Short answer
Exploded views and illustrated parts catalogs show how a product breaks into parts, how the parts relate, and which part number identifies each one. They support part identification, assembly-structure learning and repair-task grounding when diagrams are linked to parts lists. They contain no motion, force or tooling detail, and OEM copyright and dealer agreements often restrict reuse.
What these documents are
An exploded view draws a product with its parts separated along assembly axes, usually with dotted lines showing how parts fit together. An illustrated parts catalog (here meaning the manufacturer's parts publication, often abbreviated IPC, not the electronics standards body) pairs such figures with a parts list: figure number, item callout, part number, description, quantity and sometimes the models or serial ranges to which each part applies.
Two kinds of information sit in one document. The picture shows spatial relationships and what each part looks like. The list ties that picture to identity: a callout "12" on the drawing is part number "XYZ-4471" in the table. Together they form a structured map of a product, which is why they interest buyers working on repair, maintenance and assembly tasks.
What the data can support
- Part identification: matching an image of a part to its identity, name and function.
- Assembly structure: a hierarchy of assemblies, subassemblies and parts, and which parts attach to which.
- Grounding for repair language: connecting terms in a work order to named parts and figures; see work order histories.
- Spare-parts workflows: looking up the part to order from a symptom and a model; see spare parts identification.
- Document understanding: figure-to-text linking, callout detection and table extraction.
A published example of structure-from-diagrams work is IKEA-Manual, which pairs 102 IKEA objects with assembly manuals and annotates decomposed assembly parts, assembly plans, manual segmentation and 2D-3D correspondence (Wang et al., 2022). It deals with flat-pack furniture, not industrial products, but it shows how diagrams can be tied to part decomposition and plans. For the overlap with product recognition, see automotive parts catalog data.
Formats and how hard they are to parse
| Format | What you get | Parsing difficulty |
|---|---|---|
| Structured XML publications (e.g., S1000D data modules) | Text, figures and parts data as separate, coded units | Lower: structure is explicit |
| Vector diagrams with embedded text | Callouts and lines as objects | Medium: callout-to-line links must be rebuilt |
| Raster images and PDFs | Pictures only; text may be scanned | Higher: needs OCR and figure detection |
| Online dealer catalog viewers | Interactive figures and lists behind a login | Variable; terms of use often prohibit scraping |
| Parts database exports (CSV, EDI) | Part numbers, descriptions, supersessions | Easy, but no figures |
S1000D is an XML-based specification for technical publications in which data modules are stand-alone units of descriptive, procedural or operational data stored in a Common Source Database and identified by a unique data module code (S1000D). When a manufacturer authors in that structure, the metadata can be far richer than a PDF; ask whether the source files exist, not just the rendered output. See S1000D and ATA documentation for the format in more detail.
Illustrative package: parsed parts catalog
# Illustrative only - fields and values are examples, not an available dataset
package: parts_catalog_pack_v0
source: one equipment manufacturer's service parts publication, 3 product lines
figures:
format: SVG where available, else 300 dpi PNG
per_figure: figure_id, title, applicable_models, revision
callouts:
item_number: 12
bbox_or_anchor_point: pixel coordinates
leader_line_target: part region polygon (on 10% subset)
parts_list:
fields: [figure_id, item_number, part_number, description, quantity, effectivity, supersedes]
hierarchy: assembly -> subassembly -> part (parent_id per row)
linked_assets:
photographs_of_parts: optional, with part_number
cad_or_mesh: not included unless licensed separately
quality:
callout_to_row_match_rate: reported on audited sample
ocr_error_rate: reported for scanned sources
rights: AI training and internal evaluation; figures not redistributed; part numbers and trademarks as supplied
Evaluating a parts catalog
- Join check. Pick random callouts and confirm each maps to the right row in the parts list.
- Revision control. Confirm which product models and serial ranges each figure covers, and the publication date.
- Supersession. Check that replaced part numbers are tracked, or old and new parts will be treated as different objects.
- Figure quality. Inspect resolution, leader-line clarity and whether text is vector or scanned.
- Coverage. Count products, assemblies and figures against the equipment your robot would service.
- Hold-out design. Split evaluation by product line or model family so near-identical figures do not leak across splits.
Rights and provenance
Parts catalogs are authored works owned by the OEM, and distribution to dealers is often under agreements that limit copying, redistribution and use outside sales or service. Online catalog access terms may forbid automated extraction. Third-party aggregators may hold their own database rights and may not be able to license the underlying OEM content.
Ask who authored the figures, whether the supplier is the OEM or a dealer, and what the dealer agreement says. Also ask about trademarks and part-number data embedded in figures. This is not legal advice; the grant needs review. For OEM documentation access routes, see OEM service documentation access.
Limitations: what exploded views and parts catalogs cannot do
A parts diagram is a static picture of a product. It contains no motion, force, tool selection or fastener torque, and it does not teach a robot how to remove or install anything. The exploded direction is not always the disassembly sequence: drawings are laid out for legibility, and the real order may need hidden steps such as releasing a clip or draining a fluid. For sequences, see disassembly and reassembly sequences.
Diagrams are idealized: they omit wear, tolerances, adhesives and how parts look after years of use. Parts catalogs also describe configurations, so a unit modified in the field may not match. They cannot replace procedure content; see service manuals for embodied AI for the procedure side.
Finally, diagrams alone cannot train a humanoid's physical control. They give structure and vocabulary that, combined with video, sensor data or robot demonstrations, ground what a robot sees and plans.
What to buy first
If your goal is part recognition, begin with a single product family where figures, parts lists and photographs of the real parts can all be obtained, and test the pipeline end to end before widening. If your goal is repair grounding, prioritize catalogs for equipment your robot would actually service, linked to the matching service procedures. Ask for a small sample figure with its parts list and check the callout join manually. A clean sample from one product line tells you more than a large archive whose figure-to-list links were never verified.
Questions to ask suppliers
- Are you the OEM, an authorized dealer or an aggregator, and which agreement governs the catalog?
- Do you hold source files (XML, vector) or only rendered PDFs and images?
- How are callouts linked to parts-list rows, and what is the audited match rate?
- Which models, serial ranges and publication revisions does each figure cover?
- Are supersession and substitution records included?
- Can figures and part numbers be used for AI training, and under what restrictions?
- Do figures include trademarks, third-party content or customer-specific parts?
- Is part photography or CAD available, and is it licensed separately?
- How many product lines, figures and parts rows are included?
- How are updates handled when the OEM revises the catalog?
Discuss your requirements
If you can name the equipment families and the part-identification or assembly task you need, discuss your humanoid training-data requirements with SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. A request does not guarantee a match; SourceX follows up about potential sources.
Start from the humanoid training data guide or the maintenance, repair and tool-use data hub. The catalog pages on CAD and engineering drawings and field service maintenance records are related.
Discuss your humanoid training-data requirements with SourceX
Sources
- S1000D International specification for technical publications using a common source database — XML data modules and the Common Source Database.
- IKEA-Manual: Seeing Shape Assembly Step by Step — manuals paired with annotated part decomposition and assembly plans.