Case Study

The Client

An industrial plant whose production line relied on sampled, visual quality control. Defects were often caught after a large batch had already been produced, leading to rejects and rework.

Goals

Inspect every piece rather than a sample; run the check on the line without slowing it down; and keep the system working reliably without a permanent internet connection.

Challenges

The production cycle left no time to ship images to a remote server — the decision had to be made locally in milliseconds. The environment had dust, vibration and changing light levels, all of which affect visual inspection.

Our Solution

Processing on the device: The detection model runs locally on the line. No image leaves the plant, and there is no network latency or per-call cost.

PCB designed for the environment: We designed a board for the real conditions of the line — vibration, temperature, dust — with the industrial interface the existing equipment required.

Stabilised imaging: Controlled lighting and calibration so detection is not thrown off by changing light in the building.

Immediate alerting: When a defect is found the system marks the piece and alerts the operator there and then, not at the end of the shift.

Implementation

We catalogued the defect types together with the QC staff and collected samples for training.

We designed and tested the board under the real conditions of the line before installation.

We ran the system alongside the existing sampling process so results could be compared.

We handed the staff a documented retraining process for when a new product type is added.

Results

Next Steps

Extending into predictive maintenance by monitoring equipment vibration.

Feeding inspection data into a production dashboard for trend analysis.