Answer first: AQL releases a lot, not a process
Attribute sampling on finished humanoid robot wire harnesses — built on ISO 2859-1 — is a valid release mechanism for a narrow set of characteristics: continuity and point-to-point wiring against the wire list, connector seating and latch engagement, keying, dimensions and bend radius, and label/revision marking. All of these are non-destructive, verifiable at end of line, and behave like a random draw from a homogeneous population.
It is not a valid release mechanism for crimp integrity, shield termination, or torsion life at the joint zone. Those characteristics are governed by press setup, applicator wear, tooling and operator, so the defect population is driven by the process, not by the shipment. A lot can pass a normal inspection at any AQL value and still fail in the field, because the sample never touched the interface that fails.
The practical consequence for a robotics OEM moving from pilot builds into ramp: agree four batch-release gates at PO stage — first article inspection, process capability on drifting characteristics, locked critical process parameters, and a signed release record — rather than discovering at goods-in that the supplier has data for none of them.
Which harness characteristics can carry an AQL, and which cannot
The table below is the core of this page. The middle column answers a single question: can this characteristic be released by sampling the finished harness?
| Characteristic | Releasable by finished-goods AQL sampling? | Required release evidence instead |
|---|---|---|
| Continuity, point-to-point wiring vs. wire list | Yes | Electrical test results per the agreed plan, with the test program revision identified |
| Connector seating, latch engagement, keying | Yes | Outgoing inspection record with the check method defined (visual or force gauge) |
| Dimensions, bend radius, branch lengths | Yes | Dimensional results against the current drawing revision |
| Label, revision marking, traceability code | Yes | Visual record plus legibility of the lot code |
| Crimp height | Partial | Gauge reading on the sample, but the acceptance decision must come from the cross-section and pull-test record produced during that run — not from the sampled part |
| Crimp cross-section and pull force | No | Destructive verification per terminal/wire combination, at a defined frequency per setup, with specification window stated |
| Shield termination, bond/transfer impedance | No | Instrumented check per lot or per parent run; a visual sample cannot prove continuity of a shield bond |
| Torsion / flex life at the joint zone | No | Design-validation evidence, not lot release — this is a design property, not a batch property |
| Wire, terminal and tubing identity | No | Material records, UL/wire-recognition files and incoming lot traceability |
Write the sampling system into the PO, not into a conversation
Do not accept "we inspect to AQL" as a statement. The plan must name: the standard family (ISO 2859-1, inspection by attributes), the inspection level, the AQL value per characteristic, the lot definition, and the switching rules between normal, tightened and reduced inspection. If the drawing or PO is silent on any of these, the supplier's interpretation will differ from yours at exactly the moment it matters.
The lot-definition trap in ramp production
Sampling statistics only mean something if the lot is one homogeneous population. In ramp production it often is not: a lot may span two terminal reels, two press setups, a mid-batch reel change, or the transition between a day and a night shift. When that happens, the sample no longer represents the shipment. Require the lot definition to match one material and one setup combination, and require that a reel or applicator change closes the lot.
Destructive tests sit outside the AQL plan
Cross-section and pull tests consume the part. They belong to a separate process-verification frequency agreed at PO stage — typically tied to setup, shift, or a defined interval — and their records are batch-release evidence in their own right, not a supplement to the sample. Mixing the two is the single most common documentation weakness in harness batch release.
Batch-release gate 1 — first article inspection that actually closes the design
An FAI for a humanoid harness must contain, at minimum:
- Full dimensional report against the drawing, with the revision identified
- Crimp cross-section micrographs for every terminal/wire combination, assessed against the acceptance criteria of IPC/WHMA-A-620 with the revision stated
- Pull-test values per wire gauge, with the specification window and the test method shown
- Connector, keying and latch verification
- Shield-termination method and measurement result
- Electrical test results against the wire list
The FAI must be tied to a specific revision of drawing, BOM, wire list and process sheet. An FAI against "the prototype drawing" is not a production release — it is a record of a build that no longer exists.
Keep the boundary with design validation clear: FAI confirms the process reproduces the drawing. It does not re-prove torsion life or flex endurance at the joint — those are design-stage evidence and should be referenced, not re-asserted in the FAI.
Buyer action: sign the FAI as a controlled document and reference it in the PO, so that any later deviation is measurable against a fixed baseline rather than argued from memory.
Batch-release gate 2 — process capability on the characteristics that actually drift
The drifting characteristics on a humanoid harness are crimp height, pull force, seal or boot compression, and shield-bond resistance. All four are tool-wear and setup driven, which is why a single capability study performed at FAI cannot carry you through ramp.
Ask for the method rather than a headline number: which characteristic, how many subgroups, over what time window, with which specification limits, and with the study period stated. A Cpk or ppm target is a commercial agreement per characteristic, not a universal constant — the OEM and supplier should set it together, characteristic by characteristic, because the cost of capability differs across a signal crimp and a shielded power splice.
Capability resets whenever the population resets: press setup change, applicator change, tool sharpening, or a shift change with a different operator. Require the study to be repeated after any of these events.
If the supplier has no capability study at all, do not accept a verbal assurance. Ask for consecutive-lot crimp-height and pull-test records over a defined window and read the trend yourself before releasing volume. That is a slower route to the same decision, and it is auditable.
Batch-release gate 3 — critical process parameters locked and logged
Each lot should be traceable to a parameter record containing:
- Press and applicator ID, with the tooling maintenance counter
- Crimp height setpoint and tolerance window
- Strip length and conductor-damage check
- Soldering or ultrasonic welding parameters where used
- Torque values for shield clamps and strain relief
- Operator identification and qualification status
These parameters belong in a control plan that names the reaction plan when a parameter drifts out of window. "Rework and re-inspect" is not a reaction plan unless the rework method, the re-verification step, and the authority to scrap are all defined.
Ramp-specific risk: prototype lines run on bench tools; ramp lines run on automated presses. A tooling transfer is a new process, not the same process faster, and it deserves a fresh FAI. Suppliers rarely volunteer this — ask directly whether the line that made the qualification samples is the line that will make the production lots.
Traceability expectation: the lot code on the harness label must link back to the parameter log, so a field return can be narrowed to a setup window rather than to "sometime last quarter."
Batch-release gate 4 — the release record the buyer signs against
Per lot, request a defined record set:
- Certificate of conformance referencing the exact PO revision
- Outgoing inspection data — characteristics, sample size per the agreed plan, results
- Destructive-test results for that lot or its parent run
- Parameter log extract for the lot
- Deviation/waiver list with disposition and buyer approval reference
A CoC that states "conforms to specification" with no attached data does not support a batch-release decision. Require the data sheet, even if it is one page. And make the escalation rule explicit: any open deviation must appear on the release record with its disposition — nothing ships on a verbal agreement, because verbal agreements do not survive a personnel change on either side.
PPAP-style document pack: what to request and what each item is for
A full automotive-style PPAP submission is disproportionate for a low-volume pilot lot. Agree a phased pack and write the phasing into the PO.
| Document | What it is actually for | Phase |
|---|---|---|
| Process flow diagram | Shows every handling step where a conductor can be damaged | Pilot |
| PFMEA | Must name crimp-interface and shield-termination failure modes with detection and prevention controls that match the control plan | Pilot (living document) |
| Control plan | Locks the parameters, frequencies and reaction plans from gate 3 | Pilot |
| MSA / GR&R | Proves the gauges used on controlled characteristics can distinguish good from bad | Pilot, before capability claims |
| Crimp cross-section and pull-test data | The actual integrity evidence for the crimp interface | Pilot |
| Material and UL/wire-recognition files | Confirms the wire and tubing are what the drawing says | Pilot |
| Joint-zone routing, clamping and strain-relief content in the drawing package | Humanoid-specific: the harness is a motion component, not just an electrical one | Pilot |
| Full submission with capability studies and re-qualification plan | Statistical basis for volume release | Ramp |
The distinction between "we have a PFMEA" and a usable PFMEA is whether it is reviewed when a new terminal or applicator is introduced. Ask for the revision history — it tells you more than the document itself.
Change control: the PCN triggers that must reopen the gate
Any of the following must force a product change notification and a re-qualification decision:
- Wire or insulation supplier change, including a plating or strand-count change
- Terminal or connector part change — including the connector maker's own revision
- Crimp applicator, press, or tool-sharpening regime change
- Production site or line transfer, or the addition of parallel tooling
- Process parameter window change on a controlled characteristic
- Drawing or routing revision that alters joint-zone geometry
The PCN itself should state what changes, why, which part numbers and lots are affected, the form-fit-function assessment, the re-test plan, the first affected lot, and a response deadline before the change ships.
The quiet-change risk in ramp is real: suppliers under delivery pressure merge a second reel or a second qualified press into the schedule. Two presses are two processes. Require any parallel tooling to be declared and FAI'd, and put that sentence in the PO.
There is a customer-side obligation too. When the OEM revises a drawing or a routing that changes joint geometry, the torsion assumption behind the harness design is invalidated — raise the notification yourself, in writing.
2026 context — what the public data does and does not tell you
One sourced context line, with its limits stated: according to IFR figures cited by EVST, the global installed base of industrial robots passed 4 million units in 2023 and continues to grow at roughly 7% annually, with humanoids representing a fraction of that base (EVST, April 2026). That figure describes industrial robots, not humanoids, and it is a secondary citation rather than the IFR publication itself. It is useful for sizing the supplier base. It says nothing about harness reliability.
The evidence gap is worth naming plainly: there is no public, traceable dataset of humanoid harness field-failure rates or torsion-life distributions that would let anyone set an AQL value on evidence. Any ppm or life target must come from the OEM's own pilot data and be agreed contractually, characteristic by characteristic. When field data is thin, process records substitute for statistics — which is precisely why gates 1–4 carry more weight than a low AQL number on the drawing.
Buyer's acceptance checklist before you release the first ramp lot
- FAI signed against the current drawing, BOM and process sheet revision
- Crimp cross-section and pull-test data available per terminal/wire combination, with the IPC/WHMA-A-620 revision stated
- Control plan with named reaction plans, not generic rework language
- MSA/GR&R on every gauge used for a controlled characteristic
- Parameter logs retrievable by lot code
- Deviation list closed or formally dispositioned before shipment
- PCN flow agreed in writing, including parallel tooling and reel changes
Three questions expose a weak system faster than any audit form:
- "Show me the cross-section for this lot's terminal, not the qualification sample."
- "Which press made this lot, and when was it last serviced?"
- "If I order 30% more next month, what changes in your process?"
Once these records exist, the negotiation shifts to cost and second-source exposure — which is a different conversation with different evidence requirements.
Questions buyers ask at this stage
Can AQL sampling on finished harnesses replace crimp cross-section inspection?
No. Cross-section and pull tests are destructive and setup-driven; they verify the process that produced the lot. AQL sampling verifies the shipment. They answer different questions and neither substitutes for the other.
What inspection level and lot definition should be written into a harness PO?
Name ISO 2859-1 as the standard family, then specify inspection level, AQL per characteristic, the lot definition and the switching rules. Define the lot as one material plus one setup combination, so a reel or applicator change closes the lot.
What documents should a supplier provide at each batch release?
A CoC tied to the PO revision, outgoing inspection data with sample sizes, destructive-test results for the lot or parent run, a parameter log extract, and a deviation list with dispositions.
Which changes require a new PCN and re-qualification during ramp?
Wire or insulation source, terminal or connector revision, applicator or press change, tool sharpening regime, site or line transfer, added parallel tooling, and any drawing revision affecting joint geometry.
Does a first article inspection cover torsion or flex life at the robot joint?
No. FAI proves the process reproduces the drawing. Torsion and flex life are design-validation properties and must be evidenced separately before the design is frozen.
Next step
Request a batch-release document package (control plan, crimp cross-section and pull-test records, change-control flow) plus sample harnesses for first-article validation, and compare what comes back against the checklist above before you sign the first ramp lot off. The package is the fastest way to see whether a supplier's release architecture exists or is being assembled for your visit.