Вы находитесь на странице: 1из 7

Additional Theory for

Module 4: Dynamic Analysis


Eigensolution Example

Consider the matrix


1 1 0
A 1 2 1
0 1 1
This matrix is in fact the stiffness matrix for a two-element bar, with (EA/L) set equal to
1. Let us now find the eigenvalues and eigenvectors of A.

The eigenproblem is defined by the equation


Ax x
where A is as given above, x is a vector (in this case, 3-by-1) and is a scalar value.
Thus, we can rewrite the equation as
A I x 0
where I is an identity matrix, in this case the (3x3) identity.

One solution to the equation above, of course, is x = 0. This solution is trivial, though,
and we seek solutions where x 0 . Within linear algebra, there is a provision for such
non-trivial solutions. It turns out that x need not equal zero, if the determinant of the
coefficient matrix [A I] is zero. Thus, we first find values of which make the
determinant zero, then we use these values in the eigenequation to find the non-zero x
vectors. Note that these non-trivial solutions can often have physical meaning, as further
mentioned below.

So, for the A matrix above:

(1 ) 1 0
A I 1 (2 ) 1

0 1 (1 )

Set the determinant of the combined matrix above to zero to get:

det(A I ) (1 ) * [(2 )(1 ) 1] 1[1(1 ) 0] 0

The determinant can be expanded, simplified and factored to yield the three solutions (the
eigenvalues of A)
1 0
2 1
3 3
Each of these eigenvalues is now placed into the equation [A-I]x = 0, and the resulting
set is solved for x. For the example of 3, only, we get
[ A 3 ] x 3 0
(1 3) 1 0 x1 0
1 (2 3)
1 x 2 0
0 1 (1 3) x3 3 0
or
2 x1 x 2 0
x1 x 2 x3 0
x 2 2 x3 0
which has solutions
x2 = -2x1
x3 = x1
1

and x1 is any arbitrary value. Thus, the solution x1 2 is the eigenvector of A,
1

corresponding to the eigenvalue = 3.

In the case of vibrations, the eigenvalues are related to the natural frequencies of the
system, and the eigenvectors are related to the motions of the system, when the system
vibrates at the natural frequencies. In the case of buckling analysis, the eigenvalues
represent limit loads for bifurcation buckling, and the eigenvectors again represent the
deflected shapes of the structure.

Mode shapes in vibrational analysis

The eigenvalue problem derived in the course notes, [ K M 2 ]x 0 , is not set up for
efficient computational solution. In addition, if the eigenproblem is solved AS IS,
important and useful information will be missed. As a result, it is more common to solve
a related auxiliary problem, the mass-normalized problem.

To see how this works, first expand the equation above, and recognize that 2x is the
second time derivative of x. Thus we get back to our original equation
Mx Kx 0

1. We perform Cholesky Decomposition on the mass matrix M, wherein the matrix is


decomposed into a Lower Triangular matrix L and its transpose:
M LLT

2. We define the auxiliary variable q so that q LT x and thus x LT q . This leads to


LLT x Kx 0 Lq K ( LT q)

3. We next premultiply the entire equation by L-1:


~
L1 Lq L1 KLT q 0 q Kq
where
~
K L1 KLT

4. Finally, we assume q to be harmonic, just as we previously assumed x to be harmonic


q ue jt q 2 ue jt 2 q
hence,
[ K I 2 ]q 0

The final form of the eigenproblem above is now computationally easier to solve for the
squared natural frequencies, and also has the desirable property that the q vectors are
mutually orthogonal and form a spanning set, so that the q vectors can be used to describe
the motion of the system, even if the system is subsequently damped and/or forced.

The q vectors are known as the MODE SHAPES of the system, and can be used in both
harmonic response and transient response analyses. In particular, if the q vectors are
normalized to unit length (that is, each divided by its own magnitude to create an
equivalently-directed vector of unit length) the set of normalized vectors are orthonormal,
and form a spanning set for solutions to forced problems. This is the basis of the MODE
SUPERPOSITION methods of dynamic analysis. In addition, these orthonormal vectors
can be used to further decouple the dynamic equation, making it easy to incorporate
damping as either modal damping or as proportional damping.

Harmonic Analysis
Suppose we have a body known to be subject to harmonic (sinusoidal) loading. The load
may be at any frequency, so let us call this frequency . The finite element equation set
will appear as

Mx Kx F0 sin t
In harmonic analysis, we assume the response, x(t) to be harmonic as well. (Harmonic
analysis only calculates steady-state response of the structure.) In that case, we have
x x0 sin t
x 2 x0 sin t

K M x
2
0 F0


x0 K 2 M F0 1

with all of this assuming sufficient boundary conditions have been applied to allow for
the inversion. Note that the solution is relatively simple, so long as K, M do not need to
be recalculated. That is, harmonic analysis is for linear analyses, only.

Consistent Mass Matrix Formulation


Recall that the interpolation used in deriving the element stiffness matrix may be written
as
u Nd
where
u is the generalized displacement of the element
N is a set of interpolation functions, chosen by the element designer
d is the set of nodal displacements of the element

1 2
The kinetic energy of a particle is found as T mv , which for a distributed mass must
2
be modified to
1 2 1
T
2 v dm (v T v)dV
2
where
is density,
V is the volume of the particle,
and the substitution of vTv for v2 is done for purposes of dimensional consistency in the
(upcoming) linear algebra.

Using the interpolation functions above, and noting that velocity v is the first derivative
(with respect to time) of displacement u, and that the interpolation functions N are
functions of position only (not functions of time,) we can see
1
T d N T NdV d
2 element

The integral in the last equation above represents the mass of the element, consistent with
the interpolated displacements. It is, thus, the CONSISTENT MASS MATRIX.
For completeness, note that the d term is nodal displacement, and if d is assumed
harmonic, then d d 0 e jt d jd d d 2 (d d ) , which is consistent with all
of the other derivations thus far: the mass matrix will end up multiplied by 2. Also
d
note that the equation of motion will come from the derivative equation (U T ) 0
dt
for a conservative mechanical system (U is potential energy, not addressed here.) Hence
the factor of and one of the d terms will vanish in the final equation of motion, and
we end up with
[ 2 M K ]d 0 ,
exactly as in the previous derivations. Note here that the form of the matrix equation just
completed is an eigenvalue problem defined in the first section of this document: if 2 are
chosen properly, the determinant of the combined matrix will be zero, allowing d to be
non-zero.

Explicit vs. Implicit Analysis


In IMPLICIT dynamic analysis (used in Abaqus/Standard, and thus for modal and
harmonic analyses,) the finite element equations are solved at every time step, as if the
previous time steps were not important. That is, there is no true time variation taken
into consideration. Perhaps one could think of this as a situation where you might make a
plot by, instead of knowing the function y(t), you find values of y at specified values of t,
and simply connect the dots on a chart. The dots will trace a curve that indicates time
variation, but the dots themselves were each calculated using only the time value that is
current time for each dot.

In EXPLICIT dynamic analysis, time variation is taken into consideration, and the
equations are solved by numeric integration from one time step to the next. There is
more room for error when an explicit analysis is used, but the analysis is typically much
faster, because finite element equations need not be solved at each iteration.

Here is, roughly, how the explicit procedure works:


d2y
Take a simple differential equation, say, y sin x 0 . Rename the variable y to y1
dx 2
and its derivative with respect to x as y2. Then:
y1 y
dy1
y2
dx
dy 2 d 2 y
2 y1 sin x
dx dx

dy1
dx 0 1 y1 0
dy y sin x
2
1 0 2
dx
The result is just a simple restatement of the initial differential equation as a set of
coupled first-order DEs. Now, if we know the initial velocity (y2) and initial position
(y1), we can calculate their time derivatives at time 0, and numerically integrate over a
small time frame to get their values at the next time step. These then feed back into the
equation and are used to calculate new derivatives, so the process continues.

In a finite element problem, the mass and stiffness matrices (M and K) appear as
coefficients of the second derivative term and the zeroth order term, respectively. The
coefficient matrix above (containing ones and zeros, only) would have M and K in it, but
otherwise the general process is the same as outlined above, when the idea is extended to
FEA.

For linear analyses, M and K are not allowed to change as the analysis proceeds, so there
is never any need to solve FE equations to recalculate M or K. In nonlinear analysis, we
DO need to recalculate M and K periodically, which is one reason nonlinear analysis is
slow. The time savings of the explicit method becomes critical, then. Thus, Abaqus only
uses the implicit method for linear analyses.

Damping
Motions of a structure are, of course, non-perpetual. All motion must eventually cease.
In the mathematics of finite elements, we usually incorporate this effect by use of a
damping matrix.

We have been using Mx Kx F (t ) . Damping is incorporated as a term multiplying


velocity. Thus, we get Mx Cx Kx F (t ) . C is the damping matrix.

When the damping matrix C is not empty then the modal problem becomes one where the
eigenvalues are complex. This is not in itself a problem, and the complex eigenvalues
have true engineering implications. The real part of the eigenvalue represents
exponential decay of the solution, and the imaginary part represents oscillatory motion.
However, it is difficult to solve the equations to find these complex eigenvalues if C is
allowed to take any given value. Thus, experience has shown that there are some ways to
introduce damping that are better than others. The most common ways are to either input
modal damping ratios or to use proportional damping.

Modal damping ratios insert a constant damping ratio into each equation of modal
response, after decoupling the system of equations. The mathematics is covered in any
standard text on vibrations. The damping ratio is the ratio of true damping coefficient to
the theoretical critical damping ratio. A ratio of less than one is underdamped, so that
the mechanical response of the system is a decaying exponential. A ratio of unity or
greater indicates there will be no oscillatory response; a ratio of exactly one will cause
the quickest possible return to the equilibrium state.

When one uses modal damping ratios, typical values are on the order of 0.01 0.1,
though these are based on experience and it may be appropriate to use values far from the
values I have quoted.

Proportional damping utilizes weighting factors which multiply M and K to get an


approximate C. That is, we choose and and calculate C M K . Typically, a
and b are chosen based upon the first two resonant frequencies. Again, I refer you to a
text on vibrations.

Abaqus supports modal damping ratios, proportional damping, material damping


(material property definitions are used, which in effect is similar to proportional
damping) and composite damping (for composite materials, only.)

Вам также может понравиться