Tell us the problem you want to solve
Thank you. We have your message and will reply within one business day.
Near zero-defect quality at a tier-1 engine-block manufacturer
Our client is a foundry in the Southern Hemisphere that casts 129,000 iron automotive components a year, over 46,000 tonnes, at one plant. It had digitised production across the plant and wanted its engineers and managers to use the production and quality data to decide how to run the process.
About 1,000 parameters, and data in six formats
Casting an engine block runs through many steps, and about 1,000 parameters across the plant interact. The data on them sat in PLCs, the central SCADA system, Excel and CSV files, proprietary databases and handwritten forms.
When the plant had a good run, nobody could say which settings had produced it, so it could not be repeated on purpose.
One view of the plant, then prescriptions
One view of the data
We extracted the data from every department and loaded it into one warehouse: 15 months of production history, 173,000 records and 400 process variables.
Prescriptions every five minutes
DataProphet PRESCRIBE learned a model of the whole process from that history and found the operating region with the fewest defects. It sends operators setpoints every five minutes, updated from live process data, and shows each operator, engineer and plant manager the parameters that matter most for their part of the process.
After deployment
| Measure | Result |
|---|---|
| Scrap rate, first month of deployment | Halved |
| External scrap rate, within three months | Below 0.1%, ongoing |
| Defect rate, best periods | 0.5% for up to three months |
| Defect rate, long-term average | 40% lower |
| Cost | About $100k saved per month |
“We might have been able to achieve similar results in the past, but we had absolutely no clue what we did to achieve the good result. With artificial intelligence, we have a really good idea of what we need to do to improve production.”