Direct answer
Start with decisions the data must support. Define good count, reject count, running, blocked, starved, planned stop and unplanned stop consistently; capture the earliest causal event; reconcile machine and production totals; and review a small controlled reason-code list before adding more sensors.
Define the business questions
Decide whether the data must find bottlenecks, reduce micro-stops, verify output, plan maintenance or support traceability. Each purpose needs different resolution and retention.
Avoid collecting every PLC tag without an agreed use; volume can obscure the few signals that explain loss.
Create consistent machine states
Define running, ready, blocked, starved, fault, cleaning, changeover, planned stop and manual mode. Agree how a line-level state is derived when machines disagree.
Capture the first causal event before secondary alarms cascade through the line.
Control reason codes and manual entry
Use a short hierarchical list that operators can apply reliably. Review “other” and missing codes and retire categories that do not lead to action.
Where automatic classification is possible, still allow controlled correction with an audit trail.
Reconcile counts and validate data
Compare machine counts, packaging issued, good output, rejects and warehouse receipts. Test sensor double-counting, reverse movement, rework and line restarts.
Treat data-quality failures like instrument failures; record them and prevent unsupported conclusions.
Information to prepare
- Decisions and KPIs the data must support
- Machine-state definitions
- Good, reject and rework count points
- First-fault and reason-code method
- Timestamp and clock synchronisation
- Data validation and reconciliation tests
- Ownership, retention and access control
Automation Machinery
Production controls and automation routes for machine-state and line-data integration.