next up previous
Next: Stability Analysis of Nonlinear Up: DEsystem Previous: Matrix Exponential Function

Nonhomogeneous DE System

Now we consider the general solution of a nonhomogeneous ODE system with non-zero input ${\bf x}(t)$ on the right-hand side:

\begin{displaymath}
\dot{\bf y}(t)={\bf Ay}(t)+{\bf x}(t),\;\;\;\;\mbox{or}\;\;\;\;
\dot{\bf y}(t)-{\bf Ay}(t)={\bf x}(t)
\end{displaymath}

Multiply both sides by $e^{-{\bf A}t}$:

\begin{displaymath}
e^{-{\bf A}t}\dot{\bf y}(t)-e^{-{\bf A}t}{\bf Ay}(t)
=\frac{...
...eft[ e^{-{\bf A}t} {\bf y}(t) \right] =e^{-{\bf A}t}{\bf x}(t)
\end{displaymath}

Integrate both sides

\begin{displaymath}
\int_0^t\frac{d}{d\tau}\left[ e^{-{\bf A}\tau} {\bf y}(\tau)...
...(t)-{\bf y}(0)
=\int_0^t e^{-{\bf A}\tau}{\bf x}(\tau) \;d\tau
\end{displaymath}

Multiply $e^{{\bf A}t}$ on both side:

\begin{displaymath}
{\bf y}(t)=e^{{\bf A}t}{\bf y}(0)+e^{{\bf A}t}\int_0^t e^{-{...
...t}{\bf y}(0)+\int_0^t e^{{\bf A}(t-\tau)}{\bf x}(\tau) \;d\tau
\end{displaymath}

where the first term $e^{{\bf A}t}{\bf y}(0)={\bf V}e^{{\Lambda}t}{\bf V}^{-1}{\bf y}(0)$ is the homogeneous solution of the ODE system $\dot{\bf y}={\bf Ay}$, and the second term is the particular solution.

Example 1:

\begin{displaymath}
\ddot{\bf y}+2\zeta\omega_n\dot{\bf y}+\omega_n^2{\bf y}=x(t)
\end{displaymath}

Define $v=\dot{y}$,

\begin{displaymath}
\frac{d}{dt}\left[\begin{array}{c}y v\end{array}\right]
\l...
...rray}\right]+
\left[\begin{array}{c}0 x(t)\end{array}\right]
\end{displaymath}

or in matrix form:

\begin{displaymath}
\dot{\bf y}={\bf Ay}+{\bf x}
\end{displaymath}

Given $\omega_n=10$ and $\zeta=0$, we have

\begin{displaymath}
{\bf A}=\left[\begin{array}{cc}0 & 1 100 & 0\end{array}\right]
\end{displaymath}

with eigenvalue and eigenvector matrices:

\begin{displaymath}
{\bf D}=\left[\begin{array}{cc}10 j & 0 0 & -10\;j\end{ar...
...left[\begin{array}{cc}1 & 1 10 j & -10\;j\end{array}\right]
\end{displaymath}


\begin{displaymath}
e^{{\bf A}t}={\bf V}e^{{\bf D}t}{\bf V}^{-1}
=\left[\begin{a...
...10 t)/10\\
-10 \sin(10 t) & \cos(10 t) \end{array}\right]
\end{displaymath}

Given $y(0)=0$ and $\dot{y}(0)=1$, we get

\begin{displaymath}
y(t)=[1\;0]e^{{\bf A}t}\left[\begin{array}{c}0\ 1\end{array...
...f A}t}
\left[\begin{array}{c}0\ x(t)\end{array}\right]\;d\tau
\end{displaymath}

Example 2:

The motion of a tuned mass damper system shown below:

TunedMDS.png

can be described by the following two ODEs:

\begin{displaymath}
\left\{ \begin{array}{l}
M\ddot{y}=-Ky-B\dot{y}-K_d(y-y_d)-...
...dot{y}_d=K_d(y-y_d)+B_d(\dot{y}-\dot{d}_d)
\end{array}\right.
\end{displaymath}

If we define $v=\dot{y}$ and $v_d=\dot{y}_d$, then we get

\begin{displaymath}
\left\{
\begin{array}{l}
\dot{v}=\ddot{y}=[-Ky-B\dot{y}-K_d(...
...}_d=[K_d(y-y_d)+B_d(\dot{y}-\dot{d}_d)]/M_d
\end{array}\right.
\end{displaymath}

and we have

\begin{displaymath}
\frac{d}{dt}\left[\begin{array}{d}y v y_d v_d\end{arra...
...t[\begin{array}{d}0 \frac{F(t)}{M} 0 0\end{array}\right]
\end{displaymath}

which can be written in matrix form as

\begin{displaymath}
\dot{\bf y}(t)={\bf Ay}(t)+{\bf x}(t)
\end{displaymath}

where ${\bf x}=[0,\;\delta(t)/M,\;0,\;0]^T$.

We assume zero initial condition $y(t)=\dot{y}=v(t)=0$ and $y_d(t)=\dot{y}_d=v_d(t)=0$, and $F(t)=\delta(t)$, then we get the solution

\begin{displaymath}
{\bf y}={e^{\bf A}t}{\bf y}(0) {\bf Ay}+\int_0^te^{{\bf A}\t...
...A}t}\left[\begin{array}{c} 0 1/M 0 0 \end{array} \right]
\end{displaymath}

The first component of ${\bf y}(t)$ is:

\begin{displaymath}
y(t)=[1\;0\;0\;0]\;{\bf y}(t)=[1\;0\;0\;0]
e^{{\bf A}t}\left[\begin{array}{c} 0 1/M 0 0 \end{array} \right]
\end{displaymath}

which is the 1st row and 2nd column of $e^{{\bf A}t}/M$; and the third component of ${\bf y}(t)$ is:

\begin{displaymath}
y_d(t)=[0\;0\;1\;0]\;{\bf y}(t)=[0\;0\;1\;0]
e^{{\bf A}t}\left[\begin{array}{c} 0 1/M 0 0 \end{array} \right]
\end{displaymath}

which is the 3rd row and 2nd column of $e^{{\bf A}t}/M$;

The problem can also be solved using the Laplace transform method. Let $Y(s)={\cal L}[\;y(t)\;]$ and $Y_d(s)={\cal L}[\;y_d(t)\;]$, then ${\cal L}[\;\dot{y}(t)\;]=s Y(s)-y(0)=s Y(s)$ and ${\cal L}[\;\ddot{y}(t)\;]=s^2 Y(s)$ ${\cal L}[\;\dot{y}_d(t)\;]=s Y_d(s)$ ${\cal L}[\;\ddot{y}_d(t)\;]=s^2 Y_d(s)$.

Now the two DEs can be expressed in s-domain as:

\begin{displaymath}
\left\{\begin{array}{l}
Ms^2Y=-KY-BsY-K_d(Y-Y_d)-B_ds(Y-Y_d)+1\\
M_ds^2Y_d=K_d(Y-Y_d)+B_d s(Y-Y_d)\end{array}\right.
\end{displaymath}

Find the sum of the two equations:

\begin{displaymath}
Ms^2Y+M_ds^2Y_d=-(K-Bs)Y+1,\;\;\;\;\mbox{i.e.}\;\;\;\;
(Ms^2+Bs+K)Y+M_ds^2Y_d=1
\end{displaymath}

and rewrite the second equation as:

\begin{displaymath}
(M_ds^2+B_d s+K_d) Y_d=(K_d+B_d s) Y\;\;\;\;\;\mbox{i.e.}\;\;\;\;
Y_d=\frac{K_d+B_d s}{M_ds^2+B_d s+K_d}Y
\end{displaymath}

Substituting into the other equation we get

\begin{displaymath}
(Ms^2+Bs+K)Y+\frac{M_ds^2(K_d+B_d s)}{M_ds^2+B_d s+K_d}Y=1
\end{displaymath}

Solving for $Y$ we get

\begin{displaymath}
Y=\frac{M_ds^2+B_d s+K_d}{ (M_ds^2+B_d s+K_d) (Ms^2+Bs+K)+M_ds^2(K_d+B_ds)}
\end{displaymath}

Note that when $M_d=0$, the system becomes a regular damped harmonic oscillator:

\begin{displaymath}
Y=\frac{B_d s+K_d}{ (B_d s+K_d) (Ms^2+Bs+K)}=\frac{1}{Ms^2+Bs+K}
\end{displaymath}

The motion of mass $M_d$ is:

\begin{displaymath}
Y_d=\frac{K_d+B_d s}{M_ds^2+B_d s+K_d}Y
=\frac{K_d+B_d s}{ (M_ds^2+B_d s+K_d)(Ms^2+Bs+K)+M_ds^2(K_d+B_ds)}
\end{displaymath}


next up previous
Next: Stability Analysis of Nonlinear Up: DEsystem Previous: Matrix Exponential Function
Ruye Wang 2019-02-21