International Doctoral College in Fusion Science and Engineering
 Thesis catalogue
Bringing first principle transport models into the tokamak control room
PhD Code: 2016-DC-03:
  • Host institute 1: AM08-Eindhoven University of Technology (Home University) - FP8-Institut de Recherche sur la Fusion par confinement magnétique, Saint-Paul-lez-Durance, France (Home Institution)
  • Host institute 2: AM07-Aix- Marseille Université (Host University)
  • Host institute 3: AM04-ITER Organization - AM02-Dutch Institute for Fundamental Energy Research
Research fields:
  • F1. Tokamak physics for ITER and beyond
  • Prof. Marco De Baar (promotor)
  • Prof. Yann Camenen (co-promotor)
  • Dr. Clarisse Bourdelle (mentor)
  • Dr. Alberto Loarte (mentor)
  • Dr. Jonathan Citrin (mentor)
Contact Person and email: Clarisse Bourdelle -

Subject description
Background: In fusion devices such as tokamaks, the achievement of good energy confinement is a key issue. The energy, particle and angular momentum transport is dominated by turbulent mechanisms. To under-stand, model, predict and control temperature, density and rotation in existing and future tokamaks, turbulence in tokamak plasmas need to be modelled. Nonlinear gyrokinetic codes allow for detailed understanding and modeling of turbulent transport. However, their computational demand precludes their use for predictive and control issues. Predicting and controlling the confinement of energy, particle and momentum in tokamak plasmas is required to optimize the operation of existing and future machines. A turbulent model based on the quasi-linear gyrokinetic model, such as QuaLiKiz, allows for a 1 million speedup while retaining key physics [Bourdelle, Phys. of Plasmas 2007]. With such tools, one second of real experimental plasma can be modelled within 100 hours of computation time, meaning a few hours when using computers in parallel. They can be used for interpretative and predictive work. Nonetheless, they are still too slow for real time control issues. To tackle this issue, recently a database of gyrokinetic runs has been developed and regularized non-linear regression has been performed on a 5D database as a proof of principles [Citrin, Nuclear Fusion 2015].

Expected outcomes

Objective: The objective is to generalize the proof of principle previously realized of neural network emulation of first principle transport models. This will be by fitting a multilayer perceptron neural network for regression of a full dimensionality (20D) quasilinear transport model database. This will allow the reproduction of the gyrokinetic results while gaining a five order of magnitude improvement in computation time. This will mean that 1 s of plasma could be modeled within 30 ms. Such developments permit, for the first time in fusion history, to use first principle based transport models for real time control applications. The proposed PhD topic covers various aspects from fundamental turbulent transport understanding to tokamak control room real time issues. The challenges on the fundamental physics aspects are in building saturation rules using the database of linear runs that can extend to electromagnetic stabilizing effects. On the numerical methods front, the way to populate a 20 D database in an optimized way will have to be investigated, through automated execution of the codes in experimentally relevant parameter subspaces. Various regression models can also be envisaged. Finally, the PhD work will allow moving from the existing proof of principle to a universal, robust and widely used first principle turbulent transport model. The usage of such a tool will deeply modify the preparation and the optimization of planned experiments. It will also allow maximizing core temperatures thanks to physics based real time control. The model will be incorporated in RAPTOR real time control suite [Felici PPCF 2012]. The tokamak plasmas studied would be WEST in Cadarache, France, ASDEX Upgrade in Germany, TCV in Switzerland as well as extrapolation to ITER. Time line and mobility scheme (research need to be performed for at least six month in two different countries): November 2016 - May 2017 : IRFM, France, introduction to transport physics June 2017 - May 2018 : Technical University of Eindhoven, Netherlands, 20D database and neural network development and testing in real control framework June 2018 - May 2019 : IRFM, France, real time control application for WEST and other tokamaks (TCV, ASDEX Upgrade, ITER) June 2019 - October 2019 : Technical University of Eindhoven, Netherlands, PhD report redaction, defense preparation



Original document: 2016-DC-03

Search website:
Last update on 24/05/2017 -
Management login
Site map