Synthetic Fault Mode Generation for Resilience Analysis and Failure Mechanism Discovery

Traditional risk-based design processes seek to mitigate operational hazards by manually identifying possible faults and devising corresponding mitigation strategies—a tedious process which critically relies on the designer’s limited knowledge. In contrast, resilience-based design seeks to embody generic hazard-mitigating properties in the system to mitigate unknown hazards, often by modelling the system’s response to potential randomly generated hazardous events. This work creates a framework to adapt these scenario generation approaches to the traditional risk-based design process to synthetically generate fault modes by representing them as a unique combination of internal component fault states, which can then be injected and simulated in a model of system failure dynamics. Based on these simulations, the designer may then better understand the underlying failure mechanisms and mitigate them by design. The performance of this approach is evaluated in a model of an autonomous rover, where cluster analysis shows that elaborating the faulty state-space in the drive system uncovers a wider range of possible hazardous trajectories and failure consequences within each trajectory than would be uncovered from manual mode identification. However, this increase in hazard information gained from exhaustive mode sampling comes at a high computational expense, highlighting the need for advanced, efficient methods to search and sample the faulty state-space.