One approach to the optimization of building heating, ventilation and air-conditioning (HVAC) systems involves the use of intelligent agents. These Agents can be used for minimizing energy consumption, maximizing occupant comfort, fault detection, performing diagnostics, and improving system commissioning. Intelligent agents know or can learn the performance and status of the systems and equipment they monitor and can communicate and collaborate with other agents to achieve a common goal. In a previous study (Kelly 2011) (Kelly 2012), a simulation test bed was developed and used to study how agents can learn and interact to minimize energy consumption. This learning process, called system identification (Ljung 1987), makes intelligent agents ideally suited for detecting and identifying faults in building HVAC systems. This paper lays the ground work for using intelligent agents for fault detection in HVAC applications involving air handling units (AHUs). Specifically, it looks at which faults in AHU mixing boxes and cooling coils can be detected and under what conditions. Such an understanding is necessary before intelligent agents can be developed, tested, and implemented in real building HVAC systems. A simple mixing box model (Tan 2006) and a simple cooling coil model (Wang 2004) are used to represent the performance of a typical AHU, and system identification is performed using piece-wise linear approximations. Questions addressed include which variables should be used in the identification process, how many linear segments should be used for each variable, how large a fault must be in order to be detected given the inherent inaccuracies in the piece-wise linear approximation process, and over what range of external variables (e.g., environmental conditions) is fault detection possible.
Citation: Technical Note (NIST TN) - 1831
NIST Pub Series: Technical Note (NIST TN)
Pub Type: NIST Pubs
intelligent agents, building automation and control, energy management, fault detection and diagnostics