Fleet Alarm Correlation Accuracy (FACA)

GMI-RE-015Renewable EnergyIntegration-RequiredPattern O-4: Alert Quality / Signal-to-Noise Ratio

Definition

Accuracy of AI alarm correlation, assessed by sampling resolved alarm groups for whether all grouped alarms shared a root cause and no related alarm was left out. Tracked alongside over-grouping, which hides distinct faults inside one incident.

The Measurement Gap

Fleet-scale renewable operations generate alarm volumes that no human team can process individually. Correlation engines group related alarms into event clusters, presenting operators with a single "fleet event" instead of 200 individual alarms. This is essential for manageable operations, but it introduces a new failure mode: incorrect correlation.

Over-grouping occurs when the correlation engine lumps unrelated site-specific problems into a single fleet event. A vibration alarm at Site A and a temperature alarm at Site B both fire during a heat wave. The engine groups them as "fleet thermal event" and resolves them together. Site A's vibration alarm was actually a bearing problem unrelated to temperature. When the fleet event is marked "resolved" because Site B's temperature returned to normal, Site A's bearing problem is lost in the closed event record.

Under-grouping is the opposite: related alarms across sites are not recognized as sharing a common cause. A firmware bug affecting inverters from the same manufacturing batch causes voltage fluctuations at three sites. If the correlation engine treats these as three independent events, the diagnostic effort is triplicated and the root cause (batch-specific firmware) takes longer to identify.

The measurement gap is that correlation accuracy is never systematically validated after events are resolved. FACA provides that validation by retrospectively assessing whether each correlated group was correctly assembled. The decision FACA supports is whether the fleet alarm correlation engine is creating accurate event pictures or merging unrelated problems into fleet events that get resolved while individual site issues persist.

Formula

Sample correlated alarm groups after resolution. For each group, assess whether the grouping was correct: all alarms shared a root cause, and no related alarms were left out. Compute the accuracy rate. Separately track the over-grouping rate (unrelated alarms merged, hiding problems) and the under-grouping rate (related alarms fragmented, duplicating response).

Formula (Notation):

FACA = Correctly_Correlated / Total_Sampled
Over_Grouping_Rate = Over_Grouped / Total_Sampled
Under_Grouping_Rate = Under_Grouped / Total_Sampled

Variable Definitions:

Variable Definition
Correctly_Correlated groups where all alarms shared a root cause and no related alarms were excluded
Over_Grouped groups that included alarms from unrelated causes
Under_Grouped groups that missed related alarms that should have been included
Total_Sampled number of correlated groups reviewed in the measurement period

Edge Cases and Measurement Boundaries:

When the root cause is unknown after resolution, classify the group as "unassessable" and track separately rather than including it in the accuracy calculation. Single-alarm events where no correlation was applied should be excluded from FACA. Require a minimum sample of 25 correlated groups per quarter before computing FACA to ensure statistical reliability.

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Higher; more accurate grouping
Reporting Frequency
Quarterly
Category
Alarm Intelligence and Operational Awareness
← Alarm Suppression Signal Loss Rate (ASSLR)Sensor-Asset Disambiguation Ratio (SADR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment