Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
ANIL V. RAO
University of Florida
Spring 2009
ii
Anil V. Rao earned his B.S. in mechanical engineering and A.B. in mathematics from
Cornell University, his M.S.E. in aerospace engineering from the University of Michigan, and his M.A. and Ph.D. in mechanical and aerospace engineering from Princeton
University. After earning his Ph.D., Dr. Rao joined the Flight Mechanics Department
at The Aerospace Corporation in Los Angeles, where he was involved in mission support for U.S. Air Force launch vehicle programs and trajectory optimization software
development. Subsequently, Dr. Rao joined The Charles Stark Draper Laboratory, Inc.,
in Cambridge, Massachusetts. As a Draper employee, Dr. Rao led numerous projects
related to trajectory optimization, guidance, and navigation of both space flight and
atmospheric flight vehicles. Concurrently, from 2001 to 2006 Dr. Rao was an Adjunct Professor of Aerospace and Mechanical Engineering at Boston University where
he taught the core undergraduate engineering dynamics course. While at Boston University, Dr. Rao was voted the 2002 and 2006 Mechanical and Aerospace Engineering
Faculty Member of the Year and was voted 2004 College of Engineering Professor of
the Year for outstanding teaching.
c
Anil
Vithala Rao 2006
vi
Contents
1
1
3
4
.
.
.
.
.
.
.
.
9
9
9
10
12
15
42
48
53
55
55
74
84
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
Input
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
93
93
93
98
102
103
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
107
107
108
110
112
113
115
117
121
Bibliography
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
121
viii
Contents
Chapter 1
Response of Single Degree-of-Freedom
Systems to Initial Conditions
In this chapter we begin the study of vibrations of mechanical systems. Generally speaking a
vibration is a periodic or oscillatory motion of an object or a set of objects. Vibrating systems
are ubiquitous in engineering and thus the study of vibrations is extremely important.
The most basic problem of interest is the study of the vibration of a one degree-of-freedom
(i.e., a system whose motion can be described using a single scalar second-order ordinary differential equation). The generic model for a one degree-of-freedom system is a mass connected
to a linear spring and a linear viscous damper (i.e., a mass-spring-damper system). Because of
its mathematical form, the mass-spring-damper system will be used as the baseline for analysis
of a one degree-of-freedom system. In particular, the differential equation of motion will be
derived for the mass-spring-damper system. It will then be shown that the time response of
this system is the sum of the zero input response and the zero initial condition response. In this
chapter we will focus attention on the zero input response, i.e., the response of the system to a
given set of initial conditions. Several examples of single degree-of-freedom systems will then
be given. In each of these examples the differential equation will be derived and will be shown
to have the same mathematical form as the generic mass-spring-damper system.
1.1
Mass-Spring-Damper System
The most basic system that is used as a model for vibrational analysis is a block of mass m
connected to a linear spring (with spring constant K and unstretched length 0 ) and a viscous
damper (with damping coefficient c). In addition, an external force P(t) is applied to the block
and the displacement of the block is measured from the inertially fixed point O, where O is the
point where the spring is unstretched. Finally, the spring and damper are both attached at the
inertially fixed point Q. This system is shown in Fig. 11 Denoting unit vector in the direction
from O to Q as Ex and the inertial reference frame of the ground by F , the inertial acceleration
of the block is given as
F
x
a = xE
(11)
Next, the forces exerted by the spring and damper are given, respectively, as
Fs
Ff
K( 0 )us
cvrel
(12)
(13)
(14)
g
K
m
c
O
x
0
Figure 11
Block of mass m sliding without friction along a horizontal surface connected to a linear spring and a linear viscous damper.
where r and rQ are the positions of the block and the attachment points of the spring, respectively. Using a coordinate system with its origin at point O at Ex as the first principal direction,
we have
r
rQ
xEx
(15)
0 Ex
(16)
Therefore,
= kxEx 0 Ex k = k(x 0 )Ex k = |x 0 |
(17)
|x 0 | = 0 x
(18)
Finally, the unit vector in the direction from the attachment point of the spring to the position
of the block is
r rQ
(x 0 )Ex
us =
=
= Ex
(19)
kr rQ k
0 x
The force in the linear spring is then given as
(110)
Next, because the ground is already assumed to be inertial, the relative velocity between the
block and the ground is simply the velocity of the block, i.e.,
x
vrel = Fv = xE
(111)
(112)
(113)
(114)
Dropping Ex from Eq. (114) and rearranging, we obtain the differential equation of motion as
+ cx
+ Kx = P
mx
(115)
Now historically it has been the case that the differential equation has been written in a form
that is normalized by the mass, i.e., we divide Eq. (115) by m to obtain
+
x
c
K
P
+
x
x=
= p(t)
m
m
m
(116)
where p(t) = P (t)/m. Furthermore, it is common practice to define the quantities K/m and
c/m as follows:
2n
2n
K
m
c
m
The quantities n and are called the natural frequency and damping ratio of the system,
respectively. In terms of the natural frequency and damping ratio, the differential equation of
motion for the mass-spring-damper system can be written in the so called standard form as
+ 2n x = p(t)
+ 2n x
x
(117)
It is seen that Eq. (117) is a second-order linear constant coefficient ordinary differential equation. Often, the term constant coefficient is replaced with the term time-invariant, i.e., we
say that Eq. (117) is a called a second-order linear time-invariant (LTI) ordinary differential
equation. The terminology time invariant stems from the fact that, for a given input p(t) and
0 ) = (x0 , x
0 ) at the initial time t = t0 is the same as
a given set of initial conditions (x(t0 ), x(t
0 + ) = (x0 , x
0 ) at
the solution to the input p(t + ) for the initial conditions (x(t0 + ), x(t
the (shifted) initial time t = t0 + . Because of this fact associated with an LTI system, without
loss of generality we can assume that the initial time is zero, i.e., t0 = 0. Thus, when studying
the zero input response of an LTI system we can restrict our attention to initial conditions
0 ).
(x(0), x(0)
= (x0 , x
1.2
(118)
!
d
d2
2
+
2
+
n
n x = p(t)
dt 2
dt
(119)
Now let
d2
d
+ 2n
+ 2n
dt 2
dt
Then we can view the system of Eq. (117) as a system of the form
L=
Lx = f
(120)
(121)
(122)
for all constants and . Then it is seen that Eq. (121) is a linear system whose general
solution is of then form Eq. (117) is given as
x(t) = xh (t) + xp (t)
(123)
here xh (t) is the homogeneous solution (i.e., the solution for a particular set of initial conditions
0 ) = (x0 , x
0 ) with a zero input function p(t) 0) while xp (t) is the particular
(x(t0 ), x(t
0 ) = (0, 0) and an arbitrary input
solution (i.e., the solution for zero initial conditions (x(t0 ), x(t
function p(t) 0). The homogeneous solution and particular solutions are also called the zero
input response and zero initial condition response, respectively. The general solution x(t) to
a second-order LTI system is then given as the sum of the zero input response and the zero
initial condition response. Because the zero input response satisfies Eq. (117) when p(t) 0,
we have
h + 2n xh = 0
h + 2n x
(124)
x
Contrariwise, because the zero initial condition response satisfies Eq. (117) when p(t) 0 and
the initial conditions are zero, we have
p + 2n xp = p(t)
p + 2n x
x
(125)
From the preceding discussion, it is seen that studying the general response of a secondorder LTI system amounts to studying independently the zero input response and the zero
initial condition response. Consequently, the study of single degree-of-freedom vibrations
amounts to quantifying the zero input response and the zero initial condition response. In
this remainder of this chapter we study in detail the zero input response of a second-order LTI
system that arises in the study of mechanical vibrations.
1.3
We now focus on the zero input response of the second-order LTI system of Eq. (117), i.e., we
focus on the system
h + 2n xh = 0
h + 2n x
(126)
x
Suppose that we guess the solution to Eq. (126) as
xh (t) = et
(127)
where is constant that has yet to be determined. Differentiating the assumed solution of
Eq. (127) twice, we have
h (t)
x
h (t)
x
et
(128)
2 et
(129)
Substituting the results of Eqs. (128) and (129) into (126), we obtain
2 et + 2n et + 2n et = 0
(130)
Then, because et is not zero as a function of time, it can be dropped from Eq. (130) to give
2 + 2n + 2n = 0
(131)
Equation (131) is called the characteristic equation whose roots give the behavior of the zero
input response of Eq. (117). Using the quadratic formula, the roots of Eq. (131) are given as
q
q
(132)
1,2 = n 4 2 2n 42n = n n 2 1
It can be seen that the types of roots admitted by Eq. (131) depend upon the value of . In
particular, the types of roots are governed by the quantity 2 1. We have three cases to
consider: (1) 0 < 1, (2) = 1, and (3) > 1. We now consider each of these cases in turn.
It is seen from Eq. (133) that the roots of the characteristic equation for an underdamped
system are complex. Furthermore, the general zero input response for an underdamped system
is given as
q
q
xh (t) = en t c1 cos n 1 2 t + c2 sin n 1 2 t
(134)
Eq. (134) can be written as
0 ) as
constants c1 and c2 can be solved for by using the initial conditions (x(0), x(0))
= (x0 , x
follows. First, substituting the initial condition x(0) = x0 into Eq. (135), we obtain c1 as
xh (0) = x0 = c1
(136)
(137)
0 , we obtain
Applying the initial condition x(0)
=x
h (0) = x
0 = n c1 + d c2
x
(138)
Substituting the result for c1 from Eq. (136) into Eq. (138), we obtain
0 = x0 n + d c2
x
Solving for c2 we have
c2 =
0 + n x0
x
d
(139)
(140)
(141)
A schematic of the underdamped zero input response for various values of 0 < is shown
in Fig. 12.
xh (t)
= 0.05
= 0.1
= 0.2
= 0.5
Figure 12
Schematic of the zero input response of an underdamped second-order
linear time-invariant system.
Case 2: = 1 (Critical Damping)
When = 1 the zero input response is said to be
q critically damped. For critically damped
system the quantity 2 1 = 0 which implies that 2 1 = 0. The roots of the characteristic
equation for an underdamped system are then given as
1,2 = n = n
(142)
It is seen from Eq. (142) that the roots of the characteristic equation for a critically damped
system are real and repeated (i.e., the two roots are the same). Furthermore, the general zero
input response for a critically damped system is given as
xh (t) = en t (c1 + c2 t)
(143)
0 )
The constants c1 and c2 can be solved for by using the initial conditions (x(0), x(0))
= (x0 , x
as follows. First, applying the initial condition x(0) = x0 into Eq. (143), we have
xh (0) = x0 = c1
(144)
(145)
0 , we obtain
Applying the initial condition x(0)
=x
h (0) = x
0 = n c1 + c2
x
(146)
Action
1.3 General Solution to Second-Order Homogeneous LTI System
(147)
0 + n x0
c2 = x
(148)
0 + n x0 )t]
xh (t) = en t [x0 + (x
(149)
xh (t)
0
0
Figure 13
Schematic of the zero input response of a critically damped second-order
linear time-invariant system.
Case 3: > 1 (Overdamping)
When > 1 the zero input response isqsaid to be overdamped. For an overdamped system the
quantity 2 1 > 0 which implies that 2 1 > 0. The roots of the characteristic equation for
an underdamped system are then given as
q
(150)
1,2 = n n 2 1
It is seen from Eq. (150) that the roots of an overdamped system are real and distinct. Furthermore, the general zero input response for an overdamped system is given as
xh (t) = c1 e1 t + c2 e2 t
(151)
0 )
The constants c1 and c2 can be solved for by using the initial conditions (x(0), x(0))
= (x0 , x
as follows. First, applying the initial condition x(0) = x0 , we obtain
xh (0) = x0 = c1 + c2
(152)
h (t) = c1 1 e1 t + c2 2 e2 t
x
(153)
0 , we obtain
Then, applying the initial condition x(0)
=x
h (0) = x
0 = c1 1 + c2 2
x
(154)
Equations (152) and (154) can then be solved simultaneously for c1 and c2 to give
c1
c2
0
x0 2 x
1 + 2
0
x0 1 + x
1 + 2
(155)
(156)
The general zero input response for an overdamped system is then given as
xh (t) =
0 1 t x0 1 + x
0 2 t
x0 2 x
e
+
e
1 + 2
1 + 2
(157)
A schematic of an overdamped zero input response for various values of > 1 is shown in
Fig. 14.
xh (t)
= 1.1
= 1.2
= 1.5
=2
Figure 14
Schematic of the zero input response of an overdamped second-order
linear time-invariant system.
Chapter 2
Forced Response of Single
Degree-of-Freedom Systems
2.1
2.2
Recall from dynamics that the principle of impulse and momentum for a particle states that
N
F = G N G
(21)
where N G is the linear momentum of the particle as viewed by an observer in an inertial reference frame N . Suppose now that we consider the following system. A block of mass m is
connected to a linear spring with spring constant K and unstretched length 0 and a viscous
linear damper with damping coefficient c as shown in Fig. 21. The block is initially at rest
(i.e., its initial velocity is zero) at its static equilibrium position (i.e., the spring is initially un is applied. We are interested here in determining the
stressed) when a horizontal impulse P
.
velocity of the block immediately after the application of the impulse P
g
K
m
c
x
0
Figure 21
Block of mass m connected to linear spring and linear damper struck by
.
horizontal impulse P
The solution of the above problem is found as follows. First, let F be the ground. Then,
10
Ex
Ez
Ey
Origin at block
when x = 0
=
=
=
To the left
Into page
Ez Ex
(22)
Because {Ex , Ey , Ez } is a fixed basis, the velocity of the block in reference frame F is given as
F
v=
dr
x = vEx
= xE
dt
(23)
Now because we are going to apply the principle of linear impulse and momentum to this
problem, we do not need the acceleration of the block. Instead, we know that neither the spring
nor the damper can apply an instantaneous impulse. Therefore, the only impulse applied to
. Consequently, the external impulse acting on the system
the system at t = 0 is that due to P
at t = 0 is
= PEx
F=P
(24)
Furthermore, the linear momentum of the block the instant before the impulse is applied is
zero (i.e., the block is initially at rest) while the linear momentum of the block the instant after
the impulse is applied is given as
F
G = m v = mv Ex
(25)
P = mv mv(t = 0+ )
(26)
Setting
F equal to G , we obtain
P
(27)
m
The result of this analysis shows that the response of a resting second-order linear system to
an impulsive force
F is equivalent to giving the system the initial velocity shown in Eq. (27).
v(t = 0+ ) =
2.3
Suppose now that we consider the general motion of the system in Fig. 21, i.e., we consider
motion to a general force F(t). Then, recalling the result from earlier, the differential equation
of motion is given as
+ cx
+ Kx = F (t) + K0
mx
(28)
It is noted that the equilibrium point of the system in Eq. (28) is xeq = 0 , we can define the
variable y = x xeq and rewrite Eq. (28) in terms of y to give
+ cy
+ Ky = F (t)
my
(29)
(210)
,
,
t=a
t
(211)
11
The function (t) is called the Dirac delta function or the unit impulse function. It is known
that the Dirac delta function satisfies the following properties:
Z
(t a)dt = 1
(212)
(213)
where f (t) is an arbitrary function. For simplicity, consider the case where F = 1, i.e., the case
of unit impulse being applied to the system. Also, let g(t) be the response to the input (t),
i.e., consider the system
+ cg
+ Kg = (t)
mg
(214)
Let T be a value of t such that T > 0. Then, integrating Eq. (214) from zero to T , we have
ZT
ZT
+ cg
+ Kg dt =
mg
(t)dt
(215)
0
ZT
mgdt
ZT
c gdt
mg(t)|
0
(216)
mg(t)|T0
(217)
+)
lim mg(t)
= mg(0
T 0+
T 0+
lim mg(t) |T0 = lim m g(T ) g(0) = 0
T 0+
T 0+
(218)
(219)
Furthermore, because the position of the mass cannot change during the application of an
instantaneous impulse, we see that
ZT
lim
Kg(t)dt = lim Kg(0)t |T0 = lim Kg(0) = 0
(220)
T 0+
T 0+
T 0+
Using the results of Eqs. (218), (219) and (220) in Eq. (215), we obtain
+) = 1
mg(0
(221)
+ ), we obtain
Solving Eq. (221) for g(0
+) =
g(0
1
m
(222)
It is seen that, for the case where P 1, the results of Eq. (27) and Eq. (222) are identical. More
specifically, as we saw above, the effect of a unit impulsive force on a resting particle of mass
m is to provide an initial velocity of magnitude 1/m while the response of a second-order linear
system to a unit impulse function (i.e., the Dirac delta function) is to provide an initial velocity
of magnitude 1/m. Consequently, the physics of an impulsive force on a resting particle is
identical to the mathematics of the impulse response of the system to a unit impulse.
Now that we know that the response of a second-order resting system is to change the
velocity (while leaving position unchanged), we can use this fact to obtain the impulse response
g(t). In particular, assuming an underdamped system, we know that the general form of the
free response is given as
g(t) = en t (A cos d t + B sin d t)
(223)
12
q
p
where n = k/m is the natural frequency, is the damping ratio, and d = n 1 2 is
the damped natural frequency. Differentiating this last equation, we have
g(t)
= n en t (A cos d t + B sin d t) + en t (Ad sin d t + Bd cos d t) (224)
+ ) = 1/m, we have
Noting that g(0) = 0 and that g(0
A
B
=
=
(225)
1
md
(226)
(227)
g(t)
It is seen that, for an underdamped system, the impulse response is a decaying sinusoid with
a zero phase (i.e., the applied impulse did not result in a nonzero phase shift). A schematic of
the impulse response is shown in Fig. 22.
0
t
Figure 22
System.
2.4
After the unit impulse function, the next fundamental function of importance in the analysis
of vibratory systems is the unit step function. The unit step function, denoted u(t), is defined
as
u(t a) =
0
1
,
,
13
ta
t>a
(228)
Recalling the unit impulse function (t) from Eq. (211), it is seen that u(t) is related to (t)
as follows:
Zt
u(t a) =
( a)d
(229)
where is a dummy variable of integration. Now suppose we want to determine the response,
s(t), of the system of Eq. (29) to a unit step input at t = 0. The function s(t) is called the step
response and, from Eq. (29), satisfies
m
s + c
s + Ks = u(t)
(230)
d2 s
ds
+c
+ Ks = u(t)
2
dt
dt
(231)
We can obtain s(t) as follows. Consider again the relationship that holds between the unit
impulse and the impulse response, i.e.,
+ cg
+ Kg = (t)
mg
(232)
du
dt
d
Now from the fundamental theorem of calculus we have
Zt
Z
d2 t
d2 g
=
g()d
2
dt 2
d
Z
Zt
d t
dg
=
gd
dt
d
Therefore, Eq. (235) can be rewritten as
"
#Z
Zt
t
d2
d
+
c
+
K
g()d
=
u(t)
=
()d
d 2
d
(233)
(234)
(235)
(236)
(237)
(238)
Zt
g()d
(239)
In other words, the response of the system of Eq. (29) to a unit step function is the integral of
the response of the system to a unit impulse1 . We can then use the result of Eq. (239) and the
1 More generally, it is the case that the response of any linear time-invariant system to the integral of a
function f (t) is equal to the integral of the response to the original function f (t).
14
impulse response given in Eq. (227) as follows. First, for t 0 we have s(t) = 0. For t > 0, we
have
Zt
Zt
1
1
s(t) =
en sin d d =
en sin d d
(240)
md 0
0 md
Now from DeMoivres theorem we have
sin d =
eid eid
2i
(241)
Therefore,
1
s(t) =
2imd
1
=
2imd
Zt
0
h
i
en eid eid
Zt h
i
e(n id ) e(n +id )
0
#t
e(n +id )
e(n id )
+
n id
n + id 0
"
#t
id
n
e
eid
e
=
2imd n id
n + id 0
#t
"
en (n + id )eid (n id )eid
=
2imd
2 2n + 2d
0
1
=
2imd
"
(242)
q
Now, noting that d = n 1 2 , we have
h
it
1
(n id )
(n +id )
(
+
i
)e
i
)e
n
d
n
d
0
2imd 2n
h
i
1
(n id )t
(n id ) 1 e(n +id )t
=
2 (n + id ) 1 e
2imd n
h
n
oi
1
n t
=
n eid t eid t + id eid t + eid t
2 2id e
2imd n
"
!#
1
eid t eid t
eid t + eid t
n t
=
d e
n
+ d
2i
2
md 2n
(243)
s(t) =
Now we have
eid t + eid t
= cos d t
2
Using Eq. (244) together with Eq. (241), we have
h
i
1
n t
(n sin d t + d cos d t)
2 d e
md n
1
n
n t
=
1
e
cos
t
+
sin
t
d
d
d
m2n
s(t) =
(244)
(245)
It is noted that the expression in Eq. (245) is valid when t > 0. Therefore, the response of the
system of Eq. (29) to a unit step function is given as
(
0
h
i , t 0
(246)
s(t) =
n
1
n t
1
e
cos
t
+
sin
t
, t>0
d
d
2
m
n
2.5
15
Recall the standard form of the differential equation that describes the motion of a damped
single degree-of-freedom system subject from Eq. (117) as
+ 2n x = p(t)
+ 2n x
x
(247)
(248)
+ 2n x = 2n f (t)
+ 2n x
x
(249)
Suppose further that f (t) is a periodic function of the form f (t) = Aeit . We then have
+ 2n x = 2n Aeit
+ 2n x
x
(250)
The function f (t) = Aeit will be called the normalized input function.
+ 2n x = 0
+ 2n x
x
(252)
and xp (t) is the particular solution that satisfies Eq. (250). In this analysis we are interested
in determining the particular solution of Eq. (250).
where we note that the coefficient X is a function of the input frequency . Differentiating
xp (t) in Eq. (253), we obtain
p (t)
x
p (t)
x
=
=
iXeit
2
Xe
it
(254)
(255)
p (t), and x
p (t) from Eqs. (253)(255), respectively, into Eq. (250), we
Substituting xp (t), x
have
2 Xeit + i2n Xeit + 2n Xeit = 2n Aeit
(256)
Rearranging Eq. (256) gives
h
i
Xeit (2n 2 ) + i2n = 2n Aeit
(257)
Observing that eit is not zero as a function of time, it can be dropped from Eq. (257) to give
h
i
(2n 2 ) + i2n = 2n A
(258)
16
(259)
2n
X()
= 2
2
A
n + i2n
(260)
Finally, we can divide the numerator and denominator of Eq. (260) by 2n to obtain
G(i) =
X()
=
A
1
2
+ i2
(261)
The quantity G(i) is called the transfer function of the system to the input Aeit . It is seen
that the transfer function is the ratio of the amplitude of the output to the amplitude of the
input. It is seen that the transfer function of the system of Eq. (250) is a function of the
frequency, , of the input F (t) = 2n Aeit
Now since the transfer function G(i) is complex, it can be written as
G(i) = + i
(262)
where
Re [G(i)]
(263)
Im [G(i)]
(264)
where Re [] and Im [] are the real and imaginary parts of G. From complex analysis, we know
that any complex number can be written as
z = + i = |z|ei
(265)
where
|z|
q
= 2 + 2
zz
tan1
(266)
(267)
(268)
where
|G(i)|
()
G(i)G(i)
Im [G(i)]
Re [G(i)]
(269)
(270)
(271)
17
(273)
(274)
Now, by the principle of superposition we know that the particular solution of Eq. (247) to
the sum of two inputs p1 (t) + p2 (t) is the sum of the responses, i.e., if x1 (t) is the particular
solution to the input p1 (t) and x2 (t) is the particular solution to the input p2 (t), then x1 (t) +
x2 (t) is the particular solution to the input p1 (t) + p2 (t). Now suppose we rewrite the general
complex input A2n eit as
A2n eit = A2n cos t + iA2n sin t = fr (t) + ifi (t)
(275)
where
fr (t)
fi (t)
A2n cos t
(276)
A2n
(277)
sin t
Now observe that fr (t) and fi (t) are the real and imaginary parts of A2n eit , respectively.
Furthermore, observe from Eq. (273) that A|G(i)| cos(t ) and A|G(i)| sin(t )
are the real and complex parts, respectively, of the response xp (t) to A2n eit . Then, by the
principle of superposition we know that the response of Eq. (247) to fr (t) must be the real
part of xp (t) in Eq. (273), i.e.,
n
o
xr (t) = Re A|G(i)|ei(t) = A|G(i)| cos(t )
(278)
Similarly, the response of Eq. (247) to fi (t) is the imaginary part of xp (t) in Eq. (273), i.e.,
n
o
xi (t) = Im A|G(i)|ei(t) = A|G(i)| sin(t )
(279)
18
solution) and A is the input amplitude of the normalized input function f (t) = Aeit . The
frequency response to a periodic input is defined as the combination of the magnitude and
phase of the ratio of the output to the input. Recall the magnitude and phase of G(i) from
Eqs. (269) and (270). Furthermore, recall from Eq. (261) that
G(i) =
X()
=
A
1
2
+ i2
n
1/2
1
|G(i)| = G(i)G(i)
=
2
1
+ i2
n
n
where
G(i)
=
1
2
(280)
1/2
2
1
i2
n
n
(281)
i2
n
(282)
2 #2
1/2
2
+ 2
n
n
(283)
Next, the phase of G(i) can be obtained as follows. First, we know that
G(i) = G(i)
|G(i)|2
G(i)
=
G(i)
G(i)
2
i2
1
n
n
G(i) = "
#2
2
2
1
+ 2
n
n
Extracting the real and imaginary parts of G(i) from Eq. (285), we have
2
1
n
Re [G(i)] = "
2 #2
2
1
+ 2
n
n
2
n
Im [G(i)] = "
2 #2
2
1
+ 2
n
n
(284)
(285)
(286)
(287)
() = tan1
Im [G(i)]
= tan1
Re [G(i)]
n
2
1
n
(288)
19
=0
= 0.1
= 0.25
= 0.5
=1
|G(i)|
0
0
0.5
1.5
/n
2.5
Figure 23
Magnitude of Frequency Response of a Single Degree-of-Freedom Linear
System to an Input f (t) = Aeit .
The magnitude and phase of G(i) are shown in Figs. 23 and 24, respectively, for various
values of the damping ratio . It is seen from Fig. 23 that the amplitude of the response
approaches as 0, i.e.,
lim |G(i)| =
(289)
0
1
q
2 1 2
(290)
Furthermore, it is seen that as approaches zero, the value at which |G(i)| is maximum
approaches unity, i.e.,
lim arg max |G(i)| = 1
(291)
0
Turning attention to the phase of G(i) (i.e., ()), it is seen that all of the curves pass
through the point /n = 1 and = /2. Furthermore, it is seen that approaches zero and
as /n approaches zero and /n approaches , respectively, i.e.,
lim
()
(292)
lim
()
(293)
/n
/n
It is noted that, for the special case of = 0, the phase has a discontinuity at /n = 1 (this
is not shown in Fig. 24). Finally, for the case where = 0 and /n = 1 the system is at
resonance with a phase angle of /2.
20
7 /8
3 /4
|G(i)|
5 /8
/2
3 /8
/4
=0
= 0.1
= 0.25
= 0.5
=1
/8
0
0
0.5
1.5
/n
2.5
Figure 24
Phase of Frequency Response of a Single Degree-of-Freedom Linear System to an Input f (t) = Aeit .
21
(294)
(295)
where Xr and are the amplitude and phase, respectively, of xr (t). Comparing Eq. (294) and
(295) it is seen that
Xr
A|G(i)|
(296)
(297)
Therefore, the magnitude and phase of xr (t) is the same as the magnitude and phase xp (t)
where xp (t) is given from Eq. (271). Now because a complex number is defined completely
from its magnitude and phase, we have
Gr (i) = G(i)
(298)
A2n
xi (t) = Xi sin(t )
(2100)
where Xi and are the amplitude and phase, respectively, of xi (t). Comparing Eq. (299) and
(2100) it is seen that
Xi
A|G(i)|
(2101)
(2102)
Therefore, the magnitude and phase of xi (t) is the same as the magnitude and phase xp (t)
where xp (t) is given from Eq. (271). Again, because a complex number is defined completely
from its magnitude and phase, we have
Gi (i) = G(i)
(2103)
In other words, the transfer function of that the A2n sin t is identical to the transfer function
of Eq. (2103) to the input A2n eit .
22
Example 21
A collar of mass m slides without friction along a circular arc portion of a rigid massless
structure as shown in Fig. 25. The structure consists of two arms, one oriented horizontally
and the other oriented at a constant angle from the downward direction. The entire structure,
centered at point Q, translates with known horizontal displacement q(t) along a rod, where q
is measured from a track-fixed point O. A collar of mass m slides along the arc of the structure
circular part of the structure. The position of the collar relative to the structure is measured
by the angle , where is measured from the downward direction. Attached to the collar is
a curvilinear spring with spring constant K and unstretched length 0 = R. Also, a viscous
friction force with viscous friction coefficient c is exerted by the circular arc on the collar. The
spring and friction forces are given, respectively, as
Fs
Ff
=
=
K( 0 )et
cvrel
where et is the tangent vector to the track at the location of the collar and vrel is the velocity of
the collar relative to the track. Assuming no gravity, determine (a) the differential equation of
motion; (b) the static equilibrium value eq for the system; (c) the differential equation of motion
relative to the static equilibrium point found in (b); (d) the standard form of the differential
equation obtained in part (c); (e) the transfer function for /Q where is the amplitude of the
output(i.e., the amplitude of ) q(t) = (QK/2 )eit ; (f) the time response of the system to the
sinusoidal input q given in part (e).
q(t)
Viscous Friction, c
B
Figure 25
Collar of mass m moving on circular part of a structure, where the
structure slides with horizontal displacement q(t).
Solution to Example 21
(a) Differential Equation of Motion
Kinematics
Let F be fixed to the track. Then choose the following coordinate system fixed in F :
Ex
Ez
Ey
Origin at Q whenq = 0
=
=
=
to the right
out of page
Ez Ex
23
Next, let A be fixed to the structure. Then choose the following coordinate system fixed in A:
ex
ez
ey
Origin at Q
=
=
=
to the right
out of page
ez ex
Finally, let B be fixed to the direction Qm. Then choose the following coordinate system fixed
in B:
Origin at Q
er
=
along Qm
ez
=
out of page
e
=
ez er
rQ = qEx = qex
(2104)
(2105)
(2106)
z
B = e
(2107)
vQ
aQ
ex
q
(2108)
ex
q
(2109)
The velocity of the collar relative to point Q in reference frame F is obtained from the transport
theorem as
F
Bd
d
F
vm/Q =
rm/Q =
rm/Q + FB rm/Q
(2110)
dt
dt
where
B
d
rm/Q
dt
F B
rm/Q
(2111)
z Rer = R e
e
(2112)
Consequently,
F
vm/Q = R e
(2113)
The acceleration of the collar relative to point Q in reference frame F is obtained from the
transport theorem as
F
am/Q =
Bd
d F
F
am/Q =
vm/Q + FB Fvm/Q
dt
dt
(2114)
where
B
d F
vm/Q
dt
F B
Fvm/Q
R e
(2115)
z R e
= R
2 er
e
(2116)
24
Consequently,
F
2 er + R e
am/Q = R
(2117)
2 er + R e
ex R
a = FaQ + Fam/Q = q
(2118)
Kinetics
The free body diagram of the collar is shown in Fig. 26.
N
Fs
Ff
Figure 26
=
=
=
Ner
cvrel
Ff
K( 0 )et
(2119)
(2120)
(2121)
Now
et
0
vrel
(2122)
R( + )
(2123)
R
F
v FvQ = Fvm/Q = R e
(2124)
(2125)
(2126)
(2127)
(2128)
(2129)
(2130)
25
Therefore,
= m
2 er + mR e
Ner (KR + cR )e
qex mR
2 er + mR e
= m
q(sin er + cos e ) mR
2 er + mR e
= m
q sin er + m
q cos e mR
(2131)
2 )er + (m
= (m
q sin mR
q cos + +mR )e
(KR + cR )
2
m
q sin mR
m
q cos + +mR
(2132)
(2133)
It is seen that the second of these equation is the differential equation of motion. Rearranging,
we obtain
+ cR
+ KR = m
mR
q cos
(2134)
(2135)
eq = 0
(2136)
(2137)
for values of near zero. Therefore, the linearized differential equation for values of near
eq is
+ cR
+ KR = m
mR
q
(2138)
+ K = q
+ c
m
m
R
(2139)
iQKeit /
(2140)
QKeit
(2141)
n
m
m
R
R
R
(2142)
26
Qm
R
(2143)
Furthermore, let
(t) = eit
which implies
ieit
2
it
(2144)
(2145)
(2146)
Therefore,
Therefore,
h
i
eit 2 + i2n + 2n = A2n eit
2n
= 2
A
n 2 + i2n
A
2n
2n
1
m
m
Q
= 2
=
=
2
Q
R n 2 + i2n
R 1 2 + i2
n 2 + i2n
n
n
Consequently,
m
=
G(i)
Q
R
where
G(i) =
1
2
+ i2 n
(2147)
(2148)
(2149)
(2150)
(2151)
(2152)
x(t) = A|G(i)|ei(t)
(2153)
is given as
where |G(i)| and are the magnitude and phase of G(i). Now our input amplitude is
A=
Qm
R
(2154)
Qm
|G(i)|ei(t)
R
(2155)
27
Example 22
A collar of mass m slides along an inertially fixed circular track of radius R as shown in Fig. 2
7. Attached to the collar is a curvilinear spring with spring constant K and unstretched length
0 = R0 . The position of the collar on the track is measured by the angle , where is
measured from the inertially fixed downward direction. Furthermore, the contact between the
track and the collar creates a viscous friction force with friction coefficient c. The forces exerted
by the curvilinear spring and the viscous damper are given, respectively, as
Fs
Ff
=
=
K( 0 )et
cvrel
where et is the tangent vector to the track at the location of the collar and vrel is the velocity
of the collar relative to the track. Finally, attached to the other end of the spring is a massless
collar that moves with specified displacement described the angle (t), where, like , is also
measured from inertially fixed downward direction. Assuming no gravity, determine (a) the
differential equation of motion; (b) the value eq for which the system is in static equilibrium;
(c) the differential equation of motion relative to the static equilibrium found in part (b); (d)
the standard form of the differential equation obtained in part (d); (e) the natural frequency,
damping ratio, and damped natural frequency of the system (assuming that the system is
Viscous Friction, c
O
R
K
Q
Figure 27
Collar sliding on fixed circular track attached to a linear spring with
moving attachment point and viscous friction.
28
Solution to Example 22
(a) Differential Equation of Motion
Kinematics
Let F be fixed to the circular track. Then choose the following coordinate system fixed in F :
Origin at O
=
=
=
Ex
Ez
Ey
along Om when = 0
out of page
Ez Ex
Next, let A be fixed to the direction Om. Then choose the following coordinate system fixed in
A:
Origin at O
er
=
along Om
ez
=
out of page
ez
=
ez er
r = Rer
(2156)
v=
dr
dr F A
=
+ r
dt
dt
(2157)
z . Now we have
where FA = e
A
dr
dt
F A
r
(2158)
z Rer = R e
e
(2159)
v = R e
(2160)
The acceleration of the collar as viewed by an observer fixed to the track is then given as
F
a=
A
d F
d F F A F
v =
v + v
dt
dt
(2161)
Now we have
A
d F
v
dt
F A
Fv
R e
(2162)
z R e
= R
2 er
e
(2163)
2 er + R e
a = R
Kinetics
From the free body diagram, the following forces act on the collar:
N
Fs
Ff
=
=
=
(2164)
29
Now we have
N
Fs
Ff
Ner
cvrel
K( 0 )et
(2165)
(2166)
(2167)
Using the fact that et = e and that the surface is absolutely fixed, we obtain vrel = Fv. Consequently,
N
Fs
Ff
Ner
(2168)
K( 0 )e
cR e
(2169)
(2170)
(2171)
Fs = K(R( ) R0 )e = KR( 0 )e
(2172)
(2173)
(2174)
mR
(2175)
KR( 0 ) cR
(2176)
It is seen that the second of these last two equations has no unknown reaction forces and, thus,
is the differential equation. Rearranging this equation, we obtain
+ cR
+ KR = KR0 + KR
mR
(2177)
(2178)
which implies
eq = 0
(2180)
(2181)
(2182)
30
K
K
c
+
=
m
m
m
(2183)
2n =
(2184)
c
m
(2185)
c
c
=
2mn
2 mK
(2186)
(2187)
(2188)
Then
(t)
(t)
ibar eit
2
bar e
it
(2189)
(2190)
(2191)
A = PK
(2192)
i
h
it 2n 2 + i2n = A2n eit
e
(2193)
Now let
Then,
Rearranging gives
Therefore,
2n
1
= 2
= G(i)
=
2
2
A
n + i2n
1 n + i2 n
=
= KG(i)
P
AP
(2194)
(2195)
31
(2196)
x(t) = A|G(i)|ei(t)
(2197)
A = PK
(2198)
(t) = P K|G(i)|ei(t)
(2199)
(2200)
is given as
Now in our case we have
which implies that
Therefore, the response of the system to the input AK sin t is the imaginary part of (t), i.e.,
Example 23
A massless cart moves horizontally along the ground with a known displacement q(t), where q
is measured from a point O fixed to the ground as shown in Fig. 28. A block of mass m slides
along the surface of the cart. Attached to the block are a linear spring with spring constant
K and unstretched length 0 and a viscous damper with damping coefficient c. The spring
and damper are attached at point Q, where Q is located on the vertical support of the cart.
Knowing that x describes the displacement of the block relative to the cart and that gravity
acts downward, determine (a)the differential equation of motion for the system; (b) (b) the
static equilibrium value xeq for the differential equation given in part (a); (c) the differential
equation of motion relative to the static equilibrium found in part (b); (d) the standard form
of the differential equation obtained in part (c); (e) the natural frequency, damping ratio, and
damped natural frequency of the system in terms of the parameters K and c (assuming that
the system is underdamped); (f) the transfer function associated with the ratio of the amplitude
Y /Q where Y is the amplitude of the output y(t) and q(t) = QKeiat ; (g) the time response,
denoted z(t), of the system to the periodic input q(t) = QK(cos at).
x
q(t)
O
Figure 28
damper.
c
Q
Block sliding on horizontally moving cart with linear spring and viscous
32
Solution to Example 23
(a) Differential Equation of Motion
Kinematics
Let F be fixed to the ground. Then choose the following coordinate system fixed in reference
frame F :
Origin at O
Ex
=
to the right
Ez
=
into page
Ey
=
Ez Ex
Next, A be fixed to the block. Then choose the following coordinate system fixed in reference
frame A:
Origin at Q
ex
=
along Qm
ez
=
Ez
ey
=
ez ex
Now, because the block is in pure translation, the position of the support Q is given as
rQ = qEx
(2201)
Next, the position of the block relative to the upper support is given as
rP /Q = xex
(2202)
(2203)
where we note that Ex = ex . Then the velocity and acceleration of the block in reference frame
F are given as block are given, respectively, as
F
x
(
q + x)e
x
(
q + x)e
(2204)
(2205)
Kinetics
The free body diagram of the block is shown in Fig. 29.
N
Fs
Fd
mg
Figure 29
Free body diagram for block sliding on horizontally moving cart with
linear spring and viscous damper.
where
Fs
Fd
mg
N
=
=
=
=
33
Now we have
Fs
Fd
mg
N
K( 0 )us
(2206)
mgey
(2208)
cvrel
(2207)
Ney
(2209)
us
(2210)
(2211)
(2212)
(2213)
Therefore,
Fs = K(x 0 )ex
(2214)
x q
ex = xe
x
vrel = Fv FvQ = (
q + x)e
(2215)
Ex = q
ex . Therefore, the force of the damper is given as
where it is noted that vQ = q
x
Fd = cvrel = c xe
(2216)
(2217)
(2218)
(2219)
K(x 0 ) + c x
N mg
m(
q + x)
(2220)
(2221)
It is seen that Eq. (2220) has no unknown reaction forces and, thus, is the differential equation
of motion. Rearranging Eq. (2220), we obtain
+ cx
+ Kx = K0 m
mx
q
(2222)
(2223)
(2224)
Furthermore, in order to find the static equilibrium point, we need to set q(t) = 0. Substituting
the equilibrium conditions into Eq. (2222) gives
Kxeq = K0
(2225)
xeq = 0
(2226)
34
y = x xeq = x = y + xeq = y + 0
(2227)
(2228)
(2229)
(2230)
(2231)
K
c
+
y
y =
q
m
m
(2232)
K
m
(2233)
2n =
c
m
(2234)
K
c
=
m
m
(2235)
n =
The damping ratio is found by solving
for . We have
2n = 2
(2236)
(2237)
iaQKeiat
(2238)
a2 2 QKeiat
(2239)
(t) into Eq. (2232) and using the generic expressions for n and , we obtain
Substituting q
+ 2n y = a2 2 QKeiat
+ 2n y
y
(2240)
35
Then
(2241)
iaY eit
a Y e
(2242)
it
(2243)
and y
into Eq. (2240) gives
Substituting y, y,
i
h
a2 2 + i2an + 2n Y eiat = a2 2 QKeiat
(2244)
(2245)
Now let
A = Qma2 2
(2246)
Then, dropping the common factor of eit and dividing through by a2 2 + i2an + 2n
gives
A2n
Y = 2
(2247)
2
n a 2 + i2an
Then, dividing numerator and denominator by 2n gives
Y =
a
n
A
2
(2248)
a
+ i2
n
Now let
A/Q
Y
=
a
a 2
Q
1 n + i2
n
= a
(2249)
1
2
(2250)
(2251)
+ i2 n
A
Y
= G(i) = m2 G(i)
Q
Q
(2252)
(2253)
x(t) = A|G(i)|ei(t)
(2254)
A = Qma2 2 = Qm2
(2255)
36
where we recall that = a. Therefore, the time response to the input QKeiat is
y(t) = Qm2 |G(i)|ei(t)
(2256)
Then the response to the input QK cos at is the real part of y(t), i.e.,
h
i
z(t) = Re Qm2 |G(i)|ei(t)
= Qm2 |G(i)| cos(t )
2
(2257)
Example 24
A collar of mass m1 slides along an inertially fixed track. The displacement of the collar is
measured relative to the fixed point O by the variable x. The collar is attached to a linear
spring with spring constant K, a linear damper with damping coefficient c, and a rigid massless
arm of length L. Attached to the other end of the arm is a particle of mass m2 . Knowing that
the arm rotates with a constant angular rate , (a) derive the differential equation of motion for
the system in terms of the displacement x; (b) determine the equilibrium point of the system;
(c) write the differential equation in part (a) relative to the equilibrium point found in part (b);
and (d) determine the time response of the collar.
K
m1
O
P
m2
x
t
Figure 210
Solution to Example 24
(a) Differential Equation of Motion
Kinematics
First, let F be the track. Then, choose the following coordinate system fixed in reference frame
F:
Origin at point O
Ex
=
Along OP
Ez
=
Out of page
Ey
=
Ez Ex
37
Next, let R be a reference frame fixed to the arm. Then, choose the following coordinate system
fixed in reference frame R:
Origin at O
=
=
=
er
ez
e
Along m1 m2
Out of page
ez er
Then the angular velocity of arm as viewed by an observer fixed to the track is given as
F
R = ez
(2258)
r1 = xEx
(2259)
The velocity and acceleration of the collar in reference frame F are given, respectively, as
F
v1
a1
x
xE
(2260)
x
xE
(2261)
Now in order to solve this problem, we also need the acceleration of the particle attached to
the arm. The position of the particle is given as
r2 = r1 + r2/1
(2262)
where r2/1 is the position of the particle relative to the collar. Now we know that r2/1 is given
as
r2/1 = Ler
(2263)
Then, the velocity and acceleration of the particle are given, respectively, as
F
F
v2
v2
F
F
v1 + F v2/1
a1 + F a2/1
(2264)
(2265)
Now we already have F v1 and F a1 from Eqs. (2260) and (2261), respectively. Computing the
velocity of the particle relative to the collar, we have
F
v2/1 =
Now we have
d
d
r2/1 =
r2/1 + F R r2/1
dt
dt
d
r2/1
dt
F
R r2/1
(2266)
(2267)
ez Ler = Le
(2268)
v2/1 = Le
(2269)
Next, the acceleration of the particle relative to the collar in reference frame F is given as
F
a2/1 =
Rd
d F
F
v2/1 =
v2/1 + F R F v2/1
dt
dt
(2270)
d F
v2/1
dt
F
R F v2/1
(2271)
ez Le = L2 er
(2272)
38
Adding Eqs. (2271) and (2272), we obtain the acceleration of the particle relative to the collar
in reference frame F as
F
a2/1 = L2 er
(2273)
Adding Eqs. (2261) and (2273), the acceleration of the particle is given as
F
x L2 er
a2 = xE
(2274)
Then the acceleration of the center of mass of the collar-particle system is obtained as
F
a=
m2
m1 F a1 + m2 F a2
x
= xE
L2 er
m1 + m2
m1 + m2
(2275)
Kinetics
The free body diagram of the system consisting of the collar and the particle is shown in Fig. 2
11.
N
Ff
Fs
Figure 211
Free body diagram for collar attached to spring, damper, and rotating
arm with particle.
It is noted explicitly that the reaction force exerted by the arm on the collar is not included in
the free body diagram of the collar-particle system because this reaction force is internal to the
system. Consequently, the forces acting on the collar particle system are
Fs
Ff
N
=
=
=
Spring force
Force of viscous friction
Reaction force of track on system
Now we know that the reaction force N must act in the direction orthogonal to the track.
Furthermore, because the motion is planar, the force N must lie in the plane of motion. Consequently, N must lie in the direction of Ey and can be expressed as
N = NEy
(2276)
(2277)
wherevrel is the velocity of the collar relative to the track (because the track is the surface on
which the particle slides and the attachment point of the spring and damper is fixed to the
track). Therefore,
x
vrel = Fv1 = xE
(2278)
which implies that the force of viscous friction is given as
x
Ff = c xE
(2279)
Fs = K( 0 )us
(2280)
39
r1 r O
xEx
=
= Ex
kr1 rO k
x
(2281)
(2282)
Adding Eqs. (2276), (2279), and (2282), the force acting on the collar-particle system is given
as
x K(x 0 )Ex = c x
+ K(x 0 ) Ex + NEy
F = N + Ff + Fs = NEy c xE
(2283)
(2285)
(2286)
Equating Ex and Ey components in Eq. (2287), we obtain the following two scalar equations:
m2 L2 sin t = c x
+ K(x 0 ) Ex
(m1 + m2 )x
(2288)
m2 L2 cos t
(2289)
Observing that Eq. (2288) has no unknown reaction forces and all other quantities (with the
exception of x) are known, the differential equation of motion is given as
m2 L2 sin t = c x
+ K(x 0 )
(m1 + m2 )x
(2290)
Rearranging Eq. (2290), we obtain
+
x
K
m2 L2
K
c
+
x
x=
sin t +
0
m1 + m2
m1 + m2
m1 + m2
m1 + m2
(2291)
Suppose now that we define M = m1 + m2 . Then the differential equation of Eq. (2291) can be
written as
K
m2 L2
K
c
+
+
x
x=
sin t +
0
(2292)
x
M
M
M
M
(2293)
xeq = 0
(2294)
40
c
K
m2 L2
+
z
z=
sin t
M
M
M
(2295)
m2 L2
sin t
M
(2296)
(2297)
K
M
(2298)
Therefore,
2n A =
K
m2 L2
A=
M
M
(2299)
m2
m2 L2
=
K
M
2
(2300)
(2301)
f (t) = 2n Aeit
(2302)
where
is the normalized input function. The transfer function G(i) = X/A for the system of
Eq. (2301) is given as
1
G(i) =
(2303)
2
1
+ i2
n
n
where the magnitude and phase of G(i) are given, respectively, as
1/2
2
1
i2
n
n
Im [G(i)]
n
= tan1
2
Re [G(i)]
1
n
|G(i)| =
2
1
+ i2
n
n
() = tan1
(2304)
(2305)
Also, recall the time response for the system of Eq. (2301) is given as
x(t) = A|G(i)|ei(t)
(2306)
41
(2307)
K
c
+
q
q = f (t)
M
M
(2308)
(2309)
(2310)
Observing that
ei(t) = cos (i(t )) + i sin (t )
The time response of the system in Eq. (2292) is the imaginary part of q(t), i.e.,
"
#
m2
2
z(t) = Im[q(t)] =
L |G(i)| sin (t )
M n
(2311)
(2312)
42
2.6
An important problem in vibratory systems is base motion isolation. The problem of base
motion isolation is as follows. Consider an object in vibratory (i.e., connected to a linear spring
and a viscous damper) such that the spring and damper are connected at the other end to a
system that is itself vibrating. The objective is to isolate the motion of the mass from this other
system. A good example of a base motion isolation system is the suspension of an automobile
where it is desired to isolate the vibration of the automobile from undulations in the road. In
this section we derive the frequency response of a base motion isolation system.
The basic model for a base isolation system is shown in Fig. 212. The primary object is a
collar of mass m. The collar slides along an inertially fixed horizontal track. The displacement
of the collar is given by x(t) and is measured relative to a point O, where O is fixed to the track.
Attached to the collar is a linear spring with spring constant K and unstretched length 0 and a
viscous damper with damping coefficient c. Attached to the other end of the spring and damper
is a base that slides with known displacement q(t) (again, measured from the inertially fixed
point O) where q(t) is assumed to be a periodic function of the form
q(t) = Aeit
The objective of this study is to determine the frequency response of the system to the motion
of the base and to understand how this frequency response can be used to isolate the motion
of the base from the collar.
Moving Base
q(t)
O
System of Interest
K
m
P
c
x
Figure 212
We begin by deriving the differential equation of motion for the system. Choosing the horizontal shaft and an inertial reference frame (denoted F ), we can define the following coordinate
system fixed in reference frame F :
Ex
Ez
Ey
Origin at O
=
=
=
To the right
Out of page
Ez Ex
qEx
(2313)
xEx
(2314)
The corresponding velocities and accelerations in reference frame F are then given, respectively, as
F
vQ
F
Ex
q
(2315)
x
xE
(2316)
43
F
aQ
F
Ex
q
(2317)
x
xE
(2318)
Next, the free body diagram of the collar is shown in Fig. 213 (where we assume that all motion
takes place in the horizontal plane and thus there is no gravity). It is seen that the forces acting
Ff
Fs
Figure 213
on the collar are due to the spring and gravity. The spring force is given as
Fs = K( 0 )us
(2319)
In this case the length of the spring and the direction along which the spring acts are obtained,
respectively, as
us
kr rQ k = kxEx qEx k = |x q| = x q
r rQ
(x q)Ex
=
= Ex
kr rQ k
xq
(2320)
(2321)
Therefore,
Fs = K(x q 0 )Ex
(2322)
Ff = cvrel
(2323)
x q
Ex = (x
q
)Ex
vrel = F v F vQ = xE
(2324)
q
)Ex
Ff = c(x
(2325)
q
)Ex
F = Fs + Ff = K(x q 0 )Ex c(x
(2326)
Therefore,
The resultant force acting on the particle is then given as
(2327)
K(x q 0 ) c(x
(2328)
+ cx
+ Kx = c q
+ K(q + 0 )
mx
(2329)
(2330)
It is seen that the motion of the base affects the motion of the collar through both the spring
and the damper. Rewriting Eq. (2330) in standard form, we have
+ 2n y = 2n q
+ 2n y
+ 2n q
y
(2331)
44
(2332)
(2333)
(2334)
(2335)
1 + i2 n
Y
=
2
Q
1 n + i2 n
Now, using the expression for G(i) from Eq. (261), Y /Q can be written as
Y
= 1 + i2
G(i)
Q
n
(2336)
(2337)
(2338)
Where we can obtain the magnitude and phase of y(t) as follows. First, we have
Y (i) = Y (i)
G(i)
G(i)
G(i)
= 1 + i2
n
G(i)
G(i)
(2339)
n
G(i)
Then, using the expression for G(i) from Eq. (261), we have
#
"
2
1
i2
|G(i)|2 Q
Y (i) = 1 + i2
n
n
n
Expanding Eq. (2341) gives
"
Y (i) = 1
2
#
3
2
+ 2
i2
|G(i)|2 Q
n
n
(2340)
(2341)
(2342)
Using Eq. (2342), the magnitude and phase of y(t) are given as
#1/2
2
|G(i)|Q
|Y (i)| = 1 + 2
n
3
2
n
() = tan1
2
2
1 n + 2 n
"
(2343)
(2344)
45
Re (H(i))
Im (H(i))
H(i) = 1
2
Im (H(i))
Re (H(i))
2
3
+ 2
i2
n
n
(2345)
(2346)
and
2
2
+ 2
n
n
3
2
n
1
(2347)
(2348)
Therefore,
"
#1/2
2
|Y (i)|
|G(i)|
= 1 + 2
Q
n
(2349)
2
1 + 2 n
|Y (i)|
T (i) =
= "
#
2
2
+ 2
1
n
n
(2350)
The quantity T (i) is called the transmittibility and gives a measure of the amount of the input
(i.e., the motion of the base) that is transmitted to the output. We note several features of the
transmittibility function. First, it is seen that
> 1 , /n < 1
= 1 , /n = 1
(2351)
T (i) =
< 1 , / > 1
n
In other words, the motion transmitted by the base to the system is amplified for low frequencies and it attenuated for high frequencies. Therefore, when designing a base motion isolation
system the parameters and n must be chosen correctly in order to attenuate the input signal. Finally, it is seen that the phase () is zero for low frequencies and all values of , and
approach for large values of /n (although it approach very slowly when is close to
unity). Therefore, the response is in phase with the input when /n is small and is out of
phase with the input when /n is large.
46
= 0.075
= 0.1
= 0.25
= 0.5
=1
|T (i)|
0
0
0.5
1.5
/n
2.5
Figure 214
Magnitude of Transmittibility Function for a System Under the Influence
of a Base Motion Input q(t) = Qeit .
7 /8
3 /4
47
= 0.075
= 0.1
= 0.25
= 0.5
=1
()
5 /8
/2
3 /8
/4
/8
0
0
0.5
1.5
/n
2.5
Figure 215
Phase of Transmittibility Function for a System Under the Influence of
a Base Motion Input q(t) = Qeit .
48
(2352)
Square Wave
where I is the set of integers. Examples of arbitrary periodic functions include a square-wave
(see Fig. 216) and a sawtooth (see Fig. 217). It is known that any arbitrary periodic function
t
Figure 216
Square-Wave Function.
can be expressed as an infinite series of sines and cosines. This infinite series is called a
Fourier series. Suppose now that we consider a function f (t) that is periodic with period T on
the interval from zero to T . Then, in terms of a Fourier series expansion, the periodic function
f (t) can be written as
X
f (t) =
ck eikt
(2353)
k=
Sawtooth
49
t
Figure 217
Sawtooth Function.
=
=
=
=
=
1
i(k + l)
1
i(k + l)
1
i(k + l)
1
i(k + l)
1
i(k + l)
iT
ei(k+l)t
ei(k+l)T 1
e2 i(k+l)T 1
e2 i(k+l)T 1
(2354)
[1 1] = 0
ZT
ZT
ZT
X
X
ilt
ikt ilt
f (t)e
dt =
ck e
e
dt =
ck
ei(kl)t dt
(2355)
0
where
ZT
0
k=
ei(kl)t dt =
k=
h
iT
1
ei(kl)t
0
i(k l)
(2356)
50
h
i
1
ei(kl)2 1
i(k l)
(2357)
Suppose now that we let m = k l (we note that, because k and l are integers, m is also an
integer). Then when m 0 we have
i
1 h 2im
e
1 = 0,
im
(m 0)
(2358)
(2359)
Because both the numerator and denominator approach zero as m 0, we can use LHopitals
rule to obtain
lim
m0
2i e2im
2
2
e2im 1
= lim
=
lim e2im =
=T
m0
im
i
m0
2 /T
(2360)
(2361)
which implies
Z
1 T
f (t)eikt dt,
(k = 0, 1, 2, . . .)
(2362)
ck =
T 0
The expression for ck from Eq. (2362) can then be used in Eq. (2353) to obtain the Fourier
series expansion of the periodic function f (t).
It is noted that a Fourier series can be written in real form as follows. First, we note that
Z
1 T
f (t)eikt dt
(2363)
ck =
T 0
Then, we have
ck + ck =
1
T
ZT
0
Now we have
h
i
f (t) eikt + eikt dt
ei + ei
2
cos =
from which we obtain
ck + ck =
2
T
ZT
(2364)
(2365)
f (t) cos(kt)
(2366)
h
i
f (t) eikt eikt dt
(2367)
Similarly,
ck ck =
1
T
ZT
0
ei ei
2i
sin =
from which we obtain
ck ck =
2i
T
(2368)
ZT
f (t) sin(kt)dt
(2369)
ck + ck
(2370)
ck ck
(2371)
51
ck
ak + ibk
2
ak ibk
2
(2372)
(2373)
k=
= c0 +
ck eikt =
k=1
ck e
1
X
k=
ikt
ck eikt + c0 +
ck e
ck eikt
k=1
(2374)
ikt
k=1
Substituting the expressions for ck and ck into this last equation, we obtain
f (t) =
X
ak + ibk ikt X ak ibk ikt
a0
+
e
+
e
2
2
2
k=1
k=1
(2375)
Rearranging, we have
f (t) =
=
X
X
eikt + eikt
eikt eikt
a0
+
i
ak
bk
2
2
2
k=1
k=1
X
X
a0
eikt + eikt
eikt eikt
+
+
ak
bk
2
2
2i
k=1
k=1
(2376)
X
X
a0
+
ak cos(kt) +
bk sin(kt)
2
k=1
k=1
(2377)
Eq. (2377) is a real form of a Fourier series for an arbitrary periodic function f (t).
Example 25
Consider the following function:
f (t + kT ) =
1
1
,
,
kT t + kT < kT + T /2
kT + T /2 t + kT leq(k + 1)T
(k = 0, 1, 2, . . .)
(2378)
Determine both the complex and real form of the Fourier series expansion of f (t)
Solution to Example 25
From Eq. (2362), the coefficients of a Fourier expansion of a periodic function are given in
complex form as
Z
1 T
ck =
f (t)eikt dt,
(k = 0, 1, 2, . . .)
(2379)
T 0
Now because the square-wave takes on values of 1 and -1 on the intervals t [0, T /2) and
t [T /2, T ), respectively, we need to compute the integral in two parts, i.e.,
"Z
"Z
#
#
ZT
ZT
T /2
T /2
1
1
ikt
ikt
ikt
ikt
e
dt +
e
dt
e
dt =
e
dt
(2380)
ck =
T
T
0
0
T /2
T /2
52
eikt dt =
1
1
,
,
2
ik
(2381)
i
1 h ik
e
1
ik
(2382)
k = 1, 3, 5, . . .
k = 2, 4, 6, . . .
(2383)
Therefore,
Z T /2
0
eikt dt =
,
,
k = 1, 3, 5, . . .
k = 2, 4, 6, . . .
(2384)
(2385)
(2386)
Now we know that e2ik = 1. Furthermore, we can apply the result of Eq. (2383) to obtain
(
ZT
2
, k = 1, 3, 5, . . .
ik
ikt
(2387)
e
dt =
0 , k = 2, 4, 6, . . .
T /2
Substituting the result of Eqs. (2384) and (2387) into Eq. (2380), we obtain
(
4
, l = 1, 3, 5, . . .
ikT
ck =
0 , k = 2, 4, 6, . . .
(2388)
Now we consider the special case of k = 0 (which was not easily done earlier because k appears
in the denominator of the anti-derivative of eikt . In the case where k = 0, we have
ZT
Z T /2
ZT
dt = 0
(2389)
dt +
f (t)dt =
c0 =
0
T /2
It is noted that the value of c0 could have been deduced from the fact that the function is
odd. Now in order to obtain only the odd values of l in the Fourier series, we can make the
substitution
k = 2m 1
(2390)
where m = 0, 1, 2, . . .. Then, using the fact that = 2 /T , the Fourier series representation
of the square-wave function of Eq. (2378) is given as
f (t) =
2
i
ei(2m1)t
(2391)
m=
Next, using Eqs. (2370) and (2371), we can write the Fourier series of Eq. (2391) in real form
as follows. First, we have
ak = ck + ck =
4
4ikT 4ikT
4
+
=
=0
ikT
ikT
k2 2 T 2
(2392)
53
Next,
4
4
8
=
ikT
ikT
ikT
ibk = ck ck =
which implies that
8
4
1
=
i
ikT
k
Then, using the real form of the Fourier series as given in Eq. (2377), we obtain
bk =
f (t) =
X
4
sin(kt)
k
k=1
(2393)
(2394)
(2395)
2.8
Using the results of Section 2.7, we can now obtain the response of the second-order differential
equation
+ 2n x = f (t)
+ 2n x
(2396)
x
to a general periodic input f (t). First, recall the response of the system of Eq. (271) [i.e., the
particular solution] to the complex periodic input A2n eit as
x(t) = AG(i)eit = A|G(i)|ei(t)
(2397)
where |G(i)| and () were the magnitude and phase of the transfer function G(i), where
G(i) was given from Eq. (261) as
G(i) =
1
2
+ i2 n
(2398)
Correspondingly, |G(i) and () were given from Eqs. (283) and (288), respectively, as
|G(i)| = "
2 #2
1/2
2
+ 2
n
n
() = tan1
n
2
1
n
(2399)
(2400)
ck eikt
(2401)
k=
Then, because the system of Eq. (2396) is linear, the principal of superposition applies, i.e., the
particular solution to the input of Eq. (2401) is the sum of the terms in the infinite series. First,
let us determine the response of the system of Eq. (2396) to the input
fk (t) = ck eikt
(2402)
54
(2403)
Then, the response of the system of Eq. (2396) to the input of Eq. (2403) is given as
xk (t) = Ak |G(ik )|ei(kt(k))
(2404)
where |G(ik)| and (k) are obtained from Eqs. (2399) and (2400), respectively, as
|G(ik)|
(k)
2 #2
2 1/2
k
k
1
+ 2
n
n
k
2
n
1
tan
2
k
1
n
"
(2405)
(2406)
2 k
T
(k)
k =
(2407)
(2408)
Then, applying the principal of superposition and using the Fourier series representation of an
arbitrary periodic input f (t) with period T [where f (t) is given by Eq. (2353)], the response is
given as
X
X
x(t) =
xk (t) =
Ak |G(ik )|ei(k tk )
(2409)
k=0
k=0
In other words, the response of the system of Eq. (2396) to the periodic input f (t) is the sum
of the responses of Eq. (2396) to the individual periodic inputs ck eikt 2n Ak ejkt that are
the terms in the Fourier series expansion of f (t).
Chapter 3
Response of Multiple Degree-of-Freedom
Systems to Initial Conditions
We now turn our attention to vibrating systems with more than one degree-of-freedom. As
opposed to single degree-of-freedoms systems, whose dynamics are described by a single differential equation, systems with n degrees of freedom systems are described by a system of n
differential equations. Moreover, this system of differential equations is, in general, coupled
(meaning that the dynamics of each object in the system depend on one another). In this chapter we will begin the study of vibrations of multiple degree-of-freedom systems by studying systems without any time-varying external forcing. The study of unforced two degree-of-freedom
systems will itself be divided into two parts: (1) systems without damping and (2) systems with
damping.
3.1
The most basic class of systems in the study of multiple degree-of-freedom vibratory systems
is the class of undamped and unforced systems. In particular, in this section we develop a
generic mathematical model for linear time-invariant (LTI) undamped and unforced multiple
degree-of-freedom systems and develop the mathematics associated with characterizing the
response of these systems.
a1
a2
1 Ex
x
(31)
2 Ex
x
(32)
56
K1
O
K3
K2
m1
m2
x1
x2
Figure 31
Two blocks of mass m1 and m2 connected in tandem to three springs
with spring constants K1 , K2 , and K3 with corresponding unstretched lengths 10 , 20 ,
and 30 .
where Ex is the unit vector in the rightward direction. Next, the forces acting on each collar are
given, respectively, as
F1
F2
Fs1 + Fs2
Fs2 + Fs3
(33)
(34)
where we note that, because spring 2 lies between the two blocks, the force exerted by spring 2
on m1 is equal and opposite the force exerted by spring 2 on m2 (i.e., because Fs2 acts on m1 ,
Fs2 acts on m2 ). Now the forces exerted by each of the three springs are given, respectively,
as
Fs1
Fs2
Fs3
K1 (1 10 )us1
(35)
K3 (3 30 )us3
(37)
K2 (2 20 )us2
(36)
x1
(38)
(39)
x2 x1
x2
(310)
where is the length of the track. Next, the unit vectors in the directions from the attachment
points of each spring to the corresponding blocks are given, respectively, as
us1
us2
us3
Ex
(311)
Ex
(312)
Ex
(313)
We note that us2 = us3 = Ex because the attachment points of springs 2 and 3 lie ahead of
the positions of the first and second block, respectively. Then the spring forces are given as
Fs1
Fs2
Fs3
K1 (x1 10 )Ex
(314)
K3 ( x2 30 )Ex
(316)
K2 (x2 x1 20 )Ex
(315)
57
(317)
1 Ex
K1 (x1 10 )Ex + K2 (x2 x1 20 )Ex = m1 x
(318)
(319)
1 + K1 x1 K2 (x2 x1 ) = K1 10 K2 20
m1 x
(320)
Rearranging, we obtain
(321)
(322)
2 Ex
K2 (x2 x1 20 )Ex + K3 ( x2 30 )Ex = m2 x
(323)
(324)
2 + K2 (x2 x1 ) + K3 x2 = K2 20 + K3 ( 30 )
m2 x
(325)
Rearranging, we obtain
(326)
The system of two differential equations describing the motion of the two collars is then given
as
1 + (K1 + K2 )x1 K2 x2 = K1 10 K2 20
m1 x
2 K2 x1 + (K2 + K3 )x2 = K2 20 + K3 ( 30 )
m2 x
(327)
(328)
The system of differential equations given in Eqs. (327) and (328) can be written in matrix
form as
#
# "
#"
# "
#"
"
1
K1 10 K2 20
x1
K1 + K2
K2
x
m1
0
(329)
=
+
2
K2 20 + K3 ( 30 )
x2
K2
K2 + K3
x
0
m2
+ KX = b
MX
(330)
KXeq = b
(331)
Y = X Xeq
(332)
Now it is seen that the condition for static equilibrium of the system in Eq. (330) is given as
Now let
58
We then have
(333)
(334)
(335)
+ K(X Xeq ) = 0
MX
(336)
=Y
, Eq. (336) can be rewritten as
Noting that Y = X Xeq and X
+ KY = 0
MY
(337)
It is seen that Eq. (337) has a similar mathematical form to the single degree-of-freedom system, the difference being that in this case we have a matrix and column-vector differential
equation (or, alternatively, a system of differential equations) as opposed to a scalar differential equation. Consequently, the solution to Eq. (337) will itself be a column vector. For the
general case of n degrees of freedom the quantities M, K, and Y are given as follows:
m11
m21
..
.
mn1
k11
k21
.
.
.
kn1
m12
m22
..
.
mn2
k12
k22
..
.
kn2
..
.
..
.
m1n
m2n
..
.
mnn
k1n
k2n
..
.
knn
(338)
(339)
(340)
(341)
u
q
(342)
u
q
(343)
Substituting the expressions from Eqs. (341) and (343) into Eq. (337) gives
M
qu + Kqu = 0
(344)
(345)
(346)
59
Now because u is a column vector, we see that the quantities uT Mu and uT Mu are scalars.
Suppose now that we let
uT Ku
= uT Ku = (uT Mu)
(347)
= T
u Mu
Now it can be shown that the matrices M and K are symmetric and positive definite (see Appendix A for the definition of a positive definite matrix). Consequently, we have
uT Ku
T
u Mu
>
>
0 u 0
(348)
0 u 0
(349)
uT Mu
q + uT Muq = 0
(350)
(351)
+ q = 0 = q
= q
q
(352)
Mu + Ku = 0
(353)
Ku = Mu
(354)
det (M K) = 0
(355)
Furthermore, because the eigenvalues must be positive, the general solution of Eq. (352) is
given as
n
X
q(t) =
C1k cos k t + C2k sin k t = Ck cos(k t k )
(356)
k=1
2k
where
= k and the constants C1k and C2k (equivalently, Ck and k ) are determined from the
initial conditions. Observing that there will be two eigenvalues and eigenvectors in Eq. (354),
we obtain
n
X
Y(t) =
qk uk Uq
(357)
k=1
where
U
u1
q1
q2
.
.
.
qn
u2
un
(358)
(359)
Suppose now that we return to Eq. (341). Then for each eigenvector of Eq. (354) we have
Kuk = k Muk ,
(k = 1, . . . , n)
(360)
60
Now let
KU = MU
1
0
.
.
=
.
.
.
.
0
We then obtain
0
2
0
..
.
0
0
2
0
..
.
0
0
0
..
.
..
.
0
0
..
.
..
.
0
0
..
.
0
n
..
.
0
..
.
0
0
0
..
.
(361)
0
n
(362)
KU = MU
(363)
UT KU = UT MU
(364)
+ KX = MU
MX
q + KUq = 0
(365)
Returning to the original differential equation and Eq. (357), we see that
(366)
(367)
(368)
(369)
Finally, because is diagonal, we can write Eq. (369) as a set of scalar equations of the form
k + k qk = 0,
q
(k = 1, . . . , n)
(370)
Now in order to solve Eq. (370), we need initial conditions. In general we will be given initial
conditions on Y of the form
y10
y20
(371)
Y(0) = Y0 =
..
.
yn0
10
y
y
20
(0) = Y
0 =
Y
(372)
.
.
.
n0
y
61
q(0) = U1 Y(0) =
..
.
qn0
10
q
q
20
(0) = U1 Y(0) =
q
..
.
n0
q
(373)
(374)
(375)
Y(0)
(0)
Y
y10
y20
Y0 =
..
.
yn0
10
y
y
20
0 =
Y
..
.
n0
y
(376)
(377)
n
X
ak k
(379)
k=1
The eigenvalues are then the roots of the characteristic polynomial of Eq. (379), It is important to note that the eigenvalues of an undamped multiple degree-of-freedom problem should
be real and positive because otherwise the solution would not make physical sense. Finally,
the natural frequencies of the two degree of freedom system are then obtained by taking the
square-roots of the eigenvalues, i.e.,
21
22
..
.
2n
=
=
..
.
=
1
2
..
.
n
(380)
62
(k = 1, . . . , n)
(381)
Consequently,
(i M K)wi = 0,
(k = 1, . . . , n)
(382)
u2
..
.
un
..
.
=
q w1
wT
1 Mw1
q w2
wT
2 Mw2
(383)
..
.
wn
wT
n Mwn
u2
..
.
un
..
.
=
q w1
wT
1 Kw1
q w2
wT
2 Kw2
(384)
..
.
wn
wT
n Kwn
q(0)
(0)
q
q10
q20
..
.
qn0
10
q
20
q
..
.
n0
q
U1 Y(0) =
(0) =
U1 Y
(386)
(387)
63
(k = 1, . . . , n)
(388)
subject to the initial conditions given in Eqs. (386) and (387), i.e., the initial conditions for
each k = 1, . . . , n are given as
qk (0)
k (0)
q
qk0
,
k0
q
=
=
(k = 1, . . . , n)
(389)
(k = 1, . . . , n)
(390)
=
=
qk0
,
k0 /k
q
(k = 1, . . . , n)
(391)
q(t) =
(392)
(393)
where q(t) is the column vector assembled from the solution given in Step 5.
Example 31
Consider the undamped two degree-of-freedom system with the following mass and stiffness
matrices
#
"
1 0
(394)
M =
0 1
#
"
2
1
(395)
K =
1
2
Determine the solution to the undamped differential equation
+ KY = 0
MY
with the initial conditions
Y(0)
"
1
1
(0)
Y
"
2
1
64
Solution to Example 31
We will obtain the solution to this problem using the six-step procedure described in section
3.1.3. Following Step 1, we compute the eigenvalues of the weighted eigenvalue problem as
#!
# "
"
2
1
1 0
=0
(396)
det(M K) = det
1
2
0 1
Equation (396) can be rewritten as
det
"
2
1
1
2
#!
=0
(397)
(398)
=
=
1
3
(399)
2 = 3
2 =
(3100)
Following Steps 2 and 3, the eigenvectors of the weighted eigenvalue problem are obtained
from the condition
(i = 1, 2)
(3101)
[i M K] ui = 0,
For the eigenvalue 1 = 1, we have
#
"
"
1
1 2
1
u1 =
1
1
1 2
1
1
u1 = 0
"
1
1
1
1
#"
w11
w12
"
0
0
(3102)
(3103)
(3104)
It is seen that a set of values of w11 and w22 that satisfy Eq. (3104) are
w11
w21
=
=
1
1
(3105)
(3106)
(3107)
"
1
0
0
1
#"
1
1
65
=2
wT1 Mw1 = 2
(3108)
(3109)
1
2
1
2
(3110)
The second eigenvector is obtained in a manner similar to that used to obtain the first eigenvector. In particular, for the eigenvalue 2 = 3, we have
#
#
"
"
1 1
2 2
1
u1 = 0
(3111)
u1 =
1 1
1
2 2
Suppose that we denote the unnormalized second eigenvector as w2 . Then
#
"
w21
w2 =
w22
We then have
"
1
1
1
1
#"
w21
w22
"
0
0
(3112)
(3113)
It is seen that a set of values of w21 and w22 that satisfy Eq. (3113) are
w21
w22
1
1
=
=
(3114)
wT2 Mw2
which implies that
"
1
0
0
1
wT2 Mw2 = 2
(3116)
#"
1
1
(3115)
=2
(3117)
(3118)
(3119)
It is observed that u1 and u2 are unit vectors and are orthogonal with respect to both M and K,
i.e.,
#"
"
#
h
i 1 0
1
2
1
1
uT1 Mu2 =
=0
(3120)
1
2
2
0 1
2
#"
"
#
i
h
1
2 1
1
1
T
2
=0
(3121)
u1 Ku2 =
2
2
1
2
12
66
2
U = u1 u2 =
1
1
2
12
From Eq. (3122) we see that the eigenvector matrix for this example is orthogonal, i.e.,
" 1
#
1
U1 = UT =
2
1
2
12
(3122)
(3123)
(3124)
(3125)
(3126)
1
0
1
q10
T
2
2
q(0) = U Y(0) =
=
=
1
1
q20
2
12
2
#
"
#
#
"
#
"
" 1
1
1
10
2
q
T
2
2
2
(0) = U Y(0) =
=
=
q
1
3
20
1
q
12
2
2
(3127)
(3128)
Following Step 5, we can now solve the differential equations in modal coordinates, i.e., solve
k + 2k qk = 0,
q
(k = 1, . . . , n)
(3129)
2 (0)
(
q1 (0), q
q10 , q20 = 0, 2
3
1
10 , q
20 = ,
q
2
2
(3130)
(3131)
where the initial conditions are reiterated from Eqs. (3127 and (3128). Solving the differential
equation corresponding to k = 1, we have
q1 (t) = c11 cos 1 t + c21 sin 1 t
(3132)
=
=
q10 = 0
10 /1 = (1/ 2)/1 = 1/ 2
q
(3133)
Consequently,
1
q1 (t) = sin t
2
Next, solving the differential equation corresponding to i = 2, we have
q1 (t) = c12 cos 1 t + c22 sin 1 t
(3134)
(3135)
q20 = 2
20 /2 = (3/ 2)/ 3 = 32
q
(3136)
67
Consequently,
p
q2 (t) = 2 cos 3t +
p
3
sin 3t
2
(3137)
The vector solution in modal coordinates is then given from Eqs. (3134) and (3137) as
#
"
1 sin t
q1 (t)
2
q
=
q(t) =
(3138)
q2 (t)
2 cos 3t + 32 sin 3t
Following Step 6, we can now transform the solution in modal coordinates to the original coordinates (i.e., the variable Y) using Eq. (357), i.e., we can obtain Y(t) as
Y(t) = Uq(t)
(3139)
In particular, we can substitute U and q(t) from Eqs. (3122) and (3138) into Eq. (3139) to
obtain
" 1
#
1 sin t
1
2
2
2
q
Y(t) =
1
1
2
2 cos 3t + 23 sin 3t
2
(3140)
3
1
sin
t
cos
cos
3t
+
3t
2
= 21
q1 (t)
q2 (t)
Figure 32
68
y1 (t)
y2 (t)
Figure 33
The pure behavior of the modal coordinate solution as opposed to the impure behavior of the
non-modal coordinate solution is characteristic of undamped multiple degree-of-freedom systems. Essentially, in modal coordinates the solution is being viewed as a system of uncoupled
harmonic oscillators whereas in non-modal coordinates the solution is being viewed as a set
of coupled oscillators. Since the original problem is coupled (through the stiffness matrix K)
we expect that the non-modal coordinate solution will exhibit mixed (i.e., impure) behavior. On
the other hand, because the eigenvector matrix decouples the mass and stiffness matrices, we
expect that each component of the modal coordinate solution will exhibit non-mixed (i.e., pure)
behavior.
Example 32
Consider the undamped two degree-of-freedom system with the following mass and stiffness
matrices
#
"
1 0
(3141)
M =
0 2
#
"
2
1
(3142)
K =
1
3
69
"
1
1
(0)
Y
"
2
1
Solution to Example 32
We will obtain the solution to this problem using the seven-step procedure described in section
3.1.3. Following Step 1, we compute the eigenvalues of the weighted eigenvalue problem as
#!
# "
"
2
1
1 0
=0
(3143)
det(M K) = det
1
3
0 2
Equation (3143) can be rewritten as
det
"
2
1
1
2 3
#!
=0
(3144)
(3145)
7 72 4(2)(5)
7 49 40
73
1,2 =
=
=
= 1, 5/2
2(2)
4
4
(3146)
(3147)
=
=
1
5/2
(3148)
(3149)
Following Steps 2 and 3, the eigenvectors of the weighted eigenvalue problem are obtained
from the condition
(i = 1, 2)
(3150)
[i M K] wi = 0,
For the eigenvalue 1 = 1, we have
"
1 2
1
1
21 3
w1 =
(3151)
(3152)
70
We then have
"
1
1
1
1
#"
w11
w12
"
0
0
(3153)
It is seen that a set of values of w11 and w22 that satisfy Eq. (3153) are
w11
w21
1
1
=
=
(3154)
(3155)
"
1
0
0
2
(3156)
#"
1
1
=3
q
p
wT1 Mw1 = 3
(3157)
(3158)
1
3
1
3
(3159)
The second eigenvector is obtained in a manner similar to that used to obtain the first eigenvector. In particular, for the eigenvalue 2 = 5/2, we have
#
"
"
#
1
2 2
1
1
2
u1 =
w2 = 0
(3160)
1
22 3
1 2
where w2 denotes the unnormalized second eigenvector. Now we have
#
"
w21
w2 =
w22
We then have
"
1
2
1
2
#"
w21
w22
"
0
0
(3161)
(3162)
It is seen that a set of values of w21 and w22 that satisfy Eq. (3162) are
w21
w22
=
=
1
12
(3163)
(3164)
(3165)
12
"
1
0
0
2
#"
1
12
71
3
2
(3166)
3
2
(3167)
(3168)
It is observed that u1 and u2 are unit vectors and are orthogonal with respect to both M and K,
i.e.,
# q2
"
h
i 1 0
1
1
3 =0
uT1 Mu2 =
(3169)
3
3
0 2
16
# q2
"
i
h
2
1
1
1
3 =0
(3170)
uT1 Ku2 =
3
3
1
3
1
6
h
i
1
U = u1 u2 = 3
1
3
2
3
1
6
(3171)
It is seen for this example that the eigenvector matrix is not orthogonal because the eigenvectors u1 and u2 are not orthogonal, i.e.,
q
2
i
h
2
2
1
1
1
1
T
3 =
(3172)
=
0
u1 u2 =
1
3
3
3
3
18
3
2
6
q 1
"
#
2
1
1
2
1
3
3
1
q
q
3
U = 2
=
=
2
1
2 2
1
3
23
3
3
3
Following Step 5, the initial conditions in modal coordinates are given as
#
"
q10
1
q(0) = U Y(0) =
q20
#
"
10
q
1
(0) = U Y(0) =
q
20
q
(3173)
(3174)
(3175)
(3176)
=
=
q(0) = U Y(0) =
1
q20
0
2 2
3
#
#
"
#"
"
4
10
1
2
q
1
2
(0) =
(0) = U1 Y
= q 32 =
q
20
1
q
2 2
3
3
(3177)
(3178)
72
Following Step 5, we can now solve the differential equations in modal coordinates, i.e., solve
+ 2i q = 0,
q
(i = 1, 2)
(3179)
2 (0))
(q2 (0), q
p 4
10 =
3, 3
q10 , q
q
20 = 0, 23
q20 , q
(3180)
(3181)
where the initial conditions are reiterated from Eqs. (3177 and (3178). Solving the differential
equation corresponding to i = 1, we have
(1)
(1)
(3182)
c1
(1)
c2
=
=
q10 = 3
10 /1 = (4/ 3)/1 =
q
4
3
(3183)
Consequently,
p
p
4
4 3
q1 (t) = 3 cos t + sin t = 3 cos t +
sin t
3
3
Next, solving the differential equation corresponding to i = 2, we have
(2)
(2)
(3184)
(3185)
c1
(2)
c2
Consequently,
=
=
q20 = 0
p
p
20 /2 = ( 2/3)/ 5/2 =
q
2
15
(3186)
s
s
2 15
2
5
5
sin
t=
sin
t
(3187)
q2 (t) =
2
15
2
15
The vector solution in modal coordinates is then given from Eqs. (3184) and (3187) as
#
"
3
cos t
3 cos t + 4q
q1 (t)
3
=
q(t) =
(3188)
2 15
q2 (t)
sin 52 t
15
Following Step 6, we can now transform the solution in modal coordinates to the original coordinates (i.e., the variable Y) using Eq. (357), i.e., we can obtain Y(t) as
Y(t) = Uq(t)
(3189)
In particular, we can substitute U and q(t) from Eqs. (3171) and (3188) into Eq. (3189) to
obtain
q
4 3
2
1
sin
t
3
cos
t
+
3
q
3
Y(t) = 13
2 15
16
sin 25 t
15
3
q
(3190)
=
2 10
5
4
cos t + 3 sin t 30 sin 2 t
As with Example 31, again we see the key difference between the solution in modal coordinates (i.e., q) and the solution in the original coordinates (i.e., Y). First, it is seen from Fig. 34
that each modal coordinate contains only a single frequency.
q1 (t)
q2 (t)
73
Figure 34
y1 (t)
Contrariwise, examining
Fig. 33 each component of the solution Y(t) contains both frequencies
p
1 = 1 and 2 = 5/2.
y2 (t)
Figure 35
74
3.2
In section 3.1 we studied the response of undamped and unforced multiple degree-of-freedom
LTI systems. In the process of studying this class of systems, a key result was obtained that
the response was a linear combination of terms that involved the product of periodic functions
with the modal vectors where the frequencies and modal vectors were the eigenvalues and
eigenvectors, respectively, of the weighted eigenvalue problem Ku = Mu. Thus, the response
of an undamped multiple degree-of-freedom LTI system is characterized completely by mass
and stiffness matrices M and K.
We now turn our attention to unforced but damped multiple degree-of-freedom systems.
The key difference between damped and undamped systems is that the eigenvalues and eigenvectors of the weighted eigenvalue problem of the undamped system do not decouple the system into modal coordinates. Instead, the presence of damping makes it such that no general
decoupling can be obtained. However, a particular class of damping exists called modal damping for which the differential equations can be transformed to a decoupled form. In this section
we develop the general model for a damped multiple degree-of-freedom system, show why the
equations cannot be decoupled in the case of general damping, and develop the results for the
case of modal damping.
3.2.1 Model Problem: Two Blocks with Linear Springs and Dampers
Consider the system shown in Fig. 31 of two blocks of mass m1 and m2 connected in tandem
to three linear springs with spring constants K1 , K2 , and K3 and corresponding unstretched
lengths 10 , 20 , and 30 , and three viscous dampers with damping coefficients c1 , c2 , and c3 ,
respectively.
c1
O
K1
c3
c2
m1
m2
K2
K3
x1
x2
Figure 36
Two blocks of mass m1 and m2 connected in tandem to three springs
with spring constants K1 , K2 , and K3 and corresponding unstretched lengths 10 , 20 ,
and 30 , and three viscous dampers with damping coefficients c1 , c2 , and c3 .
The blocks slide along a horizontal surface of length and the displacements of each collar,
denoted x1 and x2 , respectively, are measured relative to the inertially fixed point O, where
O is located on a vertical wall located at the left end of the surface. Finally, assume that the
surface is frictionless. The objective of this part of this analysis is to derive a system of two
differential equations for the blocks in terms of x1 and x2 .
First, taking the ground as an absolutely fixed inertial reference frame (denoted F ), the
75
velocities and accelerations of the blocks in reference frame F are given, respectively, as
F
v1
v2
a1
a2
1 Ex
x
(3191)
2 Ex
x
(3192)
1 Ex
x
(3193)
2 Ex
x
(3194)
where Ex is the unit vector in the rightward direction. Next, the forces acting on each collar are
given, respectively, as
F1
F2
Fs1 + Fs2 + Ff 1 + Ff 2
Fs2 + Fs3 Ff 2 + Ff 3
(3195)
(3196)
where Fs1 , Fs2 , and Fs3 are the forces exerted by each of the three linear springs and Ff 1 , Ff 2 ,
and Ff 3 are the forces exerted by each of the three viscous dampers. Now we note that, because
spring 2 lies between the two blocks, the force exerted by spring 2 on m1 is equal and opposite
the force exerted by spring 2 on m2 (i.e., because Fs2 acts on m1 , Fs2 acts on m2 ). Similarly,
because the second damper lies between the two blocks, the force exerted by the second viscous
friction on m1 is equal and opposite the force exerted by the second damper on m2 . The forces
exerted by each of the three springs are given, respectively, as
Fs1
Fs2
Fs3
K1 (1 10 )us1
(3197)
K3 (3 30 )us3
(3199)
K2 (2 20 )us2
(3198)
x1
x2
x2 x1
(3200)
(3201)
(3202)
where is the length of the track. Next, the unit vectors in the directions from the attachment
points of each spring to the corresponding blocks are given, respectively, as
us1
us2
Ex
(3203)
(3204)
Ex
Ex
us3
(3205)
We note that us2 = us3 = Ex because the attachment points of springs 2 and 3 lie ahead of
the positions of the first and second block, respectively. Then the spring forces are given as
Fs1
Fs2
Fs3
K1 (x1 10 )Ex
(3206)
K3 ( x2 30 )Ex
(3208)
K2 (x2 x1 20 )Ex
(3207)
c1 vrel,1
(3209)
c2 vrel,2
(3210)
c3 vrel,3
(3211)
(3212)
Now we have
vrel,1
vrel,2
vrel,3
v1 FvO
v2 Fv1
F
v2 vQ
(3213)
(3214)
76
where we have taken into account the velocity of each block relative to the attachment point of
the respective damper. Now because points O and Q are absolutely fixed, we have FvO = FvQ =
0. Furthermore, using the expressions for Fv1 , Fv2 , and Fv3 from Eqs. (3191) and (3192), we
obtain
vrel,1
vrel,2
vrel,3
1 Ex
x
(3215)
2 Ex x
1 Ex = (x
2 x
1 )Ex
x
2 Ex
x
(3216)
(3217)
nd
1 Ex
c1 x
2 Ex
c3 x
(3218)
2 x
1 )Ex
c2 (x
(3219)
(3220)
(3221)
(3222)
(3223)
Rearranging, we obtain
2 + K1 x1 K2 (x2 x1 ) = K1 10 K2 20
1 + (c1 c2 )x
1 + c2 x
m1 x
(3224)
(3225)
(3226)
(3227)
(3228)
Rearranging, we obtain
2 + K2 (x2 x1 ) + K3 x2 = K2 20 + K3 ( 30 )
2 + c2 (x
1 x
2 ) + c3 x
m2 x
(3229)
(3230)
The system of two differential equations describing the motion of the two collars is then given
as
2 + (K1 + K2 )x1 K2 x2
1 + (c1 c2 )x
1 + c2 x
m1 x
1 + (c3 c2 )x
2 K2 x1 + (K2 + K3 )x2
2 + c2 x
m2 x
K1 10 K2 20
K2 20 + K3 ( 30 )
(3231)
(3232)
77
(3233)
(3234)
Now it is seen that the condition for static equilibrium of the system in Eq. (3234) is given as
KXeq = b
(3235)
Y = X Xeq
(3236)
Now let
We then have
(3237)
(3238)
(3239)
(3240)
=Y
, Eq. (3240) can be rewritten as
Noting that Y = X Xeq and X
+ CY
+ KY = 0
MY
(3241)
It is seen that Eq. (3241) has a similar mathematical form to the single degree-of-freedom
system, the difference being that in this case we have a matrix and column-vector differential
equation (or, alternatively, a system of differential equations) as opposed to a scalar differential
equation. Consequently, the solution to Eq. (3241) will itself be a column vector.
It is seen that the difference betwen Eq. (3241) and (337) on page 58 is that Eq. (3241) has
. This addition term is due to the damping that may be part of some
the additional term CY
two degree-of-freedom vibratory systems. Now, we recall from section 3.1.2 that two degreeof-freedom systems without damping [i.e., systems that satisfy Eq. (337)] can be decoupled by
determining the eigenvectors of the weighted eigenvalues problem Ku = Mu. In particular,
using the fact that the eigenvectors are orthogonal to both the mass and stiffness matrices, we
can write
UT KU = UT MU
(3243)
(3244)
78
U KU = =
"
1
0
0
2
(3245)
(3246)
Alternatively, in scalar form we had the system of two decoupled differential equations
k + k qk = 0,
q
(k = 1, . . . , n)
(3247)
Now, in the case where the system is damped, it is seen in general that the eigenvector matrix
will not diagonalize the damping matrix, i.e., in general it is the case that
UT CU C
(3248)
(3250)
As a result, in the case where U diagonalizes C, we would have n scalar equations of the form
+ k qk = 0,
k + ck q
q
(i = 1, . . . , n)
(3251)
However, because C is in general not diagonalizable by U, the second term in Eq. (3250) will
not lead to the form of Eq. (3251). In other words, when using the transformation
Y = Uq
(3252)
(3253)
where is given as
11
21
=
..
.
n1
12
22
..
.
n2
..
.
1n
2n
..
.
nn
(3254)
Because the matrix is not diagonal, the eigenvector U matrix cannot be used to obtain a modal
form for a multiple degree-of-freedom damped linear system.
(3255)
+ (M + K)Y
+ KY = 0
MY
(3256)
Suppose now that we transform Eq. (3256) using the eigenvector matrix U of the weighted
eigenvalue problem Ku = Mu, i.e., we let
Y = Uq
(3257)
79
Then, we have
MU
q + (M + K)U
q + KUq = 0
(3258)
UT MU
q + UT (M + K)U
q + UT KUq = 0
(3259)
(3260)
Suppose now that we choose a mass-normalized eigenvector matrix, i.e., U has the property
that
UT MU
UT KU
(3261)
(3262)
(3263)
Because is diagonal, it is seen that the matrix I + is also diagonal. In scalar form we have
the following two differential equations:
k + ( + k )
q
q + k q = 0,
(i = 1, . . . , n)
(3264)
Suppose now that we consider the case where + i > 0 for i = 1, 2. Furthermore, using the
earlier notation, we can write Eq. (3264) as
k + k qk = 0,
k + k q
q
(i = 1, . . . , n)
(3265)
where
k = + k ,
(i = 1, . . . , n)
(3266)
It is seen that Eq. (3265) is a system of two decoupled second-order LTI differential equations,
each of which can be solved by the techniques for single degree-of-freedom systems. The form
of damping given in Eq. (3255) is called modal damping because it is diagonalizable by the
eigenvector matrix of the weighted eigenvalue problem. Suppose now that we define
k
k
=
=
2k
,
2k k
(k = 1, . . . , n)
(3267)
(k = 1, . . . , n)
(3268)
It can be seen that the each differential equation in Eq. (3268) is in the standard form and thus
can be solved via the techniques for a single degree-of-freedom LTI.
80
Example 33
Consider the free damped two degree-of-freedom system
+ CY
+ KY = 0
MY
where the mass, damping, and stiffness matrices, are given, respectively, as follows
#
"
1 0
M =
0 2
#
"
3
2
C =
2 4
#
"
2 1
K =
1
3
Determine a system of two uncoupled differential equations of the form
+ q
+ q = 0
q
where the matrices and are diagonal.
Solution to Example 33
Using the mass, damping, and stiffness matrices given in the problem statement, it is seen that
C=
"
3
2
2
4
"
1
0
0
2
+2
"
2
1
1
3
= M + 2K
(3269)
Consequently, C is a modal damping matrix for this problem. We can then use the eigenvector
matrix associated with the weighted eigenvalue problem to decouple the original differential
equations. Now recall that the matrices M and K are the same as those used in Example 32.
In particular, recall that the mass-normalized eigenvector matrix in Example 32 is given from
Eq. (3171) on page 71 as
q
U=
1
3
1
3
2
3
16
"
1
1
3
3
T
1
U CU = q 2
1
1
1
1
(3270)
(3271)
(3272)
5
2
1
3
1
3
2
3
1
6
"
1
0
0
4
(3273)
which implies that the eigenvector matrix also decouples the damping matrix C. Now let us
return to the original system. Applying the transformation Y = Uq, we obtain
MU
q + CU
q + KUq = 0
(3274)
81
(3275)
Then, substituting the expressions for UT MU, UT KU, and UT CU from Eqs. (3271), (3271), and
(3271), respectively, we obtain
#
"
#
"
#
"
1 0
1 0
1 0
+
+
q=0
(3276)
q
q
0 4
0 1
0 52
Then, setting (q1 , q2 ) = q, Eq. (3276) is equivalent to the following two scalar differential
equations:
1 + q
1 + q1
q
2 + 4
q
q2 +
5
q
2 2
(3277)
(3278)
It can be seen that Eqs. (3277) and (3278) are decoupled, consistent with the fact that we
have a modally damped system. Finally, recall from Eq. (3269) that the damping matrix for
this example is given as C = M + 2K which implies that = 1 and = 2. Therefore, from
Eq. (3266) we have
1
1
+ 1
(3279)
+ 2
(3280)
Now the eigenvalues of the undamped problem are the same as those given in Eq. (3148) on
page 69 of Example 32, i.e., the eigenvalues are
1
2
(3281)
5
2
(3282)
Therefore,
1
2
1 + 21 = 1 + 2(1) = 1
(3283)
1 + 22 = 1 + 2(5/2) = 4
(3284)
1 and q
2
Examining Eqs. (3277) and (3278) it is seen that the coefficients multiplying the q
terms are 1 and 4, respectively. Consequently, because we have a modally damped system in
this example, determining the values 1 and 2 is equivalent to computing the transformation
UT CU.
Example 34
Consider the system of two nonlinear differential equations
cos m2 l
2 sin
+ k1 x + m2 l
(m1 + m2 )x
+ m2 g sin
cos + m2 l
m2 x
=
=
0
0
Determine (a) the static equilibrium point for the system, (b) the differential equations of motion for values of x and near the static equilibrium point found in part (a), and (c) the differential equations in modal coordinates for the case m1 = m2 = 1, l = 1, and g = 1.
82
Solution to Example 34
(a) Static Equilibrium Point
Let (xeq , eq ) be the static equilibrium point. Then we have
eq = 0
x
eq = 0
,
,
eq = 0
x
eq = 0
=
=
0 = x = 0
0 = sin = 0 = eq = 0
(3285)
Therefore, the equilibrium point for this system is (xeq , eq ) = (0, 0).
=
=
x xeq
eq
(3286)
(3287)
Then,
=
=
=
=
Furthermore, we have
cos
sin
2
(3288)
Therefore, the differential equations near the equilibrium point are given as
+ k1 x + m2 l
(m1 + m2 )x
+ m2 l + m2 g
m2 x
0
0
=
=
Dividing the first differential equation by l (in order to make the system symmetric), we obtain
m1 +m2
x
l
+ l1 x + m2
+ m2 g
+ m2 l
m2 x
=
=
0
0
m1 +m2
l
m2
m2
m2 l
#"
"
k1
l
0
m2 g
#"
"
0
0
(3289)
83
"
"
m1 +m2
l
m2
"
k1
l
(3290)
m2
m2 l
#
(3291)
0
m2 g
(3292)
det
0
1 1
0
1
#!
= det
"
(3295)
2 2
=0
(3296)
We then obtain
(2 2)( 1) 2 = 2 5 + 2 = 0
(3297)
5 25 4(1)(2)
5 17
=
=
2
2
(3298)
It is noted that 5 > 17 which implies that both 1 and 2 are positive (which they should be
because this is an undamped oscillatory system). Then we know that the equations in modal
coordinates are given as
1 + 1 q1
q
2 + 2 q2
q
(3299)
(3300)
5+
5
17
17
q1
(3301)
q2
(3302)
(3303)
84
3.3
Until now all of the theory that we have been discussed has relied on the assumption that the
mass and stiffness matrices are symmetric. In particular, the weighted eigenvalue problem
Ku = Mu produces positive real eigenvalues and orthogonal eigenvectors only in the case
where M is symmetric and positive definite while K is symmetric and positive semi-definite.
Unfortunately, the ability to obtain symmetric mass and stiffness matrices depends upon the
approach used to derive the differential equations. In more advanced courses in vibrations the
differential equations are derived using advanced methods involving Lagrangian mechanics.
However, in our study we have focused on Newtonian formulations. When using Newtonian mechanics, it is often possible to obtain a system of differential equations which, when linearized
about the static equilibrium point, results in non-symmetric mass and stiffness matrices. In
order to demonstrate this point, we will now explore an example for which this lack of symmetry exists. Now, while no systematic procedure exists from which the system can be made
symmetric, we will proceed to discuss briefly through this example how a symmetric form can
be obtained.
Example 35
A collar of mass m1 is constrained to slide along a frictionless horizontal track as shown in
Fig. 37. Attached to the collar is a linear spring with spring constant K and unstretched length
0 . The collar is attached to one end of a rigid massless arm of length l while a particle of mass
m2 is attached to the other end of the arm. Knowing that x describes the displacement of the
collar relative to the track, that the angle is measured from the downward direction, that the
spring is unstretched when x = 0, and that gravity acts downward, determine (a) a system of
two differential equations that describes the motion of the collar-particle system; (b) a system
of differential equations linearized about the static equilibrium point; (c) an alternate system
of differential equations via algebraic manipulation of the system obtained in part (a); and (d) a
system of differential equations from part (c) linearized about the static equilibrium point.
x
m1
Q
l
g
m2
Figure 37
Solution to Example 35
Kinematics
First, let F be a reference frame fixed to the track. Then, choose the following coordinate
system fixed in reference frame F :
Ex
Ez
Ey
Origin at Q
when x = 0
=
=
=
To the right
Out of page
Ez Ex
85
Next, let A be a reference frame fixed to the arm. Then, choose the following coordinate system
fixed in reference frame A:
Origin at Q
=
=
=
er
ez
e
Along QP
Out of page
ez er
The geometry of the bases {Ex , Ey , Ez } and {er , e , ez } is shown in Fig. 38.
Ey
ez , Ez
Ex
Figure 38
er
sin Ex cos Ey
cos Ex + sin Ey
(3304)
(3305)
Now, consistent with the discussion at the beginning of this problem, we establish the kinematics relevant to the system consisting of the collar and the system consisting of the collar and
the particle.
Kinematics of Collar
The position of the collar is given as
r1 = xEx
(3306)
Computing the rate of change of r1 in reference frame F , we obtain the velocity of the collar in
reference frame F as
F
v1 =
a1 =
d
x
(r1 ) = xE
dt
(3307)
d F
x
v1 = xE
dt
(3308)
86
Kinematics of Particle
The kinematics of the collar-particle system are governed by the motion of the center of mass
of the system. Consequently, in order to determine the kinematics of the center of mass of the
collar-particle system, it is first necessary to determine the position, velocity, and The position
of the particle is given as
r2 = r1 + r2/1
(3309)
Now we have
r2/1 = ler
(3310)
Then, adding Eqs. (3310) and (3309), we obtain the position of the particle as
r2 = xEx + ler
(3311)
z
A = e
(3312)
v2 =
d
d
d
r2/1 = F v1 + F v2/1
(r2 ) =
(r1 ) +
dt
dt
dt
(3313)
We already have F v1 from Eq. (3307). Applying the rate of change transport theorem to r2/1
between reference frames A and F gives
F
v2/1 =
d
r2/1 =
dt
d
r2/1 + FA r2/1
dt
(3314)
Now we have
A
d
r2/1
dt
F A
r2/1
(3315)
z ler = le
e
(3316)
Adding Eqs. (3315) and (3316), we obtain the velocity of the particle relative to the collar in
reference frame F as
F
v2/1 = le
(3317)
Then, substituting the results of Eqs. (3307) and (3317) into Eq. (3313), we obtain the velocity
of the particle in reference frame F as
F
x + le
v2 = xE
(3318)
Computing the rate of change of F v2 in reference frame F using the general expression for F v2
as given in Eq. (3313), the acceleration of the particle in reference frame F is given as
F
a2 =
F
F
d F
d F
d F
v2 =
v1 +
v2/1 = F a1 + F a2/1
dt
dt
dt
(3319)
Now we already have F a1 from Eq. (3308). Applying the rate of change transport theorem
between reference frames A and F , we obtain F a2/1 as
F
a2/1 =
Ad
d F
F
v2/1 =
v2/1 + FA F v2/1
dt
dt
(3320)
87
Now we have
A
v2/1
d F
v2/1 = le
dt
z le
= l
2 er
= e
(3321)
(3322)
2 er + le
a2/1 = l
(3323)
Finally, adding Eqs. (3323) and (3308), we obtain the acceleration of the particle in reference
frame F as
F
2 er + le
x l
a2 = xE
(3324)
r=
m1 r1 + m2 r2
m1 + m2
(3325)
Substituting the expressions for r1 and r2 from Eqs. (3306) and (3309), respectively, into
(3325), we obtain
r as
r=
m2
m1 xEx + m2 (xEx + ler )
= xEx +
ler
m1 + m2
m1 + m2
(3326)
Next, the velocity of the center of mass of the collar-particle system in reference frame F is
given as
m1 F v1 + m2 F v2
F
=
v
(3327)
m1 + m2
Substituting the expression for F v1 from Eq. (3307) and the expression for
as
Eq. (3318) into (3327), we obtain F v
F
=
v
)
x + m2 (xE
x + le
m2
m1 xE
x+
= xE
le
m1 + m2
m1 + m2
v2 from
(3328)
Finally, the acceleration of the center of mass of the collar-particle system in reference frame
F is given as
m1 F a1 + m2 F a2
F
a=
(3329)
m1 + m2
a=
a1 and
2 er + le
)
x + m2 (xE
x l
m2
m1 xE
2 er + le
)
x+
= xE
(l
m1 + m2
m1 + m2
(3330)
Kinetics
In order to solve this problem, it is convenient to analyze the following two systems: (1) the
collar and (2) the collar and the particle. The kinetic relationships for each of these two systems
is now established.
88
Kinetics of Collar
The free body diagram of the collar is shown in Fig. 39.
N
Fs
R
m1 g
Figure 39
Using Fig. 39, it is seen that the following forces act on the collar:
N
R
Fs
m1 g
=
=
=
=
The forces acting on the collar are given in terms of the bases {Ex , Ey , Ez } and {er , e , ez } as
N
R
Fs
m1 g
NEy
(3331)
Rer
(3332)
m1 gEy
(3333)
(3334)
It is noted that, because the spring is unstretched when x = 0, the length of the spring is
= x + 10 . Furthermore, it is noted in Eq. (3332) that, from the strong form of Newtons 3r d
law, the force R must lie along the line of action connecting the collar and the particle. The
resultant force acting on the collar is then given as
F1 = N + R + Fs + m1 g = NEy + Rer KxEx m1 gEy
(3335)
Then, substituting the expression for er from Eq. (3304) into (3335), we have
F1 = NEy + R(sin Ex cos Ey ) KxEx m1 gEy
= (R sin Kx)Ex + (N R cos m1 g)Ey
(3336)
Applying Newtons 2nd law to the collar by setting F1 in Eq. (3336) equal to m1 F a1 using the
expression for F a1 from Eq. (3308), we obtain
x
(R sin Kx)Ex + (N R cos m1 g)Ey = m1 xE
(3337)
N R cos m1 g
m1 x
(3338)
(3339)
89
Fs
m1 g
m2 g
Figure 310
Using Fig. 310, it is seen that the following forces act on the collar-particle system:
N
Fs
(m1 + m2 )g
=
=
=
We already have N and Fs from Eqs. (3331) and (3333), respectively. Furthermore, the force
of gravity acting on the collar-particle system is given as
(m1 + m2 )g = (m1 + m2 )gEy
(3340)
(3341)
Applying Newtons 2nd law to the collar-particle system by setting F in Eq. (3341) equal to
(m1 + m2 )F
a using the expression for F
a from Eq. (3330), we obtain
m2
2 er + le
)
x+
(l
(3342)
KxEx + N (m1 + m2 )g Ey = (m1 + m2 ) xE
m1 + m2
Equation (3342) can be rewritten as
2 er + m2 le
x m2 l
KxEx + N (m1 + m2 )g Ey = (m1 + m2 )xE
(3343)
Then, substituting the expressions for er and e from Eqs. (3304) and (3305), respectively,
into Eq. (3343), we obtain
2 (sin Ex cos Ey )
x m2 l
KxEx + N (m1 + m2 )g Ey = (m1 + m2 )xE
(3344)
+ m2 l(cos
Ex + sin Ey )
90
(3345)
N (m1 + m2 )g
cos m2 l
2 sin
+ m2 l
(m1 + m2 )x
2
sin + m2 l
cos
m2 l
(3346)
(3347)
cos
m1 x
(3348)
(3349)
(3350)
(3351)
(3352)
(3353)
A system of two differential equations that describes the motion of the collar-particle system
is then given as
cos m2 l
2 sin + Kx
+ m2 l
(m1 + m2 )x
2
sin m2 l
cos sin m2 g sin + Kx cos
cos m2 l
m1 x
2
(3354)
(3355)
(3356)
(3357)
It is seen from Eqs. (3356) and (3357) that the static equilibrium point is (xeq , eq ) = (0, 0).
Next, let
x = x xeq x
(3358)
= eq
91
Then, assuming that x and are small, we know that all terms involving products of x and
(and products involving derivatives of x and ) are negligible. Furthermor, we know that
cos = cos 1 andsin = sin . The differential equations linearized relative to the
static equilibrium point are then given as
+ Kx
+ m2 l
(m1 + m2 )x
m2 g + Kx
m1 x
(3359)
(3360)
x
0
x
K
0
m1 + m2 m2 l
=
+ K m2 g
0
m1
0
(3361)
It can be seen that the mass and stiffness matrices in Eq. (3361) are not symmetric. Consequently, this form of the linearized differential equations is not suitable for eigenvalueeigenvector analysis using the symmetric weighted eigenvalue problem. In order to use the
aforementioned techniques to decouple the differential equations, it is necessary to obtain a
symmetric form for the linearized dynamics. We now show how to obtain a symmetric form of
the linearized differential equations.
(3362)
(3363)
cos + m2 l(cos
+ sin2 ) + m2 g sin = 0
m2 x
(3364)
(3365)
Now, because the two differential equations in Eqs. (3354) and (3355) are independent and we
have obtained Eqs. (3354) and (3365) via a nonsingular transformation, the two differential
equations in Eqs. (3354) and (3365) are also independent. Consequently, an alternate system
of two differential equations describing the motion of the collar-particle system is given as
cos m2 l
2 sin + Kx
+ m2 l
(m1 + m2 )x
+ m2 g sin
cos + m2 l
m2 x
(3366)
(3367)
+ m2 l + m2 g
m2 x
(3368)
(3369)
92
where we have again used the approximations cos = cos 1 and sin = sin .
Next, dividing Eq. (3368 by l yields the system
m1 + m2
+ K x
+ m2
x
l
l
+ m2 g
+ m2 l
m2 x
(3370)
(3371)
x
0
x
(m1 + m2 )/l m2
K/l
0
=
+
0
0
m2 g
m2
m2 l
(3372)
Unlike the matrix differential equation that was obtain in Eq. (3361), it is seen that the mass
and stiffness matrices in Eq. (3372) are both symmetric. Furthermore (and equally important)
M is positive definite while K is positive semi-definite, thereby making it possible to analyze
the system in Eq. (3372) using the eigenvalue-eigenvector techniques described previously.
Chapter 4
Forced Response of Multiple
Degree-of-Freedom Systems
In Chapter 3 we studied the response of a multiple degree-of-freedom system to initial conditions. We now turn our attention to the response of multiple degree-of-freedom systems to
external time-varying forcing functions. In particular, in this chapter we start with a general
system of linear time-invariant second-order differential equations subject to a general forcing function. We then divide the analysis into two parts. In the first part of this chapter we
study the response of a multiple degree-of-freedom system to nonperiodic inputs whereas in
the second part of this chapter we study the response of multiple degree-of-freedom systems
to periodic inputs.
4.1
The general mathematical model for a forced multiple degree-of-freedom system subject to a
time-varying forcing function is given as
+ CY
+ KY = f(t)
MY
(41)
where f(t) is an external vector forcing function of time. Similar to a single degree-of-freedom
system, the function f(t) is not a function of Y and its derivatives, but is an explicit function of
time.
4.2
Now consider the case of a modally damped system and an input function f(t) that is nonperiodic. Because C is a modal damping matrix, we know that
C = M + K
(42)
Then, from Eq. (3263) we know that the eigenvector matrix U of the weighted eigenvalue
+ CY
+
problem Ku = Mu can be used to decouple the homogeneous differential equation MY
KY = 0
+ (I + )
q
q + q = 0
(43)
where Y = Uq. However, in this case we have a nonhomogeneous differential equation. Then,
from Eq. (41) we have
MU
q + CU
q + KUq = f(t)
(44)
94
(45)
(46)
where we know for a mass normalized eigenvector matrix that UT KU = . It is seen that the
system of differential equations given by Eq. (46) is also decoupled with the exception that the
right-hand side is not zero. However, as we will soon see, the fact that the right-hand side is
nonzero does not pose a computational problem for modally damped systems. In particular,
let
F1
F2
(47)
UT f =
..
.
Fn
k + ( + k )
q
qk + k qk = Fk (t),
(k = 1, . . . , n)
(48)
It is seen that Eq. (48) is a system of n scalar uncoupled second-order LTI differential equations. Consequently, for nonperiodic inputs the differential equations of Eq. (48) can be solved
using the techniques in Chapter 2. In order to see more clearly how these differential equations
can be solved, suppose that each function fk (t), (k = 1, . . . , n) is a linear combination of fundamental nonperiodic function (e.g., linear combinations of functions such as the unit impulse
function, the unit step function, the unit ramp function, etc.). Then each differential equation
in Eq. (48) can be written as
k + ( + k )
q
qk + k qk =
where
m
X
akj hj (t),
j=0
h0 (t)
h1 (t)
h2 (t)
(t)
Zt
h0 ()d
Z0t
h1 ()d
Zt
..
.
hm (t)
(k = 1, . . . , n)
(49)
(410)
hm ()d
In other words, each function hj (t), (j = 0, . . . , n) is a multiple integral of the unit impulse
function. Now suppose that we let gkj (t), (j = 0, . . . , m) be the response of the kth differential
equation in Eq. (48) to the input hj (t), (j = 0, . . . , m). Then the response of the kth differential
equation in Eq. (48) is given as
m
X
qk (t) =
akj gkj (t)
(411)
j=0
95
j=0
X
q1 (t)
m
q2 (t)
a2j g2j (t)
Y(t) = Uq(t) = U
..
= U j=0
.
..
qn (t)
m
X
j=0
(412)
Now we know that the eigenvector matrix can be written in general form
u11
u21
U=
..
.
un1
u12
u22
..
.
un2
..
.
u1n
u2n
..
.
unn
(413)
u11
u21
Y(t) =
..
.
un1
u12
u22
..
.
un2
..
.
u1n
u2n
..
.
unn
m
X
a1j g1j (t)
j=0
X
m
j=0
..
m
X
j=0
n
m
X
X
u1p
apj gpj (t)
p=1
j=0
X
m
X
n
u2p
apj gpj (t)
p=1
j=0
..
n
m
X
X
unp
apj gpj (t)
p=1
n X
m
X
u1p apj gpj (t)
p=1 j=0
X
m
n X
Y(t) =
p=1 j=0
..
n m
X X
p=1 j=0
j=0
(414)
(415)
In other words, not only is the response of the system in the original coordinates Y(t) a linear
combination of the responses to the input functions hj (t), (j = 0, . . . , n), but, due to the
fact that the elements of Y are themselves linear combinations of the elements of q (due to
the eigenvector matrix, U), this response is simultaneously a linear combination of these linear
combinations.
Example 41
Consider the forced damped two degree-of-freedom system
+ CY
+ KY = f(t)
MY
96
where the mass, damping, and stiffness matrices, are given, respectively, as follows
#
"
1 0
M =
0 2
#
"
3
2
C =
2 4
#
"
2 1
K =
1
3
Determine the response of the system to the input f(t) where f(t) is given as
#
"
(t)
f(t) =
u(t)
where (t) and u(t) are the unit impulse function and unit step function, respectively.
Solution to Example 41
Recall that the mass, damping, and stiffness matrices correspond to those of Example 41.
Furthermore, the mass-normalized eigenvector matrix of the weighted eigenvalue problem
Ku = Mu is given from Eq. (3270) on page 80
q
U=
1
3
1
3
2
3
16
(416)
Furthermore, we know from Example 41 that the system for this problem is modally damped.
Consequently, the modal coordinate equations are obtained from Eq. (46) as
+ (I + )
q
q + q = UT f(t)
Again, from Eqs. (3272), and (3273) we have, respectively,
#
"
1 0
UT KU =
5
0 2
"
#
1 0
I + UT CU =
0 4
Finally, using U in Eq. (416), we have
"
# 1
1
1
(t) + 1 u(t)
(t)
3
3
3
3
UT f = q 2
= q2
u(t)
16
(t) 16 u(t)
3
3
(417)
(418)
(419)
(420)
F2 (t)
1
X
1
1
(t) + u(t) = a10 h0 (t) + a11 h1 (t)
a1j hj (t)
3
3
j=0
s
1
X
2
1
a2j hj (t)
(t) u(t) = a20 h0 (t) + a21 h1 (t)
3
6
j=0
(421)
(422)
where h0 (t) = (t) and h1 (t) = u(t). Furthermore, the coefficients akj , (k = 1, 2), (j = 0, 1)
are given as
1
1
a10 = 3 , a11 =
3
q
(423)
1
2
a20 =
,
a
=
21
3
6
97
Suppose now that we let g10 (t) and g11 (t) be the response of the first modal coordinate, q1 (t),
to the functions h0 (t) = (t) and h1 (t) = u(t), respectively. Then, from the principle of
superposition, the total response of the first modal coordinate is given as
1
1
q1 (t) = a10 g10 (t) + a11 g11 (t) = g10 (t) + g11 (t)
3
3
(424)
Similarly, suppose that we let g20 (t) and g21 (t) be the response of the second modal coordinate,
q2 (t), to the functions h0 (t) = (t) and h1 (t) = u(t), respectively. Then, from the principle of
superposition, the total response of the second modal coordinate is given as
s
1
2
g20 (t) g21 (t)
q2 (t) = a20 g20 (t) + a21 g21 (t) =
(425)
3
6
The modal coordinate response is then given in vector form as
# 1
"
g10 (t) + 1 g11 (t)
q1 (t)
3
3
= q2
q(t) =
q2 (t)
g20 (t) 1 g21 (t)
3
(426)
Then, using the fact that Y = Uq, we obtain the response of the original system as
q "
q 1
# 1
2
2
g10 (t) + 1 g11 (t)
1
q
(t)
1
3
3
3
3 q
Y = Uq = 13
= 13
1
2
q2 (t)
(427)
(428)
It is important to observe that the solution obtained in Eq. (428) is identical to that given in
Eq. (415). In particular, for this example we have m = 1 and Eq. (415) reduces to
2 1
X X
u1p apj gpj (t)
p=1 j=0
Y(t) = 2 1
(429)
X X
=
=
1
3
1
3
,
,
u12
u22
=
=
2
3
1
(430)
q q
1
2
2
1
1
1 g21 (t)
g
(t)
20
3 3 g10 (t) + 3 g11 (t) + 3
6
q 3
Y(t) =
1 1
1
2
g
(t)
g (t)
3 20
3
3
3
6
6 21
Equation (431) simplifies to
"
Y=
1
g (t)
3 10
1
g (t)
3 10
(431)
(432)
98
4.3
Suppose now that the forcing function f(t) has the form
f(t) = Feit
(433)
where F is a constant. It is seen that the form of f(t) given in Eq. (433) is a vector periodic
function of time with input frequency . Suppose now that the damping matrix C is assumed
to be modal, i.e.,
C = M + K
(434)
Suppose further that we transform the variable Y via the eigenvector matrix U as
Y = Uq
(435)
MU
q + CU
q + KUq = Feit
(436)
UT MU
q + UT CU
q + UT KUq = UT Feit
(437)
+ q
+ q = UT Feit
q
(438)
We then obtain
which further implies
Now, assume that U has been mass-normalized. Then, because we have assumed that the
system is modally damped, this last equation can be written as
.
.
.
.
.
.
0
1
0
.
.
.
.
.
.
0
0
2
0
0
..
.
..
.
0
..
.
0
0
2
..
.
0
0
0
..
.
..
.
0
..
.
0
..
.
0
0
0
..
.
0
n
0
0
..
0
n
(439)
(440)
UT F =
..
.
fn
(441)
(k = 1, . . . , n)
(442)
Now let
k
k
2k k
(443)
2k
(444)
99
where k , (k = 1, . . . , n) and k are the damping ratios and natural frequencies for the coordinates qk , (k = 1, . . . , n). Then we can write
k + 2k qk = fk eit ,
k + 2k k q
q
(k = 1, . . . , n)
(445)
Finally, let
fk = Ak 2k ,
(k = 1, . . . , n)
(446)
(k = 1, . . . , n)
(447)
It is seen that Eq. (447) is a system of decoupled equations (i.e., modal coordinate equations)
and are in the standard form as given in Eq. (250), i.e., each equation in Eq. (447) has the form
+ 2n x = 2n Aeit
+ 2n x
x
(448)
(449)
1
2
+ i2
(450)
and |G(i)| and = () are the magnitude and phase of G(i), respectively. Then, for
each modal coordinate the solution is given as
qk (t) = Ak |Gk (i)|ei(tk )
(451)
where Gk (i) is the transfer function associated with the kth modal coordinate, i.e.,
Gk (i) =
1
2
(452)
+ i2k
k
and |Gk (i)| and k = k () are the magnitude and phase of Gk (i), respectively. Then the
solution Y(t) is obtained as
Y = Uq
(453)
where
q1
q2
q=
..
.
qn
Consequently,
Y = Uq = U
A1 |G1 (i)|ei(t1 )
A2 |G2 (i)|ei(t2 )
=
..
.
An |Gn (i)|ei(tn )
q1
q2
..
.
qn
= U
A1 |G1 (i)|ei(t1 )
A2 |G2 (i)|ei(t2 )
..
.
An |Gn (i)|ei(tn )
(454)
(455)
100
Example 42
Consider the two degree-of-freedom system
+ CY
+ KY = F
MY
where
F=
"
1
1
eit
where the mass, damping, and stiffness matrices, are given, respectively, as follows
#
"
1 0
M =
0 2
#
"
3
2
C =
2 4
#
"
2 1
K =
1
3
Determine the time response of the above system.
Solution to Example 42
Recall that M, C, and K correspond to those given in Example 33. Furthermore, recalling the
mass-normalized eigenvector matrix U from Eq. (3270) on page 80 of Example 33, we have
q
from which we obtain
U=
T
U =
1
3
1
3
1
q3
2
3
2
3
1
6
1
3
16
#
1
1
2
1
3
3
3
= q2
UT F = q 2
1
16
16
3
3
(456)
(457)
(458)
Now, recall that Example 33 was modally damped. Consequently, the unforced system can be
decoupled as given in Eqs. (3277) and Eqs. (3278). Then, using the result of Eq. (458), we
obtain the forced differential equations in modal coordinates as
1 + q
1 + q1
q
2 + 4
q
q2 + 25 q2
eit
3
q
2
16 eit
3
(459)
(460)
Also recall from Example 33 that the eigenvalues of the undamped system
1
2
(461)
5
2
(462)
=
=
1
q
(463)
5
2
(464)
101
1 = 1
(465)
2 = 4
(466)
1 and
where we recall that 1 and 2 are the coefficients associated with the terms involving q
2 , respectively. We then obtain
q
1
1
2
(467)
4
q
5
2
=2
2
5
(468)
2
f1 =
3
q
2
f2 =
A2 22
(469)
1
6
(470)
3
q
A2
A1
2
5
(471)
2
3
1
6
1
6
(472)
A2
3
q
2
5
(473)
2
3
(474)
A1 |G1 (i)|ei(t1 )
A2 |G2 (i)|e
i(t2 )
(475)
(476)
G2 (i)
1
2
1
2
+ i21
1
+ i22
2
(477)
(478)
and 1 , 2 , 1 , 2 , A1 , and A2 are as obtained earlier in the solution to this example. Finally,
the solution in the original coordinates defined by Y is given as
q "
#
2
1
q1 (t)
3
3
Y = Uq =
(479)
q2 (t)
1
1
3
102
4.4
Suppose now that we consider again a multiple degree-of-freedom LTI system subject to a
periodic forcing function, i.e.,
+ CY
+ KY = Feit
MY
(480)
Now because the input is periodic, we know that the steady-state output will also be periodic,
i.e., Y(t) will have the form
eit
Y(t) = Y
(481)
(t)
Y
(t)
Y
=
=
eit
iY
eit
2 Y
, and Y
into Eq. (480), we obtain
Substiuting the expression for Y, Y
h
i
= Feit
eit 2 M + iC + K Y
(482)
(483)
(484)
Z(i) Z = 2 M + iC + K
(485)
The quantity Z = Z(i) is called the impedance matrix. In terms of the impedance matrix we
can write
(i) = F
Z(i)Y
(486)
(487)
Y(t) = Z1 (i)Feit
(488)
103
where it is noted that Z(i) is symmetric because M, C, and K are symmetric. The inverse of
Z(i) is then given as
#
#
"
"
1
1
z22
z12
z22
z12
Z1 (i) =
=
(493)
=
2
z12
z11
z12
z11
det Z(i)
z11 z22 z12
as
Multiplying Z1 (i) by F where, we obtain Y
"
1
z22
= Z1 (i)F =
Y
2
z12
z11 z22 z12
z12
z11
#"
F1
F2
"
1
Y
2
Y
(494)
Consequently,
1
Y
2
Y
z22 F1 z12 F2
2
z11 z22 z12
z12 F1 + z11 F2
2
z11 z22 z12
(495)
(496)
2
Y
4.5
(498)
(499)
(4100)
(4101)
Consider now the undamped system shown in Fig. 41 of a block of mass M attached to two
linear springs with spring constants K1 and K2 and a second block of mass m attached in
tandem connected to the second spring. Furthermore, assume that a force F (t) is applied to
the first block.
Assuming that the unstretched lengths of the springs are 10 and 20 , the force applied to the
block of mass M is given as
F1 = F + Fs1 + Fs2 = F Ex K1 (1 10 )us1 K2 (2 20 )us2
(4102)
Now we have
1
2
kr1 rO k = x1
kr1 r2 k = x2 x1
(4103)
(4104)
104
F (t)
K2
K1
O
M
x1
x2
Figure 41
sorber.
Furthermore,
us1
us2
r1 r O
= Ex
kr1 rO k
r1 r 2
= Ex
kr1 r2 k
(4105)
(4106)
Therefore,
F1 = F Ex K1 (x1 10 )Ex K2 (x2 x1 20 )(Ex ) = K1 (x1 10 ) + K2 (x2 x1 20 ) Ex
(4107)
1 Ex , from Newtons second
Then, because the inertial acceleration of the first block is Fa1 = x
law we obtain
1 Ex
F K1 (x1 10 ) + K2 (x2 x1 20 ) Ex = M x
(4108)
which implies that the first differential equation is given as
1 + (K1 + K2 )x1 K2 x2 = F + K1 10 K2 20
Mx
(4109)
(4110)
2 + K2 x1 + K2 x2 = K2 20
mx
(4112)
It is noted that the system consisting of the block of mass M together with the first spring is
called the main system while the second system consisting of the block of mass m together
with the second spring is called the absorber. The objective of this analysis is to determine
the design that enables the absorber to absorb as much of the response of the main system as
possible.
Observing the form of the differential equations, it is seen that the differential equations
relative to the static equilibrium point are given as
1 (K1 + K2 )y1 K2 y2
My
2 K2 y1 + K2 y2
my
F (t)
(4113)
(4114)
105
where x1 = y1 y1,eq and x2 = y2 y2,eq . Suppose now that we consider the case where F (t)
is periodic of the form F (t) = F1 sin t. Then, because the system is undamped, we know that
the phase of the output will be zero which implies that
1 sin t
y1 (t) = Y
2 sin t
y2 (t) = Y
(4115)
(4116)
Then, using the results of Eqs. (4100) and (4101) we obtain y1 (t) and y2 (t) as
"
#
(K2 2 m)F1
y1 (t) =
sin t
(K1 + K2 2 M)(K2 2 m) K22
#
"
K2 F1
sin t
y2 (t) =
(K1 + K2 2 M)(K2 2 m) K22
(4117)
(4118)
where
1
Y
2
Y
"
"
(K2 2 m)F1
(K1 + K2 2 M)(K2 2 m) K22
K2 F1
(K1 + K2 2 M)(K2 2 m) K22
(4119)
(4120)
= m/M
= ratio of absorber to main system
Then Eqs. (4119) and (4120) can be written in terms of n , a , yst and as
1 (/a )2 yst
1 =
Y
[1 + (a /n )2 (/n )2 ] [1 (/a )2 ] (a /n )2
yst
2 =
Y
[1 + (a /n )2 (/n )2 ] [1 (/a )2 ] (a /n )2
(4121)
(4122)
(4123)
F1
n 2 yst
yst
=
(a /n )2
a
K2
(4124)
It is seen from Eq. (4121) that the mass m (i.e., the absorber) will absorb the motion of the
main system if a , i.e., the best natural frequency for the absorber is a = . In other
words, the best design for the absorber is one where the natural frequency of the absorber is
the same as the frequency of the forcing function F (t).
106
Appendix A
Review of Linear Algebra
A.1
Let C and R denote the set of complex and real numbers, respectively. Furthermore, let qi
C, (i = 1, . . . , n) be complex-valued scalars. These scalars can be arranged
in either
a row
or a column as
follows.
When
arranged
in
a
row,
we
can
write
q
=
q
q
.
.
.
q
.
Then
the
1
2
n
quantity q = q1 q2 . . . qn is called a row vector. Alternatively, a row vector can be written as
q = q1 , . . . , qn ,i.e.,
q = q1 , . . . , qn q1 q2 . . . qn
(A1)
q1
q2
q=
..
.
qn
(A2)
the quantity q is called a column vector. Suppose now that we consider a set of complex-valued
coefficients aij (i = 1, . . . , m; j = 1, . . . , n). Also, suppose that we arrange these coefficients as
follows:
a11
a12 a1n
a21
a22 a2n
(A3)
A=
..
..
..
..
.
.
.
.
am1 am2 amn
A=
..
.
an1
a12
a22
..
.
an2
..
.
a1n
a2n
..
.
ann
(A4)
Examining a row vector, a column vector, and a matrix, it is seen that the following are true.
First, an n-dimensional row vector is a matrix of size 1 n, i.e., if q is a row vector, then
q C1n . Next, an m-dimensional column vector is a matrix of size m 1, i.e., if q is a column
vector, then q Cm1 .
108
A.2
Types of Matrices
Identity Matrix
The most basic matrix is the identity matrix.
as
1
0
In =
..
.
0
..
.
0
0
..
.
1
(A5)
In other words, In Rnn and is such that its diagonal elements are unity while its off-diagonal
elements are zero. In index form, we can write the identity matrix as follows:
(
1 , i=j
(A6)
[In ]ij = ij =
0 , ij
where the quantity is the Kronecker delta function. We note that for any column vector q the
identity matrix satisfies the property that
Iq = q
(A7)
Transpose of a Matrix
Let A be an m n complex-valued matrix. Then the transpose of A, denoted AT , is defined as
(A8)
AT =
..
..
..
..
.
.
.
.
a1n a2n amn
It is noted that AT is obtained from A by interchanging the elements of A, i.e., the element aij
in A is equal to the element aji in AT .
12 a
1n
11
a
a
a
22 a
2n
a
21
=
(A9)
A
..
..
..
..
.
.
.
.
m2 a
mn
m1 a
a
109
Skew-Hermitian Matrix
Let A be an n n square complex-valued matrix. Then A is said to be skew-Hermitian if
T , i.e., a complex valued square matrix A is skew-Hermitian if A is equal to the negative
A = A
of the transpose of its complex conjugate. In element form, a Hermitian matrix is one such that
ji , (i, j = 1, . . . , n).
aij = a
Symmetric Matrix
Let A be a square complex-valued n n matrix, Then A is said to be symmetric if A = AT , i.e., a
square matrix is symmetric if it is equal to its transpose. In scalar form, a real-valued square
matrix is symmetric if aij = aji , (i, j = 1, . . . , n).
Skew-Symmetric Matrix
Let A be a square complex-valued matrix. Then A is said to be skew-symmetric A = AT , i.e., a
square matrix is skew-symmetric if it is equal to the negative of its transpose. In scalar form, a
matrix is symmetric if aij = aji , (i, j = 1, . . . , n).
Inverse of a Matrix
Let A be a square complex-valued matrix. Then A is said to be invertible if there exists a matrix
A1 such that
AA1 = A1 A = I
(A10)
Orthogonal Matrix
Let A be a square complex-valued matrix. Then A is said to be orthogonal if A1 = AT , i.e., a
matrix is orthogonal if its inverse is equal to its transpose. Because of the property of an
orthogonal matrix, we know that
A1 A = AT A = I
(A11)
where I is the n n identity matrix. Suppose we write an orthogonal matrix in column form as
h
i
A = a1 a2 an
(A12)
Then,
aT1
aT2
..
.
aTn
AT =
Multiplying AT by A gives
AT A =
a1
a2
an
aT1
aT2
..
.
aTn
= ij =
1
0
aT1 a1
aT2 a1
..
.
aTn a1
,
,
(A13)
aT1 a2
aT2 a2
aTn a2
i=j
ij
..
.
aT1 an
aT2 an
..
.
aTn an
=I
(A14)
(A15)
110
Determinant of a Matrix
Let A Rnn be an n n real-valued matrix. Then the determinant of A, denoted det A or |A|,
is defined as
det A = det
(A16)
..
..
..
..
.
.
.
.
an1 an2 ann
and is computed recursively as
a22
a32
..
.
an2
a23
a33
..
.
an3
a21
a22
a12 det
..
.
an1
a21
a22
a1k det
..
.
an1
a23
a33
..
.
an3
a23
a33
..
.
an3
..
.
..
.
..
.
a2n
a3n
..
.
ann
a2n
a3n
..
.
ann
a2(n1)
a3(n1)
..
.
an(n1)
(A17)
det a21 a22 a23 = a11 (a22 a33 a23 a32 ) a12 (a21 a33 a23 a31 ) + a13 (a21 a32 a22 a31 )
a31 a32 a33
(A19)
It is noted that the determinants of 2 2 and 3 3 matrices can be used as building blocks to
compute the determinant of an n n matrix. Also, it is important to understand that a square
matrix is invertible if and only if its determinant is nonzero, i.e.,
A1 exists det A 0
A.3
(A20)
A=
..
..
..
..
.
.
.
.
an1 an2 ann
111
B=
b11
b21
..
.
bm1
a11 + b11
a21 + b21
A+B=
..
.
am1 + bm1
b12
b22
..
.
bm2
..
.
a12 + b12
a22 + b22
..
.
am2 + bm2
b1n
b2n
..
.
bmn
..
.
(A21)
a1n + b1n
a2n + b2n
..
.
amn + bmn
(A22)
It is seen from Eq. (A22) that A + B = B + A. Similar to the sum of two matrices, the difference
between A and B is defined as
a11 b11
a12 b12
a1n b1n
a21 b21
a22 b22
a2n b2n
(A23)
AB=
.
.
..
.
.
.
.
.
.
.
.
am1 bm1 am2 bm2 amn bmn
It is noted that matrices can only be added or subtracted if they are the same size.
ka11
ka21
kA =
..
.
kam1
..
.
ka1n
ka2n
..
.
kamn
(A24)
a21
a22 a2p
b21 b22 b2n
.
AB =
.
.
.
.
.
.
.
.
..
..
..
..
..
..
..
.
bp1 bp2 bpn
am1 am2 amp
Pp
Pp
(A25)
Pp
=
.
.
.
..
..
..
..
.
Pp
Pp
Pp
a
b
a
b
a
b
pk
k1
pk
k2
pk
kn
k=1
k=1
k=1
p
X
alk bkm
(A26)
k=1
It is very important to note that two matrices can be multiplied only if the number of rows of A
must be the same as the number of columns of B (i.e., if A Rm,p and B Rq,n , then A and B
112
can be multiplied only if p = q). This last requirement is called conformability, i.e., two matrices
can be multiplied only if their sizes are conforming. Due to the conformability requirement, the
only case in which both AB and BA are valid operations is if A and B are both square matrices.
Finally, we note that, even in the case of two square matrices A and B it is generally the case
that AB BA (i.e., the matrix product is not commutative).
p q = pT q =
..
.
pn
q1
q2
.
.
.
qn
= p1
p2
pn
q1
i
q2
.
.
.
qn
X
n
=
pi qi
k=1
(A28)
Similarly, if p and q are both row vectors, then the dot product between p and q can be written
as
h
ih
iT
q1 q2 qn
p q = p T q = p1 p 2 p n
q1
n
(A29)
h
i q2
X
pi qi
= p1 p2 p n
.. =
. k=1
qn
We say that two vectors p and q are orthogonal with respect to the standard inner product
(i.e., dot product) if and only if p q = 0, i.e.,
p and q are orthogonal with respect to the standard inner product pT q = 0
(A30)
(A31)
where W Rnn is a weighting matrix. It is noted that two vectors p and q are orthogonal with
respect to the weighting matrix W if and only if pT Wq = 0, i.e.,
p and q are orthogonal with respect to W pT Wq = 0
A.4
(A32)
(A33)
113
The null space of A is denoted N (A). Any vector q that satisfies Eq. (A33) is said to lie
in N (A). Moreover, we say that any vector q that lies in N (A) is annihilated by A (i.e., if
q N (A), then A annihilates q). It is noted that the null space of a square matrix is nonzero
if and only if the determinant of A is zero.
Next, the range space of A is the set of all vectors p Rm (where p 0) for which there
exists a vector q Rn such that
Aq = p
(A34)
The range space of a matrix is denoted R(A). Any vector p that satisfies Eq. (A34) is said to
lie in R(A).
A.5
Let A Rnn be an nn real-valued square matrix. Then the scalar is said to be an eigenvalue
of A with eigenvector u if
Au = u
(A35)
Rearranging Eq. (A35), we obtain
u Au = 0
(A36)
(I A) u = 0
(A38)
Now we know that q = Iq where I is the n n identity matrix. Therefore, Eq. (A36) can be
rewritten as
Iu Au = 0
(A37)
Eq. (A37) can be rearranged as
Eq. (A38) implies one of two things. Either u = 0 or the vector q must lie in the null space
of the matrix I A (i.e., I A must annihilate u). The former case is the trivial solution and
hence is of no interest. Therefore, the latter case must be true. Now, in order for u to lie in the
null space of I A, the matrix I A must have a nonzero null space. Recall that I A has
a nonzero null space, it must be singular (i.e., I A does not have an inverse) and, therefore,
det(I A) = 0
det(I A) = 0
(A39)
Therefore, the condition of Eq. (A38) that leads to a nontrivial value of u is given by Eq. (A39).
Examining Eq. (A39) and using the general form for a determinant from Eq. (A17), it is
seen that det(I A) = 0 is a polynomial in , i.e., det(I A) = 0 can be written as
det(I A) = n + a1 n1 + a2 n2 + + an1 + an =
n
X
k=1
ank k = 0
(A40)
where we note that a0 1. Eq. (A40) is called the characteristic equation of the matrix A. Now,
because A is a real matrix, the coefficients (a0 , . . . , an ) must also be real.
(A41)
In general, the eigenvalues , . . . , n will not be distinct [i.e., Eq. (A41) may have repeated roots].
When an eigenvalue i of Eq. (A41) repeats itself k times (i.e., k roots of Eq. (A40) are equal to
i ), we say that the eigenvalue i has algebraic multiplicity k. For example, suppose that two of
114
the roots of the characteristic equation are equal to . Then in factored form the characteristic
equation would have the factor appear twice [i.e., we would have a factor ( )2 in the
characteristic equation]. In this case the algebraic multiplicity of the eigenvalue would be
two.
Suppose now that m n is the number of distinct eigenvalues of A and let 1 , . . . , m
be the corresponding eigenvalues of A. Next, let k1 , . . . , km be the algebraic multiplicities
of
Furthermore, let the set [1, . . . , n] be partitioned into sets Pi =
1 , . . . , m , respectively.
pi1 + 1, . . . , pi , (i = 1, . . . , m) such that
pi =
i
X
kj
(A42)
j=1
(ri Pi ), (i = 1, . . . , m)
(A43)
It is seen that the sum of the algebraic multiplicities must add to n, i.e.,
m
X
j=1
kj = n
(A44)
Furthermore, the eigenvectors associated with each partition Pi need not be distinct. Suppose
that mi ki is the number of linearly independent eigenvectors associated with each partition
Pi . Then we say that the eigenvalue i , (i = 1, . . . , m) has geometric multiplicity mi ki . In the
case where mi = ki , the algebraic and geometric multiplicities of i are the same.
u2
un
u1
u2
un
u1
(i = 1, . . . , n)
h
1 u1
u2
u1
u2
un
Now let
U
1
0
.
.
.
0
0
2
2 u2
..
.
1
0
..
.
0
(A45)
un
0
2
n un
..
.
(A46)
0
0
0
n
(A47)
(A48)
0
0
0
n
(A49)
115
The matrix U is called the eigenvector matrix of the matrix A. In terms of U, Eq. (A47) can then
be written as
AU = U
(A50)
Now because the eigenvectors of A are complete and form a basis for Rn , it is known that the
matrix U is nonsingular (i.e., U1 exists). Consequently, we can multiply both sides of Eq. (A50)
on the left by U1 to obtain
= U1 AU
(A51)
(A52)
(A53)
Now the right-hand and left-hand sides of Eq. (A53) can be written, respectively, as
i ui
Aui
i u
i
(A54)
A
ui
(A55)
(A56)
i is an
It is seen that Eq. (A56) satisfies the eigenvalue equation of Eq. (A35). Therefore, u
i
eigenvector of A with eigenvalue
A.6
Now consider the special case where A is real and symmetric, i.e., A = AT . Then for eigenvectors
ui and uj with corresponding eigenvalues i and j , respectively, we have
Aui
Auj
i ui
(A57)
j uj
(A58)
Suppose now that we multiply both sides of Eq. (A57) on the left-hand side by uTj and multiply
both sides of Eq. (A58) on the left-hand side by uTi . We then obtain
uTj Aui
uTi Auj
uTj i ui
(A59)
uTj j uj
(A60)
i uTj ui
(A61)
j uTi uj
(A62)
116
Furthermore, we know that uTj ui = uj ui and, thus, is a scalar. Consequently, uTj Aui is also a
scalar and we have
h
iT
uTj Aui = uTi AT uj
(A63)
Now because A is symmetric, Eq. (A63) implies
h
iT
uTj Aui = uTi AT uj = uTi Auj
(A64)
Substituting the result of Eq. (A64) into Eq. (A59), placing it alongside Eq. (A60), and using
the results of Eqs. (A61) and (A62) gives
uTi Auj
uTi Auj
=
=
i uTj ui
(A65)
j uTj uj
(A66)
(A67)
(A68)
Using the definition of orthogonality of vectors with respect to the standard inner product,
it is seen that Equation (A68) implies that eigenvectors ui and uj corresponding to distinct
eigenvalues i and j of a real symmetric matrix are orthogonal.
Next, assume that two of the eigenvalues of a real symmetric matrix A are a complex coni . Now we know from
jugate pair, i.e., we consider two eigenvalues i and j such that j =
earlier in this section that the eigenvector uj associated with the complex conjugate of eigeni is the complex conjugate of ui (i.e., uj = u
i ). Furthermore, from the definition of
value j =
an eigenvalue-eigenvector pair we have
Aui
A
ui
i ui
i u
i
(A69)
(A70)
Ti i ui = i u
Ti ui
u
T
i uT u
i =
ui i u
i i
(A71)
(A72)
i = uTi A
uTi AT u
ui
(A73)
i
uTi u
(A74)
Ti Aui and u
Ti ui are scalars, we can substitute the results of Eqs. (A73)
Furthermore, because u
and (A74) into Eqs. (A75) from (A76), respectively, we obtain
ui
uTi A
uTi A
ui
=
=
i
i uTi u
T
i
i u u
i
(A75)
(A76)
i 0, we have
Now because uTi u
i i )uT u
(
i i = 0
(A77)
i i = 0
(A78)
117
(A79)
i = + i
(A80)
Equation (A79) states that a complex eigenvalue of a real symmetric matrix is equal to its
complex conjugate. The only possible way for a complex number to equal its complex conjugate
is if the number is itself real. Another way of looking at this is as follows. Suppose that
(A81)
Consequently, the only way for i and i to be equal is if 0, i.e., i R. The key result
is that the eigenvalues of a real symmetric matrix are real.
Now suppose we let U be the eigenvector matrix associated with a real symmetric matrix.
Then we have
h
i
U = u1 u2 un
(A82)
Now because the eigenvectors of a real symmetric matrix are orthogonal, we can normalize
each eigenvector to unit magnitude, i.e., we can say that
kui k = 1
(A83)
Then the eigenvector matrix U is such that its columns are orthonormal which implies that U is
an orthogonal matrix, i.e.,
U1 = UT
(A84)
Then because the eigenvectors are complete (by virtue of the fact that they are orthonormal),
we know that
AU = U
(A85)
where is a diagonal matrix with the eigenvalues on the diagonal. Finally, using the fact that
U is orthogonal, we know that
U1 AU = UT AU
(A86)
Consequently,
UT AU =
A.7
(A87)
(A88)
= Bu Au = (B A)u
(A89)
where the matrices A and B are both symmetric. Eq. (A88) is called a symmetric weighted
eigenvalue problem or, simply, a weighted eigenvalue problem because the matrix B is not the
identity matrix. Rearranging Eq. (A88), we obtain
i Bui
(A90)
j Buj
(A91)
118
=
=
i uTj Bui
(A92)
j uTi Buj
(A93)
Then, using the fact that A and B are symmetric, we know that
iT
uTj Aui
h
iT
uTi Buj
h
Consequently,
(A94)
(A95)
uTi Auj
uTi Auj
i uTi Buj
(A96)
j uTi Buj
(A97)
(i j )uTi Buj = 0
(A98)
uTi Buj = 0
(A99)
Bu = Au
(A100)
uTi Auj = 0
(A101)
Equation (A99) implies that the eigenvectors ui and uj are orthogonal with respect to the matrix
B. Furthermore, by rewriting Eq. (A88) in the form
Equation (A101) implies that the eigenvectors ui and uj are orthogonal with respect to the
matrix A. In other words, for the weighted eigenvalue problem of Eq. (A88), the eigenvectors
of distinct eigenvalues are orthogonal with respect to both the matrices A and B.
i Bui
(A102)
j Buj
(A103)
(A104)
i B
A
ui =
ui
(A105)
A
ui = j B
ui
(A106)
119
Now suppose that i and j are eigenvalues of the weighted eigenvalue problem of Eq. (A88).
i . Then, from the definition of the weighted eigenvalue problem of
Furthermore, let j =
Eq. (A88), we have
Aui
A
ui
i Bui
i B
ui
(A107)
(A108)
Ti i Bui = i u
Ti Bui
u
i B
i uT B
uTi
ui =
i ui
uTi A
ui
(A109)
(A110)
(A111)
(A112)
Ti Aui and u
Ti Bui are scalars, we can substitute the results of Eqs. (A111) and
Then, because u
(A112) into Eqs. (A109) and (A110), respectively, to obtain
uTi A
ui
uTi A
ui
i uTi B
ui
i uT B
ui
(A113)
(A114)
(A115)
i i = 0
(A116)
i = i
(A117)
Equation (A117) states that a complex eigenvalue of the weighted eigenvalue problem of
Eq. (A88) is equal to its complex conjugate. The only possible way for a complex number
to equal its complex conjugate is if the number is itself real. Another way of looking at this is
as follows. Suppose that
i = + i
(A118)
It then follows that
i = i
(A119)
i and i to be equal is if 0, i.e., i R. The key result
Consequently, the only way for
is that the eigenvalues of the weighted eigenvalue problem of Eq. (A88) are real.
Suppose now that we let
h
i
U = u1 u2 un
(A120)
Then we can write
A
u1
u2
un
1 Bu1
2 Bu2
n Bun
Using the expression for U from Eq. (A120), it is seen that (A121) can be
1
0
0
0
0
2
h
i
..
..
..
.
.
0
0
n
(A121)
rewritten as
(A122)
120
Then, defining
AU =
1
0
..
.
..
.
+0
Bu1
0
2
Bu2
0
0
..
.
..
.
n
Bun
..
.
..
.
0
(A123)
(A124)
(A125)
(A126)
(A127)
Then, pre-multiplying by
wTi ,
Awi = i Bwi
(A129)
(A130)
we have
uTi Aui
wi
ui = q
,
wTi Awi
(i = 1, . . . , n)
T
wi
wi
= wi Awi = 1,
q
q
=
A
wTi Awi
wTi Awi
wTi Awi
(A131)
(i = 1, . . . , n)
(A132)
121
(i = 1, . . . , n)
(A133)
(A134)
Next, suppose we choose to normalize the eigenvectors respect to the matrix B. Then in this
case we would choose the normalized eigenvectors such that
wi
,
ui = q
wTi Bwi
T
(i = 1, . . . , n)
T
Bq
= wi Bwi = 1,
uTi Bui = q
wTi Bwi
wTi Bwi
wTi Bwi
wi
wi
(A135)
(i = 1, . . . , n)
(A136)
we see that
wi
,
ui = q
wTi Bwi
(i = 1, . . . , n)
A.8
(A137)
(A138)
Definiteness of Matrices
Let A Rnn be an n n real-valued matrix. Then A is said to be positive definite if and only if
uT Au > 0 for all u 0, i.e.,
A positive definite uT Au > 0 u 0
(A139)
Similarly, the matrix A is said to be positive semi-definite if and only if uT Au 0 for all u 0,
i.e.,
A positive semi-definite uT Au 0 u 0
(A140)
Finally, the matrix A is said to be negative semi-definite if and only if uT Au 0 for all u 0,
i.e.,
A negative semi-definite uT Au 0 u 0
(A141)
Any matrix that can neither be classified as positive definite, positive-semi-definite, negative
definite, nor negative semi-definite, is called indefinite.
122
Bibliography
Rao, A. V., Dynamics of Particles and Rigid Bodies: A Systematic Approach, Cambridge University
Press, New York.
Kreyszig, E. (1988), Advanced Engineering Mathematics, John Wiley & Sons, New York.