MOMENTUM-OPTIMIZED FEDERATED DEEP LEARNING FOR INTRUSION DETECTION IN INDUSTRIAL CONTROL SYSTEMS: AN ARCHITECTURAL EVALUATION

Abstract

Industrial Control Systems (ICS) are critical infrastructures designed to monitor and control industrial processes through field elements such as sensors, actuators, and machinery.With the increasing integration of IoT devices into ICS networks, the landscape of cybersecurity threats has expanded significantly, necessitating advanced Intrusion Detection Systems (IDS).While IDS has been widely studied, its application in ICS remains underexplored.ICS environments use specialized protocols and were traditionally isolated, relying on obscurity for security.However, increasing cyber threats have driven ICSs to integrate with IT infrastructure, introducing new vulnerabilities.Most IDS implementations are centralized, whereas traditional centralized IDS offers comprehensive data analysis but suffers from scalability issues, latency, and privacy concerns.Distributed IDS improves scalability and local responsiveness but faces challenges in data availability and model consistency.Federated IDS emerges as a promising alternative, enabling decentralized model training while preserving data privacy and enhancing robustness through collaborative learning.This paper investigates the comparative performance of centralized, distributed, and federated architectures in Intrusion Detection Systems tailored for ICS Sajath P Sathar