A Generalized Machine Learning-Based Classifier Considering Cost-Effective Features for Automated Fault Detection and Diagnosis (AFDD) of Packaged Rooftop Units - Dissertation

Aug 5, 2026

Upload by Eliana

Packaged rooftop units (RTUs) used in commercial buildings and manufacturing commonly develop soft faults that reduce cooling capacity, increase power consumption, and lower coefficient of performance, harming both equipment and energy usage. This dissertation develops a machine-learning classifier using a reduced set of nine features to detect and diagnose typical soft faults for RTUs with fixed orifice metering devices. Lab validation with the same training systems shows stronger performance than prior protocols and improved prediction as fault severity increases. Field validation at an Omaha facility confirms accurate undercharge diagnosis and supports generalizability for common soft faults, enabling industrial energy and cost savings assessments.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages164

Reading Time6 h 53 min

AccessFree to download

Abstract

Motivation and Background

Soft Faults in Packaged Rooftop Units

Proposed Method

Reduced Nine-Feature Machine-Learning Approach

Feature Set for Detection and Diagnosis

Validation and Results

Lab Testing and Performance Comparison

Field Testing at an Omaha Facility

Practical Impact

Industrial Assessment for Energy and Cost Savings

Dedication

Acknowledgments

University of Nebraska-Lincoln DigitalCommons@University of Nebraska-Lincoln | Dissertations and Doctoral Documents from University of Nebraska-Lincoln, 2024– | Graduate Studies | | --- | --- | | 12-4-2023
A Generalized Machine Learning-Based Classifier Considering Cost-Effective Features for Automated Fault Detection and Diagnosis (AFDD) of Packaged Rooftop Units
Md Rasel Uddin
University of Nebraska-Lincoln, [rasel07me@gmail.com](rasel07me@gmail.com)
Follow this and additional works at: [https://digitalcommons.unl.edu/dissunl](https://digitalcommons.unl.edu/dissunl)
Part of the Mechanical Engineering Commons | | Recommended Citation Uddin, Md Rasel, "A Generalized Machine Learning-Based Classifier Considering Cost-Effective Features for Automated Fault Detection and Diagnosis (AFDD) of Packaged Rooftop Units" (2023) . Dissertationsand Doctoral Documents from University of Nebraska-Lincoln, 2024–. 38. [https://digitalcommons.unl.edu/dissunl/38](https://digitalcommons.unl.edu/dissunl/38) This Dissertation is brought to you for free and open access by the Graduate Studies at DigitalCommons@University of Nebraska-Lincoln. It has been accepted for inclusion in Dissertations and Doctoral Documents from University of Nebraska-Lincoln, 2024– by an authorized administrator of DigitalCommons@University of Nebraska-Lincoln. A GENERALIZED MACHINE LEARNING-BASED CLASSIFIER CONSIDERING COST-EFFECTIVE FEATURES FOR AUTOMATED FAULT DETECTION AND DIAGNOSIS (AFDD) OF PACKAGED ROOFTOP UNITS by Md Rasel Uddin 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: Mechanical Engineering and Applied Mechanics Under the Supervision of Professor Robert Williams Lincoln, Nebraska December, 2023 A GENERALIZED MACHINE LEARNING-BASED CLASSIFIER CONSIDERING COST-EFFECTIVE FEATURES FOR AUTOMATED FAULT DETECTION AND DIAGNOSIS (AFDD) OF PACKAGED ROOFTOP UNITS Md Rasel Uddin, Ph.D. University of Nebraska, 2023 Advisor: Robert Williams Packaged rooftop units (RTUs) are widely used for space conditioning in commercial buildings and manufacturing facilities. The typical soft faults related to RTUs degrade the system's performance in terms of cooling capacity, power consumption, and Coefficient of Performance (COP) , detrimentally affecting both the equipment and energy consumption and the environment. Previous research in soft fault detection for rooftop units lacked classifier validation using lab and field data, developing a generalizable algorithm, and analyzing its performance across varying fault intensities. Using a simulated data library for multiple rooftop units, this study proposes a machine-learning classifier with a reduced set of 9 features (8 quantitative and one qualitative) to detect and diagnose typical soft faults in packaged rooftop units equipped with fixed orifice metering devices. An existing lab testing set consisting of the same training systems was utilized to validate the presented data-driven approach, showing significantly better performance than the existing fault detection and diagnosis protocols. In addition, the analyzed classifier’s predicting performance improves with increasing fault severity. In addition to the above lab validation, a manufacturing facility in Omaha, Nebraska, was chosen for field validation of the developed machine-learning algorithm. The proposed approach accurately predicted all the refrigerant undercharge fault cases from an RTU at that facility, although the RTU significantly differs from the RTUs with which the classifier was trained. The lab and field-testing results bolster that the considered machine-learning classifier can be generalizable, with some exceptions, for detecting the common soft faults from any rooftop unit equipped with a fixed orifice metering device. The presented classifier can be used i

FAQs

What problem does the dissertation address for packaged rooftop units (RTUs)?

It targets soft faults that degrade RTU performance, reducing cooling capacity and COP while increasing power consumption and energy/environmental impact.

How does the proposed method detect and diagnose faults?

It uses a machine-learning classifier built on a reduced set of nine features to detect and diagnose typical soft faults for RTUs with fixed orifice metering devices.

How was the approach validated in the study?

Validation included lab testing using an existing testing set based on the same training systems, followed by field validation at a manufacturing facility in Omaha, Nebraska.

What does the dissertation conclude about generalizability and usefulness?

Results show improved prediction with increasing fault severity and accurate detection of refrigerant undercharge cases in the field, supporting generalizability (with exceptions) for common soft faults and enabling industrial energy and cost savings assessments.

Aug 5, 2026

Upload by Eliana