The challenge
Plants ran separate historians with no enterprise view. Maintenance was calendar-based, and a single extrusion line failure could cost $180,000 in scrap and lost output.
What we did
Tulan built Eventstreams from plant historians and MQTT brokers into Eventhouse, handling 2.4 million events per second at peak. Anomaly detection models trained on three years of failure history ran on the stream, and Activator rules routed alerts to maintenance crews in Teams with recommended actions.
The outcome
Crews now get a median 40 minutes of warning before a likely failure. Unplanned downtime fell 34 percent in the first year, and the enterprise operations center watches all 22 plants live.
“The line tells us it's about to fail, and we fix it on our schedule instead of its schedule.”
VP, Global Operations Excellence