Journal of Fusion Energy最新文献

Diagnostics are critical on the path to commercial fusion reactors, since measurements and characterisation of the plasma is important for sustaining fusion reactions. Gamma spectroscopy is commonly used to provide information about the neutron energy spectrum from activation analysis, which can be used to calculate the neutron flux and fusion power. The detection limits for measuring nuclear dosimetry reactions used in such diagnostics are fundamentally related to Compton scattering events making up a background continuum in measured spectra. This background lies in the same energy region as peaks from low-energy gamma rays, leading to detection and characterisation limitations. This paper presents a digital machine learning Compton suppression algorithm (MLCSA), that uses state-of-the-art machine learning techniques to perform pulse shape discrimination for high purity germanium (HPGe) detectors. The MLCSA identifies key features of individual pulses to differentiate between those that are generated from photopeaks and Compton scatter events. Compton events are then rejected, reducing the low energy background. This novel suppression algorithm improves gamma spectroscopy results by lowering minimum detectable activity (MDA) limits and thus reducing the measurement time required to reach the desired detection limit. In this paper, the performance of the MLCSA is demonstrated using an HPGe detector, with a gamma spectrum containing americium-241 (Am-241) and cobalt-60 (Co-60). The MDA of Am-241 improved by 51% and the signal to background ratio improved by 49%, while the Co-60 peaks were partially preserved (reduced by 78%). The MLCSA requires no modelling of the specific detector and so has the potential to be detector agnostic, meaning the technique could be applied to a variety of detector types and applications.

The activities of the EUROfusion consortium on the development of high quality nuclear data for fusion applications include evaluations of deuteron induced reactions and related data libraries for needs of the DEMO fusion power plant and IFMIF-DONES neutron-source nuclear analyses. Molybdenum is one of the major constituents of the reference stainless steels used in critical components of these projects. While the TENDL deuteron data library was the current reference used by EUROfusion, need of its further improvement has already been pointed out. The weak binding energy of the deuteron is responsible for the high complexity of its interaction with nuclei, involving also a variety of reactions initiated by the nucleons following the deuteron breakup. Their analysis completed that of the deuteron interactions with Mo and its stable isotopes, from elastic scattering to pre-equilibrium and compound–nucleus reactions, up to 50 MeV. A particular attention has been paid to the breakup, stripping, and pick-up direct interactions which amount to around half of the deuteron total–reaction cross section. The due account of most experimental data has validated the present approach, highlighted some prevalent features, and emphasized weak points and consequently the need for modeling/evaluation upgrade.

Plasma disruptions pose an intolerable risk to large tokamaks, such as ITER. If a disruption can no longer be avoided, ITER’s last line of defense will be the Shattered Pellet Injection. An experimental test bench was created at ASDEX Upgrade to inform the design decisions for controlling the shattering of the pellets and develop the techniques for the generation of the fragment distributions necessary for optimal disruption mitigation. In an effort to analyze the videos resulting from the more than 1000 tests and determine the impact of different settings on the resulting shard cloud, an analysis pipeline, based on traditional computer vision (CV), was created. This pipeline enabled the analysis of 173 of the videos, but at the same time showed the limits of traditional CV when applied in applications with a highly heterogeneous dataset such as this. We created a machine learning-based (ML) alternative as a drop-in replacement to the original image processing code using a semantic segmentation model to exploit the innate adaptability and robustness of deep learning models. This model is capable of labeling the entire dataset quickly, accurately and reliably. This contribution details the implementation of the ML model and the current state and future plans of the project.

ITER diagnostic systems provide measurements to the Plasma Control System (PCS) in real-time. These measurements are used for plasma control and machine protection. Latency is an important parameter in the assessment of such systems. It is a time gap between capturing an external event by hardware and finishing the processing of acquired data. PCS requires the diagnostic systems to introduce a maximum total latency of 10 to 100 ms, therefore, the systems need to be tested if they meet the requirements. The system evaluated in this paper is a reference real-time image acquisition system developed as a base for ITER diagnostic systems. It consists of hardware based on the Micro Telecommunications Computing Architecture (MicroTCA) standard, developed firmware, and software. It supports cameras with various interfaces. In the paper, two cameras, with a Camera Link and 1 GigE Vision interfaces were selected to perform latency evaluation. The paper presents two methods of measuring the latency of image acquisition. The first one is based on precise time stamping consecutive stages of acquisition. This approach allows for determining which step of acquisition takes more or less time. In consequence, the software or hardware can be optimized. The other one uses LED to evaluate a particular camera, by checking the time of camera reaction to the trigger. A dedicated testing framework is developed to perform automated tests to evaluate latency. It supports collecting and analyzing the results of measurements. Besides that, a dedicated hardware is used to perform the latency tests using LED. The results and discussion of the measurements are presented in the manuscript. They show the latency evaluated using earlier proposed methods, comparing the cameras used in the image acquisition system.

The WEST (W Environment in Steady-state Tokamak) divertor serves as the primary element for heat exhaust and contributes critically to plasma control. The divertor receives intense heat fluxes, potentially leading to damage to the plasma facing units. Hence, it is of major interest for the safety of divertor operation to detect and characterize the hot spots appearing on the divertor surface. This is done through the use of infrared (IR) cameras, which provide a thermal mapping of the divertor surface. In this work, a knowledge-informed divertor hot spot detector is demonstrated, that explicitly accounts for hot spot structure and temperature repartition. A novel neural network, termed as Constrained U-Net, is proposed, which uses as input the bounding boxes of hot spots from prior automatic detection. The Constrained U-Net addresses jointly image segmentation and regression of physical parameters, while remaining compatible with the practical constraints of real-time use. The detector is trained on simulated data and applied to real-world infrared images. On simulated images, it yields a precision of 0.98, outperforming a classical U-Net, and Max-Tree. Visual results obtained on real-world acquisitions from the WEST Tokamak illustrate the reliability of the proposed method for safety studies on hot spots.

We provide a comprehensive review of the applications of the Bayesian method across various fusion devices. The progression and widespread adoption of the Bayesian method are evident in the field. Our focus is primarily on Bayesian probability theory and Gaussian process regression, aiming to offer clear definitions for each term in the formula. To facilitate understanding, we categorize the works based on the specific fusion device, enabling readers to assess the current state of development for the Bayesian method within each device. The numerous successful applications of the Bayesian method in analyzing diagnostic data from European devices underscore its significant potential and advantages.

The system proposed to measure the tritium to deuterium ratio on the International Thermonuclear Experimental Reactor (ITER) is a high-resolution neutron spectrometer, partly composed of a system of three Thin-foil Proton Recoil (TPR) spectrometers. This system works on the principle of converting neutrons into protons using a thin foil of polyethylene, which is then detected in silicon detectors to obtain the scattering angles and energy spectrum of the protons. The objective of this article is to show the benefit of artificial intelligence for improving a simple TPR system model written in Python to an accuracy approaching MCNP simulations, while significantly decreasing the computational cost. The first step was to model a polyethylene converter to obtain the energy-angle distribution of outgoing protons for a given incident neutron beam. When compared with MCNP, this simplified model was found to fail to account for proton energy and angular scattering. Therefore, in a second step, two neural networks were successfully trained to include these effects based on the output data of the TRIM code, assuming Gaussian distributions. The Python model was able to produce results very close (differences up to a few percent) to those obtained with MCNP by integrating these neural networks. To extend the study, the energy spectra of the protons could be obtained and subsequently used to obtain information on the ratio of deuterium and tritium in the plasma.

The accumulation of heavy impurities like tungsten in the plasma core of fusion devices can cause significant radiative power losses or even lead to a disruption. It is therefore crucial to monitor the tungsten impurity concentration. In this paper, we follow the integrated data analysis approach using Bayesian probability theory to jointly estimate tungsten concentration profiles and kinetic profiles from soft X-ray, interferometry and electron cyclotron emission measurements. As the full Bayesian inference using Markov chain Monte Carlo sampling is time-consuming, we also discuss emulation of the inference process using neural networks, with a view to real-time implementation.

Considering the high energy neutron environment in a nuclear fusion reactor, Reduced Activation type Ferritic/Martensitic Steels (RAFMS) containing tungsten, have been carefully curated from their surrogate Cr–Mo type Ferritic/Martensitic Steels (FMS). The substitution of molybdenum by tungsten improved the radiation stability and mechanical characteristics RAFMS. However, the effect of tungsten on the liquid metal corrosion resistance of FMS has not been well investigated. The current work attempts to estimate liquid metal compatibility by examining the surface oxides of Indian RAFMS (IN RAFMS) and its surrogate steel, P91 (9Cr-1Mo), using X-ray Photoelectron Spectroscopy. Subsequently, thermodynamic calculations have been used to establish the stability of such oxides in both ambient circumstances and liquid lead–lithium eutectic alloy (Pb–Li). The results showed that tungsten can provide a higher resistance to liquid metal attack than molybdenum because its oxides are more stable. Actual corrosion experiments with IN RAFMS and P91 were performed in liquid Pb–Li for a durations upto 2000 h, successfully validating the above stated prediction.

The understanding and control of complex systems in general, and thermonuclear plasmas in particular, require analysis tools, which can detect not the simple correlations but can also provide information about the actual mutual influence between quantities. Indeed, time series, the typical signals collected in many systems, carry more information than can be extracted with simple correlation analysis. The objective of the present work consists of showing how the technology of Time Delay Neural Networks (TDNNs) can extract robust indications about the actual mutual influence between time indexed signals. A series of numerical tests with synthetic data prove the potential of TDNN ensembles to analyse complex nonlinear interactions, including feedback loops. The developed techniques can not only determine the direction of causality between time series but can also quantify the strength of their mutual influences. An important application to thermonuclear fusion, the determination of the additional heating deposition profile, illustrates the capability of the approach to address also spatially distributed problems.