Development of an attention aware multimodal network for the identification of internal reconnection events in LTX-β
Year: 2026
Type: article
Abstract: We present the IRE Identification Multimodal Network (IIMN), a classifier to automatically detect Internal Reconnection Events (IREs) in spherical tokamaks. The IIMN utilizes data from the Lithium Tokamak eXperiment-β (LTX-β), a low-aspect-ratio spherical tokamak capable of operating with walls coated in solid or liquid lithium. As a multimodal network, the IIMN is well-suited to handling complex, multi-dimensional relationships among both image and tabular data. To train the model, 7,265 events from 2,968 shots were manually reviewed and labeled, covering nearly two years of diagnostic data from LTX-β. Each event includes key plasma parameters in identifying IREs like the time evolutions of the plasma current, average electron density, and Mirnov spectrogram. The model concatenates the results from a multilayer perceptron branch and a convolutional neural network (CNN) branch to return a one-hot encoded output that determines if the event is an IRE or not based on the event’s plasma parameters. We show that the IIMN has an accuracy of 94.4±0.9%, a precision score of 0.934±0.014, and a recall score of 0.931±0.014. IIMN is fully automated such that it could be used quickly and easily by operators in the control room for spherical tokamak operation. To realize full automation and zero human intervention, our current multimodal architecture incorporates a CNN that directly analyzes image-based data from select plasma diagnostics. By automating IRE detection, this tool can assist operators in monitoring fusion performance and developing strategies to reduce confinement losses even in between two successive plasma discharges.
Institutions: Princeton Plasma Physics Laboratory, Princeton University
Language: English
Subfield: Nuclear and High Energy Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,325
OpenAlex ID: W7225423272