Skip to content

VISION AI

Inspection that doesn’t end
at the verdict.

Vision AI inspects each product, judges whether there is a defect and of what type, and keeps every result with its evidence.

Deployment in progressAn inspection system for precision metal parts is being built as part of a smart-factory project.
Vision AI doesn’t stop at the pass-or-fail decision.
— Narration, Vision AI product film

The problem

Even a single precision part
has many surfaces to inspect.

Manual inspection varies with experience, lighting and fatigue. Curved, reflective metal surfaces are hard to image consistently, defect samples are often too scarce for conventional training, and — above all — results rarely survive as a quality history for each product and work order.

A connected inspection flow

Every verdict leaves a record — with its reason.

  1. 01Capture
  2. 02AI verdict
  • Confirmed verdictLogged
  • Needs a human eyeOperator review

Quality history

Operator reviews and quality history are all part of the record.

The inspection flow

Every moment of inspection, in one connected quality flow.

  1. 01

    Capture

    Every product is captured the same way.

  2. 02

    AI verdict

    Suspect areas and defect types are judged.

  3. 03

    Operator review

    Inspections that need a human eye are routed to an operator.

  4. 04

    Quality record

    Results and evidence build each product’s and work order’s history.

What it does

Capture, judgment, review and record — connected in one quality flow.

  1. 01

    Defect localization and classification

    Locates suspect areas, judges whether there is a defect and what type, and shows where it is. Verdicts can run in real time on an edge device beside the line.

  2. 02

    Optics and AI designed together

    Lighting and exposure are designed around the part’s shape and surface, and multi-exposure fusion and normalization reduce the effect of glare and capture variation.

  3. 03

    Normal-pattern anomaly detection

    Lines with few defect samples can start from normal images; defect data collected over time supports validation and auxiliary models.

  4. 04

    Human review built in

    Confirmed verdicts are logged; inspections that need a human eye go to an operator.

  5. 05

    Quality history

    Good parts are recorded with the evidence for why they passed, and every inspection builds the quality history of its work order.

Straight answers

Current scope and considerations

Parts and lines differ, so we agree on validation criteria before we start.

  • Today

    An inspection system for precision metal parts is being built within a smart-factory project, laying the groundwork for future unattended night-shift operation.

  • Depends on

    Detection performance is validated on site data; metrics such as miss and false-call rates are agreed in a pilot.

  • Depends on

    Connections to MES and quality systems are scoped and built per site.

  • Not claimed

    Data in the film and screens is representative exhibition demo data, not real customer information.

Applications

Starting with precision metal parts, the approach extends to other parts where surface defects decide quality.

  • Precision metal parts

    Visual inspection of machined parts with reflective, curved surfaces.

    Deployment in progress
  • Bearings and precision-machined parts

    Surface-defect inspection of repetitive geometries.

    Potential application
  • Die-cast surfaces

    Detecting defect candidates on cast surfaces.

    Potential application
  • Electronic components

    Visual inspection of small components.

    Potential application

VISION AI

Could Vision AI work on your line?

Tell us about the parts, defect types and line conditions. We’ll review imaging set-up and a pilot scope with you.