Fault detection method based on bounded error and dynamic threshold techniques

Abstract

This work presents a new fault detection method applicable to dynamic systems. This method is based on historic system input-output data and on the so-called residual signal. At each discrete time instant, the detection system checks history to locally approximate the dynamics of the faultless system and thus, obtain a prediction of the system output to calculate the residual. Likewise, the maximum threshold allowed by this residual is obtained by applying bounded-error techniques. The main contribution of this paper is the expression that allows this threshold to be obtained. Detection is based on checking whether the residual exceeds the maximum threshold determined by the method.

Set‐membership methods applied to FDI and FTC

“…In order to provide a state of the art picture of the development of set‐membership methods applied to FDI and FTC, this issue places the focus on transitioning from theoretical to practical studies. It is organized according to a double progression by goals and by techniques: the papers deal with FDI, and the papers address FTC objectives, while the main techniques used are successively related to paving methods , parameter estimation and identification , observation , control , and Networked Control Systems (NCS) .…”

“…Suárez Fábrega et al , use historical datasets from a faultless system to derive a fault detection method: residuals resulting from local affine models as well as dynamic thresholds are obtained. The fault detection of nonlinear systems working around and/or transitioning between operating points is considered and validated with a two‐tank application.…”