How a rotor line, a traction-inverter line, an assembly line and a low-volume hypercar cell were unified into a single live namespace — and what a manufacturer can take away from it. The plant is simulated; the pipeline carrying its data is not.
The problem
Shopfloor data is born fragmented.
A typical powertrain plant runs equipment from a dozen suppliers, each with its own controller, its own data format and its own island of visibility.
Station states live in PLCs. Quality measurements live in test-bench files. Downtime reasons live in an operator's head or a paper log. Answering a basic question — "what is my OEE right now, and which station is losing it?" — means stitching exports together days after the fact. By the time the answer arrives, the shift that caused it has gone home.
This reference plant was built to demonstrate the alternative: every station, every part and every event flowing through one open, queryable pipeline — visible seconds after it happens.
Data islands
Each line, cell and test bench keeps its own records in its own format — nothing joins up across the plant.
Insight arrives late
KPIs assembled from end-of-shift exports describe yesterday's problems — too late to act on today's.
No genealogy
When a defect surfaces, tracing one serial back through stations, measurements and material lots takes days.
The architecture
Unified namespace, one broker, one historian.
Open protocols at the edge, an MQTT unified namespace through a MonsterMQ broker, an indexed PostgreSQL time-series store — and live consoles reading from it in real time.
One namespaceEvery station publishes status, measurements and transactions to structured topics
One historianAll events land in PostgreSQL with partial indexes on (topic, time)
Read-only by designConsoles observe the line — control stays at Level 1 where it belongs
Replaceable layersAny layer can be swapped for another tool speaking the same open protocol
The numbers
What the pipeline sustains, around the clock.
17+
Stations across 3 lines, plus a 15-station hypercar cell
Simulated plant · real pipeline · figures reproducible from the running system
The takeaway
What a manufacturer would take from this.
The unified namespace pays for itself first.Once every station publishes to one structured topic tree, every new consumer — OEE, SPC, andon, genealogy — is a query, not an integration project.
A plain relational historian goes further than expected.With targeted indexing and hot caching, PostgreSQL serves live dashboards over millions of daily events — no specialized time-series product required to start.
Read-only Level 2 de-risks adoption.A visibility layer that never writes to the line can be deployed alongside production with zero process risk — and still feeds a MES, or replaces spreadsheets, on day one.
The pattern scales down as well as up.The same pipeline serves a 6-station line and a 15-station low-volume cell without modification — proof the architecture, not the plant size, is what matters.
The case study is running right now.
Open the live dashboard and watch the plant described above — or talk to us about applying the same architecture to a real one.