Case study
Turkish Wheel Maker Cuts Scrap Cost with ML
How a Turkish wheel manufacturer used machine learning and prescriptive analytics to cut scrap and reduce material waste.
A descriptive dashboard tells you what already went wrong. The harder, more useful model tells a line operator which process variable to change before the next wheel gets scrapped.
AI & Generative AI Practice Lead, SquareShift
A leading wheel manufacturer in Turkey, supplying some of the world's largest automobile manufacturers.
About 8% of the wheels it produced were failing quality checks and had to be scrapped before this engagement.
- Early detection — Faulty wheels forecast earlier in the production line.
- Prescriptive guidance — Process variables tuned to reduce damaged wheels.
- Real-time visibility — Dashboards track line performance and process drift as it happens.
How SquareShift delivered it.
The challenge
A leading wheel manufacturer in Turkey, supplying some of the world’s largest automobile manufacturers, was scrapping roughly 8% of the wheels it produced because they failed quality checks — a direct hit to manufacturing cost that quality control alone wasn’t catching early enough.
The team needed to see problems forming on the line before a wheel became scrap, not just count the damage after the fact.
What we delivered
Using thousands of process variables collected and stored in an AWS data lake, SquareShift built real-time dashboards that track line performance, surface patterns over time, and flag process drift as it happens.
On top of that descriptive layer, SquareShift built machine learning algorithms to forecast faulty wheels earlier in the process, and prescriptive models that suggest which process variables to adjust to bring the damage rate down.
The payoff
The manufacturer now sees process drift and forecasted defects while a line is still running, instead of discovering the damage after the wheels are already scrapped.
Prescriptive recommendations give line operators a specific variable to adjust, turning a reactive quality-control process into one that catches problems earlier in the line.
Where this work sits