Data-driven Service Optimization and Quality Management in Cyber-physical Manufacturing Systems
Open Access
- Author:
- Chen, Ruimin
- Graduate Program:
- Industrial Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- May 19, 2021
- Committee Members:
- Edward Reutzel, Outside Unit & Field Member
Soundar Kumara, Major Field Member
Timothy Simpson, Major Field Member
Hui Yang, Chair & Dissertation Advisor
Steven Landry, Program Head/Chair
Yan Lu, Special Member - Keywords:
- Additive Manufacturing
Quality Management
Machine Learning
Service Optimization - Abstract:
- Recent advancements in sensing and communication technology provide unprecedented opportunities to synchronize the additive manufacturing (AM) machine world and facilities to the cyber computational space. The new paradigm of AM cyber-physical system is a convergence of interconnectivity and intelligence to form adaptable and resilient processes in the factory of future. The potential for AM cyber-physical systems to improve productivity leads to the new wave of technological changes and triggers paradigm shifts to service optimization and quality management. Sensing technology leads to a data-rich environment and provides a unique opportunity for different learning algorithms to accelerate the development of the AM cyber-physical systems. However, realizing the full potentials of data and transforming them into useful information and knowledge depends to a great extent on the development of novel analytical methods and tools. Specifically, the transition from the conventional manufacturing systems to the novel AM cyber-physical systems brings the following challenges: 1. The emergence of sharing economy enabled by sensing data provides opportunities in acquiring, providing, and sharing access to goods and services. Novel analytical approaches are urgently needed for optimal service management. 2. Advanced sensing brings a large amount of data with nonlinear and non-homogeneous patterns, which calls for effective analytical methods to exploit acquired knowledge and extract sensitive features for process monitoring and control. 3. The presence of extraneous noises and complex interactions in modern systems prevent the extraction of hidden patterns and reveal of root cause in causal inferences from a large amount of data. The goal of this dissertation is to improve the service quality of AM cyber-physical systems. The service management in AM cyber-physical systems includes not only the service optimization between resources (i.e., between service providers and seekers), but also the quality of the service that each provider can offer. Therefore, this dissertation is aimed at developing new machine learning methodologies to enhance understanding of design-quality interactions, facilitate causality discovery, to eventually improve service management in AM cyber-physical systems. My research accomplishments include: 1. Chapter 2 developed a bipartite matching framework to model and optimize resource allocation among customers and service providers through a stable matching algorithm in AM cyber-physical systems. The framework is implemented in the customer-manufacturing allocation in cyber-physical platforms. Experimental results show that the proposed framework shows strong potentials to optimize resource allocation in the AM sharing economy. 2. Chapter 3 focused on conducting a design of experiment to investigate how design parameters (e.g., build orientation, thin-wall width, thin-wall height, and hatching spacing) interact with edge roughness in thin-wall builds. This work sheds insights on the optimization of engineering design to improve the quality of AM builds. 3. Chapter 4 targeted leveraging data to characterize and detect irregular and nonlinear patterns of signals, 2D images, and 3D voxels. We proposed heterogeneous recurrence analysis and generalized recurrence network analysis to not only capture recurrence dynamics in complex systems but also take the computational complexity into account. A tailored design of experiment study was developed to reveal the relationship between network quantifiers and design parameters (i.e., orientation, width, height, hatching pattern). The designed methodology is implemented to characterize the AM in-process layerwise data. This work enables the on-the-fly assessment of AM builds and real-time defect mitigation. 4. Chapter 5 focused on developing a knowledge-driven Bayesian network for manufacturing complex systems to identify the root cause of quality outcomes and offer a comprehensive solution. This research is aimed at aggregating machine parameters, material information, design parameters, process parameters into a Bayesian network. With the network representation, the causal relationships among variables can be identified and then be used to facilitate prediction, diagnosis, and support decision-making in manufacturing production.