13th World Conference on Earthquake Engineering

Vancouver, B.C., Canada

August 1-6, 2004
Paper No. 3086

A QUICK METHOD FOR EIGENVALUE ESTIMATION OF MULTISTORY

BUILDINGS

Mohsen Shahrouzi1

SUMMARY

A simplified method is presented to estimate desired modal frequency of a chain structure

independently from any other modes’ calculations. It is based on the developed mapping
between the chain structure and an equivalent beam model thus leading to a dimensionless
frequency equation. For the considered case of fixed-free boundary condition an estimation
technique is used to find the desired root of the dimensionless frequency equation. The missed
mass effect during the proposed mapping is taken into account to improve estimated results.
Finally the effectiveness and accuracy of the presented method is discussed via illustrative
examples of regular and irregular chain structures even with the non-uniform distribution of
mass and stiffness among their height.

Key words: Regular chain structure, Frequency equation, Missed mass effect

INTRODUCTION

Estimation of dynamic structural characteristics is of interest in many applications such as time-

consuming iterations of non-linear analyses or in the case of optimization problems where
structural model is highly changed during the process. Most of the recent methods require
exhaustive computational efforts (Clough [1], Hosseini [2]) among them Holzer method [3] may

1
Ph.D. candidate, International Institute of Earthquake Engineering and Seismology,
Arghavan 26,Koohestan 8,Pasdaran, Tehran, I.R. Iran, Tel: 2831116-9 (Ext. 143) , P.O. Box. : 19395/3913
email: [email protected]
be concerned. It estimates desired modal frequency independently from other modes, however
dependently to the first guess of the corresponding mode-shape. A modified version of that
based on energy considerations, has already been developed known as Holzer-Hosseini method
[2].
This paper aims to develop a quick simplified method to achieve an estimation of a desired
mode’s frequency for a fixed-ended chain structure independently from other modes’
calculations. It is also expected to find out the corresponding mode-shape simultaneously. Such
an approximate result may be then used as a proper starting point for some other known exact-
iterative methods aiming to achieve more efficiency and accuracy in the final results.

PROBLEM DEFINITION AND THEORY

A chain structure is defined as a system including elements of mass and stiffness which are
alternatively arranged in a sequence along a main axis. Thus, many practical structures fall in
this category such as pipelines, railways, tall process towers and so on. In this work, multistory
building frames are approximately treated as chain structures with fixed-free boundary
conditions. Such an analogy will be more accurate after condensation of rotational degrees of
freedom within each story, leading to a lumped mass model (Fig.1). Deriving chain structure’
properties may be done regarding approximate methods such as Hosseini [4].

Figure 1. Chain structure model of a building frame and equivalent uniform beam

Consider a beam with uniform mass of m per length and moment inertia of I over the whole
length L [1]. Its free-vibration equation regarding flexural behavior is given by:
∂ 4 v( x, t ) mL ∂ 4 v( x, t )
(1) + =0
∂x 4 EI ∂t
4
Applying common assumption in variables’ separation lemma, i.e. v( x ,t ) = ϕ ( x ).Y ( t ) leads to:
(2) ϕ iv ( x) − a 4ϕ ( x ) = 0
..
(3) Y (t ) + ω 2Y (t ) = 0
Whereas:
(4) ϕ ( x) = A1 sin(ax) + A2 cos(ax) + A3 sinh(ax) + A4 cosh(ax)
EI 2 EI
(5) ω = a 2 = (aL) 2
mL L mL
It is desired to find vibration frequency (or dimensionless frequency: q=aL). The fixed-free
boundary conditions are;
(6) ϕ (0) = 0
(7) ϕ ′( L) = 0
(8) EIϕ ′′( L) = 0
(9) EIϕ ′′′( L) = 0
In order to have non-zero solution for the above set of homogenous equations, its determinant
should be assigned to zero. It consequently leads to the following dimensionless frequency
equation for such a uniform property beam with Fixed-Free boundary conditions:
(10) 1 + Cos(q)Cosh(q ) = 0 , q = aL
Omitting the remained dependent coefficient between (Eq.8) and (Eq.9), the mode-shape
function is found in terms of one unknown coefficient, i.e. amplification factor A1:
sin(aL) + sinh(aL)
(11) ϕ ( x) = A1{sin(ax) − sinh(ax) + [cosh(ax) − cos(ax)]}
cos(aL) + cosh(aL)
As (Eq.10) and (Eq.11) are dimensionless, the effect of geometrical and mechanical properties
of the corresponding beam will be considered by (Eq.5).

In order to develop the relation between parameters of chain structure and corresponding
flexural beam a reverse mapping scheme is used.
Consider a flexural beam with fixed-free boundary conditions having parameters of uniform
mass m , moment of inertia EI and length L , be discretized to a lumped mass model in n equal-
height story levels of the target chain structure (Fig.2). During such a mapping, mass of each
discrete element is lumped to its center location, i.e. corresponding story level in the chain
structure. Thus the length of each element; so-called height portion of the corresponding story,
will be the same for all elements except the top and bottom ones. As a result, the mass of ground
and top level will be half the value of all other similar lumped masses. This fact is considered the
basis of definition of regular chain structure.
n
m
(12) M = mL = ∑ mi = (n − 1)m1 + 2 1 , L = H
i=0 2
L m m
(13) mi = m1 = m = mLi i = 1,2,..., n − 1 ⇒ m = i = n
n Li Ln
2
Despite the parent beam, in the discretized model the mass lumped to ground elevation does not
experience any displacements either and so its participation in modal analysis is zero. This will
cause an error, so-called the missed mass effect.

Figure 2. Discretizing a uniform beam and mapping it to a lumped mass model

Regarding reverse mapping, the similarity of mass, stiffness or height of stories in chain
structure model may not always be the case in practice. So an equivalent unit mass and height
portion for every story is defined based on the above analogy:
n −1
M
(14) M = ∑ mi + mn , m 0 =
i =1 H − 1 2 h1
h h
(15) Li = i + i +1
2 2
mi
(16) mi = , i = 1,.., n
Li
As the value of ground level mass is not determined, the height portion of this level is subtracted
from the total height in (Eq.14) in order to consider the missed mass effect. Overall unit mass,
m 0 , is then calculated using this modified height.
Regarding mode-shapes of uniform beam vibration, it is investigated that the difference between
the displacements of the first height portion with respect to the others is reduced in higher
modes. As a result, the higher the mode number, the more the missed mass modal participation
will be. So the following approximate relation is used to consider the missed mass effect for the
jth mode:
 1 mh1 j
(17) m j = m.( 1 + 2
)
M

Effect of relative changes in story mass and stiffness

Similarly, the mean stiffness of story-i and average stiffness over the whole height are defined as:
EI i
(18) EI i =
Li
n
(19) EI 0 = 1
H ∑ EI
i =1
i

Based on recently developed mapping of the continuous uniform beam to the lumped mass model, regular
chain structure can be defined under the following conditions:
(a) Story mean (unit) mass mi (Eq.16) be equal to overall average mass m0 (Eq.14)
(b) Story mean (unit) stiffness EI i (Eq.18) be equal to overall average mass, EI 0 (Eq.19)
For such a regular case, the effective length of equivalent beam, Le , is taken as structure’s height, H.
Otherwise, it is altered so that the equivalent beam frequency is as close as possible to that of the chain
structure. The overall average mass and stiffness are uniformly distributed along the equivalent beam
length to take advantage of its simple formulation. So the effect of mass/stiffness irregularity is entered
through modified story height portion, Li′ . It is increased if the mass of i th story is greater than the total
mean mass and vice versa. For the case of stiffness, however, the relation is reversed as below:
xi
mi ( ) s1
(20) Li′ = Liα i , αi = ( ) H

m0
x
EI i ( Hi ) s2
(21) hi′ = hi β i , βi = ( )
EI 0
Here, the parameter s1 is taken as 0.5 and s2 as -0.5 due to reverse relation of mass to frequency with
respect to stiffness as in Eq.5. The term (xi /H) is employed due to the fact that the higher the story
(number) is, the more its mass participation in modal frequency will be. Consequently, the total length of
equivalent beam is given as sum of the modified height portions of story levels added to half modified
height of the first level:
n
(22) Le = ∑ Li′ + 1 2 h1′
i =1
QUICK SOLUTION OF THE FREQUENCY EQUATION

The dimensionless frequency equation of uniform beam with flexural behavior (Eq.10) is of hyperbolic
type and needs considerable numerical effort to be solved. However, a simplifying scheme is employed
here to find its approximate roots. Consider both sides of Eq.10 be multiplied by non-zero term e − q , so
its left side changes to :
1 + e −2 q
(23) g (q) = e −q f (q ) = e − q + cos(q)
2
As the dimensionless frequency, q, increases, the term e − q will rapidly tend to zero. So the equation will
be simplified to:
(24) g (q ) ≈ 1
2 cos(q) = 0
Its well-known roots are given by the following binomial relation:
(25) q j = ( j − 0.5) * π
While j stands for mode number and q = al is the dimensionless frequency. Fig.3 shows good
satisfactory between roots of cos(q) and f(q).
This approximation error is reduced with j increase, so that it will be of the order 10-5 after the three first
modes. Thus, it is considered acceptable for estimation purposes. Since the frequency equation (Eq.10) is
dimensionless, exact solution of the first three modes can be saved once for ever as:
(26) q1 = 1.875104 , q 2 = 4.694091 , q3 = 7.854757

Figure 3. Comparison between roots of frequency equations g(q)=0 and cos(q)=0

ESTIMATION ALGORITHM

Based on the presented theory, the jth -mode frequency and shape of a tall multistory building
can be directly estimated through the following steps:
1. Determine the jth dimensionless frequency using (Eq.25) or (Eq.26).
2. Generate the chain structure model and determine its parameters of mass, stiffness and
story height.
3. Through mapping to uniform equivalent beam determine:
3.1. Story height portion, Li , by (Eq.15)
3.2. Mean mass, mi , for every story (Eq.16) and also total average mass, m0 (Eq.14).
3.3. Mean stiffness, EI i , for every story (Eq.18) and total average stiffness, EI 0
(Eq.19).
3.4. If m i = m 0 and EI i = EI 0 then chain structure is regular so Le = H , else select the
value of s 2 = − s1 and compute the effective beam length, Le .
4. Recover the missed mass effect as in (Eq.17) to obtain modified total mean mass in
mode-j:
m0 H 1 mh
(27) m j = (1 + 2 1 ) j
Le M
5. Modify the total mean stiffness in mode-j for the effective length by:
L
(28) EI j = e EI 0
H
6. Estimate the mode-j frequency using (Eq.5).
 7. To obtain the desired mode-shape, ϕ j ( xi ) , substitute the mapped coordinates x i in
(Eq.11) from the following relation:
h′ i
(29) xi = 1 + ∑ Lm′ − 1 2 Li′
2 m=1

ILLUSTRATIVE EXAMPLES

Here, the suitability of the proposed method is examined through some samples of regular and
irregular chain structures. In each example, the results have been compared with those of
accelerated iterative subspace method as a benchmark. The ( KN − m ) unit system is employed
for all the cases.

Regular ten story building

A 10-story building with the following specifications is modeled:
mi = 600 for i = 1,..,9 while m10 = 300 .
E = 2 * 10 9 , I i = 0.2 , hi = 3 for i = 1,..,10 .
Li = 3 for i = 1,..,9 and L10 = 3 / 2 .
4 *10 8
∀i → m i = mi Li = 200 ⇒ m 0 = m i and EI 0 = EI i = EI i hi = .
3
Thus, such a model satisfies both conditions: (a) and (b), so it is considered a regular chain structure. Each
node in this model has a rotational as well as a transversal degree of freedom. Total mass and the missed
mass are computed as below:
10
M = ∑ mi = 9 * 600 + 300 = 5700 , mlost = m0 h / 2 = 200 * 3 / 2 = 300
i =1
The results of the current approximate method have been compared with the exact one through
computed relative errors for each mode, in table 1.

Table 1. Comparison of exact vs. approximate periods obtained for regular 10-story building
regarding missed mass recovery (Average relative error = 5.5%)
Mode No. 1 2 3 4 5 6 7 8 9 10
Exact period T (s) 1.3519 .2084 .0728 .0366 .0219 .0146 .0106 .0082 .0067 .0060
Estimated T (s) 1.1668 .1908 .0697 .0364 .0224 .0154 .0112 .0086 .0068 .0055
RelativeError % -13.7 -8.4 -4.3 -0.5 -2.3 5.5 5.7 4.9 1.5 -8.3

Considering the case of ignoring the missed mass effect as in table 2, it is found that for such a
case the maximum and mean relative error have been increased from 13.7% to 25% and from
5.5% to 12.9% correspondingly. This notifies the necessity of regarding missed mass recovery
through estimation.

Table 2. The effect of missed mass on the accuracy of estimated periods

(Average relative error = 12.9%)
Mode No. 1 2 3 4 5 6 7 8 9 10
Exact period T (s) 1.3519 .2084 .0728 .0366 .0219 .0146 .0106 .0082 .0067 .0060
Estimated T (s) 1.1373 .1815 .0645 .0331 .0200 .0134 .0096 .0072 .0056 .0045
RelativeError % -15.9 -12.9 -11.0 -9.6 -8.7 -8.2 -9.4 -12.2 -16.4 -25.0
In addition, a conformation can be found between the estimated and exact mode shapes. For
example, shape of mode-7 is illustrated in Fig.4.

exact approx.

0
0 1 2 3 4 5 6 7 8 9 10
-2

-4

-6

Figure 4. Example of estimated vs. exact 7th mode shape

Irregular 10-story building

This example shows the ability of the method in modeling highly irregular chain structures. To
this end, some characteristics of the previous regular model are changed as:
h1 = 5 ,
m2 = m3 = 1400 , m10 = 600 , m6 = 1000
I 6 = 0.08, I 7 = I 8 = I 9 = I 10 = 0.10
Decreasing stiffness or increasing mass in some levels model the soft story effect. The accuracy
of approximate results with respect to exact solution is given in Table 3.

Table 3. Comparison of exact vs. approximate periods obtained for irregular 10-story building
(Average relative error = 9.0%)
Mode No. 1 2 3 4 5 6 7 8 9 10
Exact period T (s) 1.3519 .2084 .0728 .0366 .0219 .0146 .0106 .0082 .0067 .0060
Estimated T (s) 1.1668 .1908 .0697 .0364 .0224 .0154 .0112 .0086 .0068 .0055
RelativeError % -13.7 -8.4 -4.3 -0.5 -2.3 5.5 5.7 4.9 1.5 -8.3

Regular 30-story building

The effect of more distributed models has been studied in this example of a 30-story regular
building model with the same characteristics of the 1st example. As a comparison parameter, the
mean relative error has decreased from 5.5% for 10-story model to 3.9% in this case (Table 4)
while the maximum error has reduced to 13%.

Table 4. Comparison of exact vs. approximate periods obtained for regular 30-story building
(Average relative error = 3.9%)
Mode No. 1 2 10 20 28 30
Exact period T (s) 10.904 1.731 .0418 .0100 0.0060 0.0058
Estimate T (s) 10.322 1.661 .0439 .0331 0.0061 0.0054
RelativeError % -5.3 -4.0 5.0 13.0 1.7 -6.9
 CONCLUDING REMARKS

The proposed method has developed mapping of the chain structure model of a multistory
building to the uniform flexural beam. The frequency equation of corresponding equivalent
beam is dimensionless, thus independent of building structural properties except the boundary
conditions and dynamic behavior. As an outstanding point, high computational effort in solving
such a hyperbolic equation is reduced to simply evaluating a binomial for each mode number.
The missed mass effect during the mapping has been declared and treated by a mass recovery
technique. Results obtained through examples of multi-story buildings notify the importance of
regarding such an effect. The proposed mapping introduces new definition of regular structure in
height, especially for the mass/stiffness of the last story with respect to the others.
The effect of mass and stiffness irregularity was treated by modifying equivalent beam
parameters, specially its effective length. The method was then applied to high irregular
examples in the case of mass, stiffness and story height (up to 200, 80 and 60 percent
correspondingly). According to treated examples, the estimated results were in acceptable range
of accuracy even when assessing such high irregular cases including soft story effect. The mean
relative error varies from 5.5% for 10-story and less than 4% in 30-story example which
introduces better mean accuracy for taller buildings. Reasonable satisfactory was also found
between exact and estimated mode shapes.
Another advantage of the proposed method is its capability to predict both desired mode shape
and eigenvalue simultaneously but independent from other modes’ calculations. The results can
be considered acceptable for some applications such as analysis quick-checks; however, they
can be further improved being used as initial point for exact iterative methods in order to
achieve even higher accuracy and efficiency. In addition, such a direct modal estimation may be
of interest for steps of highly model-changing iterative methods such as optimization or
nonlinear analyses.

REFERENCES

1. Clough R. W. And Penzien J., Dynamics of Structures ,Mc-Graw Hill, 1975.

2. Hosseini M., Stochastic response of nonlinear MDF structures under non-stationary seismic
loading , Ph.D. Assertion (in Persian), Islamic Azad University Press, 1992.
3. Thomson W., Theory of Vibration with Application, 2nd Ed. Printice Hall, 1981.
4. Hosseini M. and Imagh_e_Naini M. R. “A quick method for estimating the lateral stiffness of
building systems”, Structural Design of Tall Buildings 1999; 8: 247-260.