Theses and Dissertations

ORCID

https://orcid.org/0009-0004-9354-0459

Advisor

Wang, Jun

Committee Member

Howard, Isaac L.

Committee Member

Ma, Junfeng

Committee Member

Gurbuz, Ali Cafer

Date of Degree

12-12-2025

Original embargo terms

Visible MSU Only 1 year

Document Type

Dissertation - Campus Access Only

Major

Engineering (Civil Engineering)

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Richard A. Rula School of Civil and Environmental Engineering

Abstract

Safety monitoring and training are two essential aspects of construction safety that help ensure workers adhere to safety standards. Technology-enabled systems in construction are becoming a common practice to enhance workers’ safety. However, technology-enabled systems still face challenges in achieving near-real-time monitoring while ensuring privacy, and current safety training methods lack real-time feedback, enhanced human-technology interactions (HTIs), and sufficient understanding of users’ technology acceptance. This dissertation addresses these challenges by developing an edge computing-based method to monitor and detect Personal Protective Equipment (PPE) in near real time while preserving data privacy, and a Virtual Reality (VR)-based safety training system that provides real-time feedback to enhance learning effectiveness and a technology acceptance model (TAM) to identify the factors affecting users’ acceptance of technology. An image dataset was developed, and Deep Learning (DL) models were trained using transfer learning and with DL models split between edge devices and servers to reduce latency. Among the five DL models considered, three single-stage You Only Look Once (YOLO) models (YOLO-v3, YOLO-v4, and YOLO-v4_tiny) achieved a mean average precision (mAP) ranging from 74.9% to 84.3% in safety glove detection; the other two DL models (YOLO-v4+VGG19, and YOLO-v4+Resnet50) achieved 89.9% mAP in hand detection and 100% mAP in safety glove classification. Furthermore, the edge computing-based method achieved 36%-38% shorter latency than the cloud computing-based method from implementation and theory perspectives. The VR-based ergonomics training incorporated real-time feedback and enhanced HTIs to improve learning effectiveness. Evaluation using effectiveness, efficacy, and TAM demonstrated higher motivation, improved knowledge gain, above-average usability score and good user experience. The TAM analysis identified the factors that affect user technology acceptance, including perceived ease of use, perceived usefulness, attitude toward use, and intention to use the training. This dissertation contributes to the body of knowledge by (i) developing edge computing-enabled DL models to enhance PPE monitoring in a time-efficient manner while maintaining data privacy, (ii) integrating real-time feedback and effective HTIs in a VR-based safety training method to improve learning effectiveness, and (iii) developing a TAM to identify factors affecting users' intention to use technology in safety training for future workforce development.

Sponsorship (Optional)

Mississippi Department of Transportation

Download

Share

COinS