Résumé:
Industry 5.0 the latest industrial revolution, advances the Smart Factory concept by emphasizing Human-Machine collaboration and sustainability. It incorporates the human aspect into industrial processes, promoting critical thinking, personalization, and adaptability, while leveraging technologies like IoT and AI for increased efficiency and productivity. However, this era also introduces a complex landscape of cyber threats. As machines, systems, and humans become interconnected, ensuring cybersecurity in smart factories becomes crucial to balance innovation and efficiency with robust security and privacy preservation measures. In response to these challenges, this doctoral research contributes innovative solutions that address the security and privacy vulnerabilities inherent in the Industry 5.0 scenario.
The first contribution of this doctoral research revolves around federated learning methodology (FL) for malware detection based on network analysis. This contribution introduced a cost-effective and efficient approach to deep-learning-based malware detection using FL methodology. This methodology addresses computational overhead and privacy concerns by leveraging network traffic data balancing emerging technologies with security and privacy to mitigate large-scale malware attacks that could undermine Industry 5.0's core principles.
The second contribution puts forth a novel privacy-preserving secure framework called PPSS, integrating blockchain with energy-efficient Proof-of-Federated Deep Learning (PoFDL) consensus protocol to optimize the process of FL in terms of preserving data privacy, enhancing system reliability, and promoting transparency. PPSS adeptly tackles the challenges associated with cyber threat detection and data privacy, specifically within the context of resource-constrained and heterogeneous industrial systems.
The third contribution focuses on developing an efficient, robust, federated cyber threat detection framework for Industrial IoTs. The approach leverages federated learning and generative adversarial networks (GANs) to enhance IDS efficiency, privacy protection, and resilience against adversarial attacks. A federated generative model was employed for data augmentation to limit the attack surface, thereby improving cyber threat detection reliability in the face of zero-day and adversarial threats.
The performance evaluation of the proposed approaches was conducted using a new cyber security dataset named Edge-IIoTset. Specifically designed for cyber threat detection in Industrial IoTs. The results showcase the efficiency and reliability of cyber threat detection under various data distribution modes.
Combining the insights from these contributions, this thesis proposes a comprehensive approach to safeguard Industry 5.0 from cybersecurity threats. Federated deep learning techniques optimize the process of knowledge sharing among participants while protecting data privacy in a resource-efficient manner. Integrating blockchain-enabled intrusion detection systems ensures the integrity and security of data exchanged among IoT-based devices. Deploying generative adversarial networks fortifies the system's resilience against zero-day and adversarial attacks.
