Anomaly detection

Detect What Normal Monitoring Can Miss

Fixed thresholds only fire once a problem is already serious. Baseline comparison identifies the drift that precedes it.

Request a Demo See Predictive Maintenance

The business problem

Thresholds are set for failure, not for change

A high-temperature alarm at 12°C tells you the chiller has already lost capacity. The three weeks of gradual rise beforehand generated no alarm at all, because every individual reading was inside limits. Threshold monitoring is necessary, but it detects events rather than trends.

Individual readings in range while the trend deterioratesAlarm fires only once service is affectedSlow drift indistinguishable from normal variationSeasonal and occupancy effects misread as faults

The IntelliGrid approach

Deviation measured against the asset’s own normal

IntelliGrid learns each asset’s normal operating envelope, including how it varies with load, weather and occupancy, then flags statistically meaningful departure from it. A reading well inside the alarm threshold can still be abnormal for that machine at that hour.

Baselines that account for load and seasonal variationStatistically meaningful deviation rather than fixed limitsMulti-signal anomalies where one reading alone is inconclusiveFindings scored and routed into the operational queue

Monitored signals

What can be watched

Coverage depends on the instrumentation present. Most deployments begin with the signals already available from the BMS and add IoT sensing where it matters most.

Temperature

Supply, return, ambient and bearing temperatures across plant and spaces.

Vibration

Signature change on rotating equipment including pumps, fans and motors.

Pressure

Differential pressure across filters, strainers and pumping systems.

Flow

Chilled water, condenser water and domestic water flow behaviour.

Electricity

Current, power factor and consumption patterns per circuit or asset.

Water

Consumption profile, including overnight flow indicating leakage.

CO₂ and air quality

Indoor air quality against occupancy and ventilation rates.

Humidity

Space conditions in humidity-sensitive environments.

Equipment runtime

Run hours, cycling frequency and duty distribution.

Security behaviour

Access and patrol patterns that depart from the expected routine.

Workflow

From normal to action

Normal Pattern Deviation AI Anomaly Risk Score Action

Business benefits

Problems visible weeks before a threshold alarm Fewer false positives from seasonal and load variation Water and energy waste detected from consumption behaviour Multi-signal confirmation before a technician is dispatched Findings that arrive with a proposed action