Predictive Maintenance and Anomaly Detection
In short
CAFMTEK's predictive maintenance uses machine learning to calculate a failure probability score for HVAC assets and alerts teams when the score passes a set threshold. Its anomaly detection compares live readings from electrical distribution boards against learned expected ranges, flags deviations and shows which parameter caused each anomaly. Predictive maintenance and anomaly detection are built into CAFMTEK as standard, not sold as a paid add-on.
Applies to: HVAC assets (predictive maintenance) and main distribution boards (anomaly detection) with BMS, smart meter or sensor data
Why condition-based maintenance
Scheduled maintenance happens at fixed intervals, whatever state the equipment is in. CAFMTEK’s predictive maintenance and anomaly detection base maintenance on how equipment is actually performing. Teams can step in before a likely failure, and avoid work that isn’t needed.
Predictive maintenance for HVAC assets
For HVAC assets, machine learning models estimate how likely the equipment is to fail.
- Evaluate: the model reviews performance trends, historical failure data and real-time metrics for each asset.
- Score: it calculates a failure probability score.
- Alert: when the score passes a predefined threshold, CAFMTEK triggers an alert.
Anomaly detection for electrical assets
For main distribution boards (MDBs), CAFMTEK detects unusual behaviour in live readings from smart meters and the BMS.
Parameters monitored
| Parameter | Unit |
|---|---|
| Total energy consumption | kWh |
| Average current | A |
| Average voltage | V |
| Total power | kW |
| Power factor | Calculated from total power, average voltage and average current |
How anomaly detection works
- Expected range modelling: the AI model uses historical data to set an expected operating range for each parameter. The ranges adjust for how the building normally consumes power, for seasonal variation, and for equipment type and load profile.
- Detection: when a real-time value moves significantly outside its expected range, CAFMTEK flags an anomaly.
- Root cause attribution: a weighted scoring algorithm measures how much each parameter contributed to the anomaly. Technicians can see which metric is driving the problem, so diagnosis is targeted.
Predicting the next failure of an MDB
Today, MDB failure prediction uses historical maintenance records. It looks for patterns such as the time between past failures and recurring component issues.
The prediction approach combines: [CONFIRM: live today or planned]
- Continuous monitoring through BMS and IoT platforms
- AI and machine learning models trained on historical failures from maintenance logs and on live BMS data
- Automated alerts and work orders through Fault Detection and Diagnostics when risk thresholds are crossed
Reliability metrics: MTBF and MTTR
CAFMTEK tracks two reliability metrics for each asset:
| Metric | What it measures | What a warning looks like |
|---|---|---|
| Mean Time Between Failures (MTBF) | Average time an asset runs between failures | Low MTBF means frequent failures. A falling trend can mean an asset is approaching failure faster than expected. |
| Mean Time to Repair (MTTR) | Average time taken to repair an asset | High MTTR means repairs take longer, so predicting failures early matters more. |
Predictive algorithms use MTBF trends to forecast when the next failure might occur, so maintenance can be scheduled before it happens.
How this works with Fault Detection and Diagnostics
- Predictive maintenance and anomaly detection show when an asset is at risk or behaving unusually.
- Fault Detection and Diagnostics turns threshold breaches into work orders and assigns them to technicians with the right skills.
Data requirements and limitations
- Accuracy depends on your data. Available data points vary from building to building. More data points and longer maintenance histories give more accurate predictions.
- Coverage today: predictive maintenance covers HVAC assets, and anomaly detection covers electrical distribution boards. [CONFIRM: any other asset types]
- [CONFIRM] Whether these modules are included in standard CAFMTEK or are add-ons.
Frequently asked questions
What is the difference between predictive and preventive maintenance?
Preventive maintenance follows a fixed schedule. Predictive maintenance uses equipment data to estimate when a failure is likely, so work can be timed to the asset’s actual condition.
Does predictive maintenance replace PPM schedules in CAFMTEK?
No. PPM schedules stay in place. CAFMTEK can suggest longer or shorter PPM intervals based on each asset’s failure history.
Why does the anomaly show which parameter caused it?
CAFMTEK’s root cause attribution scores how much each parameter, such as total kWh or power factor, contributed to the deviation. Technicians can then check the right component first.