A vision system that works on the bench and fails on the line is almost never a software problem. It is usually illumination, optics, or a sensor chosen from a specification sheet rather than against the actual scene. By the time the machine learning is being blamed, the cause is often three steps upstream.
What the work is
Sensor and optics selection
Chosen against the real scene: working distance, field of view, depth of field, motion, and the defect you actually need to resolve. Not against a datasheet comparison table.
Illumination design
Geometry, wavelength, diffusion and stray-light control. Most inspection problems that look intractable become straightforward once the lighting is right.
Imaging chain tuning
From raw sensor output through to a repeatable result: exposure, gain, calibration, and the preprocessing that determines whether downstream analysis has anything to work with.
Machine vision integration
GenICam and GigE Vision device integration, and the bridge into whatever system consumes the result: PLC, MES, or a control room platform that speaks ONVIF.
Feasibility assessment
An honest answer on whether a given inspection is achievable at the required rate and confidence, before a line is committed to it.
Where an engagement usually starts
For inspection problems we usually start with a feasibility study against your own samples, including the failure cases. The output is a written assessment of whether it can be done, at what rate, and what it would take.
What we do not do
We do not sell cameras or lighting hardware, and we have no distribution agreements. That is deliberate: it means the recommendation is not shaped by what we would earn on the parts.
Related practices
Problems here usually touch these too
03
Industrial AI & Edge Computing
Inference where the data is, sized for real hardware and real power budgets rather than a benchmark.
04
Embedded Linux & Edge Platforms
BSPs, Yocto builds, RTOS decisions and the platform work everything above it depends on.
05
FPGA & Engineering Software Tools
HDL work, hardware acceleration, and the internal tools that make an engineering team faster.
Bring us the challenge
The one that has been handed back, sits between two suppliers, or nobody can say is possible yet. A short call costs you nothing and you will speak to one of our consultants.
Engineer to engineer. No handoffs.