FIND ME ON

GitHub

LinkedIn

Design of Observer-based controllers

🌱

Control

Consider the LTIC system x˙(t)=Ax(t)+Bu(t)y(t)=Cx(t)+Du(t)\begin{align*} \dot{x}(t)&=Ax(t)+Bu(t)\\ y(t)&=Cx(t)+Du(t) \end{align*}Then the observer-based controller u(t)=Fx^(t)x^˙(t)=(A+LC+BF+LDF)x^(t)Ly(t)\begin{align*} u(t)&=F\hat{x}(t)\\ \dot{\hat{x}}(t)&=(A+LC+BF+LDF)\hat{x}(t)-Ly(t) \end{align*}is s.t. the closed-loop system exactly has its eigenvalues at the eigenvalues of A+BFA+BF and A+LCA+LC.

Stabilizability of (A,B)(A,B) and detectability of (C,A)(C,A) suffice to ensure Global Asymptotic Stability i.e. x(t)0 as tx(0)=x0x(t)\to0\text{ as }t\to \infty\quad \forall x(0)=x_{0}Note that the state feedback u(t)=Fx^(t)u(t)=F\hat{x}(t) and the observer component appear in a decoupled fashion, so they can be designed separately. This property is commonly called the separation principle.

The following result show how FF and LL in can be derived >[!thm] >(A,B)(A,B) is stabilizable if and only if there exist some P>0P>0 and zz such that AP+PA+Bx+zB<0AP+PA^{\top}+Bx+z^{\top}B^{\top}<0 Then u(t)=zP1x(t)u(t)=zP^{-1}x(t)where zP1=FzP^{-1}=F!