EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL.

16, 1177 1188 (1988)

STIFFNESS IDENTIFICATION OF A TALL BUILDING DURING

CONSTRUCTION PERIOD USING AMBIENT TESTS

MORTEZA A. M. TORKAMANI* AND AHMAD K. AHMADI'

Department of Civil Engineering, University of Pittsburgh, Pittsburgh, Pennsylvania 15261, U.S.A.

SUMMARY
Ambient response measurements were made on an eighteen-storey building at three different stages of construction to
detect any changes in the frequencies, mode shapes and stiffness with construction. The first nine frequencies and
corresponding mode shapes for each stage of construction are found. A comparison is made among these mode shapes
and frequencies and with the mode shapes and frequencies of an analytical model incorporating beams, columns, shear
walls, panels and diagonal elements. The added effects, o n frequencies and mode shapes, of non-structural elements such
as stairs, elevators, claddings and partition walls are studied. Using Improved Statistical Structural Identification, an
attempt is made to study the stiffening effect of non-structural elements by updating the stiffness matrix of the building.

INTRODUCTION
During the last two decades advancing computer technology has made modelling and analysis of high-rise
buildings an easy task. In spite of modern computational resources, neglecting non-structural elements, such
as stairwells, elevator shafts, partition walls and claddings, in the mathematical modelling of buildings is
common practice. Although the effect of a single non-structural element on the static or dynamic response of a
building may be negligible, the cumulative effect of several could be significant. Full-scale measurements of
buildings show that analytical models do not give mode shapes and frequencies which concur with test
results6 This is the direct consequence of improper modelling of some of the structural elements, neglecting
the non-structural elements and joint rotations. The results indicate that there is a need to improve the
mathematical modelling of buildings such that the gap between the modal responses of analytical and
experimental analyses is minimized.
In order to close the gap between analytical and experimental mode shapes and frequencies a building's
stiffness and mass matrices may be updated. Computation of the mass matrix of a building is generally based
on the mass of dead weight of the elements, components and systems, and is more accurate than the stiffness
matrix. Therefore, the stiffness matrix of the building should be revised. The Improved Statistical Structural
Identification Method, which is described in a companion paper,' is used in this research to revise the stiffness
matrix of a building.
This paper presents the result of a study to determine the effects of non-structural elements on the
frequencies, mode shapes and stiffness coefficients of a high-rise building. Ambient vibration measurements
were conducted on a building at three different stages of construction and after the construction was
completed. A set of mode shapes and frequencies was found from data processing of the ambient vibration
tests. A second set of mode shape and frequencies was found considering the analytical model. Then, the
Improved Statistical Structural Identification Method'. was used to revise the stiffness matrix of the
building. A third set of mode shapes and frequencies was calculated using the revised stiffness matrix. A
comparison is made among these three sets of modal responses. Descriptions of the building, experimental

* Associate Professor.
t Former Graduate Student.

0098-8847/88/091177-12$06.00 Received 27 February 1986

ii;: 1988 by John Wiley & Sons, Ltd. Revised I 1 March 1987
1178 M. A. M. TORKAMANI A N D A. K. AHMADI

setup, model of the building, modal responses resulting from the analytical model, vibration tests, revised
model, and comparison of these responses are given in the following sections.
Full-scale forced vibration tests have been used extensively to measure the dynamic response of buildings.
The most widely used methods of full-scale dynamic testing include ambient vibration and forced vibration
from harmonic vibration generators. Foutch4 and Petrovski et aL7 compared the results of these two testing
methods when applied to buildings. They found close agreement in the mode shapes and frequencies of the
buildings from both methods (using non-destructive forces in the case of vibration generator tests). However,
the duration of testing and amount of labour needed in vibration generator tests are much greater than for
ambient tests. When using the vibration generator test, additional problems and difficulties may arise if the
building is under construction, as was the building tested in this study. For these reasons only ambient
vibration testing is considered in this research.

DESCRIPTION OF BUILDING
The building selected for this study is the Six PPG (Pittsburgh Plate Glass) Place building having fifteen
storeys above ground and three underground garage levels. It is located in downtown Pittsburgh,
Pennsylvania. The exterior claddings of the building are made of laminated glass called ‘Natural Silver
Solarban 550 Clear Reflective Glass’. It consists of two layers of laminated glass each 0.25 in (6.35 mm) thick
with 0.5 ins (12.70 mm) distance between them.
The above ground portion of the building is a steel frame which is 204.75 ft (62.41 m) tall and includes a
mezzanine level, thirteen regular floors and a mechanical floor. The underground portion of the building is a
three-storey concrete frame parking garage extending 34 ft (1036 m) below ground. A typical plan of these
floors and the elevation view are shown in Figure 1.
Considering both major and minor axes of the building there are thirty-five plane frames in the building.
Some of these frames terminate at the ground floor. Three of these frames, which continue up to the roof, have
‘K’ bracing.

BUILDING MODEL
The type of mathematical model which is used to analyse the Six PPG Place building is a pseudo three-
dimensional model. The model consists of an assemblage of two-dimensional frames and shear walls that may
be arbitrarily oriented in plan. The frames are connected at each floor level by a diaphragm which is assumed
rigid in its own plane. The model has three dynamic degrees of freedom at each floor; two translations and one
rotation about the mass centre. Further discussion including the assumptions involved in this type of
modelling procedure, advantages, disadvantages, etc., are given in Reference 5.
South and west sides of the garage levels are filled by soil. The soil’s effects are modelled by discrete external
spring stiffeners. To find the modulus of subgrade reaction K , of the soil, it is assumed that the soil from ground
level to garage level one is loose sand, and from garage level one to garage level three is medium sand. In this
analysis, the lower limit of the soil properties is used to calculate K,.3
The garage levels on the east and part of the north sides are continuous. This continuation is modelled as
rigid links connected to discrete springs. K , for these two sides is computed as for the south and west.

AMBIENT VIBRATION TESTS

Three ambient vibration tests at different stages ofconstruction were performed on this building. The first test
was conducted when the main frames, floors, one stair case and freight elevator were completed. The elevators
(totally six of them), another stair case and the exterior claddings of the building were in place prior to the
second test. The third vibration test was performed after the partition walls were added. The data acquired
from these tests were processed to obtain the mode shapes and frequencies of the building at three stages of
construction.
 STIFFNESS IDENTIFICATION DURING CONSTRUCTION 1179

184' - 3"
t 1:

Elevators

Stair

--- 186' - 15''

--
A I. 1168-3" df 56'4

Figure I. A typical plan of floors and elevation view of the building

Eight Ranger Seismometers (Model SS- 1, Kinemetrics) were used to measure the responses simultaneously
at different locations. The seismometer signals, after amplification and filtering by two Signal Conditioners
(Model SC-1, Kinemetrics), are recorded on magnetic tape using an eight channel FM analogue tape recorder.
Figure 2 shows the general layout of ambient response measurement instrumentation.
As shown in Figure 3, seismometers I 4 were set and remained on the 14th floor during the vibration test,
while the other four seismometers were moved to different floors. Fifteen minutes of normal signals (velocities)
were recorded for each set-up. Throughout the testing, seismometers 1 and 5 were used to obtain information
about the N-S direction, while 2 and 6 obtained information about the E-W direction. Torsional information
was obtained by diminishing the effects of N-S modes. This was accomplished by algebraically subtracting
the signals of seismometers 3 from 4 and 7 from 8 at each end of the floor.
In the laboratory 102.4 sec of each recorded signal was digitized at a rate of 40 samples/sec, resulting in a
Nyquist frequency of 20 Hz. All digitized data were passed through the Hanning time window to minimize the
spectral spreading effect caused by the finite record length.8
A Fast Fourier Transform was computed for 4096 points of each record, resulting in 2048 spectral ordinates
and phase angles. The Fourier spectra are smoothed using five point averaging to increase the accuracy of the
spectral estimates. Figure 4 shows the smoothed Fourier spectra of the 14th floor of the third vibration test in
both N-S and E-W directions. Fourier spectra of the responses are used to determine the frequencies,
mode shapes and damping ratios (the half-power bandwidth method is used to find damping ratios) of
vibration of Six PPG Place. The first nine frequencies and mode shapes of vibration of the building are easily
detected. Frequencies resulting from data processing, along with the first nine frequencies of the analytical
model, are shown in Table I. Mode shapes obtained are representative only of the dominant directions (one-
dimensional).

PARAMETER ESTIMATION
Application of parameter estimation in a 3-D model of buildings is the subject of the companion paper.'
Therefore only a brief discussion of the identification procedure is presented. There are three steps in a
parameters estimation:
I I80 M. A. M. TORKAMANI A N D A. K. AHMADI

1 Signal

!
Conditioner
0
E

(Amplifiers
and
-i Recorder Oscilloscope

a, Filters)
C
ld

Figure 2. Ambient response measurement instrumentation

N
t

Figure 3. Layout of Ranger Seismometers

(1) determination of the form of the model and isolation of the unknown parameters;
(2) selection of a criterion function (error function); and
(3) selection of an algorithm for adjustment of the parameters.’

The model considered for this building is a pseudo three-dimensional model with 3n x 3n (n is the number of
floors) stiffness and mass matrices. To obtain an accurate mass matrix, the masses of all structural elements,
systems and components are considered. To study the sensitivity of the modal responses to changes in the
mass of the structure, the computed mass of each floor is increased by 5, 10, 15, 20 and 25 per cent. The
sensitivity study indicates that the range of frequency change is between 1 and 10 per cent, while the variation
in mode shapes is negligible. However, in the estimation of the parameters considering tests number two and
three, the mass of the mechanical floor and all other floors is increased by 50 and 5 per cent, respectively, in the
analytical model. These changes are due to construction changes, i.e. addition of non-structural elements, and
installation of heavy machinery at the mechanical floor.
The total stiffness matrix is formed from superposition of two stiffness matrices:
CKI = CK,1+ CK’l (1)
where
[K ] is the stiffness matrix of building, considering structural and non-structural elements
[ K l ] is the stiffness matrix of structural elements
[ K 2 ] is the stiffness matrix of non-structural and variable structural elements.
 STIFFNESS IDENTIFICATION DURING CONSTRUCTION 1181

Figure 4. Smoothed Fourier amplitude spectra of 14th floor response; (a) N-S direction; (b) E-W direction

[ K is constructed from a mathematical model of the building which considers known structural elements; it
is assumed that the elements of this matrix are independent of the parameters of the system. O n the other
hand, the matrix [ K 2 ] is constructed from a fictitious model which considers non-structural elements,
variable stiffness structural elements and the contribution of the joint rotations. This model passes through
1182 M. A. M. T O R K A M A N I A N D A. K. A H M A D I

Table I. Modal frequencies of mathematical model and three

vibration tests (Hz)

Mode
Directions 1 2 3

E-W 0.32089 0.94571 1.63045

Analytical N-S 0.47875 1.57720 2.14984
Tor. 0.41570 1.25126 2.02402

E-W 0.40039 1.15230 1.96289

Test No. 1 N-S 0.48828 1.58203 2.73437
Tor. 0.42969 1.25976 2.14844

E-W 044922 1.45508 2.54883

Test No. 2 N-S 0.49805 1.63086 3.14453
Tor. 0.55664 1.70898 2.83203
~~

E-W 0.46875 1.49414 2.55859

Test No. 3 N-S 0.50871 1.64062 3.24219
Tor. 0.5761 7 1.94336 2.83203

the mass centre of each floor and has two translational and one torsional degree of freedom at each floor. The
unknown parameters are the entries of the banded stiffness matrix [Kt], which is discussed in the companion
paper,' and have the following half-banded form:

The error function for parameter estimation, which is considered in this study, is
E(Pj)={ {ARl - C s l { A ~ H T C c r l - ' {(ARj-csl(AP)I +B{AP)TCc,l-'{A~)
where
{AR}={R*}-{R}i
{AP}={P}i+l- { l ' } i
{ R* ) are measured responses
{R} are analytical responses
{ P } are unknown parameters
p ( p ) is a variable weighting factor
[C,] is the covariance matrix of error in measured responses
[C,] is the covariance matrix of error in prior parameters
 STIFFNESS IDENTIFICATION DURING CONSTRUCTION 1 I83

The Gauss-Newton algorithm is used to adjust the parameters, which leads to the following two equations,

+ccrl)-"~l"ccpl,
The first equation is the Improved Statistical Structural Identification and the second is the covariance matrix
of the revised parameters.
The Improved Statistical Structural Identification is used to improve the stiffness matrix of the Six PPG
Place building at different stages of construction. All measured eigenvalues and eigenvectors (nine frequencies
and mode shapes) are used in these identification processes to estimate the 54 parameters of the building.
Because one-dimensional mode shapes are obtained from the ambient vibration test, system identification is
carried out with 1/3 of entries of each analytical mode shape. In the computation of [C,], it is assumed that all
measured responses have a standard deviation less than 4 per cent of their value. Frequencies of the improved
mathematical model resulting after three iterations are shown in Table 11.
Figures 5 to 7 are the first three, E-W direction, mode shapes of the analytical model, vibration tests and
improved models. Figure 5 shows a very good agreement between the test results (test number one) and the
mode shapes of the analytical model of the main frames, floors and one staircase. The mode shapes of the
improved model and the test results have an excellent agreement. Figures 6 and 7 show that the agreement
between the test results and the analytical model is not as good as in test number one. The mode shapes of the
improved mathematical model show a good agreement for the first mode, and the revised stiffness matrix
contributed more in matching for the higher modes. Table I and I1 are modal frequencies of the analytical
model, test results and improved analytical model. They show a good agreement between the frequencies of
the test results and the improved analytical model. Revised stiffness matrices are reported in Reference 1.

SUMMARY AND CONCLUSIONS

A pseudo three-dimensional mathematical model is used to model an eighteen-storey building. Computation
of the mass matrix, which is based on the masses of all structural elements, systems and components, is more
accurate than that of the stiffness matrix. Typical structural elements such as columns, beams, panels and
diagonal elements are considered in the assembly of the stiffness matrix.
Three ambient vibration tests at different stages of construction were performed on this building. The first
nine frequencies resulting from the analytical model and three vibration tests are shown in Table I. Mode
shapes identified from ambient vibration tests are one-dimensional. Each mode shape represents the
dominant direction of vibration of the building.

Table 11. Modal frequencies of six PPG Place from improved math-
ematical models (Hz)
~ ~

Mode
Directions 1 2 3

E-W 0.38562 1.12857 1.94006

Fir. imp. mod. N-S 0.49053 1-58254 2.56854
Tor. 0.42297 1.25239 2.14251

E-W 0.42484 1-19161 2.08325

Sec. imp. mod. N-S 0.51033 1.60416 2.97280
Tor. 0.54255 1-74588 2.43492

E-W 0.44264 1.20355 2.07885

Thd. imp. mod. N-S 0.50935 0.56481 3.15151
Tor. 0.58361 1-93231 2.41916
1184 M. A. M. TORKAMANI A N D A. K. AHMADI

2
a
-0
0
F.
-0
a
>
0

-
L
a
€
0 ...
N N N

0
.r.

-
D
c: 2
a
-0

r.
I

r.
' I
 0-0 Rnalytical - - - -0 Measured +o Improved modeL
18 -
17 -
16 .
15 -
14 -
13 -
12 -
11 -
10 -
9-
8-
7-

6-
5-
4-

3-
2-
1-
0- I
t I 7
-1.041 0.000 1.041 -1.598 0.000 1.590 -10.688 o.oao 10.688
E-W Mode 1 E-W Mode 2 E-W Mode 3
0.32089 Hz. Anoly. 0.94571 Hz. Rnaly. 1.63045 Hr. Rnoly.
0.44922 HZ. Meas. 1.45508 HZ. Meas. 2.54883 HZ. Meas.
0.42484 HI. Imp. HodeL 1,19161 Hz. Imp. Model 2.08325 Hz. Imp. Model
Figure 6. Comparison of analytical, measured and improved model mode shapes, E-W Test No. 2
1186 M. A. M. TORKAMANI A N D A. K. AHMADI
 STIFFNESS IDENTIFICATION DURING CONSTRUCTION 1187

The following conclusions are drawn from a comparison of these mode shapes and frequencies.
1. Stairs, elevators and partition walls have a significant effect on the frequencies of the building.
2. Frequencies resulting from the analytical model and first vibration test for N-S and torsional directions
are very close. The results of the second vibration test show that a small difference resulted from the addition
of the second staircase and freight elevator.
3. The effect of staircases and elevators is greater on the frequencies of the E-W direction than on those of
the torsional and N-S directions. Since the direction of the staircases in this building is N-S, their major axes
are in the E-W direction, which causes a greater effect in this direction. Also, elevator doors are facing east and
west directions.
4. Because the exterior cladding of this building is made from glass, it does not have a significant effect on
frequencies and mode shapes of the building.
5. The partition walls in this building also have a small effect on the mode shapes and frequencies.*
6. Comparison of the mode shapes resulting from vibration tests and the analytical model for lower floors
shows excellent agreement, indicating that the assumption of a fixed base is a valid one.
The improved Statistical Structural Identification Formula is used to study the effects of non-structural
elements, at three stages of construction, for Six PPG Place. Nine frequencies and corresponding mode
shapes, with only 1/3 of the entries (dominant directions) which were identified from data processing of the
vibration tests data, are used for system identification. Also, a mass sensitivity study for this building is
performed to find the effect of variation of mass on frequencies and mode shapes. In parameter estimation of
tests numbers two and three, the mass of the mechanical and all other floors is increased by 50 and 5 per cent,
respectively, for the analytical model. These changes in mass are due to the installation of heavy machinery at
the mechanical floor and the addition of elevators, staircases, partition walls, etc., which occurred after test
number one.
The following conclusions are drawn from this part of the study.
1. The Improved Statistical Structural Identification Method, along with the model, which is presented in
the companion paper,' for non-structural and structural elements can be used for parameters identification of
large degrees of freedom models without the ill-conditioning problem.
2. Because the triple matrix product [S],[C,][S]: is singular, the Least Square and Weighted Least
Square Methods can not be used for identification of three-dimensional problems. It should be noted that as
approaches zero, the matrix that must be inverted in equations (2) and (3) approaches the above singular
matrix, therefore the final value of ,!lis small but never equal to zero.
3. The rate of convergence of frequencies is very good. See Tables I and 11. The rate of convergence ofmode
shapes is slower, which could result from using 1/3 of each mode shape for system identification.
4. The effects of non-structural elements on the stiffness matrix of this building are considerable. However,
the change in stiffness for the first few floors is not significant. This may be the results of considering the fixed
base in the analytical model, which affects the degrees of freedom of the first few floors.
5. To study the effect of the variation of mass on frequencies and mode shapes, the mass matrix of the
building is increased by 5, 10, 15,20 and 25 per cent. This study shows that the frequencies of this building are
not very sensitive to variation of mass. The range of change of frequencies is between 1 and 10 per cent, with
negligible change for mode shapes.

ACKNOWLEDGEMENT
The writers are grateful for partial support from National Science Foundation Grant CEE-8206909 and to
Dr. John B. Scalzi, program director.

REFERENCES
1 . A. K. Ahmadi, 'Application of system identification in mathematical modeling of buildings', Ph.D. Thesis, in progress at University of
Pittsburgh, Pittsburgh, Pennsylvania, 1986.

' The partition walls are few in number and are located at the core of the building.
1188 M. A. M. TORKAMANI AND A. K. AHMADI

2. G. A. Bekey, ‘System identification-An introduction and a survey’, Simulation 5, 151-166 (1970).

3. J. E. Bowles, Foundation Analysis and Design, McGraw-Hill, New York, 1982.
4. D. A. Foutch, ‘A study of the vibrational characteristics of two multistory buildings’, Technical Report E E R L 76-03, California
Institute of Technology, Pasadena, CA, 1976.
5. W. E. Gates, ‘The art of modeling buildings for dynamic seismic analysis’, Workshop earthquake-resist. reinforced concrete build. constr.
University of California, Berkeley, CA 11, 857-886 (1977).
6. P. C. Jennings, R. B. Matthiesen and J. B. Hoerner, ‘Forced vibration of a 22-story steel frame building’, Technical Report E E R L 71-01,
California Institute of Technology, Pasadena, CA, and University of California at Los Angeles, Los Angeles, CA, 1971.
7. J. Petrovski, R. M. Stephen, E. Gartenbaum, and J. G. Bouwkamp, ‘Dynamic behavior of a multistory triangular-shaped building’,
Technical Report EERC 76-3, College of Engineering, 16, 1157-1 176 (1988). University of California, Berkeley, CA, 1976.
8. D. S . Samuel, Digital Signal Analysis, Hayden, New Jersey, 1975.
9. M. A. M. Torkamani and A. K. Ahmadi, ‘Stiffness identification of two- and three dimensional frames’, Earthquake eng. struct. dyn.
16, 1157-1176 (1988).