How should a plant define operating envelopes for maintenance alerts?
A vibration value that is unusual at steady load may be normal during startup, and a high temperature may be expected during a controlled process transition. Reliable condition alerts need operating context, not just one fixed threshold. Define the modes in which each measurement is meaningful and establish who can approve changes to those rules.
Define operating modes before choosing limits
For each asset class, identify stopped, starting, accelerating, normal load, reduced load, transient and shutdown conditions as relevant. Use actual control system states and verified process signals where possible. Specify minimum dwell times and data quality requirements. Separate equipment design or protective limits from advisory maintenance thresholds. Protective settings must remain under the approved engineering change process.
Build evidence based baselines
Use data from healthy operation at comparable speed, load, temperature and process conditions. Document the period selected, known maintenance events, instrument changes and excluded bad data. Where enough history exists, characterize expected variation for each operating mode. Do not use a generic number of standard deviations as an automatic alarm limit without checking the distribution, failure consequences and detection objectives.
Make alert logic understandable
- Identify the triggering measurement, its units and source tag.
- Record the mode and conditions in which the alert is valid.
- Specify persistence, hysteresis and suppression during known transitions.
- Define escalation by severity, criticality and corroborating evidence.
- Link the alert to a named reviewer and an approved maintenance investigation process.
Keep a record of rule versions and the reason for each change. An AI suggested threshold should be reviewed against engineering evidence before activation.
Test against real plant history
Replay known healthy startups, planned shutdowns, instrument faults and confirmed failure events. Count false alerts, missed events, duplicate notifications and time from detection to useful action. Check whether the alert would have been understandable to an operator on shift. Review the operating envelope after equipment modifications, process changes or sensor replacements. DOE guidance on energy management recommends defined operating and maintenance criteria for major systems, while NIST OT guidance emphasizes preserving system reliability and safety.
Example from an industrial plant
A pump bearing temperature alert triggers repeatedly during hot restarts. A review shows that the fixed threshold was calibrated only for steady load. Engineering creates a startup state with a short monitored transition, retains the existing protective trip settings and adds an escalation for sustained high temperature after the transition. Historical replay confirms fewer nuisance alerts without suppressing the known abnormal case.
Explore KANE consulting
Discuss practical maintenance and reliability improvements for your facility.
Request a consultation →Related answers: condition monitoring baselines · AI maintenance approval thresholds
Technical references
DOE Set Operating and Maintenance Criteria · NIST SP 800-82 Rev. 3 Guide to OT Security · DOE Operations and Maintenance Best Practices