You are on page 1of 112

The Story of the Koror Bridge

IABSE Bulletins

The Story of the Koror Bridge


Koror Babeldaob Bridge, also called Koror Babelthuap Bridge
or simply Koror Bridge, connects the islands of Koror and
Case Studies
Babeldaob in the Republic of Palau. The design of the bridge
began in 1974 and was based on the prevailing AASHO
Standard Specifications at that time and was supplemented by
ACI and CEB-FIP design recommendations on an as-needed
basis. When the Koror Bridge was opened to traffic in April
1977, it was the world’s longest concrete girder span. A few
1
years later, the bridge began to deflect more than had been
anticipated. The owner commissioned a Japanese engineer-
ing firm in 1985 and then a US engineering firm in 1993 to
Man-Chung Tang
conduct in-depth investigations of the structure. Both firms

M.-C. Tang
came to the same conclusion that the bridge was structurally
safe and that the excessive deflection was an unexplainable
phenomenon. Nevertheless, in order to improve the driving
quality of the bridge deck, the owner decided to repair the
bridge. The repair scheme made changes to the structural
system and added a large amount of post-tensioning force The Story of the
to the bridge. Unfortunately, less than three months after the
repair, late in the afternoon on 26 September, 1996, nineteen
and a half years after it was opened to traffic, the bridge col- Koror Bridge
lapsed. Thereafter, most of the documents were sealed as a

1
result of litigation between the various parties and the debris
was cleared. For a long time, it was impossible to study the
facts surrounding the bridge’s collapse. Only recently, through
continuous probing by a group of engineers, were these docu-
ments made accessible to researchers.

IABSE Bulletins/ Case Studies

International Association for Bridge and Structural Engineering (IABSE)


Case Studies

1
Man-Chung Tang

The Story of the


Koror Bridge

International Association for Bridge and Structural Engineering (IABSE)


Copyright © 2014 by
International Association for Bridge and Structural Engineering

All rights reserved. No part of this book may be reproduced in any form or by any means, electronic
or mechanical, including photocopying, recording, or by any information storage and retrieval
system, without permission in writing from the publisher.
ISBN 978-3-85748-136-9

Publisher:
IABSE
c/o ETH Zürich
CH-8093 Zürich, Switzerland

Phone: Int. + 41-44-633 2647


Fax: Int. + 41-44-633 1241
E-mail: secretariat@iabse.org
Web: www.iabse.org
Preface

Koror Babeldaob Bridge, also called Koror Babelthuap Bridge or simply Koror Bridge, con-
nects the islands of Koror and Babeldaob in the Republic of Palau. The design of the bridge
began in 1974 and was based on the prevailing AASHO Standard Specifications at that time and
was supplemented by ACI and CEB-FIP design recommendations on an as-needed basis.

When the Koror Bridge was opened to traffic in April 1977, it was the world’s longest concrete
girder span. A few years later, the bridge began to deflect more than had been anticipated. The
owner commissioned a Japanese engineering firm in 1985 and then a US engineering firm in 1993
to conduct in-depth investigations of the structure. Both firms came to the same conclusion that the
bridge was structurally safe and that the excessive deflection was an unexplainable phenomenon.
Nevertheless, in order to improve the driving quality of the bridge deck, the owner decided to repair
the bridge. The repair scheme made changes to the structural system and added a large amount of
post-tensioning force to the bridge. Unfortunately, less than three months after the repair, late in
the afternoon on 26 September, 1996, the bridge collapsed. Thereafter, most of the documents were
sealed as a result of litigation between the various parties and the debris was cleared. For a long
time, it was impossible to study the facts surrounding the bridge’s collapse. Only recently, through
continuous probing by a group of engineers, were these documents made accessible to researchers.

Engineering is not science. The aim of science is to search for truth. The aim of engineering is
to serve human society’s needs. Based on what they already know, scientists make discoveries
about nature that are pre-existing. Based on an accumulation of experience, engineers improve
the built environment of human beings. Experience is never complete, but engineers obviously
cannot wait for scientific discoveries of all the necessary truths they need to design and build. At
the time of the construction of the Great Wall of China, the Pyramids of ancient Egypt and many
other great structures, the laws of gravity had not yet been discovered and an understanding of the
physical properties of building materials did not yet exist. Engineers must design and build based
on their present-day experience. This was the case 2000 years ago. And so this is the case today!
The foundation of engineering is what we learn from past experience; this includes both suc-
cessful and unsuccessful experiences. Learning from our mistakes is especially useful, as it can
teach us what can be done and what cannot be done. But, for the past to be useful, we must
present the facts carefully and honestly. This is the reason why I have written this book.

Man-Chung Tang
Table of Contents

Chapter 1: A Brief History 1


Republic of Palau 1
Bidding Process 3

Chapter 2: The Original Design 5


Design Concept 5
Prestressing System 9
Structural Safety of the Bridge 9
Creep and Shrinkage 18
Prestress Loss 20
Overturning Safety Factor 20
Prediction of Creep Deflection 20

Chapter 3: Construction 23
Field Change 31

Chapter 4: Bridge Performance 33


Test Results 37

Chapter 5: The Repair 41


Repair Scheme 41

Chapter 6: The Collapse 45


Forensic Observations 45
Failure Mechanism 54
Questions Raised after the Collapse 56

Chapter 7: What Would Have Happened If No Repairs To The Bridge


Had Been Made? 57
Chapter 8: Why Was The Deflection of The Bridge So Much Larger
Than Anticipated? 59
Calculation of Creep and Shrinkage 59
Current AASHTO/LRFD Model 59
The 1990 CEB-FIP Model 60
Discussion of Creep Estimate 61
The fib 2010 Model Code (Final Draft 2014) 62
The Time Function 62
The fib 2010 Model Code 63
Creep Under Tension 64
Calibration of Creep Models 64
B3 Model 64
ACI, AASHTO and CEB-FIP 65
The Question of the Loading Age 67
Other Effects on Deflection 70
Effect of Differential Shrinkage 71
Effect of Main Pier Settlement 71
Effect of Prestress Loss 72
Effect of Cracks on Deflection 74
Interpretation of Test Results 77
Experience from Another Bridge – the Pine Valley Creek Bridge, California 78
A Worldwide Survey 81

Chapter 9: Initiation of the Collapse 83

Chapter 10: Other Failure Hypotheses 87

Chapter 11: Discussions 91


Purpose of This Book 91
The Koror Bridge Was Structurally Safe 91
Long-Term Deflection 91
Prestress Loss Was the Most Probable Culprit 92

Summary 95

Acknowledgement 97

Units 99

References 101
1

Chapter

1
A Brief History

Republic of Palau
The Koror Bridge, also called Koror Babelthuap Bridge, or Koror Babeldaob Bridge, or Palau
Bridge (see Fig. 1), is located in the Republic of Palau, a country that consists of about 300
islands in the South Pacific Ocean, and is a part of the Caroline Island group. In 1899, Spain sold
these islands to Germany. In 1914, the Germans ceded them to Japan. During the Second World
War, there were several severe battles between the US and the Japanese armies on these islands
that led to the death of over 12,000 soldiers. After the war, the United Nations trusted these
islands to the US in 1947 as the US Trust Territory of Pacific Islands. In 1979, the entire Caroline
Island group declared independence from the US. Palau did not join them and established the
Republic of Palau in 1981.

When the islands were under US trusteeship, the US navy had a military base there. The
construction of the Koror Bridge was, in part, financed by the US government.

Fig. 1: Koror Babeldaob Bridge


2 CHAPTER 1. A BRIEF HISTORY

The Koror Bridge connected the Islands of Koror and Babeldaob (Babelthuap); it was the only
fixed link between these two islands. The majority of the population lived on Koror Island but
the only airport was located on Babeldaob Island. Before the bridge was built, the main form of
transportation between the islands was by ferries (see Fig. 2a,b).

The country of Palau is very remote; it is about 800 km away from the Philippines. The tem-
perature ranges from 23oC to 32oC. The average temperature is about 28oC. The average rela-
tive humidity is about 82%. Palau receives more than 2300 hours of sunshine in a year. Annual
rainfall is about 3670 mm.

The Republic of Palau has a total land area of 459 km2, with a total population of slightly more
than 20,000. Tourism is the main economic industry.

(a)
Mariana Islands

Luzon
Philippine Sea
Saipan
Manila Rota
Philippines Gaum, U.S.

Sulu Sea Mindanao


Palau Islands Project
location
Caroline Islands
Sulu Archipelago
Celebes Sea Helen Islands
Trust
Halmahera territory
of the Pacific Islands (U.S.)
Indonesia
Vicinity Map
Not to scale

(b)

Fig. 2: Location map


BIDDING PROCESS 3

Bidding Process
The original bid documents for the construction of the Koror Bridge were prepared by the
Hawaii consulting engineering firm, Alfred Yee and Associates (AYAA) [1–3]. The original
design featured a three span prestressed concrete girder with a centre span of 560 ft (170.68 m).
The deck provided two lanes of traffic and a pedestrian path on one side. The bid documents
permitted the successful contractor to modify the design to a 760 ft (231.65 m) span bridge in
order to clear the channel and avoid building the piers in the channel. The successful bidder was
Ajoo Construction Company from Korea. Their bid price was US$ 4.5 million.

However, Ajoo was not able to start construction for many months and so the bonding company
transferred the contract to Socio Construction Company, a firm originally from Korea with
headquarters in Guam. Because the bid price was considered to be too low, the bonding com-
pany provided an extra US$ 800,000, bringing the total contract price to US$ 5.30 million [4].

Socio decided that the construction of the two deep-water piers would be too expensive because
the water was about 30 m deep and the tidal current was very swift. It would have required
heavy equipment that was not available in the area. Renting and transporting equipment from
the Philippines or other countries would have been too costly to build two relatively small piers.
Therefore, Socio decided to follow the suggestion of the engineer to extend the span to 760 ft. It
was acknowledged that $5.3 million was still reasonable for a 760 ft span bridge. Therefore,
economy was an important consideration in the redesign of the bridge [5].

Socio hired the German firm, Dyckerhoff&Widmann, AG (Dywidag) to carry out the redesign.
The redesign started with a 760 ft span. However, just short of its completion, it was found that
the 760 ft span would still require the two main piers to be placed in the water and on the slope.
The span was then increased to 790 ft (240.792 m) (see Fig. 3). The redesign was reviewed and
approved by the original designer, AYAA. In the meantime, AYAA had merged with another
firm, Leo Daly. Leo Daly was responsible for supervision during the bridge construction.

Sand and stone were the only construction materials available in Palau. Almost everything else
had to be imported including prestressing steel from the US, cement and reinforcing steel from
Japan, and the form travelers from the US [5]. The travelers were adapted from those used in the
Pine Valley Creek Bridge in California. The workers were mostly from Korea.

The Pine Valley Creek Bridge was the first long-span prestressed concrete segmental bridge
in the US, and had a main span of 450 ft (137.16 m). Its construction began in 1972 and was

Fig. 3: Koror Bridge


4 CHAPTER 1. A BRIEF HISTORY

completed in 1975. The design was based on guidelines of the American Association of State
Highway Officials (AASHO) of the US. (AASHO was later changed to American Association
of State Highway and Transportation Officials, AASHTO [6]). At that time, long-span segmen-
tal prestressed concrete bridges were relatively new in the US; the AASHO design specifications
were not complete; and the design had to be supplemented by the American Concrete Institute,
USA (ACI) [7], the German DIN (German Industry Norms) and the European ComiteEuropeen
du Beton- Federation Internationale de la Precontraine (CEB-FIP) Recommendations [8].

At the time of redesign, Japan had just completed the 236 m span Hikoshina Bridge and the 240
m span Hamana Bridge. So the span of the Koror Bridge was not outside the realm of engineers’
understanding of what was possible. The 790 ft (240.792 m) span Koror Bridge, was longer than
the Hamana Bridge by less than 1 m and is also a world record (see Figs. 1 and 3).
5

Chapter

2
The Original Design

Design Concept
Selection of the type of structural system for the Koror Bridge was based on local conditions.
Palau is a remote island whose closest neighbour is the Philippines—at least 800 km away.
Both communication and transportation were difficult at that time. Most construction materi-
als had to be imported. The population was small, and there was not much infrastructure in the
area. Under such conditions, it would have been difficult to expect the owner to implement a
high-quality maintenance program for the bridge. In addition, the area is very humid and the
air is salty. Therefore, the selection of a prestressed concrete bridge for the original design was
a reasonable and logical choice. These conditions were considered and subsequently applied to
the redesign as well.

Several alternative structural systems were studied during the preliminary design stage. The
three final options were:

1. A continuous girder on four piers with sliding bearings at all of the piers, except at one of
the main piers, where the girder was restrained in the longitudinal direction;
2. A bridge girder monolithically fixed with the main piers and sliding bearings at the side
piers, plus a sliding hinge at mid-span to allow relative longitudinal movements;
3. Same as Option 2 except that the concrete hinge was used at the main piers to reduce
possible bending of the foundation.

Owing to the bridge location any metal component such as bearings could be subject to severe
corrosion. And, even with a 790 ft (240.79 m) span, the two main piers were still very close in
proximity to the salt water which would have regularly splashed the bridge and its main piers.
Therefore, any moveable bearing at these locations was not an appropriate solution. Besides,
these bearings would have been very large and difficult to fabricate in the 1970s. Consequently,
among the three options described above, Option 1 was deemed not acceptable.

Option 3 was not selected for two reasons. First, for such a long span, the concrete hinges would
have been exceptionally large and difficult to build and, secondly, a concrete hinge would only
work after the covering concrete had cracked, which would also be unacceptable due to pos-
sible corrosion, given the environment. Thus, Option 2 was selected. The bridge girder was
6 CHAPTER 2. THE ORIGINAL DESIGN

made monolithic with both main piers (see Fig. 4). As a result, a sliding hinge that only carries
shear forces had to be used at the span centre to allow relative longitudinal movements due to
temperature, creep and shrinkage. Thus, the main span was basically a pair of large cantilevers.
Weight was an important factor in the design of such a long-span bridge. Being cantilevers,
the mid-span portion of the box girder could be slimmed down significantly to 12 ft (3.66 m)
depth because there was little bending moment and shear in this area. This reduced the bending
moment in the entire cantilever [9].

Two schemes for the side spans were considered: 410 ft (125 m) and 237 ft (72.24 m) (see Fig. 4).
The shorter side span was chosen as it was more economical and visually better. Aesthetically,
the heavier side spans were preferable because they gave the slender main span an appearance
of strength.

The main girder was very narrow for such a long span. The deck was only 31 ft 7 inch (9.63
m) wide. For such a narrow box, shear lag in the deck was not considered as significant. So the
entire cross section was assumed effective.

The span length of AYAA’s original design was 560 ft and it had a narrower box. As the span
increased from 560 ft to 790 ft, the original box was found to be too narrow so in the redesign,
the width of the box was increased to 24 ft. This resulted in two very short overhangs in the top
slab. This new width of the box was about 1/33 of the main span, which was reasonable.

The box girder was 12 ft (3.66 m) deep at the span centre and increased parabolically to about
46 ft at the main piers. The girder depth at the piers was deeper than the usual rule of thumb of
1/20 of the span. The increased depth provided higher stiffness to reduce stresses and deflection
in the bridge girder. The stiffness of the girder near the main piers was especially effective in
this respect. The increased depth near the main piers also offered a more pleasing appearance
visually. The final configuration, where the girder was very deep at the pier and very shallow
near the mid-span, gave the bridge an aesthetic appearance.

The cylinder strength of concrete specified for this bridge was f’c=5000 psi (35 MPa), which
was equivalent to a cube strength of 45 MPa. The AASHO specifications allowed the girder to
have tensile stresses under dead and live loads up to 0.21 ksi (1.5 MPa). But the design of the
Koror Bridge was more conservative and had set the allowable tensile stress to 0. The allowable

53.65 240.79 53.65

125.00 240.79 125.00

Fig. 4: Two options of end spans


DESIGN CONCEPT 7

compressive stress was 0.4 f’c = 2.0 ksi ( 14.0 MPa). The maximum allowable concrete stresses
during construction and before creep, shrinkage and prestress loss was +0.21 ksi (+1.50 MPa)
tension and 2.2 ksi (15.75 MPa) compression.

The unit weight of concrete was assumed to be 0.150 kip/ft3 (24 kN/m3). Field tests of the con-
crete mix that was used showed that the concrete itself weighted 0.1396 kip/ft3 (22.4 kN/m3);
after adding mild and prestressing reinforcement it was about 0.150 kip/ft3.

At the time of the design of the Pine Valley Creek and Koror bridges, the shear provisions in
AASHO were mainly intended for small-span girders and were not appropriate for the design
of large box girders [10]. Applying those provisions would have resulted in an “under-design”
of the web reinforcement. Therefore, the shear design of the Koror Bridge did not follow the
prevailing AASHO provisions. A truss analogy was used instead.

The webs of the Koror Bridge were 14 inch (356 mm) thick except in the end portion of the side
spans, where they were 23 inch (584 mm). For a long-span concrete bridge, it is important to
make the webs as thin as possible to reduce weight. However, in the design of the Koror Bridge,
extensive analysis was done to assure the
safety of the bridge despite its thin webs. α
T
The shear design of the Koror Bridge was V
based on a truss analogy (see Fig. 5), simi-
lar to what is called the strut-and-tie method D M
today. Based on the US design rules, only
the ultimate condition was considered. The B
maximum ultimate shear at a cross section β
was calculated by the formula according to
AASHO:

V = 1.5 VDL +2.5 VLL

The load factors used for shear were higher


than those used for bending. Fig. 5: Truss analogy
It was assumed that the verticals were in tension and were represented by the stirrups and vertical
shear tendons. The compression forces were carried by the concrete. In addition to conducting
stress calculations, the stability of the web plates was also confirmed through analysis. In real-
ity, even though the bridge had a very long span, it was also very narrow, and the shear stresses
were not high.

Based on a simplified truss analogy with 45o inclined diagonals, the bottom chord can be repre-
sented by the bottom slab of the box girder. Thus, the design shear for the webs is

V* = D sin45o = V+ T sinα – B sinβ

Here, the slope of the top chord, α, is approximately equal to the longitudinal grade of the deck,
which is 6% near the main piers. To simplify, we can assume T = B so that the shear force in the
webs can be written as:
V* = V – B (sinβ – sinα)
8 CHAPTER 2. THE ORIGINAL DESIGN

Or, the shear stress in the webs is

v* = V* / (2 h t)

where V is the total shear force over the entire cross section, and B is the total compressive force
in the bottom slab, assumed to be the smallest value under the most unfavorable conditions; h is
the distance between the centres of gravity of B and T. And β is the slope of the bottom slab and
t is the thickness of the webs; 2t represents the two webs.

The compression force in the bottom slab, B, can have significant influence in the magnitude of
the shear stress in the webs. Under ultimate conditions, B is approximately equal to M/h, where
M is the bending moment, or

V* = V – (M/h) (sinβ – sinα)

And the shear stress in the webs is:

v* = [V – (M/h) (sinβ – sinα) ] / (2h t)

According to the original design notes, the maximum factored shear force at the cross section
next to the main pier is:

V* = 5930 kips ( 26.58 MN)

and thus,

v* = 59 ksf (2.83 MPa)

= 5.5 √ f’c or 0.082 f’c

There was no provision on maximum allowable shear stress in AASHO back in 1974. However,
these shear stress values would be allowed under the current AASHTO.

According to the above equation, the higher the bending moment is, the smaller the shear stress
in the webs will be. If, for any reason, the bending moment decreased significantly, the safety of
the webs might become critical. However, in the case of the Koror Bridge, the main span girder
is a pair of cantilevers; the bending moment and shear force are coupled with each other; so, this
type of analysis is safe.

Coming back to the truss analogy, the compressive stress in the diagonals can be calculated as
follows:

σD= [V – B (sinβ – sinα)] /(2h t)

Thus, the compressive stress, not the shear stress, in the webs of the main span in the Koror
Bridge calculated this way was about 130 ksf (6.0 MPa) compression, which is only 17% of
the cylinder strength of the specified concrete. Therefore, the webs, as designed, are safe. This
method of analysis was not a part of AASHO back in the 1970s. But AASHO allowed engineers
to use a rational method of design for long-span bridges.
STRUCTURAL SAFETY OF THE BRIDGE 9

Prestressing System
Several as-built drawings are shown in Fig. 6a–h. Figure 6e to h shows the prestressing system
of the Koror Bridge. The redesign and the prestressing material were both provided by the US
office of Dywidag (Dyckerhoff&Widmann AG). Beginning with the first segmental bridge built
in 1952 until around 1980, Dywidag had exclusively used their patented, threaded high-strength
bars as prestressing tendons. They are mainly 1.25 inch (32 mm) diameter bars with an ultimate
strength of 150 ksi (1050 MPa). It has full length rolled-on threads so that it can be cut at any
point and extended using couplers. Each bar is individually sheathed and anchored. These bars
came either in a standard 12 m length or they were cut to order.

Bar tendons have a large diameter; therefore they cannot be rolled on a reel for transportation,
as in the case of strands. Therefore, in segmental construction, these bars are usually coupled to
every second segment with couplers. The bars had to be placed before the concrete was poured.
They were stressed only after the entire length of each tendon was installed. The tendon sheath
is locally enlarged to accommodate the expected longitudinal movement of the couplers during
prestressing. The strength of each bar is relatively small. There are 310 bars, arranged in four
layers in the top slab of the pier table, sometimes called “segment 0”, which is located above
the main pier. The number of tendons decreases as the segments progress towards the mid-span.

According to AASHO, the maximum allowable stress for the tendons during stressing was 0.8
fu, after anchoring and before losses, 0.7 fu, and after losses, 0.6 fu, where fu is the ultimate
strength of the tendon. Bar tendons tend to be straight, and friction loss is usually small and
negligible. The prestress loss due to steel relaxation was assumed to be 3% according to infor-
mation from Dywidag.

In addition to the longitudinal tendons, there were transverse tendons in the top slab and verti-
cal shear tendons in the webs. Transverse tendons were designed based on the deck’s truck load
requirements.

Vertical tendons in the webs were designed to carry the shear force; 50% of the non-prestressed
reinforcement was assumed to contribute to the carrying shear.

Structural Safety of the Bridge


The AASHO standard specifications in 1974 specified the ultimate load condition as

Mu = (1.30 × MDL + 2.17*MLL + M Psec) / 0.90

where 0.90 was the strength factor. The value of Mu was included in the design documents [11].

The structural system of the main span of the original bridge was a statically determinate system
and there was no secondary bending moment due to prestressing.

The bridge’s safety is verifiable with a simple calculation. Take the section about 14 m from
the main pier; this was also the cross section where the collapse began. According to the design
drawings, there were 304 Dywidag bar tendons at this location, each with a nominal diameter
of 1.25 inch and with a cross-sectional area of 1.25 sq. inch. Thus, each tendon had an ultimate
10
(a)

CHAPTER 2. THE ORIGINAL DESIGN


Fig. 6a: Basic dimensions of the bridge, copy of original design plan
STRUCTURAL SAFETY OF THE BRIDGE
(b)

11
Fig. 6b: Basic dimensions of the bridge, copy of original design plan
12
(c)

CHAPTER 2. THE ORIGINAL DESIGN


Fig. 6c: Foundation details, copy of original design plan
STRUCTURAL SAFETY OF THE BRIDGE
(d)

13
Fig. 6d: Girder details, copy of original design plan
14
(e) part 1

(e) part 2

CHAPTER 2. THE ORIGINAL DESIGN


Fig. 6e: Prestressing tendon details
STRUCTURAL SAFETY OF THE BRIDGE
(f)

15
Fig. 6f: Prestressing tendon requirement, copy of original design plan
(g)
Fig. 6g: Prestressing system and tendon layout, copy of original design plan 16
CHAPTER 2. THE ORIGINAL DESIGN
STRUCTURAL SAFETY OF THE BRIDGE
(h)

17
Fig. 6h: Prestressing system and tendon layout, copy of original design plan
18 CHAPTER 2. THE ORIGINAL DESIGN

strength of 1.25 × 150 = 187.5 kips (85.2 kN). The steel area was 1.25 × 304 = 380 sq. inch.
The steel ratio was 0.0024 and the tendon stress at its ultimate condition was 144.6 ksi. Hence
the ultimate tension capacity of all these bar tendons in the top slab was 380 × 144.6 = 54,948
kips (244.43 MN).

The bottom slab of the box girder was 24 ft (7.32 m) wide, and the allowable ultimate concrete
stress was 0.85 f’c = 0.85 × 5 = 4.26 ksi = 612 ksf (30 MPa). So, under the ultimate condition,
the depth of compression stress block d would be 54,498 / (612 × 24) = 3.71 ft. The box girder
was 45.5 ft (13.87 m) deep, and the moment arm z was about 42.94 ft (13.09 m). Thus, the ulti-
mate moment capacity Muall would be 2,340,000 ft-kips (3199 MNm), which was about 23%
larger than the ultimate bending moment of 1,880,000 ft-kips (2605 MNm) [11]. That is, it is
23% more than that required by the specifications.

The actual un-factored bending moment at this location was 1,300,000 ft-kips (1800 MNm)
[11]. The actual safety factor against failure was 1.82, which was relatively high for a bridge
carrying predominantly dead load.

A comprehensive analysis was conducted during the design stage to confirm the ultimate safety
of every cross section of the bridge.

Creep and Shrinkage


In the 1970s, AASHO did not have provisions for creep and shrinkage for long-span bridges.
The creep and shrinkage calculation of the Koror Bridge was based on the CEB-FIP 1970
Recommendations prevailing at that time. The creep coefficient was

φ = ke kb kd ks kt

where ke represents the environment. Based on local reports, the average relative humidity was
82%, accordingly, ke = 1.85; kb considers the water:cement ratio and cement content, kb =
1.25; kd considers the average loading age, here assumed as 84 days, this results in kd = 0.76;
ks considers concrete member thickness, assuming 330 mm, and so ks = 0.74. Thus, the creep
coefficient φ = 1.30 kt.

Here kt is the time function expressing the progress of the creep deformation; kt increases from
0 at time zero to 1.0 at time equal to infinite. Thus, the ultimate creep coefficient is 1.30.The
graph from CEB-FIP 1970 [8] shows that loading age can have a significant effect on the value
of the creep coefficient (see Fig. 7). The creep coefficient can change from 1.30 to 2.60, if the
loading age is changed from 84 days to 5 days.

The definition of loading age is not the age of concrete when it is first under stress, if the load
varies with time. The stress in a cross section of a segmentally constructed bridge is an accu-
mulation of many stages of loads. Each of these loads has its own loading age. For example,
if a piece of concrete is loaded with two 50 kN loads, the first 50 kN on the 20th day after the
concrete is cast and the second 50 kN on the 30th day, then the loading age of the first load is
20 days and the age of the second is 30 days. The average loading age is usually assumed as
25 days.
CREEP AND SHRINKAGE 19

2
1.8
1.6
1.7
1.4 No
rm
1.4 al c
em
ent
1.1 1.0
ke

1 Hig
h ea
rly s
tren 0.7 0.75
gth
cem
ent 0.5 0.5

0.3

0
1 3 7 14 28 56 90 180 360 Days
Age of concrete at time of loading (T = 20°C = const.)

0 100 1000 10 000


Degree of hardening D

Fig. 7: Effect of loading age as per CEB 1970

In cantilever construction, the first segment, for example, is loaded both by its own weight and
by prestressing on the third to fifth day after casting. The construction of each subsequent seg-
ment adds a certain stress to it and each of these subsequent load increments has its own loading
age. The Koror Bridge cantilever has a total of 25 segments plus the hinge and the superimposed
load, for a total of 27 individual load stages. The first load was from the first segment’s own
weight on the fifth day. On average the construction of each segment took about one week.
When the last segment was cast and prestressed, the first segment was 25 weeks old. For the
first segment, the average loading age using a straight line average was 13.5 weeks. However,
the stress caused by the weight of the first segment was much less than that caused by the last
segment, when the cantilever was longer, such that a reasonable average loading age was two
third of 27 weeks, or 18 weeks, which is 126 days.

The last segment was loaded by its own weight, the weight of the hinge and the superimposed
dead load. So the average loading age, based on the same rationale, was about 20 days because
it takes longer to place the superimposed loads.

If the loading age of all segments in the cantilever is averaged, taking a straight average of 126
and 20, it would be 73 days. The effects of creep of different segments on the general creep
deformation of the bridge cantilever are different. The effect on the first segments near the pier
was certainly much larger than that of the last few segments near the mid-span. Therefore, we
adjusted the average loading age for the entire structure to 12 weeks, or 84 days.

Certainly, this cannot be considered a “scientific” calculation today, but in 1974, before the advent
of computer application in bridge engineering, this was a rational approximation. In any case, once
the loading age is in the 80 day range, kd does not vary much—10 to 20 days either way.
20 CHAPTER 2. THE ORIGINAL DESIGN

Prestress Loss
Prestress loss due to creep, shrinkage and steel relaxation was estimated using the following
formula proposed by Fritz Leonhardt [12]:

ε Es + nφ (σc,g + σc,p)
σp,L = ------------------------------- σs,o + σp,R
nσc,p(1+ρφ) – σs,o

where ε is the shrinkage coefficient, Es is the modulus of steel, n is the ratio of modulus of steel
to concrete, φ is the creep coefficient, σc,g is the concrete stress due to permanent load at centre
of gravity of prestressing steel, σc,pis the concrete stress due to prestressing at the same location,
ρ = 0.7, as suggested by Leonhardt, σs,o is stress in prestressing steel at time zero, and σp,R is relax-
ation of steel. Based on this formula with a creep coefficient of 1.30 and a shrinkage coefficient of
0.00017, and with the average concrete stress in the top slab under permanent load and prestress of
about 30 ksf (1.46 MPa), including a relaxation of 3%, the average prestress loss was about 10%.

Overturning Safety Factor


There were numerous discussions about the appropriate value to be used as the overturning safety
factor for the Koror Bridge design. In most three-span concrete girder bridges, the side spans are
usually only slightly longer than a half of the main span, so the overturning safety factor of the
girder with respect to the piers is usually only slightly larger than 1.00. Even though the Koror
Bridge has a hinge at the span centre, it is not really much different from a three-span girder bridge
because the bending capacity of the girder at the span centre is relatively insignificant compared
to the large overturning moment. The safety factor based on the girder’s overturning moment
with respect to the main piers was 1.30 with the ballast installed in the end spans. However, some
engineers recommended that the same safety factor should be used for the other structures, such
as a retaining wall. Consequently, vertical tie-downs were added at the side piers to increase the
overturning safety factor to 1.50 (see Fig. 8). Local stones were used for the ballast.

Prediction of Creep Deflection


Because concrete bridges deflect after completion due to creep, shrinkage and relaxation of
steel, a concrete bridge must be built with a camber. In effect, the bridge is built slightly higher

Ballast

Vertical Tiedowns

Fig. 8: Ballast and vertical tiedowns


CREEP AND SHRINKAGE 21

End pier Main pier Mid-Span Main pier End pier


(a) 2

1
Top fiber stress (MPa) After 10% PT loss
0

–1

–2

–3

–4

–5 Under full prestress


–6
–200 –160 –120 –80 –40 0 40 80 120 160 200
Distance from span center (m)

Nominal bottom fiber stress


(b) End pier Main pier Mid-Span Main pier End pier
2
Bottom fiber stress (MPa)

0
–2
–4
–6 Under full
prestress
–8
–10
–12
–14
After 10% PT loss
–200 –160 –120 –80 –40 0 40 80 120 160 200
Distance from span center (m)

Nominal bottom fiber stress


(c) End pier Main pier Mid-Span Main pier End pier
1,000
Deflection due to
full prestressing
500
Deflection in mm

0 –120 –80 –40 0 40 80 120 160 200


Distance from span center (m)
Deflection due to
–500
load and full prestressing
Deflection due to
dead load
–1000

Full prestressing = 70% Fu


–1500
–200 –160 –120 –80 –40 0 40 80 120 160 200
Distance from span center (m)

Fig. 9: Condition of bridge after full prestressing: (a) top fiber stress; (b) bottom fiber stress;
(c) deflection of girder
22 CHAPTER 2. THE ORIGINAL DESIGN

than the final design grade by an amount approximately equal to that of the calculated long-term
deflection. So, after a certain period of time, the bridge will deflect to the original design grade.
This upward shape of the girder is called camber.

The deflection due to permanent loading on the Koror Bridge was estimated to be 3.46 ft (1055
mm). Stressing the tendons to 70% of their ultimate strength, that is, 577 kN per tendon after
friction and seating losses, raised the bridge to 2.49 ft (761 mm) at the mid-span, making the
theoretical elastic deflection at the mid-span to be 0.96 ft (294 mm) at the time the bridge was
completed. After a 10% prestress loss, the elastic deflection was 1.21 ft (370 mm). Based on
a creep coefficient of 1.30, the creep deflection would be 1.30 × 1.21 = 1.58 ft (481 mm) (see
Fig. 9a–c).

The common wisdom about creep and shrinkage calculations was, and still is, that it is not an
accurate science and that estimation can vary ± 30%. If our error was really 30%, our creep
deflection would, in the worst case, be 1.58 ft × 0.3 = 0.47 ft (145 mm) off, which would have
been only 0.06% of the span length. This deflection error is acceptable for a 790 ft span bridge.

The development of the predicted deflection against time is shown in Fig. 10. The deflection
on day 1000 should have been 380 mm, or about 80% of the ultimate creep deflection. And the
deflection on day 10,000 should have been 480 mm—basically reaching the estimated ultimate
creep deflection.

According to field reports [3, 13], the actual elevation at the mid-span of the completed bridge
was 67.80 ft, which suggested that the actual camber was 0.69 ft (210 mm) compared to the
theoretical elevation of 67.11 ft at the mid-span.

Today, we are not able to determine if this was the anticipated camber provided to the field. If we
assume the loading age was day 84 and the bridge was actually completed on day 210 after start
of the cantilever construction, the bridge was hypothetically 126 days older when construction
was completed. From Fig. 10, it appears that 200 mm of creep deflection should have already
taken place within these 126 days. Thus, the 210 mm actual camber would be quite close to the
theoretical camber of 481 mm – 200 mm = 281 mm at the time of bridge completion. The devia-
tion would have been only 71 mm which is acceptable for a 241 m main span.

Time (Days)
1 10 100 1,000 10,000 100,000
0
Deflection at midspan (m)

0.10

0.20

0.30

0.40

0.50

Fig. 10: Predicted creep deflection


23

Chapter

3
Construction

The bridge was supported by 16 inch × 16 inch (410 mm × 410 mm) precast concrete piles about
100 ft (33 m) long each; there were 104 piles under each main pier and 21 piles under each end
pier, totaling 250 piles. The piles under the main piers were all battered at a 1:8 angle—with half
toward the water and the other half toward the shore. The design capacity (service condition)
of each pile was 400 kips (1.82 MN). There were 40 rejected piles which were left in place and
the replacement piles were driven next to them. Due to time constraints, 15 of the replacement
piles were steel H piles, H350 × 350 × 19 × 19, which has about the same design capacity as the
original precast piles. The total capacity of the pile group was not affected.

Two test piles were loaded to about 800 kips (3.6 MN) and, each had performed well up to that
load.

Except for stones and sand, all other construction materials and equipment had to be imported
because the bridge is located in a remote area. Therefore, saving construction materials and
avoiding the use of heavy equipment was important. The most expensive construction equip-
ment for the bridge was the form travelers. There was no steel fabricator in Palau so the form
travelers had to be imported, making them even more expensive (see Fig. 11).

The construction of the superstructure began in May 1976 at the Koror side. The pier-table,
which was the first segment on top of the main pier, was 37 ft (11.28 m) long, 19 ft (5.79 m)
towards the water and 18 ft (5.49 m) towards the land. The subsequent cantilever segments were
all 15 ft (4.57 m) long except the first was 14 ft (4.27 m) long and, the second last segment was
13 ft (3.96 m) long plus a 4 ft (1.22 m) long segment with a hinge and diaphragm.

In segmental cantilever construction, it is customary to build the two cantilevers, one on each
side of a pier, simultaneously in an approximately balanced manner; so two form travelers were
required for this pair of cantilevers. To save on the number of form travelers, the original design
was based on using two form travelers. Thus, one half of the bridge was designed to be built first
and then the same form travelers were to be used to build the other half of the bridge, as shown
in Fig. 12. However, after inspecting the site conditions, the construction sequence was modified
to build both sides simultaneously to shorten the construction schedule. In the side span area, the
bridge girder was very close to the ground. The contractor first filled the area with dead corals
dredged from the ocean and used relatively simple falsework on top of the fill to build the side
24 CHAPTER 3. CONSTRUCTION

Fig. 11: Area before bridge construction

spans. The main span cantilevers were constructed segmentally by the free cantilever method
using form travelers.

Construction of the side spans was arranged in such a way that the bending moment from the
main-span cantilever was always smaller than the bending moment caused by the weight of the
side-span cantilever, which was partially resting on a falsework. This was to reduce the unbal-
anced bending moment in the main pier (see Fig. 12). To achieve this, the segments of the girder
on the land side and the water side were each divided into five groups, designated as A, B, C, D
and E groups. The landside segment group had to be completed first, before the corresponding
segments on the river side could be built.

In segmental cantilever construction, the accumulated error in elevation is corrected or com-


pensated for every time a new segment is cast such that when the bridge is complete, the actual
elevation of the bridge and the design elevation of the bridge are very close. The design eleva-
tion of the bridge is the final grade plus the required camber. Figures 13 to 16 show the various
stages of construction.

Construction of the foundation began in July 1975. The construction of the superstructure began
in February 1976. The closure pour was completed in March 1977. The bridge was opened to
traffic in April 1977.

Up to 1976, the vast majority of segmental cantilever bridges were built by either
Dyckerhoff&Widmann or under their license. All these bridges used the patented Dywidag bar
tendons. The Koror Bridge was no exception. Bar tendons are very stiff. They cannot be coiled,
so they had to be transported inside containers and, the maximum length of the bars delivered
to the site was about 40 ft (12.20 m). In order to reduce waste, the tendons for the Koror Bridge
were cut to length, as required, when they were delivered to the job site. These prestressing
tendons were coupled every two segments so that they were mostly 30 ft (9.15 m) long. Each
tendon was individually placed inside a corrugated metal sheathing. At each coupling location,
the tendon sheathing was locally enlarged to allow movement of the coupler.
(a)

Fig. 12a: Construction schedule: original

25
26 CHAPTER 3. CONSTRUCTION

(b) 15.54 53.65 240.79 53.65 15.54


Falsework Form travelers construction Form travelers construction Falsework
16.00 18.28 11.56 15.84
5.02

0 0

5 5

10 10

15 15

20 20

25 Construction schedule 25

30 30

35 35

40 40

45 45

Fig. 12b: Construction schedule: final

(a) (b)

(c) (d)

Fig. 13: Pile foundation (a, b) pile cap; (c) forming pier table walls; (d) forming pier table
27

(a) (b)

(d)

(c)

(e) (f)

Fig. 14: Pier table construction 1: (a–c) Pier-table construction; (d) vertical tendons; (e)
reinforcement at tendon anchorage; (f) placing pier table concrete
28 CHAPTER 3. CONSTRUCTION

(a) (b)

(d)

(c)

(e) (f)

Fig. 15: Pier table construction 2: (a, b) Piertable and form traveler; (c-e) The side span area had
originally about 3’-5” (1.04 m) of water. It was filled with corals dredged from the channel and the
side-span segments were built using short local support (f) Manhole in the Pier-Table Diaphragm
29

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Fig. 16a: Cantilever construction 1; Segmental Cantilever Construction: Construction began


from both main piers and proceeded towards the span center. When the two cantilevers met
at the center, one of the travelers was dismantled first. The last segment was cast using the
remaining traveler
30 CHAPTER 3. CONSTRUCTION

(i) (j)

(k) (l)

Fig. 16b: Cantilever construction 2; Cast-in-place Segmental Cantilever Construction:


Construction began from both main piers and proceeded towards the span center. The bridge
was opened to traffic in June 1977

All tendons in each segment were placed before casting concrete. As the segments were 15 ft
(4.57 m) long and the tendons were 30 ft (9.14 m) long, typically half the tendons were protrud-
ing from the segment. They were supported by a frame in front of the form traveler.

Typically, the transverse tendons of a segment were stressed as soon as the concrete had attained
sufficient strength. However, the first transverse tendon at the front end of the segment was not
stressed at that time. It was stressed with the transverse tendons in the subsequent segment. The
vertical tendons were then stressed after the transverse tendons were stressed. The longitudinal
tendons were stressed last. After the longitudinal tendons were stressed, the form traveler was
moved forward for the construction of the next segment.

A typical segment took about one week to build. The first few segments took longer because
they were deeper and the workers were learning the trade. Tendons were typically stressed about
three days after the concrete was cast. The stressed tendons were grouted before the next seg-
ment was cast.

Construction of the Koror Bridge happened efficiently except that the bridge’s deflection was
larger than anticipated. At first, the deviation was not noticeable. It increased as the length of
the cantilever became longer. It was especially noticeable in the final half of the cantilever. As a
result, the final camber was slightly less than the design value.

The original camber calculations for the Koror Bridge construction were not found. As noted
before, the final built-in camber was 0.69 ft (210 mm).
FIELD CHANGE 31

Field Change
In the original design, there was an expansion joint between the box girder and the abutment
wall at both ends of the bridge (see Fig. 17a). A gap of 2 inch was to be provided at each joint
to allow for bridge movement due to temperature and other loads. A neoprene type joint seal
was specified for these joints. However, this detail was modified, for unknown reasons, during
construction. The as-built drawings show the end of the girder monolithically connected to the
abutment wall (see Fig. 17b). This change added an unknown factor in the analysis of the bridge
structure. However, since the end spans of the bridge were rather robust, the influence was not
significant.

Expansion joint seal


6%

2˝ = 51 mm
Gap @ med temperature

Backfill
18˝ = 457 mm

(a) Abutment joint as designed


6%

Backfill
18˝ = 457 mm

(a) Abutment joint as-built


Fig. 17: Joint Detail between Abutment Wall and End of Girder
33

Chapter

4
Bridge Performance

The Koror Bridge was a long-span concrete bridge; the design live load was relatively minor
compared to its dead load. Once the bridge was completed, bending moment due to live load
was only about 8% of the bending moment due to total load (Fig. 18). It did not cause much of
a stress variation in the bridge.

The elastic deflection at the span centre under total permanent load after 10% prestress loss
was 370 mm. When this is multiplied by the creep coefficient of 1.30, this resulted in a creep
deflection of 481 mm after completion of the bridge. Figure 19 shows the time function of creep
progress according to CEB-FIP 1970. It demonstrated that most creep deformation should have
been completed around 1000 days after load application, and the progression of creep should be
as shown in Fig. 10. However the reality is quite different from the calculation.

(a) (b)

(c)

Fig. 18: Koror Bridge and its surrounding landscape: (a) The completed Bridge; (b) Bridge
May 1986; (c) Bridge and the Ferry it replaced
34 CHAPTER 4. BRIDGE PERFORMANCE

1.0

CEB-FIP 1970
m
5c
=
em 10
Kt
0.5 20
40
80

0
0 10 100 1000 10000
Days

Fig. 19: Time Function for Creep, CEB 1970. Assumed Member Thickness = 33cm

A few years after opening to traffic, the bridge began to deflect much more than anticipated. In
1985, JICA (Japan International Corporation Agency) reported that the span centre of the bridge
had deflected 2.59 ft (850 mm). A second survey in 1990 showed that the bridge deflection had
reached 3.28 ft (1030 mm).

The JICA report [14], indicated that the bridge ”due to unknown reasons”, had deflected
excessively and was still increasing 10 years after its completion; such deflection was deemed
unprecedented. But the report confirmed that the bridge was structurally sound and posed no
safety concerns.

In 1996, the owner hired another consulting engineering firm, ABAM from the US to carry out a
further investigation of the bridge [15, 16]. By that time, the deflection of the bridge had reached
5.04 ft (1.54 m) at the span centre.

The calculated creep coefficient of 1.30 may appear low; a very similar value would be obtained
if it had been calculated according to the current AASHO, ACI or CEB-FIP recommenda-
tions, as we will discuss later. The measured deflection was 5.04 ft (1.54 m) after 18 years.
If the 5.04 ft (1.54 m) deflection was due only to creep, then the creep coefficient would have
been 1.54/0.481 = 3.20 after only 18 years
instead of the calculated ultimate creep
coefficient of 1.30. Deflection at Midspan (Includes 0.21m camber)
Time of Survey Deflection
The deflected shape of the bridge based on
the survey done by JICA and ABAM are July 1977 0.00 ft 0.00m
shown in Fig. 20a and listed in Table 1. (as built)
Their comparison to the theoretical deflec- Sept. 1993 −2.31 ft −0.70 m
tion is shown in Fig. 20b.
Nov. 1985 −3.50 ft −1.07 m
If the creep coefficient was underesti- May 1986 −3.69 ft −1.14 m
mated, the prestress loss would also have Jan. 1990 −4.04 ft −1.23 m
been underestimated and, this would have
led to an additional error in the prediction Sept. 1993 −5.04 ft −1.54 m
of the deformation. Table 1. Measured Deflections
35

Top to bottom
(a) Centerline of bridge July 1977, as built camber
Theoretical grade – zero
68 Sept. 1980, centerline
Nov. 1985
67
May 1986
66 Jan. 1990
July 1993 centerline
65
64
63
62
–300 –200 –100 0 100 200 300
Deflected shape of bridge girder
All dimensions in feet 1.0ft. = 0.305 m

(b) As built camber = 0.69'


0
–300 –200 –100 0 100 200 300
–1
Top to bottom
–2 July 1977, as built camber
Theoretical grade – zero
–3 Sept. 1980, centerline
Nov. 1985
–4
May 1986
–5 Jan. 1990
July 1993 centerline
Deflection of bridge girder
All dimensions in feet 1.0ft. = 0.305 m

Fig. 20: Measured deflection of the bridge: (a) deflected shape of girder; (b) deflection of girder

The report by ABAM confirmed that the bridge was structurally sound but it could not explain
the excessive deflection. It proposed a scheme to rectify the deflection. But the estimated cost
was over the budget, so the proposed repair scheme was not carried out immediately.

There were totally five surveys of the bridge elevation made by the two consulting engineering
firms, JICA and ABAM, hired by the owner to study the bridge.

The shape of the deflection curves in Fig. 20b looks very close to a straight line, which does not
appear rational. The deflection curve under the dead load of a cantilever is not a straight line;
so the deformed shape due to creep, which is the elastic deflection multiplied by the creep coef-
ficient, was not a straight line either.

Because no survey of the bridge elevation was available at the time the bridge was completed, the
deflection curves in Fig. 20b were based on the assumption that the bridge was built to theoretical
grade. This is not quite correct, because the actual long-term deflection of the bridge should have
been measured from the bridge elevation at the time of bridge completion. Unfortunately, no survey
of the final shape of the bridge was done at the time of its completion; this is only an assumption.

From the field documentation [3] we did find a survey record of a part of the cantilever soon
after its casting and prestressing of the last box segment, Segment E6, before casting the hinge
36 CHAPTER 4. BRIDGE PERFORMANCE

segment. This record shows the deviation of the cantilever geometry from the theoretical cam-
ber. This deviation is plotted in Fig. 21a. The curve shows that the geometry of the bridge at
this time was lower than required. The concave shape of the curve indicates that the deflection
of the bridge had been more than anticipated by the theoretical analysis for many segments and
the field was trying to compensate for this excessive deflection by setting the form higher for
the subsequent segments.

This curve represents the geometry of the bridge before the hinge segment was cast. Since no
further survey data was found, it is unknown if there was any additional deviation caused by
the casting of the hinge segment and the overlay. We also do not know if the overlay had com-
pensated for part or all of the deviations. At this moment, the best we can do is to assume that
the last two operations did not cause additional deviations nor did they change the shape of the
deviation. Adding this deviation to the last surveyed deflection data of January 1990, produces
the curve in Fig. 21b which appears more rational.

The plots shown in Fig. 22 demonstrate that based on field measurements the deflection does not
appear to abate in the near future. In a report in 1993, ABAM predicted that the bridge would
continue to deflect for another 1 m over the next 85 years.

Distance from center of span (feet)


(a)
Dots represent survey data
0 50 100 150 200
Deviation (feet)

0
–0.1
–0.2
–0.3
–0.4

Distance from center of span (feet)


(b) 0 50 100 150 200
0

–0.5

–1.0
Deflection (feet)

With construction deviations


–1.5

–2.0
Without construction deviations
–2.5

–3.0

–3.5

–4.0

Fig. 21: Construction deviation: (a) deviations in camber before hinge construction; (b)
deflected shape plus deviation
TEST RESULTS 37

Time (days)
(a) 1 5,000 10,000 15,000 20,000 25,000
0

The deflection of the bridge (m) –0.20

–0.40

–0.60
Originally estimated deflection
–0.80

–1.00
Actual deflection
–1.20

–1.40

–1.60

–1.80

Time (days)
(b) 1 10 100 1,000 10,000 100,000
0

–0.20
Deflection at midspan (m)

–0.40
Originally estimated deflection
–0.60

–0.80

–1.00
Actual deflection
–1.20

–1.40

–1.60

–1.80

Fig. 22: Predicted and actually measured deflection

Test Results
Several tests were performed before the bridge repair in 1996

1. Tests showed that the cylinder strength of concrete was rather close to the design value,
5000 psi (35 MPa) [14, 17]; so the strength of the concrete was acceptable.
2. There were two tests to determine the modulus of elasticity of the concrete: the test results
were 22.1 GPa and 21.7 GPa, averaging 21.9 GPa. But that was the modulus when the con-
crete was 13 years old. According to CEB 90, the modulus of elasticity of a concrete mem-
ber of age 13 years should be about 12% higher than the 28 day modulus. Thus, the actual
28 day modulus of the concrete of the Koror Bridge should have been 21.9/1.12 = 19.6 GPa,
38 CHAPTER 4. BRIDGE PERFORMANCE

which would be about 31% lower than the AASHO value of 28.3 GPa assumed in the
original design.
3. Before the repair commenced, nine tests were performed on three tendons—three tests on
each tendon. The test results found that the average prestress loss was about 50%, which
was much higher than anticipated.

During original construction, the tendons were overstressed and anchored to predetermined
values so that the average stress in the tendons was about 70%.

The calculated prestress loss in most cantilever bridges is usually between 10% and 20%
depending mostly on the magnitude of the compressive stress in the concrete. The prestress
loss calculated for the Koror Bridge was only about 10%. It was relatively low because the top
slab stress in the Koror Bridge was close to zero under permanent load and prestressing. As a
consequence, the loss could be mainly due to shrinkage of the concrete plus the relaxation of the
prestressing steel according to the formula by Leonhardt [12].

Even though the prestress loss was uncommonly high, it did not affect the safety of the bridge.
However such high prestress losses definitely influence the deflection of the bridge.

The design cylinder strength, f’c, of the concrete of the Koror Bridge was 5000 psi (35 MPa),
and the modulus of rupture (cracking strength) was assumed as 530 psi (3.70 MPa) (7.5 √f’c,
f’c in psi).

Figure 23a shows that the top fibre of the main-span cantilever should have been in tension when
the prestress loss approached about 20%. This top-fibre stress should reach cracking strength
when the prestress loss approaches about 40%. Tests showed that the actual prestress loss before
the repair was 50%. In essence, the top slab must have cracked. However, because there are
many grouted prestressing tendons in the top slab, they would have reduced crack spacing such
that the cracks might not have been visible.

8.00
Centerline pier
6.00 50% Prestress loss
Top fiber stress (MPa)

Modulus of rupture = 3.7 MPa


4.00 40%
2.00 30%
0 20%

–2.00 10%

–4.00 0%

–6.00
–120 –110 –100 –90 –80 –70 –60 –50 –40 –30 –20 –10 0
Distance from span center (m)
Nominal top fiber stress under various values of prestress loss

Fig. 23a: Top fiber stress under different degrees of pressure loss
TEST RESULTS 39

End pier Main pier Mid-span Main pier End pier


8.00

6.00 50%
Top fiber stress (MPa)
4.00 40%

2.00 30%

0 20%

–2.00 10%

–4.00 0%

–6.00
–200 –160 –120 –80 –40 0 40 80 120 160 200
Distance from span center (m)
Nominal top fiber stress under various values of prestress loss

End pier Main pier Mid-span Main pier End pier


2
0
Bottom fiber stress (MPa)

Prestress loss:
–2
(from top down)
–4 0%, 10%, 20%,
–6 30%, 40%, 50%
–8
–10
–12
–14
–16
–200 –160 –120 –80 –40 0 40 80 120 160 200
Distance from span center (m)
Nominal bottom fiber stress under various values of prestress loss

End pier Main pier Mid-span Main pier End pier


100
Deflected shape bridge (mm)

0
–100
–200
–300
–400 Prestress loss:
–500 (from top down)
0%, 10%, 20%,
–600 30%, 40%, 50%
–700
–800
–200 –160 –120 –80 –40 0 40 80 120 160 200
Distance from span center (m)
Nominal bottom fiber stress under various values of prestress loss

Fig. 23b: Effect of prestress loss


41

Chapter

5
The Repair

Repair Scheme
VSL, a prestressing material supplier, through a general contractor, Black Micro Construction
Co., proposed an alternative repair scheme [18], for a lower price and was accepted by the
owner. Figure 24 shows the shape of the bridge before the repair. This scheme significantly
changed the bridge structure in three ways:

1. At the centre hinge, jacks were used to push the two cantilevers against each other at the top
slab level in several steps. Four 1000 t jacks introduced a large longitudinal force into the
structure (Fig. 25a). The repair design plans specified the force to be 4400 kips (18 MN) if
the foundation was stiff and 6000 kips (27 MN) if the foundation was soft. Site records are
presently not available, but the soil was quite soft, and it appears most likely that a 6000

Fig. 24: Condition of bridge before repair (1996)


42 CHAPTER 5. THE REPAIR

kip force was actually applied. The hinge was then frozen in this position by filling the gap
with concrete, thus making the bridge continuous. Neither JICA nor ABAM recommended
making the bridge continuous by filling up the centre hinge. But both the owner and the
engineer finally agreed to this scheme and ABAM was contracted to provide the design for
its implementation [15, 16].
2. After the bridge was made continuous, eight external tendons were installed inside the box
girder, each with 31, 0.5 inch diameter 270 ksi seven wire strands, running from one end
of the bridge to the other end, draped at the mid-span (see Fig. 25b). They were stressed
to 75% of the ultimate strength of the tendons, which amounted to a total of 7640 kips
(34.72 MN) of prestressing force.
3. The grey concrete wearing surface was replaced with a black asphalt wearing surface.

The repair work began in April 1996 and was completed in July 1996. Less than three months
after the repair, on 26 September 1996 around 5:35 pm, on a relatively calm afternoon, the
bridge collapsed [18].

Before the mid-span hinge was made monolithic, the main span of the bridge comprised two
independent cantilevers (see Fig. 26a). The horizontal jacking force at the hinge was fully
resisted by the main piers irrespective of the stiffness or softness of the soil, which meant that
the horizontal pier reaction was equal to the jacking force at the hinge. Thus the stress distribu-
tion in the main span girder could be easily determined. The top fibre stress near the main pier
in the main span due to the 27 MN jacking force was –4.54 MPa.

53.65 240.79/2 240.79/2 53.65


20.73 32.92 71.78 32.00 16.61 32.00 71.78 32.92 20.73
16.61

Note:
all dimens in meters.
1m = 3.2808 ft

53.65 240.79/2 240.79/2 53.65


20.73 32.92 71.78 32.00 16.61 32.00 71.78 32.92 20.73
16.61
strand tendons strand tendons
bar tendons bar tendons

Fig. 25: Repair scheme Jacking at Hinge and Adding External Tendons (step 1, top): 6000 kips
(27 MN) Jacking Force at the Hinge; (step 2, bottom): Applied external tendons (34.72 MN
force), inside the box girder
REPAIR SCHEME 43

After the bridge was made continuous, the main span became one degree statically indeterminate.
In this configuration, the horizontal stiffness of the main piers would have a significant effect on
the magnitude of the stress in the bridge girder generated by the external prestressing. Therefore,
when the 34.72 MN external prestressing force was applied to the bridge, the stress in the girder
could only be calculated with an assumed soil stiffness supporting the main pier foundations.
If the soil was infinitely stiff, the structure could be represented by the system b in Fig. 26b, in
which the horizontal reaction of the main pier was equal to the jacking force of 34.72 MN, and
the top fibre stress near the main piers was about –2.00 MPa. If the soil was extremely soft, the
structure could be represented by the system c in Fig. 26, and there would be no horizontal reac-
tion at the main piers, and the top fibre stress at the same location was about –4.0 MPa. The real
stress should have been somewhere in between. Thus, the total top fibre stress near the main piers
would have been between–6.54 MPa and –8.54 MPa due to these two repair procedures.

From another point of view, if the main pier foundation was infinitely rigid, the horizontal
reaction at the main piers under the jacking force would be 27 MN towards the water, and the
horizontal reaction under the external prestressing would be 34.72 MN towards the land. These
two reactions acted against each other so the resulting reaction would have been 34.72 – 27 =
7.72 MN towards the land which was relatively small.

If the soil was very soft, the initial horizontal reaction at the main piers would be 27 MN under the
jacking force and close to zero under the external prestressing; so the resulting pier reaction would
be about 27 MN, acting towards the land. If the soil was soft, it would probably creep under the
large pier reactions. In the end, maybe 10 years later, this reaction would be significantly reduced.

An analysis was carried out based on assumed soil characteristics of regular sand beaches. The
soil provided an elastic support to the main pier foundations. The result was about half way
between the two extremes.

53.65 240.79 53.65

(a)
A B C D

(b)

(c)

Fig. 26: Possible Boundary Conditions (units: m)


44 CHAPTER 5. THE REPAIR

Almost all previous investigators concluded that temperature did not contribute to the collapse
based on the fact that the temperature variation in the area was relatively minor. However, tem-
perature did cause additional compressive stress in the top slab. Temperature, by itself would not
be able to cause the collapse. But it might have been the last straw that broke the camel’s back.

In this area, temperatures vary approximately from 23oC to 32oC. Depending on the temperature
when the bridge was made continuous, the variance in uniform temperature of the structure
could have been any value between −9 and +9 degrees. For a 9oC uniform temperature rise, the
compression in the top slab near the main pier would have to be about –1.28 MPa.

The collapse occurred around 5:35 pm in the afternoon. The deck had been absorbing heat from
the sun for the entire day and its temperature was now at a maximum. Moreover, the repair
changed the pavement from a light-colored concrete to black asphalt, which absorbed heat much
faster. It is common knowledge that the temperature of a deck with a blacktop can rise locally
as much as 40oC higher than the rest of the structure. For a 40oC non-uniform temperature rise
in the deck slab, the top fibre stress in the girder near the main piers would have been about
–1.04 MPa.

The temperature effect was calculated based on a structural system with elastic support to the
main piers as described above.

Thus, adding all these stresses together, the total top fibre stress near the main piers could have
between –8.86 MPa to –10.86 MPa, which was rather significant.

1. 6000kips (27 MN) jacking force at top of center hinge −4.54 MPa
2. Continuous external tendons (34.72 MN) −2.00 Mpa to −4.0 MPa
3. Uniform temperature rise, assume 9°C −1.00 MPa
4. Non-uniform temperature rise, assume 40°C −1.04 MPa
Total −8.58 MPa to −10.86 MPa
Table 2. Additional Stresses in Top Slab About 14m From Main Pier Caused by the Repair
(1 Mpa = 0.145 ksi)
45

Chapter

6
The Collapse

The actual repair work began in April 1996 and was completed in July 1996. Less than three
months after the repair, on 26 September 1996 around 5:35pm, on a relatively calm afternoon,
the bridge collapsed, and a portion of the bridge fell into the water.

Witnesses described that before the collapse, there were sounds of concrete cracking and
steel rubbing against each other for about 30 minutes, then the top deck of the girder near the
Babeldaob main pier suddenly crumbled and separated from the webs while the end spans on
the Koror side rose and fell back down, and finally, the bridge span fell into the water.

The Koror Bridge was the world’s longest span at the time, so many engineers were interested
in the cause of its collapse. Several hypotheses were presented in various articles [19–23], but
none of them could satisfactorily pinpoint the reason for the failure. The task of understanding
what went wrong was further hindered by the sealing of almost all documents due to settlements
from various litigations. It was not until 2009, through the effort of a group of engineers, that
the documents were unsealed and made accessible to researchers. Unfortunately, too much time
had passed and specific evidence from the site had been removed. So, further testing was no
longer possible.

Forensic Observations
Figures 26 to 33 show the condition of the bridge after the collapse. Several clues tell the story
of the bridge’s failure [19]:

1. The deck of the Babeldaob cantilever showed signs of crushing, due to either buckling or
crushing of the concrete deck. The triangular cracking pattern in the deck is typical of con-
crete plate failure under high compression. The deck was also separated from the webs at
several locations.
2. The bottom slab of the Babeldaob cantilever near the main pier had become crushed and
dislocated downward—a sign of failure due to shear and compression. The webs between
segment 0 and segment 1 were partially dislocated and the two sides pushed past each other
(i.e., overlapping).
46 CHAPTER 6. THE COLLAPSE

Fig. 27: Koror Bridge after collapse, at 5.35 pm on Sept. 26, 1996

3. The bottom slab of the Koror cantilever near the main pier had been crushed but with
minimum vertical dislocation. This was a sign of predominant compression failure due to
bending.
4. The side span girder over the end pier at the Koror side showed signs of punching shear
failure at both sides above the end support. This type of failure, with vertical shear failure
planes at both sides of the support, indicated that the end span was lifted upward and then
fell down on the support. The portion of the end span beyond the support was too small to
cause any type of shear failure.
5. All vertical tie-downs at the Koror side pier were broken. The girder must have lifted quite
high above the side pier.
6. The Koror cantilever girder deck and bottom slab were intact except at the bottom slab of
the main pier where it was found crushed.
7. The box girder of the Babeldaob cantilever cracked from the bottom upwards, indicating
excessive positive bending moment.
8. Some deviators that had been added during the repairs were ripped off from the top slab.
9. The time of the collapse was at about 5:35 pm, shortly before sunset, when the bridge tem-
perature was hottest.
10. Corrosion was not found to be a problem. The tendons exposed after the collapse showed
absolutely no signs of corrosion.
11. There was no slippage at the construction joint between any two adjacent segments, except
the punching shear deformation at the Koror end span over the side pier as mentioned in
point 4 above.
12. Cracks perpendicular to the bridge axis were found in the ground surface near the Koror
main pier. There was no damage to the pier itself.
Figure 34 shows the condition of the bridge after the collapse from both sides.
FORENSIC OBSERVATIONS 47

Fig. 28a: Condition of box girder top slab on Babeldaob side after collapse
48 CHAPTER 6. THE COLLAPSE

Fig. 28b: Condition of box girder top slab on Babeldaob side after collapse
FORENSIC OBSERVATIONS 49

(a)

(b) (c)

Fig. 29: Condition of the pier table (a) Condition near the Babeldaob (North end) main pier
(b) looking from the West (c) Looking from the East
50 CHAPTER 6. THE COLLAPSE

(a)

(b) (c)

(d)

(e)

Fig. 30: (a–e) Condition of the Koror side cantilever (at the Koror side the main failure
occurred when the bottom slab was crushed)
FORENSIC OBSERVATIONS 51

(a)

(b)

(c)

(d)

Fig. 31: (a–d) Delamination of Koror side top slab concrete (the breaking surface between the
tendon ducts and the concrete was clean, this was an indication the bond between them was weak)
52 CHAPTER 6. THE COLLAPSE

(a)

(b)

(c)

(d)

Fig. 32: (a) Condition near the Koror (South end) main pier (b–d) Ground surface cracks
were visible near the Koror main pier after bridge collapse
FORENSIC OBSERVATIONS 53

(a)

(b) (c)

(d)

(e)

Fig. 33: (a–e) Punching shear failure at the Koror side end span (at the Koror side, punching
shear failure of the main girder occurred at both sides of the end pier)
54 CHAPTER 6. THE COLLAPSE

(a)
Side span little damage Web crushed, overlapped

Top slab buckled or crushed

8
9

12
14

17
Bottom slab crushed &
dislocated

23
25

30
Babeldaob side

33
(b) Top slab intact Double punching shear
failure at end pier
5
10
15
20
25

EB End pier
26

Crushed bottom slab


29

Koror side

Fig. 34: Condition of the collapsed bridge (a) Babeldaob side (b) Koror side

Failure Mechanism
The failure began in the top slab near the Babeldaob main pier, where the slab might have
buckled or been crushed under high compression (see Fig. 35a). When the top slab failed, the
compression originally carried by the top slab concrete was transferred to the upper portion of
the web and caused the webs to either buckle or crush in a longitudinal direction.

Failure of the web plates caused the two parts of the webs to move against each other. This
released the prestressing force in all internal and external longitudinal tendons at this location
(see Fig. 35b). This destroyed some deviators and increased shear force in the bottom slab.

After the webs failed, the entire shear force had to be carried by the bottom slab alone which was
not possible. Thus, the bottom slab failed under shear which explains the vertical displacement
at this location (see Fig. 35c).

The girder here was then only supported by the bottom slab, thus changing from a monolithic
connection to a simply supported condition with very little bending moment. The other end
of the Babeldaob cantilever was supported by the mid-span monolithic connection. Thus, the
Babeldaob cantilever became a simply supported beam. A large portion of the weight of this
half of the bridge was now shifted to the Koror cantilever through the monolithic connection
FAILURE MECHANISM 55

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 35: Sequence of failure


56 CHAPTER 6. THE COLLAPSE

at the mid-span. Because the bending moment of the girder near Babeldaob was fully released,
the bending moment at the Koror side increased by about 100%. By adding the dynamic effect
of this sudden failure, the actual bending moment in the girder at the Koror main pier became
much higher.

The design safety factor for overturning at the main pier was 1.50; so when the bending moment
in the girder on the water side at the Koror main pier increased by more than 100%, it was higher
than the 1.50 safety factor. Thus, the Koror cantilever rotated around the pier and lifted the end
span up into the air (see Fig. 35d).

The drastic increase in bending moment in the girder at the Koror main pier was higher than the
ultimate capacity of the girder cross section at this location, so it failed. The bottom slab near
the Koror main pier failed first (see Fig. 35e), because it was weaker than the top slab. With the
added external tendons, the top slab had a greater capacity to resist higher bending moment,
while the bottom slab was not designed to withstand such an excessive bending moment. This
explains why the Koror cantilever, especially the top slab, remained mostly intact.

When the Koror side girder failed near the main pier, and especially when the middle portion of
the girder fell into the water, it released a large part of the bending moment in the girder so that
the end span fell back to the ground and hit the end pier. This caused the double punching shear
failure planes in the girder on both sides of the pier (see Fig. 35f).

After becoming a simple beam, the stress pattern changed in the Babeldaob cantilever. The
stresses in the bottom slab were now under tension instead of compression as originally designed.
This tensile stress cracked the bottom slab and the lower portion of the webs. The bottom slab
was further damaged when it fell into the water.

When the Babeldaob cantilever fell into the water, it pulled the Koror cantilever with it. The
Koror cantilever did not sustain additional damage.

Questions Raised after the Collapse


There are several obvious questions after the collapse of the bridge:

1. What would have happened if no repairs to the bridge were made?


2. How did the repair fundamentally change the structure?
3. Why was the deflection so much larger than anticipated?
4. Why did the top slab of the Babeldaob cantilever near the main pier crush and initiate the
collapse?

We will study and analyze these topics in the following chapters.


57

Chapter

7
What Would Have Happened
If No Repairs To The Bridge
Had Been Made?

This is certainly a hypothetical question, because the repair was done, and the bridge collapsed.
Nevertheless, it is of interest to know what would have happened if no repairs had been made to
the bridge. There are two aspects to this question:

1. Would the bridge have been safe to use?


2. What would the geometry of the bridge look like today?

The safety of the bridge had been confirmed by the two consultants hired by the owner to assess
the condition of the bridge, JICA of Japan and ABAM of the US. They both determined that
the bridge was structurally safe. The analysis in Chapter 2 of this book also confirmed that the
ultimate capacity of the bridge girder was much higher than the actual demand. Thus the change
in the geometry would not have affected the ultimate capacity of the bridge. Consequently, we
can conclude that the bridge would have continued to be safe, if no repairs had been performed
on the bridge.

Time (days after bridge completion)


1 10 100 1000 10000 100000
0
Deflection at midspan hinge
after bridge completion (m)

–0.5
36,500 days

–1.0

–1.5

–2.0

Deflection = –2.40 m
–2.5

Fig. 36: Extrapolation of deflection


58 CHAPTER 7. WHAT WOULD HAVE HAPPENED . . .

The geometry of the bridge in the future would be difficult to predict. ABAM suggested that the
bridge might deflect another 1 m in the ensuing 85 years. Adding this to the deflection at that
time of 1.54 m, the total deflection 85 years later would be 2.54 m [16]. However, as the analysis
in the next chapter of this book shows, creep prediction models available in codes and specifi-
cations today still do not offer sufficient reliability in the estimation of long-term deformation
of long-span bridges. Nevertheless, it would be interesting to see what the worst-case scenario
would be.

There were a total of five survey results after the bridge was completed (see Fig. 20). The
creep model in all specifications today assumes that, in a semi-log scale, the time function of
creep converges to an asymptote, which represents the ultimate creep deflection of the bridge.
Our problem is that we do not know where this asymptote is. However, if we connect these
five survey points with a straight line and extend it to the time of 100 years (i.e. 36,500 days;
see Fig. 36), such an extrapolation should represent the worst scenario. The corresponding
deflection on day 36,500 is –2.40 m. Deducting the measured deflection in September, 1993 of
–1.54 m, one would anticipate the bridge to deflect another –0.86 m in 100 years.
59

Chapter

8
Why Was The Deflection of The Bridge
So Much Larger Than Anticipated?

Calculation of Creep and Shrinkage


To date, several hundred long-span prestressed concrete bridges have been built worldwide.
While there has been extensive research conducted on creep and shrinkage of concrete bridges,
it is still difficult to correctly predict their plastic deformation. The problem is that we have vari-
ous calculation models from different specifications and codes that do not match each other, as
indicated by Robertson [24] and others. Even the more popular CEB-FIP Model Code has been
revised several times, and each time the revision results in different numerical values in creep
and shrinkage predictions. This poses an on-going dilemma for practicing engineers.

Fortunately, inaccuracies in estimating creep and shrinkage deformation do not affect the safety
of structures. It is basically a geometry problem, but the unpredictable deformation does create
psychological and traffic disturbances. It is also not easy to fix!

The design of the Koror Bridge began in 1974, or about 40 years ago. Since then, the industry
has extensively researched creep and shrinkage of concrete bridges. If we had designed the
Koror Bridge according to the current specifications, what would we have been done differently?

Current AASHTO/LRFD Model


The creep coefficient according to the current AASHTO/LRFD model is

φ = 3.5 kf khc kld kc ktd

where

kf considers the concrete strength,


kf = 1/(0.67 + fc′/9) = 0.816 for f ′c = 5.0 ksi
khc considers the relative humidity,
khc = 1.58 – RH/120 = 0.90 for RH = 82%
kc considers the thickness of the concrete member,
60 CHAPTER 8. DEFLECTION OF THE BRIDGE

kc = {[45 + (t – to)] /[26 * e0.36V/S + (t – to)]} *


{(1.80 – 1.77 * e−0.54V/S)/2.587}
kc = 0.667 for V/S = 6.5″ inch and t = ∞,
ktd is the time function after load application,
ktd = (t – to)0.6 / [10 + (t – to)0.6]
ktd = 1.0 for t = ∞
kld considers the age of load application,
kld = ti−0.118 = 0.592 for ti = 84 days.

Thus, the ultimate creep coefficient would be

φ = 3.5 kfkhckldkcktd
= 3.50 * 0.816 * 0.9 * 0.592 * 0.667 * 1
= 1.52 * 0.667 = 1.02

This is even smaller than the 1.30 calculated in 1974, meaning that engineers would have
designed the bridge in much the same way if they had used today’s AASHTO specifications.
Indeed, the large deflection of the bridge would still not have been explainable.

One point of contention might still be the loading age. If we had used a loading age of seven
days instead of 84 days, kld would have been 0.795 instead of 0.592 and φ would have been 1.38.
Assuming a loading age of seven days is not correct, even φ = 1.38 cannot explain the actual
deflection of the bridge.

The 1990 CEB-FIP Model


According to the current CEB-FIP 90-99 model,
φ = φo βc(t,to)
where φo = φRH β(fcm28) β(to)
φRH = [ 1+ (1 – H/100) * α1 / (0.16 * (hc/4)1/3] * α2
hc = 2A/u
here (A/u) is the same as V/S in AASHTO.
The design cylinder strength of concrete is 35 MPa, and the cube strength should be about
45 MPa, so α1 = 0.84, α2 = 0.95, α3 = 0.88.
β(fcm28) = 5.3 / [(fcm28)/10]1/2 = 2.5
to = 84, β(to) = 1/(0.1 + to0.2) = 0.40
βc(t,to) = [(t – to)/βH + t – to]0.3
βH = 1.5 hc [1 + (0.012H)18] + 250α3 ≤ 1500α3
with hc = 330 mm, H = 82%
φRH = 1.14, we get
φo = φRH β(fcm28) β(to) = 1.14.
At time infinitive, βc = 1.0 hence φ∞ = 1.14.
DISCUSSION OF CREEP ESTIMATE 61

Adding the effect of temperature, assumed to be 28oC in average, the ultimate creep coefficient
will be φ∞ = 1.30, which is the same as what we had assumed in the original design.

For this discussion, we may add a possible deviation of 30% to this value as the upper bound
based on the prevailing common wisdom. Then we would get φ∞ = 1.69. This still cannot
explain the large deflection actually measured.

Long-span bridges are usually designed for a 100 year service life. As designers, we should try
to understand what can happen to the shape of a bridge during its designated design life. Con-
crete is different from steel in this respect. Generally, the shape of a steel bridge will remain as
built while concrete bridges will actually continue to change for many years, caused by creep,
shrinkage of concrete and relaxation of steel tendons. We must strive to sufficiently understand
the character and magnitude of these concrete properties. Both creep and shrinkage are physical
properties of concrete and there is no scientific way of calculating their value except through
experimentation. Unfortunately, there are many factors that can influence creep and shrinkage
values and it is impossible to accurately determine the effects of each of these factors. Tests for
real long-term effects of large, thick concrete members, for example, may last more than one
generation and are very difficult and expensive to carry out.

Professor Zdenek Bazant of Northwestern University, USA, has collected deflection data from
more than 60 bridges worldwide. Examination of this data has yielded one simple answer: prac-
tically every one of these bridges deflected much more than anticipated in the calculation model
in the design specifications [25–27]. Moreover, contrary to the common belief that creep will
be substantially complete within about 10 years, creep in most bridges appears to continue even
after 20 to 30 years.

Currently, there are several code-recommended calculation models for the estimation of creep
and shrinkage of concrete bridges: mainly the CEB-FIP in Europe, the AASHTO and the ACI in
the US. In addition, there are the B3, and the GL. These are not part of any bridge specification
in the US or Europe. Different models will come up with surprisingly different results. At the
moment, it is hard to decide which one should be followed!

Discussion of Creep Estimate


AASHTO, ACI and CEB-FIP1990 all interpret creep and shrinkage as follows:

εcs(t,ts) = εcso * β(t – ts) shrinkage


φ (t, to) = φ o * β(t – to) creep

In these equations, the first factor on the right hand side is the nominal shrinkage coefficient and
nominal creep coefficient which basically determines the magnitude. The second factor describes
the development of shrinkage and creep with time. Generally speaking, if the deflection is higher
than anticipated, it is because we underestimated the first factor; if the progression of the creep
and shrinkage is different from what we have anticipated, the problem is the second factor.

Actual surveys at the Koror site show that the deflection was much greater than anyone had
anticipated; therefore the first factor was incorrectly computed (Fig. 41). Furthermore, after 16
years, the deflection had not slowed down, which is contrary to what the specifications indicate,
so the second factor also is not correct.
62 CHAPTER 8. DEFLECTION OF THE BRIDGE

The fib 2010 Model Code (Final Draft 2014)


After the merger of CEB and FIP to form fib, the current creep model, fib 2010 (final draft,
2014) is slightly different from the CEB-FIP 1990-99 Model. The total creep is divided into two
components: a basic creep, φbc and a drying creep, φdc.

A calculation using the same parameters:

fcm = 45 Mpa,
h = 330 mm,
RH =80%,
to = 84 days, and
t = 100 years = 18,250 days,

shows that the total creep, φ is equal to 1.31 after 100 years. If the loading age is assumed to be
7 days, the total creep, φ is equal to 2.05 after 100 years. These values are not much different
from the values calculated according to CEB-FIP 1990 99 Model Code.

Raising the temperature will increase the adjusted age of the concrete, to and t, but the effect on
the final value of creep is not significant.

The Time Function


We will study the second factor of the creep formula, β, first.

In the ACI model, the time function, β, contains only time as a variable:

β = (t–to)0.6 / [10 + (t–to)0.6]

The value of β is plotted in Fig. 37. It shows that at about 1000 days after loading, over 80% of
the total creep should have taken place. This is significantly different from the survey data from
the Koror Bridge. The actual sur-
vey data shows that the deflection
Time (days) after 1000 days was –0.67 m, and
1 10 100 1000 10000 100000 that the deflection after 5600 days
0
was –1.33 m, which was twice as
0.2 much as the 1000 day deflection.
TIme function β

ACI 209 The value computed by the ACI


0.4 formula does not match the survey
data from the Koror Bridge. Hence,
0.6
the ACI model cannot fit the Koror
0.8 survey data.

1.0 The CEB-FIP 90 or 1990-99 model


has a similar formula for the time
1.2 function:
Fig. 37: ACI time function for creep β = [(t–to) / (βH + (t–to)]0.3
THE TIME FUNCTION 63

Where βH considers the environmental conditions,


βH = 1.5 hc [1 + (0.012H)18] +250 α3
≤ 1500 α3
In CEB-FIP 90, α3 = 1.0, in CEB-FIP 90-99, α3 = [35/fcm28]0.5, fcm28 is the 28-day cube strength of
concrete, when fcm28 = 35 MPa, α3 = 1, when the concrete strength is higher than 35 MPa, α3 < 1.
Nowadays, concrete strength for long-span bridges is seldom less than 35 MPa, hence, α3 ≤ 1,
and we can simply assume βH ≤ 1500.
Figure 38 shows the plot of time Time (days)
function β, the upper curve for 1 10 100 1000 10000 100000
βH = 1500, and the second curve is 0
for βH = 1000. The ACI curve is also CEP-FIP
0.2
included for comparison purposes. βH = 1500 top curve
TIme function β

0.4 βH = 1000 middle curve


To compare the ACI and CEB-
FIP models with the actual survey 0.6
ACI 209
results, we can assume that the 0.8
creep deflection of the bridge at
100,000 days was 2.20 m and plot 1.0
Creep
them against the survey data. The 1.2
value 2.20 m is rather arbitrary, but
this value is not significant. Irre- Fig. 38: Creep function comparison
spective of the value, the plot would
still show that neither the ACI nor CEB-FIP curves come close to the survey data (see Fig. 39a).
The curves from CEB-FIP are a little better than that of ACI as far as matching the Koror field
data is concerned, but far from the reality of the bridge’s significant and ongoing deflection after
1000 days.
We can also use another method to compare the ACI and CEB-FIP curves with the field survey
data. In November, 1985, 3000 days after completion of the bridge, JICA reported the deflec-
tion as –0.866 m, adding that to the actual construction camber of 210 mm, the total deflection
should be –1.076 m. If we plot the ACI and the CEB-FIP curves based on this survey data (see
Fig. 39b), it is clear that these curves would not be able to match the survey data.
The difference between AASHTO and ACI or CEB-FIP is that the creep behavior by AASHTO
is significantly influenced by member thickness. The creep progress reduces drastically when
the member thickness increases, as shown in Fig. 40.

The fib 2010 Model Code


According to fib 2010 Model Code, the total creep is the sum of the basic creep and the dry-
ing creep. Both of these factors are time dependent, and the time function for each of them is
different such that it is not possible to distinctively separate the time function from the creep
formula. Figure 41 shows the development of creep with respect to time based on the same
parameters of the Koror Bridge:

Loading age to = 84 days,


64 CHAPTER 8. DEFLECTION OF THE BRIDGE

(a) Time (days) Relative humidity = 80%,


1 10 100 1000 10000 100000
0 Concrete strength fcm = 45 MPa.

Even though the creep coefficient


–0.50
Actual deflection at age 100 years is not much differ-
Deflection (m)

ent from the value given by CEB-FIP


–1.00
1990, the shape of the creep curve is
ACI 209 rather different (Fig. 41). This is espe-
–1.50
cially so in a semi-log scale. It is close
CEP-FIP, βH = 1500 to a straight line.
–2.00
The creep coefficient calculated using
–2.50 fib 2010 (final draft) is still much lower
Assume creep deflection @ day 100000 = 2.20 m
than the measured deflection of the
(b) Time (days) Koror Bridge.
1 10 100 1000 10000 100000
0
–0.20
Creep Under Tension
Deflection midspan (m)

–0.40 ACI 209 As shown in Fig. 23, the top fibre stress
–0.60 in the box-girder changed from com-
Actual deflection
–0.80 pression to tension when the prestress
–1.00 loss of the tendons reached about 20%.
CEP-FIP, βH = 1500
–1.20 Indeed, field measurement of the ten-
–1.40 dons before the rehabilitation showed a
–1.60 prestress loss of about 50% which indi-
–1.80
cated that the top slab of the box-girder
Based on creep deflection @ day 30000 = –1.07 m was in tension. Creep effect of con-
crete in tension is still rather uncertain.
Fig. 39: (a, b) ACI and CEB-FIP Prediction Vs. How much influence this tension stress
Actual Measurement might have had on the creep deflection
of the bridge is a big unknown in the
analysis.
Time (days)
1 10 100 1000 10000 100000
0
1600 mm
Calibration of Creep
0.1000
0.2000
Models
Time function

1200 mm
0.3000
2V/S = 400 mm B3 Model
0.4000
800 mm
0.5000 Bazant et al. [27, 28] recalculated the
0.6000 creep prediction for the Koror Bridge
0.7000 Aashto creep based on different existing creep pre-
diction models (see Fig. 42). It shows
0.8000
that no existing model could come
Fig. 40: AASHTO creep time function versus close to predicting the magnitude
member thickness of deflections of the Koror Bridge,
CALIBRATION OF CREEP MODELS 65

(a) 1.60

1.40

1.20

Creep coefficient 1.00


Drying creep
0.80 Basic creep
0.60 Total creep

0.40

0.20

0
0 20000 40000 60000 80000 100000
Time after load application (days)
(b) 1.80
1.60 Drying creep
1.40 Basic creep

1.20 Total creep


Creep coefficient

1.00

0.80

0.60

0.40

0.20

0
1 10 100 1000 10000 100000
Time after load application (days)

Fig. 41: (a, b) fib 2010 creep estimate

including the original B3 model. The modified B3 model calculation was carried out using
3D structural models to capture the shear lag of the box-girder. They then calibrated the
model based on the survey results of the Koror Bridge and were able to find a predic-
tion curve that fits well with the survey by modifying several constants in the original B3
prediction model. However, such a drastic change in structural modeling has to go through
more scrutiny and make its way to the specifications before it can be adopted in a practical
design.

ACI, AASHTO and CEB-FIP


We have pointed out above that it is not possible for the ACI and CEB-FIP models to correctly
match the actual deflection of the Koror Bridge. However, if we modify the CEB-FIP formula to
66 CHAPTER 8. DEFLECTION OF THE BRIDGE

0 Δc = 2.20 * [(t–to) / (2800 + t–to)],


CEB (1D, SOFiSTiK) the resulting curve matches the sur-
JSCE (3D) vey data quite well (see Fig. 43a).
–0.4
Mean deflections (m)

ACI (3D)
CEB (3D) The current AASHTO model has
GL (3D)
some similarity with CEB-FIP but
–0.8 B3 (3D)
(set1) the constants and variables in the
equation are different:
–1.2 φ = 3.5 kfkhckldkcktd
B3 (3D)
(set2) where kf considers the concrete
–1.6 strength, khc considers the relative
0 2000 4000 6000 8000
humidity, kld considers the age of
t, time from construction end, days
the load application, kc considers the
CEB (1D, SOFiSTiK) thickness of the concrete member,
0 and ktd is the time function after load
JSCE (3D) application. It can also be written as:
–0.5
CEB (3D) φ (t, to) = φ o * β(t – to)
Mean deflections (m)

GL (3D)
–1.0 where φo = 3.5 kfkhckld is not affected
B3 (3D) by time, while β(t – to) = kckt
–1.5 (set1) expresses the creep deformation in
relation to time,
–2.0
B3 (3D) kc = {[45 + (t – to)] /[26 * e0.36V/S +
(set2) (t – to)]} * {(1.80 – 1.77 * e−0.54V/S)
–2.5
1 10 100 1000 10000 100000 /2.587}
log t, time from construction end, days
ktd = (t – to)0.6 / [10 + (t – to)0.6]
Fig. 42: Koror Bridge-Creep Deflection Estimated by
Various Creep Models (courtesy Zdenek Bazant) The value of ktd varies from 0.0
to 1.0. But the value of kc does not
reach 1.0; it is much smaller than 1.0 even at time equal to infinitive. Therefore, the ultimate creep
coefficient is not φo, but φo x kc,(t=infinitive).

If we assume φo Δe = 3.10, 2V/s = 800 mm, ktdxkc,(t=infinitive) is about 0.70 indicating that the
ultimate creep deflection is 3.1 * 0.70 = 2.2 m. This gives us a curve that matches the survey
data very well (see Fig. 43b).

Here, we have not really performed any calibration, except having assumed certain values for
the terms in the equations. However, the existence of such curves that so closely match the
actual data means that, by proper calibration it should be possible to modify the model to fit
the actual deflection of the Koror Bridge. Further research is needed to determine whether the
calibrated formula fits other bridges’ actual behavior as well.

However, if the ultimate creep deflection is really 2.2 m and the calculated elastic deflection is
0.37 m, the corresponding ultimate creep coefficient would be 5.95. It appears that this is a value
many engineers would not be prepared to accept today.
THE QUESTION OF THE LOADING AGE 67

Time (days)
0 10 100 1000 10000 100000
(a) 0

–0.5
Actual deflection
Deflections (m)

–1.0

ACI 209
–1.5

CEB-FIP, βH = 1500
–2.0

–2.5
Assume creep deflection @ day 100000 = 2.20 m
Green curve 2.2*(t-to) / [2800+(t-to)]

Time (days)
0 10 100 1000 10000 100000
(b) 0
Deflections at midspan (m)

–0.5

–1.0
Actual deflection

–1.5

–2.0
AASHTO creep curve
Assumptions: ϕ0 = 3.1, 2V/S = 800 mm
–2.5

Fig. 43: (a, b) ACI and AASHTO creep deflection

The Question of the Loading Age


Up to now we have been using a simplified approach for creep estimation in our discussions.
The simplified method assumed that the entire bridge was cast at the same time on falsework,
and it was then stressed and released on the day represented by the loading age. The actual
construction of the bridge was different from this assumption. The bridge was built segmentally,
one segment at a time (see Fig. 4). Therefore, the appropriate loading age is always a question
in such an analysis.
68 CHAPTER 8. DEFLECTION OF THE BRIDGE

Back in the 1970s, there was no software that could handle a step-by-step, sometimes also called
stage-by-stage, or time-history type of analysis of segmental construction sequence. The simpli-
fied method used for the design of the Koror Bridge was the common procedure to predict creep
deformation of the bridge. Today, there are many commercially available software applications
that can follow exactly the same construction process. The simplified method is generally not
used anymore.

In the design of the Koror Bridge, we had assumed that the loading age of the bridge was 84
days. The logic behind this assumption was explained in a previous chapter. But, is this assump-
tion correct? The best way to answer this question is to compare the calculated time-dependent
deflections of a segmentally constructed structure to one that is built on falsework and stressed
and released on day 84, as was assumed in the original design. Such a comparison calculation
was carried out for a simple cantilever. Two currently commercially available software applica-
tions for segmental bridges, TANGO and MIDAS, were used. These applications can analyze
the structure according to the actual construction sequence.

The calculation was based on CEB-FIP 90. The input information was as follows:

Geometry and cross section were based on the design drawings [11]; concrete cylindrical
strength = 5000 psi (35 MPa), which was equivalent to a cube strength of 45 MPa; relative
humidity = 82%; and normal weight concrete.

The input data is the same as those used in the original design of the bridge in 1974. Two groups
of calculations were performed:

Case 1: A step-by-step calculation based on the actual construction schedule of the cantilever,
assuming the dead load and the prestressing became effective on day three after casting concrete
in each segment. The result is represented by the dotted line in Fig. 44.

Case 2: Similar to the original design, we may assume that the entire bridge was cast on false-
work, and the tendons were stressed and the falsework released on the day of the loading age,
which was assumed to be
7, 14, 28, 56 and 84 days
Time (days after bridge completion)
respectively.
0 10 100 1000 10000 100000
0
In Fig. 44, day 1 on the time
Time dependent deflection at midspan
hinge after bridge completion (mm)

Segmental construction
–100
scale is always the day the
bridge is completed—the
–200 day after all the prestressing
and superimposed dead loads
–300 have been fully applied. For
Falsework construction 84 case 2, day 1 is the day when
loading age as indicated 56
–400 creep deformation of the
28
14 bridge begins.
–500 7
Time dependent deflection of a cantilever For case 1, however, because
software: MIDAS the bridge has been under
–600
construction for about 280
Fig. 44: Comparison of Loading Age (time dependent deflection) days, a portion of creep has
THE QUESTION OF THE LOADING AGE 69

already taken place during construction. Figure 44 reflects only the creep deflection after the
bridge has been completed.

Up to now, all our survey data on deflection has been measured against the as-built elevation of
the bridge. Consequently, the comparison of the different curves in Fig. 44 makes sense even
though the dotted curve only included a portion of the total creep effect of the bridge while all
other curves include the entire creep effect. The creep deflection after bridge completion was
–171 mm for a segmentally built cantilever while it was –491 mm, –439 mm, –392 mm, 351 mm
and –328 mm respectively for loading ages of 7, 14, 28, 56, and 84 days, respectively.

A step-by-step calculation is usually assumed to be more accurate, at least numerically, because


it follows the actual construction procedure of the bridge. So if we assume this is correct, the
simplified method appears to yield much too high a deflection. Even assuming a loading age of
84 days, the calculated deflection after bridge completion is almost 100% higher than the result
from the step-by-step calculation.

The effect of the loading age on the creep coefficient according to CEB-FIP 90 is denoted by
β(to), the value of which is 0.74, 0.63, 0.40 and 0.28 for loading ages of 3, 7, 84 and 300 days
respectively. Even with a loading age of 300 days, the deflection would have been –328 * 0.28/0.40
= –230 mm, which is still larger than –171 mm. Obviously, a loading age equal to or more than 300
days cannot be justified because the bridge cantilever was built in less than 300 days!

Based on this calculation, it appears that there is no appropriate loading age for a simplified
calculation that can properly match the result of a step-by-step analysis. The 84 day loading age
assumed in the original design appears to be appropriately conservative.

As mentioned above, one difference between the step-by-step analysis and the simplified method
is that at the time of bridge completion a portion of the creep had already taken place while for
the simplified method the entire creep was assumed to have begun after bridge completion.
Figure 45 shows the effect of creep on the deflection of the mid-span before completion of the
bridge. The bridge actually bends upward at the beginning of construction. This phenomenon is
quite common in segmental
Time (days after start of cantilever construction) construction because the
60
bending moment in the girder
Bridge completion
40 on day 284 after due to post-tensioning is
start of piertable usually higher than the dead
Creep induced deflection
at midspan hinge (mm)

20
load bending moment in the
0 first few segments; the bridge
0 50 100 150 200 250 300
girder tends to bend upward
–20
and consequently, the creep
–40 deformation is also bending
Stage-by-stage analysis – upwards. This upward
–60
CEB-FIP 90 software: TANGO deflection has a profound
–80 influence on the total creep
deflection of the girder. The
–100
deformation, especially the
Fig. 45: Mid-span deflection due to creep before bridge rotation of the first few
completion segments, has a higher effect
70 CHAPTER 8. DEFLECTION OF THE BRIDGE

on the final deflection of the girder at the mid-span because they are farther away from the end
of the cantilever. (The effect on the deflection of the girder at the mid-span = angle change of the
segment multiplied by the distance of the segment from the end of cantilever.) According to the
calculation, the accumulated creep deflection at the mid-span is –86 mm (downward) at the end
of construction. Adding this to the ultimate creep deflection of –171 mm which took place after
the bridge had been completed, the total creep deflection of the girder at the mid-span would be
–257 mm, which is still smaller than, but closer to the total creep deflection of –328 mm based
on a loading age of 84 days.

Figure 46 shows the development of creep deflection at the mid-span based on a stage-by-
stage (step-by-step) analysis according to various creep models. It should be noted that this
calculation is different from
Time (days) the one above. The previ-
1 10 100 1000 10000 100000 1000000 ous calculation is only for a
0 simple cantilever while the
present calculation is based
Creep deflection at midspan hinge

–50 on the actual construction of


after bridge completion (mm)

the entire bridge, including


–100 the side spans. The deflection
of the cantilever is higher
–150 because there is a rotation at
the main pier.
–200
CEB-FIP 78 It is surprising that the three
Stage-by-stage analysis CEB-FIP 90 models give very similar
–250
software: TANGO ACI 209 results. But the calculated
deflections are way too small
–300 when compared with the
Fig. 46: Creep deflection calculated by step-by-step analysis actual survey of deflection
from the site (see Fig. 47).
Time (days after bridge completion) What conclusion can we
1 10 100 1000 10000 100000 1000000 make based on these results?
0

–200
Creep deflection at midspan hinge
after bridge completion (mm)

–400
CEB-FIP 78
CEB-FIP 90
Other Effects on
–600
ACI 209 Deflection
–800
Since we are not able to fig-
–1000 ure out why the deflection
Actual deflections
–1200 of the bridge was so much
Stage-by-stage analysis larger than anticipated by
–1400 the creep models, we inves-
software: TANGO
–1600 tigated what other effects
might also contribute to the
–1800
long-term deflection of the
Fig. 47: Results of step-by-step analysis Vs. actual measurements bridge.
OTHER EFFECTS ON DEFLECTION 71

Effect of Differential Shrinkage


In general, we can assume that shrinkage only affects axial shortening without causing any
noticeable vertical deformation. In most calculations, the bridge girder is represented by a line.
The estimated shrinkage coefficient is based on a nominal thickness, represented by the term
2V/S, (or 2A/S in AASHTO). This is very close to the average thickness of the cross section.
If the bridge span is not very long, this assumption is usually acceptable because the concrete
thickness in the top slab, the webs and the bottom slab is quite similar. If the bridge span is very
long, the bottom slab of the girder near the piers will be much thicker than other parts of the
cross section. For the Koror Bridge, the cross section near the pier has a 432 mm top slab, two
356 mm webs and an 1150 mm bottom slab. The difference is certainly very significant.

The differential thickness


Time (days)
may influence the bridge
1 10 100 1000 10000 100000 1000000
deflection: its ultimate 0
shrinkage and its speed of
AASHTO shrinkage coefficient

0.000050
shrinkage.
2V/s = 500 mm
0.000100
Figure 48 shows the shrink-
age progress for the con- 0.000150
400 mm
crete member with various 0.000200
thicknesses according to 300 mm
0.000250
AASHTO. It is debatable if
the ultimate shrinkage of a RH = 80%
0.000300
200 mm
thick plate is really different
0.000350
from the ultimate shrinkage
of a thin plate when the time Fig. 48: AASHTO Time function for shrinkage
approaches infinity, but the
speed is definitely different.
A thick plate will shrink much slower than a thin plate. If we disregard their effect beyond 100
years, the AASHTO curves (see Fig. 48), should still give us a correct picture of what may
happen.

Reading from the figure, for a 200 mm thick member, the shrinkage coefficient at day 200 is
0.000223; its ultimate shrinkage is 0.000329, so its residue shrinkage after 200 days is 0.000106.
For a 500 mm thick member, its ultimate shrinkage coefficient is 0.000066; its shrinkage after
200 days is 0.000013, so its residue shrinkage after 200 days is 0.000053 which is only half of
that of a 200 mm thick member. This means that if the construction time for the bridge is 200
days, and the top slab average thickness is 200 mm while the bottom slab is 500 mm, there will
be a differential shrinkage between top and bottom slabs, which will cause vertical deflection of
the bridge girder. However, a calculation for the Koror Bridge shows that the deflection caused
by this differential shrinkage is not significant.

Effect of Main Pier Settlement


There have been questions about a possible effect on the mid-span deflection due to settlement
of the main piers. Following is an approximate analysis to answer this question.
72 CHAPTER 8. DEFLECTION OF THE BRIDGE

Each main pier was supported by 104 precast piles with a cross section of 16 inch × 16 inch
each. The piles were in general 100 ft long with a 1:8 batter. The total permanent load on the
piles was about 25,000 kips, including the weight of the footing and the seal. Thus, the axial
stress in the piles due to permanent load was about 135 ksf (6.6 MPa). The elastic shortening of
the 100 ft-long piles should be about 0.023 ft (0.0071 m).

The piles were driven to bearing so there should be no or negligible settlement at the bottom of
the piles. In other words, after completion of the bridge, the only settlement would have been
due to the plastic shortening, or creep of the piles. The piles were sufficiently old so that shrink-
age was negligible.

We have no record of the age of the piles at the time they were driven. For simplicity sake,
we may just assume a creep coefficient of about 2.50. According to this assumption, the foot-
ing would have settled about 2.50 × 0.0071 m = 0.018 m after the creep of the piles had been
completed.

The main-span cantilever was 120.4 m long and the distance from the main pier to the end pier
was 53.7 m (see Fig. 49). Conservatively neglecting any settlement at the end pier, the 0.018 m
settlement of the main pier would increase the deflection at the mid-span of 0.018 × (120.4 +
53.7)/53.7 = 0.058 m after completion of the bridge. Comparing this to the actual deflection of
1.54 m, it is relatively minor. Moreover, this deflection had already been included in our analysis
because the piles were a part of the structural model.

Effect of Prestress Loss


Nine tests on three tendons that were performed right before the repair found that the average
prestress loss was about 50%—much higher than anticipated!

The prevailing practice is that prestress loss may be calculated based on the compatibility of
the strain of the concrete surrounding the prestressing steel, plus the relaxation of the prestress-
ing steel itself. In the design of prestressed concrete bridges, the prestress loss usually ranges
from 10% to about 20% depending mainly on the level of stress of the concrete surrounding the
tendon. If the surrounding concrete is in high compression, the strain of concrete due to creep
is higher and so the prestress loss is also higher because of strain compatibility between the
concrete and the steel tendon embedded in it. In the case of Koror Bridge, the concrete stress
at the tendon level was close to zero so there should have been very little strain due to creep.
The loss should have been mainly due to relaxation. This is the conventional design procedure

53.7 m 120.4 m

Fig. 49: Geometry of the cantilever


OTHER EFFECTS ON DEFLECTION 73

of prestressed concrete bridges, and this was how the Koror Bridge was designed. Upon closer
scrutiny, however, the conventional design procedure could cause deviations:

1. Currently, in the design of a box-girder bridge, the bridge girder is represented by a series
of line elements. The nonlinear stress distribution in the top and bottom slabs due to shear
lag is taken into account through the use of effective width. This effective width is used to
calculate both stresses and deflections in all cross sections of the box-girder. The effective
width is usually a function of the ratio of the slab width to the bridge span. This method has
been used extensively and has been found satisfactory for practical purposes. In the design
of the Koror Bridge, the distance between the webs is about 6.6 m = 2 × 3.3 m, the length
of the cantilever is 120.4 m, and the ratio of slab width to cantilever length is 3.3/120.4 =
0.027. This ratio is very small so, according to the specifications, the effective width was
assumed to be the full width of the top and bottom slabs. However, this method of calcula-
tion neglected the influence of the depth of the girder and the shear lag in the webs. In a
deep girder, a vertical plane of the girder will not remain plane when only the top slab is
prestressed. When prestress is applied to the top slab of a very deep girder, the bottom por-
tion of the cross section will hardly be affected by the prestressing of the top slab. Thus,
the actual compressive stress in the top slab is higher than our calculated value assuming
a plane section. This difference can be significant, especially if the bridge is very short! In
addition, this higher local stress happens when the concrete is relatively young, so the creep
effect tends to be much higher. As a consequence, the prestress loss is higher than calcu-
lated. However, this is a local phenomenon. For a load further away from the cross section,
this effect will decrease.
2. The relaxation of the Dywidag bar tendons was based on the sketch in Fig. 50 supplied by
the supplier, Dyckerhoff&Widmann. As the tendons were anchored at about 70% ultimate
strength after stressing and were expected to be reduced to about 60%, a 3% relaxation was

Fig. 50: Relaxation of Dywidag prestressing bar tendons


74 CHAPTER 8. DEFLECTION OF THE BRIDGE

assumed in the design. The figure shows the relaxation of the prestressing steel leveling
off after about 10,000 hours, or 1.14 years. This assumption is being contested by some
researchers [2].
3. In hindsight, this graph was probably developed based on a room temperature of 20°C. The
higher temperature at the project site must have contributed to the increase of relaxation in
the prestressing steel. The average temperature in Koror is about 28°C. All tendons were
located in the top slab. The temperature of the top slab was most likely higher than 20°C in
the day time as it was under constant sunshine. So, an average temperature higher than 28°C
should be assumed. There was no test data available for relaxation of Dywidag bar tendons
at higher temperatures; a comparison of relaxation data of other prestressing steels suggests
that the relaxation at 35°C could be twice as high as the relaxation at 20°C. But increasing
the relaxation from 3% to 6% is still a small number.
4. According to Bazant et al. [2], AASHTO, ACI and CEB models do not consider the effect of
drying creep of concrete. The effect of drying creep of concrete in the top slab on prestress
loss and girder deflection could be significant. Drying creep of concrete in the top slab
would increase the loss of prestress and tend to bend the girder upwards in the early stage
of construction.

In the original design of the Koror Bridge, a 10% prestress loss was used based on the calcula-
tions at that time. Obviously, a loss of 50% would never have been expected. Most importantly,
even if we were to design the bridge today, we would have followed the current AASHTO or
CEB. The result would not have been much different!

We will have more discussion on this issue later.

Effect of Cracks on Deflection


Because the prestress loss indicated by tests was much higher than expected, people did have
doubts about the accuracy of the test results [29]. Unfortunately, we are not able to re-confirm
the test results anymore so we can only accept these values here and see how the values from the
test result may have affected the bridge deflection.

The concrete cylinder strength assumed in the design of the Koror Bridge was 5000 psi (35
MPa). Field tests during construction showed that the actual concrete strength was about 5200
psi, which was sufficiently close for practical construction purposes. According to ACI [7], the
modulus of rupture (cracking strength) of concrete can be estimated as:

fr = 7.5 √f ′c = 7.5 * √5200 = 541 psi

fr = 3.79 MPa

Figure 23 shows the relationship between top-fibre stress in the box-girder and the magnitude
of prestress loss. In the original design, the prestress loss calculated using a formula suggested
by Leonhardt [12] was about 10%. If the prestress loss had been between 10% and 20%, the top
fibre stress of the box-girder would have remained under small compression or very slightly in
tension when the prestress loss reached 20%. The allowable tension in the prevailing specifica-
tion at that time was 3√f ′c, or 212 psi = 1.48 MPa. The design was conservative.
OTHER EFFECTS ON DEFLECTION 75

However, when the prestress loss is above 36%, the top fibre stress will exceed the modulus
of rupture of the concrete, 3.79 MPa, which means that the top fibre will start to crack. When
the prestress loss is 50% as indicated by the tests, the calculated nominal top fibre stress will
be close to, and in some areas, higher than 6.0 MPa, which is about 60% above the modulus of
rupture of the concrete used.

Hair cracks in the concrete may not affect the stiffness of a concrete member much, but when
cracking progresses to a certain degree, the stiffness of the member decreases rapidly. Deflection
is inversely proportional to stiffness, so if stiffness is reduced by 50%, deflection will increase
by 100%. It is important to evaluate the effect of cracking on the deflection of the bridge. The
stiffness of a girder depends mainly on the effective moment of inertia of the girder cross sec-
tion. Therefore, it is prudent to evaluate the effective moment of inertia of the girder under the
actual cracked condition.

The moment of inertia of a bridge girder at and very close to a crack is equal to the cracking
moment of inertia, Icr, after the tensile fibre stress is equal to or exceeds the modulus of rupture, fr.
PCI suggests that it can be approximately calculated as

Icr = np Aps dp2 [ 1 – 1.6 √(np p)]

The moment of inertia of the girder between the cracks does not change and is equal to the
transformed moment of inertia of the cross section. In order to calculate the deflection, an
effective moment of inertia of a cracked girder is usually estimated by the formula suggested
by Branson [6], as

Ie = Icr + (Mcr/Ma)3*(Itr−Icr)

where Icr is the moment of inertia of the cracked transformed section, Itr is the moment of inertia
of the uncracked transformed section, Mcr is the bending moment corresponding to the modulus
of rupture, fr, and Ma is the actual bending moment in the girder.

Obviously, the values of Ma and Mcr vary along the length of the girder because the girder has a
variable depth. To get an idea of what this means, we can take a cross section, say, section 21,
as an example. This cross section is in the main span located at 33 ft (10.06 m) from the centre
of the main pier. The dimensions are as follows: top slab 1.417 ft × 329.36 ft wide, the webs are
1.167 ft thick each, the bottom slab is 3.60 ft thick, and the total girder depth is 45.08 ft. There
are 298 tendons in the top slab; with each at 1.25 inch nominal diameter with a 1.25 square inch
cross-sectional area, the ultimate strength is 150 ksi. For this cross section, the transformed
moment of inertia is Itr = 79,900 ft4 (2264 m4), the gross cross section, Ig = 69,100 ft4 (1958 m4),
and Icr can be estimated by the formula [30],

Icr = np Aps dp2 [ 1 – 1.6 √(np p)]

Icr = 2450 ft4 = 688 m4

With Itr = 2264 m4 and Icr = 688 m4, we can plot the relationship between Ie and the ratio (Ma/
Mcr) as shown in Fig. 51 according to the Branson formula. It shows that the effective moment
of inertia Ie of the member decreases rapidly when the maximum moment exceeds the crack-
ing moment. For Ma/Mcr = 1.50, the effective moment of inertia is 1155 m4, or approximately
76 CHAPTER 8. DEFLECTION OF THE BRIDGE

2500
Itr = 2264 ft4
2000

Moment of inertia I
1500
Ie
1000

Icr = 688 ft4


500

0
0 1 2 3 4 5 6 7
Ma / Mcr

Fig. 51: Effective Ie of a cracked girder

half of the uncracked moment of inertia. This means, when the maximum moment of a beam is
50% higher than the cracking moment, the effective stiffness of the beam will decrease by 50%.
Consequently, the deflection of the beam will increase by 100%.

Notably, Branson’s formula was written for a non-prestressed beam. In a non-prestressed beam,
the nominal extreme fibre stress is equal to

f = M y/I = M/S

where S is the section modulus of the cross section and y is the distance between the neutral axis
and the extreme fibre. Thus, the term Ma/Mcr can also be expressed as the ratio of the extreme
tensile fibre stresses under maximum moment fmax and the modulus of rupture, fr, with both
stresses calculated based on the uncracked cross section. Thus,

(Mcr/Ma) = fr/fmax

where fmax is the total stress at the extreme fibre.

In the case of the Koror Bridge, the cause of cracking was not due to external loads but due to
loss of prestress, and the above formula may or may not be directly applicable. Nevertheless, it
should be at least an approximation. Thus, with fmax = 6.2 MPa when the prestress loss reaches
50%, and fr = 3.79 MPa,

(Mcr/Ma) = fr /fmax = 3.79/6.20 = 0.61.

Thus, Ie = 1040 m4, and Ie/Itr = 0.46

which means the effective stiffness of the girder could have been reduced by more than half.

What would happen to the deflection if the stiffness of the bridge girder was reduced by 50%?
We will look at that next.
INTERPRETATION OF TEST RESULTS 77

Interpretation of Test Results


As discussed in Chapter 4, based on the test results, the actual 28 day modulus of the concrete in
Koror was probably about 21.9/1.12 = 19.6 GPa, which is 31% lower than the AASHTO value
assumed in the original design. The prestress loss was found to be close to 50%.

Table 3 shows various scenarios of the deflection based on these actual test results. The first col-
umn represents what had been used in the design. The second column is based on the test results,
assuming the prestress loss was 50% and the modulus was 19.6 GPa. With a creep coefficient
of 1.69 according to CEB 90 as explained in Chapter 8, the ultimate creep deflection would be
1329 mm, which is 3.45 times the original estimate.

The third column assumes a 50% prestress loss had caused cracks in the top portion of the con-
crete girder and thus reduced the girder stiffness by 50% as described above. The total creep
deflection after closure would have been 2577 mm which would have been reached after the
creep had been completed. This might be 100 years later. The survey result of 1300 mm after 16
years appears to fit this estimate.

We do not have any record on the development of these cracks. Nor do we know how fast these
cracks progressed. As the prestress loss increased, the degree of cracking also increased which,
in turn, reduced the girder stiffness and caused the elastic deflection of the girder to increase.
Since creep deflection is approximately equal to the product of elastic deflection and the creep
coefficient, this would have triggered further deflection of the girder. The resulting deflection
could have been significantly larger and prolonged in time.

Assumptions As Designed Based on actual test results


Prestress (PT) loss 10% 50% 50%++
Concrete Modulus (GPa) 28.3 19.6 19.6
Assumed creep coefficient 1.3 1.69+ 1.69+
Elastic dead load deflection (mm) –1055 –1532 –3064
Deflection due to full PT +761 +1099 +2198
Deflection due to PT loss –76 –550 –1100
Total elastic deflection –370 –983 –1966
Creep deflection –481 –1661 –3221
Creep deflection before closure
(assumed 20%) +96 +332 +644
Deflection after closure –385 –1329 –2577
+
According to CEB 90 as explained in Chapter 8;
++
Assume cracks in top slab and webs caused a 50% reduction in stiffness.
Table 3. Deflection at Mid-Span Under Various Assumptions
78 CHAPTER 8. DEFLECTION OF THE BRIDGE

Experience from Another Bridge – the Pine Valley Creek


Bridge, California
An interesting exercise would be to find out what happened to other long-span prestressed con-
crete bridges. Do they have excessive deflection also? As mentioned in the last chapter, Bazant
et al. have studied over 60 bridges worldwide and found that most of these bridges deflected
more than anticipated in the design [25].

The Pine Valley Creek Bridge in California was designed and built about two years before the
Koror Bridge. It was the first long span prestressed concrete bridge in the US. Its construction
began in early 1973 and was completed in 1975. The bid documents and original design were
prepared by the California Department of Transportation. After bidding, the bridge was rede-
signed as a “value engineering proposal” by Dyckerhoff and Widmann, Inc. for the contractor,
a joint venture of S.J. Groves and Dyckerhoff and Widmann. The writer, as Chief Engineer of
Dyckerhoff and Widmann, was responsible for the final design of this bridge as well.

(a) 42´
1´ 10´ 24´ 6´ 1´
Profile grade
16˝

10˝

–2% Barrier railing


16˝
19´
10˝

9´ 6˝ 4´ 5˝ 14´ 2˝ 4´ 5˝ 9´ 6˝

(b) 1691´ LT. Bridge


1741´ RT. Bridge
BB

EB

255´ 340´ 450´ 380´ 266´


285´ 286´
115´ 115´ 225´ 225´ 225´ 225´ 155´ 155´ Cantilever

ABUT. 1

ABUT. 6
450´

PIER 2
PIER 5

PIER 4
PIER 3

Fig. 52: Pine Valley Creek Bridge: (a) cross section; (b) elevation
EXPERIENCE FROM ANOTHER BRIDGE – THE PINE VALLEY CREEK BRIDGE, CALIFORNIA 79

The Pine Valley Creek Bridge is located on Interstate Highway I-8 between San Diego and
El Centro in southern California. It crosses the Pine Valley, about 65 km northeast of San
Diego and about 450 ft above the Pine Valley Creek. It consists of two parallel constant 19 ft
deep box-sections along the entire bridge length. It has two superstructures. The span lengths
of the bridge are as shown in Fig. 52. The bridge deck is about 450 ft above the Pine Valley
Creek.

The creep coefficient of the Pine Valley Creek Bridge was estimated to be 2.50 in the design.
The difference between the 2.50 value here and the value of 1.30 used in Koror Bridge was due
to the differences in humidity and loading age (see Fig. 54). For the Pine Valley Creek Bridge,
the relative humidity in the area is about 40%, so the coefficient of environmental condition is
ke = 3.07 which was 1.85 in Koror; the average loading age of the concrete in the Pine Valley
Creek Bridge is younger because the cantilevers are shorter, so the age coefficient is kd = 0.88
instead of 0.76. Thus, by proportion, the creep coefficient of Pine Valley compared to that of the
Koror Bridge should be

1.30 * (3.07 * 0.88)/(1.85 * 0.76) = 2.50.

No deflection survey was done after the Pine Valley Creek Bridge was opened to traffic in 1975.
However from the three images in Fig. 53 it can be concluded that creep deformation has not been
a problem for this bridge. The first picture was taken right after the bridge was opened to traffic in
1975. The second picture was taken in 1998, 23 years later. The third picture was taken in 2011,
36 years after its completion. The edge of the bridge deck appears to remain the same in all three
pictures, indicating that there is no noticeable creep deformation after the bridge was completed.

The Koror Bridge was designed right after the design of the Pine Valley Creek Bridge so the
design procedure was basically the same. Both utilized the same Dywidag prestressing system
with the 150 ksi, 1.25 inch nominal diameter threadbars. Even though the contractors were
different, they were both advised by the same group of engineers from the same company,
Dyckerhoff&Widmann, Inc. However, there are three structural differences between the two
bridges (see Fig. 54).

The Pine Valley Creek Bridge has a constant cross section along the entire length of the girder.
Relatively speaking, the mid-span portions of the Pine Valley Creek Bridge are significantly
stiffer than the Koror Bridge. This probably helped avoid a local dip as appeared in the mid-span
of the Koror Bridge.

The span of the Pine Valley Creek Bridge is much smaller, 137 m versus 241 m. The Pine Valley
Creek Bridge is a continuous girder while the Koror Bridge had a hinge at mid-span. As a result,
the absolute elastic deflection is much smaller.

For the Pine Valley Creek Bridge, the bending moment in the girder was adjusted before clo-
sure to the same bending moment diagram, as if the structure was built on falsework. This
was to avoid the less reliable calculation of moment redistribution due to creep and shrinkage.
This, however, is not important for the Koror Bridge because its main span consists of a pair
of cantilevers, which are statically determinate. Statically determinate structures do not have
moment redistribution. The side spans of the Koror Bridge are one degree statically indetermi-
nate, but the bridge girder is very stiff in both side spans; hence its deformation is negligible.
80 CHAPTER 8. DEFLECTION OF THE BRIDGE

(a) (b)

(d) Elevation of pier


42´ 38´ 42´
(c)

19´
Inclined
column

60´
Struts
Pier shaft Pier 2 110´
(column) Pier 3 280´
Pier 4 280´
Pier 5 80´

Fig. 53: Pine Valley Creek Bridge (a) at time of completion in 1975; (b) in 1998, 23 years
after completion; (c) in 2011, 36 years after completion; (d) Longitudinal view

3.5

3.0
(environmental condition)
Coefficient ke (FIP 1970)

2.5

2.0

1.5

1.5

1.0

0
30 60 60 60 70 80 90 100
Relative humidity %

Fig. 54: Effect of relative humidity CEB1970


A WORLDWIDE SURVEY 81

A Worldwide Survey
While the Pine Valley Creek Bridge performed very well, Bazant et al. [25] have collected
data on deflection records of over 50 bridges worldwide, which shows almost every bridge had
deflected more than anticipated (see Fig. 55). The deflection in the data collected does not seem
to approach any asymptote. This means, the problem is wide spread.

102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0
Zuari Bidge Zuari Bidge Zuari Bidge Zuari Bidge Zuari Bidge
Span C Span E Span F Span H Span J
40 100 120 m 120 m 120 m 120 m
years 120 m
0.4 KB 0.3 Pelotas 0.1 Goa, 0.1 Goa, 0.1 Goa, 0.1 Goa, 0.1 Goa,
Bridge River 1986 1986 1986 1986 1986
241 m 189 m
0.8 Palau, 1977 0.6 Brazil, 1966 0.2 0.2 0.2 0.2 0.2
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0
Zuari Bridge Zuari Bridge Zuari Bridge Zuari Bridge Nordsund
Span L Span M Span O Span P Parrots Bru
120 m 120 m 120 m 120 m Ferry 142 m
0.1 Goa, 0.1 Goa, 0.1 Goa, 0.1 Goa, 0.2 Bridge 0.2 Konaru 0.2 Norway,
1986 1986 1986 1986 195 m Bridge 1971
U.S.A, 101.5 m
0.2 0.2 0.2 0.2 0.4 1978 0.4 Japan, 1987 0.4
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 Savines 0 Savines 0 0 0
Bridge Bridge
Span K Span J
Savines Savines Savines
0.2 Stenung- 0.1 Savines 0.15 77 m 0.1 77 m 0.1 Bridge 0.1 Bridge 0.1 Bridge
sundsbron France, France,
Bridge 1960 1960 Span h Span g Span f
94 m Span I 77 m 77 m 77 m 77 m
0.4 Sweden 0.2 France, 1960 0.3 0.2 0.2 France, 1960 0.2 France, 1960 0.2 France, 1960
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0
Zvíkov-Otava Zvíkov-Otava Zvíkov-Vltava
Bridge Bridge Bridge
Hinge 1, Hinge 2, Zvíkov- Hinge 4
0.1 Savines 0.1 Savines 0.15 0.1 84 m 0.1 84 m 0.1 Vltava 0.1 84 m
Bridge Bridge
Span b Tunstabron Czech Rep., Czech Rep., Bridge Czech Rep.,
Span c 1963
77 m 77 m 107 m 1963 1963 Hinge 3, 84 m
0.2 France, 1960 0.2 France, 1960 0.3 Sweden, 1955 0.2 0.2 0.2 Czech Rep., 1963 0.2
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0

Tsuki-
0.15 Urado 0.1 yono 0.1 Koshi- 0.1 Maastricht 0.2 0.2 0.1 Alnöbron
La Lutrive La Lutrive
Bridge Bridge razu 112 m Hinge 2, Bridge Hinge 1
230 m 84.5 m 19.5 m Netherlands, 131 m Hinge 3, 131 m 134 m
0.3 Japan, 1972 0.3 Japan, 1982 0.2 Japan, 1987 0.2 1968 0.4 Switzerland, 1973 0.4 Switzerland, 1973 0.2 Sweden, 1964
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0

Källösu- Källösu-
0.1 Alnöbron 0.1 Alnöbron 0.1 Alnöbron 0.1 Alnöbron 0.1 0.08 ndsbron 0.08 ndsbron
Grubben-
Hinge 2 Hinge 3 Hinge 4 Hinge 5 Vorst Bridge Hinge 1 Hinge 2
134 m 134 m 134 m 134 m Netherlands, 107 m, 107 m,
0.2 Sweden, 1964 0.2 Sweden, 1964 0.2 Sweden, 1964 0.2 Sweden, 1964 0.2 121 m, 1971 0.16 Sweden, 1958 0.16 Sweden, 1958
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0

Raven-
Veprek
0.2 stein 0.1 Heteren 0.07 Bridge 0.07 Empel 0.07 Narrows 0.07 Captain 0.08 Wessem
Bridge Bridge Bridge Bridge Cook Bridge
139 m 121 m 125 m 120 m 97.5 m Bridge 76.2 m 100 m
Netherlands, Netherlands, Czech Rep., Netherlands, Australia, Australia, Netherlands,
0.4 1975 0.2 1972 0.14 1995 0.14 1971 0.14 1966 0.14 1966 0.16 1966
102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104 102 103 104
0 0 0 0 0 0 0

Želivka Želivka
Decín Bridge Bridge Art
0.07 Gladesville 0.07 Bridge 0.02 Hinge 1 0.02 Hinge 2 0.1 Macq- 0.05 Gallery 0.03
Arch 300 m 104 m 102 m 102 m uarie Bridge Victoria
Australia, Czech Rep., Czech Rep., Czech Rep., Bridge Australia, Bridge
0.14 1962 0.14 1985 0.04 1968 0.04 1968 0.2 Australia, 1969 0.1 1961 0.06 Australia, 1966

Fig. 55: A Collection of Deflection Data of Concrete Bridges Worldwide Courtesy Bazant,
et al and Concrete International
83

Chapter

9
Initiation of the Collapse

There is a widely held belief that the collapse was initiated at the top slab near the Babeldaob
side main pier. Witness reports also confirmed this theory. However, experts disagree on why the
failure occurred at this location. Here, we will try to address this issue:

There were about 300 tendons in the top slab near the main pier. They were arranged in four lay-
ers. Each tendon was placed inside a 47 mm diameter metal duct (see Fig. 56). Bonding of these
ducts to the concrete is usually weaker, especially at the underside of the ducts (see Fig. 57). Due
to the Poisson’s effect, a high compressive stress in the longitudinal direction, after the repair,
had certainly accelerated the separation of the concrete from the tendon ducts. Creep and shrink-
age could have further aggravated this problem. These effects are time dependent. This explains
why the failure occurred almost three months after the repair was done. As a result, the concrete
in this area might form layers between the rolls of ducts. These thin concrete layers can buckle
under relatively small axial compression.

Before the repair, the longitudinal stress in the top slab was in compression but the stress was
very low. But the excessive prestress loss caused high-tensile stresses in the top slab. These
tensile stresses far exceeded the tensile capacity (modulus of rupture or cracking strength)
of the concrete, thereby causing the concrete to crack. These were transverse cracks whose

Fig. 56: Tendon arrangement in top slab at pier table


84 CHAPTER 9. INITIATION OF THE COLLAPSE

Fig. 57: Possible delamination of top slab

planes were vertical, approximately perpendicular


to the longitudinal axis of the bridge (see Fig. 58).
These cracks did not affect the safety of the bridge
as long as the top slab remained in tension.

The repair introduced relatively high compression in


the top slab. As shown in Table 2, the added com-
pressive stress in this area could have been as high as
10 MPa, if the effect of local temperature had been Cracks
included. Subtracting the nominal tensile stress of
about 6.0 MPa, the top fibre stress after the repair
could have been close to 4.0 MPa.

Thus, when the repair introduced a high compressive


force here and reversed the stress from tension back
to a relatively high compressive state, the cracked
concrete in the thin layers could no longer resist the
compressive force and it either crushed or buckled.

The critical elastic buckling stress, σcr, of a concrete Fig. 58: Cracked and uncracked con-
plate (see Fig. 59), can be expressed by the following crete column
formula:

σcr = k {π2E/[12(1-ν2)]} (t/b)2

where k considers the boundary condition of the plate, E is the modulus of elasticity and ν the
Poisson ratio of concrete, b is the width of the plate and t is the thickness of the plate.

With all other values being equal, the buckling stress was proportional to the square of the thick-
ness. If five layers of thin planes had formed in the top slab making the average thickness of each
plate 1/5th of the total thickness, the buckling stress would have been reduced to only 1/25th
of the original buckling stress. If unequal layers had formed, the problem would be worse. The
85

vertical cracked planes would have


further reduced this capacity in
resisting axial compressive stress,

b
just as the same as a column that has
a series of horizontal crack-planes
(see Fig. 58).

The compressive stress in the top t


slab of the original design was very
Fig. 59: Plate buckling
low, so no restraining reinforce-
ment was required. The repair,
however, introduced relatively high compressive stress in this area, so the assumption in the
original design was no longer applicable. This fact was apparently not considered in the repair
design.

The repair introduced a relatively high compression in the top slab, which had already cracked
extensively before the repair and together with local stresses due to temperature, caused the
concrete to crumble (see Fig. 58).

Structural details are designed to resist the forces acting on them. When we change the internal
forces in the structure, the original details may no longer work. A careful investigation must be
undertaken before such a repair can begin. In the original design of the Koror Bridge, the longi-
tudinal stress in the top slab was very low so no restraining reinforcement was required. If there
had been restraining reinforcement in the vertical direction to tie the possible layers together, the
buckling load would have been higher and the slab might not have failed.
87

Chapter

10
Other Failure Hypotheses

Many people shared their opinion on why and how they believed the Koror Bridge collapsed,
other than what we described in this book. We have investigated most of these theories and
found them to be incomplete. Following are several of them:

1. Shear failure of the webs: Because the webs of the box-girder at the Babeldaob side crashed,
some engineers believed that the failure was instigated by shear failure of the webs. If
the bridge was designed based on today’s British Code, Chinese Code or German Code,
the webs would likely have been thicker. However, pictures of the bridge after the failure
showed that the webs on the two sides of the failure plane had actually crushed against each
other, which indicated that the webs had failed under compression, not shear. Shear failure
would have shown a completely different picture (see Fig. 29).
Even though the centre span of the Koror Bridge was large, overall the bridge was very nar-
row. It was designed to carry only two traffic lanes and one pedestrian path. The actual shear
force and shear stress in the webs of the Koror Bridge were relatively low. The possibility
of a shear failure was rather remote.
2. Preparation of construction joints: There were no shear keys at the vertical interface of
the construction joint of the Koror Bridge. The bulkhead surface at the joint was cleaned
before casting of the subsequent segment took place. The necessity of shear keys at the
construction joints of a cast-in-place, prestressed concrete segmental bridge is still debat-
able. However, if a failure did occur due to the lack of shear capacity of the webs at the
construction joint, the failure surface should have shown a vertical slippage at the joint. We
have not observed any joint slippage in the Koror Bridge before or after the collapse. Thus
we can conclude that the preparation of the construction joint could not have contributed to
the failure of the bridge.
3. Damage of the deck plate during repair of the pavement: There was a suggestion that the
failure was instigated due to damage of the deck plate during repair and resurfacing of
the deck. Originally, the bridge had a concrete overlay which cracked extensively after
18 years of service. It was replaced by an asphalt blacktop as part of the repair. During
the removal of the original overlay the deck concrete might have been damaged. We have
no record of those possible damages. However, if these damages were the main cause of
failure, the bridge should have collapsed right at the time the damages took place, not two
months later.
88 CHAPTER 10. OTHER FAILURE HYPOTHESES

4. Large horizontal force from soil pressure at the abutments: In the original design, the ends
of the bridge were separated from the abutment wall. The separation was to be 2 inch at
medium temperature. An expansion joint seal was to bridge the gap between the bridge end
and the abutment. However, this detail was changed to a monolithic connection during the
construction (see Fig. 17). There was a speculation that, after the bridge girder was made
continuous by the repair, the soil pressure from the abutment backfill might have introduced
a high compression force in the box-girder, thus causing the girder to fail. However, our
calculations showed that the active earth pressure was not sufficient to cause high stress in
the girder.
5. Internal force redistribution of the dead load and original prestressing due to creep: After
the bridge was made continuous by the repair, the structural system of the bridge changed
from being two independent cantilevers in the main span to being a continuous girder (see
Fig. 60a, b, d). This system change could cause a redistribution of internal forces in the
structure due to creep of the concrete, but this redistribution takes time to realize. The
bridge collapsed about 90 days after it was made continuous and the bridge was 18 years
old at that time, which means that the loading age for the redistribution was 18 years, or
6570 days. Based on CEB 90-99, the ultimate creep coefficient for such a redistribution
would be 0.50 at time infinite and the creep coefficient 90 days after load application would
be 0.24. The effect of redistribution can be estimated approximately by the formula:

Ft = [ϕt/ (1+0.8 ϕt)] F0

Ft = 0.20 F0 for ϕt = 0.24

53.65 240.79 53.65

(a)
A B C D
(b)

(c)

(d)

18.60
53.65 240.79 53.65 18.60

Fig. 60: Structural Systems due to possible changes in boundary conditions (a) as Designed;
(b) After Repair; (c) if Main Piers Settled Under Horizontal Reaction; (d) if Both Abutments
Were Restrained.
89

which means only about 20% of the redistribution would have taken place. This is rather
insignificant.
6. Horizontal settlement of the main piers: After the bridge was made continuous by the re-
pair, there would have been a horizontal reaction at the main piers. Horizontal displacement
of these main piers could have created additional stress in the girder (see Fig. 60c).

Each of the main piers was supported by 104 piles with a 1:8 batter. Because the soil was not
very stiff, when the piers were subjected to a horizontal force, they could have displaced hori-
zontally. This settlement would have been partially instantaneous and partially long term. The
repair applied a jacking force of 6000 kips (27MN) to push the two halves of the bridge apart.
This horizontal force had to be resisted by the main piers. The bridge was then changed to a
continuous girder after the gap at the hinge at mid-span was filled with grout. The subsequent
external prestressing of 7640 kips (34.72 MN) would partially offset the horizontal pier reaction
from the jacking force. As a result, the horizontal reaction and thus the settlement of the piers
would have been quite small.
91

11
Discussions

Purpose of This Book


The purpose of this book is to present the facts about the Koror Bridge from a practicing engi-
neer’s point of view. That is, to explain how the bridge was designed, how it was built and what
happened after it collapsed. It also tries to theorize any differences in outcome were we to design
this same bridge today according to the current codes and specifications.

This book is not meant to be a forensic research report, but rather, a way to help future research-
ers in understanding the possible causes of the collapse of the Koror Bridge. It may be long
before we can fully comprehend what happened. Presenting what we know now may encourage
researchers to take steps to finally solve this intriguing riddle.

The Koror Bridge Was Structurally Safe


For 19 years, from its completion in 1977 to its collapse in 1996, and before the repair, there
were no safety issues on the Koror Bridge. The structural safety of the bridge had never been
called into question. The two consulting firms, JICA and ABAM, commissioned by the owner to
investigate the bridge had both come to the conclusion that the bridge was structurally safe. Our
calculations in previous chapters also indicated the safety factor of the bridge under ultimate
capacity was higher than required.

Therefore, we can conclude that the bridge would have had no problem if the long-term deflection
had not been excessive. There would have been no need for any repair and consequently no reason
for the bridge to collapse. Therefore, making sure that future bridges do not have unacceptable
excessive deflection is of utmost importance. Unfortunately, it appears that none of the current
relevant codes and specifications, ACI, AASHTO and CEB-FIP (now fib), could have predicted the
large deflection of the Koror Bridge, even if we were to design the bridge again today.

Long-Term Deflection
The main issue therefore is certainly the long-term deflection. There are various hypotheses of
relaxation, shrinkage and creep besides those that the codes would indicate. Unfortunately, each
92 DISCUSSIONS

model gives a different value of deflection for the same bridge structure. Estimation of long-
term deflection is not like the calculation of stress distribution in the structure, where it is usu-
ally possible to increase the accuracy of the result by further refining the model. Using a more
complex model does not necessarily mean the result will be more accurate in the estimation of
long-term deflections.

Furthermore, long-term deflection is a value we have to determine with sufficient accuracy in


the design stage so that we can build the bridge to the properly cambered elevation. Unlike
allowable stresses where we can build in safety factors, long-term deflection must be sufficiently
accurate so the bridge will neither have an excessive camber nor an excessive deflection.
Estimation of the long-term deflection for a bridge is either right or wrong and not up for inter-
pretation! Obviously, a good approximation is accepted as right in engineering!

There are numerous hypotheses of relaxation, creep and shrinkage in the literature. But each
of them gives a different result for the same structure. That puts the practicing engineer in a
dilemma. On the one side, our legal system requires the engineer to satisfy the codes and speci-
fications. (This is certainly much more serious in the US than in other countries!) While on the
other side, the state-of-the-art knowledge on creep, shrinkage and relaxation is full of uncertain-
ties, without consensus among leading researchers. Consequently, calculating the long-term
deflection based on various hypotheses does not have much meaning at all. If each of them gives
us a different deflection, what should the practicing engineer select and use as construction
camber?

If we are to design any long-span prestressed concrete bridge in the future, we have to be able
to estimate the prestress loss and the creep deflection to within an acceptable range. During the
design of the Koror Bridge, we were well aware that long-term deflection estimation was not
an exact science. But the prevailing belief was that the deviation should be within a range of
about ±30%. The calculated long-term deflection was 481 mm. Adding a 30% deviation would
only bring it to 625 mm, which would have been perfectly acceptable after a built-in camber for
the calculated value of 481 mm. Even a 100% deviation would have resulted in the elevation
at the mid-span being 481 mm too high at the time of completion and 481 mm too low after
all relaxation, creep and shrinkage had taken place. A 481 mm deflection over a span of 241 m
would mean a deflection to span ratio of 1/500, which would be noticeable but still acceptable.
The bridge has a 6% grade at both main pier locations, so the design elevation of the deck at
mid-span was about 12 ft (3.66 m) higher than the deck at both main piers. A 481 mm (1.58 ft)
deflection would not have been a critical issue at all.

Prestress Loss Was the Most Probable Culprit


From the foregoing analysis, it appears that the main problem may not have been creep and
shrinkage but the magnitude of prestress loss. High prestress loss caused high-tensile stress in
the top portion of the girder. When this high tensile stress exceeded the modulus of rupture,
cracks started to form. Depending on the severity of the cracks, the stiffness of the girder might
have been significantly reduced causing a very high deflection.

Calculations show that when the prestress loss in the Koror Bridge reached about 50%, the
nominal tensile stress at the top fibre of the girder cross section would have reached 6.0 MPa,
PRESTRESS LOSS WAS THE MOST PROBABLE CULPRIT 93

which would have been more than 50% above the modulus of rupture of the concrete, which
is 3.79 MPa. It is obvious that the stiffness of the girder must have been severely reduced.
Elastic deflection is inversely proportional to the stiffness, so the actual elastic deflection of
the Koror Bridge must have been significantly higher than what we have estimated under the
assumption of an un-cracked girder. It is difficult to accurately calculate the reduction in stiff-
ness of the girder due to cracks caused by the tensile stresses in the Koror Bridge girder. But, as
the Branson formula indicates, the effective moment of inertia of a girder decreases rapidly as
the maximum bending moment exceeds the cracking moment (see Fig. 56). In other words, the
effective moment of inertia of a girder decreases rapidly when the maximum fibre stress exceeds
the modulus of rupture.

As shown in the third column of Table 3, when we assumed that the effective moment of iner-
tia of the bridge girder was 50% of the full moment of inertia, the magnitude of the calculated
long-term deflection appeared to match the magnitude of the measured deflection of the bridge
quite closely.

Obviously, assuming that the effective moment of inertia of the girder was reduced by 50% is
only an approximation. But it appears that the assumption is rational.

If this is the case, our focus should be to find the reason why the loss of prestresss was so high.
It is far beyond the 10% to 20% that most engineers might typically anticipate.

Prestress loss in a tendon is basically due to the relaxation of the steel tendon itself and the
change in strain of the surrounding concrete:

σLoss = σsteel relaxation+ Esteel * Δεconcrete

In this formula, Esteel is a known value, relaxation of the steel tendons is determined by tests,
but usually it can be predicted to within a practical limit. The last term, Δεconcrete represents the
variation in the strain of the concrete surrounding the tendon, this is the biggest unknown. But
this value should be independent of what kind of steel tendon we used.

The patented threadbar tendons used in the Koror Bridge has an ultimate strength of only 150
ksi, or 1050 MPa. Today, seven wire strands are almost exclusively used for prestressed con-
crete bridges. These strands have an ultimate strength of 270 ksi, or nominal 1860 MPa, which
is much higher than that of the bar tendons. Because a large part of prestress loss is due to the
compatibility of the strain of the steel tendon and the surrounding concrete, the strain, and
consequently the loss of stress should remain the same no matter what steel tendons are used.

In the case of the Koror Bridge, assuming that the relaxation was 8%, then the prestress loss due
to other losses would have been 42%. This would be equivalent to 0.42 * 0.7 * 150 ksi = 44.1
ksi (304 MPa). If we had used seven wire strands, stressing them to 70% of the initial prestress
would be 0.7 * 270 ksi = 189 ksi (1303 MPa) and the loss would have been 44/189 = 0.23, or
23%. Adding back the 8% relaxation would result in a total loss of 31%. Therefore, today’s
bridges using strand tendons should perform better.

If we modify the prestress loss value in the second column in Table 3 from 50% to 31%, while
keeping the concrete modulus at 19.6 GPa, the creep coefficient at 1.6, the elastic deflection at
–1532 mm, the ultimate long-term deflection after closure would be –990 mm. Subtracting the
94 DISCUSSIONS

actual camber of 210 mm from this, the bridge would have been 780 mm below the design grade
line at the mid-span. This would amount to a deflection to span ratio of 1:309, which would have
been much better.

So, hypothetically, if we had used seven-wire strands for the Koror Bridge, the prestress loss
would have been reduced to 31%. Under a 31% prestress loss, the fibre stress in the top slab
would have been less than 2.5 MPa, which would have been below the modulus of rupture of
3.79 MPa (see Fig. 23), and the top slab would not have cracked; the stiffness of the girder
would not have been drastically reduced; and the elastic deflection, and consequently, the long-
term deflection would have been much smaller.
95

Summary

It appears that the only conclusion we can make based on the analysis we have done up to now
is that, none of the codes and specifications, ACI, AASHTO or CEB-FIP, can predict the large
long-term deflection of the Koror Bridge. The problem is the same today as it was in 1974, when
the Koror Bridge was designed.

The main culprit seems to be the prestress loss, which was much higher than these specifica-
tions had predicted. On the other hand, no engineer, even today, would accept a prestress loss
of 50% as rational. But this was the result of actual tests. The prestress loss is affected mainly
by the long-term creep and shrinkage of the surrounding concrete, so we will have to be able to
estimate the creep and shrinkage of concrete more accurately before we can better estimate the
prestress loss. Further research into this problem is recommended.

Based on what happened in the Koror Bridge, we may offer a few suggestions for engineers to
consider in the design or repair of long-span prestressed concrete bridges:

1. Increase the amount of prestress: Increasing prestressing can have two beneficial effects:
firstly, it will reduce the elastic deflection, and consequently, the long-term creep deflec-
tion no matter what the creep coefficient may be; and secondly, it reduces the possibility of
high-tensile stress caused by uncertainty in the estimation of prestress loss. In most cases,
the cost of additional prestressing is not significant when compared to the overall cost of
the bridge. However, note that excessive prestressing is also not advisable because this may
cause congestion and a higher compressive stress and thus higher creep deformation, which,
in turn, also increases prestress loss.
2. Use higher-strength steel for the tendons: Because almost all new bridges are using seven
wire strands for post-tensioning nowadays, this suggestion may not be relevant. Still, it may
be useful to keep in mind as an option. As indicated in the last chapter, the percentage of
prestress loss for a higher-strength tendon is lower.
3. Provide an additional camber to the bridge: In addition to the theoretically calculated value,
providing an additional upward camber is more visually appealing. A bridge that bends
upward looks better than a bridge that droops downwards.
4. Consider the use of a 3D model: If the geometry of the girder is more extraordinary, for ex-
ample, for exceptionally deep or wide box-cross sections, supplement the 2D analysis with
a 3D analysis may offer a clearer picture of the actual stress distribution in the structure.
96 SUMMARY

5. Assure the applicability of the existing structural details: When we perform a repair of an
existing bridge, we are usually changing the stress in the structure. Therefore, we must
carefully review the original details to understand whether they are suitable for the modified
stress conditions. It would be most ideal to redesign the bridge and compare the new design
to the old design to identify discrepancies or issues.
6. Consider the history of the original bridge: A repair design must carefully consider the his-
tory of the existing structure. Obviously, any bridge would only require repair if it has been
distressed in some way. It is highly likely that part or all of the bridge was weakened before
the repair.
97

Acknowledgement

The writer would like to thank Alfred Yee, Khaled Shawwaf, Ray Zelinski, Zdenek Bazant, Qiang
Yu, John Lee, and Chuck Seim for their valuable discussions. Thanks to Delan Yin, Lionel
Bellevue, Yu Deng and Yue Li for their help in the numerical analysis and preparation of the
sketches, and thanks to Pam Ching for her effort in reading and editing the manuscript.

Following are the participants in the design and construction of the original bridge:

Owner
Trust Territory of the Pacific Islands, Saipan, Mariana Islands, Micronesia
Koichi Wong, Director of Public Works and Contracting Officer
Bill Burmeister, Chief Inspector

Engineer of Record
Alfred A. Yee & Associates, Honolulu, HI, USA
Alfred A. Yee, President
Fred Masuda, Chief Design Engineer
Ray Zelinski, Project Resident Engineer

Contractor
Socio Construction Company, Seoul, South Korea
B. W. Chung, President
J. D. Lee, Project Manager
B. Y. Chung, Site Engineer

Designer and Field Engineers


Dyckerhoff&Widmann, New York, NY, USA
Man-Chung Tang, Vice President & Chief Engineer
Khaled Shawwaf, Project Manager, Design
Don Ward, Western Region Chief Construction Manager
Simeon Crosier, Construction Project Engineer
The above list of individuals does not include those involved with condition surveys, inspection,
repair design and repair execution.
99

Units

The bridge was designed using the US system of measurements.


Readers who are not familiar with these units can use the conversion table below:
1 ft = 1 ft = 0.3048 m
1 ft = 12” = 12 inches
Example: 5 ft 6” = 5 feet 6 inches
1 kip = 1000 US pounds = 4.4482 kN
1 US ton = 2 kips = 2000 US pounds
1 ksi = 1 kip per square inch = 144 ksf = 6.897 MPa
1 ksf = 1 kip per square foot = 0.0479 MPa
1 MPa = 0.145 ksi = 20.88 ksf
1 MN= 1000 kN = 224.809 kips
1 m = 3.2808 ft
1 MNm = 737.55 kip-ft
1 cubic meter = 35.315 cubic feet
1 kip per cubic foot = 157.09 kN per cubic meter
101

References

[1] Yee, A.A. Record span box girder bridge connects Pacific Islands. Concrete International
1979; 1(6): 22–25.
[2] Private correspondence with Alfred Yee, 2011, 2012.
[3] Private correspondence with Ray Zelinski, 2011, 2012.
[4] Private correspondence with John Li, 2012.
[5] Engineering News Records. Deep, fast water spawns record span, New York, July 14
1977.
[6] American Association of State and Transportation Engineers. LRFD Bridge Design
Specifications, Washington, D.C., 1998.
[7] American Concrete Institute Committee 209. Prediction of creep, shrinkage and tempera-
ture effects in concrete structures. reported by Subcommittee II, Report No. ACI209R-82,
ACI Publication SP-76, 1982 and subsequent revisions.
[8] CEB-FIP. International Recommendations for the design and construction of concrete
structures, Paris/London, 1970, 1978 and 1990.
[9] Tang, M.C. Koror-Babelthuap Bridge – a world record span. International Conference of
Large Structures, Proceedings, ASCE Annual Convention and Exposition, Chicago, IL,
October 1978.
[10] Tang, M.C. Shear design of large box girders. ACI Symposium, SP 42–14, 1973.
[11] Design drawings by Dyckerhoff and Widmann, Inc. for construction of the Koror
Babelthuap Bridge, July 1975.
[12] Leonhardt, F. Spannbetonfuer die Praxis, 3rd edn. Wilhelm Ernst & Sohn: Berlin, Germany;
1973.
[13] Koror Bridge Project Diary by Ray Zelinski.
[14] Japanese International Cooperation Agency. Present Condition Assessment of the Koror-
Babelthuap Bridge, February 1990.
[15] ABAM Engineers Inc. Basis for Design, Koror-Babeldaob Bridge Repairs, Bridge
Condition and Survey Report, 1993.
[16] Berger/ABAM Engineers, Inc. Koror-Babeldaob Bridge Modifications and Repairs, 100%
Submittal Design Drawings, October 1995.
[17] Louis Berger International Inc. Evaluating Quality Defects of Specific Projects, Report to
the Department of Interior, 1988.
[18] Pacific Sunday News. Palau Bridge Story, Guam, October 1996.
102 REFERENCES

[19] SSFM Engineers, Inc. Preliminary assessment of Koror Babelthuap Bridge Failure for the
United States Army Corps of Engineers, Honolulu, HI, 21 October 1996.
[20] Pilz, M. The Koror-Babeldaob Bridge in the Republic of Palau, History and Time Depen-
dent Stress and Deflection Analysis. Dissertation, Imperial College, Department of Civil
Engineering: London, June 1997.
[21] Tang, M.C. The collapse of the koror Bridge. Proceedings, FIB Congress, Osaka, Japan,
October 2002.
[22] McDonald, B., Saraf, V., Ross, B. A spectacular collapse: the Koror Babeldaob (Palau)
balanced cantilever prestressed post-tensioned bridge. Proceedings of the 27th Conference
on Our World in Concrete & Structures, Singapore, August, 2002.
[23] Burgoyne, C., Scantlebury, R. Why did Palau bridge collapse? The Structural Engineer
2006; 6: 30–36.
[24] Robertson, I.N. Correlation of creep and shrinkage models with field observations. The
Adam Neville Symposium, American Concrete Institute Special Publication SP-194, 2000.
[25] Bazant, Z.P., Hubler, M.H., Yu, Q. Excessive creep deflection – an awakening. Concrete
International 2011; 33(8): 44.
[26] Bazant, Z.P., Huber, M.J., Yu, Q. Pervasiveness of excessive segmental bridge deflections:
wake-up call for creep. ACI Structural Journal 2011; 108(6): 766–774.
[27] Bazant, Z.P., Yu, Q., Li, G.H. Excessive long-time deflection of prestressed box girder. I:
Record-span bridge in Palau and other paradigms. ASCE Journal of Structural Engineering
2012; 138(6): 676–686.
[28] Private correspondence with Qiang Yu, 2012.
[29] Private correspondence with Khaled Shawwaf, 2011, 2012.
[30] PCI. Prestressed Concrete Bridge Design Handbook. Prestressed / Precast Concrete
Institute: Chicago; 1998.
[31] fib, Model Code 2010, Final draft, fib Bulletin 65, The International Federation for
Structural Concrete, Paris/London, 2014.
[32] Bazant, Z.P., Yu, Q., Li, G.H., Klein, G.J., Kristek, V. Excessive deflections of record–span
prestressed box girdeer. Concrete International 2010; 32(6): 44–52.
[33] Bazant, Z.P., Yu, Q., Li, G.H. Excessive long-time deflection of prestressed box girder. II:
Numerical analysis and lessons learned. ASCE Journal of Structural Engineering 2012;
138(6): 687–696.
[34] Yu, Q., Bazant, Z.P., Wendner, R. Improved algorithm for efficient and realistic creep
analysis of large creep-sensitivie concrete structures. ACI Structural Journal 2012; 109(5):
665–675.
About the Author Case Studies
Objective:
To provide in-depth information to practicing
Man-Chung Tang, PE stuctural engineers in reports of high scientific
Dr.-Ing., Dr.-Ing. Eh, Hon. Dr. Litt., Hon. Architect, and technical standards on a wide range of
Hon. Professor structural engineering topics.
Member, National Academy of Engineering, USA,
IABSE Bulletin Editorial Board:
Foreign member, Chinese Academy of Engineering, PRC
J. Sobrino, Spain (Chair); H. Subbarao, India
Chairman of the Board, T.Y. Lin International (Vice Chair); M. Bakhoum, Egypt; C. Bob,
Romania; M. Braestrup, Denmark; M.G.
Bruschi, USA; R. Geier, Austria; N.P. Hoej,
Switzerland; S. Kite, Hong Kong; D. Laefer,
Ireland; R. Mor, Israel; H.H. (Bert) Snijder,
The Netherlands; R. von Woelfel, Germany.

Topics:
Structural analysis and design, dynamic analy-
sis, construction materials and methods, project
management, structural monitoring, safety
assessment, forensic investigations, mainte-
nance and repair, and computer applications.

Readership:
Practicing structural engineers, teachers,
researchers and students at a university level, as
well as representatives of owners, operators and
builders.

Publisher:
The International Association for Bridge and
Structural Engineering (IABSE) was founded
as a non-profit scientific association in 1929.
Today it has more than 3500 members in over
90 countries. IABSE’s mission is to promote
the exchange of knowledge and to advance the
practice of structural engineering worldwide.
IABSE organizes conferences and publishes
the quarterly journal Structural Engineering
International, as well as conference reports
and other monographs, including the SED,
Case Study series and Guidelines. IABSE also
presents annual awards for achievements in
structural engineering.

For further Information:


IABSE
c/o ETH Zürich
CH-8093 Zürich, Switzerland
Phone: Int. + 41-44-633 2647
Fax: Int. + 41-44-633 1241
E-mail: secretariat@iabse.org
Web: www.iabse.org
About the Author Case Studies
Objective:
To provide in-depth information to practicing
Man-Chung Tang, PE stuctural engineers in reports of high scientific
Dr.-Ing., Dr.-Ing. Eh, Hon. Dr. Litt., Hon. Architect, and technical standards on a wide range of
Hon. Professor structural engineering topics.
Member, National Academy of Engineering, USA,
IABSE Bulletin Editorial Board:
Foreign member, Chinese Academy of Engineering, PRC
J. Sobrino, Spain (Chair); H. Subbarao, India
Chairman of the Board, T.Y. Lin International (Vice Chair); M. Bakhoum, Egypt; C. Bob,
Romania; M. Braestrup, Denmark; M.G.
Bruschi, USA; R. Geier, Austria; N.P. Hoej,
Switzerland; S. Kite, Hong Kong; D. Laefer,
Ireland; R. Mor, Israel; H.H. (Bert) Snijder,
The Netherlands; R. von Woelfel, Germany.

Topics:
Structural analysis and design, dynamic analy-
sis, construction materials and methods, project
management, structural monitoring, safety
assessment, forensic investigations, mainte-
nance and repair, and computer applications.

Readership:
Practicing structural engineers, teachers,
researchers and students at a university level, as
well as representatives of owners, operators and
builders.

Publisher:
The International Association for Bridge and
Structural Engineering (IABSE) was founded
as a non-profit scientific association in 1929.
Today it has more than 3500 members in over
90 countries. IABSE’s mission is to promote
the exchange of knowledge and to advance the
practice of structural engineering worldwide.
IABSE organizes conferences and publishes
the quarterly journal Structural Engineering
International, as well as conference reports
and other monographs, including the SED,
Case Study series and Guidelines. IABSE also
presents annual awards for achievements in
structural engineering.

For further Information:


IABSE
c/o ETH Zürich
CH-8093 Zürich, Switzerland
Phone: Int. + 41-44-633 2647
Fax: Int. + 41-44-633 1241
E-mail: secretariat@iabse.org
Web: www.iabse.org
The Story of the Koror Bridge
IABSE Bulletins

The Story of the Koror Bridge


Koror Babeldaob Bridge, also called Koror Babelthuap Bridge
or simply Koror Bridge, connects the islands of Koror and
Case Studies
Babeldaob in the Republic of Palau. The design of the bridge
began in 1974 and was based on the prevailing AASHO
Standard Specifications at that time and was supplemented by
ACI and CEB-FIP design recommendations on an as-needed
basis. When the Koror Bridge was opened to traffic in April
1977, it was the world’s longest concrete girder span. A few
1
years later, the bridge began to deflect more than had been
anticipated. The owner commissioned a Japanese engineer-
ing firm in 1985 and then a US engineering firm in 1993 to
Man-Chung Tang
conduct in-depth investigations of the structure. Both firms

M.-C. Tang
came to the same conclusion that the bridge was structurally
safe and that the excessive deflection was an unexplainable
phenomenon. Nevertheless, in order to improve the driving
quality of the bridge deck, the owner decided to repair the
bridge. The repair scheme made changes to the structural
system and added a large amount of post-tensioning force The Story of the
to the bridge. Unfortunately, less than three months after the
repair, late in the afternoon on 26 September, 1996, nineteen
and a half years after it was opened to traffic, the bridge col- Koror Bridge
lapsed. Thereafter, most of the documents were sealed as a

1
result of litigation between the various parties and the debris
was cleared. For a long time, it was impossible to study the
facts surrounding the bridge’s collapse. Only recently, through
continuous probing by a group of engineers, were these docu-
ments made accessible to researchers.

IABSE Bulletins/ Case Studies

International Association for Bridge and Structural Engineering (IABSE)

You might also like