Focus On : Real Time Control of Plasmas

If JET had live news reports, this might be a typical broadcast during a JET pulse:

"The high temperature plasma is evolving correctly, confined by the magnetic fields of the JET machine. Now powerful microwaves and particle beams are being switched on to increase the temperature of the plasma up to hundreds of millions of degrees! At this moment, we can see fusion reactions occurring. Oh! what's that? What has happened? There was a flash in the plasma but the machine has brought it under control again! From the experts at the control desk I understand there was a sort of fast growing perturbation in the magnetic field, a magnetic island as they call it. Quite unpredictable they say, like turbulence in the air. But even before we spotted the danger, JET's automated systems had recognised it and reacted: the heating was switched off briefly until the perturbation vanished. Look, there are even more fusion reactions now than before!" (Fig. 1)

During the early years of fusion research, the parameters of high temperature plasmas were severely limited by the design of the experimental facility and its power sources. The experimental scenario was 'hard-wired', including some basic real-time feedback features, and used elementary electronic control over a few key parameters. The output data were usually shown on oscilloscope screens and photographed (Fig. 2). Advances in computing in the eighties enabled experimental scenarios to be 'pre-programmed' and the resultant data to be stored digitally for subsequent analysis. The very fast response of today's computing and control systems allows us to move towards extensive real-time control and data analyses. The real-time tools are enabling us not only to precisely tailor key plasma parameters and keep them under control, but also to run several consecutive experiments within a single plasma discharge. This latter feature is very significant now that JET's plasma discharges may last tens of seconds - and in future superconducting facilities where plasmas could potentially extend over tens of minutes.

In principle, real-time control allows instantaneous modification of actions according to changes in observations. There are many examples of real-time control in nature - response to light is an elementary example. Pursuing a moving target is another example, requiring much more sophisticated real-time control. Even just now your eyes, brain and arm were performing a quite complex feedback process when you last moved your mouse to position the cursor on the screen.

What is the correct way of designing a real-time control system in the technical world? A very general outline is given in Fig. 3. A sensor measures the changes in a control parameter over time. Some control parameters, eg magnetic field perturbation, correspond directly to experimental measurements. Others, eg normalised plasma pressure, require the sensor signal to be calculated from several independent measurements - just like the post-pulse physics analysis, except that the sensor signal is needed in real time. This is quite challenging in terms of both hardware and software performance.

Real-time feedback control is achieved by comparing a sensor signal to a desired reference value that is pre-set within the experiment scenario. The difference between the two - the error - serves as an input to the controller. The controller can rapidly modify performance of the actuators (every 10 milliseconds at JET) in order to minimise the error.

The relationship between the action of the actuators and the sensor measurements (the system response) is not straightforward. Indeed, the plasma behaviour can be quite complex and involve many disturbances. A process model is needed to predict the response but process models based on plasma physics equations alone cannot yet fully predict plasma behaviour (Fig. 4). Consequently, dedicated experiments are run to help to identify plasma responses so that a reliable process model can be implemented. Also worthy of note are the disproportionate levels of power required to drive the actuators and the response measured by the sensors: while the former is in the order of millions of watts at JET, the latter is often less than milliwatts, which means that the power scale differs for more then nine orders of magnitude.

In the previously mentioned example of computer mouse control, the mouse cursor position is sensed by your eyes, the target area is the reference, your brain is the controller and your arm the actuator. Instead of the plasma environment there is a mouse, computer and monitor between the actuator (arm) and the sensor (eyes), with a much more predictable behaviour.

Actuators must be designed so that they have enough power to change the quantities measured by sensors, but possibly without modifying other characteristics of the system. The controller, on the other hand, should be designed so that it can respond to errors within an appropriate time, usually referred to as a deadline. In today's plasma physics, the controller commonly consists of a PID (Proportional-Integral-Derivative) element as used in many industrial process controllers, eg in chemical plants. However, more sophisticated controllers based on multiple-input multiple-output models, state-space models, and neural networks are being developed.

In magnetic confinement fusion research, the earliest examples of real time control were in the sensing and control of the magnetic fields used to keep the very hot plasma away from the vessel walls. Feedback stabilises the confining magnetic fields, counteracting plasma forces that randomly disturb the configuration. The magnetic field is monitored by tiny magnetic probes and from their data the plasma boundary position is calculated (Fig. 5). The distance of the plasma boundary from points within the vacuum vessel produces a Sensor measurement. The controller takes the errors in these distances and drives large poloidal coils (the actuator) to correct the magnetic fields. Scientists realised in the early sixties that without this feedback control, high temperature plasmas would never survive for more than a few tens of milliseconds. Of course, the feedback at that time was completely hard-wired and analogue (ie based on resistive, inductive and capacitive elements and amplifiers, not on digital processors).

Further real time control appeared a bit later - a feedback control based on plasma density (Fig. 6). Precise fuelling of high temperature plasmas is essential to keep an optimal plasma density, but it is very difficult to predict how much fuel will be required because of gas absorbtion into, or released from, the walls and materials within the tokamak chamber. Therefore, the most reliable method to keep plasma fuelling within the required limits is to use a real time control of the fuelling valve (the actuator) based upon the density measurements (the sensor). This density feedback was quite a technological milestone for fusion research as there is no diagnostic signal that directly corresponds to plasma density. This had to be calculated from measurements of the plasma refractive index.

Today, powerful digital data acquisition systems (Fig. 7) allow us to implement a wide range of real time controls and JET is at the forefront of this progress. For example, we can control in real time the gradient of plasma temperature from the edge to the centre of the plasma, and how it evolves in time. The same can be also done for magnetic field helicity (Fig. 8). This allows us to control particle transport barriers that significantly reduce losses of heat and thus improve plasma confinement. Concentration of different chemical elements in the plasma or occurrences of regular magnetic structures can also be influenced in real time. Additionally, there are event-driven controls that can immediately modify heating and/or fuelling in response to fluctuations in magnetic fields or excessive radiative losses from the plasma.

JET makes real-time measurements of neutrons, magnetic flux, plasma temperature, density, helicity, X-ray, UV, visible and IR radiation, etc. We do real-time analysis of magnetic fields, confinement, spectral lines, chemical composition, and profiles of temperature, density and current. There are over 500 signals involved, updating every few milliseconds! We use an ATM computer network (like telephone companies use in their backbone exchanges) to deliver sets of signals (datagrams), from each source to the appropriate destinations.

Magnetic coils, gas valves, Neutral Beam injectors (NBI - Fig. 9), pellet injectors, Ion Cyclotron Resonance Heating (ICRH) and Lower Hybrid Current Drive (LHCD) microwave systems can all act as Actuators in JET. In other words, their performance can be modified in real time in response to instantaneous measurements and calculations. By combining these Actuators, a large variety of plasma scenarios can be tuned and stabilised. Thanks to this feature, JET can drive experiments starting from basic physics studies (with very simple and symmetric plasma set-ups) through to identity/similarity experiments which model other facilities, up to reactor-like (ITER-like) high power plasma scenarios. Notice that in identity/similarity experiments, JET can - thanks to its real time capacities - mimic plasma conditions of other magnetic fusion facilities (eg the German ASDEX-U, Japanese JT-60U or American DIII-D) and thus confirm and even enhance their results.

In most of the above applications, the real-time control must be fast enough to keep up with the plasma evolution. That is, the response time of the feedback system is a critical parameter. In our scheme, it is the process model that indicates the allowable delay for the response time. When exceeded, the control is lost. Even worse, some actuators might produce effects which cause operational delays. For example, exaggerated gas influx can aggravate the vacuum properties of the vessel.

Some plasma processes such as the diffusion of helium (Fig. 10) are quite slow (in the order of a few hundred milliseconds) and do not require rapid response times. Others are much more rapid, for example the magnetic perturbations which can evolve in only a few milliseconds. The latter are thus quite demanding on the electronics of the real-time hardware, namely on the high-power electronics that drive the actuators.

Fortunately, in general, with bigger fusion facilities the allowable time delay increases, and at the same time computer technology keeps evolving. It is thus probable that in future fusion reactors plasma will be controlled in real-time by very sophisticated methods and algorithms. The precise tailoring of the key plasma parameters in both space and time will play a crucial role in developing a continuous and economic source of fusion energy.

"JET is developing new real-time techniques for measuring and controlling fusion plasmas to maximise performance and minimise internal disturbances. Real-time control will be essential for future long-pulse reactors such as ITER."

Rob Felton, Real Time Measurement and Control Systems Manager

Main Author: Jan Mlynar, Public Relations Officer