Durham School of Architectural Engineering and Construction


First Advisor

Dr. David Yuill

Date of this Version



A DISSERTATION Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Doctor of Philosophy, Major: Architectural Engineering, Under the Supervision of Professor David Yuill. Lincoln, Nebraska: November, 2021

Copyright © 2021 Amir Ebrahimifakhar


This dissertation describes a large-scale investigation of heating, ventilation, and air-conditioning (HVAC) fault prevalence in commercial buildings in the United States. A multi-year dataset with 36,556 pieces of HVAC equipment including air handling units (AHUs), air terminal units (ATUs), and packaged rooftop units (RTUs) was analyzed to determine values for several HVAC fault prevalence metrics. The primary source of data for this study comes from three commercial fault detection and diagnostics (FDD) providers. Since each FDD provider uses different terms to refer to the same fault in an HVAC system, a mapping function was created for each FDD provider’s dataset, to convert the fault reports to a single standardized fault identifier. The fault identifier is taken from a standard taxonomy that was created for this purpose.

Since the commercial FDD software outputs are inherently subject to some level of error, i.e., they could have false negatives and false positives, a field study was conducted to gain greater insight into the commercial FDD software results. Two buildings from among the buildings of one of the FDD providers were selected. The RTUs serving these two buildings were monitored for about two weeks using our installed data loggers. The actual faults in these buildings were identified using methods that we developed or selected from the literature. The results of the field study were compared with the FDD provider fault reports.

This study also proposes a data-driven FDD strategy for RTUs, using machine learning classification methods. The FDD task is formulated as a multi-class classification problem. Seven typical RTU faults are discriminated against one another as well as the normal condition. Nine classification methods were applied to a dataset of simulation data, which was split into a training set and a test set. The performance of the classifiers for individual faults was characterized using true positive rate and false positive rate statistical measures. The relative importance of input variables was analyzed, and is also discussed in the dissertation.

Advisor: David Yuill