Anomaly DetectionWeld InspectionSurface InspectionVideo QC/QAAI Assistant
HOME/PRODUCTS/LINE WELD
LINE WELD · Weld Vision Inspection

Inspect every weld —
more precisely than the human eye

LINE WELD fuses weld imagery with process signals (current, voltage, power) to classify porosity, undercut, spatter and cracks in real time. Every verdict, coordinate and severity grade is recorded against the cell/part ID — inspection becomes quality data.

98.6%
defect detection rate
0
escaped defects (6 months)
<50ms
verdict per weld
INPUT

Synchronized image + signal

Bead imagery and current/power waveforms aligned to the weld trigger

FUSION

Image × signal fusion

Surface defects and internal-quality signatures judged together

CLASSIFY

Classify & locate

Defect type + pixel coordinates + bead geometry

RECORD

Verdict & traceability

Per-ID verdicts and evidence images pushed to MES

/01 How it works

Image alone misses things. Signal alone misses things.

A clean surface can hide internal porosity; a normal waveform can coexist with spatter. LINE WELD fuses both sources in a single verdict, lifting detection rate and confidence together.

01 · SYNC

Synchronized capture

Bead images and current/voltage/power waveforms collected per weld, aligned to the trigger signal

02 · FUSE

Fusion analysis

A verdict model combining image features (shape, surface) with signal features (energy, stability)

03 · LOCATE

Classify & measure

Porosity/undercut/spatter/crack classification, pixel-level localization, bead width/height measurement

04 · TRACE

History & interlock

Per-weld verdict with evidence image stored; NG triggers PLC interlock and MES quality records

* Evidence images are stored with every verdict — ready-made material for customer quality audits and claim responses.

FUSION

Image × signal verdicts

Bead appearance and weld-energy patterns are judged together. Images catch surface defects; signals reveal signatures of internal quality issues like lack of fusion.

GEOMETRY

Quantified bead geometry

Beyond pass/fail: bead width, height and continuity are measured numerically. Drifting geometry warns the process before defects appear.

HYBRID

Normal-learning + few labels

Anomaly detection trained on normal welds, topped with a type classifier built from the few defect samples you have — minimal data burden.

ADD-ON

No robot or fixture rework

Existing robots and fixtures stay untouched; only cameras and signal taps are added. Zero impact on cycle time.

/02 Use Cases

Use Case Analysis

Laser, arc or spot — the fusion-inspection principle stays the same.

Battery pack · Laser welding

Zero escaped defects for six months

Challenge

Busbar laser welds were 100% visually inspected, yet micro-porosity and undercut escaped to downstream 2–3 times a month — a direct fire/recall risk for batteries.

Approach

Coaxial camera imagery fused with laser-power signals; every weld verdict mapped to the cell ID for full traceability.

Result

Zero escapes in six months. Inspectors moved to verdict auditing and process improvement; weld history became audit-ready evidence.

Automotive body · Spot welding

From sampling to 100% of thousands of spots

Challenge

Thousands of spot welds per body were managed only by destructive sampling, leaving quality gaps between lots.

Approach

Normal-pattern learning on weld current/resistance waveforms enabled per-spot real-time verdicts with instant location flags.

Result

Full-population monitoring reduced destructive testing, and electrode-wear quality drift is now managed proactively.

Heavy industry · Arc welding

Reliable verdicts in fume and arc glare

Challenge

Thick-plate arc welding defeated conventional vision systems due to fume and intense arc light.

Approach

Process-specific optics and filtering, with arc signals as a supporting verdict axis in a fusion configuration.

Result

Verdict reliability held under fume/glare, and welder-to-welder quality variation became visible in data.

* Figures and scenarios reflect representative deployments and may vary by process and data conditions.

/03 Specifications

Specs & Integration

Weld types
Laser · Arc (MIG/TIG) · Spot — optics and verdict models per process
Cameras
Area/line-scan, coaxial or side-mounted; process-specific lighting and filter design included
Signal inputs
Weld current · voltage · laser power · trigger (taps on existing power source/controller)
Detected items
Porosity · undercut · spatter · crack · bead geometry (width/height/continuity) — 40+ types
Verdict speed
Under 50ms per weld/shot; add-on with no takt impact
Deployment
Edge GPU unit · on-premise server
Integrations
PLC NG interlock · reject signal · MES quality history · evidence-image archive
/04 FAQ

Frequently Asked Questions

Can it detect internal porosity?
Open surface porosity is detected directly from imagery. Internal porosity is judged from anomalous weld-energy signatures; for joints requiring the highest assurance we propose a combined setup with NDT such as ultrasonic testing.
Do we need to modify robots or fixtures?
No. The system is an add-on of camera brackets and signal taps — no impact on cycle time or existing control logic.
Can verdicts follow our weld specification?
Yes. Your weld spec (allowable defect size and grading) is mapped directly to verdict thresholds, and pass/fail criteria are calibrated with your quality team during the PoC.
We only have a handful of defect samples.
Normal weld data is enough to start, since the base model learns normal welds. Your few defect samples are used for type classification and validation.
Get Started

See the detection rate on your own weld specification

Tell us about your process and data environment — we will come back with a feasibility summary and expected impact.