DO-200B aeronautical data versus ML training sets
Informational only, not professional or legal advice. Facts are linked to the source we checked; statuses can change after the date above.
High-intent query: is DO-200B the standard for machine-learning training data? The cited FAA pages answer no. A program that feeds aeronautical data into an airborne ML function may still need both a DO-200B data-process argument and an EASA-style learning-assurance argument. Those are different packages.
Public starting points: the DO-200B explainer, data assurance for ML avionics, and the EASA Level 1/2 kit.
What AC 20-153B means by aeronautical data
FAA AC 20-153B describes an acceptable means, but not the only means, for showing compliance for equipment with an installed aeronautical database. It defines aeronautical data as data used for aeronautical applications such as navigation, flight planning, flight simulators, terrain awareness, and other purposes. It is not airborne system configuration files or lookup tables used to adapt equipment to the aircraft.
Data-quality characteristics named in the AC include accuracy, resolution, assurance level, timeliness, completeness, traceability and format. An LOA confirms compliance with RTCA/DO-200B.
The B revision adapted RTCA/DO-330 to provide better structure and consistency for database production tool qualification. That is tool qualification for database production, not model training. See DO-330.
What EASA asks of ML data sets
EASA bases learning assurance on three separate and independent data sets for training, validation and testing, and treats a complete operational design domain as a prerequisite to an adequate level of quality of the data sets. Those sentences are about trained models, not navigation databases.
What the FAA roadmap does not do
The FAA AI Safety Assurance Roadmap discusses applying existing standards. We have not found a public FAA sentence that says a training set is an “aeronautical database” under AC 20-153B.
Listings
MOA Software REIN and MathWorks AI Verification appear in this directory’s airborne-ML articles because their cited pages name EASA or airborne deep-learning examples. They are not DO-200B LOA products. The airborne AI certification kit collects vendor questions from the same FAA and EASA pages.
What this is not
DO-200B is not DO-178C software assurance and not an AI-model certificate. After AC 20-153B, RTCA/DO-200A is no longer an acceptable means of compliance for new TSO and installation approvals, except where it is already part of the certification basis.