A Secure and Explainable AI-Based Framework for IIoT with Privacy-Prioritized Model Aggregation

Abstract

Industrial Control Systems (ICS) serve as indispensable components across diverse human industries, encompassing vital sectors like Oil and Gas companies, Power Grids, and Transportation. The evolution of technology has witnessed a paradigm shift in communication within ICS, transitioning from traditional systems to wireless communication for swift monitoring and control. However, this technological advancement has simultaneously exposed ICS to the online world, amplifying the susceptibility to cyber attacks and threats. Given the pivotal role of ICS in human life, establishing a robust framework to ensure the security and privacy of these systems has become an imperative. The repercussions of failing to secure ICS extend beyond data breaches, potentially endangering human lives and the environment. This thesis meticulously addresses the omnipresent threat of cyber-attacks on critical infrastructures, centering on the development of an advanced security-privacy framework tailored for ICS. The framework is designed to adeptly detect Generative Adversarial Attacks (GAAs), facilitate explainable attack attribution, and uphold privacy through the implementation of Federated Learning (FL). Innovative solutions presented in this thesis transcend traditional cybersecurity measures, providing comprehensive enhancements for critical infrastructures that include power grids and industrial systems. Beyond merely fortifying model robustness against adversarial attacks, the proposed frameworks are characterized by a steadfast commitment to prioritizing privacy. Moreover, it furnishes effective tools for attack attribution and anomaly prediction within dynamic and complex environments.