engiAI inspection
Inspecting plastic parts: camera + laser + AI (YOLO)
A working industrial architecture in which every component does its own job.
Laser + camera — measuring the geometry
Laser triangulation: the laser projects a line (a profile) onto the part and the camera films it from a known angle. From the curvature of the reflected line the system works out the height and shape at every point. As the part passes under the laser line, a full 3D profile is assembled. This is metrology — control of dimensions, warping and sink marks.
Camera + YOLO — finding defects
YOLO (You Only Look Once) is a neural network that detects objects in a 2D image. It is trained to recognise cosmetic defects — scratches, cracks, inclusions, short shots, burrs, stains — and to box them in real time. This is classification and detection, not measurement.
Optics on transparent and glossy parts
Triangulation relies on the beam reflecting off the surface. On transparent, glossy and very dark plastics part of the light enters the material and refracts, so false reflections and noise appear in the data. Such surfaces need preparation: a matting spray, special lighting or a different shooting angle.
For transparent parts (lenses, optics, clear housings), where refraction has to be taken into account on purpose, separate methods are used — deflectometry and schlieren imaging.
The neural network, meanwhile, works with the image the camera delivers, whatever the nature of the light: it looks for defect patterns in the pixels rather than analysing the physics of reflection or refraction.
How it is assembled on the line
The part travels along the conveyor. A laser profilometer captures the 3D geometry and compares the dimensions with the reference. In parallel, industrial cameras with different lighting (direct, side, back) film the surface while the detection model finds cosmetic defects in real time. The results of both channels are combined: the part passes only if it is sound in geometry and in appearance alike. Rejects are pushed off the line automatically.
