(199g) Input Design for Active Fault Diagnosis: Recent Developments, New Results, and Future Directions

Faults can occur in any real-world system, and reliable diagnosis has become increasingly complex. This is largely owing to factors such as growing system complexity and increasingly stringent requirements on safety, availability, and performance. Traditional methods for fault detection and diagnosis rely on nominal input-output data, which can contain insufficient information to support reliable conclusions. Recent years have witnessed a growing interest in active fault diagnosis, which addresses this issue by injecting input signals specifically designed to reveal the fault status of the system.

In this work, we provide an overview of state-of-the-art methods for input design for active fault diagnosis and discuss the primary considerations in the formulation and solution of the input-design problem. Based on a recent review of the field (Heirung and Mesbah, 2019), we classify the modern methods into three separate classes: probabilistic, set-based, and energy-bounded. Probabilistic methods are commonly applied when a system is subject to stochastic measurement noise and disturbances or when parametric uncertainty can be specified with probability distributions (e.g., Blackmore and Williams, 2006). These probabilistic design methods generally involve minimizing or eliminating the overlap between predicted output distributions, often with the goal of minimizing the probability of misdiagnosis. Set-based methods can be applied when the uncertain quantities are described using sets, such as polytopes and zonotopes (e.g., Nikoukhah, 1998; Scott et al., 2014). These approaches typically involve the design of inputs that fully separate the output prediction sets, which enable guaranteeing diagnosis. Finally, the methods that involve energy-bounds on the uncertainty (Campbell and Nikoukhah, 2004) can be applied when the uncertainty is primarily in the power density of the exogenous signals.

Based on this classification, we contrast solution methods for the most common case of linear models. We further present variations and extensions to the problem of active fault diagnosis, including the significantly more difficult case of nonlinear models, which is receiving increasing attention. Finally, we suggest avenues for future research in this rapidly evolving field.

Blackmore, L. and Williams, B.C. “Finite horizon control design for optimal discrimination between several models.” In Proceedings of the IEEE Conference on Decision and Control, pages 1147–1152. San Diego, CA, 2006.

Campbell, S.L. and Nikoukhah, R. “Auxiliary Signal Design for Failure Detection.” Princeton University Press, 2004.

Heirung, T.A.N. and Mesbah, A. “Input design for active fault diagnosis.” Annual Reviews in Control, in press, 2019.

Nikoukhah, R. “Guaranteed active failure detection and isolation for linear dynamical systems.” Automatica, 34(11):1345–1358, 1998.

Scott, J.K., Findeisen, R., Braatz, R.D., and Raimondo, D.M. “Input design for guaranteed fault diagnosis using zonotopes.” Automatica, 50(6):1580–1589, 2014.