Building fault detection data to aid diagnostic algorithm creation and performance testing

Jessica Granderson,Yan Chen,Guanjing Lin,Ari Harding,Piljae Im

doi:10.1038/s41597-020-0398-6

Abstract

It is estimated that approximately 4–5% of national energy consumption can be saved through corrections to existing commercial building controls infrastructure and resulting improvements to efficiency. Correspondingly, automated fault detection and diagnostics (FDD) algorithms are designed to identify the presence of operational faults and their root causes. A diversity of techniques is used for FDD spanning physical models, black box, and rule-based approaches. A persistent challenge has been the lack of common datasets and test methods to benchmark their performance accuracy. This article presents a first of its kind public dataset with ground-truth data on the presence and absence of building faults. This dataset spans a range of seasons and operational conditions and encompasses multiple building system types. It contains information on fault severity, as well as data points reflective of the measurements in building control systems that FDD algorithms typically have access to. The data were created using simulation models as well as experimental test facilities, and will be expanded over time.

Highlights

Background & SummaryBuildings use 40% of primary energy globally, and account for 33% of direct and indirect carbon emissions from fuel combustion[1]
Algorithms developed to perform automated fault detection and diagnostics (FDD) use building operational data to identify the presence of faults and isolate their root causes
As buildings become more data rich, and as data science comes to buildings, FDD is of increasing relevance to the building community

Summary

Background & Summary

Buildings use 40% of primary energy globally, and account for 33% of direct and indirect carbon emissions from fuel combustion[1]. Algorithms developed to perform automated fault detection and diagnostics (FDD) use building operational data to identify the presence of faults and (in some cases) isolate their root causes. The dataset described in this article contains operational building heating ventilation and air-conditioning (HVAC) data, paired with validated ground-truth information as to the presence and absence of faults. A preliminary illustration of use of the dataset to compare and contrast FDD algorithm performance accuracy and identify performance gaps is documented in[11]. This initial dataset will be expanded over time to cover a larger range of operational conditions, fault types, and seasons. It will be evolved to include a larger set of HVAC systems, chiller and boiler plants, dual-duct AHUs, terminal variable air volume (VAV) boxes, and terminal fan coil units

Methods

Method of fault imposition

Findings

Code availability