Högskolan i Skövde

The growth and the complexity scale of Cyber-Physical Systems (CPSs) are ever-evolving due to the fast expansion of networked applications in smart-x systems, which are overseeing critical infrastructures such as the smart-grid. These smart networked systems use a network of embedded sensors, platforms and actuators to perceive and affect a physical process that typically requires guaranteed quality-of-service performances provided by safety-critical applications. The confluence of sensors, platforms and networks is also nourishing the expansion of the emerging Internet of Things (IoT) area. However, these developments lead to increased surfaces that are vulnerable to cyberattacks.Since the capability of attackers and the trust in networked-components are subject to substantial variability, a dynamic-vulnerability assessment is advocated in this study, in contrast to traditional static-approaches. 

Recent advances in data analytics prompt dynamic data-driven vulnerability assessments, whereby data contained and produced by CPS cyber-components include hidden traces of vulnerability fingerprints. However, the imprecise nature of vulnerability assessment and the huge volume of scanned data call for computational intelligence techniques to analyse such data. We first investigate computational models to capture semantic properties related to vulnerability concepts revolving around CPS components. This study reveals salient metrics and related measurements used to quantify CPS component vulnerabilities. We show the potential of applying fuzzy-logic techniques to diagnose vulnerability, and infer objective vulnerability scores. Then, we examine computational methods to extract meaning from text by mining online public-repositories of published vulnerabilities and discovering potential vulnerability-matches in a given CPS infrastructure. Graph-mining techniques are also explored to identify critical-assets of CPS infrastructure to weigh vulnerabilities, considering topological structures and functional features. 

In this proposal, we explore the state of the art and highlight the drawbacks of current research approaches in CPS vulnerability assessment area, based on which, we build our research questions with the purpose to piece together solution elements for the stated problem. In doing so, computational intelligence techniques such as fuzzy-logic and machine-learning, are investigated in order (a) to reduce existing security management gaps induced by ad-hoc and subjective vulnerability auditing processes, (b) to narrow further the risk window induced by discoverable vulnerabilities, and (c) to increase the level of automation in vulnerability analysis, at various levels of the CPS architecture.