| dc.description.abstract | We interact with a variety of Cyber Physical Systems (CPSs) in our everyday lives directly through consumer goods such as automobiles and smart homes as well as indirectly through examples including energy infrastructure and manufacturing systems. CPSs have further potential in a variety of applications with commonly identified grand challenge problems including advanced/smart power grids, autonomous transportation, and improved biomedical and healthcare systems. However, traditional, analytically-based design techniques often fall short for addressing these problems and providing the high level of autonomy required, leading to the introduction of data-driven techniques such as machine learning to the CPS domain. While machine learning is a well-studied field, existing applications are largely in non-safety critical applications where the consequences of failure are relatively low. Introduction of these techniques to the safety- and mission-critical applications common in the CPS domain, combined with the increasing scale and complexity of CPS design, has exposed the need for new methods of safety assurance.
This dissertation presents several related model-based safety assurance techniques including: (1) A formal analysis technique which uses Colored Petri Net models to verify timing constraints of real-time, component-based software, (2) A CPS development platform which integrates safety assurance processes into the model-based development cycle with special consideration for the demands of data-driven software development, (3) A technique for automated construction of safety assurance arguments based on the instantiation and composition of argument patterns with information sourced from existing system design models, and (4) A methodology for modeling potential hazard escalation paths and dynamic estimation of the risk posed by these hazards which is used to periodically reevaluate assurance arguments as a system operates under changing internal and environmental conditions. | |
