Three families: equipment state (running, stopped, fault, cycle count), process parameters (temperature, pressure, speed, energy), and quality-to-batch linkage (work order, batch number, inspection result). Energy is typically sampled every 1–15 minutes, process parameters every 1–60 seconds depending on the machine, and equipment state on event. The interval follows the purpose — higher is not better.
Frequency is a cost decision. High-rate collection adds storage and network cost and a great deal of noise. For energy accounting and carbon conversion, 15 minutes is enough; for process traceability and fault diagnosis, seconds are required. Fix the purpose before fixing the tags.
Two collection paths. Machines with an interface (PLC, drive, meter) are read over Modbus or OPC UA; older machines get a meter, sensor or counter, and state is inferred from indirect signals such as current draw or rotation. In brownfield plants most of the effort sits in the second path.
Data quality beats data volume. Unsynchronised timestamps, mismapped tags and data loss on power interruption are the three recurring faults. Run two weeks of stability validation against manual meter readings before any analysis, or every later conclusion inherits the error.
Write the boundary down: collect only what relates to production, energy and quality, and stay out of personal data and unrelated systems. Putting that in the agreement protects the plant and lowers resistance on the shop floor.
| Equipment state | Running / stopped / fault / cycle count, event-triggered |
|---|---|
| Process parameters | Temperature, pressure, speed — 1–60 seconds |
| Energy | One reading every 1–15 minutes |
| Protocols | Modbus / OPC UA direct read, or retrofitted sensors |
| Validation | Two weeks against manual meter readings |