Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


29311.pdf (5.33 MB)

ETD Abstract Container

Abstract Header

Semi-supervised Ensemble Learning Methods for Enhanced Prognostics and Health Management

Shi, Zhe
http://orcid.org/0000-0003-1654-1138
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1522420632837268

Abstract Details

2018, PhD, University of Cincinnati, Engineering and Applied Science: Mechanical Engineering.
Advances in data acquisition and storage technologies have enabled the easy accumulation of a large amount of training data from many real-world applications in industry. However, assigning labels to all those samples can be expensive because the labeling process usually requires human expertise and is time-consuming. In prognostics and health management (PHM) applications, the labels assigned to the condition monitoring data represent the general health condition of a monitored system, which can serve as a performance indicator(normal vs. faulty) or a product quality metric (good vs. bad). In practice, it is common to have a small portion of labeled samples and a large number of unlabeled samples, however, without a sufficient amount of labeled training samples, most supervised learning methods are unable to produce an accurate generalized model. In addition, most of the learning methods have their own pros and cons. When dealing with the same data set, the result can be quite different based on the method being used. Then, a question arises: how to extend the capability of the learning algorithm to fully explore the information behind the entire data set to generate a stronger hypothesis. The research presented in this dissertation focuses on exploring the methodology of utilizing unlabeled or partially labeled samples with a group of weak learners to improve the performance of current learning methods for PHM applications. Semi-supervised learning (SSL) and ensemble learning (EL) are two important learning paradigms that were developed almost in parallel but with different philosophies. SSL focuses on improving generalization performance of learning models by including unlabeled data, while EL takes advantage of a cluster of diverse models to promote final prediction accuracy. The core research lies in discovering the inherent correlation between labeled samples and unlabeled samples and increasing diversity of ensemble weak learners to ultimately minimize prediction error. Research supports that it is promising to combine SSL and EL as a semi-supervised ensemble learning (SSEL) framework to further improve the capability of each other. The SSEL methods have been successfully applied to several classical machine learning and pattern recognition problems, such as image classification and natural language processing. The success of these applications encourages the study and development of effective SSEL methods for PHM applications.
Jay Lee, Ph.D. (Committee Chair)
Raj Bhatnagar, Ph.D. (Committee Member)
Mark Schulz, Ph.D. (Committee Member)
David Thompson, Ph.D. (Committee Member)
Wenyu Zhao, Ph.D. (Committee Member)
125 p.
Mechanical Engineering
Regression; Classification; PHM; Semi-supervised Ensemble Learning; Unlabeled samples; Insufficient training data set

Recommended Citations

Citations

  • Shi, Z. (2018). Semi-supervised Ensemble Learning Methods for Enhanced Prognostics and Health Management [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1522420632837268

    APA Style (7th edition)

  • Shi, Zhe. Semi-supervised Ensemble Learning Methods for Enhanced Prognostics and Health Management. 2018. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1522420632837268.

    MLA Style (8th edition)

  • Shi, Zhe. "Semi-supervised Ensemble Learning Methods for Enhanced Prognostics and Health Management." Doctoral dissertation, University of Cincinnati, 2018. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1522420632837268

    Chicago Manual of Style (17th edition)

ucin1522420632837268
2,754
© 2018, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.