Smart building applications rarely experience widespread adoption due to the prohibitive cost of porting them to different buildings. We created a schema, called Brick, that defines an ontology for sensors, subsystems and relationships among them. We demonstrate the effectiveness of Brick by using it to represent six diverse buildings, comprising of 17,700 data points, and running eight unmodified applications on these buildings.
Research has shown that contemporary thermostats are inadequate to keep occupants satisfied in offices as they can be difficult to use and provide insufficient control. We provide occupants with a personalized web portal through which they can view HVAC status, change temperature settings and send feedback to building manager. We study the impact of use of our software thermostat by 220 registered occupants across 21 months.
Sensors already deployed in the field do not have a standard naming convention or ontology to describe them. We collect information for 180,000 sensors across 55 buildings at UCSD, and use their metadata to identify the type of sensor and provide them a uniform name across the system. We first group similar sensors using hierarchical clustering, and then use active learning with random forest classifier to label the sensors.
Modern fault management systems for HVAC are rudimentary, and fail to detect pertinent faults that may lead to equipment damage or energy wastage. We build a web services based fault management framework for better information integration and flexibility. We also develop a comparative data mining based fault detection technique to identify many faults missed by the traditional system.
Traditional building management systems do not provide adequate support for storing sensor data and contextual metadata. We created BuildingDepot, a RESTful web service based building data store that integrates information from thousands of sensors across a building. We incorporate useful features such as role based access control, sensor tagging for search and contextual information, and RESTful APIs for third party apps.
We participated in a competition hosted by Department of Energy for improved building energy efficiency. We wrote two proposals: (a) Innovative methods for incentivizing building owners and managers to adopt energy efficiency retrofits. (b) Proposal for improving sustainability practices in research laboratories in universities.
Occupants of modern buildings typically connect to enterprise WiFi network when they are in the building. We tap in to the authentication logs of the WiFi network to determine when occupants connect to access points near their offices. We use this as a proxy for occupancy, and use it to control the HVAC system.
As CMOS technology continues to shrink, process variation leads to changes in performance from chip to chip. We characterize the variation in power consumption of Intel Core i5-540M processors using specially instrumented Calpella and Sandy Bridge platforms from Intel for accurately measuring CPU power.
We develop a low cost, battery powered, wireless occupancy sensor that uses a combination of PIR and door sensors. They have 97% accuracy of detection in single room offices. We connect this information to the building management system and control HVAC based on occupancy.
We created an extremely low cost, non-intrusive, wireless energy meter to monitor and actuate plug loads within buildings. The design is modular, and contains a programmable microcontroller. We exploit the meter for managing plug loads during demand response events and estimating energy wastage in offices.
We use a specially instrumented Calpella motherboard from Intel to breakdown power consumption of individual components in a computer - processor, memory, solid state disk and ethernet. We design an in house hardware harness and use National Instruments DAQs for power measurement. We model the power consumption of motherboard components using Linux performance counters.
802.11b WiFi clients listen to packets even when they are not addressed to them. We use a software defined radio to show significant power savings can be obtained in crowded spaces when the client sleeps after receiving just the header of the packet. QualNet simulations show promising power savings in different environments.
We use prediction of handoff in cellular network as an example application to show that rough set theory is effective in reducing feature set. We simulate a cellular network on Matlab and extract features for prediction. We use rough set theory to reduce the feature set and neural network for handoff prediction.