SolutionsIndustriesTechnologiesR&DProjectsAboutBlog  Robotics Integration  Industrial Automation  Mechanical Engineering  AI for Manufacturing  Digital Twin  Machine Vision  Physical AI  Factory AutomationCareersContactTalk to an engineer

Solution — Machine vision

Lighting first. Everything else follows.

Inspection, metrology and vision guidance engineered from the physics of the defect outward — with a false-reject rate we measure, state and hold in production.

PASS 0.994 POROSITY 0.971 SCRATCH 0.883 CAM-01 · 5MP · 12 mm · RING LIGHT · 38 ms/PART
FIG. — Inspection frame with classificationsSHEET RL-410 · REV D

Approach

Most failed vision projects were lost at the illumination stage.

A scratch on a machined face, a pore in a casting and a smear on a printed label each require a different light. Choose the illumination and optics correctly and a classical algorithm often suffices. Choose them last and no amount of deep learning rescues the image.

We prototype on your actual parts — good ones and, critically, the defective ones — before quoting a system. The deliverable of that study is a measured detection rate and false-reject rate on a sample set, which is the only honest basis for a specification.

Deep learning is deployed where classical methods genuinely cannot generalise: variable textures, cosmetic judgement, or defect classes that resist explicit description. Even then it is bounded by a confidence threshold tied to a documented quality plan.

Scope

What is included.

Delivered as a defined scope with acceptance criteria, not as a time-and-materials estimate that drifts.

  • 01Feasibility & imaging studyon your good and bad parts
  • 02Illumination designring, dome, dark-field, coaxial, DOAL
  • 03Optics & camera selectionresolution vs field of view vs speed
  • 04Classical inspectionblob, edge, template, subpixel
  • 05Deep learning inspectionsegmentation and classification
  • 06Dimensional metrologygauge R&R capable measurement
  • 07Code reading1D, 2D, DMR verification to ISO 15415
  • 08OCR / OCVdate, lot and marking verification
  • 093D visionstructured light, stereo, laser profile
  • 10Vision-guided roboticshand-eye calibration, pose estimation

Applications

Where this is used.

Surface defect inspection

Scratches, porosity, contamination and cosmetic defects at line rate with a documented false-reject rate.

Dimensional measurement

Non-contact metrology with gauge R&R studies, not just a number on a screen.

Assembly verification

Presence, orientation, count and correct-part checks that catch errors before value is added.

Vision-guided robotics

Hand-eye calibrated pose estimation for picking, placing and adaptive path correction.

3D & bin picking

Structured-light and stereo perception for randomly presented parts with collision-free grasping.

Traceability & code reading

Barcode, data matrix and OCR with grading to ISO 15415 for regulated traceability.

Stack

Tools and platforms.

MVTec HALCONCognexKeyenceBaslerOpenCVPyTorchTensorRTNVIDIA JetsonStructured lightLaser profilometryTelecentric opticsGigE VisionGenICamISO 15415Gauge R&R

FAQ

Questions about machine vision

Can you inspect a defect we cannot reliably describe?

Often, yes — that is exactly where deep learning earns its place. What we need is a sample set that includes the defect and, just as importantly, the acceptable variation. If your own inspectors disagree with each other on borderline parts, we resolve that first, because an inconsistent standard cannot be automated.

What accuracy can you achieve on measurement?

It depends on the field of view and optics. With telecentric optics and a controlled setup, subpixel edge measurement in the range of a few microns is realistic on a small field. We validate any measurement claim with a gauge R&R study on your parts rather than quoting a sensor datasheet.

How fast can the system run?

Typical inline inspection runs 20 to 60 milliseconds per part including acquisition, inference and result handling. Faster is achievable with multi-camera parallelism and GPU inference; we model the timing budget as part of the feasibility study.

Do we need to send you parts to get a quote?

For anything beyond code reading or presence checking, yes. An imaging study on real parts is the difference between a specification we can commit to and a guess. It is a short, low-cost engagement and it is the single best predictor of project success.

Next step

Ready to scope your next automation programme?

Send us a drawing, a cycle-time target or a line layout. Our engineers respond with a technical assessment — not a brochure.