amachine learning -based approach for managing cyber -physical systemes
No Thumbnail Available
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The increasing prevalence of Cyber-Physical Systems (CPS) demands adaptive systems capable
of responding to dynamic environmental changes. Traditionally, ensuring quality of service
in CPS involves pre-planned activities at design time, posing challenges in highly dynamic
contexts. This thesis investigates the integration of Machine Learning (ML) to bolster CPS
adaptability, emphasizing security and privacy.
The primary objective is to employ ML algorithms for real-time monitoring and dynamic
adaptation of CPS behavior in response to environmental changes. Key contributions of this
thesis encompass a meta-model for adaptive CPS, providing a conceptual foundation to comprehend
and model interactions within CPS. In the second contribution, a ML-based architecture
is proposed, enhancing CPS functionalities, including data and resource management, fault
prediction, and proactive maintenance. A novel Collaborative ML-based architecture, founded
on the second contribution, is introduced, highlighting scalability, security, heightened data
privacy, and its suitability for resource-constrained environments. This Collaborative ML approach
encompasses Federated Machine Learning (FML) and Split Machine Learning (SML).
Validation is conducted through a case study on fault prediction in industrial CPS, and the experimentation
results showcase promising outcomes.