Production of a FAIR Tokamak Equilibria Database for Analysis and Machine Learning

To advance data-intensive scientific discovery in the field of tokamak research, the EFIT-AI project has assembled and curated a database of equilibrium reconstructions for a variety of plasmas. The database features more than 7,000 DIII-D discharges (2018-22) and the full history of NSTX(-U) experiments (13,000 discharges). The majority of reconstructions are provided with high resolution and all available diagnostics directly fit by EFIT, but subsets with lower resolutions and more or less constraints are also available to examine sensitivity. This brings the total collection of equilibria to more than 6 million. The data is organized according to the ITER IMAS data schema using the open source implementation in OMAS [https://gafusion.github.io/omas]. The latest version of EFIT can read and write directly in this format, with validation checking, for efficient database generation. Changes to the database are tracked using Git and Data Version Control (DVC). A supplemental collection of python tools provide Findable, Accessible, Interoperable, and Reusable (FAIR) data principles. OMAS supplies an interface for ITER supported tools, mappings to other workflows are available through the OMFIT framework [https://omfit.io], and additional scripts offer data extraction for different use cases.