Semester

Summer

Date of Graduation

2025

Document Type

Thesis (Campus Access)

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Industrial and Managements Systems Engineering

Committee Chair

Thorsten Wuest

Committee Member

Zhichao Liu

Committee Member

Zeyu Liu

Abstract

Manufacturing is undergoing a digital transformation of systems and processes – commonly known as Industry 4.0. This transformation is fueled by data which enables many of the smart manufacturing technologies like artificial intelligence (AI), digital twins, and augmented reality applications. However, manufacturing data, especially machine tools and other shopfloor applications, is infamously difficult to acquire and work with for a variety of reasons, including limited resources, data scarcity, privacy and cybersecurity concerns, high dimensionality, imbalanced target classes, and frequent missing values to name a few. These issues not only persist but are abundant in advanced manufacturing processes like additive manufacturing (AM). This makes working with AM process data difficult for practitioners who wish to utilize data-driven solutions like AI and machine learning (ML) for process improvement.

This thesis explores the deployment of data-driven approaches to evaluate a representative challenging dataset obtained from a unique, state-of-the-art hybrid additive manufacturing (HAM) machine. This unique manufacturing system employs two advanced manufacturing processes to construct additive structures using carbon fiber-reinforced polymer (CFRP) materials. The machine tool has a dual orifice head with a material extrusion nozzle and a tape laying fiber guide with a rotational axis. These orifices utilize existing technologies such as material extrusion AM (MEX-AM) and automated fiber placement (AFP) respectively, to create a novel HAM process. The relationship between the input process parameters, material microstructure changes during object construction, and final part properties is an established framework in AM. Through leveraging multi-scalar process-structure-property (PSP) relationships, predictions about physical and mechanical characteristics of constructed objects can be generated. These predictions can be utilized to provide an in-situ closed-loop solution to quality issues in a myriad of AM technologies. Before constructing this closed-loop quality system, it is important that the process parameters which impact the desired results are monitored and their data is intentionally collected. In this analysis, a feature selection approach is evaluated using a ML pipeline to robustly evaluate preprocessing techniques and their impact on the analysis of raw manufacturing data with imbalanced classes and data leakage. This study quantifies the stability and reliability of selected features using binary classification model accuracy, precision, recall, F1-scores, and the area under the receiver operator characteristic (AUC-ROC) curve. The results of this study illustrate the importance of intentionally applying the appropriate preprocessing techniques depending on the practitioner’s dataset characteristics.

Available for download on Friday, July 31, 2026

Share

COinS