Professional Documents
Culture Documents
www.elsevier.com/locate/compstruc
Abstract
Existing test results of full-scale in-service masonry arch bridges are analysed to determine appropriate material
properties for the modelling of this structural type. Three-dimensional nonlinear ®nite element models of three masonry
arch bridges are generated using a commercially available ®nite element package. The behaviour of the masonry is
replicated by use of a solid element that can have its stiness modi®ed by the development of cracks and crushing. The
®ll is modelled as a Drucker±Prager material, and the interface between the masonry and the ®ll is characterised as a
frictional contact surface. The bridges are modelled under service loads, and the model results are compared to the
results of a program of ®eld testing of the structures. It is found that the assumption of a reasonable set of material
properties, based on visual observations of the material and construction of the structure, implemented through a
program of three-dimensional nonlinear ®nite element analysis enable good predictions of the actual behaviour of a
masonry arch bridge. Ó 2001 Elsevier Science Ltd. All rights reserved.
0045-7949/01/$ - see front matter Ó 2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 4 9 ( 0 1 ) 0 0 1 0 9 - 2
2646 P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662
2. Construction of masonry arch bridges most widely distributed and cited method for the analy-
sis of masonry arch bridges, the mechanism method,
A typical masonry arch bridge, the Grith Bridge in proposed by Heyman [3] is universally accepted by ac-
Dublin, Ireland, is shown in Fig. 1. Although much of ademic researchers and has almost no proponents
the structure is visible, a signi®cant part of the structure among bridge owners. However, an adaptation of the
is concealed. The main support at the abutment of the mechanism method has gained some currency among
bridge is only rarely available for inspection. The depth bridge owners in the British Isles for the capacity analy-
of the arch barrel visible on the face of the structure is sis of arch bridges.
not necessarily the same as the depth within the struc- Two wide-spread methods for the assessment of
ture. The properties of the ®ll are generally unknown, masonry arch bridges are currently available. The ®rst,
and the ®ll has been shown in testing to have a sub- known as the MEXE method, is a semi-empirical
stantial in¯uence on the ultimate strength of the struc- method, including multiple reduction factors based on
ture [2]. Nearly all stone arch bridges have stone backing conditions noted in a visual inspection.
behind the abutments in some form, either loose cob- The MEXE method is incorporated into the UK
blestones, or dressed, coursed unbonded masonry to some standard for the assessment of bridges [1]. The method
unknown depth within the ®ll. The structural properties is crude at best, and the rational basis of the reduction
of the backing are substantially dierent from the re- factors has never been well established. Other limita-
mainder of the ®ll, and the depth or extent of the back- tions of the method are well known. It is extremely
ing is rarely known in the course of the assessment of a conservative, and because the basic formula for allow-
stone bridge. able axle load is based on a parabolic arch, the
method becomes grossly over-conservative for shapes
of arch other than parabolic, especially segmental or
elliptical arches. The method is widely used because
3. Assessment of masonry arch bridges it is extraordinarily simple to apply, but it is gener-
ally unsuitable to the management of an important
Although analysis procedures for steel and concrete population of bridges because it will result in unwar-
bridges, such as determination of eective widths of ranted bridge postings, unwarranted rejections of heavy
concrete slab bridges or determination of girder distri- vehicle permits, and unwarranted strengthening of
bution factors for steel or prestressed concrete girders, bridges.
are ®rmly established, no such accepted procedure for The other commonly occurring method for ma-
the rational analysis of masonry arch bridges exists. The sonry bridges is a computer program known as A R C H I E
walls due to the self-weight of ®ll material are made masonry stiness, etc., were determined by adjustment
consistent with the coecient of lateral earth pressure at of the ®nite element model to match the testing results.
rest, k 1 sin u, by setting the Poisson's ratio of the Once a set of properties was determined, however, the
®ll material such that: same properties and support conditions were applied to
all three of the bridges. Modelling results, using the basic
1 sin u properties and procedures outlined in Section 7, showed
m :
2 sin u reasonable predictions of crown and abutment de¯ec-
tion and the ability of the three-dimensional ®nite ele-
The loading on a masonry arch bridge is a combi- ment modelling to predict dierences in the bridge
nation of self-weight loading and trac loading. The response due to dierences in bridge geometry (width,
proportion of loading due to the self-weight is signi®cant depth/span ratio, ring thickness, etc.). Some of the
and indeed much of the strength of these bridges is due modelling results are described by Fanning et al. [17].
to the stresses induced in the masonry material due to An important conclusion was also reached directly from
self-weight eects. In modelling this type of bridge an the testing program in that these visibly similar bridges
initial gravity-loading step is applied to generate the were found to have similar responses to load. As a re-
in situ stresses, both longitudinal and transverse, to sult of this program, the values of modulus of elasticity
which the bridge is subjected. Subsequent loading events, and compressive strength were recommended for use
such as the passing of a truck over the bridge, use the in modelling stone bridges [18] using an elastic frame
equilibrium solution from the gravitational load step as analysis representation of the mid-line of the arch barrel.
a set of initial conditions. It is dicult to approximate The strength values given in this reference are conser-
accurately the in situ stress state of a given bridge as the vative compared to the values recommended by Hendry
structure may have been subjected to a complex stress [19]. However the strength values were meant to be
history, which may have resulted in cracks that would conservative for rating by an approximate rational
alter its response. In these circumstances a visual in- procedure. The modulus of elasticity values are used in
spection may be useful in identifying signi®cant struc- the modelling described in this paper, and larger
tural cracking that can subsequently be included in an strength values are used, similar to Hendry's recom-
initial load step simulation prior to modelling of any mendations, as this paper describes a more accurate
trac event. analysis method.
The boundary conditions for a masonry arch bridge A three-span stone arch bridge of rough random
are a result of the bridge itself and the ®ll material stone construction, located in Adams County, Pennsyl-
interacting with the surrounding soil medium. In ap- vania, USA, was tested in August 1998. A photograph
plying boundary conditions to numerical models, it is of the bridge is shown in Fig. 5. Measurements taken
important that additional stresses are not induced due to under static truck loading included crown de¯ections
over-constraining the model. In general terms, the ®ll in the main span, both at the roadway centreline and
material should be considered to provide longitudinal one of the edges, and crown de¯ections in one of the
restraint to the arch barrel, while the base of the arch approach spans. Horizontal longitudinal displacements
and the spandrel walls are supported vertically. The of a pier and abutment were also measured, as was the
speci®c boundary conditions for the three dimensional transverse lateral displacement of the crown of the arch
models of the bridges under study are discussed in Sec- and of the spandrel walls over one of the piers. This set
tion 7 of this paper. of measurements was taken for two load levels: one from
an empty truck and one from a full truck. The results of
this testing program are presented in the following sec-
6. Investigation: testing tion along with the modelling results.
Two very similar elliptical arch canal bridges in
A series of similar stone arch bridges was tested Dublin, Ireland were subjected to truck loads to de-
in Ohio, USA, by the second author. The results are termine measurable responses. Photographs of these
described in detail in Refs. [8,16]. In this study, three bridges are shown in Figs. 1 and 6 respectively. The
bridges of similar geometry, time of construction, and Killeen Road Bridge has a span of 9.23 m, a rise over the
materials were tested within the space of one week. The abutments of 2.67 m, a width of 6.30 m, and an arch ring
bridges were tested by construction of a reference frame thickness of 48 cm. The arch ring is constructed of
beneath the bridge structure, installing linear variable limestone on the face and in the barrel, with joints about
dierential transformers (LVDT) on the reference frame 1 cm thick. The spandrel walls are also of ashlar lime-
to measure displacements of the structure, and loading stone construction, with joint thickness of approxi-
the structure with a vehicle of known weight. Other than mately 1 cm. The bridge has suered extensive damage
the basic geometry of the structures, the modelling pa- in the form of large longitudinal cracks in the haunches,
rameters such as support conditions, ®ll, properties, which were repaired about ®ve years prior to the testing
2650 P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662
Fig. 5. John's Burnt Mill Bridge, Adams County, PA Bridge no. 56.
of the structure. The Grith Bridge has a span of 9.49 construction, with joint thickness of approximately
m, a rise over the abutments of 2.67 m, a width of 7.85 1 cm.
m, and an arch ring thickness of 45 cm. The arch ring is For the Killeen Road Bridge, DCLVDTs having a
constructed of granite on the face, while the remainder linear range of 1:27 mm, mounted on a purpose-built
of the arch ring is limestone, with joints about 0.5 cm reference frame underneath the structure, were used to
thick. The spandrel walls are also of ashlar limestone measure displacements. A DCLVDT mounted in a
P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662 2651
similar installation is depicted in Fig. 7. The resolution the bridges being based on visual inspections and re-
of these analog instruments is practically in®nite, but commendations outlined by Boothby [18].
limited by the quantization of the data acquisition sys- The choice of element to be used for the arch barrel
tem and electronic noise. A 16-bit analog/digital con- and the spandrel walls is limited in A N S Y S V 5 . 5 because
version board was used in data acquisition resulting in a the smeared crack model is only available as an option
resolution of 0:04 lm over the instrument range. The for the Solid65, eight noded isoparametric solid element.
root mean square of the electronic noise is estimated to Eight-noded isoparametric elements, with a Drucker±
be about 2 lm, but the large number of samples taken Prager material model were used to de®ne the ®ll ma-
mitigates this error considerably. Absent a full statistical terial. The ®nite element mesh employed for Adams
analysis of the experimental data, 1 lm can be taken as County Bridge is shown in Fig. 8.
a practical limit on the resolution of the instruments in In each case, an elevation of the bridge geometry was
this testing program. Displacement measurements were constructed by photogrammetry from digital photo-
taken at the centreline of the roadway at the crown, the graphs of the bridges and extruded in the transverse
two abutments, designated north and south, and at one direction to generate the solid model. In all three cases,
of the haunches. The bridge was loaded with a two-axle loading was symmetrical about the centreline of the
truck, full, half-loaded, and empty. The weights of the bridge; hence only a half model was generated. The ex-
truck and the peak abutment, crown and haunch dis- tent of the ®ll model was limited to that amount of
placements are displayed in Table 1 (Panel a). A more material contained between the spandrel walls. In all
extensive instrumentation and testing program was cases the thickness of the spandrel walls was measurable
conducted at the Grith Bridge, with the instruments at the road surface and was assumed constant over their
mounted on a professionally assembled scaolding sys- full heights. The depth of the arch barrel visible on the
tem. Truck weights and peak displacements at the face of the structure was assumed to be constant across
centreline of the bridge are shown in Table 1 (Panel b). the width of the bridges. The ®ll material was assumed
Additional displacements measured at abutment, haunch, to be uniform through its depth and no backing to the
and crown on the edge of the bridge, which will be arch ring or internal buttressing of the arch was con-
discussed in the next section in connection with the sidered.
modelling of the structure, were also recorded. The masonry and ®ll material properties used in each
of the models are documented in Table 2, and follow
the general guidelines given in Table 3, which are in
7. Modelling results turn based on the recommendations of Boothby [18]. The
Adams County Bridge (Fig. 5) consists of random rub-
Finite element models of the Adams County Bridge, ble limestone masonry construction with wide joints. In
Killeen Road Bridge, and Grith Bridge, were con- recognition of the coarseness of the construction, a value
structed. Consistent modelling techniques were used for of 1.5 GPa, was used for the modulus of elasticity of the
each bridge with material properties for the masonry of masonry material. The compressive and tensile strengths
2652 P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662
Table 1
Peak displacements of Killeen Road and Grith Bridges (Panels a and b)
Truck weight (kg) Abutment peak displace- Haunch peak displacement Crown peak displacement
ment (lm) (lm) (lm)
Test FEM Test FEM Test FEM
Panel a
Killeen Road Bridge
10,700
F <10 <15 73/82 96 131/127 148
R <10 <15 71/87 89 124/145 155
14,800
F <20 <20 86/98 107 204/212 162
R <20 <20 125/134 143 259/342 228
23,600
F <10 <10 114/113 137 228/240 190
R 32 35 237/268 255 510/552 387
Panel b
Grith Bridge
Truck A 16 ± 117 ± 156 ±
13,500
Truck A 48 ± 374 ± 509 ±
30,000
Truck B 32 78 172 237 270 319
21,100
Truck B 36 ± 252 ± 351 ±
26,100
Truck B 41 48.3 331 387 429 544
31,200
F: Front axle, R: Rear axle.
Table 2
Material properties for ®nite element models
Young's modu- Poisson's ratio Density Tensile strength Compressive
lus (GPa) (kg/m3 ) (MPa) strength (MPa)
Masonry
Adams 1.5 0.3 2200 0.2 6
Killeen 10 0.3 2200 0.5 10
Grith 10 0.3 2200 0.5 10
Cohesion Angle of friction Angle of dilatancy
(MPa) (deg) (deg)
Fill
3
Adams 15 0.3 1700 1 10 35 20
3
Killeen 15 0.23 1700 1 10 44.43 44.43
3
Grith 15 0.23 1700 1 10 44.43 44.43
the dierence between the applied loading and the in- 7.1. Adams County Bridge results
ternal forces generated, was calculated. The solution was
deemed to have converged once the out of balance force The axle load pattern for the empty and full truck
was less than 0.5% of the applied loading. The ANSYS used in the testing program is illustrated in Fig. 9(a),
solution algorithm also includes automatic time stepping along with the approximation of these loads used in the
which enables automatic reduction and expansion of the ®nite element model. The loads were applied as pres-
substep size used in each load step dependent on the sures to elements of the sizes shown in the ®gure. Dur-
severity of the nonlinear response. Converged solutions ing the development of the model, some adjustments
using a maximum of 20 substeps per load step, 26 itera- were made for the full truck, static load only, in order
tions per substep and a 0.5% force convergence value to match abutment and crown displacements recorded
were obtained for all service load tests simulated. The during the testing program. These adjustments were
displacements of unconverged solutions were inspected primarily the reduction of the masonry stiness from 3
to determine whether the nonconvergence was due to GPa to the ®nal value used and the addition of exten-
a global collapse of the structure, a local collapse, or sions of the piers and abutments below the springing
numerical instability. line. The empty truck results reported were obtained
Self-weight eects, used for the initial in situ stress without further modi®cations to the model. A summary
state, were accounted for by specifying acceleration of the results for the full and empty truck loading is
due to gravity to the model in an initial loadstep. No provided in Table 4.
eorts were made to model the complex stress history, or Figs. 11 and 12 show the comparison of displacement
shakedown, that each of these bridges would have ex- results for the static loading of the bridge at midspan of
perienced prior to simulating the response of a truck the main span. To obtain good matching results, it was
passing over the bridges. However, subsequent results necessary to incorporate some eect of previous loading
are reported for the ®rst passage and later crossings by loading and unloading the bridge with a moving load
of a vehicle, providing a simple way of accounting for equivalent to the loaded truck. Fig. 10 shows the results
some of these eects. from the initial load pass for the main span at the crown,
The axle patterns and weight distributions for the both at the centreline of the bridge and at the two edges,
trucks used in the ®eld tests are described in Fig. 9(a)± while Fig. 11 shows the results for the second passage
(c). Appropriately distributed nodal forces or element of the same loading. All other results shown are for the
surface pressures applied to the roadway surface were second load pass. Fig. 12 shows the modelled and mea-
used in the ®nite element models to represent the weight sured crown de¯ections of one of the approach spans;
distributions. In each case, simulation of the moving Fig. 13 shows the response of the pier and the abutment.
vehicle, which travelled typically at less than 20 km/h, Fig. 14 exhibits the measured lateral responses at the
was undertaken by a sequential series of static loadsteps crown of the arch barrel in the main span and at the ®rst
with the equilibrium solution at the end of one loadstep interior pier. The model predicts ± and experimental
resulting in a set of initial conditions for the subsequent results con®rm ± the presence of small lateral move-
load step. ments at the pier and at the crown of the main span.
An important aspect of the numerical models is the These movements are indicative of the transverse eects
representation of the interfaces between the masonry in the bridge, which give rise to signi®cant transverse
and the ®ll material. In constructing the models, con- tensile stress. The model predicts formation of cracks in
tact surfaces were included at these interfaces so that the arch barrel in line with the inside face of the spandrel
movement of the ®ll material relative to the masonry wall, which are also observable on the bridge structure.
was possible. This proved particularly important in
generating the initial stress state. Under gravity load-
ing, both the masonry and the arch settle. The low 7.2. Grith Bridge results
stiness of the ®ll relative to the arch results in greater
settlement of the ®ll material. Unless a contact inter- The de¯ected shape of the arch barrel and spandrel
face is provided, the ®ll material tends to hang from walls when Truck B, fully loaded, has its rear axle at
the surrounding masonry, generating signi®cant tensile mid-span is plotted in Fig. 15. The de¯ected shape of the
stresses in the cohesionless ®ll and resulting in di- bridge demonstrates the important three-dimensional
culties in converging to an equilibrium solution. In the eects that contribute to the strength and stiness of the
absence of a coecient of friction, the contact surfaces bridge. The maximum de¯ection occurs at the centreline
slide freely relative to each other. Although the service of the bridge with the spandrel walls stiening the outer
load response was found to be insensitive to the level of edges of the arch barrel. These transverse eects are
friction speci®ed, a value of 0.4 for the coecient of evident in the three bridges analysed and are considered
friction between the masonry and the ®ll was speci®ed to be the source of longitudinal cracking that is evident
in the analyses. in many arch bridges. The formation of these longitu-
P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662 2655
dinal cracks and their location (either at the spandrel The numerical and experimental responses at mid-
walls or nearer the centreline of the bridge) is primarily span, at the centreline of the bridge, as the fully loaded
due to the relative stiness of the barrel in the transverse truck passes over the bridge are plotted in Fig. 16. The
direction and the spandrel walls and the tensile strength maximum de¯ection measured in the test was 0.43 mm
of the masonry. compared to 0.54 mm predicted by the ®nite element
Fig. 9. Truck axle patterns: (a) Adams County Bridge, (b) Grith Bridge and (c) Killeen Road Bridge.
2656 P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662
Fig. 9. (continued)
Fig. 10. Adams County Bridge ± main span results, ®rst load pass.
Fig. 11. Adams County Bridge ± main span results, second load pass.
A second pass of the same truck results in an increase in higher under ``Pass 2'' conditions as a result. The model
the numerical response, shown in Fig. 20(a) and (b). results, ``Pass 2'', are consistent with the test results and
The numerical and test responses for empty and half the fact that the bridge had been repaired due to lon-
loaded truck are plotted in Figs. 22 and 23 respectively. gitudinal cracking. The peak de¯ections measured dur-
Two sets of numerical data are included. In both cases ing testing and predicted by the numerical models are
``Pass 1'' represents the response of the model when not summarised in Table 1.
subjected to prior loading. The results labelled as ``Pass The study of the test and numerical responses for
2'' are those predicted when both the empty and half-full Killeen Road Bridge highlight the complexity of ma-
truck traverse the bridge after the fully loaded truck. sonry arch bridges and also the importance of transverse
The fully loaded truck, which causes the longitudinal eects in determining their responses. A bridge that has
cracks plotted in Fig. 21, reduces the strength and sti- been in service for a signi®cant period of time may have
ness of the bridge and the predicted de¯ections are been subjected to loading that require that the presence
2658 P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662
Displacement (mm)
of longitudinal cracking be included in a predictive ®ll and the masonry requires consideration of contact
model. However if longitudinal cracking can be detected stiness, and the probability of sliding at vertical inter-
by in situ inspection, as was the case for Killeen Road faces. The distribution of ®ll pressures to the arch barrel
Bridge, the response of the bridge system remains pre- produces signi®cant transverse bending within the arch
dictable once this cracking is represented, either by a ring and complex bending of the spandrel walls, in ad-
discontinuity in the model or by an initial loading event. dition to the characteristic response of the arch. The
transverse bending may induce cracking in the arch
barrel. Cracking due to transverse bending modi®es the
8. Conclusions stiness of the arch and may alter its response to load-
ing.
Masonry arch bridges under truck loading exhibit a Commercially available three-dimensional ®nite ele-
complex three-dimensional response. The distribution of ment routines can be implemented to predict these ef-
tire pressures through the ®ll requires consideration of fects. However, it is necessary to incorporate nonlinear
three-dimensional eects. The interaction between the response of the ®ll material, cracking and crushing of the
P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662 2659
Fig. 16. Grith Bridge ± numerical and test crown displace- Fig. 18. Grith Bridge ± numerical and test haunch displace-
ments for fully loaded truck. ments for fully loaded truck.
2660 P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662
Fig. 20. Killeen Road Bridge ± haunch and crown de¯ections Fig. 22. Killeen Road Bridge ± haunch and crown de¯ections
measured along centreline (full truck): (a) haunch de¯ections measured along centreline for empty truck: (a) haunch de¯ec-
and (b) crown de¯ections. tions and (b) crown de¯ections.
masonry, and contact and possible sliding at the ®ll- In spite of the complexity of some of the unknown
masonry interface to arrive at a model that reproduces factors in assessment of a masonry arch bridge, such as
the response of a bridge with reasonable ®delity. overall thickness of the arch barrel, nature of the ®ll
P.J. Fanning, T.E. Boothby / Computers and Structures 79 (2001) 2645±2662 2661
References
[17] Fanning PJ, Boothby TE, Roberts BJ. Longitudinal and [20] Melbourne C, Begimgil M, Gilbert M. The load test to
transverse eects in masonry arch bridge assessment. collapse of a 5 m span brickwork arch with tied spandrel
Construct Building Mater 2001;15:51±60. walls. Arch Bridges. London: Thomas Telford; 1995.
[18] Boothby TE. Load rating of masonry arch bridges. p. 509±18.
J Bridge Engng 2001;6(2):79±86. [21] Melbourne C, Walker PJ. Load tests to collapse of
[19] Hendry AW. Masonry properties for assessing arch model brickwork masonry arches. Proceedings, 8th Inter-
bridges. Department of Transport, Transport and Road national Brick and Block Masonry Conference, vol. 2.
Research Laboratory Contractor Report 244. Transport New York: Elsevier Applied Science; 1988. p. 991±
Research Laboratory, Crowthorne, England: 1990. 1002.