This project is part of a multi-lab effort with Oak Ridge National Laboratory (ORNL), Lawrence Berkeley National Laboratory (LBNL), and Pacific Northwest National Laboratory (PNNL) in the area of adaptive and autonomous controls. In commercial buildings, it is estimated that annual building energy use can be cut by an average of 29%, or ~4-5% of overall national energy consumption through corrections to existing buildings controls infrastructure and resulting improvements to operating efficiency.1 For example, equipment faults and control errors in commercial buildings waste an estimated 1 quadrillion BTU of primary energy annually. Algorithms developed to perform automated fault detection and diagnostics (AFDD) use building operational data to identify the presence of faults and isolate their root causes. As a result, AFDD techniques are necessary in determining when corrections to building operations are needed and are a critical feature of adaptive and autonomous controls. The elimination of rules-based algorithms is an active area of research in buildings controls. As buildings become more data rich, analytics to improve the models developed for identifying and calculating faults, as well as the overall accuracy of these FDD solutions and their resulting energy efficiency, is of increasing relevance to the buildings community. Approaches investigated include first-principle or white-box (model-driven), black-box (data-driven), or gray-box (i.e. a combination of physical and empirical models). The most effective approach will depend in part on the building type, existing automation infrastructure, and type of fault under investigation. For heating, ventilation, and air conditioning (HVAC) systems, for example, BTO aims to achieve 20% end-use energy savings through research and development of novel adaptive and autonomous control solutions. This comprehensive multi-lab core effort aims to contribute to this goal through tasks that include the research and development of early-stage AFDD approaches for both buildings with and without existing automation systems, exploration of more effective integration schemes of sensor and metering data to improve the accuracy of models and the utilization of fewer sensors than are required by traditional rule-based methods, and data curation along with performance testing and benchmarking for field testing and verification of the approaches developed through this effort, as well as by researchers and members of the buildings community.
As part of this multi-lab effort, NREL will investigate fault detection and diagnostics for small- to medium-sized commercial buildings (up to 50,000 square feet), which are limited in their options due to the typical lack of existing building automation infrastructure. Specifically, NREL, in partnership with GE Global Research and Purdue University, will develop a novel model-based AFDD platform that leverages whole-building, physics-based energy models to provide fault detection and diagnosis. The project combines the building energy modeling capabilities of EnergyPlus and the OpenStudio ecosystem, DOE’s flagship energy modeling software tools, with the analysis and diagnostic capabilities of GE’s Predix cloud computing platform. The AFDD algorithm will model the expected energy performance of a building, detect the presence of abnormal (faulty) behavior by examining the degree of deviation between expected and actual building performance, and diagnos the most likely fault type by comparing the observed performance with a library of fault signatures developed from OpenStudio models of common equipment and control faults.
DOE Technology Manager: Marina Sofos
Lead Performer: Stephen Frank, National Renewable Energy Laboratory
1 N. Fernandez et al. Impacts of Commercial Building Controls on Energy Savings and Peak Load Reduction. PNNL Report 2017. http://buildingretuning.pnnl.gov/publications/PNNL-25985.pdf