Short answer
Cable harness inspection data is mostly photographs of crimps, terminals, connectors, routing and labels, paired with pass or fail decisions from human inspectors. IPC/WHMA-A-620 sets acceptability criteria and three product classes, so labels need the class attached. Harness work is largely manual, so image archives are often thin and inconsistent, and they show assemblies, not the handling that built them.
A manual product with codified acceptance
Wire harnesses are assembled largely by hand: cutting, stripping, crimping, inserting terminals into connector housings, routing, taping and labeling. Automation covers cutting and crimping; the rest is dexterous, variable work on flexible parts. Inspection is therefore also human-heavy, with continuity testing for electrical function and visual checks for workmanship.
That structure shapes the data. IPC/WHMA-A-620 covers materials, methods, tests and acceptability criteria for crimped, mechanically secured and soldered interconnections and related assembly of cable, wire and harness assemblies. The IPC store lists Revision F as current, it relies on process control methodology, and it has three product classes, with the user responsible for selecting the class (IPC/WHMA, 2025). The standard is paid, so suppliers who use it can tell you which clauses they apply; you rarely get the clauses themselves. This page sits in the inspection cluster of the humanoid data section.
What harness inspection data can and cannot support
It can support models that check crimp appearance, terminal presence and seating, connector housing integrity, wire color order, label legibility and routing against a reference. It supports assembly completeness verification in particular: is every cavity populated, is every clip present, is the boot seated.
It cannot support electrical judgment. A crimp that looks acceptable can fail a pull test, and an acceptable-looking joint can hide a cold connection. Cross-sections, pull-force tests and continuity results are separate data. Check separately whether the supplier's criteria for inspection frequency, cross-sections or X-ray come from the standard or from company rules; do not assume the standard supplies them.
Harness data does not show how to handle flexible wire. Insertion of terminals, routing slack and tape wrapping are contact-rich manipulation tasks on deformable objects. Images of finished harnesses tell a robot what the result should look like and nothing about the motions or forces needed. Footage of workers doing these steps is a different asset, discussed in human procedural mistake videos.
Defect families worth naming in a request
| Family | Examples | Why it is hard to see |
|---|---|---|
| Crimp | Strands outside barrel, insulation in crimp, over- or under-crimp, short stripped length | Small features, shadow, strand reflectance |
| Terminal | Bent, damaged, wrong terminal, not fully seated | Pose-dependent, occluded by housing |
| Connector | Cracked housing, missing seal, unpopulated cavity, wrong keying | Dark plastic, deep cavities |
| Wire | Nicked insulation, wrong gauge or color, kinks | Color accuracy, lighting |
| Routing and securing | Tension, bend radius, clamp or tape placement | Needs board or fixture context |
| Marking | Wrong, missing or illegible label | Requires text reading, glare |
The families are Illustrative groupings for specification; the actual list depends on your harness types and the class you select.
Illustrative package
Illustrative; values are synthetic.
| Element | Example specification |
|---|---|
| Harness identity | Family H-30, revision B, variant 04 |
| Class basis | Product class stated per harness; revision of the standard stated |
| Views | Terminal close-up (per crimp), connector face (per cavity), full-length routing view |
| Imaging | Fixed macro station, diffuse light, color reference card in frame |
| Labels | Defect family, location (circuit number), severity, accept or reject |
| Test link | Continuity and pull-test result for the same unit |
| Cross-section link | Optional cross-section image for a subset |
| Splits | By harness family and by shift, not by image within one unit |
Per-crimp images matter because defects live at the crimp; a full-harness photo cannot resolve a strand outside the barrel. See multi-view inspection imagery for view planning.
Why archives are often thin
Harness lines seldom have inline vision on every step. Where cameras exist, they sit at crimp presses (crimp-height or force monitoring plus optional vision) or at a final check station. Many plants rely on human visual inspection with no image capture, and where images exist they come from phones or QA photos taken for non-conformances. Those photos are valuable for defect examples but cover rare cases, not the passing majority. A common pattern is plenty of defect narratives, few normal images, no consistent angle. Where the supplier also makes boards, the defect image dataset guide explains how to combine sources.
That points to commissioned capture, in which a supplier photographs a planned set of good and defective assemblies at a fixed station. It depends on site access, production disruption and what the supplier agrees to; treat it as a scoping conversation. Existing nonconformance records can guide the defect list; see nonconformance reports.
Preparation before use
Group images by unit so a harness's crimps are never split across training and evaluation. Normalize file naming so the circuit number and view are machine-readable. Record the lens and working distance, because crimp features are small and scale matters when a threshold is expressed in millimeters. Where a color reference card is in frame, keep it unmasked; cropping it out removes the only way to check color drift.
Finally, map the supplier's reject reasons to your defect families and keep both. Plant vocabularies ("bad crimp", "strands out") rarely match a clean taxonomy, and the original wording often carries information a mapped code loses.
Quality and acceptance criteria
- Class and revision recorded so labels have a stable meaning; see acceptance standards as label definitions.
- Defect location tied to a circuit or cavity number, so a label can be re-checked.
- Balanced views: at least one image per crimp, not only a summary shot.
- Color fidelity: wire color is a label in its own right; require a color reference or calibrated capture.
- Annotator expertise: harness inspectors, not generalist labelers, should adjudicate borderline crimps.
- Held-out harness families for evaluation, so a model cannot recognize a harness instead of a defect.
Rights and sensitivity
Harness designs are customer-owned in many contract settings, and pin-outs, part numbers and wire lists can reveal product architecture. Images of connectors show markings and keying that identify the customer's component choices. Ask who owns the design, whether the customer must approve licensing, and whether the images can be limited to the crimp or terminal area. Cropping reduces exposure but does not settle ownership. See inspection images linked to outcomes for linking decisions responsibly. This is not legal advice.
For electronics boards inside the same product, see PCB and solder joint inspection data, which has a different standard and image type.
Limitations and extra preparation
Harness inspection is often done by eye, so archives tend to be thin: images exist mainly where a dispute, a customer complaint or a training exercise prompted someone to take a photo. Those images are biased toward unusual defects and away from routine passes. Crimp cross-sections, pull-test results and continuity tests capture properties a surface image cannot show, and they are rarely stored alongside the photos. Wire colors and connector housings vary by supplier, so an archive from one customer program may not cover the variants your robot will see.
The data supports inspection and recognition; it does not show how to route a wire, seat a connector or apply the insertion force that a robot would need, which requires robot, teleoperation or instrumented-rig data. Plan for extra preparation: naming the product class for each label, linking images to test results where they exist, removing customer drawings and part numbers, and grouping images by assembly so that one unit's crimps do not appear in both training and evaluation.
Questions to ask suppliers
- Which harness families and connector types are represented, and in what volumes?
- Which product class and revision of IPC/WHMA-A-620 governs acceptance?
- Do images exist at crimp level, or only at whole-harness level?
- How were defects found: inline vision, human inspection, test failure or customer return?
- Are continuity and pull-test results linked to the unit?
- Is there a sample of passing units, imaged the same way as defective ones?
- Who owns the harness designs, and do customer agreements allow licensing images?
- Can the supplier produce a planned capture of specified defects under fixed lighting?
Discuss your requirements
If you need cable harness or connector inspection data from US manufacturers, describe the harness types, defect families and permitted uses to SourceX. SourceX reviews the requirement and follows up about potential sources, and checks the data and the supplier's licensing permissions. Whether a commissioned capture is feasible is a case-by-case discussion, and a request does not guarantee a match.
Discuss your humanoid training-data requirements with SourceX
Sources
- IPC/WHMA-A-620F Requirements and Acceptance for Cable and Wire Harness Assemblies — IPC/WHMA-A-620 Revision F scope, acceptability criteria for crimped, mechanically secured and soldered interconnections, three product classes chosen by the user.