Professional Documents
Culture Documents
MDOF
Dynamical
Systems
191
Lecture 19: MDOF DYNAMICAL SYSTEMS
TABLE OF CONTENTS
Page
19.1. Introduction 193
19.2. A Two-DOF Lumped-Parameter System 193
19.2.1. Equations of Motion . . . . . . . . . . . . . . . 193
19.2.2. Matrix Form of EOM . . . . . . . . . . . . . . 194
19.2.3. Undamped Natural Frequencies and Modes . . . . . . . 195
19.3. Generalization to Arbitrary Number of DOF 196
19.3.1. Linkage to Linear Algebra Eigenproblem . . . . . . . 196
19.3.2. Generalized Mass and Stiffness . . . . . . . . . . . 197
19.3.3. Eigenvector Normalization Criteria . . . . . . . . . . 198
192
19.2 A TWO-DOF LUMPED-PARAMETER SYSTEM
19.1. Introduction
We move now to dynamic structural systems with Multiple Degrees of Freedom (abbreviation: MDOF).
For simplicity we focus on linear, viscous-damped, lumped parameter models. All explicit examples
have only two DOF, but that is sufficient to show how such models are treated using matrix algebra.
Once the matrix equations are exhibited in compact form, the extension to an arbitrary, but finite,
number of DOF is readily done.
;; (a) (b)
k1 c1
Static equilibrium
position .
Fs1 = k1 u1 Fd1 = c1 u 1
u 1 = u1(t)
Mass m1
p1 (t) p1 (t) ..
FI1 = m 1 u1
Fs2 Fd 2
Static equilibrium k2 c2
position
Fs2 = k2 (u2u1) . .
Fd 2 = c2 (u 2 u1)
u 2 = u2(t)
Mass m2
p2 (t) ..
p2 (t) FI2 = m 2 u 2
x
Figure 19.1. Two-DOF, lumped parameter, spring-mass-dashpot dynamic system.
Consider the lumped-parameter, mass-spring-dashpot dynamic system shown in Figure 19.1(a). It has
two point masses m 1 and m 2 , which are connected by a spring-dashpot pair with constants k2 and
c2 , respectively. Mass m 1 is linked to ground by another spring-dashpot pair with constants k1 and
c1 , respectively. The system has two degrees of freedom (DOF). These are the displacements u 1 (t)
and u 2 (t) of the two masses from their static equilibrium positions. These displacements are assumed
small. Known dynamic forces p1 (t) and p2 (t) act on the masses. Our goal is to formulate the equations
of motion (EOM), and to express them in matrix form.
193
Lecture 19: MDOF DYNAMICAL SYSTEMS
Next express the spring, dashpot and inertial forces in terms of displacement, velocities and accelera-
tions. Note that for the spring and dashpot forces we must use relative displacements and velocities,
respectively, whereas for inertia forces total accelerations must be used:
m 1 u 1 k1 u 1 c1 u 1 + k2 (u 2 u 1 ) + c2 (u 2 u 1 ) + p1 = 0,
(19.3)
m 2 u 2 k2 (u 2 u 1 ) c2 (u 2 u 1 ) + p2 = 0.
Finally, collect all terms depending on the DOF u 1 and u 2 and their temporal derivatives in the LHS
while moving everything else to the RHS. Sign convention used in mechanics: mass-times-acceleration
terms in LHS should be positive. Following that convention we get
m 1 u 1 + c1 u 1 + c2 u 1 c2 u 1 + k1 u 1 + k2 u 1 k2 u 2 = p1 ,
(19.4)
m 2 u 2 c2 u 1 + c2 u 2 k2 u 1 + k2 u 2 = p2 .
194
19.2 A TWO-DOF LUMPED-PARAMETER SYSTEM
def
Here D() = K 2 M is called the dynamic matrix.
Equation (19.12) states the free vibrations eigenproblem for an undamped MDOF system. For nontrivial
solutions U = 0 the determinant of the dynamic matrix must vanish, whence
C(2 ) = det D() = det K 2 M = 0. (19.13)
This is the characteristic equation for free undamped vibrations. For a two-DOF system, C(2 ) is a
quadratic polynomial in 2 , which will yield two roots: 12 and 22 .
The next section shows that under certain conditions on K and M, which are satisfied here, those roots
are real and nonnegative. Consequently the positive square roots
1 = + 12 , 2 = + 22 , (19.14)
195
Lecture 19: MDOF DYNAMICAL SYSTEMS
are also real and nonnegative. These are called the undamped natural circular frequencies. (Qualifiers
undamped and circular are often omitted unless a clear distinction needs to be made.) As usual
they are measured in radians per second, or rad/sec. They can be converted to frequencies in cycles
per second, or Hertz (Hz), by scaling through 1/(2 ), and to natural periods in seconds by taking their
reciprocals:
i 1 2
fi = , Ti = = , i = 1, 2. (19.15)
2 fi i
How about the U vector in (19.12)? Insert 2 = 12 there and solve for U = 0. (This operation
is possible because K 12 M is singular by definition.) Call that solution vector U1 . Repeat with
2 = 22 and call that vector U2 .
Vectors U1 and U2 are mathematically known as eigenvectors. Since U1 and U2 may be scaled by
arbitrary nonzero factors, we must normalize them to make them unique. Normalization criteria often
used in dynamic analysis are discussed in 19.3.3.
For this particular problem the eigenvectors are called undamped free-vibrations natural modes, a
mouthful often abreviated to just modes, on account of the physical interpretation discussed later. The
visualization of an eigenvector as a motion pattern is called a mode shape.
Thanks to matrix notation, the extension to an undamped, unforced MDOF system with n degrees of
freedom is immediate.
In this general case, matrices K and M are both n n. The characteristic polynomial is of order n in
2 . This polynomial provides n roots labeled i2 , i = 1, 2 . . . n, arranged in ascending order. Their
square roots are the natural frequencies of the system.
The associated vibration modes are the eigenvectors of the vibration eigenproblem. They are called
Ui before normalization. Normalization schemes are studied in 19.3.3.
where i is a natural frequency index that ranges from 1 through n. Second, rename the symbols that
appear in (19.16) as follows: K A, M B, i2 > i , Ui xi . This allows us to rewrite (19.16)
as
A xi = i B xi , i = 1, 2, . . . n. (19.17)
This is the canonical form of the generalized algebraic eigenproblem that appears in Linear Algebra
textbooks. If both A and B are real symmetric, (19.17) is called the generalized symmetric algebraic
eigenproblem. If, in addition, B is positive definite (PD), it is shown in those textbooks that
(i) All eigenvalues i are real
(ii) All eigenvectors xi have real entries
Property (i) can be further strengthened if A enjoys additional properties:
196
19.3 GENERALIZATION TO ARBITRARY NUMBER OF DOF
(iii) If A is nonnegative definite (NND) and B is PD, all eigenvalues are nonnegative real
(iv) If both A and B are PD, all eigenvalues are positive real
The nonnegativity property is obviously important in vibration analysis since natural frequencies are
obtained by taking square roots of the eigenvalues, which appear as squared frequencies in (19.16).
If a squared frequency is negative, its square roots are imaginary. The possibility of negative squared
frequencies arise when studying dynamic stability and active control systems, but those topics are
beyond the scope of this course.
If matrix B in (19.17) is the identity I, (19.17) reduces to the standard algebraic eigenproblem
A xi = i xi , i = 1, 2, . . . n. (19.18)
If A is real symmetric, (19.18) is called the standard symmetric algebraic eigenproblem, or symmetric
eigenproblem for short. Since the identity matrix is obviously symmetric and positive definite, the
aforementioned properties (i) through (iv) hold for (19.18).
Remark 19.1. Some higher level programming languages such as Matlab, Mathematica, and Maple can solve
eigenproblems directly through their built-in libraries. Of these 3 commercial products, Matlab has the most
extensive capabilities, being able to process the generalized eigenproblem form (19.17) directly. On the other
hand Mathematica only accepts the standard eigenproblem (19.18) as built-in function. Similar constraints may
apply to hand calculators with matrix processing abilities. To overcome those limitations, there are transformation
methods that convert (19.17) into (19.18). Only a couple of such schemes will be mentioned:
(I) If B is nonsingular, premultiply both sides of (19.17) by its inverse: B1 to get B1 A xi = B1 i B xi = i xi .
This has the standard form A xi = i xi , in which A = B1 A. The eigenvalues and eigenvectors of A are
those of (19.17).
(II) If A is nonsingular, premultiply both sides of (19.17) by its inverse: A1 to get A1 A xi = A1 i B xi =
i xi . This has the standard form B xi = (1/i ) xi , in which B
= A1 B. The eigenvectors of B
are those of
(19.17) but its eigenvalues are reciprocated.
Note that matrices A or B produced by these schemes will not generally be symmetric, even if A and B are.
However the eigenvalue-eigenvector properties listed previously will be preserved. There are fancier reduction
methods that preserve symmetry (for example, those based on the Cholesky factorization of either A or B), but
those noted above will be sufficient for this course.
Remark 19.2. For small number of DOF, say 4 or less, it is often expedient to solve for the eigenvalues directly
by finding the roots of the characteristic polynomial introduced in (19.13):
in which = 2 and D = K 2 M = K M is the dynamic stiffness matrix. This procedure will be followed
in subsequent Lectures for 2-DOF examples. It is particularly useful with hand calculators that can compute
polynomial roots but cannot handle eigenproblems directly. Eigenvectors my be then obtained as illustrated in the
next Lecture.
K Ui = i2 M Ui , (19.20)
197
Lecture 19: MDOF DYNAMICAL SYSTEMS
(I) The mass matrix M is positive definite (PD) whereas the stiffness matrix K is nonegative definite
(NND). As a consequence all eigenvalues i = i2 are nonnegative real, and so are their positive
square roots: i 0.
(II) All frequencies are distinct: i = j if i = j.
Under assumptions (I) and (II), the theory behind the algebraic eigenproblem says that for each i
there is one and only one eigenvector Ui , which is unique except for an arbitrary nonzero scale factor.
An eigenvector Ui scaled as per some normalization condition (for example, unit Euclidean length) is
called a normal mode. We will write
Ui = ci i , i = 1, 2, . . . n. (19.21)
Here i is the notation for the i th normal mode. This is obtained from Ui on dividing through by a
scale factor ci , chosen according to some normalization criteria.
Normal modes enjoy the following orthogonality properties with respect to the mass and stiffness
matrices
iT M j = 0, iT K j = 0, i = j. (19.22)
If i = j, the foregoing quadratic forms provide two important quantities called the generalized mass
Mi and the generalized stiffness K i :
Mi = iT M i > 0, K i = iT K i 0, i = 1, 2, . . . n, (19.23)
These are also called the modal mass and the modal stiffness, respectively, in the structural dynam-
ics literature. The squared natural frequencies appear as the ratio of generalized stiffnesses to the
corresponding generalized masses:
Ki
i2 = , i = 1, 2, . . . n. (19.24)
Mi
This formula represents the generalization of the SDOF formula n2 = k/m to MDOF.
The definition of generalized mass Mi in (19.22) provides one criterion for eigenvector normalization,
which is of particular importance in modal response analysis. If the scale factors in (19.21) are chosen
so that
Mi = iT M i = 1, i = 1, 2, . . . n, (19.25)
the normal modes i are said to be mass-orthonormal. If this is done, (19.24) reduces to
i2 = K i = i K i , i = 1, 2, . . . n. (19.26)
Remark 19.3. If there are coincident frequencies (i.e., multiple eigenvalues appear), assumption (II) no longer
holds. Eigenvector extraction becomes more complicated. For example if 12 = 22 , any linear combination of U1
and U2 is also an eigenvector. Instead of an a single vector we have to deal with a subspace spanned by two or
more vectors. This case is treated in detail in the Linear Algebra literature, and will not be reviewed further here.
Suffice to say that, if assumption (I) holds, it is always possible* to construct a distinct set of eigenvectors that
satisfy the orthogonality properties (19.22) as well as the definitions (19.23) of generalized mass and stiffness.
198
19.3 GENERALIZATION TO ARBITRARY NUMBER OF DOF
Example 19.2. Consider the two-DOF system developed in 19.2.1. As numerical values we take
m 1 = 2, m 2 = 1, c1 = c2 = 0, k1 = 6, k2 = 3, p1 = p2 = 0. (19.27)
3
12 = = 1.5, 22 = 6. (19.31)
2
The undamped circular natural frequencies (to 4 places) are 1 = 3/2 = 1.225 and 2 = 6 = 2.449. If the
physical system of units is consistent,these are expressed in radians per second.
To convert to cycles per second,
or Hertz, divide by 2 : f 1 = (2)1 3/2 = 0.1949 Hz, and f 2 = (2 )1 6 = 0.3898 Hz.
The calculation of eigenvectors is carried out in the next Lecture.
Note that no units are specified. It is tacitly understood that a consistent set of physical units, SI or English, is used througout.
199