Case study / Manufacturing / Worked example

Checking Every Part on a Moulding Line.

A manufacturer of moulded and trimmed parts supplies automotive customers from a plant near Sriperumbudur. Quality is checked by pulling parts from the line to a sample table. Most defects are caught, but some reach the customer, and each complaint costs a sorting visit. This is how we would check every part on the line with a camera.

Business
Moulded parts manufacturer, 3 lines
Runs on
Sample-table checks, paper records
We add
Line camera, trained defect model
Kept as is
Machines, inspectors, final audit
Vision inspection screen showing six moulded parts on a conveyor, four marked pass and two rejected for burr and short moulding
The inspection screen: each part is marked pass or reject with the defect type, and reject reasons are counted for the shift. Illustrative screen with sample data.

About this case study. This is a worked example, built from the kind of operation we work in. The business described is a composite and is not a named client, the screens show sample data, and we quote no result figures we cannot source. The commitments at the foot of the page are the ones that hold on every engagement.

01 / The situation

Where the Business Is Today

An inspector checks a sample from every batch against a limit board. The check is careful, but it covers a small share of what the line produces, and it happens after the parts are made.

When a trim die starts to wear, burrs appear slowly. The sample table notices some hours later. In between, the line has produced parts that will be sorted by hand, or found by the customer.

The inspectors know every defect by sight. That knowledge is exactly what a vision system needs, and it is the starting point of the work.

02 / The build

What We Build

The exact scope is fixed after two to four days on site. For an operation like this one, it usually comes down to these parts.

01

Camera and lighting on the line

An industrial camera and controlled lighting over the conveyor, with a small computer beside the line. Nothing on the machine is changed.

02

A model trained on your parts

We photograph good and defective parts from your own production, and your inspectors label the defects: burr, short moulding, missing hole, surface scratch.

03

Pass or reject for every part

Each part is checked as it passes. Rejects trigger a light, a buzzer or a diverter, depending on what the line allows.

04

Reject reasons by shift

Every reject is saved with its picture and reason, so the quality head sees which defect is rising and on which line.

05

Early warning

When one defect type crosses a set rate, the supervisor is notified to check the tool before a full batch is lost.

06

Trial before trust

The system first runs beside the sample table, and its decisions are compared with the inspectors' until the plant is satisfied.

How it fits together

  1. 01Camera and lighting over the line
  2. 02Computer beside the line runs the model
  3. 03Pass or reject signal to the line
  4. 04Reject pictures and reasons for the quality team
03 / The change

What Changes Day to Day

 TodayWith the system
How many parts are checkedA sample from each batchEvery part on the line
When a defect is seenAt the sample table, after the batchAs the part passes the camera
Tool wearNoticed when rejects pile upFlagged when one defect starts rising
Record of rejectsTally marks on paperPicture, reason, time and line
InspectorsSorting and samplingTeaching the system and auditing it
04 / The project

How the Project Runs

The same stages on every engagement. The full walk-through is on the How We Work page, and the wider service is described under Manufacturing software and factory AI.

What we commit to on every engagement
Discovery call30 min, free
Days on your floor before a proposal2–4
Scope, timeline and priceFixed
First working demo on your dataWeek 1
Typical time to productionWeeks 6–12
Existing systems replacedZero
05 / FAQ

Questions Owners Ask

How many sample images are needed?

It depends on the part and the defect. We start with what a few days of production provides, and add images of rare defects as they occur. The on-site visit is where we judge whether a defect is visible to a camera at all.

Will it work for all our part variants?

The model is trained per part family. A first scope covers the highest-volume parts on one line, and further variants are added with their own images.

Does it need an internet connection?

No. The model runs on a computer beside the line, and pictures are stored on a server inside the plant.

06 / More

More Case Studies

Does This Sound Like Your Business?

Tell us what your week looks like. Thirty minutes on a call is usually enough to tell whether there is something worth building together. The call is free, and we respond within one business day.