Integrated scenario in JET using real-time profile control
ABSTRACT The recent development of real-time measurements and control tools in JET has enhanced the reliability and reproducibility of the relevant ITER scenarios. Diagnostics such as charge exchange, interfero-polarimetry, electron cyclotron emission have been upgraded for real-time measurements. In addition, real-time processes like magnetic equilibrium and q profile reconstruction have been developed and applied successfully in real-time q profile control experiments using model based control techniques. Plasma operation and control against magnetohydrodynamic instabilities are also benefiting from these new systems. The experience gained at JET in the field of real-time measurement and control experiments operation constitutes a very useful basis for the future operation of ITER scenarios.
- SourceAvailable from: produtronica.pucpr.br
Article: On Control[show abstract] [hide abstract]
- [show abstract] [hide abstract]
ABSTRACT: Interaction of waves in the ion cyclotron range of frequencies (ICRF) with a plasma has a number of key properties that make them attractive beyond pure heating. First, the waves can interact resonantly with either the plasma ions or electrons. In the case of ion cyclotron damping, a small number of resonant ions are often accelerated to high energies. These ions, apart from heating the bulk plasma via Coulomb collisions, can increase fusion reactivity, affect plasma stability and drive current. They have also been invaluable in diagnostic applications and simulations of fusion-born 3.5 MeV alpha-particles. The second key property of ICRF waves is the transfer of wave momentum to the plasma. This allows one to drive current, affect plasma rotation and induce radial transport of the fast-ions with toroidally directed waves. Finally, ICRF power deposition is rather narrow and its location can be externally controlled, which has important applications in improving the plasma performance, affecting the local plasma transport and providing a tool for plasma transport studies. Representative examples from present-day tokamak experiments are reviewed to highlight the available capabilities.Plasma Physics and Controlled Fusion 01/2003; · 2.37 Impact Factor
Integrated scenario in JET using real time profile control
E. Joffrin1, F. Crisanti2, R. Felton3, X. Litaudon1, D. Mazon1, D. Moreau1,
L. Zabeo1, R. Albanese4, M. Ariola5, D. Alves6, O. Barana7, V. Basiuk1,
M. Becoulet1, J. Blum8, T. Bolzonnella7, K. Bosak8, J.M. Chareau1, M. de Baar9,
P. de Vries3, P. Dumortier10, D. Elbeze1, J. Farthing3, H. Fernandes6, C. Fenzi1,
R. Giannella1, K. Guenther3, J. Hardling3, N. Hawkes3, T. C. Hender3,
D. F. Howell3, P. Heesterman3, F. Imbeaux1, P. Innocente7, L. Laborde1,
G. Lloyd3, P. J. Lomas3, D. McDonald3, J. Mailloux3, M. Mantsinen11,
A. Messiaen10, A. Murari7, J. Ongena10, F. Orsitto2, V. Pericoli-Ridolfini2,
M. Riva2, J. Sanchez12, F. Sartori3, O. Sauter13, A. C. C. Sips14, T. Tala15,
A. Tuccillo2, D. Van Ester10, K.-D. Zastrow3, M. Zerbini2 and contributors to the
JET EFDA programme*
1Association EURATOM–CEA, CEA Cadarache, 13108 Saint-Paul-lez-Durance, France.
2Associazione EURATOM-ENEA sulla Fusione, C.R. Frascati, Frascati , Italy.
3Euratom/UKAEA Fusion Association, Culham Science Centre, Abingdon, Oxon OX14 3DB, UK.
4Associazione EURATOM/ENEA/CREATE,DIS, Universita` di Napoli Federico II, Napoli, Italy
5Associazione EURATOM/ENEA/CREATE, DIMET, Universita` di Reggio Calabria, Italy
6Instituto Superior Técnico, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
7Consorzio RFX - Associazione Euratom-Enea sulla Fusione, Corso Stati Uniti 4, I-35127 Padova.
8Universite de Nice-Sophia Antipolis, UMR 6621, Parc Valrose, 06108 Nice Cedex 02, France
9FOM-Rijnhuizen, Ass. Euratom-FOM, TEC, PO Box 1207, 3430 BE Nieuwegein, NL.
10Association Euratom-Belgian State, Royal Military Academy, Belgium.
11Association Euratom-Tekes, P.O.Box 2200, FIN-02015 HUT, Finland
12Associacion EURATOM-CIEMAT para Fusión, Avenida Complutense 22, E-28040 Madrid, Spain
13Association EURATOM-Confédération Suisse, EPFL, 1015 Lausanne, Switzerland
14IPP-EURATOM Assoziation, Boltzmann-Str.2, D-85748 Garching, Germany
15Association EURATOM-TEKES, VTT Processes, P.O. Box 1608, FIN-02044 VTT, Finland.
The recent development of real time measurements and control tools in JET
has enhanced the reliability and reproducibility of the relevant ITER scenarios.
Diagnostics such as charge exchange, interfero-polarimetry, Electron Cyclotron
Emission (ECE) have been upgraded for real time measurements. In addition, real
time processes like magnetic equilibrium and q profile reconstruction have been
developed and applied successfully in real time q profile control experiments using
model based control techniques. Plasma operation and control against MHD
instabilities are also benefiting from these new systems. The experience gained at JET
in the field of real time measurement and control experiments operation constitutes a
very useful basis for the future operation of ITER scenarios.
In the past years, the preparation of ITER scenarios [1, 2] has been the main
focus of tokamaks experimental activity and operation. As the studies are progressing,
the elaboration and operation of these scenarios demand the integration of always
more control parameters to fulfil the requirements for a fusion reactor. The active
control of plasma shape, current and pressure profiles , radiation, magneto-hydro-
dynamic (MHD) instabilities, etc… now appears as the most important development
to provide to the ITER scenarios the necessary relevance for burning plasma
operation. Attempts of real time control of internal transport barriers (ITB) , MHD
instabilities [5, 6] and current profile  have already produced promising results in
various large devices such as DIII-D, JT-60, and Tore Supra.
In this context, JET has developed since 2001 a comprehensive set of real time
diagnostics, control tools and simulation facilities for the operations of the reference
ITER scenario and advanced tokamak scenarios. This enhancement project
undertaken under the European Development Fusion Agreement (EFDA) is now
playing a decisive role in the operation of the JET device. In particular, ITER relevant
plasma scenarios have been improved in JET in their reliability and stability thanks to
the use of the real time control tools [8, 9, 10]. This confirms that in ITER a very
large effort will be devoted to the development of real time control tools. Among the
recent tools developed in JET, the real time equilibrium, together with real time
electron and ion temperature and current profiles measurements, has dramatically
enhanced the experimental work on the integration of advanced tokamak scenario and
in particular the development of the techniques to control in real time the q and
pressure profiles simultaneously .
This paper reports the technical and scientific achievements made in JET in
this domain. The developments of the new real time tools and algorithms are first
described together with respect to their relevance to real time control experiments.
Practical examples realised on relevant ITER scenario are given and the methodology
used in JET to prepare and execute real time profile control experiments for advanced
tokamak scenario is highlighted as well as the modelling activity for these
experiments. Finally, the benefit of real time control on plasma safety (such as
disruption or plasma MHD instabilities) and operation is also illustrated.
2. Developments of real time measurements and control systems in JET
To achieve extended burn with a fusion gain Q close to 10 with duration sufficient
to reach stationary conditions key physics issues  related to plasma performance
need to be fulfilled. For the relevant plasma scenario, four physics issues can be
- The control of confinement (or H factor) at sufficiently high density (n~0.8
nG) to produce the requisite fusion power and Q value. According to the
scenario, this issue is closely related to the control of the q profile (see below
in section 3)
- The control of power (loss power) and particle exhaust to ensure acceptable
levels of helium (or ashes), plasma impurities and heat load on the divertor
target. This also encompasses the control of ELMs to ensure adequate lifetime
of the in-vessel components.
- The control of global magneto-hydrodynamic (MHD) instabilities (such as
neo-classical tearing modes or resistive wall modes) and the plasma control to
reduce the effect of disruptions.
- The control of ?-particles losses via collective instabilities to enable the
transfer of ?-particle energy to the thermal plasma.
In present day tokamak, the last item can only be partially investigated using ICRH,
for example. But it requires D-T operation to be fully tested. On the other hand, all
other items are necessary together to give to the operation scenario its relevance for
burning plasma operation. As a result of the above list, the active control of a plasma
discharge will require the use of a wide range of real time sensor parameters and
Furthermore, the development of plasma operation scenario offering the
prospect of establishing reactor relevant steady state operation has motivated the use
of active profile control and has also put more demands on the flexibility of plasma
shaping, heating and current drive systems . Although the detailed conditions for
the creation of Internal Transport Barriers (ITBs) are still uncertain , the aim of
these tools is to improve the stability and reliability of this mode of operation by the
control of current and pressure profile simultaneously . In real discharges, small
deviations from the reference scenario may indeed increase with time. Therefore
feedback control of non-linearly linked parameters such as the q and pressure profile
or the confinement and loss power will be needed.
For all these reasons, JET has developed in the last two years an ambitious
enhancement programme of real time measurements and control tools (figure 1) with
the ultimate aim of assisting the development of the ITER relevant scenario. A large
number of key diagnostics have been upgraded to produce real time measurements
routinely (Table I). Real time processes such as real time equilibrium and profile
mapping have also been implemented (Table II). This was made possible by the
recent improvements in diagnostic reliability and also, by the rapidly growing
capability of computers and communication networks. The upgrades were selected by
their expected potential value to the scenario integration and the main experimental
For the confinement parameters, the new fast calculation has been based on the
JET flux boundary code XLOC  used for plasma shape control. Using magnetic
and diamagnetic data it produces plasma parameters like the diamagnetic energy
(Wdia), internal inductance (li), and plasma separatrix geometry in less than 1ms [15,
For the line integrated density the interferometer has been equipped with new fast
ADCs (Analog Digital Converter) and a new fringe jump algorithm corrector  has
been installed and validated . Together with the real time faraday rotation data
 they are Abel-inverted in less than 10ms [20, 21] to infer the density and q
profile using a plasma external geometry based on the XLOC data. To improve
further the reconstruction of the current density profile, a new Grad-Shafranov solver
 called EQUINOX  has been developed, validated and installed on JET and
also in Tore Supra . This new real time equilibrium code computes the magnetic
equilibrium and density profiles in less than 10ms for each time step. Another version
of this code includes internal flux measurements from the far-infrared polarimeter as
input to compute the current density profile with more accuracy. The flux map issued
from the EQUINOX solver is also used to reconstruct the profile data such as density
and electron and ion temperature onto the plasma flux grid.
To complement the real time measurement of the current profile, new fast ADC
have also been installed on the motional Stark effect (MSE) diagnostic  and the
pitch angles are now produced in real time with a source rate of about 25ms. In the
near future, these data will be processed in real time using the EQUINOX geometry to
compute a q profile independently from the q profile inferred from polarimetric data
For the electron and ion temperature and rotation real time profile, both the 96-
channel electron cyclotron emission radiometer [27, 28] and the charge exchange
diagnostic  have also been upgraded and connected to the communication
network,. The latter is synchronised to the neutral beam and is processing the five-
Gaussian spectral analysis in less than 50ms. These data are all re-mapped onto the
flux grid from EQUINOX and the ITB criterion ?*Te and ?*Ti  are also inferred
from this procedure for the control of the pressure profile during ITBs.
Other relevant data such as MHD magnetic signals, neutron signals, heavy
impurity lines from X-ray and visible spectroscopy , radiated power from
bolometry are complementing this ensemble. In addition some specific processing
such as ELM detector (using D? signal), Zeff and thermal energy calculations, have
been included. All algorithms of diagnostic data processing have been tested and
validated on a large number of pulses to guarantee their robustness during the
experiments. For instance, the equilibrium code EQUINOX has been tested against
EFIT on a database of more than 500 discharges with a large variety of magnetic
configurations, plasma current and toroidal field strength.
All data produced as well as the actuators data (Neutral Beam injection, Lower
Hybrid wave, Ion Clyclotron Resonance Heating, gas and pellet fuelling) have been
connected to an ATM (Asynchronous Transfer Mode) and Ethernet computer
communication network (figure 1). They are available in a Real Time Signal Server
(RTSS) and the experimental control algorithms are implemented in a Real Time
Central Controller (RTCC) . This unit is also being upgraded to facilitate the
routine use of so-called multi-input multi-output (MIMO) control schemes which is
required for current and pressure profile feedback control in particular.
3. Real time control feedback experiment for ITER scenarios
In the last campaigns, JET has strengthened its programme on the validation
of ITER scenarios. The real time control systems have played an increasing role in the
integration and reliability of the scenarios relevant for the next step. In line with the
performance assessment  three different scenarios have been considered in JET as
relevant for future ITER operation: a) the inductive ELMy H-mode scenario, b) the
steady state scenario, and c) the so-called “hybrid” advanced tokamak scenario. The
main characteristics of each of these scenarios are briefly described below together
with the feedback control scheme developed for each of them.
a- The inductive ELMy H-mode scenario
The inductive ELMy H-mode scenario in ITER  is aimed at producing Q~10
for a limited burn time of about 400s with q<1 in the plasma core. It will be run at
q95=3 ?N~1.8, and H98y~1 close to the Greenwald density (typically at 0.85 nG) and
would be conducted in ITER at high field (5.3T) and current (15MA) at a triangularity
at the separatrix of about 0.48.
Figure 2 shows an ELMy H-mode inductive scenario where the radiation
fraction has been feedback controlled by Argon injection for more than 5s. This
scheme achieves the control of the conducted and convected power on the target
plates. The scenario has been run with an ITER-like plasma configuration (?=0.4) at
90% of the Greenwald density and high frequency (~40Hz) type I ELMs. The
confinement is not dramatically degraded by the impurity injection (H89=0.91 and
?N=1.5). The feedback scheme includes the filtering of the radiation fraction and uses
both integral and derivative gains. After two seconds, the feedback controller
stabilises the pulse to the requested value of 60% of radiated power, which correspond
to a loss power of 7MW and a deposited power of about 3MW/m2 for this magnetic
configuration. In another experiment with this particular scenario  a second
feedback loop has been also coupled to control the H factor with deuterium fuelling.
This scenario has the potential to integrate simultaneous feedback control of the
confinement, the loss power and possibly the ELM frequency to control the loss
power during the ELMs and mitigate their effect on the target erosion.
b- The steady state scenario.
In the steady state non-inductive ITER scenario at Q~5 the total current at
current flat top phase is generated non-inductively by additional current drive and a
dominant fraction of bootstrap current (typically more than 50%). The q profile in the
plasma core is non-monotonic and typically between the rational surfaces 2 and 3. In
ITER, it would be run with q95=5 to 6, ?N~2.8, H98y close to 1.6 and n/nG~0.8. The
plasma current would be 9MA to maximises the bootstrap current and, with the help
of high confinement, to make this discharge steady state with Vloop~0.
Figure 3 shows a prototype in JET of a steady state scenario  with real
time control of the ion temperature gradient R/LTi with the beam power. The target
value of R/LTi has been set to 24 which, in JET, corresponds to a “non-stiff” profile
. The “no ITB” reference value is also indicated for comparison. This pulse uses
LHCD to sustain the reversed q profile after it has been pre-formed in the early phase
before 4s. At this time a wide ITB (R~3.6m) is created as qmin reaches the q=3 surface
. A second more internal ITB (R~3.3m) is also present. Real time control of R/LTi
help maintaining the outer ITB tills the end of the pulse using a proportional-integral
controller. The electron temperature profile is also showing a very steady electron
ITB as illustrated by the ?*Te criterion  in figure 3. The modest strength of these
ITBs (?*Te ~0.02) is probably preventing the accumulation of the impurities in this
discharge as revealed by the impurity analysis . Although this regime is still
operated at low density (n/nG=0.4) and not fully non-inductive (Vloop=50mV, with
35% LH-current, 35% bootstrap current and ~15% NB-current), it provides an
adequate target for implementing the control of the q profile together with the
pressure profile up to the technical limit at JET(~20s).
c- The “hybrid” advanced tokamak scenario
In this more recent mode of operation, current drive power and bootstrap
current drive a substantial fraction of the total current. The burn time in ITER would
be therefore increased significantly with respect to the inductive scenario to about
1000s. For this regime, the core q profile lies between 1 and 1.5 with a magnetic shear
close to 0. In ITER, this scenario would be operated with q95=4, ?? =2.8 and H98y
close to 1.5 and n/nG=0.8 with a plasma current of 12MA.
The hybrid scenario has been achieved recently in JET in an experiment
attempting to produce identity and similarity experiments with ASDEX-Upgrade .
Figure 4 presents an example of the hybrid scenario produced in JET  using the
feedback control of the normalised beta ?N with the beam power. The requested ?N
waveform is made of two plateaux at ?N=2 and ?N=2.8 with the idea to test the
confinement behaviour as the power increases. In this scenario, only low amplitude
3/2 and 4/3 NTM modes (both island size of the order of 3cm) are observable during
the whole discharge without major deleterious effect on the confinement . This
feedback control scheme is very relevant to this regime since the ?N real time control
could be used for preventing the growth of NTMs when they start degrading the
confinement. In this scenario, the q profile is close to q=1 as evidenced by the regular
n=1 m=1 fishbone activity throughout the discharge. The ELMs are type I, but high
frequency (~30Hz). The ELM frequency increases slightly during the second power
phase. The non-inductive current fraction of this discharge is of the order of 46%
shared equally between bootstrap and beam current. This particular discharge reached
only 50% of the Greenwald density in a low triangularity configuration (?~0.2).
However, in similar experiment have achieved 85% of the Greenwald density using
an ITER-like shape configuration with ?=0.45 . The real time control of the q
profile and of NTMs at high-normalised beta and high triangularity are therefore
amongst the most favoured control schemes considered for future experiments.
The new real time systems developed at JET have started to contribute to the
integration of ITER scenarios. As a result, the reliability of these scenarios has been
improved significantly. However, these experiments are limited to a simple real time
network using the control of one output by one actuator. In particular for the steady
state scenario, they have also revealed the need for the feedback control of profile
which requires the use of several controlled outputs by several actuators (multi-input
multi-output). Within the real time project, JET has started to develop the control
techniques and the necessary algorithms for achieving this goal; this is described in
the next section.
4. Feedback experiment using real time profile control
a. Design of a real time controller for tokamak
For the design of multi-input multi-output controller for real time profile
control, let us first consider the layout of a general control system for tokamak as
presented in figure 5. For simplicity, all the transfer functions are linearised and
represented by their Laplace transforms. This layout can be mainly divided in two
blocks: the plant and the controller. In the plant, the plasma transfer function K(s)
relates the inputs X(s) to the actuators to the outputs Y(s) measured by the sensors. In
the controller, the operator sets up the reference YREF(s), the signal conditioning F(s)
(such as filtering when required), and the gain matrix transfer G(s) which function can
be expressed in three different terms as:
C(s) . .s)
where C(s) is the control gain matrix, and ?i and ?d the integral and derivative gains
respectively also in matrix form.
In this process, the plasma transfer functions (or kernel) K(s) is most of the
time unknown, but can be identified either from power modulation experiments or by
simulation using a predictive transport code such as CRONOS  or JETTO .
Real time profile control experiments have therefore motivated the activity on plasma
transport modelling with the ultimate goal of replicating the control experiments.
Although a full plasma transport modelling cannot yet fully replace the experiments
for the design of controllers, they bring useful information on the dominant
parameters for the design of a feedback control experiment. Once the plasma transfer
function is identified, it is possible to simulate the control feedback loop and
determine the most appropriate combination of controller gains (C,??i and ?d) in G(s)
to ensure closed loop stability.
This layout does of course apply to the feedback control experiments with a
single actuator and output. In the case of the control of ?N by NBI power presented
figure 4 for example, the kernel K(s) has been modelled by a transfer function of the
form: K(s) = ?N(s)/PNBI(s)=(0.116.s+0.57) / (0.29s+1).
From this model, the coefficients of G(s) can be determined either by try and error
simulation or using experimental techniques like the Ziegler-Nichols method . In
the case of this control experiment (figure 4), the control gains in G(s) are scalars and
made of a proportional and integral term with C=8, ?i =27 (and ??d =0).
b. Experimental design of a controller for the real time control of the q profile.
For real time q profile control experiments with LHCD, NBI and ICRH
powers as actuators, the general method is to build a linear Laplace response model
around the target state to be controlled . In this case the response matrix is built
Q(s) = K(s) . P(s)
in Laplace form, where Q represents the safety factor difference vector and P the
input power difference vector with respect to a reference discharge. Here, the kernel
K(s) is determined experimentally using step or modulation experiments of the
actuators. In this procedure each actuator is stepped up or down in three different
pulses and the input power P and output Q differences are measured in their steady
state limit after about one resistive time (i.e. for s=0). Figure 6 shows the step
experiment in the case of the neutral beams. This experiment is performed on the
same type of discharge as the long ITB discharge shown figure 4. (BT=3T, Ip=1.8MA,
n=2.5. 1019 m-3). After a few seconds, the real time q data are measured and the
variation of q inferred for each five of the chosen control points (at r/a=0.2; 0.3; 0.4;
0.6; 0.8). This experiment is repeated for each three actuators. The singular value
decomposition (SVD) expansion is performed on K(0) to identify the most significant
singular values and avoid over-determination:
K(0) = W(0) . ?(0) . V+(0)
In this decomposition ?(0) contains a diagonal matrix of singular values named ?~???.
The SVD expansion is truncated to the highest singular values and provides the so-
called steady state de-coupled modal input ?(0)=V(0)+.P(0) and output
?(0)=W(0)+.Q(0) . From this analysis, it is possible to invert the truncated
diagonal matrix ?~???, and obtain a feedback control with a controller transfer
function of the form:
Here, QREF is the q profile reference to achieve, ?d =0 and ?i a is constant chosen
empirically but close to typical the current diffusion rate.
It is important to note that this method has been generalised in reference 11 to
include the control of the pressure profile using the ?*Te criterion as an input and also
includes the use of appropriate basis functions for the output and input profiles (i.e. q
and power deposition profiles).
c. Experiments with real time control of the q profile.
As a test of principle, this procedure has been first applied to the control of a
pre-defined q profile of 5 points (r/a=0.2; 0.4; 0.5; 0.6; 0.8) with one actuator only,
namely the LH power . In this case the accessible targets are of course reduced to
one family of profiles, so the reference points have been chosen close to the family
inferred from the SVD analysis. The experiment is realised during an extended LHCD
phase of 15s like those used to pre-form the q profile for the creation of an ITB
(Ip=1.3MA, BT=3T, n=2.51019m-3). The Kernel K(s) is identified from a simple LH
power step experiment. The matrix K has in this case a size of [5x1] and ? is reduced
to a scalar and ?i=1s. The real time q profile data are issued from the Abel inversion
of the polarimetric data as described in reference and in section 2. Figure 7a shows the
behaviour of the q profile traces together with their references and the LH power
produced by the controller. The q profile reaches steady state and is maintained for
about two resistive times. The LH-deposition profile (figure 7b) calculated by the ray-
tracing code DELPHINE  included in CRONOS is consistent with the gains of the
control matrix: i.e. the maximum deposited power corresponds to the maximum gain
at r/a=0.5. With this technique, reversed shear q profiles are also accessible and have
also been achieved in steady state conditions by changing the reference value of the q
After this first encouraging result, the SVD technique has been applied to the q
profile control using three actuators (i.e. LHCD, NBI and ICRH). This time, the
determination of the steady state plasma response is determined from one reference
discharge and three dedicated step down experiments (one for each actuator) as
explained in sub-section b. The Kernel K(0) is in this case a [5x3] matrix. Two out of
three singular values have been retained by the SVD analysis in matrix ? (indicating
that the accessible q profiles is only a 2-parameter family) .
Figure 8a shows the resultant feedback waveforms together with the demand
produced by the controller and the time traces of q at r/a=0.5. Figure 8b illustrates the
evolution of the q profile during the controlled phase (from 7 to 13s), demonstrating
that the selected gains were adequate and the technique effective on a time scale that
approaches the current diffusion time scale . Figure 9 shows the non-inductive
current components generated by LH and the beams at 51sas calculated by JETTO.
The separate depositions of these non-inductive currents (LHCD at mid-radius and
NBCD in the plasma core) indicate that the q profile is controlled at two different
radial points. This is consistent with the results from the above SVD analysis
indicating that the accessible q profile targets are restricted to a two-parameter profile
family. This successful experiment represents a step forward in view of future
application combining the q and pressure profile as an input in the controller.
d. Space-state method modelling for the optimisation of the controller
In order to describe the dynamic of a system to be controlled, the state-space
control method is often used in many areas since it has the advantage of handling
more than one control input or more than one sensed output . If P is the input to
the system (NBI, ICRF and LHCD) and Q the outputs (including the q profile
measurements at 11 radial locations, q95 and the loop voltage), the linear plasma
response can be expressed as: ? = A.? + B.P(1)
Q = H.????
where ? is the state variable and A, B, H are matrices. Singular value decomposition
is again used on Q to determine the principal components of the system and optimise
the number of state variables ? describing the system. A and B are then determined by
classical system identification techniques.
Figure 10 shows the modelling of the experiment described in the previous
section. The SVD technique has determined that 4 state-space variables are sufficient
to describe this system. The matrices A and B have been calculated from the step
down experiments presented in the previous section. Knowing this model it is then
possible to calculate the transient plasma response of the plasma using the Laplace
transform on equation (1) and (2), leading to:
.BA) I . H.(s
The plasma response function (depending of s) is now represented in a linear form
and can be used in a simulated control loop to determine the most appropriate PID
gains (?i and ?d) in G(s). This modelling is also a useful tool for the future design of
controller for the simultaneous feedback of the q and pressure profile.
5. Real time feedback control for plasma operation
Real time feedback control tools have also been applied in JET to the
operation of scenarios with regard to MHD instabilities and disruption in particular.
The active control of MHD modes deleterious for the confinement (such as NTMs)
has already been achieved . Real time control has also shown is efficiency in the
control of resistive wall modes in DIII-D .
Figure 11 shows an example of an NTM controlled by power step downs in
JET. This kind of experiment is not relying on the design of a PID controller but uses
a simple event driven produced by a threshold level on the input MHD signal. In this
cases the root mean square signal of the n=2 amplitude triggers a step down of the
NBI and ICRH power as it exceeds a level of 0.1 volt corresponding to an island size
of 6cm. When the NTMs induced by the sawtooth crashes reaches the threshold level
the additional power (NBI and ICRH) is stepped down. The same power level is re-
applied when the n=2 signal goes back to zero i.e. when the NTM vanishes. The NTM
strikes again a second time before being completely stabilised at 17s. Note that the
stored energy comes back to a higher level than before the NTM onset. At the end of
this sequence, an m=2 n=1 NTM is destabilised and the power is stepped down thus
preventing a likely disruption from occurring.
Disruption are currently avoided at JET by the real time step down of the main
heating when the pressure gradients are becoming too steep during an ITB scenario
for example . This is generally achieved using the neutron rate as an indicator to
limit the pressure peaking and will be further developed using the new real time
measurements. Another on-going effort is devoted to the classification of disruptions.
A recent work is aiming at classifying the disruptions in JET using the neural network
technique . This neural network is using plasma parameters as input (such as
plasma current, internal inductance, radiated power, etc…) from the past four years of
disruptions in JET. The results suggest that this technique could be used ultimately for
the prediction and control of disruption, which will be an essential asset in a tokamak
of the size of ITER.
Real time control contributes as well to the operation of specific heating
schemes and fuelling such as 3He minority heating  and pellet injection . An
example of this is presented in reference  where the concentration of 3He is being
controlled in real time at a requested level to optimise the ICRH in a mode conversion
experiment with ITBs. In this particular experiment the controller required a
derivative term (i.e. ?d?0) to account for the elapsed time between the opening of the
valve and the penetration of the gas in the discharge which is of the order of 300-
400ms in this case. The 3He concentration has been controlled in a satisfactory
manner even during the dynamic phases produced by the onset and vanishing of ITBs.
This experiment could also be used as base for the control of the fuel mixes in future
All these subjects are naturally highly relevant to the operation of the next step
and will certainly be further developed in JET to improve the safety and operation of
the main plasma scenarios.
The recent development of real time measurements and control tools in JET
has enhanced the integration of the relevant ITER scenarios. These new facilities are
now used routinely in the JET experimental campaigns and offer to scientific
community a unique integrated set of real time diagnostics and processes for the
control of plasma scenario. In the past two years, challenging diagnostics such as the
charge exchange diagnostic have been implemented in real time. Ambitious real time
codes and processing such as the Grad-Shafranov solver EQUINOX have been
successfully installed and validated. All these real time data are now strongly
contributing to the reliability, reproducibility and protection of the plasma scenario in
The recent campaigns have further developed three relevant scenarios for the
next step, namely the inductive scenario, the “hybrid” advanced tokamak scenario and
the steady state non-inductive scenario. Specific real time control networks have been
worked out for all three scenarios and have improved their reliability and
Real time control tools have been more specifically applied to the advanced
tokamak scenario since they can assist efficiently in sustaining internal transport
barrier in fully non-inductive plasma. For that reason specific model-base multi-
variable techniques have been proposed for controlling the q and pressure profiles
simultaneously. The JET experiments on q profile feedback control have validated
these techniques and will provide a sound basis for future experiments to produce
long (~20s) steady state discharges with Vloop=0 using q and pressure profile control.
Last but not least, the real time tools are also indispensable in preventing
plasma instabilities developing such as neo-classical tearing modes or resistive wall
modes. The real time control of the plasma fuel mix has also been achieved and this
foresees future control of the D-T mix in a burning plasma experiment.
It is now demonstrated that real time measurements and control can play an
increasing role in the integration of the scenarios relevant for ITER, and in the
operation of a burning plasma experiment.
This work has been performed under the European Fusion Development
Agreement (EFDA). It is a great pleasure to acknowledge the technical support of the
UKAEA teams involved in the preparation and realisation of the real time
measurement and control (RTMC) project. The invaluable contributions of the
Associations involved in this project (Association Euratom-CEA in Cadarache, the
University of Nice, Consorzio RFX Associazione Euratom ENEA sulla Fusion in
Padova, Associazione Euratom ENEA sulla Fusione in Fracati, Instituto Superior
Tecnico Lisbon) have also been essential in this undertaking.