All case studies
Kellner Werke, Manufacturing
Predictive maintenance on machines older than the cloud
Retrofitted sensor data joins decades of ERP history in one lakehouse, so maintenance gets scheduled instead of triggered.
- less unplanned downtime
- A fifth
- Platform
- Apache Spark, Databricks, MongoDB, Docker
The situation
Production machines predated any notion of connected instrumentation. Maintenance was scheduled by calendar and performed by reaction, and the ERP history that would have explained failures was never joined to anything.
What we did
- Retrofitted vibration and temperature sensors on the highest-value assets first, rather than instrumenting the whole floor on principle.
- Joined the new telemetry to two decades of ERP work orders in one lakehouse, which made the failure history usable as training data for the first time.
- Scored assets daily and pushed the result into the existing maintenance planning tool, so no technician had to learn a new system.
Where it landed
Maintenance moved from calendar-driven to condition-driven on the instrumented assets. The rollout to the remaining floor is being run by the client's own engineering team.