Definition
Measures the percentage of rack positions experiencing inlet temperatures that deviate from facility-average thermal performance by more than the defined threshold for more than the defined persistence window during AI-optimized cooling operation.
The Measurement Gap
AI cooling optimization systems adjust chiller setpoints, CRAH fan speeds, and airflow distribution at minute or sub-minute intervals to minimize facility-level energy consumption. This produces measurable PUE improvement. What PUE cannot reveal is whether the optimization is thermally uniform across all rack positions or whether it is silently concentrating thermal stress in specific locations. A cooling AI that reduces average inlet temperature by 1°C while allowing three rack positions to run 8°C above the recommended range is improving PUE while accelerating equipment degradation in those positions. The consequence (elevated hardware failure rates in specific racks, potentially weeks or months after the operating condition began) is entirely invisible to facility-average metrics.
Decision Supported:
Is the cooling AI creating localized thermal stress that accelerates equipment degradation in specific rack positions, even while facility-level PUE improves?
Formula
Plain language: Count the percentage of rack positions where inlet temperature exceeds the facility average by more than 3°C for more than 25% of the measurement period. Separately compute average excess temperature across those positions as a severity companion.
Notation:
TMDI = (Rack_positions_with_deviation_above_threshold / Total_monitored_positions) × 100
TDS = (1/N) × Σ max(0, T_inlet_i − T_recommended_max) for all positions i with deviation
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Rack_positions_with_deviation_above_threshold | Positions where (T_inlet_actual − T_inlet_recommended) exceeds facility-average deviation by more than Δ°C for more than P% of the measurement period | Count |
| Total_monitored_positions | All rack positions with inlet temperature sensors | Count |
| Δ | Deviation threshold above facility average | °C (proposed: 3°C) |
| P | Persistence threshold | % of measurement period (proposed: 25%) |
| T_inlet_recommended | ASHRAE A1 recommended inlet range upper bound | °C (27°C) |
| TDS | Thermal Deviation Severity companion metric | °C average excess |
| N | Number of positions with deviation above threshold | Count |
| T_inlet_i | Actual inlet temperature at position i | °C |
| T_recommended_max | ASHRAE A1 recommended maximum inlet temperature | °C (27°C) |
Proposed formula, pending validation pilots.