Semantic-assisted Anomaly Detection for Cyber-Physical Systems

Oct 28 2024 12:00-1:00pm. WWH 335

Dr. Chenglong Fu

Software and Information Systems

Abstract:

Modern critical industrial infrastructures increasingly rely on Cyber-Physical Systems (CPS), which enable advanced features like remote and automated control. However, the integration of CPS introduces significant risks, as these systems are potential targets for cyber-attacks that could result in catastrophic consequences. To mitigate these risks, various anomaly detection methods have been developed to help operators identify and address security issues before severe damage occurs. While effective, these methods often require domain-specific expertise to manually define and extract physical invariants that characterize normal CPS behavior, making them costly and difficult to scale. To address these limitations, we propose developing a novel method to enhance CPS security and resilience by providing a more efficient, automated approach to threat detection. Our method leverages recent advancements in large language models (LLMs) to infer physical invariants directly from CPS specification documents, which are then integrated with deep learning-based correlation mining techniques. By utilizing the built-in knowledge of physics and engineering embedded in pre-trained LLMs, this approach aims to reduce the reliance on manual expertise in extracting physical invariants, thereby improving scalability and reducing costs associated with CPS anomaly detection.

Dr. Chenglong Fu is Assistant Professor of Software and Information Systems at UNC Charlotte’s College of Computing and Informatics.