Every part inspected.Every defect caught.
NanoQuantumAI puts AI cameras on your production line to catch scratches, cracks, misprints and missing parts in real time, traces every part, and finds out why defects happen.
- Part #48,214reviewpossible burr at edge · confidence 71% → quality team
- Part #48,213pass
- Part #48,212defectsurface scratch 2.4 mm · rejected to bin B
- Part #48,211pass
Root-Cause AI: scratches rose 4x after 11:20 am, matching tool change on Press 2. Suggested: inspect Press 2 die.
A tired eye at hour seven misses what a camera never will.
Manual checks see a sample. Defects that slip through show up later as customer complaints, returned batches and audits. One lens per line closes that gap.
See every line, shift and product at once.
Pass, defect and review rates update with every part. Plant heads see the whole floor, quality managers drill into the shift where the trouble started.
- Live rates per linePass, defect and review percentages with parts per minute and trend.
- Compare shifts fairlySame product, same rules, so shift A and shift B are judged on equal terms.
- Alerts on driftGet notified when a defect rate crosses the limit your team set.
Defect rate by hour
By shift
| Line | Speed | Pass | Defect | Review |
|---|---|---|---|---|
| Line 1 · Bottle caps | 600/min | 99.6% | 0.3% | 0.1% |
| Line 2 · Blister packs | 180/min | 99.1% | 0.6% | 0.3% |
| Line 3 · Brackets | 240/min | 99.2% | 0.6% | 0.2% |
| Line 4 · PCB assemblies | 45/min | 98.7% | 0.9% | 0.4% |
Judged at line speed. Rejected before packing.
Each part is imaged as it passes the camera and judged in milliseconds. Bad parts are kicked to a reject bin or the line controller gets a signal, with no change to your line speed.
- Inspection at full speed15 to 60 ms per image, up to 1,200 parts per minute per camera.
- Automatic rejectionAir jet, pusher or diverter triggered through PLC output.
- Uncertain parts go to peopleLow-confidence results are set aside for your quality team, never guessed.
Your defects, your limits, written down.
Every defect type has reference images and a severity rule set by your quality team. The inspector follows those rules exactly, on every shift.
- Reference images per defectScratches, cracks, burrs, dents, misprints, missing components and more.
- Measured limitsReject a scratch over 1.5 mm, accept one under it. Limits in mm, not opinions.
- Approved by qualityRule changes need sign-off and every change is logged.
Not only what failed. Why it started failing.
Root-Cause AI lines up defect spikes against machine events, tool changes, shifts and material batches, then tells you where to look first.
- Spike detectionSpots when a defect type rises above its normal rate on any line.
- Linked to real eventsTool changes, die swaps, coil and resin batches, shift handovers.
- Evidence attachedEach finding comes with the images and timings behind it.
Likely causes, ranked
Every part has a picture and a verdict on file.
When a customer complaint arrives, search the batch and show the image of every part shipped, with the time, camera, model and decision.
- Search by batch, part or timeFind any part in seconds, across lines and months.
- Audit-ready reportsBatch reports in PDF and CSV for IATF 16949, ISO 9001 and GMP audits.
- Kept as long as you needRetention set per product, with tamper-evident records.
- Part
- #48,212 · Bracket BR-220
- Inspected
- 14 Oct · 11:42:07.318
- Camera
- CAM 3A · model v14
- Finding
- Surface scratch 2.4 mm · 97% confidence
- Action
- Rejected to bin B
New product on Monday. Inspected by Wednesday.
Your quality team teaches new products and new defect types with a few dozen labeled images. No code, no data science team, no waiting on a vendor.
- Label in the browserDraw a box around the defect, pick its type, done.
- Checked before it goes liveAccuracy is tested on images the model has never seen.
- Versioned modelsRoll back to the previous model with one click if results change.
One line. Four weeks. Accuracy you can check.
The pilot runs next to your current inspection, so you compare our results with your own before anything is rejected automatically.
Mount and collect
Camera Kit mounted on one line. We collect images of good and bad parts at full speed.
Teach and set rules
Your quality team labels defects and sets reject limits in the Defect Library.
Shadow mode
The inspector judges every part but rejects nothing. Results are compared with your checks.
Accuracy report
Catch rate, false reject rate and review rate reported openly, before you decide.
Your quality team stays in charge.
The AI inspects. Your people decide what counts as a defect and what happens to it.
- Uncertain results go to a personParts below the confidence limit are set aside for your quality team, never passed on a guess.
- Reject rules are yoursEvery limit is set and approved by your team, with a log of who changed what and when.
- Your images stay yoursImages and models belong to you and are never used for other customers.
- Cameras watch products, not peopleEvery camera is aimed at parts on the line. Workers are not recorded.
Talks to your controllers and factory systems.
Inspection runs on an edge computer at the line. Decisions never wait for the internet.
Questions from the shop floor
Do we need special cameras?
We recommend a Camera Kit matched to your part size, line speed and surface finish, because lighting decides most of the accuracy. If you already have industrial cameras with GigE Vision or USB3 Vision output and steady lighting, we test them during the pilot and use them when they are good enough.
How fast can it inspect?
A single inspection point runs at 15 to 60 ms per image depending on resolution and the number of defect types. That covers most lines up to 1,200 parts per minute per camera. Faster lines are split across two cameras or a line-scan camera.
How many images do we need to start?
Usually 30 to 50 images of good parts and 10 to 20 examples of each defect type. Rare defects can start with fewer examples and improve as the line sends uncertain parts to review and your team labels them.
Does it work offline on the shop floor?
Yes. Inspection runs on an edge computer next to the line, so pass and reject decisions never wait for the internet. Results and images sync to the dashboard when the connection is back.
Can it connect to our line's controllers?
Yes. Reject signals go out through digital I/O or directly to Siemens, Allen-Bradley, Mitsubishi and Omron PLCs over OPC UA, Modbus TCP or EtherNet/IP. Results can also be pushed to your MES or ERP over REST or MQTT.
How long does a pilot take?
Four weeks on one line. Week one is camera mounting and image collection, week two is training with your quality team, weeks three and four run in shadow mode next to your current inspection so accuracy is measured against your own results.
Who owns the images and models?
You do. Inspection images and the models trained on them belong to you and are never used to train models for other customers.
Turn the knob to your line speed.
A 5% manual sample checks 5,760 of them. NanoQuantumAI checks all 115,200, and keeps a picture of each one.