Professional Documents
Culture Documents
Abstract
The objective of this work was to develop a method enabling the optimal design of arch bridges based on their
modal characteristics. The relationship between the resonant frequencies, a function of mass and stiness, and the
load carrying capacity of the arches was investigated. Natural frequency and dynamic direct implicit time
integration analyses were performed to determine the resonant frequencies and analyse the responses to impact
loading. Their collapse loads, under dierent loading regimes, were also investigated using elasto-plastic non-linear
®nite element analysis. The collapse loads were found to be related to the resonant frequencies and an optimal
design could therefore be achieved. Under vertical loading, arches with span to rise ratios between 4 and 5 had both
the highest resonant frequencies and collapse loads. The proposed method is simple and applicable to other
structures as a design optimization tool. # 1998 Elsevier Science Ltd. All rights reserved.
The usual modal analysis of an undamped ®nite el- For this analysis the steel comprising the arches was
ement system was carried out by performing the sol- assumed to be in a state of plane stress. It was mod-
ution of the classical eigenvalue problem presented in elled as an elasto-plastic continuum with yield being
Eq. (1). governed by the von Mises criterion [11] whilst obeying
A. Bensalem et al. / Computers and Structures 68 (1998) 461±472 463
the normality rule [12]. In terms of principal stresses, If no strain hardening is allowed, dierentiating the
s1 and s2, the yield criterion can be represented in yield criterion gives Eq. (11) below.
dimensionless form by Eq. (3). @F @F @F
2 2 dF dsx dsy dtxy 0
11
s1 s1 s2 s2 @sx @sy @txy
ÿ 1
3
sy sy sy sy This may also be expressed in the form given by
If strain hardening is taken into account then the uni- Eq. (12).
axial yield stress, sy increases with plastic strain. aT ds 0
12
3.3.1. Incremental stress±strain relationships Multiplying both sides of Eq. (8), the expression for
In the following section the salient equations dE, by aT D, gives Eq. (13).
involved in the adopted elasto±plastic constitutive aT DdE aT DDÿ1 ds aT D dl a
13
model are summarised.
T
The total strain increment dE is the sum of dEe repre- Noting that, according to (12), a ds is equal to zero
senting the elastic component of the strain and dEp and that D[D]ÿ1=1, the following expression, Eq. (14)
representing the irreversible plastic component of the may be obtained for dl:
strain as shown in Eq. (4).
aT DdE
dl
14
dE dEe dEp
4 aT Da
The elastic strain increment is calculated in the conven- Knowing the value of dl the incremental stress±strain
tional way as Eq. (5) below. relationship in the plastic region is given by Eq. (15).
dEe Dÿ1 ds
5 DaaT D
ds D ÿ T dE
15
a Da
Here, ds is the incremental stress vector and D is the
elasticity matrix. In this formulation, it is assumed that A bi-linear stress±strain relationship was adopted and
the material is isotropic and the familiar equations in an incremental iterative procedure was used [12, 13]. A
terms of Young's modulus and Poison's ratio are used load factor of 0.5 of the applied load of 42 kN was
to calculate each element of the D matrix. Using the used at the ®rst increment of each analysis. This incre-
normality rule, a relationship between plastic strain mental load factor was applied for the other incre-
increment, dEp and the corresponding stress increment, ments and automatically reduced if convergence was
ds is obtained. Following on from this the incremental not reached at subsequent increments. This was
plastic strain dEp is normal to the yield surface at the achieved by using the previous incremental load factor
point considered, and is given by Eq. (6). and storing the current state of the arch, e.g. stiness
matrices, for the converged increment. A maximum of
dEp dla
6
30 iterations at each increment were allowed in all ana-
We can also further de®ne the yield function as shown lyses. Using dierent values for the tolerance of 5, 4, 3,
in Eq. (7). 2 and 1%, it was found that the results were similar. A
tolerance of 2% was adopted. The element stiness
@F @F @F @F
aT ; ;
7 was computed at the ®rst iteration of the ®rst incre-
@s @sx @sy @txy ment and at each of the subsequent iterations if con-
The yield criterion function, F de®nes the yield surface vergence was not obtained in the ®rst iteration.
and dl is a positive constant of proportionality, as yet
unde®ned, known as the plastic multiplier. Using the
above equations, the total strain increment dEp can be 4. Results
written as shown in Eq. (8) with the total stress incre-
ment then calculated by Eq. (9). The following sections presents the results of the
three analyses carried out namely: natural frequency
dE Dÿ1 ds dl a
8 analysis, implicit dynamic analysis, and non-linear sta-
tic analysis for collapse load determination.
ds DdE ÿ dlD a
9
The yield function can be expressed in terms of stress, 4.1. Natural frequency analysis
s and strain hardening parameter, h. This is shown by
Eq. (10). The mode shapes of the four lowest modes of vi-
bration are presented in Fig. 1 for arches having
F
s; h f
s ÿ k
h 0
10 s = 2, 4, and 8. The ®rst and second modes of
464 A. Bensalem et al. / Computers and Structures 68 (1998) 461±472
vibration were independent of span to rise ratio, s: form of vibration underwent mode conversion to
they are longitudinal and vertical single bending modes double vertical bending as s increased. Fig. 2 shows
respectively. The third mode of vibration for s = 2 the resonant frequencies of the arch rings in relation to
and s = 3 was a double longitudinal bending. This their span to rise ratios. This shows that the resonant
The ®rst mode of vibration was not seen in the re- the strip of 84 kN where yielded zones covered most
sponse signals. This was mainly because both the exci- of the arch. The onset of yield started at similar loads
tation force and measured response were in the vertical implying that stier arches having 4 R sR 5 behaved in
direction. To excite this mode the force and response a more ductile manner. This is desirable in practice as
must have a horizontal component. Purely vertical it leads to a safer design avoiding the inherent dangers
input excitation would do no work in an orthogonal of brittle failure.
direction and could not therefore excite any associated The particular correlation between stiness and load
modes of vibration. carrying capacity implicit in Fig. 6(a) is only applicable
for this geometry. Future research to develop design
4.3. Non-linear static ®nite element analysis for collapse charts covering a wider range of parameters;
load estimation span:thickness, span:width may be worthwhile. The
correlation discussed here would, in general terms
The arches' load carrying capacities were investi- apply universally but in details would change.
gated using a non-linear ®nite element model. The
arches were loaded with: a strip, or line load at the 4.3.2. Vertical strip, or line load at one 1/4-span point
crown, a strip, or line load at one 1/4-span point, or a A strip, or line load was applied vertically to one 1/
uniformly distributed vertical load. The results are pre- 4-span point. The results presented in Fig. 7 show that
sented in the following sections. the collapse loads were practically independent of span
with the exception of the semicircular (s = 2) case,
4.3.1. Vertical strip, or line load at the crown where the collapse load was higher than that for the
The arches were loaded by a strip load applied verti- remaining arches. This was expected because vertical
cally at the crown. Consideration of symmetry allowed load was better transmitted to the nearest support for
half the arch to be analysed. Fig. 6(a) shows the load± lower span to rise ratios. The dierence between col-
de¯ection plots for the arches. As s increased to 5 the lapse loads over the range 2 < s < 8 was only 13%.
collapse load increased; for s>5 the collapse load As mentioned in Section 4.3.1 above, the relationships
decreased. This re¯ected the ®ndings obtained using implied by Fig. 7 depend, for their numerical values,
the natural frequency and dynamic time integration upon geometrical factors, e.g. the span : width and the
analyses. It should be noted that s = 4 and s = 5 ring thickness.
arches had very similar collapse loads. Their corre-
sponding resonant frequencies for the vertical bending 4.3.3. Horizontal strip, or line load at one 1/4-span
mode were almost equal. The collapse load increased point
by 30% when s increased from 2 to 5. Arches with A horizontal strip, or line load was applied at a 1/4-
4 R sR 5 had stier load±de¯ection plots. Fig. 6(b) span point. As shown in Fig. 8 there was a signi®cant
shows the load±strain curves for arches' intrados dierence between the collapse loads of the arches as s
crown point; again it can be seen that the s = 5 arch increased from 2 to 8. The highest collapse load, under
was stier. The semicircular (s = 2) arch was the least this regime, was obtained for the s = 8 arch. The same
sti most notably above an applied total force along span to rise ratio had a higher resonant frequency for the
A. Bensalem et al. / Computers and Structures 68 (1998) 461±472 467
Fig. 5. Instantaneous spectra: responses measured at the crown (a) Ratio, s = 2 to s = 5; (b) Ratio, s = 6 to s = 8.
®rst mode of vibration which was horizontal bending. supports as s increased. In practice, such horizontal
Compared to the semicircular (s = 2) arch, the collapse forces can be generated on arch bridges by braking or
load of the s = 4 arch was almost doubled. For the cornering vehicles. Design charts, based on parametric
s = 5 arch the capacity tripled over that obtained for the studies involving load and dierent arch geometries
s = 2 case as shown in Fig. 8. The resonant frequency of should, for future research, be derived from this analysis.
the horizontal bending mode for the s = 5 arch increased This could provide a 3D design surface version of Fig. 8
by 120% compared to that obtained for the s = 2 arch where stiness and capacity eects could be correlated
as was shown in Fig. 2. The increase in collapse load was over a wider range of variables than were possible within
due to the way the applied forces were transmitted to the this paper.
468 A. Bensalem et al. / Computers and Structures 68 (1998) 461±472
Fig. 6. (a) The eect of a vertical concentrated load at the crown. (b) Load±strain curves at the crown under vertical concentrated
crown load.
4.3.4. Uniformly distributed vertical load over the range 3 < s < 4: the collapse load for the
A uniformly distributed vertical load was applied to s = 3 arch was approximately 50% higher than that
the arches. Fig. 9 shows the load±de¯ection curves. The obtained for the semicircular (s = 2) arch. The arches'
lowest collapse loads were obtained for the s = 2 and load carrying capacities were all higher than those perti-
s = 8 arches. The highest collapse loads were obtained nent to vertical line, or strip loading at the crown only;
A. Bensalem et al. / Computers and Structures 68 (1998) 461±472 469
this was shown in Fig. 6. This was due to the con®ne- soil back®lling on the system's behaviour under both
ment of the thrustline within the arch under the in¯u- dynamic and static load. This study could be extended
ence of the uniformly distributed load. In the absence of to other materials such as reinforced concrete and
this uniformly distributed load the eccentricity of the brickwork, blockwork, or stonework masonry which
thrustline has more scope to deviate from the arch's neu- are extensively for arch bridges, tunnels, culverts, and
tral axis precipitating earlier onset of the eventual col- vaulted structures.
lapse mechanism. It should be emphasized that although the ®ndings
From the analyses conducted, arches with span to of this paper were obtained solely from steel arches,
rise ratios between 4 and 5 can be adopted, as they the analysis can be extended to arches made with other
provided the highest load carrying capacities, lower materials as long as the loading levels stay within the
de¯ections, and less mass of material used in their con- serviceability limit. An extensive experimental testing
struction. Large cost savings on raw material alone programme is under way at Napier University to study
could be made by adopting the right choice of span to the method's applicability to arches constructed with
rise ratio for accurate construction. This optimization brickwork and concrete materials, both experimentally
method proved useful in reaching such design de- and numerically.
cisions. The method was simple and required only pre- Currently available data enables valuable compari-
liminary natural frequency analysis which was very sons to be drawn with the theoretical results presented
cheap in terms of computer processing time. For refer- here. Small scale tests to collapse on timber arches
ence, the natural frequency analysis occupied approxi- over 0.75 m span were carried out by Fair®eld and
mately 5 and 7% of the total CPU time used for the Ponniah [16]. Semicircular (s = 2) pro®les had collapse
200 time steps required for the dynamic, and non-lin- loads of only 75% of those pertaining to a ¯atter
ear static analyses, respectively. s = 4 arch for load points above one springer. For 1/
4-span loading the s = 2 arch had only between 20
and 45% of the load carrying capacity of an s = 4
5. Comparisons with Currently Available Data arch under identical load. For each pro®le tested mini-
mum collapse loads were found for loading at points
An experimental programme will be necessary to slightly closer to the crown than one 1/4-span point.
verify the ®ndings. It will also be of interest to study This was in accordance with the ®ndings presented
the eects of spandrel walls, and possible compacted from the ®nite element analysis on bare steel arches.
A. Bensalem et al. / Computers and Structures 68 (1998) 461±472 471
Arch model tests to collapse, at various scales, have capacities of such arches. A span to rise ratio between 4
been carried out by Hendry and a wide range of co- and 5 resulted in higher collapse loads under concen-
workers [17]. The failure loads are not directly compar- trated vertical loads as would be the case in practice. It
able as consistency of dimensions in¯uencing both fail- is not yet clear whether this increase was due to better
ure mechanism and collapse load were not achieved transmission of load to the supports or an increase in
between test sets. Dierent arch ring thicknesses or ®ll overall stiness and hence, capacity. An experimental
depths were used for each span to rise ratio making study is needed to verify the theoretical ®ndings. In this
normalization of the data for comparative purposes way, an optimum design can be achieved by balancing
extremely dicult. However, changing the ring thick- the total weights of the arches, which depend on their
nesses of Hendry's arches to render them consistent shapes, and the desired stiness. This optimization
for the three s values tested (s = 6.4, 3.4, and 2) and method is simple and requires only natural frequency
running the data through a typical mechanism package analysis of the structure. The method can be generalized
(ARCHIE) [18] allow better comparison of the collapse and extended to other types of structures such as: tun-
loads. The following are based on line loading across nels, curved beams, truss structures, culverts, or brick-
the span at one 1/3-span point. At s = 6.4, collapse work sewers, to assist engineers making preliminary
loads in the range 1.75 kN to 2.00 kN were recorded design decisions.
for the 0.038 m thick bare arch ring. This is an actual
experimental value [17] and no attempt at normaliza-
tion has been made. At s = 3.4 the collapse load was
between 2.6 kN and 2.7 kN for the same ring thick-
ness. At s = 2 the collapse load was between 1.1 kN References
and 1.2 kN for the same ring thickness. Data for
s = 3.4 and s = 2 were generated by ARCHIE [18] in [1] Hinton E, Afonso SMB, Rao NVR. Some studies on the
an attempt to compare dierent sets of experiments. optimisation of variable thickness plates and shells.
Whilst not wishing to ascribe undue accuracy to ana- Engng Comp 1993;10:291±306.
lytically derived collapse load predictions, the trend of [2] Hinton E, Osakca M, Rao NVR. Structural shape
maximum collapse load occurring at span to rise ratios optimisation of vibrating shells and folded plates using
between s = 3 and s = 4 is borne out by the data. two-noded ®nite strips. Engng Comp 1993;10:139±57.
This compares well with the authors' ®nite element [3] Bhatt P, Bensalem A. Behaviour of a reinforced con-
analysis. crete ¯at slab over the column support designed using
nonelastic stress ®elds. Struct Engng Rev 1996;201±12.
As a ®nal comparison the authors chose optimiz-
[4] Bhatt P, Bensalem A. Behaviour of reinforced concrete
ation data presented by Tadjbakhsh [19]. Adopting slabs designed using non-elastic stress ®elds. In:
Euler±Bernoulli buckling theory Tadjbakhsh aimed to Computational methods & experimental measurements.
produce optimized, momentless arch pro®les. In his Elsevier Science, Oxford, 1991:471±480.
buckling equations maximum buckling loads arose for [5] Bushnell D, Bushnell WD. Approximate method for the
pro®les in the range 2.7 < s < 3.3. The buckling load optimum design of ring and stringer stiened cylindrical
at s = 2 was 85% of the maximum at 2.7 < s < 3.3. panels and shells with local, inter-ring, and general buck-
These results are similar to those arising from the ling modal imperfections. Comp & Struct 1996;59:489±
authors' optimization presented above. Whereas 527.
Tadjbakhsh's work [19] continued by producing [6] Kang KJ, Bert CW, Stritz AG. Vibration and buckling
analysis of circular arches using DQM. Comp & Struct
momentless pro®les, these were felt to be impractical
1996;60:49±59.
and undesirable. Simple steel sections would suce for [7] Bathe KJ. Finite element procedures in engineering
the near-optimum pro®les presented where the key analysis. Prentice±Hall, Englewood Clis, New Jersey,
variable was span to rise ratio. Reliance on a design 1982.
being momentless may be risky given inherent con- [8] Ramirez RW. The FFT fundamentals and concepts.
struction problems such as member lack of ®t on site, Prentice±Hall, Englewood Clis, New Jersey, 1985.
temperature changes, and basic tolerances on section [9] Sibbald A, Bensalem A, Fair®eld CA. NDT of arch
manufacture. bridges. Insight, J Brit Inst NDT 1995;37:864±70.
[10] Bensalem A, Fair®eld CA, Sibbald A. Non-destructive eval-
uation of the dynamic response of a brickwork arch.
Proc Instn Civ Engnrs J Struct & Bldg 1997;122:69±82.
6. Conclusions
[11] Slater RAC. Engineering plasticity: theory and appli-
cation to metal forming. Macmillan, London, 1977.
The eect of dierent span to rise ratios both on the [12] Owen DRJ, Hinton E. Finite elements in plasticity: the-
modal characteristics and the collapse load were success- ory and applications. Pineridge Press, Swansea, 1980.
fully studied for arch rings. A relationship was found [13] Elnashai S, Izzudin BA. Modelling of material non-line-
between the resonant frequencies and the load carrying arities in steel structures subjected to transient dynamic
472 A. Bensalem et al. / Computers and Structures 68 (1998) 461±472
loading. J Earthquake Engng & Struct Dynamics. [16] Fair®eld CA, Ponniah DA. Model tests to determine the
22:1993:509±532. eect of ®ll on buried arches. Proc Instn Civ Engnrs J
[14] Bensalem A, Fair®eld CA, Sibbald A. Non-destructive Struct & Bldg 1994;104:471±82.
testing for condition based maintenance of arch bridges. [17] Hendry AW. Structural masonry. Macmillan, London,
Proc 8th Int Cong Condition Monitoring & Diagnostic 1990:246±263.
Engng Management 1995;2:503±9. [18] Smith FW. Load path analysis of masonry arch bridges.
[15] Bensalem A, Fair®eld CA, Sibbald A. The response of PhD thesis. Dundee Univ., 1991.
brickwork arches to dynamic loading: a parametric [19] Tadjbakhsh IG. Stability and optimum design of arch-
study. In: Proc Int Conf Structural Dynamics, Vibration, type structures. Int J Solids & Struct 1981;17:565±
Noise and Control, SDVNC'95 1995;1:209±14. 574.