Machine vision · Inspection
Automated visual inspection does not start with the camera: it starts with the defect to detect. We define what is accepted, what is rejected and with what evidence, then select lighting, optics, camera and algorithm. The cell integrates with the line, PLC or MES and is validated with real parts, including good and bad ones.
Defect, presence, dimension and assembly inspection with a repeatable, measurable criterion.
What it solves
Comparison
The technology is chosen for the defect, not the trend. A simple rule can be more robust than a neural network; 3D imaging can solve what 2D cannot see.
| Criterion | 2D vision | 3D vision | AI |
|---|---|---|---|
| What it measures | Color, texture, presence and planar position | Height, volume, profile and orientation | Patterns that are hard to describe with rules |
| Best for | Labels, surfaces, assembly and codes | Stacked parts, welds, volume and wear | Variable defects, textures and sorting |
| Limit | Does not see height or depth | More calibration and more data | Needs labeled samples and validation |
In detail
Lighting solves more defects than the camera. We define geometry, angle, diffusion and filter before choosing resolution and lens.
The system must distinguish a good part from a bad one with a criterion the customer can explain and audit. We measure false positives and false negatives before accepting the cell.
The camera does not live alone: it receives a signal, inspects, decides and returns a result. Integration defines cycle time and system stability.
Cost includes camera, optics, lighting, mounting, software, integration, validation and maintenance. Payback comes from rework, scrap, claims and inspection hours.
Industries
Related guides
Frequently asked questions
A rule is preferable when the defect can be described with measurements, color or presence. AI is justified when the defect varies and cannot be expressed with simple thresholds. Many cells combine both.
It is calculated from the field of view and the minimum defect: the defect must cover several pixels. Higher resolution does not compensate for poor lighting or optics.
Yes, with a trigger, encoder and compatible exposure times. Line speed and processing time define whether inspection is in-line or at a buffer station.
With a set of good, bad and borderline parts, and a measurement of false positives and false negatives. Validation is documented before acceptance.
Yes. Trigger, result, reject, recipe and record signals are defined. The interface is agreed with the plant automation owner.
Optics, lighting and mounting are designed for the environment. Cleaning, frequency, spare parts and periodic verification criteria are defined.
Yes. A trial with real parts is the recommended step to confirm lighting, resolution, algorithm and criteria before quoting the cell.
Yes. Assessment and integration are coordinated from Córdoba for Argentina, Chile, Peru, Bolivia, Paraguay and Uruguay.
Next step · Gist Point
Send photos or a video and we will define whether a vision trial, an assessment or a complete cell makes sense.