Faststream Vision builds industrial machine vision systems that run on a production line: imaging chain, illumination, edge inference, PLC handshake and reject actuation. The engineering that decides whether a vision cell works is rarely the model. It is the lighting geometry, the trigger timing, the cycle-time budget and the handshake with the controller that already runs the line.
A defect that is invisible in the image cannot be found by any model. Most failed inspection pilots are lighting failures wearing an algorithm costume — a scratch that only shows under low-angle illumination, a print defect that vanishes under a dome, a specular surface that saturates half the frame.
So the imaging trial comes before the model. Faststream photographs real parts with real defects under several illumination geometries, and only then decides what the detection problem actually is. Frequently it turns out to be a classical vision problem that does not need a learned model at all, which is a better outcome: deterministic, explainable and free of drift.
Scratches, dents, inclusions, porosity, contamination and coating faults.
Gauging, edge and hole position, gap and flush, profile against tolerance.
Components fitted, fasteners present, seals seated, contents correct.
Orientation, sequence, mating and connector engagement.
Print quality, label placement, skew, legibility and code grading.
Date, lot and serial reading, and verification against expected content.
Shade matching, gloss, weave and finish consistency.
Bead geometry, continuity, spatter and undercut.
AI visual inspection is automated optical inspection that uses learned models, running on edge hardware, to classify or locate defects in images captured on a production line. A complete system includes cameras and optics, controlled illumination, encoder-locked triggering, inference hardware, a PLC handshake and reject actuation.
No, and it is worth checking. Many problems — dimensional gauging, presence and absence, code reading — are solved more reliably by classical vision, which is deterministic, explainable and does not drift. Learned models earn their place where defect appearance varies in ways a rule cannot capture.
Through the PLC. The system takes a trigger, returns a pass or fail result within the cycle-time budget, and drives reject actuation through digital I/O or fieldbus. Results and images are passed to MES or a historian for traceability.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.