MAD-VAMCHS
The MAD-VAMCHS project aims to develop systematic methods and tools to identify vulnerabilities and allocate defenses in so-called Cyber-Physical-Human Systems (CPHS). These refer to large-scale, complex cyber-physical infrastructures—such as operational technology (OT) systems—where interaction with human operators plays a crucial role.
The focus is on identifying fundamental, inherent vulnerabilities, rather than implementation-specific flaws, in order to determine which components and data streams must be protected against adversaries capable of observing, learning, and adapting.
The research groups involved in the project have strong expertise in modeling and control of dynamical systems and energy systems, as well as in learning theory and artificial intelligence. The project builds on this expertise and aims to develop methodologies and tools for systematically identifying vulnerabilities and allocating protective measures in CPHS, such as electricity and gas networks.
The technical work is divided into two work packages:
WP1 (Model-based scenario):
In this work package, we assume that both the operator and the attacker have access to accurate models of the physical processes and OT systems. This represents a worst-case scenario and enables the identification of the most critical components requiring protection. We develop a modeling framework and define security metrics that can guide the allocation of protective measures.
WP2 (Learning-based scenario):
In this work package, we assume that the attacker has access only to (potentially leaked) partial data or logs, while the operator has either accurate models or large amounts of historical data. We investigate which vulnerabilities identified in WP1—and potentially additional ones—can also be detected in these scenarios using modern machine learning and AI techniques. For example, the analysis may reveal which datasets are particularly important to protect. Furthermore, we aim to apply explainable AI and sensitivity analysis to support human operators in understanding and effectively utilizing the results of the data-driven analysis.
