 Event Information |
|
Interpretable Convolutional Neural Network-based Diagnostics for Rotating Machinery
Abstract: Rotating machines with faults can be diagnosed using vibrational characteristics extracted by signal processing techniques requiring parameter setting. Convolutional neural networks can eliminate the need for parameter setting, but their internal operations lack interpretability, impeding physical validation. This dissertation presents a method for changing the internal operations of a convolutional neural network to emulate envelope analysis, a specific type of signal processing. Envelope analysis extracts signatures associated with mechanical impact on defective rolling components. The developed network identifies the signatures that change with speed, which are verified through a feature selection process. Case studies demonstrate the feasibility of using the network for fault diagnosis. About the Presenter:
This Event is For: Public • Clark School • All Students • Graduate • Faculty • Post-Docs • Alumni
|
Namkyoung Lee received a B.S. degree in electronics from Hanyang University, Seoul, Korea, in 2015 and an M.S. degree in computer science from the Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea in 2017. He is a Ph.D. candidate in Reliability Engineering in the Department of Mechanical Engineering at the University of Maryland, College Park, MD, USA. His research interests include data-driven prognostics and health management systems for rotating machinery and diagnostics and prognostics of automotive components. His advisors are Dr. Michael H. Azarian and Prof. Michael Pecht.
