Residential HVAC fault detection using a system identification approach

Abstract Faults in building HVAC systems can have a significant impact on system efficiency, energy consumption and occupant comfort. This paper introduces a new, data-driven automated building HVAC fault detection method that uses a recursive least-squares model approach. System identification is performed using synthetic time-series data from an advanced residential building simulation program. To produce the model, only data on indoor and outdoor air temperatures are required. Model parameters are then observed in real time. During normal system operation, these parameters converge to stable values. Faults can be detected when the model parameters deviate from their converged values. The new fault detection approach has the distinct advantage of being computationally efficient while not requiring detailed building and HVAC models. These features make the new fault detection approach tractable between large and small HVAC systems.