Spatiotemporal Analytics for Advanced Manufacturing
Open Access
- Author:
- Zhang, Siqi
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 27, 2024
- Committee Members:
- Steven Landry, Program Head/Chair
Yan Lu, Special Member
Lingzhou Xue, Outside Unit Member
Edward Reutzel, Outside Field Member
Timothy Simpson, Major Field Member
Hui Yang, Chair & Dissertation Advisor - Keywords:
- Feature learning
engineering knowledge
process monitoring
quality control
additive manufacturing
melt pool
additive manufacturing
melt pool
morphological modeling
feature learning
engineering knowledge
process monitoring
quality control
generative adversarial networks
spatiotemporal analytics - Abstract:
- The proliferation of big data from various sensing has created unprecedented opportunities to model, monitor, and control advanced manufacturing systems. However, the generated data often have complex structures across both spatial and temporal dimensions and reflect process variations caused by a multitude of interacting factors within the manufacturing system. These characteristics pose significant challenges for these systems to be effectively modeled, monitored, and controlled. Hence, this dissertation attempts to develop innovative methodologies to address the above challenges and advance manufacturing intelligence to a higher level. This body of research will enable and assist in 1) better handling of complex-structured data, 2) effective extraction of process features pertinent to quality patterns guided on engineering principles and diagnosis of issues prior to post-build quality analysis; 3) generative modeling of melt-pool images across a variety of scenarios. The research accomplishments include: • Characterization of Complex Morphology for Manufacturing Process Monitoring: Chapter 2 presents a multiscale modeling framework to represent and characterize 3D morphological variations of melt pool. This framework enables the extraction of multiscale morphological features over oversimplified geometric properties to track process variations and detect anomalies during fabrication. • Deep Learning for Quality Analysis guided on Engineering Principles: Chapter 3 proposes a novel feature learning framework to integrate engineering knowledge into neural networks to delineate quality patterns and extract related process features. A 3D neighborhood model is further designed to characterize spatiotemporal dynamics from engineering-guided features for quality analysis. This framework not only improves image analytics via engineering-guided learning but also enables the investigation of process-quality relationships due to spatial-temporal effects. • Generative AI for Manufacturing Intelligence: Chapter 4 develops a process-informed adversarial learning framework to predict high-dimensional dynamics of the melt pool during the additive manufacturing process. By fusing historical imaging streams and process parameters, this framework can predict both the spatial dynamics of melt pool in individual images and temporal dynamics along the scanning history embedded in previous frames. It provides data-driven insights into the underlying mechanisms of melt-pool dynamics under various process conditions and has the potential to elevate manufacturing intelligence to a new level.