I.Tumer, NSF CMMI-SYS #1562027 (2016)

Many engineered systems that form the fabric of our society, such as power grids, commercial aircraft, or self-driving cars, are by nature difficult to model and analyze. In such complex systems, many interacting subsystems are often independently designed, managed and engineered. This "design-by-subsystem" approach reduces the cost and complexity of the design process, but prevents the discovery of unexpected, emergent interactions between subsystems until the system is integrated and tested in operation. This award explores new methods of failure analysis for complex engineered systems by focusing on how the failure modes depend on the connections between subsystems and the variables and parameters governing the physics-based behaviors of those subsystems. While in previous research network models of complex engineered systems have been introduced to address this issue, the fundamental limitation remains the exclusive reliance on physical and functional dependencies between components. This research aims to produce a set of indicators to assist engineers in quantifying the failure tolerance of complex engineered systems without going through the expense of a complete system simulation, which in many cases is neither feasible nor practical.

This project specifically introduces a new approach to robust design based on failure tolerance of interdependent networks. The approach explicitly addresses both physical architecture and behavioral, physics-based relations as contributing to the system's tolerance to failure. The methodology incorporates a mesoscale property of complex networks, namely modularity, to understand how the architecture of components and their physics-based relations have an important function of creating stability and failure tolerance. The project will first introduce a new conceptualization of robust design based upon the relational stability between design elements. This represents an important conceptual shift in thinking about robust design, moving away from component reliability alone towards consideration of the connectivity of the behavior of parts. The project will then introduce a network-based methodology and simulation environment capable of calculating the robustness of a complex engineered system. A set of measures will be introduced, which, while being single quantities, holistically capture the failure behavior of the complex engineered system not just at phase transition between nominal behavior and failure but before and after that as well.