science
Understanding turbulence at the boundaries of magnetically confined plasmas in tokamak devices is fundamental to fusion research. Researchers call this boundary the Last Closed Flux Surface (LCFS). The LCFS is where the line shape of the magnetic field transitions from a ‘closed’ to an ‘open’ state. A “closed” region is where the field lines do not intersect the material surface, forming a closed magnetic flux surface. The “open” regions are where the magnetic field lines end up intersecting the material surface. This causes a rapid loss of particles and energy that reach these magnetic field lines. Around this boundary there is a region of enhanced turbulence-driven transport across magnetic field lines called ‘perpendicular’ transport. The transport that occurs in this region is important because of its role in plasma confinement. This transport is also important. This is because device operators influence how they deal with heat rejection and particle loading on intersecting material surfaces. Researchers use a technique called Gaspuff Imaging (GPI) to visualize the phenomena occurring at and around plasma boundaries in both space and time. GPI produces videos that researchers can analyze to study a type of turbulence called “blobs.”
impact
One of the challenges when analyzing GPI data is the number of frames in the GPI video. About 1 million frames per experiment. This is too much video data for the human eye to analyze. Researchers evaluate the properties of the blobs using traditional data analysis approaches. However, these methods either yield only average properties of blobs or use custom, non-standard workflows, making them difficult to use. Machine learning offers a new solution that provides per-blob tracking per frame. Machine learning technology excels at identifying and tracking objects in images, and there are many models available for this task. Blob tracking provides detailed temporally and spatially resolved information about blobs and serves as a powerful tool for estimating blob statistics. Therefore, it may be a useful tool for the design and operation of future fusion power devices. The study, published in the journal Nature Scientific Reports, was his 37th most downloaded physics paper among nearly 1000 physics papers published in 2022. All publications, data, models and code are publicly available to make the research as widely useful as possible. Available.
summary
GPI injects a small amount of neutral gas into the plasma. The technique then utilizes a line of sight tangential to the magnetic field to capture visible light resulting from the interaction of the plasma with the gas cloud. The scientist can then analyze the video from his GPI to study turbulence with filamentary structures called blobs. The blob’s radially outward movement widens the fusion device’s exhaust channels, thereby reducing the peak heat and desired particle flux to the diverter’s plate. However, the same movement can also increase unwanted plasma interactions with other components facing the plasma. Evaluation of blob size, velocity, and frequency allows evaluation of particle and energy fluxes to plasma-facing material surfaces in tokamak.
An interdisciplinary team of researchers from the Massachusetts Institute of Technology’s Center for Science and Fusion, the Department of Civil and Environmental Engineering, the Laboratory for Computer Science and Artificial Intelligence (CSAIL), and the Swiss Plasma Center of the Federal Institute of Technology Lausanne (EPFL), is part of EPFL’s Tokamak à We have developed and compared computer vision methods in machine learning, such as optical flow, to track blobs of GPI data from Configuration Variables (TCVs). Researchers implemented a novel application of four well-known, standardized, and benchmarked tracking methods trained on synthetic GPI data to replicate human blob identification as closely as possible. Two of these methods show excellent agreement with human labeling on real GPI data and successfully predict the theory-defined blob regime, consistent with the results of conventional methods. Did. This demonstrates the effectiveness of machine learning approaches in blob tracking applied to important research investigations. The machine learning method predicts the blob dynamic regime based on size estimation, consistent with conventional methods, and thus contributes to the validation of theoretical transport models for plasma boundaries. The specific ability to track individual blobs allows for blob-by-blob analysis of very large datasets and, ultimately, particle confinement, one of the key issues on the road to practical fusion energy. contribute to a better understanding of The researchers also published datasets and benchmarks with the aim of lowering the barrier to entry for tokamak plasma research, thereby significantly broadening the community of scientists and engineers who can contribute to the endeavour.
fundraising
This work was supported by the Department of Energy’s Office for Science, the Fusion Energy Science Program, and the Swiss National Science Foundation. This work is carried out within the framework of the EUROfusion Consortium and is funded by the European Union through the Euratom research and training programme.
