FAMOUS - Federated Anomaly Modelling and Orchestration for modUlar Systems

Important results from the project

The purpose of the project was:The development of a federated protocol and models for fault detection with intermittent connected vehicles - 3 federated models for on-board anomaly detection were developed where convergence for intermittent connectivity was analyzed; leverage the underlying modularity of vehicles for model accuracy improvement - with the data analyzed no true improvement on building specilized models was found, therefore focus was shifted on new survival models; and develop a scalable and flexible vehicle edge analytics prototype - edge device built and tested

Expected long term effects

This prototype was tested in a truck and ML models for anomaly detection were developed tailored to this prototype. The outcome was an architecture to build edge devices for the vehicle industry and for MLOps. The developed models for anomaly detection are very promising and are being leveraged in other use-cases. New was to develop survival models for continuous distributions have been proposed and are used in further research projects.

Approach and implementation

We had quarterly workshops where each partner showed their progress and where next steps were discussed. Such workshops were a key to sync within the partners, show progress, address impediments and plan next steps. Ad-hoc meeting, if possible on site, were organized when needed, improving the cross collaboration between the parties. - In retrospective this was a very fruitful collaboration even more if taking into consideration that the project started at the height of the COVID pandemic, where no IRL meetings were possible.