A model-based approach to HVAC fault detection and diagnosis

This thesis develops and experimentally evaluates a sequential model-based detector for detecting actuator failures in HVAC systems.  Specifically, this thesis considers actuator failures which result in the actuator valve sticks in an unknown (but constant) position. A first order heat-equation model is assumed to model interactions between adjacent rooms, which is used to formulate a hypothesis testing problem assuming known inter-room thermal parameters. The detector is formulated to provide performance that asymptotically bounds both the probability of miss and the probability of false alarm. A wireless sensor network was deployed and employed in coordination with the campus SCADA system to both demonstrate the detector performance and identify model parameters.  Multiple testing scenarios are considered where the inter-room dynamics are physically altered (by opening and closing windows and doors).  The time required to accurately determine the inter-room thermal parameters is studied, where parameter identification is performed using a time-series least-squares approach. Results indicate that when the model parameters are accurately known, the detector performance is acceptable; however, inaccurate modeling assumptions lead to significant detector performance degradation.