Definition
Percentage of known degradation mechanisms for a component or component population that the AI predictive monitoring system actively monitors, measured against the component's full failure mode profile from FMEA/RCM analysis.
The Measurement Gap
When AI predictive monitoring recommends extending a PM interval based on condition data, the recommendation's validity depends on whether the monitoring covers all the ways the component can fail. A vibration sensor detects bearing wear. A temperature sensor detects overheating. An oil analysis program detects lubricant degradation. But the component may also be subject to stress corrosion cracking, fatigue, or electrical insulation breakdown that none of these sensors detect. No legacy metric measures the coverage gap between what the predictive monitoring system actually monitors and the full failure mode profile from the component's FMEA or RCM analysis. PM completion rate tracks whether PMs get done. Equipment reliability metrics track failure rates. Maintenance Rule (a)(2) criteria track system performance. None of them measure whether the monitoring that justified the PM deferral covers the full set of degradation mechanisms the component is susceptible to. The invisible failure mode is that AI recommends extending a PM interval because the monitored parameters look healthy, while an unmonitored degradation mechanism progresses undetected.
Decision Supported
When the AI recommends extending a PM interval based on condition monitoring data, does the monitoring cover all the ways the component can fail, or only the mechanisms the sensors happen to measure?
Formula
Plain language: For each component where AI has recommended a condition-based PM interval extension, list every degradation mechanism from the component's FMEA or RCM analysis. Determine which mechanisms the predictive monitoring system can detect (based on the sensor types, data sources, and model capabilities). Divide detectable by total. The ratio tells you what fraction of the component's failure risk the AI is actually watching.
Notation:
DMMCR_c = DM_monitored_c / DM_total_c
Population composite:
DMMCR = Σ_c (w_c × DMMCR_c) / Σ_c w_c
Variables:
- DM_monitored_c = number of degradation mechanisms for component c that the predictive system actively monitors (count, a mechanism is "monitored" if the system has a validated detection capability for that mechanism type)
- DM_total_c = total number of degradation mechanisms identified in the FMEA/RCM analysis for component c (count)
- w_c = safety significance weight of component c (1.0 safety-related, 0.7 risk-significant, 0.4 BOP)
- c = each component where AI-recommended condition-based interval extension is active
Sign convention: DMMCR = 1.0 means every degradation mechanism is monitored. Lower is worse. A component with DMMCR of 0.4 has 60% of its failure modes unmonitored: the AI's "healthy" assessment is based on 40% visibility.
Aggregation period: Per PM interval review cycle (when the AI recommends extending or maintaining the interval).
Edge Cases
- Degradation mechanism has negligible contribution to failure risk: weight by mechanism risk contribution if quantitative FMEA data exists; otherwise count equally (conservative)
- Predictive model detects a mechanism indirectly (e.g., bearing wear detected via vibration, which also correlates weakly with electrical insulation degradation): count as monitored only if the indirect detection is validated. Correlation is not causation, especially in nuclear
- Component with no FMEA/RCM analysis: DMMCR is undefined. The AI shouldn't be recommending interval extensions for components whose failure mode profile hasn't been analyzed
- Minimum: evaluate for every component where AI-driven interval extension is active (no sampling. Census)
Proposed formula, pending validation pilots.