Вы находитесь на странице: 1из 10

Soil Dynamics and Earthquake Engineering 16 (1997) 503-512

© 1997 Elsevier Science Limited


All rights reserved. Printed in Great Britain
PIhS0267-7261(97)00009-2 0267-7261/97/$17.00
ELSEVIER

Dynamic characterization of a long span bridge:


a finite element based approach
M. L. Wang*, G. Heo* & D. Satpathi*
*Department of Civil and Materials Engineering, University of Illinois, Chicago, IL-60607, USA
*Chungbuk National University, Chungbuk, South Korea

(Received 15 October 1996; revised 3 April 1997; accepted 9 April 1997)

Aging bridges coupled with increasing traffic loads are producing a severe toll on the
nation's infrastructure. This has made it necessary to take a closer look at the health of
existing bridges and develop automated damage identification methods if possible.
Recent works in the field of structural dynamics have shown that damage detection
techniques utilizing parameters like mode shapes, modal frequencies and damping
ratio~; can be used to identify damage in structural systems. It is, however, important to
be able to establish a baseline model for the structure first, and then a model updating
technique can be utilized to evaluate the condition of the structure from time to time.
It is with this goal in mind that the authors have decided to establish the process for
obtaining a baseline model for a long span bridge. Based on the actual design
drawings of a bridge, finite element (FE) models of the bridge in question are
developed using SDRC-IDEAS. Three models of the bridge are simulated using
Normal Mode Dynamics solver in SDRC-IDEAS to obtain the modal parameters of
interest, in this case the modal frequencies and the mode shapes. A modal assurance
criteria (MAC) is utilized to compare the different simulated mode shapes and, finally,
the modal frequencies that have been obtained from the FE analysis are compared to
frequencies that have been obtained from some preliminary field tests. © 1997
Elsevier Science Limited.

Keywords: dynamic testing, model analysis, finite element model, assurance criteria,
SDRC-IDEAS, long span bridge

I INTRODUCTION introduced structural integrity monitoring using digital


processing of vibration signals. They suggested an alter-
According to the National Bridge Inventory (NBI) database, native method for overall assessment of platform structural
50% of all bridges in this country are structurally deficient, integrity. The amount and distribution of the changes in
with steel bridges showing the greatest need for remedial natural frequencies were used for monitoring the structure
action. The estimated cost of repair is around $90 billion, u with relatively few measurements taken at the selected
The situation is further complicated by increasing traffic positions. Integrity monitoring was based on the assumption
loads moving at faster speeds and natural calamities in the that each load-carrying member was involved in the overall
form of earthquakes and hurricanes which subject the struc- vibration modes. Abdel-Ghaffar and Masri proposed struc-
tures to very high levels of dynamic loading. Customarily, tural health monitoring of long span cable-supported
most bridges are inspected once every two years. 2 The bridges. 4 A number of similar comparisons of finite element
visual inspection of long bridges is a cumbersome process model with experimental data have been reported for var-
but is still the dominant method of inspection technique for ious types of bridge construction. 6-8 A number of
cracking, corrosion and other forms of deterioration. It is approaches for correcting and updating finite element
thus imperative that a an automated system of bridge inspec- models have also been presented. 9-12
tion and monitoring be developed. Many researchers have The current article is a part of an ongoing research effort to
been trying to develop the structural monitoring system characterize the serviceability of different bridges. The
using dynamic parameters. 3-5 For example, Begg et al. 3 increasing age of the structures in conjunction with increased
503
504 M. L. Wang et al.

l .f__

40 m 40m I I

Fig. 1. A typical span of the bridge.

traffic load has raised some serious issues and questions espe- 2 FEM ANALYSIS OF A LONG SPAN BRIDGE
cially in the aftermath of a large number of bridge collapses all
around the world. This article is the first phase of a two part The bridge selected for this modeling is approximately
approach to structural dynamics based damage monitoring 960 m long and the deck of the bridge is 30 m above the
technique. Only the issues of finite element (FE) modeling river High Flood Level (HFL) shown in Fig. 1. It is made up
of a typical long span bridge have been discussed. Developing of 8 spans, each of which are 120 m in length. Because of
an FE model of such a large structure, however, involves symmetry, only three of the spans will be modeled and
making simplifying assumptions at different stages. analyzed using SDRC-IDEAS. 13"14 A typical span is
However, each step introduces approximations which shown in Fig. 1. The FE model includes pier sections,
contribute to the deterioration of the accuracy and reliability detailed end and center sections and simplified mid
of the model. Inaccuracies in the prediction of the sections. It takes advantage of shell and solid elements
dynamic behavior also result from the inability to model types along with 1-D beam elements that are used to
non-linearity, damping mechanisms, coupling effects, and represent the truss system underneath the bridge. Generic
boundary conditions. Uncertainties in the material and geo- material properties for the analysis are specified as shown in
metrical properties and meshing errors also generate large Table 1.
discrepancies between the structure and its modeling. Three sets of modeling and simulation are carried out:

13.5 13.5
1
0 . 2"i, ? 0.2~ 2.4,,_

1.2 ! rlillil
, J, 0.4 ,l
9.46 _t___ Symmetric
0.034 v
Symmetric
0.02
9.0 1.0 [ 0.012

All Units : m
(a) Cross Section of Box Beams (b) Cross Section of I Beams

10 10 10 10, 20 *

7 *'bTa~
dc ~ Pf3 Pf4 Symmetric
Or, 0n i[ ,
All Units : m
a: 0.632 Pn = 207.40 Kg/m3
b: 1.785 Pf2 = 256.00 Kg/m3
c: 2.570 Of3 = 334.28 Kg/m3
d: 3.356 Pf4= 381.30 Kg/m3
e: 4.142 pf5= 1232.01 Kg/m3

(c) Transformed Side Section


Fig. 2. Section view and transformed section.
Dynamic characterization of a long span bridge 505

Table 1. Material properties of the bridge 2. the geometry instead of being continuous is made up
Material Young's Density Poisson's of discontinuous sections of varying thickness as
modulus (GPa) (Kg/m 3) ratio shown in Figs 1 and 2;
3. the sections comprising of the box beams are trans-
Concrete and 22.1 2400 0.2
formed into equivalent shells with the same I~. I~ is
asphalt
Steel 200.0 7850 0.3 ignored in this transformation process;
4. to ensure that the mass distribution remained the same
the mass density of the shell material was adjusted
based on the following equationlS:
1. preliminary analysis, where a large number of sim-
plifying assumptions were made at the modeling
stage. This analysis is motivated by the necessity to
~--Pilri (1)
Peq = ix x _}_ Ivy"
obtain rough numbers for the actual field testing of
the structure. The parameters of interest are the mode
shapes and modal frequencies; . finally, the concrete and asphalt deck is transformed
2. detailed analysis, where the structure is modeled to into an equivalent thickness of steel by using the sec-
incorporate a higher degree of complexity and bound- tion transformation principle. The bridge model
ary conditions and is to serve as the actual model representing the three spans have been modeled as a
against which the e~:perimental results are to be continuous structure with appropriate boundary con-
compared; and ditions applied at the pier supports. It is assumed that
3. side load, where non-dimensional masses are added to the end support acts as a pin and the intermediate
the sides of the detaiLled model to account for the supports simulate a roller support as shown in Fig.
architectural arches on the sides of the bridge 3. The simplified bridge analysis provides a quick
shown in Fig. 1. overview of the natural frequencies of the structure.
This is not meant to be an exact analysis, but the
intention is to obtain approximate numbers.
2.1 Preliminary analysis

The complicated bridge geometry is simplified with two 2.2 Detailed modeling
goals in mind: reduction of the degrees of freedom and
saved modeling time. In this modeling process the following The detailed model as shown in Fig. 4 is developed with the
assumptions have been made: intention of establishing a more accurate representation of
the bridge's actual geometry and boundary conditions. The
1. the height of the truss varies linearly instead of being truss system underneath the travel way is modeled in detail
an arch as shown in Figs 1 and 2; along with the concrete support/pier section. The mid span

Thin Shell
Simplified Bridge Mesh
~ - ~ Roller I Material Thickness m I
, ~ \ I Ft 4.142 I
~,m \ F2 3 356

t/'~"~Z,< ~'>'~,- ~ / F4 1.785

" v " " ~ ~ , ~)~M ~ F4 Roller Tranfonned Section

108 Elements ~ a~, -,-c.,..7--.


~

Elements - 108
Restraints - 4 Places :1 Pinned 3 Roller Pinned
Boundsry Conditions - Pinned X, Y, Z (Translation) Restrained
X, Y, Z (Rotation) Free
Rolller Y, Z (Translation) Restrained X (Translation) Free ) .
X,Y (Rotation) Restrained Z (Rotation) Free
Material - Generic Isotropic Steel
Thickne.ms - Per Transformed Section Table

Fig. 3. Simplified bridge model.


506 M. L. Wang et al.

Detailed Bridge
Mesh (w/o Boundary Conditions)

p Left End
/
~ . ~ MidSpan

Restraints - 256
I Paris Ellmnent Type
Pier Solid
Ends Shell end Beam
Center Shell and Beam
Md Shell (Transformed)

Fig. 4. Detailed bridge model (w/o boundary conditions).

uses the transformed section from the simplified bridge and right end, center, and mid section) measured 40 m
model. An added element type is used to attach the center in length and 27 m in width. The vertical distance
and end sections to the piers. This element type is called a between the top surface of the pier and the bridge deck is
rigid constraint. The rigid constraint allows the superstruc- 1 1 . 6 6 m.
ture to be attached to the substructure (pier in this case) The left and right end consisted of 2-D shell elements (72
thereby allowing for a uniform load transfer. The truss elements) and 1-D beam elements (105 elements). This
system includes bottom and top box beams along with strin- decision was taken to preserve the geometrical accuracy
ger beams to add stiffness. Cross bracings have been com- of the actual structure. Beam elements using the material
pletely ignored in this analysis. The bridge model measured properties of generic isotropic steel are used to model the
360 m from the left end to the right end. Each section (left box beams; 0.25 m thick shell elements with the property of

End Span (Underside)


Mesh
2 - D Shell

Top Box ~

sodom , o x

Element Type Material Number of Elements

1 - D Beam Generic Isotroplc Steel 105


2 - D Shell Concrete 72

Fig. 5. End span model.


Dynamic characterization of a long span bridge 507

Center Span Underside


Mesh (w/o Boundary Conditions) s.~,

Beam Elements

Element Type Material Number of Elements

t - O Beam Generic Isotroplc Steel 455


2 - D Shell Concrete 312

Fig. 6. Center span model.

concrete are used to model the reinforced concrete and is shown in Fig. 6. The pier section as shown in Fig. 7 was
asphalt deck of the superstructure. The detailed model of modeled using 1439 3-D solid elements. The material
the end span is shown in Fig. 5. assigned to this section is concrete.
The center span is made up of 2-D shell elements (312 The boundary conditions (Fig. 8) for the bridge are
elements) and I-D beam elements (455 elements). Beam introduced at the foundations (pier base) and at the attach-
elements identical to those described in the previous section ments to the piers. The pier bases were attached by clamping
is used to model the box beams and identical shell elements or restraining all six degrees of freedom. The truss to pier
are used for the deck. The detailed model of the center span attachment points are clamped. This is based on the actual

Pier
Mesh

Solid
3 - D
Elements

Element T y p e Material Number of Elements

3 - D Solid Concrete 1439

Fig. 7. Pier section model.


508 M.L. Wane et al.

Detailed Bridge
Mesh (w Boundary Conditions)

Boundary Conditions -
C l a m p e d - X , Y , Z (Trans. and Rot.) Restrained
Restrained - X,Y, Z (Trans. and R o L )

Fig. 8. Detailed bridge boundary conditions.

site observation of this structure. Final adjustments to the from the three different sets of analysis is given in Table 2.
boundary conditions have to be made at the model updating The results clearly show that the frequencies are very clo-
stage. sely spaced, a characteristic that is typical of large struc-
tures. This is because of geometric non-linearity and large
2.3 Detailed model with side load degrees of redundancy that is typically characteristic of
these class of structures.
In this analysis the same detailed model that is developed in Comparison between the simplified model and detailed
the previous section is used. However, the loading model shows some differences that can be explained on the
conditions are changed by adding non-dimensional masses basis that follows. Initial low frequencies are a result of the
to the side of the structure. This is done to evaluate the effect modeling simplifications that resulted in a loss of stiffness in
of the architectural arches on the side of the bridge. comparison to the actual structure. It is, however, observed
A schematic view for the side arch structure of the bridge that from the fourth mode onwards, the modal frequencies
is show in Fig. 9. The mass of this structure was calculated are higher than those obtained from the detailed model. This
on the basis of its volume and density and was added as a can be explained on the basis of the modeling assumptions.
non-dimensional mass to the edge shell elements. The mass Long span structural systems are characterized by closely
per unit area for the additional side arch structure is calcu- spaced modal frequencies due to the redundancies and non-
lated as 70.65 Kg/m 2. linearity of the structure. Modeling simplifications resulted
in a loss of some degrees of redundancy and non-linearity in
the simplified model thereby increasing the spacing between
3 ANALYSIS OF RESULTS subsequent modes. The actual objective behind the simpli-
fied analysis is not to obtain accurate results but to get a
The outcome of any modal analysis consists of the follow- rough idea of the modal frequencies and mode shapes prior
ing modal parameters: modal frequencies, modal shapes and to carrying out a modal test on the bridge.
modal damping ratios. Incorporating damping into a finite The side load analysis of the detailed model is done
element model is possible but extremely difficult and has because it was felt that the addition of side loads would
not considered in the present analysis. Thus, the results that allow torsional modes to develop earlier in comparison to
are obtained from the analysis are mode shapes and modal the detailed model. The most effective way to establish that
frequencies. A comparison of the different modal frequencies will be to compare the mode shapes.

I 40 m I 40 m I 40 m
,-'~ ~ - - ; ~ -'~ "1 15.26 m

2.0 m -~-
-f 11.66 in 0.009 rn

Fig. 9. Schematic view o f the side arch structure o f the bridge.


Dynamic characterization of a long span bridge 509

Table 2. Comparison of the first 10 modal frequencies from the three different analyses

Freq. (Hz) 1st 2n,:l 3rd 4th 5th 6th 7th 8th 9th 10th
SM 0.34 0.76 1.46 2.23 3.17 4.13 5.04 5.06 5.10 6.85
DM 1.23 1.99 2.11 2.16 2.23 2.62 2.76 2.84 2.92 2.94
DMwSL 1.19 1.91 2.02 2.10 2.13 2.54 2.65 2.73 2.81 2.87
SM: simplified model; DM: detailed model; DMwSL: detailed model with side load.

Table 3. Comparison of typical mode shape MAC between the detailed model without side load and detailed model with side load

Mode 1st bending 2nd bending 1st torsion 2nd torsion


1st bending 1.000 0.001 0.000 0.000
2nd bending 0.001 0.999 0.001 0.005
1st torsion 0.000 0.001 0.986 0.000
2nd torsion 0.000 0.005 0.000 0.999

An effective way to compare two sets of structural yield a value greater then 0.9 and uncorrelated modes will
dynamic data is to come up with some kind of a correlation yield a value less then 0.005. M A C is not affected by a
measure. Popular measures of the correlation are the fre- scalar multiple.
quency relative errors and the modal assurance criterion Table 3 and Figs 10 and 11 show no differences in mode
(MAC), which is sometimes referred to as the modal shapes between the detailed model without side load and
correlation coefficient and is defined as detailed model with side load. This means that the addition
of side loads does not lead to earlier development of tor-
n 2
sional modes.

MAC(/,j) = (2)
4 EXPERIMENTAL VERIFICATION
j=l j=l
Some preliminary tests have been performed on the bridge,
where a means analytical data and e means experimental results from which are reported in this article. Traffic is
data. M A C is calculated to quantify the correlation between closed on the bridge and the bridge is excited by a truck
measured mode shapes during the different tests and to traveling across the bridge. The response of the bridge is
check the orthogonality of measured mode shapes during recorded in the form of strain, deflection and acceleration
a particular test. M A C makes use of the orthogonality prop- data as ~( function of the truck speed and the bridge span.
erties of the mode shapes l:o compare either two modes However, a full-fledged modal test has not been performed
from the same test or two modes from different tests. If and thus the only data that is pertinent to this article are the
the modes are identical, a scalar value of one is calculated power spectral densities (PSD) of the acceleration response
using the MAC. If modes are orthogonal, a value of zero is of the structure as shown in Fig. 12. The schematic diagram
calculated. Ewins 16 points out that correlated modes will indicating the positions of the accelerometers on the bridge
is shown in Fig. 13. The power spectral densities obtained
from the tests failed to represent the entire frequency range
MAC
! obtained from the analysis. In the channel A data, a peak
is observed at 1.7 Hz followed by a stronger peak at 3.15
0.8
Hz. The 1.7 Hz frequency is close enough to the second
OI- mode of the FE model. The 3.15 Hz corresponds to the
a.4.
eleventh mode in the FE model. The channel B data
contains a peak at 2.14 Hz and 2.5 Hz. The 2.14 Hz fre-
a.2~ quency corresponds to the fourth mode and the 2.5 Hz
a:
corresponds to the sixth mode. The first mode frequency
as predicted by the finite element model is missing in
,l
the experimental data. Similarly, modes between 2.5 Hz
Mode
and 3.0 Hz are conspicuously absent. The authors are of
a 0
the opinion that this is a consequence of the nature of
the excitation used. The spectrum generated by a truck
Fig. 10. Comparison between detailed model without side load passing over a bridge is far from being a wide band
and with side load (typical first Lwo bending and torsional modes). excitation. The excitation frequencies in this case are
510 M.L. Wang et al.

Detailed ModelwithoutSideLoad Detailed Modelwith Side Load


~ ::~ M1 : 1.23Hz

M3:2.11.z

M4 : 2.16Hz
~ 4~IIIvMd : 2.10Hz

M5: 2.13

M7 : 2.76 Hz
~ M7: 2.65Hz

~ ::~M8 : 2:84Hz

Fig. 11. Mode shape of the bridge.

dependent on the characteristics of the suspension system of 5 CONCLUSIONS


the loading truck. Instrumenting the loading truck to deter-
mine its force spectrum can provide a conclusive answer to Three sets of modeling and simulation are carried out. In the
this discrepancy. preliminary analysis, a large number of simplifying assump-
Although, it would be beneficial to be able to charac- tions are made at the modeling stage. In the subsequent
terize the structural dynamic properties of a system from detailed analysis the model incorporated a larger degree of
ambient vibration response, there are some significant complexity. The model of the superstructure included more
difficulties involved in the process. The first and fore- details and the piers were added to model the behavior more
most amongst them is the fact that ambient vibration accurately. The detailed model with side load where non-
data are susceptible to more noise and this makes the dimensional masses are added to the sides of the detailed
task of mode identification more difficult. Secondly, as model to account for the architectural arches on the sides of
discussed earlier, depending on the spectrum that is gener- the bridge is performed to investigate the effect of adding on
ated by the passing traffic, the modes that are necessary or massive architectural loads to the structure. The analysis as
important for developing a damage detection scheme detailed above serves to streamline the entire process of
may be missing. This is a very severe drawback for the modal analysis and testing of large span structures. The
ambient vibration method because the ultimate objective simplified model serves to establish a preliminary basis
of this effort is to be able to utilize the process for damage for performing a modal test on the structure. It is recom-
assessment. mended that this process be carried out prior to running a
Dynamic characterization of a long span bridge 511

A.O Up

,At2 AI0 A8;


i /(A6 A4 ,e-~
"A2~ .
L South
MAC North

"A5
AI
,5 Down

Fig. 13. Schematic diagram for the position of the accelerometers.


PWR ,~= ~l. hl=~ LIN 101"11
X: 3. 150Hz Y" I I . 28,,*V
excitation. This may prove to be a serious problem in the
case of large flexible structures characterized by closely
Power Spectrum from the Sensor spaced very low frequencies. To be able to use ambient
in Vertical Direction traffic excitation it is necessary to do a force identification
study of traffic movement. The authors are unaware of any
l0 •

t research in this area at the moment.

ACKNOWLEDGEMENTS
i.i 4

mV
The authors acknowledge Dr. Choi and Dr. Lee for pro-
viding the drawings of the bridge and also furnishing the
,s[t necessary experimental data. We also acknowledge the help
of Dr. Walter Gerstle and Robert Valerio of the Department
of Civil Engineering at UNM for their valuable suggestions
0l 0 PWR SP R ~e,~ L|N I 0Hz and assistance with the finite element modeling. This project
x: 2. 5 2 5 H z Y: 3 . 922mV
was partially supported by NSF grant #9622576. Dr. S.C.
Liu is the program manager.
Power Spectrtml from the Sensor
in Lateral Direction
Fig. 12. Power spectral densities.
REFERENCES
test. The detailed analysis which is more time consuming 1. Silano L. G. (Ed.) Bridge Inspection and Rehabilitation--A
and elaborate when used in conjunction with experimental Practical Guide, Wiley, New York, 1992.
modal data serves as the basis for developing the structural 2. Caltrans. ABME Structure Maintrance Procedures, Caltrans,
integrity of a system. On the basis of the results that had Division of Structures, Office of Structure Maintrance and
been obtained from the FE analysis it may be inferred that a Investigations, Investigations Branch, 1995.
3. Begg, R. D., Mackenzie, A. C., Dodds, C. J. & Loland, O.
simplified model may be a good starting point but needs to Structural Integrity Monitoring Using Digital Processing of
validated against experimental data. Subsequently, depend- Vibration Signals, Offshore Technology Conference, Dallas,
ing on the performance of the simplified model, it may be TX, 1976.
necessary to update and refine the model, adding more com- 4. Abdel-Ghaffar, A. M. & Masri, S. F. Dynamic Response and
plexity to the model in a step by step fashion and validating Structural Health Monitoring of Long-Span Cable-Supported
Bridges. In Proceedings Cable Stayed Bridges, Recent
it against the experimental data at each step. The method Developments and their Future, Yokohama, Japan, 1991.
that has been enumerated in the article can be considered to 5. Alampalli, S., Fu, G. & Dillon, E. W. On the u~e of measured
be some kind of an FE model updating technique. vibration for detecting bridge damage. In Conference Pro-
Although the experimen~:al results obtained from the ceedings 7, 4th International Bridge Engineering Confer-
bridge are very preliminary, they bring up certain issues ence, Vol. 1, August, 1995.
6. Ventura, C. E., Felber, A. J. &Prion, H. G. L. Seismic
pertaining to the health monitoring of structures. In spite evaluation of a long span bridge by modal testing. In Pro-
of the experimental data being incomplete, a comparison ceedings of the 12th International Modal Analysis Confer-
of the modal frequencies obtained from the experimental ence, Honolulu, HI, 1994, pp. 1309-1315.
data and the FE analysis shows fairly good agreement. 7. Rotter, T., Kohoutek, R. & Marusiak, G. Modal analysis of
The incompleteness of the experimental results brings up rail bridge in Mlada Boleslav. In Proceedings of the 12th
International Modal Analysis Conference, Honolulu, HI,
the limitations of using ambient traffic excitation to monitor 1994, pp. 1316-1320.
a large structural system. As has been explained earlier 8. Aktan, A. E., Chuntavan, C., Lee, K.-L. & Toksoy, T. Struc-
this is due to the limited spectral content of this type of tural Identification of a Steel Stringer Bridge, Transportation
512 M. L. Wang et al.

Research Record 1393, National Academy Press, Washing- Proceedings of the 1991 European Conference on New
ton, DC, 1993. Advances in Computational Structural Mechanics, 1991.
9. Minas, C. & Inman, D, J. Correcting finite element models 12. Tong, M., Liang, Z. & Lee, G. C. Correction criteria of finite
with measured modal results using eigen structure assign- element modeling in structural dynamics. Journal of Engi-
ment methods. In Proceedings of the 6th International neering Mechanics, 1992, 118, 4.
Modal Analysis Conference, April, 1988. 13. SDRC. Exploring I-DEAS Simulation, Vols I and I1, 1996.
10. Simmermacher, T., Zimmerman, D. C., Reese, G. M. & 14. SDRC. Exploring I-DEAS Design, Vols I and II, 1996.
James, G. H. The effects of finite element grid density on 15. Farrar, C. R., Duffey, T. A., Goldman, P. A., Jauregui, D. V.
model correlation and damage detection of a bridge. In & Vigil, J. S. Finite Element Analysis of the 1-40 Bridge over
Proceedings of the 36th A1AA/ASME/ASCE/AHS/ASCE Struc- the Rio Grande, Los Alamos National Laboratory, LA-
tures, Structural Dynamics, and Materials Conference, 1995. 12979-MS, Los Alamos, 1996.
11. Reynier, M. & Nedjar, D. Control of finite element models: 16. Ewins, D. J. Modal Testing: Theo~ and Practice, Research
the M.A.T. software (Model Adjustment Technique). In Studies Press LTD, Wiley, New York, 1994.

Вам также может понравиться