Manufacturing Analytics on Databricks
Case Study
OEE and Predictive Maintenance on Databricks for an Industrial Manufacturer
A manufacturer running several plants across Europe had production data trapped inside each site. Every plant had its own historian, its own MES conventions and its own definition of downtime — so group-level questions about equipment effectiveness could only be answered with spreadsheets emailed between sites.
Dateonic connected plant-floor systems to a governed Databricks lakehouse, standardized the asset and downtime model across sites, and delivered comparable OEE metrics plus the clean sensor history needed for predictive maintenance.
Turnover
10+ months
Industry
Manufacturing
Technology:
Azure Databricks, Delta Lake, Unity Catalog
The challenge
- Every plant counted differently. Downtime categories and shift calendars varied by site, so OEE figures were not comparable.
- Plant-floor data stayed on the plant floor. Historian and MES data was used for local monitoring and rarely left the site.
- ERP context was disconnected. Production orders and material data lived apart from machine telemetry.
- Maintenance was reactive. Vibration and temperature data existed but was not retained in a form usable for modelling.
- No group-level visibility. Comparing line performance across plants was a manual exercise repeated every month.
Our approach
Common asset model. We built a hierarchy — site, line, machine, component — and mapped each plant’s local naming into it. Without this, cross-plant comparison is impossible no matter how good the pipelines are.
Standardized downtime taxonomy. Working with plant managers, we agreed one set of downtime reason codes and mapped every local category into it, so availability losses mean the same thing everywhere.
Integration by system type. Historian tags, MES events and ERP master data were each given an ingestion pattern appropriate to their volume and update behaviour.
The solution
Medallion architecture. Bronze holds raw historian, MES and ERP feeds; silver holds normalized machine states, production events and the common asset model; gold holds OEE and quality marts.
Standardized OEE. Availability, performance and quality are computed once, centrally, using the agreed taxonomy and shift calendars — so a number means the same thing in every plant.
Retained sensor history. High-frequency vibration and temperature data is stored in Delta with the resolution needed for condition modelling, rather than being downsampled and discarded.
Governance. Unity Catalog gives plant teams full access to their own site and controlled access to group-level comparisons, with lineage from every reported metric back to source events.
Terraform-managed environments. Dev, test and production are provisioned as code, with production jobs run by service principals.