Sensor-based Modeling and Analysis of Advanced Manufacturing Systems for Quality Improvements
Open Access
- Author:
- Imani, Farhad
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 09, 2020
- Committee Members:
- Hui Yang, Dissertation Advisor/Co-Advisor
Hui Yang, Committee Chair/Co-Chair
Timothy W. Simpson, Committee Member
Saurabh Basu, Committee Member
Edward William Reutzel, Outside Member
Steven James Landry, Program Head/Chair - Keywords:
- Additive Manufacturing
Artificial Intelligence
Smart Manufacturing
Machine Learning
Process Monitoring and Control
Quality Control - Abstract:
- Advanced sensing is increasingly invested in modern manufacturing systems to cope with the complexity and enhance information visibility, thereby leading to data-rich environments. Generated data provide unprecedented opportunities to investigate system dynamics and further improve quality monitoring and control for advanced manufacturing in real-time. However, high-dimensionality and complex structures of sensing data pose significant challenges. Realizing full potentials of sensing data depends to a great extent on the development of novel analytical methods and tools for effective modeling, monitoring, and control of manufacturing systems. The research objective of this dissertation is to develop new learning methodologies for real-time quality monitoring and control of complex manufacturing systems. This body of research will enable and assist in 1) understanding the effect of process conditions on quality of manufacturing builds, 2) extracting sensitive features and characterizing patterns of image data, 3) diagnosing defects in low-volume and highly-customized production settings, and 4) handling high dimensional spatiotemporal data. My research accomplishments include: 1) Process mapping and monitoring of porosity in additive manufacturing (AM): In Chapter 2, spectral graph theory and multifractal analysis are developed to quantify the effect of process conditions on lack of fusion porosity in builds made using AM process, and subsequently, to detect the onset of process conditions that lead to lack of fusion porosity from in-process sensor data. 2) Multifractal and lacunarity analysis for nonlinear pattern characterization: In Chapter 3, the joint multifractal and lacunarity analysis is designed to resolve local densities and characterize the filling patterns in image profiles. Further, we derive the composite quality index by computing Hotelling T2 statistics from multifractal and lacunarity features for defect detection and characterization in ultra precision machining (UPM) and AM image profiles. 3) Image-guided variant geometry analysis of layerwise build quality: In Chapter 4, we develop a tailored deep neural network (DNN) framework that learns the broad geometrical diversity of images from builds made with AM. The proposed methodology leverages the computer-aided design (CAD) file to register the region of interest (ROI) in each layerwise image. Next, we propose a dyadic partitioning method to delineate variant ROI into distinctive regions with the same size and in multiple scales. Then, we leverage the semiparametric spatial model to characterize the complex spatial patterns in subregion ROIs. Finally, a DNN is designed to learn incipient flaws from spatial characterization images. 4) Spatiotemporal Gaussian process for AM quality monitoring: In Chapter 5, a novel spatiotemporal Gaussian process (STGP) is introduced to model the standard geometric profile within ROIs and capture layer-to-layer spatiotemporal deviations for quality monitoring. Finally, we leverage the STGP model to develop new monitoring charts, namely, the STT2 and STLR tests, for the anomaly detection in AM processes. This framework enables on-the-fly assessment of AM build quality.