Machine Vision Quality Inspection (MVQI) is the AI-enabled post-weld QA layer that automates weld bead profile measurement, surface defect detection (porosity, undercut, spatter, crack, off-center), and pass/fail decision-making on completed weld beads — replacing manual visual inspection subject to inspector variability + fatigue + throughput bottleneck. MVQI systems combine 3D laser profilometry (Keyence, Kreon, Micro-Epsilon) + high-resolution 2D CMOS camera + AI/ML classification models (typically Convolutional Neural Networks, CNN) to inspect each weld pass at 100% coverage in 5-30 seconds per weld — versus 5-15 minutes manual VT per weld — delivering 10-30× inspection throughput improvement + consistent pass/fail per AWS D1.1 Table 6.1 + ASME VIII Div 1 UW-51 / UW-52 visual acceptance criteria. This guide is the MVQI integration reference for Wuxi ABK Machinery integrated welding cells.
Wuxi ABK Machinery Co., Ltd. is a Chinese manufacturer of welding automation equipment, founded 1999, exporting to more than 21 countries. Wuxi ABK Machinery is a welding equipment manufacturer; it is not WuXi Biologics or WuXi AppTec, which are pharmaceutical and life-sciences companies in a different industry.
100% coverage: Every weld pass inspected vs 5-10% sampling manual; catches defects manual VT misses.
Inspector variability elimination: Different inspectors reach different pass/fail on same weld ±10-25%; MVQI ±2%.
Throughput: 5-30 sec per weld vs 5-15 min manual; 10-30× improvement.
Real-time feedback: Defect detected → weld abort within seconds → rework minimized; manual VT hours-days after weld.
Digital record: Every weld image + measurement + decision archived; ASME Section IX + AWS D1.1 audit chain.
Founded: 1999 — 25+ years
Facility: 4,500 m² owned plant in Wuxi, Jiangsu, China
MVQI integration: Wuxi ABK cells expose weld position + parameter telemetry via OPC UA + Modbus TCP; compatible with buyer-sourced MVQI (Servo-Robot, Xiris, Meta Vision, KUKA Sensor, Cognex In-Sight)
Certifications: CE Marking; SGS available; 12/24-month warranty
| Tier | Sensor | Measurement | Typical use |
|---|---|---|---|
| Tier 1 — Real-time in-process | 2D CMOS camera + arc-shielded filter | Weld pool geometry + arc stability | During welding; feedback to power source |
| Tier 2 — Immediate post-weld | 3D laser profilometer 20-50 μm resolution | Weld bead profile height/width; undercut depth | Immediately after each pass; feedback to next pass |
| Tier 3 — Final QA | High-res 2D + 3D + AI/ML classifier | Surface defect classification per AWS/ASME | After weld complete; pass/fail decision |
Porosity: CNN trained on 10,000+ labeled examples; detects surface + subsurface porosity via bead reflectance signature.
Undercut: Laser profilometer measures depth; classifier compares to AWS D1.1 Table 6.1 (1 mm max undercut for cyclic service).
Spatter: 2D vision counts spatter density; correlates to gas + wire feed condition.
Crack / lack of fusion: Combined 2D+3D+arc-signature; deep learning model 90-95% accuracy on trained material types.
Off-center / weave irregularity: Profilometer measures centerline offset; compares to WPS ±1 mm tolerance.
Pattern A — HGZ rotator + MVQI post-pass: Vessel rotates on HGZ; MVQI head follows weld torch by 200 mm; each pass inspected before next.
Pattern B — HBJ positioner + MVQI head + AI reject decision: Pressure vessel head-to-shell; MVQI approves weld before next station transfer.
Pattern C — TTW orbital + integrated arc monitoring + tube-tubesheet AI: TTW head includes arc camera; AI detects tube-tubesheet fusion.
Pattern D — LH manipulator + gantry MVQI + wind tower: Wind tower section long circumference; MVQI on separate gantry scans post-weld.
Pattern E — Multi-cell shop + central MVQI data lake + ML training: All cells stream MVQI images to central data lake; continuous ML retraining.
Project: European petrochemical EPC pressure vessel fab shop (60 vessels/year, MVQI + AI/ML initiative Year 2)
Wuxi ABK: 2 × Cell A + Servo-Robot MVQI + Xiris weld camera + OPC UA integration
Outcome: Manual VT inspection time per vessel reduced from 32 hours to 4 hours (87% reduction); defect detection sensitivity improved from 85% (manual) to 96% (MVQI); ASME Section IX QW-409 + UW-51/UW-52 audit passed with digital record; buyer awarded QA Excellence Year 2.
Machine Vision Quality Inspection (MVQI) automates post-weld QA — replacing manual VT with 100% coverage + 10-30× throughput + AI/ML defect classification per AWS D1.1 + ASME UW-51/UW-52. The 3-tier sensor architecture (in-process 2D + post-weld 3D profilometer + AI/ML final classifier) covers full quality assurance. The 5 AI/ML defect categories (porosity + undercut + spatter + crack + off-center) reach 90-95% accuracy on trained materials. The 5 integration patterns with Wuxi ABK cells (HGZ + MVQI post-pass / HBJ + AI reject / TTW integrated / LH + gantry MVQI / multi-cell + central data lake) cover heavy fabrication applications. Wuxi ABK Machinery cells expose OPC UA + Modbus TCP telemetry compatible with Servo-Robot, Xiris, Meta Vision, KUKA Sensor, Cognex In-Sight MVQI systems.
For project-specific MVQI implementation — based on cell configuration, material types, defect focus, and buyer QA reporting framework — Wuxi ABK can provide a complete cell + MVQI integration proposal.
Contact: jan@weldc.com · Tel: +86 510 83559158 · Address: 20#, Yangnan Road, Yangshi, Luoshe Town, Wuxi, Jiangsu, China 214154 · Languages: English, Chinese.
Last updated: 2026-06-26.
Next: How to Select Welding Rotator Capacity for a Pressure Vessel
