Whole-Building Fault Detection: A Scalable Approach Using Spectral Methods
For building management system operators, this provides a scalable method to enhance fault detection without requiring new data sources, though the improvement is incremental over existing rules-based approaches.
This paper extends rules-based fault detection for buildings by leveraging spectral properties of the Koopman operator to decompose diagnostic signals, enabling multi-scale temporal and spatial analysis that improves classification effectiveness. The approach maintains compatibility with existing knowledge bases while offering new diagnostic capabilities.
In this paper, an extension to rules-based fault detection is demonstrated utilizing properties of the Koopman operator. The Koopman operator is an infinite-dimensional, linear operator that captures nonlinear, finite dimensional dynamics. The definition of the Koopman operator enables algorithms that can evaluate the magnitude and coincidence of time-series data. Using spectral properties of this operator, diagnostic rule signals generated from building management system (BMS) trend data can be decomposed into components that allow the capture of device behavior at varying time-scales and to a granular level. As it relates to the implementation of fault detection (FDD), this approach creates additional spatial and temporal characterizations of rule signals providing additional data structure and increasing effectiveness with which classification techniques can be applied to the analysis process. The approach permits a knowledge base to be applied in a similar manner to that of a rules-based approach, but the introduced extensions also facilitate the definition of new kinds of diagnostics and overall provide increased analysis potential.