EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
1458. Contrastive Transfer Learning Based on Damage Propagation for Remaining Useful Life Prediction
Invited abstract in session MA-4: Recent Methodologies in Explainable AI (XAI) 1, stream Recent Advancements in AI .
Monday, 8:30-10:00Room: 1001 (building: 202)
Authors (first author is the speaker)
| 1. | Seunghwan Song
|
| Department of Industrial and Management Engineering, Korea University | |
| 2. | Jeena Son
|
| Department of Industrial and Management Engineering, Korea university | |
| 3. | Junyong Lee
|
| Department of Industrial and Management Engineering, Korea University | |
| 4. | Jun-Geol Baek
|
| School of Industrial Management Engineering, Korea University |
Abstract
Remaining useful life (RUL) prediction is a crucial task in prognostics and health management (PHM), gaining increasing importance in industrial applications. However, the accuracy of RUL predictions in real-world industrial can be compromised due to data imbalance and discrepancies in data characteristics caused by varying operational conditions. Inspired by the idea of transfer learning, this study proposes a novel RUL prediction framework that utilizes damage propagation and contrastive transfer learning. At first, the proposed method differentiates between health and degradation states within source domain data on a window basis, employing the augmented Dickey-Fuller (ADF) test and acceleration graphs. Then, it predicts the RUL by training on the history subset of degradation state data to maximize the similarity with the multi-scale feature representation of future subset. In the target domain, a comparable strategy is employed but with adjusted sensitivity towards the acceleration graph for more precise state differentiation, enabling the application of the model learned in the source domain to predict the RUL in the target domain. Finally, the proposed method is proved to be effective through two public run-to-failure datasets and one real-world dataset. Demonstrated through experimental results, the proposed method shows superior predictive performance and high transferability, indicating its potential to address the challenges in industrial environments.
Keywords
- Reliability
- Forecasting
- Artificial Intelligence
Status: accepted
Back to the list of papers