GigE Vision and USB3 Vision cameras
GigE Vision and USB3 Vision cameras stream directly into a Gatepoint gateway. Lens, mount and trigger stay as they are.
Vision & Inspection
Teach the cameras you own to see more
We add AI defect detection to the industrial cameras and smart cameras already on your line — keeping the mounts, guarding and validation you paid for, and wiring every verdict into your PLC and MES.
The problem
The camera is fine. The software is out of its depth.
Most lines we walk already have cameras. A smart camera went in years ago to confirm a label or count fasteners, and it still does that well. Then the job grew: the same station is now expected to catch weld spatter, thread damage and cosmetic scuffs — defects that change shape from part to part. Tolerances get widened, false rejects climb, and eventually someone bypasses the station on second shift.
Replacing the whole cell means new brackets and guarding through a safety review, re-validation with your customer or quality system, and days of line time. Usually the mount, the trigger and the lens are sound. What the station lacks is judgement — a model that has learned what your good parts and your defects actually look like.
How it works
What we reuse, and what we add
Smart-camera image output
Where a smart camera runs your measurement tools, we take the images it streams or pushes and add AI classification beside its result.
Mechanics and guarding stay
Brackets, shrouds and guarding that passed your safety and validation reviews stay in place. We add compute, not a new cell.
PLC handshake and reject timing
We map into your trigger, encoder and reject logic over digital I/O, EtherNet/IP, PROFINET or Modbus TCP, so the reject gate behaves as maintenance expects.
MES and traceability
Verdicts and image references flow to your MES or historian over OPC UA or MQTT, attached to the part ID. See integrations.
Shadow mode before switchover
The new model runs beside the current tool, logging every disagreement until your quality team signs off.
Retrofit or replace
When replacing the camera is the better answer
We would rather say so early than retrofit something that cannot work. These are the signs that a camera, a light or the station should change:
- Motion blur. Rolling-shutter or analog cameras on a fast conveyor smear the edges a model needs.
- Too few pixels. A defect under about 3 pixels across at the current field of view won't be found reliably.
- No contrast. Ambient light or the wrong angle hides the defect before the camera sees it.
- The wrong view. No existing camera can see the face where the defect occurs.
- End-of-life hardware. No spares, no updates, no way to get images out.
Then we quote a partial upgrade — a Lensnode V on the existing mount, or a new light — or a full AI visual inspection cell, with the cost of each. Our field note Retrofit or replace? covers the decision in depth.
Process
From line audit to switchover
-
Line audit and image capture
We record images from your cameras at production speed, review triggers, lighting and PLC logic, train a baseline and give you a written verdict per station.
-
Shadow run on your line
A loaner Gatepoint taps the image stream without touching reject logic, judged against success criteria agreed in writing.
-
Integration and cutover
PLC handshake, MES fields and operator screens, then a planned switchover during a scheduled stop. The original logic stays available as a documented fallback.
-
Care after go-live
Perflens Care watches for drift and retrains the model as materials, suppliers and tooling change.
Deliverables
What a retrofit delivers
- Station-by-station audit with a retrofit, upgrade or replace recommendation
- Gatepoint gateway installed on DIN rail in your existing panel or a compact enclosure
- Trained model and model card, validated on held-out images from your own line
- Updated I/O map and PLC changes, reviewed with your controls engineer
- MES or historian integration carrying part ID, verdict, defect class and image reference
- Shadow-mode report listing every disagreement between the old tool and the new model
- A tested fallback procedure, as-built documentation, operator training and ONNX model files you own
Compatibility
What a retrofit typically connects to
- Camera interfaces
- GigE Vision and USB3 Vision (GenICam-compliant); smart cameras that can stream or push their images
- Streams per gateway
- Up to 4 on Gatepoint G2, up to 12 on Gatepoint G4
- PLC integration
- Isolated digital I/O, EtherNet/IP, PROFINET, Modbus TCP/RTU
- MES and historians
- OPC UA, MQTT with Sparkplug B, REST and webhooks
- Decision latency
- Under 40 ms frame-to-I/O for typical inspection models
Honest fit
Is a retrofit right for your line?
A good fit when
- Your cameras are global-shutter, in focus and aimed at the defect, but the software misses defects or rejects good parts
- The station is validated and repeating that validation would cost more than the inspection problem itself
- You run several identical stations and want one model maintained across all of them
- You need verdicts and reject images in the MES for traceability
Probably not when
- The existing cameras are analog, rolling-shutter on fast motion, or out of support
- No current camera position can see the defect
- The line will be rebuilt within a year — design the new cell properly instead
See it in context
Where retrofits fit
Weld-nut verification on existing cameras
A stamping line's own cameras plus a PLC handshake, in a modeled automotive scenario.
Read more
Automotive & mobility
Fastener, weld and surface checks on lines with years of validated vision.
Read more
Metals & fabrication
Weld seams, spatter and surface defects that rule-based tools struggle to judge.
Read moreFAQ
Questions we hear often
Do we have to stop the line to install a retrofit?
Not for the evaluation. During shadow mode the gateway only reads images and never touches reject logic. The cutover itself is planned around a scheduled stop, and the original logic stays available as a fallback.
Can our existing smart-camera tools keep running?
Yes, and they often should. A common pattern keeps dimensional checks on the smart camera and adds the AI model for cosmetic or variable defects, with both results combined in the PLC.
Which cameras can you work with?
GigE Vision and USB3 Vision industrial cameras, and smart cameras that can stream or push their images. If a camera only talks to a closed PC application, the Feasibility Sprint checks whether its images can be reached; sometimes a new camera on the same mount is cheaper.
Does a retrofit change our validated process?
It adds a decision to it, so it goes through your change control. We supply the evidence that process needs — the shadow-mode comparison, the model card and the FAT/SAT records — so the review is about data rather than opinions.
What does a retrofit cost?
It starts with a $9,500 Feasibility Sprint, fully credited if you go ahead within 90 days. Because cameras and mechanics are reused, retrofits usually cost less than a new cell; you can buy the system outright or subscribe from $2,950 per station per month. See pricing.
Find out what your cameras can already see
A three-week Feasibility Sprint tests images from your existing stations and tells you, in writing, whether to retrofit, upgrade or replace.