Thermal
Intelligence
The ELPRO system unifies sensor data from Critical Rooms in SMRT, performs logic processing, and delivers operational insights into a single platform. It automatically tags anomalies, predicts failures, and opens avenues for energy optimization across the facility.
System Capabilities
Predictive Trending
The RANSAC algorithm predicts limit violations up to 14 days in advance. View these priority alerts in the Dashboard.
Smart Noise Filter
Advanced MAD (Median Absolute Deviation) logic filters out harmless sensor noise. Test sensitivity in the Tuning Lab.
Energy Savings
Identify rooms running significantly below target temperature.
Logic Pipeline
Raw Data
10-min intervals
Pre-Process
MAD Calculation
RANSAC Regression
Logic Gates
Check Traps (A-D)
Panic Switch Check
Classification
Assign Bucket
(1 to 5)
Impact Matrix
Critical Failure
Bucket 1
Predictive Alert
Bucket 2
Instability
Bucket 3
Efficient
Bucket 4
Daily Rhythm
Sensor Measurement
Cloud Ingestion
Logic Engine Batch
Report Refresh
Algorithm Glossary
Activation Logic (Stream A)
A short-term volatility audit. It compares 24-hour moving averages against a 7-day baseline. If the deviation exceeds 5% (while temp > 25°C), it triggers an "Activation". This tracks acute spikes that may not yet trigger a trend alert.
The "Panic Switch"
Standard trends use 14 days of history. However, if the slope exceeds 0.15 and temperature is above 25°C, the system discards history and calculates trend based only on the last 4 days to react faster to runaways.
The "Triple Net"
We use three distinct traps to catch instability:
• Trap B: High frequency noise (MAD).
• Trap C: Large amplitude swings.
• Trap D: Comb patterns (stable base, periodic spikes).
Cooling Veto
If a room is trending up in the long term, but the 6-hour short-term gradient is negative (cooling active), the system suppresses the trend alert. This prevents false positives during active recovery cycles.
Signal Generator
Simulates HVAC failure or gradual heating over 5 days.
Simulates "hunting" control loops or cycling compressors.
Random sensor jitter. Use MAD threshold to filter this out.
Real-time Visualization
Engine Result
Tuning Parameters
Global
Hard Limits
Instability
Legacy V1
Suppression
Trend Logic
Stream A: Activation Logic
Acts as an Event Audit for short-term spikes (24h vs 7-Day trends). Think of this as the "Check Engine Light"—a counter for recent, acute volatility.
Use the "Fortnightly Triggers" tab to view this history. High counts warrant investigation even if the room status is currently normal.
Stream B: Risk Logic
Determines Prioritized Status based on long-term (14-Day) analysis. Think of this as the "Warning Lights"—signaling overall priority and systemic risk.
Classification Categories (Buckets)
1. Trending Rooms (Highest Priority)
Sensors showing a clear path toward failure or that have already failed.
- Limit Exceeded: Temp ≥ 28.0°C.
- Predictive Forecast: RANSAC predicts failure within 14 days.
- Panic Switch: Slope > 0.15 and Temp > 25.0°C.
2. Oscillating Rooms (Control Issues)
Sensors exhibiting "hunting" or instability.
- High Variance: MAD > 0.4 & Crossings > 1.
- Large Amplitude: Swing > 5.0°C.
- Rapid Fluctuation: Swing > 3.0°C & Crossings > 3.
3. Low Zone Rooms (Efficiency)
Stable, cold rooms consistently below the Energy Target of 20.0°C.
Navigating the Power BI Tabs
| Tab Name | Risk Category | Description |
|---|---|---|
| Overview | Summary | Complete distribution of all categories. |
| Fortnightly Triggers | Stream A | Audit count of short-term volatility events. |
| Trending Rooms | Stream B (High) | Sensors predicted to fail or currently exceeding limits. |
| Oscillating Rooms | Stream B (Med) | Sensors confirmed unstable due to multi-factor patterns. |
| Low Zone Rooms | Efficiency Status | Sensors running efficiently below 20.0°C. |