Summary
The aim of the project is to design and develop an online learning-based fault diagnosis engine with adaptation capabilities. This engine will monitor and analyse data arriving in real time from critical infrastructure (CI) systems, to accurately detect a potential fault and effectively isolate and identify its exact location. Modern society relies heavily on the availability and smooth operation of CI systems, such as electrical power systems, water distribution systems and telecommunication networks. In such large-scale, complex engineering systems when a failure occurs due to faults, it can have severe societal, health and economic consequences. The sequential arrival of data in CI systems calls for a fault diagnosis engine with adaptive behaviour to achieve and maintain optimal performance. However, the vast majority of existing work falls short on this requirement. This project will incorporate online learning capabilities to achieve adaptability and will also address class imbalance, a major challenge for learning systems, arising from the fact that faults are low probability events. Online class imbalance learning (OCIL) is an emerging research topic focusing on the combined challenges of online learning and class imbalance. We will shed light on supervised OCIL as very few methods currently deal with this problem and address for the first time the unsupervised and semi-supervised OCIL problems. The proposed algorithms will be evaluated in realistic fault diagnosis datasets from industrial partners and in an advanced Smart Buildings simulator allowing us to run sensor fault scenarios in large-scale multi-zone buildings. Furthermore, a prototype on sensor fault diagnosis will be delivered that will be evaluated on a physical Smart Buildings testbed to enable its efficient testing under realistic conditions. Overall, this novel and interdisciplinary project will provide invaluable insights on incorporating learning capabilities in CI systems fault diagnosis.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/867433 |
| Start date: | 01-10-2019 |
| End date: | 14-10-2021 |
| Total budget - Public funding: | 157 941,12 Euro - 157 941,00 Euro |
Cordis data
Original description
The aim of the project is to design and develop an online learning-based fault diagnosis engine with adaptation capabilities. This engine will monitor and analyse data arriving in real time from critical infrastructure (CI) systems, to accurately detect a potential fault and effectively isolate and identify its exact location. Modern society relies heavily on the availability and smooth operation of CI systems, such as electrical power systems, water distribution systems and telecommunication networks. In such large-scale, complex engineering systems when a failure occurs due to faults, it can have severe societal, health and economic consequences. The sequential arrival of data in CI systems calls for a fault diagnosis engine with adaptive behaviour to achieve and maintain optimal performance. However, the vast majority of existing work falls short on this requirement. This project will incorporate online learning capabilities to achieve adaptability and will also address class imbalance, a major challenge for learning systems, arising from the fact that faults are low probability events. Online class imbalance learning (OCIL) is an emerging research topic focusing on the combined challenges of online learning and class imbalance. We will shed light on supervised OCIL as very few methods currently deal with this problem and address for the first time the unsupervised and semi-supervised OCIL problems. The proposed algorithms will be evaluated in realistic fault diagnosis datasets from industrial partners and in an advanced Smart Buildings simulator allowing us to run sensor fault scenarios in large-scale multi-zone buildings. Furthermore, a prototype on sensor fault diagnosis will be delivered that will be evaluated on a physical Smart Buildings testbed to enable its efficient testing under realistic conditions. Overall, this novel and interdisciplinary project will provide invaluable insights on incorporating learning capabilities in CI systems fault diagnosis.Status
CLOSEDCall topic
WF-01-2018Update Date
17-05-2024
Images
No images available.
Geographical location(s)
