Model,ClusterandCompare(MCC)-ClusterandCompare(.PDFVIP

Data Driven Investigation of Faults in HVAC Systems with Model, Cluster and Compare (MCC) Balakrishnan Narayanaswamy?, Bharathan Balaji?, Rajesh Gupta?, Yuvraj Agarwal?? ?University of California, San Diego ?Carnegie Mellon University ?{muralib, bbalaji, rgupta}@cs.ucsd.edu, ?yuvraj.agarwal@cs.cmu.edu Abstract 1 Introduction The complexity of modern HVAC systems leads to de- Buildings are large and complex systems with thousands vice mis-con?guration in about 40% of buildings, wasting of moving parts including fans, water pumps, dampers and upto 40% of the energy consumed. Fault detection methods many more [2]. Improvements in sensing, communication generate excessive alarms leading to operator alert fatigue, and control infrastructure offer hope that increased instru- faults left un?xed and energy wastage. Sophisticated fault mentation and ?ne grained control will improve operational detection techniques developed in the literature are seldom ef?ciency [19]. However, this also increases the number of used in practice. We investigate this gap by applying vari- devices that are a part of the HVAC system and its complex- ous fault detection techniques on real data from a 145,000 ity. This increase in complexity also increases susceptibility sqft, ?ve ?oor building. We ?rst ?nd that none of these al- to faults and con?guration errors. In total, it is estimated that gorithms are designed to capture control loop con?guration 40% of buildings have mis-con?gured devices [20] and that faults. We develop a novel algorithm, Model, Cluster and there is a 40% energy savings potential from ?xing building Compare (MCC) that is able to detect anomalies by automat- faults [22]. In addition to energy waste, Heating, Ventilation ically modeling and clustering similar entities in an HVAC and Air Conditioning (HVAC) faults also lead to occupant sys