Industries · InspectionAI
Built for inspection programmes across physical products.
Each programme names the defect classes a model is trained on — and what a camera cannot do for that industry.
01 — Six programmes
Electronics and PCB assemblySolder defects, missing components, alignment errors.
Boards are flat, fixtured and repeatable — the easiest geometry a camera ever gets.
Defect classes a model is trained on
- Missing component
- Misalignment
- Solder bridging
- Insufficient solder
- Tombstoning
- Wrong part
The boundarySurface and placement only. It does not replace electrical test, and it cannot see a fault under a BGA.
Semiconductor equipment and wafersSurface contamination, particles, dimensional tolerance.
Controlled lighting and a rare, small, expensive defect population — where consistent machine attention beats a tiring eye.
Defect classes a model is trained on
- Particle contamination
- Scratch
- Residue
- Edge chip
- Pattern anomaly
The boundaryOptical inspection at the resolution your camera provides. Sub-optical defects need metrology.
Automotive componentsSurface scratches, weld quality, conformance.
High volume, mixed surfaces and a taxonomy your quality system usually already defines.
Defect classes a model is trained on
- Surface scratch
- Dent
- Weld porosity
- Undercut
- Spatter
The boundarySurface appearance. Not a substitute for radiographic or ultrasonic weld testing, and it does not measure dimensions.
Solar and PV panelsMicro-cracks, discolouration, delamination.
Large, repetitive surfaces where human attention degrades fastest.
Defect classes a model is trained on
- Micro-crack
- Cell discolouration
- Delamination
- Busbar defect
- Snail trail
The boundaryWhat is visible in the capture you provide. Electroluminescence reveals cracks a normal camera cannot.
Packaging and consumer goodsLabel placement, seal integrity, print defects.
Very high line speed and very low defect rate — the audit trail matters as much as the detection.
Defect classes a model is trained on
- Missing label
- Skewed label
- Print defect
- Seal gap
- Wrong artwork
- Fill level
The boundaryAppearance, not content. Barcode and text verification is a different technology.
Medical device manufacturingSeal integrity, burrs, particulates, tolerance.
A regulated environment where the record of who decided what is worth as much as the decision.
Defect classes a model is trained on
- Seal integrity
- Burr
- Particulate
- Surface finish
- Assembly error
The boundaryA decision-support tool. Not a validated device, no regulatory clearance, and it does not replace your qualified process.
02 — Not on this list?
The question is whether a person can see it.
If a trained inspector can tell good from bad by looking, a model can usually be taught the same call.
The defect is visible, has a name, and two inspectors would mostly agree on it.
Visible but rare, lighting-dependent, or often disputed. Solvable, but the dataset takes longer.
Internal, dimensional, or below what the camera can resolve.
Talk to INSAIT