Data-driven fault identifiability analysis for discrete-time dynamic systems

Abstract

Fault identifiability plays a key role in the fault diagnosability performance analysis of a system. However, considering the need for an accurate model, the design of control systems and fault diagnosis systems based on fault identifiability performance suffers from a limited application range because accurate models are rarely available for complex processes or plants, whereas substantial quantities of offline and online data can be obtained from systems with ease. This situation constitutes the motivation to develop a data-driven analysis method for determining the identifiability of faults in dynamic systems. The fault identifiability analysis problem is reformulated as the quantification of the difficulty associated with estimating an unknown value from a regression analysis perspective. Moreover, the proposed identification scheme for a stable kernel representation provides a direct way to identify the matrices required to compute the proposed measure, the estimability, which effectively reduces engineering costs in practical applications. Finally, the proposed method is applied to a vehicle lateral dynamic system to exemplify how to analyse the fault identifiability performance of a dynamic system.