Set-based fault diagnosis for uncertain nonlinear systems

Automated fault diagnosis algorithms aim to identify the root cause of a fault after it is detected, which is crucial for determining a safe and effective response. Set-based methods are attractive for this task due to their unique ability to provide formal guarantees that the observed data is inconsistent with candidate fault models. However, such diagnosis methods have scarcely been applied to nonlinear systems because nonlinearity often makes the required inconsistency tests either very conservative or very computationally demanding using existing techniques. This paper proposes a new set-based fault diagnosis algorithm for nonlinear systems and demonstrates improved diagnosis speed and accuracy compared to existing set-based methods. The method tests for inconsistency using a new set-based observer enabled by recent advances in discrete-time reachability analysis.