Cybersecurity in Industrial Networks: Artificial Intelligence Techniques Applied to Intrusion Detection Systems

Abstract

Industrial control systems (ICS) operate on serialbased networks which lack proper security safeguards by design. They are also becoming more integrated to corporate networks, creating new vulnerabilities which expose ICS networks to increasing levels of risk with potentially significant impact. Despite those risks, only a few mechanisms have been suggested and are available in practice as cybersecurity safeguards for the ICS network layer, maybe because they might not be commercially viable. Intrusion detection systems (IDS) are typically deployed in the corporate networks to protect against attacks since they are based on TCP/IP. However, IDS are not used in serialbased ICS networks yet. This study examines and compares modern Artificial Intelligence (AI) techniques applied in IDS that are potentially useful for serial-based ICS networks. The results showed that current AI-based IDS methods are viable in such networks. A mix of AI techniques would be the best way forward to detect known attacks via rules and novel attacks, not previously mapped, via supervised and unsupervised techniques. Despite these strategies' limited use in serial-based networks, their adoption could significantly strengthen cybersecurity of ICS networks.