Definition
Frequency with which AI-generated maintenance schedules create concurrent equipment out-of-service configurations that individually pass Maintenance Rule (a)(4) risk assessment but collectively produce risk states in the upper quartile of the plant's historical risk distribution.
The Measurement Gap
AI scheduling optimizers balance resource utilization across maintenance crews, tool availability, and outage windows. The optimal schedule from a resource perspective may concentrate maintenance activities in ways that create elevated concurrent equipment out-of-service configurations. Each individual maintenance activity passes the Maintenance Rule (a)(4) risk assessment, but the concurrent configuration may produce a risk state that individual assessments do not capture. No legacy scheduling metric measures the frequency with which AI-generated schedules create elevated-risk concurrent configurations. Schedule compliance tracks whether work gets done on time. Resource utilization tracks crew efficiency. The online risk monitor calculates instantaneous risk but does not track the scheduling system's tendency to create elevated-risk configurations. The invisible failure mode is that the AI scheduler systematically favors resource-dense schedules that concentrate maintenance activities, producing elevated concurrent risk more frequently than manual scheduling did, and no metric detects the pattern because each individual activity passes its risk screen.
Decision Supported
Is the AI scheduler systematically producing risk-dense schedules because resource optimization favors concentrating maintenance activities, or does it distribute maintenance to maintain low concurrent risk?
Formula
Plain language: For each AI-generated weekly schedule, compute the peak plant risk state (using the online risk monitor or (a)(4) risk assessment tool) for the concurrent maintenance configuration the schedule creates. Classify each peak state against the plant's historical risk distribution: low (below 50th percentile), moderate (50th-75th), elevated (75th-90th), high (above 90th). Count the fraction of scheduled weeks where the peak state is elevated or high.
Notation:
CMRDI = Weeks_elevated_or_high / Total_weeks
Variables:
- Weeks_elevated_or_high = number of scheduled weeks where the AI-generated concurrent maintenance configuration produces a peak risk state above the 75th percentile of the plant's historical risk distribution (count)
- Total_weeks = total weeks in the measurement period (count)
- Risk state computed from the plant's online risk monitor or (a)(4) risk assessment tool for the AI-recommended maintenance configuration
- Historical risk distribution = 2+ years of pre-AI weekly peak risk states (establish baseline)
Sign convention: CMRDI = 0.0 means no elevated-risk weeks. Higher is worse. By definition against a historical distribution, the expected baseline rate is 0.25 (25% of weeks in the 75th+ percentile). CMRDI significantly above 0.25 means the AI scheduler is creating risk-dense weeks more often than historical practice.
Aggregation period: Monthly rolling.
Edge Cases
- Risk monitor model limitations (not all concurrent configurations modeled): use bounding estimates for unmodeled configurations. Do not assume zero risk increment
- Emergent work changing the AI schedule mid-week: evaluate both the as-planned and as-executed configurations. CMRDI measures the scheduling tendency, not the execution reality
- Plant in an outage: exclude outage weeks (outage risk management follows different logic and is covered by OAWCI)
- Minimum: all online operation weeks (census, no sampling)
Proposed formula, pending validation pilots.