Skip to content
Zester
An engineer working on a laptop beside test machinery
All applications

Machining An auto-components machine shop (con-rods, crankshafts)

Inspecting machined parts with the senior inspector's eye

Edge visual inspection that learns from the inspector's accept and reject decisions, instead of from a rulebook nobody wrote.

Stage
Pilot-ready
Sector
Machining
Capabilities
Computer VisionEdge AI
Deployment
Edge: Jetson Orin + 2 cameras, ring light, works offline.
Integrations
Line PLC reject gate (Modbus), Excel / Tally batch export
RING LIGHTINSPECTOR LABEL
Signal path. Illustration, not a site photograph.

The knowledge at risk

It lives in the head of the senior inspector.

He tilts a con-rod under the light and knows a machining mark from a crack. His defect categories are in his head, not in the quality manual.

Shop-floor constraints

  • Oily, reflective surfaces that defeat naive vision.
  • Parts arrive in mixed batches with frequent changeovers.
  • Customer audits need a traceable record per part.

What we build

  1. 01

    Build the cell

    Two cameras and controlled lighting at the inspector's bench. The part sits where he already puts it.

  2. 02

    Shadow the inspector

    For the first weeks the system only records. His accept, reject and defect name become the labels.

  3. 03

    Suggest, then check

    The model proposes a call. The inspector confirms or corrects in one tap.

  4. 04

    Keep the record

    Every part gets images, a decision and a defect code, exportable per batch.

Today: what it improves

  • A second pair of eyes on every part, including the night shift.
  • Per-batch defect records for customer audits.
  • Newer inspectors trained against real, labelled examples.

Over time: how it learns

A defect library in the inspector's own categories that grows every shift.

  • D1Multi-angle images per part under fixed lighting
  • D2The inspector's accept / reject and defect name
  • D3Corrections where he disagreed with the model
  • D4Batch, machine and tool context

Recorded on site and owned by the plant. The system uses it to keep learning your floor.

Architecture

  1. 01

    Capture

    Two industrial cameras, ring light, part-present sensor

  2. 02

    Edge

    Jetson Orin: inference, image store, offline queue

  3. 03

    Model

    Defect detector fine-tuned on the inspector's labels

  4. 04

    Record

    Per-part images, decision, defect code, batch

  5. 05

    Surface

    Bench display, reject gate signal, batch export

What we will measure

  • Escaped defects fall.

    PILOT MEASUREMENT REQUIRED: customer returns per 10,000 parts before and after.

  • The model agrees with the inspector.

    PILOT MEASUREMENT REQUIRED: precision and recall against his labels.

  • Inspection keeps pace with the line.

    PILOT MEASUREMENT REQUIRED: seconds per part at the bench.

Run a plant with an expert like this?

Tell us whose judgement your floor cannot run without. That is where a pilot starts.