Graduation Semester and Year
Summer 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Industrial Engineering
Department
Industrial and Manufacturing Systems Engineering
First Advisor
Chen Kan
Second Advisor
Victoria Chen
Third Advisor
Jay Rosenberger
Fourth Advisor
Xin Liu
Abstract
Complex systems, e.g., advanced manufacturing systems, are largely associated with dynamic and transient behaviors, resulting in condition changes and anomalies. Sensor-based condition monitoring is critical in detecting anomalies and supporting process monitoring and performance improvement for complex manufacturing systems. Traditional sensor-based monitoring approaches primarily focus on one-dimensional (1D) signals and two-dimensional (2D) images, which are limited in their ability to capture high-resolution spatial patterns pertaining to anomalies induced by systems’ condition changes, especially subtle ones. Recent advancements in three-dimensional (3D) sensing present a unique opportunity to address this limitation by enabling the capture of 3D point cloud data with micro-level resolution. Despite its potential, analyzing point cloud data presents challenges due to the factors such as unstructured patterns, high dimensionality, and large data volume. Most existing methods lack the ability to fully explore the unstructured point clouds and extract both local and global geometric features for modeling, monitoring, and anomaly detection of complex systems. This dissertation aims to fill the gap by developing novel point cloud representation learning methodologies. The contributions of this dissertation are three-fold: (1) spatial domain point cloud representation learning – developing recurrence graph-based models to characterize and represent geometric patterns in both 2D and 3D unit cells of complex structures, such as metamaterials; (2) spectral domain point cloud representation learning – extending point cloud learning into the spectral domain using graph diffusion wavelets, which enables multiscale analysis to effectively isolate and highlight subtle or localized defects; and (3) designing both in-situ and ex-situ point cloud fusion methods for anomaly detection and process mining in manufacturing. These methods provide effective solutions for quality assurance of complex structures, e.g., additively manufactured metamaterials. The proposed framework enhances the interpretability and adaptability of point cloud analytics in advanced manufacturing.
Keywords
3D Point Cloud, Spatial Domain Analysis, Spectral Domain Analysis, Quality Assurance, Anomaly Detection, Additive Manufacturing
Disciplines
Industrial Engineering
License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Yang, Yujing, "Representation Learning of Point Cloud Data for Process Mining and Anomaly Detection in Complex Systems" (2025). Industrial, Manufacturing, and Systems Engineering Dissertations. 245.
https://mavmatrix.uta.edu/industrialmanusys_dissertations/245