Approximately 50 diagnostics will be installed on ITER, distributed in 20 ports, on the vacuum vessel surface, and in the divertor. These diagnostics will measure more than 100 parameters necessary for control of the plasma and first wall processes in order to achieve the required goals, and to gain the knowledge needed for future reactor designs. Many diagnostic projects are now moving forward thanks to support of teams working cooperatively around the world, with several components being already manufactured, delivered to ITER site and installed.

The diagnostics on ITER will be subject to new challenges unprecedented in today’s tokamaks. The diagnostics will operate in a nuclear environment which requires the design to mitigate the transmutation, radiation damage, and thermo-electric effects, as well as to cope with nuclear heating. The diagnostics will be subject to very limited or no maintenance, and they need to be designed with very high reliability and/or redundancy. The diagnostic components installed inside the vacuum vessel require rugged design to withstand e.g. bakeout conditions and exposure steam.

Since ITER is a nuclear facility, the design, manufacture and installation of the diagnostic components is subject to safety and quality requirements, in particular for installation on the nuclear confinement barriers such as the vacuum vessel, and vacuum vessel feedthroughs and windows.

This contribution will describe how these challenges and requirements are successfully met in the diagnostic designs through standardization, quality requirements, dedicated R&D, and design for protection and replacement of critical items. The contribution will also outline the design of the diagnostic ports, engineered to accommodate several diagnostic systems and their services, whilst maximizing the nuclear shielding performance and respecting the weight limit.

In magnetic fusion devices, the measurement of plasma quantities is essential to ensure the control of the plasma discharge, to increase plasma performance while ensuring safe operation, and to perform advanced physics analysis. Meeting such requirements is directly linked to the diagnostics capabilities. On any tokamak, a set of about 50 complementary diagnostics are implemented for such purpose. Among this list of measurement systems, the magnetic diagnostics play a central role in magnetic fusion devices. They contribute directly to the plasma position measurement and shape control. They are also used for advanced physics measurements to assess the plasma performance such as the plasma energy content. Finally, they provide information about magnetic perturbations in the magnetic configuration through the measurement of magnetic fluctuations. As a result, magnetic diagnostics must observe demanding specifications because their performance directly affects margins and therefore operational flexibility near machine limits.

Inverse problems involve the conversion of line-integrated data into the spatially localized information which is essential for detailed analysis. The most prevalent example is the CT scan which has revolutionized noninvasive medical diagnosis. In magnetic confinement experiments, these techniques are utilized for a variety of diagnostics ranging from soft X-ray to interferometry to neutron diagnosis in fusion plasmas. The constraints on locations for detector installation in toroidal devices challenge the state of the art of inverse problems by rendering them extremely underdetermined. In this lecture we focus primarily on the application of these inverse methods to bolometer diagnostics for the measurement of total radiated power from the plasma. Resistive and imaging bolometers are briefly introduced and their employment for numerous inverse problems are described. The calculation of the projection (or geometry) matrix and its uses for both forward modelling (synthetic diagnostics) and tomography are explained. Various inversion techniques are elucidated including singular value decomposition for a 1D inversion of resistive bolometer data [1], a novel 2D relative gradient smoothing technique applied to a traditional arrangement of two resistive bolometer arrays a poloidal cross section [2], Phillips-Tikhonov regularization used for the 1D inversion of interferometry data [3], the 2D inversion of data from an imaging bolometer with a tangential view of a tokamak [4] and for 3D tomography in a helical device with multiple imaging bolometers [5]. Finally, the recent application of SART and Bayesian techniques to improve 2D tomography of data from an imaging bolometer with a tangential view in a compact tokamak will be reviewed [6].

[1] B.J. Peterson et al., Plasma Phys. Contr. Fusion 45 (2003) 1167.
[2] D. Zhang et al., Nucl. Fusion 61 (2021) 116043.
[3] Y. Ohtani et al., Plasma Phys. Contr. Fusion 62 (2020) 025029.
[4] J. Jang et al., Rev. Sci. Instrum. 89 (2018) 10E111.
[5] R. Sano et al., Rev. Sci. Instrum. 87 (2016) 053502.
[6] F. Federici, Ph.D. Thesis, University of York (2023) Ch. 3.

The collection, processing, and interpretation of data on fusion experiments sits at the very core of fusion research. Scientific data is a key product of our experiments and, if cataloged and stored correctly, will persist as a valuable resource long after the devices that created it are decommissioned.
The first step along any reproducible data processing pipeline consists of metadata capture, including data providence. When analyzing a magnetics signal it is important to know not only the signals values but also the signals’ metadata, for example, the name and location of the diagnostic, any calibration factors or upstream data processing that has be applied to the signal, and the signals units. This additional metadata is of vital importance and must be stored alongside the signal itself such that the data retains its value in years to come. To facilitate this endeavor the Fusion community has developed a standard Data Dictionary. This Dictionary defines a number of Interface Data Structures IDSs. These IDSs standardize the storage and distribution of Tokamak data ether between users or between codes. Students will be introduced to the Integrated Modeling and Analysis Suite IMAS and a number of key IDSs used for storing and consuming Tokamak data.
Digital signal processing applications are diverse, ranging from the relatively simple down-sampling, filtering, and feature extraction all the way to advanced multi-diagnostic inference schemes that make use of Bayesian priors used to disentangle errors from underlying signals.
As digital signal processing is required by many scientific and industrial fields there exists a rich ecosystem of mature, open-source applications that are of great benefit to the Data Scientist.
This Lecture will introduce the IMAS Data Dictionary alongside number of popular digital signal processing techniques sourced from the Python community. Students will have the opportunity to learn some of these techniques firsthand by taking part in the ITER School Fusion data science challenge. This mini competition will provide students with the opportunity to apply data science techniques learnt during the ITER school to the solution a real-world fusion inference problem. Students will be guided through the setup of an appropriate data science environment using Python and associated data science libraries.
Participation will require the students to each bring a laptop with them to the ITER School.

Operating fusion plasma involves a multi-objective task that aims to maintain stability while achieving desired outputs such as fusion power, efficiency, and confinement quality. This leads to an optimization problem for these objective functions, requiring control strategies to find optimal solutions. Generally, it is assumed that a control solution can be derived from the physics model describing the system. However, the physics-based model of fusion plasma is often too complex for real-time application or may be unavailable due to its strong nonlinearity and time-variant nature. Adaptive control and data-driven (ML/AI) schemes are effective in addressing these challenges. By categorizing the plasma system into finite states and utilizing adaptive control policies, the complexity of control is reduced. Concurrently, ML/AI models enable fast physics solvers and diagnostic processing, allowing the integration of enriched physics information into real-time control systems. These tools facilitate the derivation of optimal control solutions for desired fusion plasma conditions in real time. This talk will introduce this concept, perspective, and progress we made on the control of plasmas. In addition, promising examples will be presented, including optimized high-confinement plasma operation with actively controlled or avoided edge instability and tearing instability in DIII-D and KSTAR tokamaks.

The scattering of electromagnetic radiation by electrons is known as Thomson scattering. Thomson scattering has been a cruical plasma diagnostics on fusion devices since the beginning of fusion research. It is capable of delivering the kinetic profiles of both electron temperature and density at the same time. Thomson scattering can be used for diagnostic purposes in two regimes, the coherent regime, where the collective electron phenomena are observed and the incoherent regime, where scattering of individual electrons is measured. In the lecture, I will touch the basic principles of Thomson scattering. I will then be showing the different types of instrumental strategies how to use them for a diagnostics purpose.
LIDAR Thomson scattering technique uses backscattering geometry and the time of flight methodology to measure profiles at distinct positions. A more typical setup used for Thomson scattering uses a high power laser probe beam fired through a plasma. The scattered light is collected and transported to a spectrometer (or set of polychromators) where it is spectrally resolved. As a detection part, either CCD camera or Avalanche photodiodes are used. Polychromators using high performance interference filters (with extreme rejection of laser light) and Avalanche photodiode coupled with pulsed Nd:YAG lasers are the most commonly used diagnostic setup nowadays. Commercially available high power laser frequencies are in range of tens Hz and thus do not provide sufficiently temporal resolution for studies of fast plasma events. In recent years, a custom made high repetition laser (up to 100 kHz) has been developed and is being improved through ongoing R&D development. Thomson scattering diagnostics should be always designed for a specific device and a required set of parameters. The extremely low cross-section of this diagnostics makes its design process challenging. The data processing should take into account all necessary inputs like scattering geometry, calibration constants and special effects in case they cannot be neglected (relativistic blueshift, depolarisation). Using a synthetic forward and backward models of diagnostic can help in both design stage and processing data stage.

The US Department of Energy (DOE) makes substantial investments in the production and collection of massive amounts of scientific data through supporting the user facilities and scientific software. The high-performance computing (HPC) resources supported by the Office of Advanced Scientific Computing Research (ASCR) provide an ideal platform for applying scientific machine learning (SciML) on these massive data to accelerate scientific discoveries. However, an efficient, scalable, federated algorithm is necessary to apply SciML to distributed data produced at scientific user facilities. There is a push at Oak Ridge National Laboratory (ORNL) to develop the next generation of smart laboratories (https://www.ornl.gov/intersect), locally developing connections between experimental and computational facilities at ORNL. This talk will focus on introducing bib and deep data analysis methods for scientific discoveries in large scale connected facilities.

Electron Cyclotron Emission (ECE) and Reflectometry Diagnostics in Magnetic Fusion Devices: An Overview
In this lecture, the fundamental principles, instrumentation, and applications of two pivotal diagnostic techniques in magnetic fusion research: Electron Cyclotron Emission (ECE) and Reflectometry, will be explored. These diagnostics are essential for understanding and optimizing the behavior of high-temperature plasmas in fusion devices such as tokamaks and stellarators.

Electron Cyclotron Emission (ECE): ECE diagnostics utilize the emission of electromagnetic radiation by electrons gyrating around magnetic field lines at the electron cyclotron frequency. This technique provides localized measurements of the electron temperature profile with high spatial and temporal resolution. ECE operation across different harmonics and frequency ranges, and the methods used to interpret ECE data, will be discussed. Additionally, the design and operation of ECE radiometers, including heterodyne and Michelson interferometer systems, and recent advancements in ECE technology will be highlighted.

Reflectometry: Reflectometry diagnostics use the principle of wave reflection, similar to radar, to infer electron density profiles and fluctuations. The basic principles of reflectometry, including the propagation and reflection of electromagnetic waves in plasmas, and the interpretation of phase and time-delay measurements, will be explained. The lecture will address various types of reflectometry systems, such as fixed-frequency, frequency-modulated continuous-wave (FMCW), and correlation reflectometry, along with their respective advantages and limitations.

Throughout the lecture, the complementary nature of ECE and reflectometry in providing comprehensive insights into plasma parameters will be emphasized. Examples from current fusion experiments, illustrating how these diagnostics contribute to the understanding of plasma confinement, stability, and turbulence, will be presented. Furthermore, the challenges and future directions in the development of ECE and reflectometry diagnostics, particularly in the context of next-generation fusion devices like ITER, will be discussed.
This lecture aims to equip students and researchers with a solid understanding of ECE and reflectometry diagnostics, fostering their ability to apply these techniques in their own research and contribute to the advancement of magnetic fusion energy.

Data-driven approaches, especially deep learning utilizing neural networks (NNs), have brought tremendous progress in fusion research over the last decade. One of the outcomes is acceleration in integrated simulations of fusion plasmas. We perform integrated simulations for a variety of purposes, including understanding physical phenomena, predicting plasma performance and designing future fusion devices. A turbulent transport model, which is a component of the integrated codes and has a role in calculating fluxes driven by turbulence, is often a computational bottleneck in the integrated simulations. First-principle-based reduced turbulent transport models show reasonable agreement with experiments. They, however, take several hours to days with parallel computation to give steady-state temperature and density, which determine fusion power, despite the fact that they are reduced models. To accelerate calculations of turbulent fluxes, several NN-based turbulent transport models have been developed. Such models successfully predict turbulent fluxes much faster than the conventional reduced models, and furthermore, they can emulate the first-principle simulations. In this lecture, integrated simulations are introduced in terms of turbulent transport and the development of NN-based turbulent transport models for integrated simulations is presented. Also, outcomes obtained by bringing the results of the first-principle simulations with the NN models into integrated simulations will be discussed.

The accurate measurement of plasma parameters is crucial for understanding and controlling fusion plasmas in magnetic confinement devices like tokamaks. This lecture will introduce the principles and applications of interferometry and polarimetry in plasma diagnostics. Interferometry allows for precise measurements of electron density, while polarimetry provides information on magnetic field structures (current profiles) and density. The theoretical foundations of these techniques are based on the fact that the refractivity of plasmas as a medium is a function of plasma density and magnetic field in cold plasmas. By measuring the variations in phase and polarization of electromagnetic waves in plasmas, plasma parameters can be determined. Recent developments have shown that for burning plasmas like ITER, the finite electron temperature effect on refractivity must be considered. These effects arise not only from electron pressure but also from the relativistic electron mass. The practical implementation of interferometry and polarimetry in MST reversed field pinch, C-Mod, DIII-D, and EAST tokamaks will be presented to illustrate recent advancements, key challenges, and future research directions.

Neutron diagnostics is one of the most important ones in burning plasma devices. The major subjects of neutron diagnostics are plasma performance evaluation, burn control, and physics understanding of the burning plasmas. The major targets of neutron diagnostics are the total neutron emission rate, time-integrated neutron yield, the neutron source profile, and the neutron spectrum. In this lecture, the basic neutron emission process in the plasma is introduced. Moreover, typical neutron diagnostics are explained based on the ITER and other fusion devices. The total neutron emission rate is measured with a neutron flux monitor (NFM), a divertor neutron flux monitor (DNFM), and a micro-fission chamber (NFM), which employs U-235 or U-238 fission chamber and is installed inside the vacuum vessel or near the vacuum vessel. The time-integrated neutron yield is measured with a neutron activation system (NAS), where the capsule with a small amount of materials, so-called activation foils, is transferred to the irradiation points near the plasma. The shot-integrated neutron yield is derived from the radio-activities of those activation foils. In ITER, irradiation points are located in the upper port, equatorial ports, under the divertor cassette, and near the inboard blanket. The neutron emission profile is measured with horizontal and vertical fan arrays of neutron detector arrays. The neutron detector combing of an organic scintillator, diamond detectors, and U-238 fission chambers is employed in ITER. A high-resolution neutron spectrometer system consisting of time-of-flight spectrometers, diamond neutron spectrometers, and recoil-proton spectrometers is planned. The absolute calibration of the detection efficiency for the total neutron yield is one of the most important issues for neutron diagnostics, especially for in-vessel neutron diagnostics because in-vessel neutron diagnostics do not have any collimator and are sensitive to neutrons from the whole plasma.

 

Visible spectroscopy has been widely used as a fundamental diagnostic method in fusion-relevant magnetically confined plasma and will become more critical in ITER. In this seminar, we will focus on active spectroscopy with the aid of a neutral beam, such as charge-exchange recombination spectroscopy (CHERS, CXRS, CXS), motional stark effect spectroscopy (MSE), and beam-emission spectroscopy (BES). The CXRS has been developed to measure ion temperature, flow velocity, and impurity concentration using the low-Z impurity lines such as Helium, Beryllium, Carbon, and Neon since 80′. Recently, this technique has been applied to measure bulk flow velocity or isotope ratio profiles (bulk-CXRS), fast ions injected by the neutral beam (FIDA, FICXS), and ion velocity distribution (fast-CXS) with high time resolution to study phase-space dynamics. Since these techniques rely on the charge exchange reaction with a neutral beam, the charge exchange cross section should be large enough to overcome the background emission. Therefore, the heating neutral beam with a positive ion source has been used as a diagnostic beam because the beam energy (40 ~50 keV/amu) matches the peak of the charge exchange cross section. The charge exchange cross section significantly drops at higher beam energy (for example, one order of magnitude at 200 keV/amu), the energy of the heating neutral beam in the ITER (~ 500keV/amu) is too high and cannot be used as a diagnostic beam. The diagnostic neutral beam with lower energy is indispensable in ITER for CXRS. The MSE has been widely used to measure the pitch angle of the local magnetic field and safety factor profile by detecting the change in the angle of polarized Ha or Da emission due to the motional stark effect. In contrast to CXRS, the high beam energy becomes a significant advantage for the MSE because the motional stark effect becomes larger as the beam’s velocity increases. The other active and passive spectroscopy are also discussed in this seminar.

Plasma exhibits complex and diverse phenomena, and various analytical techniques have been devised for its understanding. A crucial aspect of these analytical techniques is how data is processed and presented for human comprehension to extract characteristic relationships and anomalies among variables hidden within the raw data of complex and diverse dynamic experiments and simulations. This lecture introduces data visualization methods using virtual reality technology and interactive analytical techniques. The concept of virtual reality was proposed by the US company VPL in 1989, and since then, there have been rapid technological innovations, leading to the development of large-scale immersive virtual reality devices and head-mounted displays with wide field of view, high resolution displays, and spatial audio at affordable prices, flooding the market with virtual reality content. Virtual reality is being utilized as one of the forms of entertainment, such as in the metaverse. The potential of virtual reality extends beyond entertainment to significant applications in the scientific field. The lecture touches upon the history and workings of virtual reality devices and introduces several applications to plasma science.

The large quantity of data generated by infrared imaging diagnostics during the operation of long-pulse machines makes it cumbersome for humans to cope with the real-time analysis of thermal events. There is a strong need for a process able to detect and analyze thermal events automatically and in real-time, for feedback control and machine protection. Real-time image processing using machine learning (ML) and artificial intelligence (AI) offers a solution for enhancing the operational safety and efficiency of fusion machines such as ITER.

Infrared cameras are employed to continuously capture thermal images of the reactor’s interior, which are then analyzed using deep learning algorithms, particularly convolutional neural networks (CNNs). These models enable the automatic detection and classification of thermal anomalies, providing rapid and accurate responses to potential faults or inefficiencies.

The lecture will cover the development and implementation of AI-driven image processing systems, highlighting their ability to significantly reduce the reaction time to potential hazards. By training ML models on datasets of thermal images, these systems are able to identify patterns of interest. This capability is crucial for maintaining the stability and safety of the reactor, as it may allow for the prompt addressing of any deviations from normal operating conditions. The real-time processing aspect ensures continuous monitoring and immediate intervention, minimizing the risk of damage and operational downtime.

The principles and methodologies discussed in this lecture are mainly developed and tested using data from the WEST tokamak, which serves as an ITER-relevant test bench. They are designed for and adapted to other fusion machines, and in particular ITER that will require a protection system from day one. The session will provide an in-depth technical exploration of these AI systems, their operational framework, and their impact on the future of fusion energy research. Attendees will gain an understanding of how ML and AI can be leveraged to create safer and more efficient fusion reactors.

In order to implement feedback controllers for plasma regulation in tokamaks, a reliable estimation of the plasma state is required in real time. Moreover, this plasma state estimation is the initial condition for any possible plasma-evolution forecasting capability. Therefore, the quality of the estimated state at time t has a very strong impact on the quality of the forecasted state at time t + ∆t. Unfortunately, the estimation of plasma scalar and profile properties in real tokamaks is limited by the level of measurement noise in the system and by the availability of reliable internal plasma diagnostics with adequate spatial and temporal resolution. This will be particularly problematic in ITER and future fusion reactors because the level of nuclear noise due to fusion-induced neutron flux will be much higher than that found in present tokamaks. In addition, it is anticipated that the number of diagnostics in future fusion reactors will be dramatically reduced due to the need to maximize the blanket area for production of electricity, thereby reducing the space for diagnostics. The state estimator (observer) plays a critical role in control design when the state of the to-be-controlled system is not directly measurable or only noisy measurements are available. The closed- loop observer design challenge is to make the estimated state converge to the actual state by regulating the tradeoff between model prediction (the core of the observer is a predictive model) and available noisy measurements. The observer filters any measurement noise not consistent with the physics of the plasma evolution, i.e., not predicted by the predictive model embedded in the observer. Moreover, by exploiting the physics knowledge embedded in the predictive model, closed-loop observers can provide a real-time estimation of the plasma state from a reduced set of diagnostics. This lecture will provide an overview of the diagnostic needs for real-time control and will discuss how the raw diagnostic data, which is limited and noisy, can be combined with plasma response models in the form of observers to reconstruct in real time the plasma state needed for feedback control. Moreover, the possibility of leveraging the observer architecture as a residual generation mechanism to detect and isolate in real time faulty diagnostics and enable fault-tolerant control schemes will also be discussed.

Reliable analysis of experimental data is a key step to study the physics of fusion plasmas. The goal is to maximally exploit the information provided by the large amount of experimental data from a multitude of diagnostics of present day and future fusion devices. Since all measurements suffer from statistical and systematic uncertainties, a data analysis method benefits from its capability to process any kind of uncertain measurements. Data analysis within a Bayesian framework allows for consistently combining and processing any kind of uncertain information from measurements as well as modelling. It utilizes probability theory for most reliable estimation of physics quantities and their uncertainties. The foundation, recipes and illustrating examples of Bayesian inference will be presented in a first part of the presentation.
A second part will address the coherent combination of data from heterogeneous diagnostics and modelling codes being a major challenge in nuclear fusion research is [1]. Measured data from different diagnostics often provide information about the same subset of physical parameters (redundancy). Additionally, information provided by some diagnostics might be needed for the analysis of other diagnostics (complementarity). A joint analysis of complementary and redundant data in a Bayesian framework allows, e.g., to improve the reliability of parameter estimation, to increase the spatial and temporal resolution of profiles, to obtain synergistic effects, to consider diagnostics interdependencies and to find and resolve data inconsistencies. Modelling codes may provide additional physical information allowing for an improved treatment of ill-posed inversion problems. The concept of Integrated Data Analysis (IDA) [2] in the framework of Bayesian probability theory will be introduced and contrasted with conventional data analysis. Applications from nuclear fusion research will highlight various aspects of IDA and the respective benefits.
References
[1] R. Fischer, A. Dinklage, and E. Pasch, ”Bayesian modelling of fusion diagnostics”, Plasma Phys. Control. Fusion 45 1095-1111 (2003)
[2] R. Fischer, et al., ”Integrated data analysis of profile diagnostics at ASDEX Upgrade”, Fusion Sci. Technol. 58 675-684 (2010)