Adversarial Transfer Learning for Machine Remaining Useful Life Prediction

Page view(s)
53
Checked on Jan 29, 2025
Adversarial Transfer Learning for Machine Remaining Useful Life Prediction
Title:
Adversarial Transfer Learning for Machine Remaining Useful Life Prediction
Journal Title:
2020 IEEE International Conference on Prognostics and Health Management (ICPHM)
Publication Date:
07 September 2020
Citation:
M. Ragab, Z. Chen, M. Wu, C. K. Kwoh and X. Li, "Adversarial Transfer Learning for Machine Remaining Useful Life Prediction," 2020 IEEE International Conference on Prognostics and Health Management (ICPHM), Detroit, MI, USA, 2020, pp. 1-7, doi: 10.1109/ICPHM49022.2020.9187053.
Abstract:
Remaining useful life (RUL) prediction is a key task for realizing predictive maintenance for industrial machines/assets. Accurate RUL prediction enables prior maintenance scheduling that can reduce downtime, reduce maintenance costs, and increase machine availability. Data-driven approaches have a widely acclaimed performance on RUL prediction of industrial machines. Usually, they assume that data used in training and testing phases are drawn from the same distribution. However, machines may work under different conditions (i.e., data distribution) for training and testing phases. As a result, the model performing well during training can deteriorate significantly during testing. Naive recollection and re-annotation of data for each new working condition can be very expensive and obviously not a viable solution. To alleviate this problem, we rely on a transfer learning approach called domain adaptation to transfer the knowledge learned from one labelled operating condition (source domain) to another operating condition (target domain) without labels. Particularly, we propose a novel adversarial domain adaption approach for remaining useful life prediction, named ADARUL, which can work on the data from different working conditions or different fault modes. This approach is built on top of a bidirectional deep long short-term memory (LSTM) network that can model the temporal dependency and extract representative features. Moreover, it derives invariant representation among the working conditions by removing the domain-specific information while keeping the task-specific information. We have conducted comprehensive experiments among four different datasets of turbofan engines. The experiments show that our proposed method significantly outperforms the state-of-the-art methods.
License type:
Funding Info:
This work is supported by the A*STAR Industrial Internet of Things Research Program under the RIE2020 IAF-PP Grant (grant no: A1788a0023).
Description:
© 2020 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
ISBN:
978-1-7281-6286-7
978-1-7281-6287-4
Files uploaded:

File Size Format Action
phm-darul.pdf 1.14 MB PDF Open