Definition
Cumulative count of AI-recommended inspection deferrals on a corrosion circuit, weighted by the AI's own reported uncertainty in the remaining-life estimate and the elapsed fraction of the original inspection interval: detecting whether a sequence of individually rational deferrals is accumulating circuit-level risk that no single decision authorized.
The Measurement Gap
AI inspection optimization makes one deferral decision at a time. Each decision is defensible: the model says remaining life is adequate, the operator agrees, the next interval is scheduled. What no decision authorizes, and no legacy metric measures, is the compounding effect of many deferrals on the same circuit. Inspection adherence rate measures whether scheduled inspections happened; it goes up when deferrals are approved. Mean time between inspections is gameable for the same reason. Corrosion rate tracks damage but lags by inspection cycles. RBI program compliance reports whether the program ran, not whether the program's assumptions remained valid after the AI started revising intervals. None of these metrics has a notion of cumulative uncertainty: the property that grows each time the model's prediction window is extended without empirical verification.
Decision Supported:
Has AI-driven inspection optimization deferred enough inspections on this circuit that cumulative uncertainty exceeds the risk tolerance, and if so, which circuits need immediate inspection regardless of the AI's current recommendation?
Formula
Plain language: For every deferral the AI has approved on this circuit since the last physical inspection, multiply the AI's reported uncertainty by the fraction of the original interval that has elapsed. Sum across all deferrals. The result is in uncertainty-years.
Notation:
IDRA = Σ ( σ_RUL_d × t_since_last_inspection_d / t_interval_original_d )
Variables:
| Variable | Definition | Unit |
|---|---|---|
| σ_RUL_d | Standard deviation of the AI's remaining-useful-life prediction at the time of deferral d (the model's own reported uncertainty) | years |
| t_since_last_inspection_d | Time elapsed since the last physical inspection at the point of deferral d | years |
| t_interval_original_d | Original (pre-AI) inspection interval for this circuit (per API 510/570/653 or RBI program) | years |
| d | Index of deferral event on the circuit since the last completed physical inspection | n/a |
IDRA resets to 0 for the circuit on completion of a physical inspection.
Proposed formula, pending validation pilots.