Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


37983.pdf (6.75 MB)

ETD Abstract Container

Abstract Header

A Transfer Learning Methodology of Domain Generalization for Prognostics and Health Management

Yang, Qibo
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1613749034966366

Abstract Details

2020, PhD, University of Cincinnati, Engineering and Applied Science: Mechanical Engineering.
In recent years, prognostics and health management (PHM) using machine learning methods have been widely developed for industrial applications. However, the traditional methods assume that training and testing data should come from the same probability distribution, which can hardly be satisfied in real-world applications due to the variability of domains, which is the discrepancy among machines and the discrepancy among working regimes. To deal with this issue, this study proposes a transfer learning methodology for the systematic deployment of domain generalization models for PHM tasks. This methodology includes the following parts. (1) A modeling approach of domain-generalization-adversarial neural network (DGANN) is studied to reduce the impact of the variability of domains, which can be flexibly used in the applications of fault detection, fault diagnosis, and fault prognosis. (2) A data assessment method called domain generalizability assessment (DGA) is proposed to evaluate the relationship among domains on knowledge transfer before domain generalization modeling. (3) A model assessment method called model generalization assessment (MGA) is presented to evaluate the generalization performance of models in real-world applications. Then the proposed methodology is exemplified by several industrial cases, including (1) fault detection of hard drives in cloud data centers, (2) fault diagnosis of rolling element bearings, and (3) fault prognosis of industrial robots.
Jay Lee, Ph.D. (Committee Chair)
Manish Kumar, Ph.D. (Committee Member)
Jing Shi, Ph.D. (Committee Member)
David Siegel, Ph.D. (Committee Member)
137 p.
Mechanical Engineering
Prognostics and health management; Transfer learning; Domain generalization; Maximum mean discrepancy; Neural networks

Recommended Citations

Citations

  • Yang, Q. (2020). A Transfer Learning Methodology of Domain Generalization for Prognostics and Health Management [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1613749034966366

    APA Style (7th edition)

  • Yang, Qibo. A Transfer Learning Methodology of Domain Generalization for Prognostics and Health Management. 2020. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1613749034966366.

    MLA Style (8th edition)

  • Yang, Qibo. "A Transfer Learning Methodology of Domain Generalization for Prognostics and Health Management." Doctoral dissertation, University of Cincinnati, 2020. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1613749034966366

    Chicago Manual of Style (17th edition)

ucin1613749034966366
258
© 2020, some rights reserved.Creative Commons License A Transfer Learning Methodology of Domain Generalization for Prognostics and Health Management by Qibo Yang is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by University of Cincinnati and OhioLINK.