antıkytera
Lynx

The line
that never blinks.

Autonomous visual inspection at production speed. Lynx sees microscopic scratches, casting flaws, paint runs and weld defects on every unit that passes — continuously, identically, without fatigue.

Lynx inspection station with an industrial camera and controlled lighting imaging an aluminium casting on a production line fixture
A Lynx station imaging a casting inline — camera, controlled lighting, edge compute.Installed at a manufacturing customer's line
How it works
01

Capture

Fixed cameras and controlled lighting image each unit at line speed, from the angles that matter.

02

Detect

Models trained on your parts segment defects at pixel level and separate them from cosmetic noise.

03

Decide

Each unit is classified against your acceptance thresholds — pass, rework, reject — with a reason.

04

Act

The verdict goes straight to the PLC or MES: reject gates, rework routing, live process alerts.

Inside the station

One camera, one verdict, every part — before it leaves the cell.

Nothing waits for a lab, a sampling window or a shift review. The unit is imaged, graded against your acceptance criteria and routed — while it is still moving.

CAM 01VERDICT STREAMPASS · 0.998PASS · 0.994REJECT · porosity 0.41mmpass → next stationreject gate → reworkEDGE INFERENCE · < 200 ms

Graded, not just flagged

Every finding carries a class, a size and a confidence — so a 0.2 mm cosmetic mark and a structural flaw don't share a verdict.

Reasoned rejects

The operator sees why a unit failed, with the pixels highlighted. No black-box stops on the line.

Traceable per serial

The image, the verdict and the model version are stored against the unit for warranty and recall questions later.

Beyond pass/fail

The defects you catch are worth less than the ones you prevent.

Because Lynx sees every unit, it sees the trend before the scrap does. A tool wearing, a bath drifting, a supplier batch shifting — these appear as a slow change in defect distribution hours or days before anything crosses a threshold.

Process signals, not just QC

Defect type, position and severity are trended per station, shift, tool and supplier lot.

Early warning

Alerts fire on drift in the distribution, not on a single bad unit — the way an experienced line lead would read it.

Root cause with evidence

Every alert links back to the actual images behind it, so the argument on the shop floor is short.

Sampling tells you what probably happened. Lynx tells you what did.

100% inspection coverage

Every unit, every shift. No sampling plans, no statistical hope, no missed batches.

Pixel-level accuracy

Defects far below reliable human detection, graded by severity rather than a binary pass/fail.

Consistent by construction

The same part gets the same verdict at 3am on a Sunday as at the start of a Monday shift.

Drift detection

Defect patterns are trended over time so you see a tool wearing or a bath drifting before scrap appears.

Plugs into your line

PLC, MES and SCADA integration for real-time reject routing and full traceability per serial.

Retrains on your data

New part, new finish, new defect class — the model is extended without rebuilding the cell.

100%
Of produced units inspected rather than sampled
< 200 ms
Typical decision latency per unit at line speed
24/7
Operation with no inspector fatigue or shift variance
Deployment & fit
Hardware
Industrial area or line-scan cameras with controlled lighting, specified per part and cycle time.
Compute
Edge inference at the cell; no dependency on plant network or cloud availability.
Line integration
PLC over OPC UA or digital I/O; MES and SCADA for traceability and reporting.
Training data
Starts from a few hundred labelled samples per defect class; improves with production data.
Environment
Rated for factory conditions — vibration, dust, coolant mist, ambient light swings.
Pilot
Four to eight weeks from part study to a measured pilot on a single station.
Where it's applied

Production lines and castings · automotive components and paint finish · welds and coatings in energy and construction · packaging and print quality in consumer goods · anywhere a high volume of items has to look right, every single time.

Questions we get

We have very few defect samples. Is that a problem?

It's the normal case. We combine anomaly-style learning on good parts with synthetic and augmented defect data, then sharpen on real rejects as they occur.

Will it stop the line on false positives?

Thresholds are yours to set. Most customers start with Lynx advising alongside existing QC, then hand it authority once the confusion matrix is agreed.

Do we need to rebuild the line?

No. Lynx is a station: cameras, lighting, edge box. It is installed around your existing conveyor or fixture.

What about parts a human still has to judge?

Route those to Iris, where an inspector works under AI guidance and the report is generated automatically.

Deploy Lynx on your line.

Send us a part and your defect catalogue. We'll come back with a feasibility read before you commit to a pilot.

Book Your Demo →