Semi-supervised approach to identifying and locating faults and defects

Assist the expert analyzing the data to enable him to focus on the crucial elements, i.e. the data indicating anomalies.

Internship project – summer 2020

Student : Olivier Bloch
Company: Bentley Systems

Bentley provides specialized software for all types of infrastructure projects worldwide.

The proposed project aims to develop a generic anomaly detector based on an adversarial autoencoder, which automatically learns the distribution of normal elements. Subsequently, any element deviating from this distribution can be qualified as abnormal, even if the system has never seen such a fault during its training.

Overall, the aim is to support the expert analyzing the data, enabling him to concentrate on the crucial elements (the data indicating anomalies) in the knowledge that identifying >95% of the data presents no problem. The project would initially focus on sewer and rail data, but could later be applied to a wide range of situations. Bentley envisages future joint use of multiple modalities (visual, infrared, etc.) to enhance the method’s performance and versatility.

Let’s keep in touch!

Would you like to be informed about IID news and activities? Subscribe now to our monthly newsletter.