma DENsity Simulator) model into the TCV Plasma Control System (PCS).

• The Model (RAPDENS):
  • Solves a 1D flux-surface averaged electron density transport equation coupled with 0-D neutral inventories (vacuum chamber and wall).
  • Incorporates source terms for ionization, recombination, recycling, and fueling (gas puffing, NBI, pellets).
  • Uses a control-oriented approach with cubic B-splines for spatial discretization and Crank-Nicolson for time stepping.
• The Observer (Multi-rate EKF):
  • Data Fusion: Fuses high-frequency FIR data (200 kHz, filtered to 10 kHz) with lower-frequency Thomson Scattering (TS) data (60 Hz).
  • State Estimation: Estimates the spline coefficients of the $n_e$ profile and neutral inventories.
  • Adaptive Transport: A key innovation is the real-time estimation of the electron pinch velocity-to-diffusivity ratio ($\nu/D$). Using the steady-state form of the transport equation and real-time TS data, the observer dynamically updates the $\nu/D$ profile to match observed particle pumpout or peaking effects.
  • Fault Tolerance: The EKF includes a "Diagnostic Faults Detect and Correct" block. It uses the innovation residual to identify and correct FIR fringe jumps or offsets, effectively reconstructing corrupted signals using the physical model and valid TS data.
• Control Scheme:
  • Integrated into the SAMONE (Supervisory Actuator Manager) framework.
  • Uses PI controllers with anti-windup and feedforward terms to modulate gas valve flux (and NBI power in some cases).
  • Controls derived quantities: line-averaged density within the Last Closed Flux Surface ($NEL_{LCFS}$), local density at specific radii, and the critical edge density fraction ($f_{crit,edge}$).

3. Key Contributions

1. Decoupling Upstream and SOL Conditions: The observer successfully reconstructs the density profile strictly within the LCFS, filtering out the SOL density contribution that plagues traditional interferometer-based control. This allows for independent control of upstream conditions during divertor detachment studies.
2. Real-Time Adaptive Transport Modeling: The paper demonstrates the first real-time estimation and adoption of the time-varying $\nu/D$ ratio. This significantly improves spatial accuracy, particularly in capturing profile peaking changes induced by different heating schemes (ECH vs. NBI).
3. Robustness to Diagnostic Failures: The system demonstrates the ability to reconstruct density profiles and control plasma states even when FIR channels suffer from fringe jumps, by relying on the model and available TS data.
4. Device-Agnostic Proximity Control: The observer enables the control of the critical edge density fraction ($f_{crit,edge}$) based on magnetic flux surfaces rather than specific diagnostic chord locations. This allows for consistent disruption avoidance strategies across different machines (TCV, AUG, JET).

4. Experimental Results

The methodology was validated through several TCV experiments:

• Detachment Studies (Shot #80728, etc.):
  • Traditional Control: Failed to maintain constant upstream density ($NEL_{LCFS}$) as SOL density pickup increased with nitrogen seeding and leg length changes.
  • RAPDENS Control: Successfully maintained $NEL_{LCFS}$ at the target despite varying divertor geometries and impurity seeding, proving the decoupling of upstream and SOL conditions.
• Local Density Control below ECH Cutoff (Shot #82913, #82893):
  • Controlled the central density ($n_e(\rho=0)$) to stay below the ECH cutoff limit ($4.0 \times 10^{19} m^{-3}$) despite disturbances from ECH and NBI heating.
  • ECH Effects: Observed "particle pumpout" (flattening of the profile) driven by Trapped Electron Mode (TEM) turbulence, confirmed by GENE gyrokinetic simulations. The observer adapted the $\nu/D$ ratio to track this.
  • NBI Effects: Observed profile peaking due to inward pinch and core ionization.
  • The observer provided accurate local density references, allowing the controller to react to heating-induced transport changes.
• High-Performance H-Mode Disruption Avoidance (Shot #80525, #82876):
  • Simultaneously controlled the toroidal beta ($\beta_{tor}$) and the critical edge density fraction ($f_{crit,edge}$).
  • Achieved high performance ($\beta_N \approx 2.15$, $f_{GW} \approx 0.80$) while maintaining a safe distance from the disruption boundary.
  • Demonstrated that the observer-based edge density metric is independent of plasma shape changes and diagnostic geometry, unlike raw FIR signals which drifted during ramp-down.
  • Successfully reconstructed corrupted FIR signals in real-time.

5. Significance and Outlook

• Future Reactor Readiness: The work addresses critical needs for ITER and DEMO, where diagnostics will be limited, and operation must be close to physical limits. The "device-agnostic" nature of the control metric is crucial for cross-machine scalability.
• Integrated Control: The ability to estimate transport coefficients and reconstruct profiles in real-time paves the way for Model Predictive Control (MPC) strategies that can actively manage particle transport, not just density levels.
• Robustness: The fault-tolerant capability ensures that control systems can continue operating during diagnostic failures, a requirement for reliable power plant operation.
• Future Work: The authors plan to implement inhomogeneous boundary conditions to improve edge reconstruction accuracy further and integrate the observer with other kinetic estimators (RAPTOR, TORBEAM) for full kinetic equilibrium reconstruction.

In summary, this paper demonstrates a significant leap in tokamak control capabilities, moving from simple integral density control to sophisticated, model-based profile control that is robust to diagnostic errors and adaptable to complex plasma physics phenomena.