*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Zhaoyi Xu
(Advisor: Prof. Joseph H. Saleh]
will propose a doctoral thesis entitled,
Deep Prognostic and Transfer Learning for Accurate Remaining Useful Life
Prediction with Uncertainty Quantification
On
Tuesday, May 25 at 11:00 a.m.
Bluejeans: https://bluejeans.com/876523583
Abstract
Unexpected failures in engineering systems or equipment often lead to significant disruptions and
losses. A key output of equipment prognostic is the estimation of remaining useful life (RUL) of the system
under consideration. Accuracy in RUL prediction is important to sustain equipment reliability, reduce total
maintenance costs, and prevent unexpected failures. This thesis addresses two prevalent challenges in
data-driven RUL prediction related to model accuracy and model generalization.
In Part I, this thesis addresses two aspects of the model accuracy challenge in data-driven RUL
prediction, namely the robustness to noise in sensor data and prognostic datasets, and the nonstationarity
or time-dependency of system degradation and RUL prediction given sensor data. A highly
accurate RUL prediction model is developed with uncertainty quantification, which integrates and
leverages the advantages of deep learning and nonstationary Gaussian process regression (DL-NSGPR).
The model is then subjected to critical evaluation, and its performance benchmarked against other datadriven
RUL prediction models. Computational experiments show that the DL-NSGPR significantly
outperforms other current best-in-class models, and the etiology for this performance differential is
identified and discussed.
In Part II, currently work-in-progress, this thesis will address select aspects of the model generalization
challenge. Two hypotheses for transfer learning related to RUL predictions are proposed, one related to
parameter transfer, and one to domain adaptation. Part II will design computational experiments and test
both hypotheses. It will then compare and benchmark the performance of the two proposed transfer
learning approaches. The best-in-class, if any, will be subjected to further critical assessment and its
potential for generalization examined.
Committee
Prof. Joseph H. Saleh – School of Aerospace Engineering (advisor)
Prof. Dimitri Mavris– School of Aerospace Engineering
Prof. Eric Feron– School of Aerospace Engineering
Dr. Evangelos Theodorou – School of Aerospace Engineering
