Automated fault detection and diagnosis of HVAC subsystems using statistical machine learning

The faulty operation of Heating Ventilation and Air Conditioning (HVAC) systems in commercial buildings can waste vast amounts of energy, cause unnecessary CO2 emissions and decrease occupant thermal comfort, reducing productivity.   We propose a new method of automating Fault Detection and Diagnosis (FDD), based on the modelling of operational faults in HVAC subsystems, using techniques from statistical machine learning and information theory.  Discovery of interrelationships between groups of sensors by analysing the level of Information Transfer present can help fine tune the simulation inputs and improve model accuracy. We present results of the detection and diagnosis of faults from an occupied commercial office building in Newcastle, Australia and using data from the ASHRAE 1020 fault detection project (Norford, Wright et al. 2000).