Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


38271.pdf (5.64 MB)

ETD Abstract Container

Abstract Header

Application and Performance Enhancement of Intelligent Cross-Domain Fault Diagnosis in Rotating Machinery

Ainapure, Abhijeet Narhar
http://orcid.org/0000-0003-2924-6322
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1623164772153736

Abstract Details

2021, MS, University of Cincinnati, Engineering and Applied Science: Mechanical Engineering.
In recent years, intelligent data-driven techniques for health identification and diagnostics of rotating machines, have started to gain increasing attention for optimizing maintenance practices, enhancing operational safety, and reducing unnecessary costs. Specifically, deep learning methodologies have been popularly developed because of their ability to automatically extract meaningful features from raw measurement data. Owing to this, they are suitable for industrial applications, where feature engineering knowledge and domain expertise might be limited. Despite their effective development, a major drawback of traditional approaches lies in the general assumption that the training and test dataset are acquired from similar distributions, i.e., same mechanical system/component under identical conditions. However, in real-world applications, the diagnostic model developed using labeled train data (referred as source domain) can be applied on a different (but related) unlabeled test data (referred as target domain). In such scenarios, where there exists a distributional shift, the generalization ability of the model is seriously affected leading to poor fault diagnosis performance. This cross-domain diagnostic issue is further enhanced in harsh practical conditions with environmental noise and reduced data availability. To minimize the distributional gaps and improve the model’s generalization process, a domain adaptation methodology within the deep learning framework has been proposed in this study. Multiple convolutional operations coupled with maximum mean discrepancy metric, facilitate automatic extraction of domain-invariant features for promising diagnostic performance across domains. For further enhancing the model’s robustness and maintaining high levels of generalization, noisy condition labels are introduced during the model training process. Using the proposed approach, promising results can be obtained even with strong noise interference in testing data, as well as in the practical cases where data quality is compromised. In this thesis, the proposed techniques are validated on both open-source and industrial rotating machinery datasets.
Jay Lee, Ph.D. (Committee Chair)
Jay Kim, Ph.D. (Committee Member)
Manish Kumar, Ph.D. (Committee Member)
77 p.
Mechanical Engineering
Fault Diagnosis; Deep Learning; Domain Adaptation; Convolutional Neural Network; Noisy Labels; Rotating Machines

Recommended Citations

Citations

  • Ainapure, A. N. (2021). Application and Performance Enhancement of Intelligent Cross-Domain Fault Diagnosis in Rotating Machinery [Master's thesis, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1623164772153736

    APA Style (7th edition)

  • Ainapure, Abhijeet Narhar. Application and Performance Enhancement of Intelligent Cross-Domain Fault Diagnosis in Rotating Machinery. 2021. University of Cincinnati, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1623164772153736.

    MLA Style (8th edition)

  • Ainapure, Abhijeet Narhar. "Application and Performance Enhancement of Intelligent Cross-Domain Fault Diagnosis in Rotating Machinery." Master's thesis, University of Cincinnati, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1623164772153736

    Chicago Manual of Style (17th edition)

ucin1623164772153736
253
© 2021, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.