Anomaly detection for Building Service Components using performance data

The efficient operation of building systems is important energy efficiency, comfort and safety. Determining when maintenance is required or when a fault has occurred is the focus of this work. We show how to use available performance data in a methodology for improved maintenance scheduling through anomaly detection. We apply two statistical prognostic techniques – Particle Filters and Gaussian processes – to sensed data from two HVAC components to illustrate the methodology. We demonstrate that both methods identify occasions when maintenance should be carried out for one of the components, but are less successful on the second component.