Real-time Estimation of Building System State
This project aims to develop an agile asset management tool that can be integrated into legacy building management systems. The tool estimates in real-time the state of various building components. From our deep first-principles knowledge of building components we first create a taxonomy of possible failure modes and their likelihoods. We then mine the historical data archives using machine learning techniques to compile a library of faulty and normal signatures of building components for each mode in our taxonomy.
Matching the current state in run-time to these signatures will drive fault detection and isolation algorithms, as well as predictive maintenance strategies. This project uses supervised machine learning techniques to extract statistically significant signatures from monitoring datasets for a wide range of building components. It provides the building owners with enhanced visibility, extended knowledge, and smarter decision-making in building management functions.
This project will result in energy consumption saving, O&M cost reduction, and improved functionality. Algorithms and software will be developed to automatically integrate and analyze both structured and unstructured data from facilities design and operations, and thus enable facilities managers to be more proactive in improving building occupant comfort, conducting efficient resource allocation, and reducing wasted energy through machine learning and optimal control techniques.