Securing Industrial Control Systems from Cyber-Attacks: A Stacked Neural-Network based Approach

Abstract

Demanding scientific evolution and undisrupted resource requirement of consumers signified the amalgamation of mechanical production, mass production, and digitalized production for the fourth industrial revolution, “Industry 4.0.” Critical infrastructures that operate and govern industrial sectors and public utilities, such as water desalination plants, smart grids, and gas pipelines, incorporated this cognitive-mechatronic augmentation for the seamless integration of software, control components, and production employees to increase the productivity scale. Although connectivity, automation, and optimization made industrial sectors realize the full potential of smart manufacturing, the inclusion of supervisory control and data acquisition systems into cyberspace expanded the attack vectors that made industrial control systems the prime target for cyber-attackers. Conventional security solutions, such as firewalls, traditional intrusion-detection systems, and antivirus, have been proposed and developed by the research community acted as a proficient line of cyber-defense. However, protecting critical infrastructures from heterogeneous cyber-attacks for resilient operability still pose a significant research challenge. In addition, although machine learning and deep-learning-based intrusion-detection models have been proposed and optimized in the literature, operational viability still poses a significant setback for real-time intrusion detection on industrial control systems. By considering the limitations identified in the literature, a stacked deep-learning model is proposed and validated over laboratory-scale industrial datasets. Furthermore, this article provides an overview of cyber-physical systems, conventional security solutions, and their challenges in identifying unseen exploits. As a concluding remark, JARA: a hybrid opensource deployment-ready intelligent intrusion-detection system, has been presented that feasibly detects the HnS IIoT malware when deployed on a Linux virtual machine.