MFFD digital-thread tour
MFFD digital-thread tour
A 10-minute walk through the Multi-Functional Fuselage Demonstrator data in shepard.
The MFFD (Multi-Functional Fuselage Demonstrator, JEC World Innovation Award 2025) upper shell was manufactured at DLR ZLP Augsburg in a chain of robotic process steps — AFP tapelaying, stringer/bridge/spot welding, NDT thermography. Every stop below is real process data from that campaign, imported into shepard: ~13 500 DataObjects, hundreds of sensor channels, welding videos, thermography frames, and the robot cell’s kinematic model.
Prereq: sign in at https://shepard.nuclide.systems (any demo account,
e.g. bob). Every link below is clickable once signed in.
Stop 1 — The gallery
https://shepard.nuclide.systems/collections
All MFFD process-step collections at a glance — names, descriptions, and per-collection DataObject counts (8483 AFP tracks, 1031 bridge-welding executions, 744 thermography frames, …). This is the “what’s in the box” view a new colleague sees first.
Stop 2 — AFP tapelaying at scale
https://shepard.nuclide.systems/collections/019ed455-66f4-7aea-8cb3-5c0b34a737df
The FAIR landing page of the biggest collection: 8483 track DataObjects from the real cube3 export, with a citation card (plain text / BibTeX / RIS / CSL), a DMP download, RO-Crate export, a Regulatory Evidence Pack, and an honest metadata-completeness score. The sidebar tree lazily loads the full daily-log hierarchy without breaking a sweat.
Stop 3 — One track, full provenance
Track 9 (Run 29995): a single AFP layup run as a citable dataset — semantic annotations linking it back to its cube3 origin, attributes, its timeseries reference, and the provenance activity trail. Everything you see here is addressable by one stable UUID.
Stop 4 — The track’s sensor data
The TimeseriesReference scopes the shared container down to this track’s ~190 channels and its exact time window. Pick channels, plot them, and jump into the view recipes (Trace3D below) — all pre-scoped, no manual time-window typing.
Stop 5 — The channel inventory
https://shepard.nuclide.systems/containers/timeseries/019ede2a-60ec-7ac1-899d-3fe4c6263cbb
The AFP tapelaying TimeseriesContainer itself: 192 distinct channel identities — robot TCP positions, tape temperatures, consolidation forces, motor temperatures, cut counters. This is the raw signal richness a single manufacturing cell produces.
Stop 6 — Trace3D: the thermal trail
The robot TCP path (X/Y/Z) rendered in 3D and coloured by tape temperature — 30 seconds of layup as a heat trail in space. A view recipe binds channels to axes; no plotting code, no CSV export, no notebook.
Stop 7 — Trace3D: consolidation force
Same idea, different question: the TCP path coloured by tape consolidation force during a Track-9 layup window. The force applied to each ply, visible along the actual path it was applied on — the kind of view a process engineer sketches on a whiteboard, live from the recorded data.
Stop 8 — NDT thermography heatmap
An OTvis thermography frame from the NDT step, streamed straight from the uploaded instrument file and rendered as an interactive amplitude/phase heatmap. The file is the reference — no conversion step, no exported PNG.
Stop 9 — Spot welding, machine-readable
A spot-welding scope project (.svdx). shepard’s file-format plugin parsed the
instrument file on upload and materialised 213 urn:shepard:svdx:* semantic
annotations on the file reference — welding parameters as queryable facts,
not bytes in a blob.
Stop 10 — The welding video
Stringer welding on camera: a 4K HEVC upload (1.04 GB) that plays in the browser because shepard transcoded an h.264 proxy on the side. Scrub the timeline — the metal surface and weld head are the same seconds the sensor channels recorded.
Stop 11 — The robot, animated
https://shepard.nuclide.systems/scene-graphs/play/019f16d8-2fc7-76e8-b066-f713e2fb713d
The KUKA KR210 R2700/2 from the MFFD AFP cell as a live kinematic model — the URDF file reference plus six joint telemetry channels, bound by a mapping-recipe template. Press play and the KUKA-orange arm re-runs the recorded trajectory. (Index of all recipes: https://shepard.nuclide.systems/scene-graphs.)
Stop 12 — The vocabularies underneath
https://shepard.nuclide.systems/semantic/vocabularies
The controlled vocabularies powering every annotation you saw: CHAMEO, DataCite, Dublin Core, PROV-O, Material OWL, metadata4ing, schema.org, SKOS and more — seeded into the internal semantic store so annotation keys are terms, not typos.
Stop 13 — Find it again
https://shepard.nuclide.systems/search?q=tapelaying
Search across the instance brings you back to any of it — the entry point your colleague uses next week when they only remember the word “tapelaying”.
Where to go deeper
- In-app help: https://shepard.nuclide.systems/help — task pages for uploading, annotating, plotting, SPARQL, templates, and more.
- Timeseries plotting, Render a view, Query with SPARQL, Upload and annotate video, Importing from DLR cube3
- Reference docs: the per-primitive pages under Reference cover every payload kind and endpoint used on this tour.