The challenge
Surface defects such as scratches, slivers and roll marks were being detected by inspectors at the exit of the mill, inconsistently and too late to prevent downgrading of entire coils.
Our approach
We engineered the optics, lighting and camera mounting for the mill environment, built an edge-computing pipeline and trained a deep-learning classifier on labelled production images. The system integrates with the Level 2 automation to tag defect positions on the coil map and trigger alarms.
The outcome
Defect detection rate and consistency improved markedly, coil downgrading fell, and the client extended the system to the galvanising line.
Representative engagement. Client identity and certain project details are withheld or generalised under confidentiality agreements.