2026-05-24T01:13:52Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-2023202025-08-18T22:22:09Zoai_pmh:alloai_pmh:repo_items

202320 https://hdl.handle.net/2286/R.2.N.202320 http://rightsstatements.org/vocab/InC/1.0/ All Rights Reserved 2025 190 pages Doctoral Dissertation Academic theses en Ouidadi, Hasnaa Guo, Shenghan Starly, Binil Kannan, Arunachala Mada Ko, Hyunwoong Arizona State University Partial requirement for: Ph.D., Arizona State University, 2025 Field of study: Manufacturing Engineering Defects are major quality concerns that hinder the broad adoption of additive manufacturing (AM) in critical industries such as aerospace, maritime, and energy, where errors are fatal. With all the advancements achieved in the field of artificial intelligence (AI), leveraging this technique to improve AM processes has become particularly important. Nevertheless, the growing use of AM in personalized production causes scarcity in historical data, making it challenging to train AI models effectively. The present dissertation has two main objectives. First, it develops novel AI-driven models to improve in-situ AM quality monitoring, defect segmentation, and process planning. Second, it proposes knowledge transfer as an efficient practice to mitigate AI models’ generalization problems when data is scarce or unlabeled. Knowledge transfer is a technique that allows models to reuse or “transfer” knowledge acquired from one task, dataset, or domain to another related but different one. This approach provides many advantages in saving data collection and annotation efforts. The present work proposes three methodologies to successfully transfer knowledge between different customized parts, namely, online/incremental learning, transfer learning, and unsupervised adversarial domain adaptation. Furthermore, it leverages generative adversarial networks, a type of generative AI models that use a knowledge transfer mechanism during the training process, to help manufacturers obtain an assessment of their products’ quality before starting the printing process. This dissertation uses real data from three AM processes: direct energy deposition, laser powder bed fusion, and aerosol jet printing. The case studies presented as part of this research demonstrate the benefits of the proposed methodologies in improving smart AM processes and products’ quality despite the challenges induced by increased personalized production. Artificial Intelligence Engineering Additive Manufacturing Artificial Intelligence defect detection Knowledge Transfer Quality control Smart Manufacturing Machine Learning Solutions for Knowledge Transfer in Smart Additive Manufacturing Processes