Attacks on Data-Driven Process Monitoring Systems: Subspace Transfer Networks

• Published in: IEEE transactions on artificial intelligence Vol. 3; no. 3; pp. 470 - 484
• Main Authors: Jiang, Xiaoyu, Ge, Zhiqiang
• Format: Journal Article
• Language: English
• Published: IEEE 01.06.2022
• Subjects: Adversarial attack; Artificial intelligence; data-driven models; Generators; Perturbation methods; poisoning attack; Process control; Process monitoring; process monitoring system (PMS); Sensors; subspace transfer; Training
• ISSN: 2691-4581; 2691-4581
• DOI: 10.1109/TAI.2022.3145335

With the rapid development of information technology, intelligent upgrading of the manufacturing industry has broken the closed environment of traditional industrial control systems (ICS); thus, the information security of ICS has been seriously threatened. As part of ICS, the process monitoring system (PMS) is heavily subject to external risks. Data-driven PMSs have been widely used as initial lines of defense to ensure ICS safety. Once the PMS is under attack, the consequences on the whole ICS will be unimaginable. Unfortunately, the safety issues of the PMS have received inadequate attention. This article reveals PMS's vulnerabilities through effective attacks. A novel method called subspace transfer network (STN) is proposed to conduct adversarial and poisoning attacks on the PMS simultaneously. Then the attack task flow is defined and explained to make online adversarial attacks and data poisoning on PMS. Meanwhile, aiming at two poisoning goals, targeted and untargeted attacks of STN are designed, respectively. Finally, the PMS's fragility is verified in two industrial benchmarks.