Fusion Engineering and Design 86 (2011) 1026–1029
Contents lists available at ScienceDirect
Fusion Engineering and Design
journal homepage: www.elsevier.com/locate/fusengdes
A MARTe based simulator for the JET Vertical Stabilization system
Teresa Bellizio a,∗ , Gianmaria De Tommasi a , Nicola Risoli a ,
Raffaele Albanese a , Andrè Neto b , JET–EFDA Contributors1
a
b
Associazione EURATOM-ENEA-CREATE, University di Napoli Federico II, Via Claudio 21, 80125 Napoli, Italy
Associação EURATOM/IST, Inst. de Plasmas e Fusão Nuclear – Laboratório Associado, Instituto Superior, Técnico, P-1049-001 Lisboa, Portugal
a r t i c l e
i n f o
Article history:
Available online 9 April 2011
Keywords:
Vertical stabilization
Model-based design
Software validation
a b s t r a c t
Validation by means of simulation is a crucial step when developing real-time control systems. Modeling
and simulation are an essential tool since the early design phase, when the control algorithms are designed
and tested. This phase is commonly carried out in off-line environments such as Matlab® and Simulink® .
A MARTe-based simulator has been recently developed to validate the new JET Vertical Stabilization
(VS) system. MARTe is the multi-thread framework used at JET to deploy hard real-time control systems.
This paper presents the software architecture of the MARTe-based simulator and it shows how this
tool has been effectively used to evaluate the effects of Edge Localized Modes (ELMs) on the VS system.
By using the simulator it is possible to analyze different plasma configurations, extrapolating the limit of
the new vertical amplifier in terms of the energy of the largest rejectable ELM.
© 2011 Elsevier B.V. All rights reserved.
1. Introduction
To achieve high performance in tokamaks, plasmas with
elongated poloidal cross-section, characterized by large vertical
instability growth rates, are needed. Such elongated plasmas are
vertically unstable, hence position control on a fast time scale is
an essential feature of all machines. The achievement of the fast
time performance is strictly dependent on the flexibility and reliability of the real-time systems that operate the plant during the
experiment. In large experimental plants like JET [1] or ITER [2], it
is crucial to have an architecture that supports model-based development to validate software modules against a plant model. In this
way the risks related to the development of complex plant control systems are minimized. Furthermore, simulation tools can be
effectively adopted also during the deployment of real-time systems. Indeed, off-line testing of the full real-time system permits to
debug the code and to validate the real-time version of the control
algorithms before running them on the plant [3]. Such an approach
permits to minimize the risk of malfunctions and to reduce the
time needed for the commissioning on the plant, yielding a cost
reduction. Eventually, by using such a simulation environment, it
is possible to perform off-line analyses addressed to the fine tuning
of the control algorithms for specific operative scenarios. In order
∗ Corresponding author. Tel.: +39 3482443209.
E-mail address: teresa.bellizio@unina.it (T. Bellizio).
1
See the Appendix of F. Romanelli et al., Proceedings of the 22nd IAEA Fusion
Energy Conference 2008, Geneva, Switzerland.
0920-3796/$ – see front matter © 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.fusengdes.2011.02.076
to adopt such an approach the real-time framework has to satisfy
some key requirements: in particular it must allow to run the realtime code in an off-line (non real-time) environment, interfacing
it with a plant simulator. On the other hand, reliable plant models
must be available.
A MARTe-based simulator has been recently developed at JET
tokamak, and it has been used to validate the new JET Vertical Stabilization (VS) system [4]. MARTe [5] is the multi-thread framework
used at JET to deploy hard real time control systems. Thanks to
the modularity of its software architecture, MARTe allows to interface the real-time control system with a C++ version of the plasma
magnetic linear model. It also allows to use different linear models (corresponding to different plasma equilibria) in different pulse
phases, in order to simulate a complete JET pulse.
This work describes the software architecture of the MARTebased simulator. As an example case study, it is shown how this
tool has been effectively used to evaluate the effects of Edge
Localized Modes (ELMs) on the VS system [6]. In particular, a drop
of energy due to an ELM occurrence is simulated by a variation of
both poloidal beta and internal inductance [7]. These parameters
are considered as a disturbances applied to the plasma linearized
model. By using the simulator it is possible to analyze different
plasma configurations, extrapolating the operational limit of
the new vertical amplifier in terms of the energy of the largest
rejectable ELM.
The paper is structured as follows: the VS simulator is introduced in the next section, while in Section 3 the case study is
presented. Eventually some conclusive remarks are drawn in Section 4.
T. Bellizio et al. / Fusion Engineering and Design 86 (2011) 1026–1029
1027
Fig. 1. Block diagram of the VS software architecture.
2. The Vertical Stabilization simulator
The new VS control system represents the first MARTe based
control system that has been successfully developed and deployed
at JET [8]. Within the MARTe environment, the end users are
required to define and implement algorithms inside a well defined
software block named Generic Application Module (GAM), which
is executed by the real-time scheduler. The JET VS system was
implemented by using MARTe under the Real Time Application
Interface (RTAI)/Linux operating system. The adoption of a standard
framework for the development of real-time systems, model-based
design and validation is an effective approach to reduce the time
needed for commissioning a new system on the plant.
By exploiting the MARTe modularity it is possible to add different modules to the structure of the VS system, in order to
implement a complete closed-loop simulator that permits to study
the VS behaviour. The VS simulator has been very useful first during the commission phase of the new VS system, especially to tune
the controller parameters and to study the behaviour of the new
control law.
In the following we present both the software architecture of
the VS simulator and its human–machine interface.
2.1. Software architecture
The VS simulator is MARTe based and as mentioned before, it
mimics the same structure of the VS system. By adding four different modules to the structure of the VS system it is possible to
implement a complete closed-loop simulator of the same code that
runs on the real plant. These four GAMs are enabled by the user
only when performing offline validation and are disabled when the
VS runs on the real plant. In particular the four simulation GAMS
depicted in yellow in Fig. 1 have been added:
• State–space GAM: it allows to simulate the plant behaviour by
receiving as input the voltage applied by the Enhanced Radial
Field Amplifier (ERFA) and producing as outputs the estimation
of the plasma vertical velocity and the amplifier current. The
state–space model can be configured by using the CREATE-L [9]
code. This GAM allows to load different linearized model to take
into account the different plasma configurations, allowing to simulate a complete JET pulse.
• Waveform generator GAM: it allows to add inputs that are not
modified by the closed-loop, e.g., the plasma current, which is an
input for the simulator and it is not modified by the VS system.
• ERFA logic GAM: it allows to simulate hysteretic characteristic of ERFA and adds some noise to simulate a real acquired
signal.
• Noise GAM: the plasma vertical velocity measurement used for
the vertical stabilization in JET, is reconstructed by means of a
suitable linear combination of flux and field time derivative measurements. This reconstructed signal is obviously affected by an
error. The main sources are: the violation of hypotheses assumed
by the reconstruction algorithm; noise due to power electronics;
noise due to measurement instrumentation and signal conditioning electronics; noise due to plasma activity. The noise GAM allows
to add the noise signal to the plasma vertical velocity computed
by the model, in order to model all these sources of uncertainty.
2.2. Human–machine interface
At JET the plant control systems are configured using a distributed system named Level-1. This system is used by the expert
users to set up all the VS system parameters before the experiment.
MARTe provides a web interface which enables the browsing
of its internal components, allowing the user to navigate into the
GAMs’ structure and check the values of the parameters loaded in
the VS system.
The human–machine interface of the simulator has been
designed as similar as possible to the VS Level-1 interface. Thanks
to this choice, by using the simulator, the user can access to all the
parameters available on the plant during the experiment.
The Level-1 interface for MARTe based systems is divided in two
levels:
(1) Real-time executor level, which allows the user to load the
MARTe skeleton configuration; here the user can specify what
GAMs are to be executed together with their basic parameters. It is important to note that this level is common to all
MARTe-based applications;
(2) Application level, which is customized for the VS system. This
level is designed to allow the user to set each controller parameter before the experiment and to configure the GAM parameters
using fully featured graphical user-interface.
The user interface of the simulator presents the same structure of the Level 1. This interface is realized by using the Matlab
GUI Application. As shown in Fig. 2 (yellow blocks) this interface is
structured in two levels: the former allows to load the machine configuration file, the latter allows to change it by setting the controller
parameters and to add the different linearized model of plasma.
In particular:
• Set VAM & others: allows to change the parameter of the Vertical Amplifier Manager (VAM) GAM and to add as waveform the
signals that are not modified by the closed-loop.
T. Bellizio et al. / Fusion Engineering and Design 86 (2011) 1026–1029
5000
0
-5000
15.184
15.185
15.186
15.187
15.184
(×108)
15.185
15.186
15.187
15.184
15.185
15.186
15.187
1000
500
0
-500
4
2
JG10.261-4c
ZPDIP (MAm/s)
IEFRA (A)
VERFA (V)
1028
0
-2
Time (s)
Fig. 2. MATLAB Graphics User Interface of the simulator.
Fig. 4. Effect of an ELM event on the VS parameters.
• Set Scheduler: allows scheduling the experiment because every
JET discharge is logically divided into a number of time windows. As shown in Fig. 3, for each time windows it is
possible to set several control mode and several controlled
variables.
• Set Model: allows to load different linear models (corresponding
to different plasma scenarios) in different pulse phases, allowing
to simulate a complete JET pulse.
• Set Control: allows to set all the parameters of the controller which
are independent from the scheduling of the pulse. These global
parameters are set by using a waveform editor
• Set ERFA Logic: allows to set the amplifier parameters.
3. A case study: larger rejectable ELM
The VS simulator is useful to study the operational limits of the
VS system in the presence of very strong localized MHD plasma
instabilities, named ELMs. ELMs manifest themselves as strong
magnetic perturbations associated with a burst of D-alpha radiation and a loss of particles and energy from the plasma periphery.
Moreover an ELM event is characterized by a loss of the diamagnetic energy that is strictly related to a variation of poloidal beta
and the relationship is given by [7]:
W =
3
0 R0 Ip2 ˇ
8
Fig. 3. Interface of the simulator to set the controller parameters for each time windows.
T. Bellizio et al. / Fusion Engineering and Design 86 (2011) 1026–1029
(×104)
1.0
0.5
0
-0.5
-1.0
VERFA
15.184
15.188
15.192
15.196
IERFA
1000
SLOW
0
-1000
-2000
FAST
15.184
15.188
15.192
15.196
4
VSEL
2
JG10.261-5a
Simulation versus Experiment
a
0
- 2
15.184
15.188
15.192
15.196
15.200
Time (s)
b
2000
IERFA Simulated
IERFA Experiment
VD2*10
1000
0
Since the poloidal beta and internal inductance variations are
strictly related to a loss of diamagnetic energy, with these simulations we are able to find the maximum controllable ELM from the
VS system in terms of the maximum diamagnetic energy loss by
multiplying the disturbances for a factor ˛.
The tolerable poloidal beta (ˇ) drop (ˇ) scales with 1/Ip ,
whereas the tolerable energy drop (W ∝ ˇIp2 ) scales with Ip .
ELM transients are characterized by fast dynamics (hundreds of
s) followed by a slow ˇ drop (tens of ms).
In pulse #78452 with Ip = 3MA, and a relatively high growth rate
of the vertical instability ( = 200 s−1 ), the VS system tolerated a
considerable energy drop (|W| > 1.5 MJ) with an excursion of the
ERFA current (|IERFA | = 2.5 kA) well below its operational limit.
Simple extrapolations based on scaling laws and more accurate
simulations based on the CREATE-L model show that the tolerable energy drop for a 4 MA plasma would have been well beyond
2 MJ, with a dramatic improvement with respect to the previous VS
system with the old radial field amplifier FRFA.
As shown in Fig. 5(a) the ELM effect on the VS system in characterized of two phases. The first one is a fast phase in which the
simulated trace is very close to the experimental behaviour.
On the contrary during the slow phase the simulated behaviour
is very different from the experimental one. As shown in Fig. 5(b)
this experimental behaviour is essentially due to a shape controller
effect that is not taken into account by the VS simulator.
4. Conclusions
Slow Phase
-2000
15.20
15.21
15.22
15.23
15.24
JG10 261-5b
-1000
15.19
1029
15.25
Time (s)
Fig. 5. (a) Comparison between experimental data (black line) and simulation (red
line) during an ELM event; (b) effect of the divertor voltage during the slow phase
of the ELM event.
where Ip is the plasma current, R0 is the major radius and ˇ is the
variation of poloidal beta.
Because the perturbation affects the magnetic fields creating
a strong variation in the plasma speed measurement, the VS sees
an ELM as a rapid increase of plasma speed (ZPDIP) followed by a
rapid inversion and a slower decay (Fig. 4). This causes the firing
of ERFA and a resulting vertical excursion of the plasma, in some
cases associated with loss of control.
For these reasons it is very important to characterize the
behaviour of the VS system in terms of the energy of the largest
rejectable ELM.
Since an ELM event creates a strong variation in the plasma
speed measurement, it can be modelled as a disturbance for the
VS system. In particular, by using the CREATE-L model, a representation of the plant behaviour is given in the state space form.
A characterization of ELMs by means of poloidal beta and internal
inductance variations has been carried out via simulation, using
both experimental magnetic signals and models. By considering
the identified quantities as disturbances for the system the closed
loop simulations have been performed by using MARTe simulator. In particular, the inputs of the system are the amplifier voltage
and the identified quantities, instead the outputs are the amplifier
current and the estimation of the vertical velocity.
A MARTe-based simulator for the JET VS system has been presented. This simulator has been effectively used to test off-line the
VS software during the commissioning of the system. Furthermore
this simulator has been used to assess the system performance by
tuning the controller parameters. Since the VS simulator is MARTe,
and exploiting the modularity of this framework in future it will
be to study the coupling between the VS system and the shape
controller at JET, by simply adding the shape controller GAM.
Acknowledgments
This work, supported by the European Communities under the
contract of Association between EURATOM and ENEA/CREATE, was
carried out within the framework of the European Fusion Agreement. The views and opinions expressed herein do not necessarily
reflect those of the European Commission.
References
[1] Fus. Sci. Technol. 53 (2008) (special issue).
[2] Nucl. Fus. 47 (2007) (special issue).
[3] G. Ambrosino, M. Banfi, G. Carannante, G. De Tommasi, A. Mandelli, A. Pironti,
et al., Rapid Prototyping of Safety System for Nuclear Risks of the ITER Tokamak,
IEEE Transactions on Plasma Science 38 (7) (2010) 1662–1669.
[4] F. Sartori, A. Barbalace, A.J.N. Batista, T. Bellizio, P. Card, G. De Tommasi, et al.,
The PCU JET Plasma Vertical Stabilization control system, Fusion Engineering
and Design 85 (3–4) (2010) 438–442.
[5] A.C. Neto, F. Sartori, F. Piccolo, R. Vitelli, G. De Tommasi, L. Zabeo, et al., MARTe: a
Multi-Platform Real-Time Framework, IEEE Transactions on Nuclear Science 57
(2) (2010) 479–486.
[6] H. Zohm, Plasma Phys. Control. Fusion 38 (1996) 105–128.
[7] O. Barana, E. Joffrin, A. Murari, F. Sartori, Real-time determination of confinement
parameters in JET, Fusion Engineering and Design 68 (6) (2003) 697–701.
[8] T. Bellizio, G. De Tommasi, P. Mc Cullen, A. Neto, F. Piccolo, F. Sartori, et al., The
Software Architecture of the New Verical Stabilization System for the JET tokamak, IEEE Transactions on Plasma Science 38 (9) (September 2010) 2465–2473.
[9] R. Albanese, F. Villone, Nucl. Fus. 38 (5) (1998).