Theses and Dissertations

ORCID

https://orcid.org/0009-0007-5310-9981

Issuing Body

Mississippi State University

Advisor

Tian, Wenmeng

Committee Member

Bian, Linkan

Committee Member

Wang, Haifeng

Committee Member

Lee, Seunghan

Date of Degree

12-13-2024

Original embargo terms

Visible MSU only 6 months

Document Type

Dissertation - Campus Access Only

Major

Industrial and Systems Engineering

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Department of Industrial and Systems Engineering

Abstract

This dissertation aims to develop effective methodologies towards enhanced intellectual property protections for data sharing frameworks in metal-based additive manufacturing (AM). Currently, many small-to-medium sized manufacturers face data availability challenges due to the prohibitive high cost to collect, process, and analyze large amounts of process-related data for AM. Because these manufactures rely heavily on small-scale data, it can be difficult for them to effectively train complex machine learning (ML) algorithms, which are commonly used for AM process monitoring. One popular solution is to develop collaborative data sharing frameworks, where multiple independent AM users can share their data to increase the size and diversity of available training data. By aggregating multiple, small-scale datasets, the collection of AM users can develop a more accurate and generalizable ML model, improving each stakeholder’s quality assurance capabilities. However, sharing AM process data outside of the user’s organization can raise serious intellectual property concerns, as sensitive design features are embedded within thermal process data that is commonly used for ML-based process monitoring and defect detection. To overcome this gap, this dissertation explores the development of two robust de-identification measures that can effectively remove or obscure confidential design attributes in the thermal process data, while simultaneously maintaining a high level of data usability. The first proposed algorithm leverages existing computer vision-based de-identification approaches, commonly used in facial image anonymization, and adds a unique adaptive component to adapt these frameworks to the AM-specific feature space. The second approach leverages state-of-the-art generative modeling to facilitate an image swapping framework, which directly replaces original, unprotected images with synthetic samples that disrupt the underlying geometric pattern in the data. Overall, the goal of this dissertation is twofold, aiming to (1) provide practitioners with a stronger understanding of the importance and impact of intellectual property in metal-based AM and (2) develop enhanced data privacy and intellectual property protections for metal-based AM, without compromising the usability of the data for process monitoring and defect detection applications.

Download

Share

COinS