Robust fault detection and isolation via a diagnostic observer

Abstract

We present two methodologies for the design of robust fault isolation observer for linear uncertain systems. The proposed fault isolation observer is robust to structural uncertainties by producing disturbance decoupled residuals. The first method uses a direct eigenstructure assignment scheme to accomplish a diagonal transfer function between the faults and the residuals. The second method is carried out through transformation of the linear system under consideration into its special coordinate basis (SCB) form. Once the system is in SCB form, we propose a disturbance decoupled fault detection observer (DDFDO) which is combined either with Beard–Jones detection filter (BJDF) theory, or input estimation results. This will lead to the final proposed robust fault detection filter. Finally, two numerical examples are given in order to illustrate the validity and effectiveness of the proposed fault detection and isolation (FDI) strategy. Copyright © 2000 John Wiley & Sons, Ltd.

Robust multiple‐fault detection and isolation: A gradient flow approach

“…The aim of this example is to show that the proposed gradient flow approach allows one to cope with singular and/or critical cases that cannot be treated by standard and, to our knowledge, more recent eigenstructure assignment methods (e.g. [10,24]). The system is a double-effect pilot plant evaporator described by the following matrices: Owing to the fact that there are three measurements and four independent disturbances and faults, it clearly results that the necessary condition (4) does not hold true.…”

“…As a consequence, the method in [10] is not applicable, while other algorithms (e.g. [24]) allow one to derive robust fault detection observers only. Note that the system is detectable and an admissible observer spectrum is Figure 1, for the sake of completeness, the step responses of residual-to-fault transfer matrices, computed using the algorithm of Section 3.1 and, respectively, the SCB-based design algorithm of Xiong and Saif [24], are shown.…”

“…[24]) allow one to derive robust fault detection observers only. Note that the system is detectable and an admissible observer spectrum is Figure 1, for the sake of completeness, the step responses of residual-to-fault transfer matrices, computed using the algorithm of Section 3.1 and, respectively, the SCB-based design algorithm of Xiong and Saif [24], are shown. As expected, our gradient flow solution has determined a reasonable 'detection and isolation filter' in a case when the exact multiple-fault isolation cannot be obtained.…”

“…As expected, our gradient flow solution has determined a reasonable 'detection and isolation filter' in a case when the exact multiple-fault isolation cannot be obtained. As a matter of fact, the approach of Xiong and Saif [24], based on detection conditions only, is not capable of achieving any degree of fault isolability (see Figure 1, right). Figure 2 illustrates the transient behavior of the scaled temperature parameterT (t) = T (t)/ .…”

“…Step responses of G rf (s) computed via gradient flow algorithm of Section 3.1 (continuous line) and via SCB-based design algorithm of Xiong and Saif[24] (dashed line).…”