FAMOUS - Federated Anomaly Modelling and Orchestration for modUlar Systems
|Coordinator||Scania CV AB|
|Funding from Vinnova||SEK 8 922 297|
|Project duration||November 2020 - December 2023|
|Venture||Electronics, software and communication - FFI|
|Call||Electronics, Software and Communication - FFI - June 2020|
Purpose and goal
The goals of the FAMOUS project are: to develop a federated protocol and models for fault detection with intermittent connected vehicles that guarantees converegence of the federated models; to integrate the federated models with a hierarchical clustering based on the underlying modular system of Scania´s vehicles; and to develop a scalable and flexible vehicle edge analytics solution for efficient development, testing, and deployment of models as well as data streaming for selected time-series sensor signals.
Expected results and effects
The planned results are to develop federated anomaly detection methods optimized for edge computing and evaluated on injected faults and on a test fleet with methods that maps anomaly classes to known or undiscovered faults. The federated anomaly detection method will be developed for a diverse vehicle fleet. Finally, the project will build a scalable vehicle edge device prototype for orchestration of federated learning, model deployment and model testing.
Planned approach and implementation
The project will be executed with one full time post-doc researcher and two part time professors from LiU for two and a half years. Crosser will have the equivalent of 0.4 full-time development engineers for two and a half years and Scania’s resources will be equivalent to 3.28 full-time data scientists and development engineers for two and a half years. The project consists of five work packages from which Crosser and LiU will lead one each and Scania the remaining three.