No label defined (Q4525864)

A complete procedure for designing robust residual generators in the frequency domain is proposed. Consider a class of systems as follows: \[ \begin{aligned} \dot x&= A(\theta)x+B(\theta)u+E(\theta)f+G(\theta)d,\\ y &=C(\theta)x+D(\theta)u+F(\theta)f+H(\theta)d, \end{aligned} \] where \(x\in \mathbb{R}^n\), \(y\in \mathbb{R}^{n_y}\), \(u\in \mathbb{R}^{n_u}\), \(f\in \mathbb{R}^n\), \(d\in \mathbb{R}^{n_d}\) is the state, the output, the control, the fault input (actuator and sensor failures), the disturbances (measurement noises, structured uncertainties) vector, respectively, \(A,\dots, H\) are appropriately dimensioned matrices. \((A,C)\) is supposed observable. \(\theta\) is an unknown vector characterizing parameter uncertainties. It belongs to a convex set \(\Theta =\{\theta :\theta\in \mathbb{R}^p\), \(\theta^-\leq\theta\leq\theta^+\}\). Unknown inputs are supposed bounded, i.e., \(d\in\Omega_d\). Consider a nominal system given by the matrices \(A^o,\dots , H^o\) for a fixed \(\theta\). First, the residual generator is constructed by using an \(H_{\infty}\) parametrized observer for the nominal system: \(\dot{\widehat x}=A^o\widehat x+B^ou+L(y-C^o\widehat x)\), \(r=y-C^o\widehat x\), where \(r\) is the residual vector. The matrix \(L\) is determined to guarantee the requirements: a) asymptotic stability for \(f=0\); b) minimization of disturbance effects; c) maximization of fault effects, all within the frequency range \(\omega\in [\omega_o, \omega_f]\). The matrix \(L\) is designed by using the LMI approach. The faults are detected by using the evaluation function \(J(r)=({1\over{\omega_f-\omega_o}}\int_{\omega_o}^{\omega_f}\|r(j\omega)\|d\omega)^{1/2}\). Then, parameter uncertainties are considered. Their worst effects on residuals \(r\) is evaluated by computing the optimal threshold signal, i.e., \(J_{th}=\)sup\(_{\theta\in\Theta, d\in\Omega_d, f=0} \{J(r)\}\) depending on the operating conditions. This is achieved by using Kharitonov polynomials and graphical techniques in the associated uncertain Nyquist diagram. A simulation example is supplied.