Camera-based AOI for discrete manufacturing

Catch defects at
200 parts per minute.
Trained in a day.

Procunit reads your line cameras and calls pass/fail in real time — trained on as few as 50 labeled defect images. No machine vision PhD, no month-long recalibration.

Request a Pilot See How It Works

200+ parts/min 50-image training < 8ms latency

AI vision inspection system on a high-speed manufacturing conveyor line, with red bounding boxes overlaid on defective metal parts

1 in 12

defective parts escape inspection when lines run above 150 parts per minute

Human inspection breaks down at speed. Rule-based AOI breaks down when conditions change.

Human inspectors catch roughly 80% of defects on slow lines. Push past 120 ppm and escape rate triples — not because inspectors stop caring, but because there are not enough milliseconds per part. Rule-based AOI systems require a dedicated vision engineer to recalibrate every time lighting shifts, part geometry changes, or a new defect class appears. Procunit is not that system.

How It Works

From line camera to reject signal in under 8ms

Camera Frame

0.5ms capture

Mount or reuse existing line cameras — GigE, USB3, RTSP all supported

Procunit Inference

6ms inference

Model runs on-premises on a standard industrial PC; no GPU required

PLC Reject Trigger

<2ms signal

Pass/fail signal fires to your PLC reject system within 8ms total

200+

parts/min

Throughput on a single camera

images

Minimum training set

8ms

end-to-end

Camera frame to PLC signal

<0.3%

FP rate

False positive rate, pilot data

What Procunit Detects

Five defect classes. One model. Your line.

Surface scratches and scoring

Sub-millimeter scratch detection on polished, anodized, or painted surfaces

Dimensional drift

Out-of-tolerance geometry caught before a go/no-go gauge sees it

Assembly errors

Missing fasteners, wrong connector orientation, absent labels

Coating and finish anomalies

Thin spots, bubbles, skip zones in paint or powder coat

Porosity and voids

Subsurface voids visible in die-cast and injection-molded parts

Machined metal component with two AI inspection bounding boxes: one red marking a surface defect, one green on a passing area

From the field

What quality engineers say

We ran a 2-week pilot on our door latch stampings. Escape rate dropped from 1.4% to under 0.2%. Our previous AOI took four days to set up — this took an afternoon.

D. Ferreira

Quality Engineering Manager — Velcroft Industries

Our previous AOI system required a dedicated vision engineer every time we changed part families. With Procunit, our quality team labels 50 images and retrains overnight — no outside help.

L. Nakamura

Process Engineer — Haltern Precision

Run your first inspection in one shift.

We ship a pilot kit. You supply the camera feed and 50 defect samples. Most lines have a live pass/fail signal before the end of the week.