Diagnosis of stochastic systems

Addressing uncertainty is a major challenge for accurate fault detection and diagnosis (FDD), given that most system models are inherently uncertain/imprecise. This chapter examines stochastic diagnostic inference from the first-principles point of view of model inversion. It also examines the three main approaches to stochastic diagnostic inference: model-driven, data-driven and hybrid. The chapter explains the strengths and weaknesses of these approaches, and the prospects for future developments. First-principles model-based methods use physical laws for describing the dynamic behavior of the monitored system to generate a model, and then these methods use an algorithm that inverts that model for diagnostics inference. The chapter describes the main model types and their corresponding inversion algorithms. The approaches are grouped in terms of traditional FDD methods and Bayesian inversion methods.