A plant that finally saw itself.
OPC UA + MQTT + a unified namespace turned a 24-machine line from "we think we're running well" into "we know we are".
A precision-components plant had millions of rupees of equipment and zero idea what its actual OEE was. The supervisor's whiteboard said one thing; the operators said another; the customer's quality complaint said something else entirely. We instrumented the floor, built a unified namespace, and put the truth on a screen.
What changed in the months after we shipped.
Equipment everywhere. Data nowhere.
The plant ran 24 machines across three lines — Siemens PLCs, Rockwell drives, a couple of older units running Modbus. The numbers people argued about (OEE, throughput, downtime) were collected by hand at end of shift, on a clipboard.
Quality issues that originated in machine-state changes only surfaced in customer complaints two weeks later. Maintenance was calendar-driven, which meant either too early or too late, never on time.
- No real-time visibility into machine state
- OEE calculated end-of-day, manually, with errors
- Customer-side defects untraceable to floor events
OPC UA on every controller. MQTT to the cloud. Unified namespace in the middle.
We installed an industrial edge gateway per line, exposing every PLC over OPC UA — Siemens S7, Rockwell EtherNet/IP, and the Modbus units, all unified. Every signal got a canonical address in a unified namespace.
An on-prem MQTT broker (Sparkplug B) carries the data with sub-second latency. A live dashboard shows OEE per machine, throughput, downtime reasons, and predictive alerts. Anomaly detection runs at the edge — vibration models that catch a bearing failing before it stops the line.
- Edge gateways with protocol bridging
- Sparkplug-B MQTT for sub-second telemetry
- OPC UA unified namespace per ISA-95
- Predictive maintenance ML on edge compute
OEE went up. Downtime came down. Quality stopped being a mystery.
Six months after rollout, OEE on the primary line moved from 62% to 84%. Unplanned downtime dropped 40%. Quality complaints became traceable — when a defect hits the customer, the plant can pull machine state for that exact batch in under a minute.
Maintenance shifted from calendar-based to condition-based. The bearings that used to fail in the middle of a shift now get replaced during planned downtime.
We finally know what our floor is doing. The customer complaints stopped feeling random.
What we reached for.
