A repeated alarm may be noise, a real defect or poor configuration. Its occurrence count alone cannot establish which one indicates failure. Put the signal in the context of operating conditions, production events and maintenance work. Crucially, analyzing history to improve maintenance does not authorize deleting a safety alarm.
Build a consistent timeline
Collect asset identifiers, timestamps, operating conditions and failure events. Check time zones and system synchronization before concluding that a signal preceded downtime. Distinguish startup, stable running, normal shutdown and fault shutdown. A startup spike may be normal for one system and abnormal for another.
Move from ranking to a technical hypothesis
- Rank recurring, persistent and clustered alarms around downtime.
- Compare failure and non-failure periods to see whether the signal is equally common during healthy operation.
- Check a plausible physical mechanism with operations and maintenance.
- Connect findings to work orders and field observations without confusing correlation with cause.
- Define the action, owner, escalation criteria and how effectiveness will be verified.
Fictional example: temperature before downtime
Temperature rises appear before several motor shutdowns. Before adopting an indicator, the team checks whether episodes also coincide with normal high load, a faulty sensor or blocked ventilation. If inspections confirm a cooling defect, the signal may support a defined maintenance task. Follow-up then compares defect frequency, false positives and downtime over comparable exposure.
What the ArcelorMittal Mines case publishes
Vizory’s published case reports more than $3 million in recurring gains from improved shovel availability driven by alarm analysis. That claim belongs to that specific engagement. The page describes the problem and solution but does not publish a raw history here to reproduce every step. Inventing a client timeline or presenting the gain as typical would therefore be misleading.
Alarm changes require a separate framework
HSE states that an alarm should be useful, relevant and associated with a defined response and sufficient time to act. Changing priority, threshold, delay or suppression can affect that role. Changes must follow site management-of-change procedures and competent review. Maintenance analysis does not replace protection studies or safety validation.
Measure value, not just a lower count
Track signals that trigger useful action, confirmed defects, response time and false positives. Also examine missed defects. A lower alarm count is not evidence of lower risk if essential signals have disappeared. A good program improves decision quality and records what changed, why and with what approval.