A crimp defect found before the harness is assembled costs roughly 10 times less to address than the same defect found during harness-level testing. The ratio is documented across automotive tier-supplier cost-of-quality literature and holds across connector families, wire gauges, and production environments. For a wire harness plant running multiple connector types across three shifts in Malaysia — where a mix of automotive and contract EMS customers demands variants ranging from 2-cavity power connectors to 64-cavity signal harnesses — the practical translation is direct. The cost of one rework event at the harnessing table, re-terminating a 30-cavity connector at 20 to 40 minutes of skilled-operator time plus potential housing replacement, exceeds the cost of a full shift of crimp-station vision inspection.
Manual pull testing, the default crimp verification method at most press stations, samples roughly 0.5 to 1% of crimps produced per shift. The remaining 99% ship on the basis of process capability data from the last tool-qualification run. When crimp-tool wear, wire-lot variation, or ambient humidity shifts the crimp-height distribution between qualification events, pull-test sampling may not detect the drift until a batch outside tolerance has already shipped downstream.
Vision inspection at the crimp press inspects every crimp as it is made and writes each result to a timestamped quality record. The shift in inspection coverage — from 1% to 100% — also changes the character of the quality data produced. Instead of a point-in-time pass or fail on a sampled unit, the record shows a trend line across the shift: where the crimp-height distribution was at hour one, how it moved at the reel change in hour four, and what the press was doing at hour six when ambient temperature climbed in the afternoon.
Where wire harness defects originate
Harness defects cluster across three stations: the crimp press, the insertion board, and the routing table. Each produces distinct failure modes with distinct visual signatures.
Crimp station failure modes
| Defect mode | Visual indicator | Inspection approach |
|---|---|---|
| Crimp height out of spec (too high or too low) | Terminal profile height deviation vs reference | Sub-pixel 2D profile measurement at press exit |
| Partial wire insertion | Strands visible outside or behind crimp barrel | Combined backlight and forward-illumination view |
| Wire gauge mismatch | Barrel fill ratio, visible strand count at entry | Multi-illumination at high magnification |
| Terminal contamination or corrosion | Surface discoloration, foreign material on contact face | Colour and texture classification |
| Insulation damage at barrel edge | Melted trace or cut insulation at barrel entry | Oblique lighting at macro focal length |
| Insulation gap — wire too short | Gap between insulation edge and barrel entry exceeds tolerance | Side-view measurement at press exit |
Insertion board failure modes
| Defect mode | Visual indicator | Inspection approach |
|---|---|---|
| Terminal not fully seated | Rear-face position offset vs housing datum | Back-face or side-entry camera, per cavity |
| Wrong terminal in cavity | Terminal body profile mismatch to cavity template | Model-based profile matching |
| Missing terminal | Unoccupied cavity | Per-cavity presence detection |
| Seal or grommet mispositioned | Seal face proud or recessed beyond tolerance | Side-profile measurement at seal entry |
| Terminal rotated in cavity | Orientation offset from locking feature axis | Rotational position measurement |
Routing and labelling failure modes
| Defect mode | Visual indicator | Inspection approach |
|---|---|---|
| Wire colour routing violation | Colour mismatch at branch point vs harness diagram | Sequential colour verification per branch |
| Cross-routing between circuits | Wire path deviation from reference layout | Multi-camera route comparison at routing board |
| Label absent or incorrect | OCR/OCV failure on circuit-ID or connector-ID label | Character-level OCR at labelling station |
| Missing protective seal on connector housing entry | Seal absent from cavity entry | Presence detection per housing |
Why sampling fails in high-mix harness production
Statistical sampling plans work when the production run is long enough for sampling frequency to carry meaningful coverage. High-mix harness production does not provide that run length. A plant producing 40 harness variants across three shifts may change variants every 20 to 50 units. A sampling plan calibrated on 500-unit runs covers a statistically different population when variant changes happen every few minutes.
Defect probability is not constant within a shift. Crimp-tool wear progresses as a function of cycle count, not elapsed time. A tool within specification at shift start may produce marginal crimps at the 1,500th cycle. Wire-lot variation — reel-to-reel gauge tolerance, insulation diameter variance, surface lubrication — shifts the crimp-height distribution between reels. Pull-test sampling taken at hour one does not represent what the press is producing at hour six when a new reel came in and ambient temperature changed.
A timestamped vision record from every cycle shows when a drift started and at what cycle count it crossed a warning threshold. Quality engineers can trace a crimp-height trend to a specific tool change, a specific reel, or a specific time event. That trace is not recoverable from a pull-test log. Pull tests are point-in-time checks that document a result but not a trajectory.
Crimp height measurement at the press
Crimp height is the primary predictor of crimp connection reliability under vibration and thermal cycling. A terminal crimped 0.1 mm too low over-compresses wire strands past their elastic recovery limit — the strands work-harden at the compression zone and become brittle under repeated flex. Crimped 0.1 mm too high, the terminal does not cold-weld the strands sufficiently, leaving a contact interface with elevated resistance and reduced mechanical retention. Both variants produce crimps that pass static pull tests under controlled conditions and fail in service under automotive vibration profiles (20–200 Hz) or thermal cycling between -40°C and +125°C.
Accurate crimp-height measurement requires a fixed optical geometry. The camera, lens, and illumination source must be mounted with consistent positioning relative to the crimp press anvil, and that geometry must be confirmed after every press maintenance event. A vision system on a flexible arm or adjusted between shifts by press operators introduces measurement error that invalidates inter-shift comparisons. The geometry confirmation — a calibration verification against a reference artefact of known crimp height — is part of the quality record for the production run, not an afterthought recorded on paper.
HyperQ AI Vision deployed at the crimp station measures terminal profile on every press cycle using a fixed-focus lens and structured-light illumination. Each measurement result is written against the terminal cavity identifier and the press cycle counter. When the crimp-height trend moves toward the process lower warning limit — 80% of the tolerance band consumed — the system issues a press-inspection alert before out-of-spec units accumulate. The alert is a schedule event, not an emergency stop: the press operator inspects tooling and clears the alert, or escalates to a tool change before the reject limit is reached.
Connector seating verification
Terminal seating is the inspection mode most often missed by end-of-line electrical testing. A terminal inserted 0.5 mm short of its locked position may still make electrical contact under the spring force of the connector housing, passing continuity and resistance tests at end of line. That terminal fails in service when the connector is mated and de-mated repeatedly — as happens at vehicle service connectors, door modules, and diagnostic port harnesses — or when vibration gradually backs the terminal out of the housing cavity.
Back-face imaging places a camera to view the rear face of the connector housing after each terminal insertion. A seated terminal sits within a defined position window relative to the housing rear datum. An unseated terminal presents at a measurably different depth. The inspection runs in under 100 milliseconds per cavity and does not constrain insertion board throughput at up to approximately 15 insertions per minute.
Sealed connectors with rubber grommets require a side-entry camera to verify seal face position after insertion. A grommet protruding past the housing face creates a compression mismatch when the mating connector is engaged, leaving a leak path that defeats the IP sealing rating. This failure mode is undetectable by pull test and invisible to electrical continuity testing.
Routing verification before taping
After insertion and before taping, harnesses are laid out at a routing board where each branch is positioned according to the current harness drawing. Routing errors at this stage — a colour-coded wire in the wrong branch, a reversed polarity pair, a circuit transposed between two identical-gauge wires — are visually detectable and cost only a re-route to fix. After taping, the same errors require complete harness disassembly.
A multi-camera routing verification station compares the actual wire colour sequence at each branch point against the master harness diagram loaded for the current part number. The comparison covers the full harness layout, not individual connectors in isolation. Crossed-circuit errors created at the routing stage are invisible to crimp-station and insertion-board inspection because those stations check individual terminated ends, not the assembled routing path.
The station switches between harness part numbers by loading the corresponding reference layout from the production scheduling system. The reference update runs automatically on a part-number scan at the routing board; no operator re-programming is required between variants in the same product family.
The MY wire harness production context
Malaysia's wire harness sector spans two supply segments with different quality-consequence profiles.
Automotive harness plants supply OEMs and tier-one integrators with safety-critical circuits: restraint system wiring, powertrain control harnesses, ADAS sensor connections. These applications sit under functional safety requirements that mandate per-unit inspection records and defect traceability to individual harness serial numbers. A field return traced to a wiring defect without a corresponding production inspection record creates liability exposure that extends across the entire lot shipped within the tooling window.
Contract EMS harness plants produce signal and power wiring for industrial equipment, consumer electronics, and medical devices. The defect consequence is typically a functional failure rather than a safety event, but customer incoming inspection requirements and warranty terms often translate similar defect-detection obligations into contract specifications.
Pre-crimp vision inspection serves both segments with the same station architecture. The distinction is in specification and record granularity: automotive harness applications require per-cavity records and tighter crimp-height tolerances; EMS applications can run on batch traceability with wider tolerances on non-safety-critical circuits. The station hardware is the same; the configured inspection recipe differs.
Cross-industry evidence for the architecture
The per-unit inspection architecture — vision checks at each processing station, timestamped results written to the quality record — operates in a Tier-1 automotive parts supplier running over 8,000 part variants at 11,520 units per day across 6 production lines. In that deployment, station-by-station vision inspection replaced end-of-process sampling across the full variant mix. The variant-switching logic that handles part-number transitions in that context is the same logic that manages connector-type changes at a harness insertion board: the incoming part number triggers a recipe load from the production scheduling system, and the inspection parameters update without operator intervention.
The record architecture — per-unit result, station identifier, timestamp, and batch lot number in a single quality record — is the same whether the inspected unit is a pressed automotive component or a terminated connector cavity. The architecture is described at HyperQ AI Vision.
What commissioning a crimp-station vision system actually requires
A crimp-station vision system requires deliberate mechanical and optical choices that determine whether the measurement is reliable across a full shift. Mounting a camera above the press and pointing it at the terminal is where the work starts, not where it ends.
The camera mount must be attached to a rigid structure that does not move with press vibration. A camera bolted to the press frame inherits the press vibration profile — typically 5 to 15 Hz on a hydraulic crimp press — and the resulting image blur degrades sub-pixel measurement accuracy. The mount must attach to a separate structure, typically the floor or a freestanding frame isolated from the press, and the geometry between the lens focal plane and the crimp anvil face must be verified with a reference artefact at the start of each production run.
The illumination geometry determines which defect modes are detectable. Crimp height profile measurement uses structured-light or coaxial illumination that creates a high-contrast edge at the crimp barrel boundary. Insulation damage and contamination detection uses oblique lighting at 15 to 30 degrees from the surface plane to create shadow contrast at surface anomalies. A single lighting setup that tries to serve both requirements simultaneously is generally a compromise that performs poorly on both. Stations designed with multiple lighting zones — switched in sequence or captured as separate image channels — detect a wider defect mode range at the cost of a longer cycle time per terminal.
Recipe qualification for each connector family requires sample crimps across the expected specification range: several units at the nominal crimp height, several at the lower warning limit, and several at the reject threshold. The model's detection boundary is set against those samples, not against a theoretical specification. A recipe qualified only against nominal crimps will produce erratic results when process drift moves the distribution toward the warning limit, because the model has not seen what a marginal crimp looks like on that specific terminal geometry under that specific illumination.
These setup requirements are the conditions that make the measurement meaningful. A crimp-station vision system commissioned without this rigour produces data that cannot be trusted for drift detection or recall scoping.
What vision inspection does not replace
Pre-crimp and insertion-board vision does not replace end-of-line electrical testing. Vision confirms crimp geometry and terminal seating before the harness is assembled. It does not measure circuit continuity across a completed multi-point wiring path, detect pinched wires under harness tape, or identify contact-resistance problems on a mated connector under rated load.
The production architecture that closes the defect window requires both layers. Vision covers the modes electrical testing cannot reach: individual crimp geometry, pre-mated terminal seating, routing branch errors before taping. Electrical test covers the modes vision cannot reach: completed circuit continuity, cross-circuit shorts created during assembly, insulation breach from routing clip pressure, and contact resistance on mated connectors.
Running crimp-station vision without end-of-line electrical test leaves a real defect window open on safety-critical harnesses. Running end-of-line electrical test without crimp-station vision ships marginal crimps that passed electrical detection because the crimp geometry was borderline rather than open-circuit. Both layers are required. Vision inspection that is not integrated into the quality record — if results are only displayed on a station screen without being written to a database — provides detection without traceability, and cannot support a recall scope determination when a field return arrives.
Building the quality record from station data
The value of per-unit crimp-station inspection extends beyond the detection event. The quality record built from station data answers three questions that end-of-line sampling cannot: which production window produced the defect, which process variables correlated with the shift, and how many units fall within the suspect window. The data pipeline from station inspection events to batch records in an eQMS — including the record schema and the linkage to lot numbers and serial identifiers — is described in a separate post on vision data pipelines into eQMS and MES.
When a field return arrives with a crimp failure, the trace runs from the harness serial number to the station record, to the press cycle range, to the tooling log, to the wire reel lot. That trace bounds the recall scope to the affected window rather than the entire shipment lot. On a plant producing 5,000 harnesses per shift, the difference between a shift-level and a 200-unit window-level scope represents a substantial exposure difference.
Station records also drive process improvement. A crimp-height trend chart plotted against press cycle count reveals when tooling wear is accelerating beyond the normal rate. That information, used proactively, converts an unplanned press stoppage into a scheduled tool-change event. The cost of the downtime is predictable; the cost of a batch-level reject is not.
Send crimp samples from your worst-performing connector type — five units that passed incoming inspection and five that failed or returned from the field. We will run a crimp-height and seating feasibility assessment within 2 weeks, identify which failure modes the station camera can close, and specify the measurement resolution and lighting configuration required for your connector geometry. No contract is required until the detection specification has been agreed and demonstrated on your parts: Book the crimp feasibility assessment.