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

Real-time control of tokamak plasmas encompasses sustaining a high-performance stationary state, avoiding disruptions, and managing ramp-up and ramp-down phases. Real-time estimation and control of electron density is fundamental for monitoring and controlling particle confinement, heating efficiency, exhaust conditions, impurity concentration, fusion power, and proximity to the density limit. Building on the integration of a multi-rate observer based on RAPDENS into the TCV control system, this study explores its application to density profile control for detachment studies, ECH, and NBH L-mode plasmas, and high-performance H-mode scenarios. TCV experiments demonstrate the observer’s capability to support detachment studies in complex divertor geometries, controlling the line-averaged density within the last-closed flux surface while rejecting interferometer pick-up from Scrap-Off Layer density in the divertor. The estimated density profile enables local control of central density in ECH/NBH L-mode plasmas below cutoff conditions; heating-induced profile peaking modification is treated as a disturbance to the control task. Real-time estimation of time-varying transport coefficients, such as the pinch velocity-to-diffusivity ratio, improves model predictive capabilities, and the underlying turbulent transport is characterized via linear and non-linear gyrokinetic simulations with GENE. Simultaneous control of edge-normalized density and toroidal beta in H-mode plasmas is then demonstrated, yielding good confinement, scenario reproducibility, and a diagnostics-independent edge-density metric, while avoiding density limits and diagnostic faults propagation.

💡 Research Summary

This paper presents a comprehensive development and experimental validation of a novel model‑based real‑time observer for electron density profile estimation and control on the TCV tokamak. The observer builds on the RAPDENS framework, a physics‑based 1‑D flux‑surface‑averaged electron density transport model coupled with 0‑D neutral and wall particle balance equations. By embedding this model in a multi‑rate Extended Kalman Filter (EKF), high‑frequency far‑infrared interferometer (FIR) data and low‑frequency Thomson scattering (TS) measurements are fused to produce a spatially resolved density profile together with real‑time estimates of the transport coefficients: the diffusion coefficient D(ρ,t) and the pinch velocity ν(ρ,t).

The implementation integrates the observer into the TCV Plasma Control System (PCS). At each control cycle the RAPDENS model predicts the evolution of ne(ρ) using the current estimates of D and ν, while the EKF correction step incorporates the latest FIR line‑integrated density and TS point measurements. The resulting profile is reduced to derived quantities such as line‑averaged density inside the last‑closed flux surface (LCFS), central density, edge‑normalized density, and toroidal beta (βt). These quantities feed anti‑windup PI controllers that modulate gas puffing, pellet injection, and neutral beam injection (NBI) to achieve the desired density set‑points.

Three experimental campaigns demonstrate the observer’s versatility.

1. Detachment studies – In divertor configurations with strong SOL contributions, the FIR interferometer picks up density from the scrape‑off layer, corrupting conventional control. The multi‑rate observer separates the SOL component using the reconstructed profile, allowing precise control of the LCFS‑averaged density while avoiding the density limit. This enables stable detachment without triggering disruptions.

2. ECH/NBI L‑mode experiments – Central density is kept below the cutoff for electron cyclotron heating (ECH) by controlling a specific radial point of the profile. Heating‑induced peaking is treated as a disturbance; the observer‑based PI loop compensates in real time. Simultaneously, a pronounced “particle pump‑out” is observed when NBI is applied. Linear and nonlinear gyrokinetic simulations with the GENE code reproduce the effect, attributing it to a reduction of the pinch term and enhanced turbulent diffusion.

3. High‑performance H‑mode – Simultaneous regulation of edge‑normalized density and toroidal beta is achieved, yielding reproducible H‑mode discharges with βN≈2.15 and a Greenwald fraction of ≈0.80. The observer uses only a subset of FIR channels yet attains edge‑density accuracy comparable to offline TS, and automatically corrects fringe‑jump errors, improving robustness against diagnostic faults.

Key advantages of the observer‑based approach over traditional line‑averaged density feedback are: (i) the ability to target density at arbitrary radial locations; (ii) real‑time adaptation of transport coefficients, which reduces model prediction error during rapid actuator changes (e.g., heating switches, pellet injections); (iii) intrinsic fault tolerance, as the EKF can reject or reconstruct corrupted diagnostic signals, preventing propagation of errors through the control system.

The authors argue that such a model‑based, multi‑rate observer is essential for future reactors like ITER and DEMO, where diagnostic access will be limited and reliability is paramount. By providing a physics‑consistent, real‑time estimate of the plasma state and its transport properties, the system enables advanced feed‑forward and feedback control strategies, improves disruption avoidance, and supports high‑density, high‑performance operation. The work thus establishes a new paradigm for real‑time plasma state estimation and control in magnetic confinement fusion.