A stair-step probabilistic approach for automatic anomaly detection in building ventilation system operation

HVAC systems contribute to a large part of energy consumption in buildings and studies suggest that savings up to 30% can be achieved by utilising the potential of FDD methods which aim to identify faults and their root causes. In particular, model-based FDD are becoming more useful as the modelling and simulation of complex building systems have been eased due to advancements within the field. However, methods often lack the ability of effectively distinguishing between healthy and abnormal operation and some are highly subject to human evaluation. Bang et al. proposed a model-based fault detection method for automatic identification of abnormal energy performance on a daily basis in building ventilation units using a statistical definition of abnormality based on the Chernoff bound. The method enables the fault detection process to be automated which removes the need for human evaluation. However, the method is governed by linear interpolation leading to uncertain identification of abnormal operation and imprecise probability calculations, thereby triggering the need for modifications. This work upgrades the model-based fault detection method by introducing a stair-step approach to more accurately identify abnormal behaviour. The outcomes of the upgraded approach are reported for a case study building and evaluated in comparison with the original method. The improved method shows correct identification of abnormal periods and detected the precise day of a faulty occupancy counter. Moreover, it shows that the ascribed probabilities of the original approach are consequently lower for the two analysed ventilation units by an average of 13 and 15% points, respectively.