Definition
Degree to which condition monitoring AI severity scores correspond to actual urgency across different asset types and failure modes.
The Measurement Gap
Condition monitoring platforms generate severity scores on a numerical scale, typically 1 through 10, and maintenance supervisors use these scores to prioritize responses. The implicit assumption is that severity means the same thing across all asset types: a score of 7 on a crusher bearing should carry the same urgency as a 7 on a SAG mill gearbox. In practice, this assumption is rarely validated. Legacy alarm management metrics count alerts and track response times, but they do not test whether the severity scale is consistent across equipment populations. Priority compliance metrics measure whether high-priority work orders were addressed within target timeframes, but they do not question whether the priority assignment was correct in the first place. Vibration trending and oil analysis reports track condition trajectories for individual assets without asking whether the thresholds that trigger alerts produce comparable urgency signals across different machine types. The result is that maintenance crews develop informal, experience-based rules about which severity scores to trust on which equipment, creating an undocumented, person-dependent translation layer between the AI output and the actual scheduling decision.
Decision Supported: When the AI says severity is 7 out of 10, should the maintenance supervisor shut down now, schedule for next shutdown, or keep monitoring, and does the answer depend on the asset type?
Formula
Plain Language
For each combination of severity score and asset type, calculate the median actual time-to-failure from the moment the alert was issued to the moment the component failed or was replaced. Compare that median to the time-to-failure that the severity score is supposed to represent. The closer the actual matches the expected, the higher the alignment score. Weight the overall score by the number of alerts in each combination so that common scenarios carry more influence.
Notation
SUAS(s, a) = 1 - |Median_TTF(s, a) - Expected_TTF(s)| / Expected_TTF(s)
SUAS_weighted = Sum(n_s,a x SUAS(s, a)) / Sum(n_s,a)
Variables
| Variable | Definition | Unit |
|---|---|---|
| s | Severity score on the condition monitoring platform's scale (typically 1 to 10) | Integer |
| a | Asset type classification (e.g., crusher, SAG mill, pump, conveyor) | Category |
| Median_TTF(s, a) | Median actual time-to-failure for alerts at severity s on asset type a | Hours |
| Expected_TTF(s) | The time-to-failure that severity level s is designed to represent, per vendor documentation or site calibration | Hours |
| n_s,a | Number of resolved alerts at severity s for asset type a | Count |
| SUAS(s, a) | Alignment score for a specific severity/asset combination | Ratio (0 to 1) |
| SUAS_weighted | Overall weighted alignment score across all combinations | Ratio (0 to 1) |
Proposed formula, pending validation pilots.