Choose timezone
Your profile timezone:
CD Technical Meeting (VI5): Combined talks: Interactive Tools & Digitals Twins
→
Europe/London
,
Description
1. Interactive Tooling for Tokamak Data Annotation and Exploration
Samuel Jackson
Effective interpretation of tokamak experimental data requires not only scalable analysis pipelines but also intuitive interfaces that enable physicists, machine learning practitioners, and data engineers to rapidly label, explore, and contextualise large shot catalogues. We present two open-source tools developed at UKAEA that address complementary stages of this workflow, from structured data annotation through to interactive exploratory analysis.
The first tool, “Toktagger”, is a web-based annotation platform for curating labelled datasets from tokamak diagnostics. It provides an intuitive interface for browsing shots, inspecting diagnostic data, and applying consistent labels through a customisable tagging system. The tool supports interactive multi-modality labelling of time series traces, spectrograms, and camera data. Integrated machine learning model support enables data driven workflows to automate the annotation cycle, accelerating the creation of high-quality training datasets for downstream tasks such as disruption precursor analysis and regime classification.
The second tool, “NiceShot!” is an interactive web-based dashboard for exploring tabular shot metadata across entire experimental campaigns. It provides dimensionality reduction projections of the shot space, pairwise scatter plots, sortable data tables, and per-shot time-trace visualisation, all updated live through a flexible filtering system. Additional features include clustering & outlier detection methods, as well as feature attribution plots to support interpretable analysis of data-driven models, helping physicists understand which plasma parameters drive shot-level predictions. The tool connects directly to different backend data access layers, making deployment against existing data infrastructure straightforward.
Together, these tools form a practical pipeline connecting broad exploratory analysis, filtering and quality analysis, through to supporting the human-in-the-loop annotation workflows leading to high quality datasets. We will demonstrate both tools on MAST/MAST-U data and discuss lessons learned, open challenges, and opportunities for wider community contribution.
2. Building digital twin experiences of fusion facilities with OpenUSD
Rui Costa
Making fusion energy commercially viable is a complex, multidisciplinary challenge that depends on teams with very different expertise working through integrated, traceable workflows. In this talk we show how UKAEA is using OpenUSD (Universal Scene Description) as the foundation for custom digital twin experiences and scalable data pipelines: an open framework for composing 3D worlds that can bring together engineering geometry, metadata, real-time operational data and simulation results in a unified, extensible environment. Because OpenUSD is an open standard that can support many applications and deployment routes, it provides flexibility in how these digital twin workflows are built and used; in our case, we use NVIDIA Omniverse for the real-time collaboration, visualization, simulation and interoperability features it provides, as well as our access to OVX hardware. Built on OpenUSD, NVIDIA Omniverse enables interactive, multi-user digital worlds that help teams explore, understand and act on complex information more quickly. We present our approach to integrating user tools and data sources with OpenUSD, and how we leverage Omniverse experiences of our fusion research facilities to combine live data streams, fast simulation predictions and rich 3D context. The result is a composable framework for building actionable digital twin applications that accelerate insight, improve cross-domain communication and support faster, better-informed decision-making.
Participants
2