Simulation of magnetic control of the plasma shape on the DEMO tokamak
M. Ariolaa,b,∗, A. Pirontic,b , R. Ambrosinoc,b , M. Matteid,b , W. Biele , T. Frankef,g
a Dipartimento
di Ingegneria, Università degli Studi di Napoli Parthenope, 80143 Napoli, Italy
b Consorzio CREATE, via Claudio 21, 80125 Napoli, Italy
c DIETI, Università degli Studi di Napoli Federico II, 80125 Napoli, Italy
d Dipartimento di Ingegneria, Università degli Studi della Campania “Luigi Vanvitelli”, 80131 Aversa, Italy
e Institute of Energy and Climate Research, Forschungszentrum Jülich GmbH, Germany
f Max-Planck-Institut für Plasmaphysik, Garching, Germany
g EUROfusion Programme Management Unit, PPPT, Boltzmannstr. 2, D-85748 Garching, Germany
Abstract
About 85% of the primary energy is currently obtained from fossil sources. In the next future, nuclear fusion can significantly
contribute to energy production: the fuel is in principle unlimited and radioactive waste are short lived. Next steps in fusion
research are represented by the two tokamaks ITER, which is under construction, and DEMO, which is in its conceptual design
phase. In this paper we focus on a specific aspect of DEMO design, that is indeed crucial for tokamak safe operation: plasma
vertical and shape control. It is well known that a plasma with elongated cross-section exhibits a vertical instability that needs to be
feedback controlled. Typically on operating tokamaks, and in ITER, this task is accomplished using in-vessel actuator coils. Since
in the present DEMO design these coils are not foreseen, hereafter we assumed that all the actuator coils, located outside the vessel,
are used at the same time to guarantee both vertical stabilization and shape control, resorting to a suitable geometric decoupling.
The performance of the controller is shown in simulation using a nonlinear evolution code during a plasma H-L back-transition.
Keywords: Plasma Vertical Stabilization, Magnetic Control, Magnetic Sensors
1. Introduction
This paper describes the preliminary design of a position,
current and shape control for DEMO tokamak, based on the
availability of magnetic sensors. The controller is designed
basing on the CREATE-L model [1] of the DEMO 2017 SingleNull configuration. The design follows the guidelines presented
in [2], and it consists of various loops: i) a Vertical Stabilization
fast controller; ii) a shape controller; iii) a PF and CS current
controller; and iv) the plasma current controller. The controller
performance has been assessed in the presence of a given set
of events. In this paper we will present the results for a
so-called loss of power which occurs during a plasma H-L
back-transition. In the closed-loop simulations, simplified yet
realistic models of the actuators and of the magnetic diagnostics
are included.
2. Description of the plant and controller architecture
The interaction between the plasma and conducting
structures surrounding can be approximately described around
an operational point by the following linearized time-invariant
model
L11 ẋPF + L12 ẋe + E1 ẇ = uPF + kVS 1 uVS 1
(1a)
L12 ẋPF + L22 ẋe + R22 xe + E2 ẇ = 0 ,
(1b)
∗ Corresponding
author. E-mail: ariola@uniparthenope.it
Preprint submitted to Elsevier
where we use the fact that in DEMO the active coils are
superconductive and hence there is no resistive term in the first
equation. All the quantities indicate deviations with respect to
a nominal value. With the vector xPF we denote the currents
in the 11 active coils (6 PF coils and 5 CS coils); with the
vector xe the currents in the passive circuits and the plasma
current. The vector w contains the two parameter li and β p ,
which are assumed to be disturbances, as explained in [1].
Finally the vector uPF contains the voltages provided by the
main converters, which have the task to supply the currents
needed for scenario and shape control, whereas uVS 1 is the
voltage provided by the fast converter acting on the imbalance
circuit (see Fig. 1), whose task is to supply the current needed
for vertical stabilization. The imbalance circuit VS 1 makes use
of 4 PF coils; the vector kVS 1 in (1a) is used to select the coils.
The main converters V P2 − V P5 and the imbalance circuit
converter VS 1 act on the same coils PF2 − PF5. Hence
differently from other tokamaks, where there are coils dedicated
to the vertical stabilization of the plasma configuration, in this
case it is not possible to have a structural decoupling between
the plasma current, position and shape position controller, and
the vertical stabilization controller. Since it is convenient to
have the possibility to decouple these two controllers, and
to design them separately, in this paper we will resort to a
geometric decoupling based on a suitable decomposition of the
state subspace constituted by the 11 PF and CS coil currents.
As shown in Figure 2, the controller has the following
structure:
January 17, 2019
been vertically stabilized with a suitable separate loop, it acts
as a disturbance for this feedback system. Indeed the dynamic
evolution of the eddy current is usually much faster than the
evolution of the current flowing in the PF and CS coils; hence a
singular perturbation approach can be used to reduce the model
as it has been shown in [3]. Moreover the presence of the
plasma produces only slight modifications to the inductance
matrix L11 . Under these assumptions, the system equations are
then reduced to
Figure 1: The imbalance circuit used for DEMO vertical stabilization. This
circuit uses 4 PF coils. The imbalance current Iimb is given by Iimb = IPF2 +
IPF3 − IPF4 − IPF5 .
I p controller
IFF
✲ ❞✲
✻
references
Shape
controller
✲❄
❞✠
✲
✻
IFB
PF & CS
current
controller
✲ ❞✲
✻
V
Tokamak
L11 ẋPF = uPF + kVS 1 uVS 1 .
(2)
The equation of the current decoupling controller are given by
reference on I p
❞✛
Ip
geometrical
✲
descriptors
✲
PF & CS currents
✲
uPF = L11 Λ(xPF,re f − xPF )
(3)
where Λ = I/τ, being τ an assigned value determining the
time constant of a first order circuit which describes the map
between the references and the actual values of the PF and CS
coil currents. Substituting (3) in (2), the closed loop system
equations (again neglecting the eddy currents and the plasma
current) are
Vertical
controller
Figure 2: A schematic representation of the plasma control feedback scheme.
IFF indicates the scenario feedforward currents, whereas IFB indicates the
feedback current deviations; V are the total voltages provided by the main
converters.
ẋPF = −ΛxPF + ΛxPF,re f + bVS 1 uVS 1 ,
(4)
−1
where bVS 1 = L11
kVS 1 . Recalling that there are 11 active coils,
the controllability subspace from the input uVS 1 is given by
h
i
XVS 1 = R bVS 1 ΛbVS 1 · · · Λ10 bVS 1 = R[bVS 1 ] .
a) Plasma current, position and shape controller. This
controller also integrates with the supervisor providing the
feedforward (IFF ) currents needed to track a desired plasma
scenario. This controller is in turn divided in three components:
i) Current decoupling controller; ii) Shape controller; iii)
Plasma current controller.
b) Vertical stabilization controller.
It follows that the space of the active coil currents
⊥
is decomposed in two subspaces XVS 1 and XVS
1 where
⊥
dim(XVS 1 ) = 1 and dim(XVS 1 ) = 10. From this, it is clear
that the VS 1 power supply will generate currents which are
proportional to the vector bVS 1 ; this fact will be used in the
shape controller design.
The current decoupling controller acts on an intermediate
timescale; its task is to evaluates the voltages to be applied
to the plant in order to track the reference PF and CS coil
currents which are the sum of the scenario feedforward currents
(IFF ) and the feedback current deviations (IFB ) coming from
the shape controller to compensate for unforeseen disturbance.
The shape controller acts on a slower timescale with respect
to the current decoupling controller. The plasma current
controller has the aim of keeping the plasma current close
to its reference minimizing the cross-coupling with the shape
controller. Finally, the vertical stabilization controller acts on
the fastest time scale and uses as actuator the imbalance circuit
converter.
3.2. Design of the vertical stabilization controller
The vertical stabilization controller is designed on the basis
of overall system equation (1) considering the presence of the
current decoupling controller (3). Hence the equations read
L11 ẋPF + L12 ẋe + L11 ΛxPF + E1 ẇ
= L11 ΛxPF,re f + kVS 1 uVS 1
(5a)
L12 ẋPF + L22 ẋe + R22 xe + E2 ẇ = 0 .
(5b)
From equation (5a) it is clear that the effect of the current
decoupling controller is to introduce a resistance matrix (equal
to L11 Λ) in the PF and CS coil circuits. As it is shown for
example in [3], when the coils used to vertically stabilize the
plasma are superconductive, a constant gain feedback loop on
the plasma vertical speed is sufficient to obtain a stable closed
loop system; this is no more true when the stabilizing coils are
resistive. In these cases the proportional gain on the vertical
loop has to be combined with a proportional gain on the currents
of the active coil circuits (see for example [4] for the JET
tokamak, and [5] for the ITER tokamak).
In DEMO the current generated by the VS 1 converter is
spread (with different weights) to the four coils PF2 − PF5,
3. Controller design
3.1. Design of the current decoupling controller
The design of the current decoupling controller is done on
a plasmaless model, by neglecting the presence of the eddy
currents in such a way that in dry discharges (i.e. discharges
with no plasma) the current references are tracked with a certain
accuracy. When the plasma is present, assuming that it has
2
Therefore from (8), recalling that Λ = 1τ I, we obtain
x̃˙ PF = −Λ x̃PF + Λ x̃PF,re f
g = Cg1 T 2 x̃PF + Cg1 T 1 xVS 1 .
(10a)
(10b)
The plasma-wall gap controller is chosen to be a multivariable
PI controller in the form
I
x̃PF,re f = KP (gre f − g) + KI L−1
∗ (gre f − g) .
(11)
s
The current component xVS 1 in (10b) is generated by the VS 1
converter; it acts as a disturbance for the plasma-wall gap
controller. Finally the reference to the current decoupling
controller is given by (9b)
Figure 3: The six gaps that are controlled in DEMO
xPF,re f = T 2 x̃PF,re f .
and, because of the inductive coupling, to all the other coils.
Our choice was to feedback together with the vertical speed
the projection of the actual PF and CS coil current on the XVS 1
subspace. Hence the voltage applied to the VS 1 converter is
given by
3.4. Design of the plasma current controller
The last loop to be designed is a PI controller on the plasma
current. The gains of the PI controllers have been obtained by
means of a parametric optimization aimed at minimizing the
cross-coupling with the other loops, especially with the shape
controller. The settling time has been chosen in the order of
40–50 s.
uVS 1 = k1 ż + k2 T 1T xPF = k1 ż + k2 xVS 1 ,
(6)
where z indicates the plasma centroid vertical position and T 1 =
bVS 1 /kbVS 1 k is an orthonormal basis of XVS 1 .
4. Simulation results
The controller performance has been assessed in the presence
of various events; due to the lack of space, hereafter we will
only show the case of a so-called loss of power which occurs
during a plasma H-L back-transition. This disturbance is
modeled as a β p drop of about 0.4 in 4 s while li is more or
less constant.
The closed-loop simulations have been carried out under the
following conditions:
3.3. Design of the shape controller
The plasma shape in DEMO is controlled by means of six
plasma-wall gaps (see Fig. 3), whose deviations with respect to
the nominal values can be described by the linear equation
g = Cg1 xPF + Cg2 xe + Fg w .
(7)
It can be indeed demonstrated (see [6]) that choosing the gaps
in a suitable way, it is possible to control the overall plasma
boundary by means only of a “few” gaps. Once the plasma
is stabilized by the vertical stabilization controller, on a slow
timescale the dynamic system describing the evolution of the
gaps can be approximated by
ẋPF = −ΛxPF + ΛxPF,re f
g = Cg1 xPF .
• taking into account voltage saturations on the PF and CS
voltages;
• including a simplified, yet realistic model for the actuators,
consisting of a first-order filter plus delay in the form
e−sτ1
1+sτ2 , where the values for the delay τ1 and the time
constant τ2 have been derived from a study conducted in
ITER [7];
(8a)
(8b)
The references on the PF and CS coil currents are used to
control the selected plasma-wall gaps. Since we want a minimal
interaction between the shape and the vertical stabilization
controllers, the current references generated by the shape
⊥
controller should lie on the subspace XVS
1.
Let T 2 be a matrix whose columns are an orthonormal basis
⊥
of the XVS
1 subspace. Recalling that T 1 is a basis of XVS 1 it
holds that T 2T T 1 = 0. Hence we can write
xPF = T 1 xVS 1 + T 2 x̃PF
xPF,re f = T 2 x̃PF,re f .
• assuming that the various geometric controlled quantities,
i.e. the centroid vertical position and the six gaps, are
reconstructed basing on the measurements of magnetic
sensor located inside the vessel;
• absence of noise on the sensor measurements.
Fig. 4 and Fig. 5 show the results of a linear simulation in
presence of the loss of power disturbance, in terms of controlled
gap deviations and control power. Under the conditions
specified before (use of magnetic in-vessel sensor with no
noise), the disturbance is fully controllable, but with a peak
active power of about 700 MW. It remains an open issue at
this time whether and how the DEMO operator will be able
(9a)
(9b)
From (9a) it is straightforward to obtain that
x̃PF = T 2T xPF .
3
Figure 4: Deviation of the controlled gaps during a loss of power
Figure 6: Closed-loop behavior in the presence of a loss of power using
the CREATE NL code. The figure shows the plasma boundary before the
disturbance is applied (in blue) and the boundary after 4 s, when the plasma–
wall distance reaches its minimum of about 6 cm
power is very large and the possibility of providing such
large amount needs to be ascertained.
The results of the closed-loop simulations are useful to direct
the possible modifications of the DEMO design, in order to
converge to a DEMO configuration which could guarantee the
best possible reliability.
Figure 5: Control power during a loss of power
to provide this amount of control power at any time. Hence the
conclusion on the controllability of the plasma will depend on
the provision of control power. The controller performance
have been then tested using the nonlinear evolution CREATENL [8] model. The nonlinear simulation confirms the linear
prediction: the minimum plasma-first wall distance is of
about 24 cm at equilibrium and becomes of about 6 cm after 4
seconds the disturbance is applied, as shown in Fig. 6, where
the plasma boundary is shown at t = 0 and at t = 4 s.
Acknowledgments
This work has been carried out within the framework
of the EUROfusion Consortium and has received funding
from the Euratom research and training programme 20142018 under grant agreement No 633053. The views and
opinions expressed herein do not necessarily reflect those of
the European Commission.
5. Conclusions
References
In this paper we have presented the preliminary design of a
current, position and shape controller for the DEMO tokamak.
The controller performance have been tested using the model
of the DEMO 2017 Single-Null configuration. At the present
stage there are many critical issues that need to be tackled in
order to ascertain whether the present DEMO configuration is
controllable or not:
[1] R. Albanese and F. Villone. The linearized CREATE-L plasma response
model for the control of current, position and shape in tokamaks. Nucl.
Fus., 38(5):723–738, May 1998.
[2] M. Ariola and A. Pironti. Magnetic Control of Tokamak Plasmas. Springer,
Second edition, 2016.
[3] M. Ariola and A. Pironti. An Application of the Singular Perturbation
Decomposition to Plasma Position and Shape Control. Eur. J. Control,
9:410–420, 2003.
[4] M. Lennholm et al. Plasma control at JET. Fus. Eng. Design, 48(1–2):37–
45, Aug. 2000.
[5] G. Ambrosino, M. Ariola, G. De Tommasi, and A. Pironti. Plasma
Vertical Stabilization in the ITER Tokamak via Constrained Static Output
Feedback. IEEE Trans. Contr. Sys. Tech., 19(2):376–381, Mar. 2011.
[6] A. Pironti and A. Portone. Optimal choice of the geometrical descriptors
for tokamak plasma shape control. Fus. Eng. Des., 43(2):115–127, Dec.
2008.
[7] E. Gaio, R. Piovan, V. Toigo, and I. Benfatto. The control system of
the ITER vertical stabilization converter. Fus. Eng. Des., 66–68:719–725,
Sept. 2003.
[8] R. Albanese, R. Ambrosino, and M. Mattei. CREATE-NL+: A robust
control-oriented free boundary dynamic plasma equilibrium solver. Fus.
Eng. Des., 96–97:664–667, Oct. 2015.
• in our design we rely on the availability of magnetic
sensors located inside the vessel. It is still an open
issue whether these sensors can survive the harsh
DEMO conditions. The use of alternative sensors, such
as microwave reflectometry and other diagnostics like
polarimetry/interferometry needs to be investigated;
• the present list of disturbances to be rejected
controller is still preliminary. Anyway some
disturbances seem to be hardly controllable: the
gets very close to the first wall and/or the needed
by the
of the
plasma
control
4