Search

The inadequacies of basic physics models for disruption prediction have induced the community to increasingly rely on data mining tools. In the last decade, it has been shown how machine learning predictors can achieve a much better performance than those obtained with manually identified thresholds or empirical descriptions of the plasma stability limits. The main criticisms of these techniques focus therefore on two different but interrelated issues: poor “physics fidelity” and limited interpretability. Insufficient “physics fidelity” refers to the fact that the mathematical models of most data mining tools do not reflect the physics of the underlying phenomena. Moreover, they implement a black box approach to learning, which results in very poor interpretability of their outputs. To overcome or at least mitigate these limitations, a general methodology has been devised and tested, with the objective of combining the predictive capability of machine learning tools with the expression of the operational boundary in terms of traditional equations more suited to understanding the underlying physics. The proposed approach relies on the application of machine learning classifiers (such as Support Vector Machines or Classification Trees) and Symbolic Regression via Genetic Programming directly to experimental databases. The results are very encouraging. The obtained equations of the boundary between the safe and disruptive regions of the operational space present almost the same performance as the machine learning classifiers, based on completely independent learning techniques. Moreover, these models possess significantly better predictive power than traditional representations, such as the Hugill or the beta limit. More importantly, they are realistic and intuitive mathematical formulas, which are well suited to supporting theoretical understanding and to benchmarking empirical models. They can also be deployed easily and efficiently in real-time feedback systems.

In the last years, it has become apparent that detecting disruptions with sufficient anticipation time is an essential but not exclusive task of predictors. It is also important that the prediction is accompanied by appropriate qualifications of its reliability and it is formulated in mathematical terms appropriate for the task at hand (mitigation, avoidance, classification etc.). In this paper, a wide series of rule-based predictors, of the Classification and Regression Trees (CART) family, have been compared to assess their relative merits. An original refinement of the training, called noise-based ensembles, has allowed not only to obtain significantly better performance but also to increase the interpretability of the results. The final predictors can indeed be represented by a tree or a series of specific and clear rules. Such performance has been proved by analysing large databases of shots on JET with both the carbon wall and the ITER Like Wall. In terms of performance, the developed tools are therefore very competitive with other machine learning techniques, with the specificity of formulating the final models in terms of trees and simple rules.

Nowadays, disruption predictors, based on machine learning techniques, can perform well but they typically do not provide any information about the type of disruption and cannot predict the time remaining before the current quench. On the other hand, the automatic identification of the disruption type is a crucial aspect required to optimize the remedial actions and a prerequisite to forecasting the time left for intervening. In this work, a stack of machine learning tools is applied to the task of automatic classification of the disruption types. The strategy is implemented from scratch and completely adaptive; the predictors start operating after the first disruption and update their own models, following the evolution of the experimental program, without any human intervention. Moreover, they are designed to implement a form of transfer learning, in the sense that they identify autonomously the most important disruption classes, generating new ones when necessary. The results obtained are very encouraging in terms of both prediction performance and classification accuracy. On the other hand, regarding the narrowing of the warning times, some progress has been achieved, but new techniques will have to be devised to obtain fully satisfactory properties.

The last years have witnessed on JET a significant increase in the use of digital cameras, which are routinely applied for imaging in the IR and visible spectral regions. One of the main technical difficulties, in interpreting the data of camera based diagnostics, is the presence of movements of the field of view. Small movements occur due to machine shaking during normal pulses while large ones may arise during disruptions. Some cameras show a correlation of image movement with change of magnetic field strength. For deriving unaltered information from the videos and for allowing correct interpretation, an image registration method, based on highly distinctive Scale Invariant Feature Transform (SIFT) descriptors and on the Coherent Point Drift (CPD) points set registration technique, has been developed. The method ensures a reliable determination of the alignment parameters and their corresponding uncertainty. The algorithm incorporates a complex procedure for rejecting outliers. The method has been applied for vibrations correction to a representative set of videos collected by the JET wide angle infrared camera. It proved to provide reliable image vibration correction when the camera is operated in full view. Another application has been developed for the JET fast visible camera, where the image rotation artefacts occur due to the sensitivity of the image intensifier to the magnetic field (above a certain threshold). The method has proved to be able to deal with the images provided by this camera frequently characterized by low contrast and a high level of blurring and noise. It has also the advantage of using directly the input images as they are provided by the cameras without any pre-processing step. The parameters of the algorithm are the same for all videos in the database.

An original spot detection method for large surveys of videos in Joint European Torus (JET) is presented. The method can be used for the automatic identification of spots in JET infra-red (IR) videos and for the assessment of the long-term trends in their evolution. This method has been applied to the analysis of a large database of JET IR images collected during the last campaigns of operation with the International Thermonuclear Experimental Reactor-like wall. The evolution of the spots and their properties, such as size and distribution, can be correlated with macroscopic events, in particular series of intentional disruptions. On the other hand, care must be taken in the interpretation of the results because there is some evidence of toroidal asymmetries in the spot distribution.

In the past, the analysis of JET polarimetry measurements were carried out only for the vertical channels using a polarimetry propagation code based on the Stokes vector formalism [1] , [2] . A new propagation code has been developed therefore for the horizontal chords to simulate and interpret the measurements of the Faraday rotation and Cotton–Mouton phase shift in JET. The code has been used to develop a theoretical study to the effect of the input polarization on the eventual quality of the measurements. The results allow choosing the best polarization to optimize the polarimetric measurements for the various experiments.

Machine learning tools have been used since a long time ago to study disruptions and to predict their occurrence. On the other hand, the challenges posed by the quality and quantities of the data available remain substantial. In this paper, methods to optimize the training data set and the potential of kernels-based advanced machine learning tools are explored and assessed. Various alternatives, ranging from appropriate selection of the weights to the inclusion of artificial points, are investigated to improve the quality of the training data set. Support vector machines (SVM), relevance vector machines (RVMs), and one-class SVM are compared. The relative performances of the different approaches are initially assessed using synthetic data. Then they are applied to a relatively large database of JET disruptions. It is shown that in terms of final results, the optimization of the training databases proved to be very productive. Further, the RVM algorithm performs well when it is trained on a small set of discharges compared to the traditional methods.

A new compact gamma-ray spectrometer based on a Silicon Photo-Multiplier (SiPM) coupled to a LaBr3(Ce) crystal has been developed for the upgrade of the Gamma Camera (GC) of JET, where it must operate in a high intensity neutron/gamma-ray admixed field. The work presents the results of an experiment aimed at characterizing the effect of 14 MeV neutron irradiation on both LaBr3(Ce) and SiPM that compose the full detector. The pulse height spectrum from neutron interactions with the crystal has been measured and is successfully reproduced by MCNP simulations. It is calculated that about 8% of the impinging neutrons leave a detectable signal of which less than < 4% of the events occur in the energy region above 3 MeV, which is of interest for gamma-ray spectroscopy applications. Neutron irradiation also partly degrades the performance of the SiPM and this is mostly manifested as an increase of the dark current versus the neutron fluence. However, it was found that the SiPM can be still operated up to a fluence of 4×1010 n/cm2, which is the highest value we experimentally tested. Implications of these results for GC measurements at JET are discussed. © 2017 Istituto di Fisica del Plasma - CNR. Published by IOP Publishing Ltd on behalf of Sissa Medialab.

A loss of vacuum in a vessel, containing or not dust, is the typical case study considered in the STARDUST-UPGRADE facility of the Quantum Electronics and Plasma Group of the university of Rome Tor Vergata. This kind of accident was simulated numerically, without including the presence of dust, for two mass flow rates and three different inlet ports (C, E and F). Numerical settings are explained and the results obtained in each case are shown and discussed. At the end of the work, conclusions about what seen and further foreseen developments of this research are presented.

Video cameras have recently become diagnostic tools widely used on Joint European Torus (JET) for fusion plasma diagnostic and control. Since video streams are usually compressed for storage, MPEG-2 compressed domain information is processed to obtain a very fast and reasonably accurate 2-D motion estimation of the video scenes for the JET diagnostics, whose computational costs are prohibitively high. These methods can be used for the manipulation of the large JET video databases and, in specific cases, even for real-time data processing. Plasma instabilities, which can trigger harmful disruptions, are detected and tracked by means of motion segmentation. Motion segmentation is used as a key contrivance to allow very fast optical flow estimation for the determination of the deuterium ice extrusion velocity of JET pellet injector. Experimental validation is performed on significant JET video data.

The far-infrared polarimeter at JET is affected by an anomaly that makes difficult the interpretation of both Faraday and Cotton-Mouton effect measurements. The anomaly is clearly displayed during calibration operations in the absence of plasma: As the polarization of the probing beam is rotated, the phase shift of the polarimetric signal with respect to the interferometric signal is not constant, as expected, and changes significantly. It affects all the polarimetric measurement channels and has so far been removed by an empirical preprocessing of the raw data. It can be ascribed to a nonideal behavior of some optical components. Looking for a possible explanation of the anomaly, in this paper, we analyze the optical setup of the JET polari-interferometer according to the laws of classical polarization optics. At first, the optical characteristics of the recombination plates are analyzed in detail. Although they produce ellipticity in the transmitted and reflected beams, the results show that the recombination plates should not be responsible of the anomaly of the polarimeter. Then, the dielectric waveguides used to transfer the recombined beams from the torus hall to the detectors are, for the first time, considered as a possible origin of the anomaly. The anomalous behavior is expected to be mainly originated by reflections on metal mirrors, which may produce rotations of the polarization of the beams. A calculation has been performed in order to analyze the effects of a rotation of the polarization of the recombined beam on the detector signals. As a result, a rotation of the polarization along the line could explain the anomaly. We also suggest some simple and feasible tests, which are useful to give an experimental support to this conclusion, and discuss possible modifications of the optical setup to remove or greatly reduce the anomaly in future measurements.

Visible light and infrared cameras allow the acquisition of plasma movies during discharges. A great amount of data from video-movies is stored for every discharge, and therefore, automatic methods...

Dynamical systems are often considered immune from memory effects, i.e., the dependence of their time evolution on the previous history. This assumption has been tested for two phenomena in nuclear fusion that are believed to sometimes show sensitivity to the previous history of the discharge: disruptions and the transition from the L mode to the H mode of confinement. To this end, two neural network architectures, tapped delay lines and recurrent networks of the Elman type, have been applied to the Joint European Torus (JET) database to extract these potential memory effects from the time series of the available signals. Both architectures can detect the dependence on the previous evolution quite effectively. In the case of disruptions, only the ones triggered by locked modes seem to be influenced by the previous history of the discharge. With regard to the L-H transition, memory effects are present only in the time interval very close to the transition, whereas once the plasma has settled down in one of the two regimes, no evidence of dependence on the previous evolution has been detected.

An optical flow method is applied to the study of several fusion plasma relevant issues, including plasma wall interactions. A multi-resolution coarse-to-fine procedure is used in order to cope with large displacements of objects between consecutive frames, characteristic of plasma images captured by JET fast visible camera. Occlusion modeling is also implemented. The method is able to provide good results for JET fast visible camera images which can be affected by saturation, discontinuous movement, reshaping of image objects, low gray-level in-depth resolution. Significant experimental cases concerning pellet injection, plasma filaments and MARFEs are analysed. The method is able to provide the real velocity for objects moving close to structures.

In modern tokamaks, visible and infrared video cameras are becoming more and more important in monitoring plasma evolution during fusion experiments. Analyzing these images in real time can provide relevant information for controlling plasma and improving machine safety. The real-time image processing capability of the cellular nonlinear/neural network-based chips that are available nowadays has been applied to several tasks, both at Frascati Tokamak Upgrade (FTU) and at Joint European Torus (JET). The successful applications range from the identification of plasma instabilities, such as multifaceted asymmetric radiations from the edge (MARFEs), to the determination of the strike-point position in the divertor and to the detection of the so-called ldquohot spots.rdquo

This paper reports on the three new Single Crystal Diamond (SCD) detectors installed at JET for the 2008 Restart campaign. The yields of total neutrons produced during DD plasma pulses as well as the time dependent neutron emission have been measured. The three detectors present some main new developments as for their fabrication characteristics, embeddings and electronic chains. The first detector is 200 μm thick and is embedded in paraffin. The second detector is 104 μm thick and is covered by a thermally evaporated 3 μm thick 6 LiF film in order to detect the total and the 14-MeV neutrons. In addition, it is surrounded by a 2.5-cm thick polyethylene shield to enhance its thermal neutron response. The third detector is operated in an innovative way that is with a single low capacitance super-screened cable and its whole electronic chain put outside the JET Torus Hall. Furthermore it uses fast electronics, suitable to the fast diamond response pulse (

In this work, we describe the solution developed by the gamma ray camera upgrade enhancement project to improve the spectroscopic properties of the existing JET γ-ray camera. Aim of the project is to enable gamma-ray spectroscopy in JET deuterium-tritium plasmas. A dedicated pilot spectrometer based on a LaBr3 crystal coupled to a silicon photo-multiplier has been developed. A proper pole zero cancellation network able to shorten the output signal to a length of 120 ns has been implemented allowing for spectroscopy at MHz count rates. The system has been characterized in the laboratory and shows an energy resolution of 5.5% at Eγ = 0.662 MeV, which extrapolates favorably in the energy range of interest for gamma-ray emission from fast ions in fusion plasmas.

The Joint European Torus (JET) neutron profile monitor ensures 2D coverage of the gamma and neutron emissive region that enables tomographic reconstruction. Due to the availability of only two projection angles and to the coarse sampling, tomographic inversion is a limited data set problem. Several techniques have been developed for tomographic reconstruction of the 2-D gamma and neutron emissivity on JET, but the problem of evaluating the errors associated with the reconstructed emissivity profile is still open. The reconstruction technique based on the maximum likelihood principle, that proved already to be a powerful tool for JET tomography, has been used to develop a method for the numerical evaluation of the statistical properties of the uncertainties in gamma and neutron emissivity reconstructions. The image covariance calculation takes into account the additional techniques introduced in the reconstruction process for tackling with the limited data set (projection resampling, smoothness regularization depending on magnetic field). The method has been validated by numerically simulations and applied to JET data. Different sources of artefacts that may significantly influence the quality of reconstructions and the accuracy of variance calculation have been identified.

Artificial diamond neutron detectors recently proved to be promising devices to measure the neutron production on large experimental fusion machines. Diamond detectors are very promising detectors to be used in fusion environment due to their radiation hardness, low sensitivity to gamma rays, fast response and high energy resolution. High quality “electronic grade” diamond films are produced through microwave chemical vapour deposition (CVD) technique. Two CVD diamond detectors have been installed and operated at joint European torus (JET), Culham Science Centre, UK. One of these detectors was a polycrystalline CVD diamond film; about 12 mm2 area and 30 μm thickness while the second was a monocrystalline film of about 5 mm2 area and 20 μm thick. Both diamonds were covered with 2 μm of lithium fluoride (LiF) 95% enriched in 6Li. The LiF layer works as a neutron-to-charged particle converter so these detectors can measure thermalized neutrons. Their output signals were compared to JET total neutron yield monitors (KN1 diagnostic) realized with a set of uranium fission chambers. Despite their small active volumes the diamond detectors were able to measure total neutron yields with good reliability and stability during the recent JET experimental campaign of 2006.

In fusion devices, the surface temperature of plasma facing components is measured using infrared cameras. This method requires a knowledge of the emissivity of the material, the reflected and parasitic fluxes (Bremsstrahlung). For carbon, the emissivity is known and constant over the detection wavelength (∼3–5 μm). For beryllium and tungsten, the reflected flux could contribute significantly to the collected flux. The pulsed photothermal method described in this paper allows temperature measurements independently of both reflected and parasitic fluxes. A local increase of the surface temperature (Δ T ∼ 10–15 K) introduced by a laser pulse (few ns) results in an additional component of the photon flux collected by the detector. Few μs after the pulse, a filtering of the signal allows to extract a temporal flux proportional only to the variation of the emitted flux, the emissivity and Δ T . The ratio of simultaneous measurements at two wavelengths leads to the elimination of Δ T and emissivity. The range of application increases for measurements at short wavelengths (1–1.7 μm) with no limitation due to the Bremsstrahlung emission.

To meet the requirements of the scientific programme, the EFDA JET real-time measurement and control project has developed an integrated set of real-time plasma measurements, experiment control and communication facilities. Traditional experiments collected instrument data during the plasma pulse and calculated physics data after the pulse. The challenge for continuous tokamak operation is to calculate the physics data in real-time, keeping up with the evolution of the plasma. In JET, many plasma diagnostics have been augmented with extra data acquisition and signal-processing systems so that they can both capture instrument data for conventional post-pulse analysis and calculate calibrated, validated physics results in real-time. During the pulse, the systems send sampled data sets into a network, which distributes the data to several destinations. The receiving systems may do further analysis, integrating data from several measurements, or may control the plasma scenario by heating or fuelling. The simplest real-time diagnostic systems apply scale factors to the signals, as with the electron cyclotron emission (ECE) diagnostic's 96 tuned radiometer channels, giving the electron temperature profile. In various spectroscopy diagnostics, spectral features are least-squares-fitted to measure spectra from several lines of sight, within 50 ms. Ion temperatures and rotation speed can be calculated from the line widths and shifts. For diagnostics using modulation techniques, the systems implement digital-signal processing phase trackers, lock-in amplifiers and filters, e.g., the far infrared (FIR) interferometer samples 15 channels at 400 kHz for 30 s, i.e., six million samples per second. Diagnostics have specific lines of sight, spatial channels, and various sampling rates. The heating/fuelling systems have relatively coarse spatial localisation. Analysis systems have been developed to integrate the basic physics data into smaller sets of controllable parameters on a common geometry, e.g., temperature, density and safety factor profiles with values at 10 points of normalised radius. The EFDA real-time project is essential groundwork for future reactors such as ITER, and has successfully involved many scientific and technical staff from several institutions. The facility is now frequently used in experiments.

A multichannel far‐infrared (FIR) polarimeter has been installed in RFX, a Reversed Field Pinch (RFP) plasma experiment, to measure the poloidal magnetic field profile. The polarimeter uses a CH3OH FIR laser operating at λ=118.8 μm. Faraday rotation measurements on five of six parallel diagnostic chords are used in preliminary investigations of poloidal field profiles. The experimental results are generally found in good agreement with the μ&p model predictions. The choice of the profile of μ = μ0j·B/B2 is discussed. For the reconstruction of the magnetic field profiles a numerical filamentary current equilibrium code is described, where polarimetric data are included as constraints. An alternative method based on the best‐fit of a three‐parameter μ profile to the polarimetric data is also reported. Both methods provide reliable reconstructions of the plasma magnetic field and the results indicate the existence in RFX of hollow μ profiles.

In this paper, the algorithms to determine the safety factor q and density profiles in real time are presented. They have been designed to be implemented in JET, in the framework of the real time enhancements for the control of the internal transport barriers. In this method, only the signals of the magnetic diagnostics and of the interferometer polarimeter are used. Instead of solving the equilibrium equation, which at the moment does not seem fully compatible with real time requirements, the topology of the magnetic surfaces is determined on the basis of the external magnetic measurements. Then the density and poloidal field profiles are found inverting the interferometric and polarimetric measurements, respectively. Once the poloidal field has been obtained, the q profile is inferred. The density profiles obtained have been checked against the JET LIDAR diagnostic, showing very good agreement. On the other hand, the q profiles have been compared with the results of several versions of the equilibrium code EFIT, proving that the algorithms developed can identify with good precision both monotonic and reversed shear profiles. An optimized version of the code has also been written and it has been verified that the computing time is compatible with the requirements of JET real time system. The present algorithm is being implemented and is used during JET experimental campaigns.

We present a preliminary investigation of the plasma column displacement in the RFX reversed field pinch (RFP) experiment, carried out with different diagnostics. Determinations of the plasma position have been obtained by a set of external pick-up coils, a multichannel far infrared (FIR) polarimeter and a soft x-ray (SXR) tomographic camera. These diagnostics provide complementary information about the plasma position. In particular, the location of the polarimetric chords allows a direct estimate of the horizontal magnetic axis shift. The analysis of the isoemissivity SXR surfaces also allow investigations of the differential shift of the various magnetic surfaces. A comparison of the results of these three diagnostics is reported, and their agreement is found to be satisfactory and consistent with our expectations.

The deliberate use of biological warfare agents (BWA) and other pathogens can jeopardize the safety of population, fauna and flora, and represents a concrete concern from the military and civil perspective. At present, the only commercially available tools for fast warning of a biological attack can perform point detection and require active or passive sampling collection. The development of a stand-off detection system would be extremely valuable to minimize the risk and the possible consequences of the release of biological aerosols in the atmosphere. Biological samples can be analyzed by means of several optical techniques, covering a broad region of the electromagnetic spectrum. Strong evidence proved that the informative content of fluorescence spectra could provide good preliminary discrimination among those agents and it can also be obtained through stand-off measurements. Such a system necessitates a database and a mathematical method for the discrimination of the spectral signatures. In this work, we collected fluorescence emission spectra of the main BWA simulants, to implement a spectral signature database and apply the Universal Multi Event Locator (UMEL) statistical method. Our preliminary analysis, conducted in laboratory conditions with a standard UV lamp source, considers the main experimental setups influencing the fluorescence signature of some of the most commonly used BWA simulants. Our work represents a first step towards the implementation of a spectral database and a laser-based biological stand-off detection and identification technique.

Recently, surveying large areas in an automatic way, for early detection of both harmful chemical agents and forest fires, has become a strategic objective of defence and public health organisations. The Lidar and Dial techniques are widely recognized as a cost-effective alternative to monitor large portions of the atmosphere. To maximize the effectiveness of the measurements and to guarantee reliable monitoring of large areas, new data analysis techniques are required. In this paper, an original tool, the Universal Multi Event Locator, is applied to the problem of automatically identifying the time location of peaks in Lidar and Dial measurements for environmental physics applications. This analysis technique improves various aspects of the measurements, ranging from the resilience to drift in the laser sources to the increase of the system sensitivity. The method is also fully general, purely software, and can therefore be applied to a large variety of problems without any additional cost. The potential of the proposed technique is exemplified with the help of data of various instruments acquired during several experimental campaigns in the field.

This paper describes the soft X-ray (SXR) cross-sectional distributions measured in the European Reversed Field Pinch (RFP) experiment RFX by means of a tomographic diagnostic. Due to the non-invasive character of this diagnostic, the plasma radiation emissivity pattern can be reconstructed. The emissivity depends on various crucial parameters and can be used to study several plasma properties. To reconstruct SXR emissivity in RFX, three different inversion schemes have been compared: the Cormack technique, based on the Bessel-Fourier series expansion, a maximum entropy algorithm and a hybrid method where the previous techniques have been merged together. In this paper the first survey of the SXR emissivity properties in the RFP configuration have been presented. Measurements have been taken in a variety of plasma conditions. The measured SXR data has been compared with the results of a one dimensional impurity diffusion model, which predicts the absolute SXR emissivity, with good agreement. In particular the effect of mode locking on plasma equilibrium in the standard Multiple Helicity (MH) states has been studied. This has given a significant contribute to the description of the plasma MHD structure, both in MH regime and in the Quasi Single Helicity (QSH) state, whose signature is the growth of a sizable m=1 structure in the emissivity distributions, and which has been experimentally discovered thanks to the tomography. A good correlation between SXR tomographic reconstruction and magnetic measurements has been found.

This paper illustrates the realisation and testing of the multichannel far-infrared polarimeter for the RFX machine, a plasma confinement experiment with the Reversed Field Pinch configuration. The diagnostic is described and the its main sections are outlined: the laser system operating at λ = 118.8μm, the FIR optical line, the beam splitter and the beam detection units, the signal processing electronics and the motion electronics for the calibration halfwave plates. Emphasis is placed on the work performed on the polarimeter to make it operational. We illustrate the preliminary tests of the diagnostic equipment and the activity performed to bring the diagnostic into reliable operation, in particular the reduction of the noise produced by magnetically induced mechanical vibrations during machine shots. An example of the first Faraday angle measurements performed on five chords are presented and discussed. The measured Faraday angle was also compared to a theoretically calculated value, based on a simple plasma model, with satisfactory agreement between measured and predicted data.

Particulate matter (PM), emitted by vehicles in urban traffic, can greatly affect environment air quality and have direct implications on both human health and infrastructure integrity. The consequences for society are relevant and can impact also on national health. Limits and thresholds of pollutants emitted by vehicles are typically regulated by government agencies. In the last few years, the interest in PM emissions has grown substantially due to both air quality issues and global warming. Lidar-Dial techniques are widely recognized as a cost-effective alternative to monitor large regions of the atmosphere. To maximize the effectiveness of the measurements and to guarantee reliable, automatic monitoring of large areas, new data analysis techniques are required. In this paper, an original tool, the Universal Multi-Event Locator (UMEL), is applied to the problem of automatically indentifying the time location of peaks in Lidar measurements for the detection of particulate matter emitted by anthropogenic sources like vehicles. The method developed is based on Support Vector Regression and presents various advantages with respect to more traditional techniques. In particular, UMEL is based on the morphological properties of the signals and therefore the method is insensitive to the details of the noise present in the detection system. The approach is also fully general, purely software and can therefore be applied to a large variety of problems without any additional cost. The potential of the proposed technique is exemplified with the help of data acquired during an experimental campaign in the field in Rome.

In a Tokamak the configuration of the magnetic fields remains the key element to improve performance and to maximise the scientific exploitation of the device. On the other hand, the quality of the reconstructed fields depends crucially on the measurements available. Traditionally in the least square minimisation phase of the algorithms, used to obtain the magnetic field topology, all the diagnostics are given the same weights, a part from a corrective factor taking into account the error bars. This assumption unduly penalises complex diagnostics, such as polarimetry, which have a limited number of highly significant measurements. A completely new method to choose the weights, to be given to the internal measurements of the magnetic fields for improved equilibrium reconstructions, is presented in this paper. The approach is based on various statistical indicators applied to the residuals, the difference between the actual measurements and their estimates from the reconstructed equilibrium. The potential of the method is exemplified using the measurements of the Faraday rotation derived from JET polarimeter. The results indicate quite clearly that the weights have to be determined carefully, since the inappropriate choice can have significant repercussions on the quality of the magnetic reconstruction both in the edge and in the core. These results confirm the limitations of the assumption that all the diagnostics have to be given the same weight, irrespective of the number of measurements they provide and the region of the plasma they probe.

Disruptions in tokamak devices are inevitable and can severely damage a tokamak device's wall. For this reason, different protection mechanisms have to be implemented. In the Joint European Torus (JET), these protection systems are structured in different levels. At the lowest level are those systems that are responsible for protecting the machine's integrity, which must be highly reliable. More complex systems are located at higher levels; these higher-level systems have been designed to take action before low-level systems. Since the installation of the new metallic wall in JET, new protection systems have been being developed to improve the overall protection of the device. This work focuses on a software application-a disruption predictor-that detects an incoming disruption. This software application simulates the behavior of a real-time implementation. In recent years, efforts have been devoted to developing and optimizing a reliable system that is capable of predicting disruptions. This has been accomplished by the novel combination of machine-learning techniques based on supervised learning methods. Disruptions must be predicted early enough so that the protection systems can mitigate the effects of disruptions. This paper summarizes the software development of the JET disruption predictor. This software simulates the real-time data acquisition and data processing. It has been an essential software tool to both optimize the disruption prediction model and implement a simulator of the real-time predictor.

Internal magnetic measurements are essential to obtaining reliable and accurate magnetic reconstructions in the interior of the plasma column in Tokamaks. In the last years, polarimetry has been increasingly used to provide global constraints to equilibrium codes. Joint European Torus (JET) polarimeter has four lateral channels, whose arrangement is similar to the topology of the diagnostic foreseen in ITER. A statistical analysis of JET polarimetry measurements have been provided in the past for only one vertical channel (CH3) using a polarimetry propagation code based on the Stokes vector formalism. A new propagation code has therefore been developed for the lateral channels to simulate and interpret the measurements of the Faraday rotation (FR) and Cotton–Mouton (CM) phase shift in JET. In this paper, a complete analysis of the integral form of the Stokes equation for the lateral channels is presented. In particular, the analysis shows that there is a strong interaction between the FR and CM effect even at low currents and low magnetic fields. The code results of both FR and CM have been used to estimate the line-integrated density and a proper benchmarking with experimental data has been performed. Two approximations to obtain the electron density in real time, using only plasma current in case of FR measurements and also toroidal magnetic fields in case of CM measurements, are proposed.

The reconstruction of the magnetic fields, which produce the confinement of the plasma, is an essential ingredient of tokamaks, since they affect both the operation of the devices and the interpretation of the physics results. This work reports a preliminary investigation to determinate whether the position of last magnetic flux surface can be evaluated from the images collected by JET visible cameras. To this end, the frames of some JET visible cameras have been analyzed with the phase congruency method to extract the position of the emission at the boundary on the high field side. The results of the comparison between the optical reconstruction of the plasma boundary obtained from the cameras and the separatrix position derived from the equilibrium code EFIT has been performed. Depending on the measurements used as inputs to EFIT, the difference between the two estimates of the separatrix position is below 10 cm.

This paper presents a very high-speed image processing algorithm applied to multi-faceted asymmetric radiation from the edge (MARFE) detection on the Joint European Torus. The algorithm was built in serial and parallel versions and written in C/C+ using OpenCV, cvBlob, and LibSVM libraries. The code implemented was characterized by its accuracy and run-time performance. The final result of the parallel version achieves a correct detection rate of 97.6% for MARFE identification and an image processing rate of more than 10 000 frame per second. The parallel version divides the image processing chain into two groups and seven tasks. One group is responsible for Background Image Estimation and Image Binarization modules, and the other is responsible for region Feature Extraction and Pattern Classification. At the same time and to maximize the workload distribution, the parallel code uses data parallelism and pipeline strategies for these two groups, respectively. A master thread is responsible for opening, signaling, and transferring images between both groups. The algorithm has been tested in a dedicated Intel symmetric-multiprocessing computer architecture with a Linux operating system.

In this paper, a new denoising method, based on the wavelet transform of the noisy signal, is described. The method implements a variable thresholding, whose optimal value is determined by analyzing the cross-correlation between the denoised signal and the residuals and by applying different criteria depending on the particular decomposition level. The residuals are defined as the difference between the noisy signal and the denoised signal. The procedure is suitable for denoising signals in real situations when the noiseless signal is not known. The results, obtained with synthetic data generated by well-known chaotic systems, show the very competitive performance of the proposed technique.

In Tokamaks, the use of cameras as diagnostics has increased remarkably in the last years. One of the main technical difficulties, in interpreting the data of camera-based diagnostics, is the presence of movements and vibrations of the field of view, consequence of various plasma phenomena, and modes of operation. Typically, these movements results in rigid transformations of the frames. In this paper, various information theoretic indicators (correlations, entropies, and mutual information) are described and their potential to detect the vibrations and to register images is investigated. The indicators are applied to a representative set of videos collected by Joint European Torus wide-angle infrared camera. For movement detection, the Tsallis entropy is the indicator with the best performance, providing a success rates in excess of 86% and an almost equal number of false and missed detections. Its application to image registration looks also very promising.

This article shows a reliable method to determine the number and position of local perturbations in the plasma emissivity from projections. The method is based on machine learning and in particular on the approach of conformal predictors. The developed technique is an alternative to tomography in thermonuclear plasmas.

The increasing importance of providing reliable polarimetric measurements in real time, for both machine protection and plasma control, has motivated the development of a quick version of the calibration algorithms for JET polarimeter. This new code, which interprets the calibration procedure performed before each shot, is based on a physical equivalent model of the diagnostic optical path and is valid for any operational regime of JET. It provides results before the plasma breakdown, and, with its estimates of the optical paths parameters, the polarimetric measurements have an accuracy more than sufficient for real-time purposes. New approximate equations have been validated in order to obtain the line integrated density from the newly calibrated horizontal chords, so that also these polarimetric measurements can also be used for density feedback and machine protection. The availability of reliable polarimetric measurements in real time opens new perspectives also to the determination of the plasma boundary, the magnetic equilibrium, and their use in advanced feedback control schemes.

Tomography is a technique which is widely applied to fusion plasmas as it can provide improved understanding of plasma emissivity distributions. It is challenging because of the sparse nature of data available from the measured plasma projections. An optimised version of robust and fast tomographic algorithm based on the Tikhonov regularisation constrained to Minimum Fisher Information is presented in this contribution. A new regularisation matrix enforcing preferential emissivity smoothness along magnetic flux surfaces is introduced. The paper also details application of advanced numerical methods which lead to a substantial decrease in computation time. Subsequent implementation of fast presolvers of the inverse problem further contributes to the algorithm's efficiency and also an improved stability of the tomography reconstruction. Finally, reliability and performance of the tomography algorithm is exemplified by the reconstruction of soft X-ray data evolution following tungsten ablation into a JET plasma. The resulting speed of reconstruction is compared to other referenced tomographic algorithms.

Recent advances in data mining allow the automatic recognition of physical phenomena in the databases of fusion devices without human intervention. This is important to create large databases of physical events (thereby increasing the statistical relevance) in an unattended manner. Important examples are the L/H and H/L transitions. In this contribution, a novel technique is introduced to automatically locate H/L transitions in JET by using conformal predictors. The focus is on H/L transitions because typically there is not a clear signature in the time series of the most widely available signals to recognize the change of confinement. Conformal predictors hedge their prediction by means of two parameters: confidence and credibility. The technique has been based on binary supervised classifiers to separate the samples of the respective confinement modes. Results with several underlying classifiers are presented.

A recent investigation of an anomaly in the polarimetric measurements at the Joint European Torus (JET) is reported. A calculation, based on the Stokes vector and the Mueller matrix method, has been developed to simulate the effects of the waveguide on the polarization of the propagating beams. These effects consist in both rotations of the polarization and alterations of the ellipticity, and affect both the probing and the reference beams. The model that is assumed here is made up of a rotated retarder followed by a rotator, with three free parameters to be optimized. A very good agreement between the resulting calculated curves and the experimental calibration diagrams is obtained for the four vertical channels of JET polarimeter. Variations of both phase shift and amplitude ratio are fitted by the model. This result supports the hypothesis that the waveguide system, which transfers the beams from the torus to the detectors, is the main source of the anomaly.

This paper presents a high-speed image processing algorithm applied to multifaceted asymmetric radiation from the edge detection on Joint European Torus. The algorithm has been developed in C/C++ language using the OpenCV library for image processing and the LibSVM library for pattern classification and has been tested on a dedicated Intel 64-bit Linux computing platform. The code implemented achieves at the same time high accuracy and good run-time performance. The final version presents a correct detection rate of 93.3% and an average image processing rate of 650 frames per second. A complete analysis of each image processing module is presented to illustrate the overall characteristics and performance of the algorithm.

This work summarizes the latest results on prediction with newly developed estimators based on statistical significance. These predictors implement conformal predictions and have been applied to classification tasks for data of the TJ-II stellarator. In particular, different adaptations to solve a 5-class image classification problem for the TJ-II Thomson scattering (TS) are presented. Off-line (nearest neighbour and support vector machines based) and real-time (SVM based) versions of conformal predictors have been developed. In all cases, if the classifications are reliable, the predicted images are incorporated to the training dataset for future predictions. The nearest neighbour classifier (NNC) obtains a success rate of 97% with confidence 0.96 and a mean credibility of 0.61. The CPU time to predict shows a linear dependence with the number of images in the training set ( t = 0.519 n + 100.212 s). The SVM classifiers are used in the one versus the rest approach. The off-line version provides a success rate of 99%, a confidence of 0.99 and an average credibility of 0.55. The CPU time also follows a linear law with the number of images in the training set ( t = 15.023 × 10 −3 n + 4.523 s). The real-time classifier achieves a success rate of 96% and a mean confidence and credibility of 0.99 and 0.53, respectively. In this case, after 395 classifications, the CPU time per image to classify remains constant: 89.7 ± 14.1 ms.

In magnetic confinement Nuclear Fusion video cameras have become routine diagnostics, which can produce gigabytes of data per discharge. New tools and methods are required to manipulate the frames of these videos to obtain the required information. New algorithms have been developed, which implement pattern recognition methods to search the repositories of videos on the basis of their content and not on their address. Real time automatic analysis of videos has motivated the development of machine learning tools to classify objects in images. The identification and tracking of objects in the videos of Tokamak diagnostics requires the computation of image descriptors, which are insensitive to rotation, translation and rescaling. The Hu moments have proved to be quite effective in performing this image analysis task (© 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim)

In many domains where images are produced and acquired, the field of view of cameras can be subject to oscillations and movements, which can induce errors in the interpretation of the frame contents and can even jeopardize the analysis of the videos. The problem is particularly severe in applications such as nuclear fusion, in which, typically, no stable and reliable reference points exist within the camera fields of view to register the frames. A nonadditive form of entropy S(q), which is more sensitive to long-range correlations than the Shannon entropy, has been applied to the problem of automatically detecting such camera movements in the videos of a JET wide-angle infrared camera. A systematic analysis of the results, covering more than 110 000 frames, has been undertaken, and the results obtained, reaching a total success rate of almost 97%, are more than satisfactory.

The performance of intelligent data acquisition systems relies heavily on their processing capabilities and local bus bandwidth, especially in applications with high sample rates or high number of channels. This is the case of the self adaptive sampling rate data acquisition system installed as a pilot experiment in KG8 B correlation reflectometer at JET. The system, which is based on the ITMS platform, continuously adapts the sample rate during the acquisition depending on the signal bandwidth. In order to do so it must transfer acquired data to a memory buffer in the host processor and run heavy computational algorithms for each data block. The processing capabilities of the host CPU and the bandwidth of the PXI bus limit the maximum sample rate that can be achieved, therefore limiting the maximum bandwidth of the phenomena that can be studied. Graphic processing units (GPU) are becoming an alternative for speeding up compute intensive kernels of scientific, imaging and simulation applications. However, integrating this technology into data acquisition systems is not a straight forward step, not to mention exploiting their parallelism efficiently. This paper discusses the use of GPUs with new high speed data bus interfaces to improve the performance of the self adaptive sampling rate data acquisition system installed on JET. Integration is sues are discussed and performance evaluations are presented.

An experiment with the aim of determining the neutron field characteristics for high performance discharges is being carried out on the JET tokamak. The neutron field parameters (yield, fluence, energy distribution) at a specific location outside the JET Torus Hall are to be measured simultaneously by three independent techniques (super-heated fluid detectors or “bubble detectors”, activation, and time-of-flight). Eventually the results are to be compared with neutron transport calculations. The bubble detector and activation measurement location is situated at the end of a vertical collimated line of sight, behind the TOFOR [1] time-of-flight spectrometer. The field-of-view of the neutron detectors can be varied by means of a pre-collimator.

In order to qualify materials and develop integrated scenarios for ITER, Joint European Torus (JET) is going to operate with a new wall consisting of beryllium in the main chamber and tungsten in the divertor. These new materials will require particular care when the machine operates, considering that they are much more vulnerable than the present combination of graphite and stainless steel. Early detection of hotspots, which are regions of the first wall where the temperature approaches dangerous levels, is considered one essential element fin the safety strategy. In this paper, cellular nonlinear networks (CNNs) are applied to the task of detecting hot spots in the infrared images of JET wide-angle camera. The ability of the CNNs to process the pixels of the images in parallel makes this technology a very good candidate for this task. Various algorithms are presented, which can locate the hot regions in any part of the image with a temporal resolution on the order of 60 ins, which is considered adequate for safety purposes at JET. In addition to the frame-by-frame static identification of the hot spots, their evolution in time is also followed to determine if they approach dangerous parts of the vacuum vessel. The potential of the CNNs would therefore allow for the implementation of alternative protection strategies, such as following the increase and displacement of the hot spots inside the entire vacuum vessel and identifying particles dropping into the plasma.

A data driven methodology to automatically derive a Fuzzy Logic Classifier (FLC) only on the basis of the raw signals available, is proposed. The first step is a feature selection performed with the approach of Classification and Regression Trees (CART), to extract the variables in the database which are the most critical for the problem under study. Then a CART is produced using only the previously selected features and is provided to a fully automated algorithm which determines the membership functions and the most appropriate rules to reproduce the classification tree obtained with CART. The resulting FLC attains good performance in terms of generalization and classification, still providing a set of rules which can be easily interpreted in order to achieve a first, intuitive understanding of the phenomenon involved. To assess the potentiality of the approach, the method has been applied to a synthetic database provided for the NIPS 2003 feature selection competition and to a real classification problem.

The importance of predicting the occurrence of disruptions is going to increase significantly in the next generation of tokamak devices. The expected energy content of ITER plasmas, for example, is such that disruptions could have a significant detrimental impact on various parts of the device, ranging from erosion of plasma facing components to structural damage. Early detection of disruptions is therefore needed with evermore increasing urgency. In this paper, the results of a series of methods to predict disruptions at JET are reported. The main objective of the investigation consists of trying to determine how early before a disruption it is possible to perform acceptable predictions on the basis of the raw data, keeping to a minimum the number of ‘ad hoc’ hypotheses. Therefore, the chosen learning techniques have the common characteristic of requiring a minimum number of assumptions. Classification and Regression Trees (CART) is a supervised but, on the other hand, a completely unbiased and nonlinear method, since it simply constructs the best classification tree by working directly on the input data. A series of unsupervised techniques, mainly K-means and hierarchical, have also been tested, to investigate to what extent they can autonomously distinguish between disruptive and non-disruptive groups of discharges. All these independent methods indicate that, in general, prediction with a success rate above 80% can be achieved not earlier than 180 ms before the disruption. The agreement between various completely independent methods increases the confidence in the results, which are also confirmed by a visual inspection of the data performed with pseudo Grand Tour algorithms.