Case Studies

Our deployments

Real industrial deployments with measurable outcomes — Cognex, Keyence, Baumer cameras and 3D profilometers combined with AI/Deep Learning for quality control in series production.

Photo coming soon

Vision Inspection Keyence CA-HL04MX (linear) + CA-HF2100M (area)

Battery Can Inspection — 460 pcs per Cycle

Vision system integration, algorithm development, cycle-time optimisation.

Challenges

—460 cylindrical cans (ø21 × 74.5 mm) per tray — extreme throughput requirements
—Detection of micro-defects on sidewalls, bottom, internal clip and inside walls
—Tunnel-effect on cylindrical interiors limits illumination and contrast
—Linear camera synchronised with two independent servo axes (XY)
—Carrier cleanliness, content count and orientation verified between cycles

Solution

80 vision-camera sets across stations: line-scan WT inspection + matrix sidewall/bottom + high-resolution internal
Three resolutions used selectively (5 MP / 21 MP / VGA) — each defect class scanned at optimal pixel size
Full empty- and full-tray validation, plus repacking control integrated with Eco Cell platform
Automatic NG ejection without line stop

Business Impact

18.4 s

Cycle time / tray (460 pcs)

≥98%

Technical availability (3-shift operation)

100%

Inline inspection — no statistical sampling

Photo coming soon

AI / Deep Learning Cognex IS3816MX

Workpiece Carrier Loading Control — AI Classify

Vision system integration, AI model training, loading verification on a fast assembly line.

Challenges

—Both presence AND correct position of small electronic components must be verified
—Components have no clean geometric signature — classical edge/blob tools fail
—Many product references on one line — representative training set needed for each
—High line throughput limits available analysis time per cycle
—Variable specular highlights from PCB surfaces and shielding

Solution

ViDi Classify model trained on curated multi-reference dataset to recognise correctly loaded vs mis-loaded carriers
Cognex IS3816MX with dedicated lighting tuned to suppress highlights
Result fed to PLC as per-pocket OK/NG decision before downstream process
Model retrainable for new patterns without changing hardware

Business Impact

Classification instead of geometric rules

↓OEE

Losses reduced by faster reaction to mis-loads

100%

Per-pocket inspection coverage

Photo coming soon

AI / Deep Learning Cognex D902C

Solder Joint Quality — Deep Learning Inspection

Two-sided solder coverage check on bobbin pins after dip-soldering.

Challenges

—Very small inspection field around each pin — sub-millimetre features
—Dip-soldering produces irregular tin formations — no single geometric reference exists
—Large unit-to-unit variability defeats classical thresholding methods
—Two-sided inspection required for full coverage assessment
—False negatives must be minimised to protect downstream electrical reliability

Solution

Two images per part (front/back) with controlled coaxial lighting
ViDi Deep Learning classifier evaluates whether solder volume on each pin is within spec
Borderline cases routed for operator review; clear OK/NG handled automatically
Model retrainable for new PCB variants without code changes

Business Impact

<0.5%

False reject rate — matches manual expert inspector

2×

Both sides in one cycle — no repositioning

↑

Defect data fed back to upstream wave-solder process

Photo coming soon

Vision Metrology Baumer VCXG.2-82M monochrome

High-Precision Holder Positioning — sub ±10 µm

Visual measurement-based positioning of a holder relative to micro-pins, accuracy below ±10 µm.

Challenges

—Required positioning accuracy below ±10 µm at the pin corners
—Very small reference features with low contrast against the holder
—System must remain stable under industrial vibration and lighting drift
—Optical limits caused by small physical size of the pins
—Repeatability guaranteed across long production shifts

Solution

Monochrome high-resolution camera with telecentric optics — parallax eliminated
Sub-pixel corner detection on pin tips combined with a calibrated coordinate frame
Two-step strategy: coarse pre-positioning, then fine vision-driven correction below ±10 µm

Business Impact

<±10 µm

Positioning repeatability

↑yield

Higher first-pass yield on critical micro-mechanical operation

vs.

Higher repeatability than operator-tuned mechanical fixtures

Photo coming soon

Vision Guidance Cognex IS7802 2 MP

Adaptive Sealant Application & Inspection

Two-cycle vision: dynamic XY/angle measurement, then post-application bead quality control.

Challenges

—Component position varies cycle-to-cycle — XY and angle must be measured live
—Robot path for liquid sealant must adapt continuously to the measured geometry
—Post-application QC must judge a continuous fluid bead — width and continuity
—Even small head deviations cause leaks — micro-tolerances on the path

Solution

Cycle 1: vision system computes X, Y and angle of the part, sends robot the adjusted dispensing path
Cycle 2: same camera re-photographs the bead and verifies continuity, width and correct routing
Both inspections share a single camera & fixture — minimal added hardware footprint

Business Impact

±0.3 mm

Path correction accuracy

Leakage rate post go-live

100%

Automatic per-unit bead documentation

Photo coming soon

AI / Deep Learning Cognex IS3805M

Steering Gearbox Contamination Detection — AI

AI-based detection of foreign objects on machined surfaces of steering gearboxes.

Challenges

—Customer faced sporadic foreign objects (screws, centring pins) on working surfaces
—Small visual difference between accepted and contaminated states
—Three images per part needed — different angles plus 4-bolt fastening verification
—System must detect contamination AND missing/loose bolts simultaneously

Solution

Three-image inspection: two for foreign-object detection, one for 4-bolt check
ViDi Deep Learning trained on representative contamination examples from production
Combined attributive + AI logic — defect type recorded for each rejection

Business Impact

100%

Automated EOL inspection replaced 100% manual check

False positives across 3-month production validation

Detects known and previously unseen contamination types

Photo coming soon

EOL Inspection Cognex IS8505MP 5 MP

Final Product EOL Inspection — Multi-Defect

Combined attributive inspection of plugs, connectors, coolant ports, label DMC and OCR matching.

Challenges

—Many defect types on one product: pin position in plugs, connector damage, coolant ports
—Simultaneous DMC read + OCR comparison against printed label
—Many connector and label variants per single product reference
—Final station — escapes here go directly to the customer

Solution

One 5 MP camera programmed with multiple inspection regions per cycle
DMC read and matched against OCR-decoded text on the label sticker
Defect class and location logged per unit for traceability
Reference changeover <2 s including full parameter and DMC config reload

Business Impact

100%

EOL coverage — major reduction in customer complaints

<2 s

Changeover — no operator action on product switch

DMC

Full traceability per serial number, MES-queryable

Have a similar project?

Get in touch