Electronics & PCB
Solder joint quality, missing SMD components, scratches on coated boards.
Quality Inspection
AI-Inspect
Sub-millimeter defect detection at line speed.
A ruggedized edge-AI vision unit that inspects every part, every cycle - catching scratches, dents, misprints, missing components and surface defects in real time, without slowing the line.
95-99%+
Detection Accuracy
<80 ms
Inspection Time
OK / NOK
Real-time Sort
Surface
& Defect Inspection
Live on the line
Product Overview
Inspection toolkit
Every measurement your line needs, on one platform. From outline matching and grayscale analysis to OCR, barcode decoding, dimensional gauging and high-speed position correction - configure each station with the exact toolkit your part needs, then deploy the trained recipe to the edge GPU on the line.
Capabilities
What AI-Inspect does on your line
Surface defect detection
Scratches, dents, porosity, weld spatter, paint runs, contamination - pixel-level segmentation flags exactly where the defect is.
Presence / absence checks
Missing screws, clips, labels, stickers, gaskets, washers - verified before the part reaches the next station.
Dimensional gauging
Sub-pixel measurement against CAD nominal with pass/fail tolerances and full SPC trend data.
Print & OCR verification
Date codes, batch numbers, barcodes, serials, logos - read, validated and reconciled with the MES record.
Edge inference, no cloud
Inference under 80 ms on the station GPU. No internet dependency, no production data leaves the plant.
Closed-loop learning
Every NOK feeds the next training cycle. Models improve weekly without re-engineering the station.
Sample detections
OK vs NOT-OK, side by side
Real captures from deployed lines, graded the same way every cycle.
OK - within specNOT-OK - flagged
01Size variation
OKvsNOT-OK
Bounding box flags parts that fall outside the expected dimensional envelope.
02Feature exist
OKvsNOT-OK
Confirms a required feature - thread, slot, marking - is present on every part.
03Missing area
OKvsNOT-OK
Detects chipped, broken or missing material along an expected boundary.
04Counting
OKvsNOT-OK
Counts repeated features and flags any deviation from the target count.
05Mismatch
OKvsNOT-OK
Catches color, orientation or part-type mismatches in assemblies.
06Angle
OKvsNOT-OK
Measures angular position against a reference and flags off-axis parts.
07Pitch
OKvsNOT-OK
Verifies spacing between repeated features against tolerance.
08Width
OKvsNOT-OK
Sub-pixel width gauging against CAD nominal with pass/fail bands.
How it works
Four simple steps, every single part
- Step 01
Sense the part
A sensor or line signal lets the system know a new part has arrived.
- Step 02
Take the picture
An industrial camera with steady lighting captures a clean, consistent image every time.
- Step 03
Check with AI
The trained model studies the image in milliseconds and marks anything that looks off.
- Step 04
Pass or reject
A clear OK or NOT-OK is sent to the line, so good parts move on and defects are stopped instantly.
Where it's deployed
Built for real production lines
Automotive parts
Casting porosity, machined surface finish, dimensional checks scanned per single part.
Pharma & packaging
Tablet shape & color, blister fill, label and date-code OCR verification.
Weld quality
Bead width, undercut, spatter and porosity scoring per joint, traceable per VIN.
Sheet metal
Scratches, dents, draw marks, and edge burrs on stamped panels.
Hydraulic & seals
Seal seating, O-ring presence, surface integrity on machined housings.
Integrations
Plays nicely with your stack
AI-Inspect ships with connectors for the most common shop-floor and IT systems, so production data flows where you already track it.
PLC (Siemens, Allen-Bradley, Mitsubishi)OPC-UAMQTTModbus TCPREST / WebhookMES (SAP, Ignition, Tulip)Power BIGrafana
Flexible
Deployment
On-prem, edge, or air-gapped
Every Advitiix product runs fully on-prem. No production data leaves the plant unless you explicitly opt in to cloud dashboards.
FAQ
Questions we hear from plant heads
How fast can AI-Inspect be integrated into our existing line - is it plug and play?
Yes. The hardware is pre-configured and ships ready to mount. A typical station is bolted to the line, wired to the PLC (24V I/O or OPC-UA / Modbus TCP), and online in 1-2 days. Model tuning runs in parallel during a 1-2 week pilot, so you see live OK/NOK results from day one without disrupting the existing PLC or MES logic.
Can it run without internet?
Yes. Inference, logging and PLC signalling run fully on-prem. Cloud dashboards are opt-in.
What's the typical ROI timeline?
Most lines see payback in 6-12 months from reduced rework, lower scrap, and lower QC headcount.
Can one station inspect multiple part types?
Yes - multiple models load per recipe and switch on PLC signal in milliseconds.
Put AI-Inspect on a line you care about.
Send us a few sample parts or a short video of the station - we'll come back with a feasibility plan and a deployment estimate within a week.