Predictive maintenance caught a fault nobody could hear

Open the full-size figure (new tab)

Measured consumption against the forecast for an air compressor. A work order is created at the third deviation above the threshold, several weeks before consumption has tripled. Illustration using synthetic data.

Comparing measured power consumption with a forecast based on normal operation reveals gradual faults weeks before they cause downtime. The alert goes straight into Dalux FM as a work order.

A gradual fault goes unnoticed in everyday operations

One of two air compressors in Veterinærbygningen developed a leaking seal. The whining sound in the plant room grew a little louder each week, and power consumption rose from around 25 kW to around 75 kW. Nobody noticed, because each day seemed much like the last. A fixed alarm threshold would only have been triggered once the damage was done.

A forecast instead of a threshold

The analysis uses time series already held in the Smart Campus platform. A period of normal operation provides a baseline. This is used to calculate an hourly forecast incorporating the machine’s weekly pattern. Hours when the measured value exceeds the forecast plus a threshold are counted as deviations. When more than two deviations occur on the same day, the analysis creates a work order in Dalux FM, and the operations team receives the alert where they already work. Only one work order can be open at a time.

The analysis runs as an Azure Function.

Source: Messom Digital, ‘Predictive maintenance at NMBU’ (in Norwegian)