The Behavior of Dynamic Systems: 1.0 Where Are We in The Course?

The Behavior of Dynamic Systems
Professor Sanjay Sarma

December 4, 2007
1.0 Where are we in the course?

In the first two-thirds of the course, we concentrated on the generation of differential
equations for systems consisting of rigid bodies, springs and dashpots. In the last third of
the course, we look at the solutions of these equations.
System
Kinematics
Kinetics & Constitutive
Particle
September
September
System of particles
September
October
Rigid Bodies
September
October
Lagrangian formulation
November
Oscillations
December
2.0 Solving the Differential Equations: Overview

Generally, dynamic systems yield non-linear 2nd-order ODEs. Thats the good news:
no partial differential equations and no third order or higher equations.
Non-linear systems hide insights. The bad news is that these equations are often non
linear. That would have been especially bad news fifty years ago when computing was
still in its infancy. Today, we can rewrite any ODE in such a way that it can be solved
by brute-force methods using advanced computers. The only problem is it is hard to
draw generalizable insights from these brute-force numerical techniques. (Phase plots
are a feeble attempt to draw some insights from non-linear equations.)
Local linear analysis yields rich insights. A theorem called the Hartman-Grobman The
orem (which you can forget about now that you have read this) tells us that if we linearize
a non-linear equation about an equilibrium point, then insights about stability etc. that we
can draw from local behavior (read small perturbations) are valid for the underlying non
linear system. This fact gives us a path forward and makes available the vast landscape of
tools available for the analysis of linear systems.
Linearization involves two steps. First, find all the equilibrium points. Second, intro
duce small perturbations about each equilibrium point that you are interested in and
rewrite the equations of motion in terms of these perturbations, remembering along the
way to strike out any quadratic terms.
Solve the linear system. There is a large body of work for solving systems of linear equa
tions with constant coefficients. The way forward is easy.
December 4, 2007
Cite as: Sanjay Sarma, Nicholas Makris, Yahya Modarres-Sadeghi, and Peter So, course materials for 2.003J/1.053J
Dynamics and Control I, Fall 2007. MIT OpenCourseWare (http://ocw.mit.edu), Massachusetts Institute of Technology.
Downloaded on [DD Month YYYY].
1. Regardless of whether there is a forcing function, start with the characteristic

equation.
Inserting an exponential solution as a guess always yields a characteristic equa
tion. The roots of the characteristic equations are all you need to proceed forward.
They tell you if the system is stable, marginally stable or unstable. They tell you if
the system is oscillatory and if so, what the frquency is. Oh, and they tell you how
quickly the system dies down (i.e., the damping factor).
If the system is free, reconstitute the exponential solution with the roots in place in
the exponent and solve for the initial conditions.Write out the solution of the homog
enous equation.
If the system is oscillatory, it will come out automatically in the equation because
there will be imaginary parts in the exponent.
2. If the system is forced, determine the roots from the characteristic equation as
before.
But then, guess a particular solution. We will limit our analysis to harmonic excita
tion of the form p o e
it
. The guess will always be of the same form with a phase
lag thrown in: x( t ) = Ue
it i
Add the homogeneous and solution and the particular solution, plug in the initial
conditions, and you are ready to go!
3.0 Linearization
Consider the inverted pendulum system we discussed in
class, shown in Figure 1. The equation of motion is:
2
2
ml + ka sin cos mgl sin = 0 .
(EQ 1)
3.1 Equilibrium points
We find the equilibrium points by setting and to
eq
zero and solving for . (That is the definition of an

equilibrium point.) Solving gives us the following equi
librium points:
FIGURE 1. Inverted pendulum
eq
1 = n . In other words, when the pendulum is

perfectly vertical pointing either upwards or downwards.
December 4, 2007
eq
2, 3 = acos mgl
--------- . in other words, when the pendulum leans to the left or right enough
2
ka
that torque from gravity exactly counterbalances the torque from the spring. Keep in
mind that we are assuming that:
mgl
--------- 1 ;
2
ka
(EQ 2)
otherwise the spring is too weak to ever counter gravity.

In class, we guessed at the phase plot (correctly) as shown in Figure 2
mgl
acos ------- 2
ka
acos mgl
------- 2
ka
/2
/2
FIGURE 2. Phase plot for the spring-loaded inverted pendulum
3.2 Small Perturbations

As described before, linearization is useful around equilibrium points. Simply put, for
each equilibrium point, create a new perturbation variable which we will asume to be
small. Rewrite the equations of motion in terms of these perturbation variables, perform
ing the following steps along the way:
a)Use Taylor Series to expand any function to the first order (linear terms).
b)Delete any quadratic terms that arise in terms of these perturbation variables or their
derivatives.
In the case of the inverted pendulum, this is straightforward. Consider any equilibrium
eq
eq
point i and introduce a perturbation variable i such that = i + i . Here are some
simplifications for any equilibrium point:
sin (
eq
(EQ 3)
(EQ 4)
+ ) sin
eq
+ cos
December 4, 2007
eq
(EQ 5)
3
cos (
When
eq
eq
eq
+ ) = cos sin
eq
(EQ 6)
= 0 , we can see that

sin(0 + ) sin 0 + cos 0
(EQ 7)
and
cos (
eq
+ ) = cos 0 sin0 1 .
Remember: Use Equations 7 and 8 only when
eq
(EQ 8)
= 0 !!! With these tools, we can linear

eq
ize the equations of motion for the inverted pendulum. Lets linearize about 1 = 0 .
Equation 1 simplifies to:
2
2
ml + (ka mgl) = 0 .
(EQ 9)
3.3 Stability
We test for stability by simply generating the characteristic equations and looking at the
roots. Where does the characteristic equation come from? It comes from using a guess
solution of the form ( t ) = Ae
2 2
in the equation of motion. Inserting it into Equation 9

2
for example gives us: ml + (ka mgl) = 0 . The solutions of this characteristic
equation are:
2
(ka mgl)- .
= ---------------------------2
ml
But wait, Condition 2 tells us that must be imaginary. This is good news because purely
imaginary roots imply that the system is oscillatory and therefore marginally stable.
Heres the summary:
If the real part of the roots is positive, the system is unstable.
If the real-part of the roots is zero, the system is marginally stable.
If the real-part of the system is negative, the system is asymptotically stable.
If the imaginary part of the roots is non-zero, the system will display oscillatory behav
ior.
December 4, 2007
4.0 The Free Response of Linear Systems
All linearized systems of the type we have

looked at in this course will eventually reduce
to mass-spring-damper systems of the type
shown in Figure 3 with an equivalent mass,
equivalant damping and an equivalent stiffness.
The equation of motion of such a system is sim
ply:
mx + cx + kx = 0 .
k
m
c
FIGURE 3. Mass-spring-damper system
(EQ 10)
For example, the linearized inverted pendulum is simply a spring-mass-damper system of

2
equivalent mass ml and equivalent spring stiffness (ka mgl) . This system has no
damping. Equation 10 is therefore the canonical system for all single-degree-of-freedom
linear systems of the sort we expect to see in our analysis. Equation 10 has constant coef
ficients and it is a homogenous equation because the right-hand-side is zero.
The solution approach to this canonical equation is a repeat of what we saw in Section 3.3;
we are doing exactly the same thing but now for the general system. Proposing a solution
t
of the form x( t ) = Ae , the characteristic equation comes out to be:

2
m + c + k = 0 .
(EQ 11)
For systems where c=0, we saw that the natural frequency will be:
k- .
n = + --m
(EQ 12)
We will use this term n to simplify our writing here-forward. We call it the natural fre
quency of the system or the undamped natural frequency of the system. Dividing both
sides of Equation 11 by m, the equation reduces to:
2
c
k
2
c
2
+ ---- + ---- = + ---- + n = 0 .
m
m
m
(EQ 13)
c
We now introduce some new terminology. We define = -------------- as the damping factor.
2 km
Take it on faith for the moment. Equation 13 reduces to:
2
+ 2 n + n .
(EQ 14)
The solutions this quadratic equation are:
December 4, 2007
2 n 4( n ) 4 n
2
= ---------------------------------------------------------------- = n n 1 .
2
(EQ 15)
For positive and n the real-parts of the roots, as you can see, will always be less than
zero. So if we have a nice system, stability is assured. Now lets understand the behavior
in more detail.
If < 1 , then the system will have complex roots with negative real parts.
If = 1 , then the system will have purely real and repeated negative roots.
If > 1 , then the system will have purely real and distinct negative roots.
Each of these situations results in different behavior. Lets examine them.
4.1 Underdamped systems: <1
If < 1 , the roots are imaginary. We can force the issue by rewriting as:
= n i n 1 . Furthermore, if we define d = n 1 , where we call

d the damped natural frequency, then Equation 15 can be written simply as:
= n i d .
(EQ 16)
t
Inserting this back into our guessed solution x( t ) = Ae , we get the general solution as:
x( t ) = A 1 e
( n )t + i d t
+ A2 e
( n )t i d t
= e
( n )t
[A 1 e
i d t
+ A2 e
i d t
];
(EQ 17)
in other words, one term for each root. This can also be converted into sines and cosines
using Eulers Formula to get something of the form:
x( t ) = e
( n )t
[B 1 cos d t + B 2 sin d t] .
(EQ 18)
The insight here is that you get a sinusoidal curve of exponentially diminishing amplitude.
The circular frequency of the sinusoid is d .
Inserting the initial conditions x(0) = u o and x (0) = v o and solving for B1 and B2 we
get the particular solution:
x( t ) = e
( n )t
v o + n u o
--------------- sin d t .
u 0 cos d t + -----------
December 4, 2007
(EQ 19)
4.2 Critically damped systems: =1

When =1, we have two equal negative, real roots: n and n again. When you
have two equal roots, the solution takes a slightly different form (from 18.03):
x( t ) = A 1 e
(n )t
+ A 2 te
(n )t
(EQ 20)
Solving once again for A1 and A2 using the initial conditions the initial conditions
x(0) = u and x (0) = v , we get the final particular solution:
o
x( t ) = e
( n )t
[u o + n u o t + v o t] .
(EQ 21)
4.3 Overdamped systems: >1
When >1 we get two real, negative, unequal roots of the form:
n n 1 .
(EQ 22)
2
For convenience, we will us the short-form for n 1 . So the roots are

n . The general solution is of the form:
x( t ) = A 1 e
( n + )t
+ A2 e
( n )t
(EQ 23)
which we can massage into the form:

x( t ) = e
( n )t
[B 1 cosht + B 2 sinht] .
(EQ 24)
Solving for initial conditions x(0) = u o and x (0) = v o , we get the particular solution:
x( t ) = e
( n )t
(v o + n u o )
u o cosht + -------------------------------- sinh t .
w
December 4, 2007
(EQ 25)
5.0 The Forced Response of Linear Systems
The right-hand-side of an equation of the form

mx + cx + kx = 0 is obviously zero, and such
equations are called homogeneous equations.
The solution is called the free response. How
ever, when the right-hand-side is not zero, we
refer to the solution as a forced response, and
we refer to that RHS term as the forcing func
tion. For example, in the case of the system in
Figure 4, the equation of motion is:
k
m
p(t)
c
FIGURE 4. Forced mass-spring-damper
mx + cx + kx = p ( t ) .
(EQ 26)
The complete solution of this equation, as you learned, is the sum of the homogeneous
solution and the particular solution. The homogeneous solution is simply the same as what
we derived in the previous section (depending on the damping ratio, look at Sections 4.1,
4.2 or 4.3). We will concentrate on the particular solution to the forcing function.
The forcing function can be any function. It can be a constant force, a linearly increasing
force, a saw-tooth function, a sinusoid or a general squiggle over time. We will not con
sider all these functions. We will limit our analysis to harmonic (sinusoidal) forcing func
tions in this course. There are ways to compute th response to more general forcing
functions but that is what 2.004 is for!
5.1 Solving for Harmonic Forcing Functions

Assume that p ( t ) = p o cos t . The equation of motion is then:
mx + cx + kx = p o cos t .
(EQ 27)
One way to solve this equation is to replace the forcing function with the complex func
tion p o e
it
. This is complex, but the real-part of this is simply p o cos t . Let us also call
x the complex version of x. Then the equation becomes:
mx + cx + kx = p o cos t .
(EQ 28)
i(t )
Now, we will propose a solution x( t ) = Ue

. (Why the alpha? Because we will
need the extra constant. Intuitively, because we know the solution will lag the forcing
function.) Inserting this solution, Equation 28 will become:
2
U( m + ic + k)e
Cancelling out e
i(t)
i(t )
= po e
i(t)
(EQ 29)
from both sides and re-arranging, we get:
December 4, 2007
Ue
p o
po k
Uo
- = ------------------------------------------------------------------------------,
= ---------------------------------------=
2
2
2
m + ic + k
(1 r ) + i(2r)
(1 r ) + i(2r)
(EQ 30)
where U o = p o k is the static displacement, i.e., the displacement of the spring if the
forcing function had had the same amplitude but were static in other words of zero fre
quency.
The denominator of the RHS of this equation is a complex number which can be
expressed in complex polar form as:
2
(1 r ) + i(2r) =
2 2
(1 r ) + (2r) e
2r
------------------
i atan
(1 r 2 )
complex modulus
(EQ 31)
phase
From this, we can conclude that:

Uo
-,
U = ----------------------------------------------2 2
2
(1 r ) + (2r)
(EQ 32)
2r -
.
= atan ----------------2
(1 r )
(EQ 33)
and
By the way, we define the dynamic magnification factor D as follows:

U
1
D = ------ = ------------------------------------------------ .
Uo
2 2
2
(1 r ) + (2r)
(EQ 34)
D is also referred to as gain.
5.2 Insights from D and

The frequency response of a system is the behavior of the D and the when the excitation
frequency, usually expressed in ratio form as r, is varied from low to high. When the plots
use logarithmic r and logarithmic D, they are called Bode plots. They are shown in Figure
5. From the Bode plots of the spring-mass-damper system we can see that:
The system goes bananas near r =1. This is called resonance. The more the damping,
the less the resonance.
December 4, 2007
D=101
=0.05
=0.3
D=100
=0.8
=0.7
D=10-1
D=10-2
r=10-1
r=100
r=10+1
180o
90o
=0.7
=0.3
=0.05
r=10-1
r=100
r=10+1
FIGURE 5. Bode plots
As the damping increases, the maximum dynamic amplitude reduces and the peak
moves to the left. This jives with our sense that damping reduces the damped natural
2
frequency: d = n 1 .
When the damping factor exceeds 0.7, there is no resonant peak.
The response always lags the excitation.
December 4, 2007
10
At small values of the damping factor, the lag goes from 180o to 0o when the frequency
ratio goes from just below 1 to just above.
Important piece of trivia: we have plotted the log of the gain against r in the upper of the
two graphs. Traditionally, people use 20 times the log of the gain. This is referred to as
decibels (dB). In other words, decibels of gain = 20lo D . Why do they do this? Just
because it is a more precise unit and it gives you greater sensitivity in communicating with
each other. So when you hear someone talk about a 20 dB difference, it means that the
gain is 10 times higher.
5.3 Putting it all together

The analys above tells us that the particular solution is:
Uo
2r -
- cos t atan ----------------.
x p ( t ) = ----------------------------------------------2
2
2
2
(1
r
)
(1 r ) + (2r)
(EQ 35)
The homogenous solution, from Section 4.0, assuming an underdamped system is:
xh( t ) = e
(n )t
[B 1 cos d t + B 2 sin d t] .
(EQ 36)
For a critically damped or an overdamped system, it would be different.

The complete solution is x( t ) = x h ( t ) + x p ( t ) where we need to plug in the initial condi
tions to solve for B1 and B2.
The key insight is that initial conditions affect the homogenous solution. If there is any
damping, the homogenous solution will die away. It is therefore also called the transient
response. The particular solution stays as long as you have an excitation. It is therefore
called the steady-state response as well.
6.0 Conclusion
So there you have it. In the first two-thirds of the class, you derived the equations of
motion of rigid-body dynamic systems with springs and dampers thrown in. In the last
piece, you looked at how the equations behave. And guess what, you discovered stability,
damping and oscillation. Onwards to 2.004 now!
December 4, 2007
11
Appendix A
You might wonder about imaginary numbers, stability, oscillation, etc. Here I provide a
few simple comments which might help you piece together a number of concepts you
have learned over the years.
A.I.
History
The following history of infinite series is taken from Wikipedia, which is an excellent
source, believe it or not. I have listed the cultures which contributed to this topic. I pro
duce this because I want you to know that these concepts are not all that obvious, but if
you ponder them, you can make them somewhat instinctive. Many many mathematical
concepts have similar histories spanning centuries and continents.
The Pythagorean philosopher Zeno considered the problem of summing an infi
nite series to achieve a finite result, but rejected it as an impossibility: the result
was Zeno's paradox [Greece, 450 BC]. Later, Aristotle proposed a philosophical
resolution of the paradox, but the mathematical content was apparently unresolved
until taken up by Democritus and then Archimedes. It was through Archimedes's
method of exhaustion that an infinite number of progressive subdivisions could be
performed to achieve a finite trigonometric result. Liu Hui independently
employed a similar method several centuries later [China, 300 AD].
In the 14th century, the earliest examples of the use of Taylor series and closelyrelated methods were given by Madhava of Sangamagrama [India, 1400 AD].
Though no record of his work survives, writings of later Indian mathematicians
suggest that he found a number of special cases of the Taylor series, including
those for the trigonometric functions of sine, cosine, tangent, and arctangent. The
Kerala school of astronomy and mathematics further expanded his works with var
ious series expansions and rational approximations until the 16th century [India,
1500 AD].
In the 17th century, James Gregory [Scotland] also worked in this area and pub
lished several Maclaurin series [Scotland]. It was not until 1715 however that a
general method for constructing these series for all functions for which they exist
was finally provided by Brook Taylor [England], after whom the series are now
named.
A.II.
Infinite Series
You probably know this, but the English mathematician Brook Taylor came up with what
is now referred to as the Taylor Series: a representation of a function as an infinite sum of
terms calculated from the values of its derivatives at a single point. If a = 0 then the
series is also called the Maclaurin Series, a Scottish Mathematician.
f '(a)
f ''(a)
2 f '''(a)
3
f( x ) = f ( a ) + ------------ (x a) + ------------ (x a) + ---------- --- (x a) .
1!
2!
3!
December 4, 2007
(EQ 37)
12
The Taylor Series does not converge or converge to the right value for every function. It
does in the neighborhood of a for a large class of functions called analytic functions. It is
true everywhere for a smaller class of functions called entire functions.
The Taylor expansion can be used to approximate functions in a neighborhood of a, as
we did in linearization. Usually we use the first few terms and ignore the rest.
The Taylor expansion can also be used to understand imaginary numbers. To under
stand this, we first state that the exponential, sine and cosine functions are all entire
functions.
A.III. Sine, Cosine, Exponential Series and the Imaginary Link

The infinite series for sine and cosine are respectively:
3
x-- --x x
si ( x ) = x --+ -- ---- +
3! 5! 7!
2
(EQ 38)
x
x x
co ( x ) = 1 ---- + ----- ---- + .
2! 4! 6!
(EQ 39)
Just the first 4 terms in sine, as shown above, actually approximate it to within 3 parts in a
million! So you can get arbitrarily close if you want to.
The series for the exponential is:
2
x
x
x
x
e = 1 + x + ----- + ----- + ----- + .
2! 3! 4!
(EQ 40)
You can see an obvious similarity between Equations 38, 39 and 40. The first two seem to
want to merge to the third, but they dont actually add up, do they? In other words,
x
e co ( x ) + si ( x ) and e co ( x ) si ( x ) . Now lets invent a new and odd concept:
the imaginary number, i = 1 .1 It was the Swiss mathematician Euler who noticed that,
armed with the concept of imaginary numbers, we can generalize the exponential series to
include the sine and the cosine. He wrote down the Euler Formula in 1748:
x
e = co ( x ) + i si ( x ) .
(EQ 41)
Go ahead, try it outit works. The beauty of this is that now, we can express exponential
and oscillating functions under one complete umbrella.
1. The concept was first introduced by Cadano and Bombelli, both Italians, in the 1500s.
December 4, 2007
13

i ---
Imaginary
e
i ---
2
Imaginary
e 4
i ---
e 6
+i
i( o ))
t=0
time
Real
-1
0,0
t=1.6
t=2 +1/2
0,0
+1
Real
+1
-i
amplitude
amplitude
/2
2
3/2
a) Imaginary root
FIGURE 6. Exponentials
A.IV.
time
time
3
a) Real root
and how they behave in the complex plane
How Roots Capture Oscillations and Decay (Obvious Stuff)
When we solve the characteristic equation, we get roots which can be real or complex. If a
ibt
ibt
root is complex, say a + ib , the solution will have the form e . The function e can be
thought of as a rod of unit length which is pinned at, and rotating about, the origin of the
complex plane as shown in Figure 6(a) with an angular speed of b.This of course is what
you know and love as the polar form of a complex number. To make our lives easier for
the purposes of this explanation, assume that b = 1 . As t increases from 0, the rod rotates
counterclockwise as shown. The projection of the rod on the real-axis, which is the real
portion of the complex number, is co ( t ) . The projection on the imaginary axis is si ( t ) .
All obvious, but I just want to make sure everyone is on the same plane. It isnt complex
(well it is.) Remember that you have two roots, and there will be two such terms. If the ini
tial conditions are zero, then the nett answer will be purely realthe imaginary portions
will cancel out in the added-up final term.
If a root is real, say , the rod doesnt rotate. It stays on the real-line and simply increases
or decreases in length over time. There are two options. If is positive, then e
December 4, 2007
explodes
14
on you and you have instability. If the root is negative, then it exponential decay exponen
tially over time as shown in Figure 6(b) (where we assumed = 1 ).
December 4, 2007
15

The Behavior of Dynamic Systems: 1.0 Where Are We in The Course?

Uploaded by

Copyright:

Available Formats

The Behavior of Dynamic Systems: 1.0 Where Are We in The Course?

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Behavior of Dynamic Systems: 1.0 Where Are We in The Course?

Uploaded by

Copyright:

Available Formats

The Behavior of Dynamic Systems

Professor Sanjay Sarma

1.0 Where are we in the course?

Kinetics & Constitutive

2.0 Solving the Differential Equations: Overview

Downloaded on [DD Month YYYY].

1. Regardless of whether there is a forcing function, start with the characteristic

. The guess will always be of the same form with a phase

lag thrown in: x( t ) = Ue

3.1 Equilibrium points

We find the equilibrium points by setting and to

zero and solving for . (That is the definition of an

FIGURE 1. Inverted pendulum

1 = n . In other words, when the pendulum is