The Harvard Center for Green Buildings and Cities Research Team has filed a patent application for the following work: An Open Platform for Automated HVAC Fault Detection and Smart Building Applications. This research and application provides an automated fault detection and diagnostics (FDD) solution that transforms energy-efficient building operational data into actionable information by detecting and tracking energy inefficiencies and system faults. It gives valuable insights on building performance in real time so that facilities can identify, prioritize, and fix faults quickly to avoid costs.

According to the Department of Energy, more than $10 billion US dollars can be saved through FDD and energy information systems for commercial HVAC systems every year. In China, it is estimated that there are more than 30% energy savings potential out of total energy consumption in large-scale commercial buildings, according to Tsinghua University. CGBC’s new research aims to improve FDD by using a robust data-driven algorithm for FDD that takes uncertainties in operations into account. It can detect abnormalities in whole building energy consumption; detect faults on the component-level of HVAC and other mechanical system operations; and estimate energy savings potential of what-if scenarios in building commissioning and retrofitting projects.

CGBC’s web-based forecasting application was developed from this research to allow users without coding background to build Gaussian Process (GP) models to make predictions and evaluate the impact of certain variables. It also has a decentralized architecture that is salable and extensible to easily incorporate new data sources, enhancing user experience by optimizing the design specifically for FDD. The tool’s forecast gives an estimate of how significant the impact of the investigated parameter is on total energy consumption. This rapid estimate could be useful in early decision-making in building commissioning/retrofitting projects.  Engineers can now use this information to select control variables that have large impact to optimize. The proposed GP modeling process is a direct and rapid modeling method based on actual data, and it avoids input configuration and calibration.