Definition
Proportion of predictive maintenance interventions that address a recurring failure pattern without root-cause investigation.
The Measurement Gap
Predictive maintenance excels at detecting degradation and triggering timely replacement, but this capability creates a subtle problem: when the system reliably predicts and replaces a failing component, there is no burning platform for root-cause investigation. The component gets replaced on schedule, the asset returns to service, and the predictive system resets its monitoring baseline. Legacy metrics reinforce this cycle. Schedule compliance improves because replacements happen on time. Availability stays high because the predictive system catches degradation before failure. Component life tracking shows consistent intervals, which appears to validate the maintenance strategy. Repeat failure metrics in most CMMS/EAM configurations count consecutive failures on a single serial-numbered asset, missing the pattern when the same failure mode recurs across an asset class or fleet. Defect elimination programs rely on someone noticing the pattern and initiating an investigation, but predictive maintenance removes the pain signal that would trigger that investigation. The cost of this masking effect accumulates silently: the organization pays for recurring component replacements, each individually justified by the predictive model, while a one-time root-cause fix could eliminate the pattern entirely.
Decision Supported: Is the predictive maintenance program replacing components on schedule while a chronic design, installation, or process root cause persists, and how much is the recurring replacement costing compared to a one-time root-cause fix?
Formula
Plain Language
Identify all distinct combinations of asset class, component type, and failure mode that have had at least one AI-triggered replacement. For each combination, count how many AI-triggered replacements occurred within a rolling 24-month window. Any combination with three or more replacements in that window is classified as a "masked" pattern, meaning the predictive system is managing a recurring problem rather than preventing it. Divide the number of masked combinations by the total number of AI-intervention combinations to get the masking rate. Separately, for each masked combination, calculate the cumulative replacement cost and subtract an engineering estimate of what a one-time root-cause fix would cost, to quantify the recurring cost penalty.
Notation
PMR = (Masked_combinations / Total_AI_intervention_combinations) x 100
Recurring_cost = Sum(Total_replacement_cost per masked combo) - Sum(Estimated_single_RCA_fix_cost per masked combo)
Variables
| Variable | Definition | Unit |
|---|---|---|
| Masked_combinations | Distinct (asset_class, component, failure_mode) combinations where AI-triggered replacements occurred 3 or more times within 24 months | Count |
| Total_AI_intervention_combinations | All distinct (asset_class, component, failure_mode) combinations with at least one AI-triggered intervention | Count |
| Total_replacement_cost | Cumulative parts, labor, and downtime cost for all replacements within a masked combination | Currency |
| Estimated_single_RCA_fix_cost | Engineering estimate of the one-time cost to address the root cause for a masked combination | Currency |
| Recurring_cost | Aggregate cost penalty of continued replacement versus root-cause elimination | Currency |
Proposed formula, pending validation pilots.