Advisors:
Dr. Jianjun Shi, School of Industrial and Systems Engineering, Georgia Tech
Committee members:
Dr. Roshan Joseph, School of Industrial and Systems Engineering, Georgia Tech
Dr. Jing Li, School of Industrial and Systems Engineering, Georgia Tech
Dr. Yajun Mei, School of Industrial and Systems Engineering, Georgia Tech
Dr. Kamran Paynabar, School of Industrial and Systems Engineering, Georgia Tech
Dr. Hao Yan, School of Computing, Informatics, and Decision Systems Engineering, Arizona State University
Date and Time: 10:00 am (EST), Wednesday, May 5th, 2021
Meeting URL: https://bluejeans.com/324530406
Meeting ID: 324 530 406 (BlueJeans)
Abstract:
Nowadays, the multistage manufacturing processes (MMPs) are usually equipped with complex sensing systems. They generate data with several unique characteristics: the output quality measurements from each stage are of different types, the comprehensive set of inputs have distinct degrees of influence over the process, and the relationship between the inputs and outputs is sometimes ambiguous, and multiple types of faults repetitively occur to the process during the operation. These characteristics of the data lead to new challenges in the data analytics of MMPs so that the existing state-space modeling approach is not suitable.
In this thesis, we conduct three studies to tackle the new challenges from MMPs. In the first study, we propose a feature ranking scheme that ranks the process features based on their relationship with the final product quality. Our ranking scheme is called sparse distance correlation (SpaDC), and it satisfies the important diversity criteria from the engineering perspective and encourages the features that uniquely characterize the manufacturing process to be prioritized. The theoretical properties of SpaDC are studied. Simulations, as well as two real-case studies are conducted to validate the method.
In the second study, we propose a holistic modeling approach for the MMPs, aiming at understanding how intermediate quality measurements of mixed profile outputs relate to sparse effective inputs. This model can identify the effective inputs, output variation patterns, and establish connections between them. Specifically, the aforementioned objective is achieved by formulating and solving an optimization problem that involves the effects of process inputs on the outputs across the entire MMP. This ADMM algorithm that solves this problem is highly parallelizable and thus can handle a large amount of data of mixed types, obtained from multiple stages.
In the third study, a retrospective analysis method is proposed for multiple functional signals. This method simultaneously identifies when multiple events occur to the system and characterizes how they affect the multiple sensing signals. A problem is formulated using the dictionary learning method and the solution is obtained by iteratively updating the event signatures and sequences using.
]]>