Safe Human-Robot Interaction

Qin Lin, PhD | Biography
Department of Electrical Engineering and Computer Science, AIR Lab

Keywords: AI, Robotics, Safe Control, Cyber Security

Description

The protection of industrial control systems (ICS) for public infrastructures such as power, water treatment, and transportation systems is of utmost importance due to significant damage an attack may cause. Machine learning techniques can be a powerful tool to profile normal physical process of an ICS for further intrusion detection. Our new tool, called automata learning, can deliver interpretable model and detection results. The abnormal sensors or actuators can be localized due to the interpretability.

Business Applications
  • Digital Twin Modelling: Data-driven approach for reverse engineering of legacy industrial control systems
  • Intrusion Detection: Anomaly detection for cyber attacks
  • Recovery Control: Resilient control for industrial control systems, recovery control from cyber attacks
 
Case Study

 

  • We have successfully validated our tool in a real industrial control system testbed called Secure Water Treatment System (SWAT) in Singapore University of Technology and Design.
  • Our model learning and detection are in real-time. Compared with state-of-the-art machine learning models such as SVM and deep neural networks, we have a higher detection success rate and significantly less training time.
  • False alarms and shutting down industrial control systems for a further check are costly in practice. Our graphical model captures temporal and causality features in the system. The model and the detection results are understandable and explainable to system operators. They can also validate our model using domain knowledge.