You are on page 1of 11

Case Study

Application of Topological Optimization to Bridge Design


B. Briseghella1; L. Fenu2; C. Lan3; E. Mazzarolo4; and T. Zordan5

Abstract: Recently, structural optimization has become an important tool for structural designers, because it allows a better exploitation of
material, thus decreasing a structure’s self-weight and saving material costs. Moreover, structural optimization helps the designer to find
innovative design solutions and structural forms that not only better exploit material but also give the structure greater aesthetic value from an
architectural point of view. In this article, the seismic retrofitting of a bridge originally designed in reinforced concrete is illustrated, showing
how lightening the bridge superstructure, rather than reinforcing the already completed foundations and abutments, allowed these latter features
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

to resist greater seismic actions as required in the recent update of the Italian seismic code. Therefore, besides using the steel-concrete composite
typology, the bridge superstructure was lightened through structural optimization. After having optimized the thickness of webs and flanges, it
was necessary to further lighten the bridge superstructure by removing unexploited material from the bottom flange through the insertion of large
cavities. For this purpose, topology optimization is shown to be a powerful tool that allowed the designer to find that the hole shape was basically
elliptic, thus suggesting their regularization as ellipses. Comparisons were made between several design solutions, each characterized by
a specific volume reduction of the bottom flange. Identification of the highest-performing solutions through computer-aided procedures led to
a weight reduction of 40% with respect to the design solution in reinforced concrete. Retrofitting the already existing foundations and abutments
to satisfy the updated provisions of the new seismic code was thus avoided by defining an innovative layout of arch bridges with holes in the
bottom flange, which has never been used before. DOI: 10.1061/(ASCE)BE.1943-5592.0000416. © 2013 American Society of Civil
Engineers.
CE Database subject headings: Bridges; Design; Conceptual design; Optimization.
Author keywords: Bridge design; Conceptual design; Topological optimization; SIMP method.

Introduction the Piave River near Venice, Italy (Zordan et al. 2006, 2010). The
bridge was originally designed to be made of prestressed concrete
Although the conceptual design of a bridge depends on the [Fig. 1(a)], and both the foundation and abutments were already
designer’s intuition and ability to recognize the role of members in partially built when, in 1999, the contractor’s financial failure caused
transferring weight and loads to the earth, currently, structural the interruption of construction work [Figs. 2 and 3(a)].
optimization is not only an important tool for sizing structure Before work resumed, the Italian seismic code (Presidenza del
members but also for helping the designer find the most suitable Consiglio dei Ministri 2003) was updated together with a new
shape of a structure from a structural and an architectural point of seismic classification of the Italian territory. It prescribed higher
view (Majowiecki 2007, 2008; Samartin 1995). Structural optimi- acceleration values [peak ground acceleration ðPGAÞ 5 0:25g re-
zation is nowadays common in mechanical and aeronautical ferred to a return period of 475 years and site class B, corresponding
engineering, and in recent years, it has been progressively adopted to a very dense soil with S-wave velocity 360e800 m=s], requiring
for structural-engineering applications, such as multistory buildings a much higher increase in the resistance (35%) of the already existing
and long-span bridges (Allahdadian and Boroomand 2010; Huang foundations. Given the owner’s absolute requirement of not mod-
and Xie 2008; Neves et al. 1995; Stromberg et al. 2010). ifying the bridge profile, lightening the bridge superstructure became
The topic of this paper is a topological optimization problem essential to save the already existing foundations and abutments
faced during the design of a 500-m multispan arch bridge built over without reinforcing foundations with high work costs. With this aim,
the design solution of using a composite deck with concrete slabs and
1
Professor, College of Civil Engineering, Fuzhou Univ., Fuzhou steel arches was finally chosen [Fig. 1(b)]. Moreover, it was possible
350108, China (corresponding author). E-mail: bruno@fzu.edu.cn to further lighten the steel arch not only by optimizing the thickness
2
Professor, Dept. of Structural Engineering, Univ. of Cagliari, 09124 of the webs and the top and bottom flanges but also by removing
Cagliari, Italy. inefficient material through the creation of large holes in the bottom
3
Postdoctoral Researcher, Progettazione e Pianificazione in Ambienti flange. In this way, the bridge profile remained unchanged, but the
Complessi, Univ. Università Iuav di Venezia, 30135 Venice, Italy. weight reduction allowed the existing foundations to resist the re-
4
Postdoctoral Researcher, Progettazione e Pianificazione in Ambienti quired magnitude of seismic actions according to the new Italian
Complessi, Univ. Università Iuav di Venezia, 30135 Venice, Italy. seismic code.
5
Professor, College of Civil Engineering, Tongji Univ., Shanghai Therefore, the problem of choosing the most suitable shape of the
200092, China.
holes in the bottom flange had to be considered, as often happens in
Note. This manuscript was submitted on September 20, 2011; approved
on May 14, 2012; published online on May 16, 2012. Discussion period design problems where the optimal design solution has internal
open until January 1, 2014; separate discussions must be submitted for boundaries created by holes. Finding the optimum shape of cavities
individual papers. This paper is part of the Journal of Bridge Engineering, in a two- or three-dimensional continuum body is a topological
Vol. 18, No. 8, August 1, 2013. ©ASCE, ISSN 1084-0702/2013/8-790–800/ optimization problem (Eschenauer and Olhoff 2001) that presents
$25.00. difficulties when using a standard optimization approach.

790 / JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013

J. Bridge Eng. 2013.18:790-800.


Fig. 1. (a) Typical cross section of original project; (b) typical cross section of retrofitted composite deck solution (units: m)
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. Original solution as planned and as built (shaded) before work interruption

Fig. 3. (a) Some parts of the bridge built before the introduction of the new seismic regulation; (b) retrofitting joint details on the piers

In recent years, intuitive approaches to topology optimization The shape of the hole edges was then regularized to give the two holes
based on the concept of removing inefficient material have been an actual elliptic form in the first case and to obtain large holes by
introduced (Bendsøe and Kikuchi 1988; Burns 2002; Christensen merging the two ellipses in the second case.
and Klarbring 2008; Rozvany 2009). The two most popular ones Both the single- and double-hole design solutions are optimum
are the solid isotropic material with penalization (SIMP) method solutions achieved from topology optimization that differ from one
(Bendsøe 1989; Sigmund 2001) and the evolutionary structural another regarding their assigned VR values.
optimization (ESO) method (Edwards et al. 2007; Xie and Steven Therefore, it was necessary to choose the best design solution,
1992). identified in terms of minimum weight and maximum overall
In this paper, topology optimization was carried out through the structural performance, as well as in terms of a combination of these
SIMP method, which is based on the assumption that the stiffness competing objectives.
matrix of each element is proportional to its density. The optimization procedure allowed the designer to identify
Different values of volume reduction (VR) led to different optimum the optimum design solution through the introduction of a spe-
solutions, each one with a different impact on the appearance of the cific optimization index, also including a penalty exponent, able
bridge and its aesthetic value. In particular, topology optimization to give a score to the general performance of different layout
identified two main families of holes: the first one, for lower values of solutions.
VR, had two holes with approximately elliptic form (the bigger the In the approach to structural design used to retrofit this bridge,
holes, the higher the VR), one on each side of the middle of the arch; conceptual design and structural optimization become mutually and
and the second one had the two holes merging into a large hole, until, profitably related to one another, with structural optimization ap-
for very high values of VR, the bottom flange was separated longi- plied as an effective design tool. This is an interesting feature of this
tudinally in two parts. Naturally, in practical structural design, holes study: the solution of a serious technical problem that could worsen
with irregular form, such as the one obtained in this paper through the aesthetic value of the bridge with a heavier reinforcement of its
topology optimization, cannot be accepted for construction reasons. abutments and foundations instead resulted in an improvement of

JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013 / 791

J. Bridge Eng. 2013.18:790-800.


the original design from both engineering and architectural points of In addition to permanent loads (self-weight plus dead loads at-
view. This was allowed by the use of topological optimization, tributable to guard rails, asphalt, etc.), live loads attributable to pe-
a design tool that nowadays should be used more frequently by destrian (4 kN=m2 ) and truck (30 kN=m uniformly distributed load
structural designers also in practical engineering. plus 3 3 200 kN concentrated loads) loading were taken according
to the Italian code (Ministero dei Lavori Pubblici 1990) (see Fig. 5).
To maximize both the bending and shear forces, five different
Finite-Element Model positions for the tandem load were considered (0, 1/4, 1/2, 3/4, and 1
times the span length). All loads were amplified by safety factors (1.4
Because fixed restraints between the steel-concrete composite su- for dead loads and 1.5 for live loads) to satisfy the ultimate limit
perstructure and concrete piers were used [obtained by means of state (ULS) conditions. Because each lane was 3 m wide, three
anchoring systems and bolted joints; Fig. 3(b)], it was sufficient to different lane loads were considered as acting simultaneously, the
analyze the bridge deck only. A finite-element (FE) model was first one with 100% of the previously defined loads, and the other
implemented using the general-purpose software Ansys 11.0 and ones with 50 and 35% of the loads as imposed by the regulation code
integrated Ansys Parametric Design Language (APDL), which (Fig. 5). To account for the eccentricity effect, four different lane
allowed a parametric definition of the model geometry. Because
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

combinations were considered for each truck load position (Table 1),
rigid connection between the steel deck and concrete abutments is for a total of 16 different load combinations considered during the
assumed, fixed restraints at both ends of the steel superstructure were whole optimization process.
assumed in the model, simulating the hogging bending moment of
the retrofitting joint on the piers and abutments. The FE model of the Adopted Optimization Procedure
steel superstructure before hole insertion in the bottom flange
through topology optimization (starting FE model) is shown in Because the aim of retrofitting the bridge was deck lightening,
Fig. 4. structural optimization was performed in two stages. In the first
Both the concrete deck and the steel plates were modeled with stage, after having defined the bridge geometry according to the
quadrilateral eight-node shell elements. An average mesh size of owner’s requirements of maintaining the profile of the original
0.40 m was chosen so as to ensure a sufficiently refined mesh, bridge in prestressed concrete, the thicknesses of the webs and
suitable for element removal during the topology optimization flanges were minimized. Structural optimization was performed to
procedure (TOP). The total number of nodes is 37,406, with 12,240 size the thicknesses of the webs and flanges, assumed as discrete
elements, among which 4,320 belong to the deck slab, 4,320 variables, to minimize the self-weight of the deck.
belong to the webs, and 960 and 2,640 belong to the top and Because the stress level in the bottom flange was quite low and
bottom flanges, respectively. deck lightening was insufficient to allow seismic verification of the
Steel S355, which according to Eurocode 3 (European Committee foundations, unnecessary material was removed from the bottom
for Standardization (CEN) 2005) has a yield stress fyk 5 355 MPa, flange through topology optimization in the second stage. A design
was used (corresponding to a design yield stress fyd 5 338 MPa after solution with holes was thus obtained. Different configurations of
application of a safety factor gM0 5 1:05). The steel elastic modulus holes were obtained, varying based on the percentage of VR as-
was assumed to be 200 GPa, with an elastic shear modulus of 76.9 GPa signed by the designer.
and a Poisson ratio of 0.3.
The concrete flange was modeled by assuming two different
Identification of Optimal Flanges’ Thickness
values of the concrete elastic modulus: 30 GPa in the bridge region
for a length of a half-span about the midspan and 15 GPa in the The thicknesses of the webs and flanges were optimized by using
remaining regions close to the abutments, thus reduced by a factor of the design optimization tool implemented in Ansys 11.0. It provides
2 to take into account concrete cracking in the hogging moment a zeroth-order method where the constrained minimization problem
regions close to the span ends (International Federation for Struc- is first converted to an unconstrained one by means of penalty
tural Concrete 1993). Linear elastic behavior for the whole structure functions and is then solved using Powell’s modified method
was assumed. (Powell 1964). Plate thicknesses were assumed as discrete design

Fig. 4. Numerical model of the bridge before optimization

792 / JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013

J. Bridge Eng. 2013.18:790-800.


Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 5. Live load distribution

Table 1. Considered Lane Load Factor a low stress level. For this purpose, topology optimization was used,
Lane 1 Lane 2 Lane 3 Crowd 1 Crowd 2 because it is particularly appropriate when the design problem is the
Combination (%) (%) (%) (%) (%) optimized insertion of cavities in a continuum body (the bottom flange
in this specific case). It was performed through the SIMP method,
1 100 50 35 100 0 which is a very efficient structural optimization approach that
2 50 100 35 100 100 has demonstrated its effectiveness in a large number of examples
3 35 100 50 100 100 (Bendsøe and Sigmund 2003). It is also the method implemented
4 35 50 100 0 100 in many commercial tools (OptiStruct, Genesis, MSC/Nastran, Ansys
11.0, etc.) performing topology optimization (Rozvany 2009).
Topology optimization through the SIMP method was perfor-
variables whose optimum arrangement was found by minimizing the med using Ansys 11.0, in which the design variables are internal
plate’s total weight with the conditions that the stress level was lower pseudodensities that are assigned to each ith finite element in the
than an allowable value and the plate thickness higher than a min- topological problem. Based on the hypothesis about the relationship
imum value (to avoid local stability problems). Four different between the variation of material properties and density, the stiffness
regions of the steel deck were located where the thicknesses of the matrix of each element is assumed to be proportional to hE, where
web, bottom flanges, and top flanges had to be identified (Table 2). E is the actual elastic modulus; h 5 rq is the internal pseudodensity
An optimization problem with 12 discrete variables (corresponding of the element; r is the relative density compared with the actual
to the plate thicknesses to be optimized) was hence defined. density of the material and continuously varying between 0 and 1;
Because the optimum solution was found to depend on the initial Eef 5 hE is the effective elastic modulus, lower than E in structure
values of the plate thickness, different initial values were tried to regions with relative density r lower than 1; and q is a penalty
avoid local minimum solutions. exponent whose effect, for values sufficiently higher than 1 (nor-
Table 2 shows the values of plate thickness at different design steps: mally q . 3), is that of avoiding in the optimal solution elements
the initial thickness values (see Table 2, Initial), the ones obtained from with intermediate values of the effective elastic modulus. From
the optimization procedure (see Table 2, Best), and finally, the actual a structural point of view, all of the previous assumptions mean that
thicknesses chosen by the designer taking into account both opti- the elements with a r near 0 make very little contribution to the
mization results and plate thicknesses available on the market (see global stiffness matrix (and therefore to the model compliance), so
Table 2, Chosen). Moreover, Table 2 shows that the stress levels of the that the effect of their removal is negligible.
initial and actual thickness arrangements are practically the same, When topology optimization is performed using the SIMP
while optimization reduces the steel volume by 29%. method as a minimum compliance design, a material distribution
problem arises. If the problem is discretized using finite elements,
numbered as i 5 1, . . . , N, and u and f are the displacement and load
Topological Optimization Process
vectors, respectively, compliance C 5 f T u is minimized (Bendsøe
Although the previous optimization procedure was able to con- and Sigmund 2003).
siderably lighten the bridge deck, seismic verification of the Additionally, u and f are related to one another through the global
foundations required a further reduction of the deck self-weight, stiffness matrix K, as KðEef i Þu 5 f, where Eef i is the effective elastic
achievable by means of a VR of the steel bottom flange of at least modulus of each element i defined as Eef i 5 hi E. Hence, if V is the
20%. The low stress level in large regions of the bottom flange total volume of the structure after topology optimization, assigned as
(Fig. 6) suggested a further significant decrease of superstructure a percentage of the actual volume V0 of the structure before the TOP,
self-weight, attainable by removing inefficient material working at minimization of compliance C leads to the following:

JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013 / 793

J. Bridge Eng. 2013.18:790-800.


Table 2. Steel Flange Thickness Identification from Design Variable Optimization Process
Design T1 T2 T3 T4 W1 W2 W3 W4 B1 B2 B3 B4 smax Volume
set (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (mm) (MPa) (m3)
Initial 20 20 20 20 15 15 15 15 20 20 20 20 256 27.2
Best 14.6 10.2 10.3 14.5 11.7 10.3 10.1 11.6 21.7 16.5 10.9 16.5 249 19.4
Chosen 15 10 10 15 12 10 10 12 20 16 10 16 251 19.2
Note: B 5 bottom flange; W 5 web plates; T 5 top flange; numbers (1, 2, 3, and 4) refer to identified thickness groups (cf. Fig. 5).
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 6. View of the bottom flange with two regions with unexploited material to be removed through insertion of cavities

Fig. 7. Pseudodensities (h) from topological optimization with different imposed VR

min f T u (1)
Eef i , ui

and leads to the pseudodensity value hi in each element for the


following:

P
N
0 # hi # 1 and V¼ hi Vi # ð1 2 VRÞV0 (2)
i¼1

After having evaluated the volume V0 of the bottom flange before


inserting the holes, topology optimization was performed by minimizing
compliance C (that is, maximizing stiffness) for different given per- Fig. 8. Definition of TOP models from topological results
centages p of reduction of the volume of the original bottom flange (from
15 to 100%), thus obtaining the range of solutions reported in Fig. 7.
The sensitivity of the optimized solution to mesh size was removing material with a pseudodensity lower than a given value
evaluated. Even if the pseudodensity contour was nearly unchanged, (Fig. 8).
computational time increased considerably as mesh size was re- Fig. 7 shows that for values of VR lower than 30% or higher than
duced. A suitable compromise between accuracy and analysis time 40%, topology optimization clearly suggested design solutions with,
was found with an adopted average mesh size of 0.40 m. Two main respectively, two holes or one hole. The TOP models were obtained
families of holes were identified: the first family, for low values of by choosing a pseudodensity limit for material removal able to
VR, used two approximately elliptically shaped holes, and the guarantee the same amount of reduction of the corresponding to-
second family used the two holes merged to form a single hole. pological optimization solution (within a given tolerance of 5%).
Based on topology optimization solutions, different layouts On the contrary, for values between 30 and 40%, and notwith-
(hereafter identified as TOP models) were then obtained by standing the penalty exponent introduced by the SIMP method,

794 / JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013

J. Bridge Eng. 2013.18:790-800.


Fig. 7 shows that the hole shape can be greatly affected by the value Through this procedure, a new set of design solutions derived
of the pseudodensity below which material must be removed. Hence, from topology optimization was identified (hereafter identified as
from each of the two design solutions obtained by means of topology DOP models) that was more easily realizable and had better aesthetic
optimization for VR of 30 and 40%, two different layouts of the value (Table 4).
holed flange were obtained (Table 3): one with two distinct holes
and the other one with only one hole, depending on the limit value of Numerical Results of Topological and Design
the pseudodensity below which material must be removed. Optimization Process Models
At the end of this procedure, a set of design solutions (TOP
models) obtained from topology optimization was finally available Two sets of design solutions generated through the insertion of
for the designer. cavities in the bottom flange were obtained (Fig. 10), each one for
a given value of VR of the bottom flange. The same procedure shown
in Figs. 8 and 9 was then applied, thus first obtaining the set on the
Design Optimization Process left column of Fig. 10 directly through topology optimization and
The TOP adopted in the previous section, Topological Optimization then obtaining the set on the right column of Fig. 10 through reg-
ularizing the cavities and giving them an elliptical form.
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

Process, offers an overall overview of the unexploited material, but


the shapes of the cavities directly obtained from topology optimi- The structural response caused by the insertion of holes with
zation are not suitable from both a construction and an architectural different shapes can be evaluated from the data in Figs. 11 and 12,
point of view. Nevertheless, the TOP results suggested that the de- where the stress and deformation state are plotted as functions of
signer regularize the holes by means of ellipses. the VR of the bottom flange.
To obtain the elliptical holes, two different procedures were After having calculated the average stress sei in each element
defined, depending on whether one or two distinct holes (for lower and the weighted average of all the sei with respect to the area Aei of
values of VR) characterize the TOP models. If topology optimiza- the respective element, the global average stress s is as follows:
tion led to two distinct holes, the minor and the major axes of the two P 
ellipses were, respectively, identified as the maximum transverse i s e i × Ae i
s¼ (3)
and longitudinal distance between the hole edges; the ellipse center Atot
was then located at the middle of the major axis [Fig. 9(a)].
When a unique hole was obtained instead, the minor axis of the In Fig. 11, both the maximum stress level along the bottom flange
two ellipses merging into each other was identified, as in the previous and the percent variation of the average stress s are plotted.
case. The center of the ellipse was then located in the midspan of such It can be observed that the higher the VR, the higher the stress
an axis, and the major axis was defined to constrain the ellipse to cross level, in terms of both maximum and average stress. When two
the hole edge close to the bridge ends [points A and B in Fig. 9(b)]. holes merge into each other, an additional compliance and stress

Table 3. Volume Reduction in Topological Optimization and Corresponding Values in Topological Optimization Process Models
Model T0 T1 T2 T3a T4a T3b T4b T5 T6 T7 T8 T9 T10
VR (TOP) (%) 0 15 20 30 40 30 40 50 60 70 80 90 100
Chosen hlim — 0.8 0.4 0.2 0.2 0.8 0.8 0.4 0.5 0.2 0.2 0.2 —
VR final (%) 0 15.4 21.8 27.8 32.1 35.6 45.6 51.6 62.1 70.5 81.7 89.9 100
Layout Double hole Single hole

Fig. 9. Definition of DOP models from topological results: (a) double-hole layout; (b) single-hole layout

Table 4. Geometrical Property of Design Optimization Process Models


Model D0 D1 D2 D3a D4a D3b D4b D5 D6 D7 D8 D9 D10
VR (%) 0 16.3 24.5 34.5 48.5 42.5 58.5 59.3 64.2 76.2 84.7 87.8 100
X0 (m) — 17.5 18 17 15 19 19 17.5 17.5 16.5 18 30 30
AX (m) 0 6.5 9 10 12 12.5 17 16.5 17.5 18.5 23 41 ‘
BY (m) 0 2 2 2.5 2.5 2.5 2.5 2.5 2.5 2.8 3.1 3.1 3.1
Layout Double hole Single hole
Note: VR 5 volume reduction; X0 5 ellipse center position; AX 5 major ellipse’s axis; BY 5 minor ellipse’s axis.

JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013 / 795

J. Bridge Eng. 2013.18:790-800.


Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. Von Mises stress envelope for each optimized model

Fig. 11. Von Mises stress versus VR: (a) maximum stress value; (b) global average stress variation (%)

relaxation occur, although the VRs for single- and double-hole Optimization Index Formulation
layouts are almost the same.
Analogously, Fig. 12 shows that, for close values of VR, deflections Because a set of optimum layouts with holes in the bottom flange was
of the design solutions with two distinct holes appear to be lower than obtained from the topology optimization for different values of VR,
deflections of the design solutions with only one hole, meaning that a specific optimization index (OI) was introduced to give the designer
compliance is increased when two holes merge into one. a specific tool for identifying the most suitable design solution.

796 / JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013

J. Bridge Eng. 2013.18:790-800.


Such an index had to take into account VR (and therefore weight In the following, both VR and RI are expressed as a percent, thus
reduction) together with the structural response in terms of both the varying between 0 and 100%; although in this study, a significant
stress and deformation level. value of VR is at least 20%.
Two indexes summarizing the overall behavior of the whole Consider the Cartesian plane where VR is the x-axis and RI the
structure were identified: Von Mises stress, averaged throughout y-axis. The difference VR 2 RI represents the distance between the
the whole steel superstructure, was considered as representative of plane bisectrix and the RI curve. By scaling the difference VR 2 RI
the stress level, whereas deflection at midspan was considered as through the coefficient 1=VR for each value of VR, the distances
representative of the deformation level. between the RI curve and the plane bisectrix results decreased to
Two response indexes (RI) were then identified in terms of ðVR 2 RIÞ=VR, so that a curve whose distance from the abscissa
the percentage variation of the deformation and stress level with axis is VR 2 ðVR 2 RIÞ=VR can be obtained [Fig. 13(a)]. Con-
respect to their corresponding values obtained in the starting model versely, a curve whose distance from the bisectrix is VR 2
(model 0 in Fig. 10) under the same loading condition ðVR 2 RIÞ=VR and whose distance from the abscissa axis is
ðVR 2 RIÞ=VR can also be obtained [Fig. 13(b)]. This latter curve is
si 2 s 0 di 2 d0
RIðs, iÞ ¼ RIðd, iÞ ¼ (4) therefore the OI curve defined as in Eq. (4) and therefore is expressed
s0 d0
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

as a percent like VR and RI.


Because the retrofitting objective was lightening the bridge
where d 5 midspan displacement; s 5 Von Mises stress averaged deck, it was then worth modifying the scaling coefficient 1=VR
throughout the whole steel superstructure; i 5 ith model under through a penalty exponent able to favor design solutions with
consideration (cf. Fig. 10); and 0 5 starting model. higher VR. For this purpose, after the introduction of the penalty
Although the design objective was the reduction of superstruc- exponent, the scaling coefficient became ð1=VRÞb , where values
ture weight, the increase in VR causes an increase of both the stress of the penalty exponent between 0 and 1 favor design solutions
level and deflections, whose control was therefore a competing with higher VR, while values higher than 1 favor design solutions
requirement with respect to VR. with higher performances but higher self-weight and are there-
Hence, a way of identifying the most suitable design solution among fore not convenient for lightening the bridge superstructure
all of these different optimized layouts with holes had to be defined. [Fig. 13(c)].
For this purpose, the variation of structural response with respect Hence, the updated expression of the OI through the introduc-
to the variation of VR can be compared by defining OI 5 RI=VR or, tion of the penalty exponent is the following:
conversely, its complement to 1
RIðiÞ 1 1
OIðiÞ ¼ 1 2 ¼ ½VRðiÞ 2 RIðiÞ × (5) OIðiÞ ¼ ½VRðiÞ 2 RIðiÞ × (6)
VRðiÞ VRðiÞ ½VRðiÞb

Fig. 12. Deck’s displacement versus VR: (a) maximum deflection value; (b) deflection variation (%)

Fig. 13. Graphical interpretation of the optimization index

JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013 / 797

J. Bridge Eng. 2013.18:790-800.


The application of the penalty exponent b . 0 to the scaling In Fig. 14(b), the ratio RðbÞ 5 ð1=VRÞb =ð1=VRÞ is plotted for
factor 1=VR therefore results in lower values of OI for b , 1 and in different values of b. It can be observed that, for b 5 2, the ratio RðbÞ
higher values for b . 1. is roughly 0 for almost every value of VR. On the contrary, for
From Fig. 14(a), it can be observed that for b . 1 the scaling b 5 0:5, the ratio RðbÞ is a steep function of VR, meaning that the
factor ð1=VRÞb as a function of VR tends to become a bilinear curve scaling factor with penalty exponent 0.5 becomes much higher than
(the more bilinear the curve, the higher the b), with a vertical branch the one without the penalty exponent, even for low values of VR,
for VR → 0 and a horizontal branch whose values are almost 0 in the becoming 10 times 1=VR for VR 5 100%; an intermediate trend is
range of VR values that are significant in the design of the bridge observed for b 5 0:8, with R initially slightly steep and then even
under consideration. For b , 1, on the contrary, the scaling factor less steep until the value of only 2.5 times 1=VR is reached for
ð1=VRÞb plotted as a function of VR approaches a hyperbola for VR 5 100%.
b → 1, tending to become a horizontal line for b → 0. Hence, although values of b less than 1 generally favor design
Hence, whereas for b . 1, design layouts with holes are not fa- solutions with significant VR, values of b less than but suitably close to
vored and those with low values of VR are conversely favored already 1 are able to favor design solutions with intermediate values of VR.
for b-values slightly higher than 1, for b , 1, design solutions with Finally, because the expression for OI refers to a specific structural
high VR are instead favored: the higher the VR, the lower the b.
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

response in terms of stress or deformation, a global optimization index

Fig. 14. (a) Scaling factor 1=VR with penalty exponent b versus VR; (b) coefficient RðbÞ versus VR

Fig. 15. Normalized GOI versus VR for different values of b

798 / JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013

J. Bridge Eng. 2013.18:790-800.


(GOI) considering both of these features of structural response can
also be defined. By giving the same weight to both the deformation
and stress level, a GOI averaging the two OIs that refer to stresses and
deformations is then defined as follows:

GOIðiÞ ¼ ½OIðs, iÞ þ OIðd, iÞ=2 (7)

Identification of the Best Design Solution

The GOI allows the identification of the optimum solution that


best balances material saving and overall performance of the
Fig. 16. Best TOP and DOP models versus b and VR structure.
Fig. 15 shows the trend of GOI as VR varies for some values of
b ranging between 0 and 1.2. Fig. 16 reports the maximum values of
the curves of the TOP and DOP models of Fig. 15, that is, the relative
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

optimum design solutions obtained by both the TOP and DOP.


As expected, even for values of b slightly higher than 1, design
solutions without any VR, coinciding with the starting model with-
out holes, but meaningless from the optimization point of view, are
favored. On the contrary, design solutions with intermediate values
of VR identifying two disconnected holes are favored by values of
b varying between 1 and 0.75. For still lower values of b, material
saving is instead the prevailing aspect, and design solutions with the
two holes merging into each other are found. For b → 0, this latter
Fig. 17. Best identified models versus b type of design solution tends to the layout with two parallel ribs
unconnected by the bottom flange.
Fig. 17 reports the best design solutions obtained through both
Table 5. Optimized Model Performance with Respect to Model 0 (without the TOP and DOP and identified by means of GOI for different
Hole) values of the penalty exponent b.
Values as percentage Model D4a Model 10 Taking into account these results, the design solution D4a with
a VR of nearly 50% was finally chosen. On one hand, it is a good
VR (bottom flange) 48.4 100
compromise solution between the two competing requirements of
VR (steel superstructure) 15 32
reducing self-weight without worsening structural performance
Dsbot. flange 124 162
(see Table 5). On the other hand, it is an innovative structural so-
Dstot 111 130
lution that shows a clear conceptual design. In fact, through insertion
D displacement 10:1 132
of elliptical holes, the identified layout is effective in reducing the

Fig. 18. Completed optimized superstructure makes clear the correspondence between shape and structural response: (a) FE model maximum stresses;
(b and c) some pictures of as-built structure

JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013 / 799

J. Bridge Eng. 2013.18:790-800.


seismic action on foundations and abutments and is also very in- Bendsøe, M. P. (1989). “Optimal shape design as a material distribution
novative from an architectural point of view, increasing the bridge’s problem.” Struct. Multidiscip. Optim., 1(4), 193–202.
aesthetic value (Fig. 18). Bendsøe, M. P., and Kikuchi, N. (1988). “Generating optimal topologies in
Such a design result can only be obtained as the fruit of design structural design using a homogenization method.” Comput. Meth. Appl.
Mech. Eng., 71(2), 197–224.
experience as well as of structural and architectural sensibility
Bendsøe, M. P., and Sigmund, O. (2003). Topology optimization: Theory,
that currently, in structural design, cannot get along without methods, and applications, Springer, Berlin.
numerical tools. Burns, S. A. (2002). Recent advances in optimal structural design, ASCE,
Reston, VA.
Christensen, P. W., and Klarbring, A. (2008). An introduction to structural
Conclusions
optimization, Springer, Dordrecht, Netherlands.
Edwards, C., Kim, H., and Budd, C. (2007). “An evaluative study on ESO
A design procedure to optimize the superstructure of the S. Donà and SIMP for optimising a cantilever tie–beam.” Struct. Multidiscip.
Bridge for its retrofitting is described. Optim., 34(5), 403–414.
Retrofitting the San Donà Bridge, which was required after the Eschenauer, H. A., and Olhoff, N. (2001). “Topology optimization of
foundations and abutments were constructed because of the updated continuum structures: A review.” Appl. Mech. Rev., 54(4), 331–381.
Downloaded from ascelibrary.org by NDSU LIBRARY on 05/28/14. Copyright ASCE. For personal use only; all rights reserved.

Italian seismic code with higher accelerations, was an opportunity European Committee for Standardization (CEN). (2005). “Design of steel
for additionally improving the aesthetic value of the bridge while structures, part 1-1: General rules and rules for buildings.” Eurocode 3,
reinforcing the already-built foundations. Brussels, Belgium.
A conceptual approach to the design problem of having higher Huang, X., and Xie, Y. (2008). “Topology optimization of nonlinear struc-
tures under displacement loading.” Eng. Struct., 30(7), 2057–2068.
seismic forces on the already-built foundations led to lightening the
International Federation for Structural Concrete (FIB). (1993). CEB-FIP
bridge, thus giving up the notion of completing the bridge with Model Code 90—Design of concrete structures. Thomas Telford, London.
prestressed concrete. A lighter steel deck with a new form obtained Majowiecki, M. (2007). “Ethics and structural reliability in free-form design
through topology optimization that lightened the deck by removing (FFD).” J. Int. Assoc. Shell Spat. Struct., 48(4), 29–50.
inefficient material working at a low stress level was then designed. Majowiecki, M. (2008). “The free form design (FFD) in steel structural
The importance of an appropriate conceptual design is thus ev- architecture—Aesthetic values and reliability.” Steel Constr., 1(1), 3–15.
ident, as even a troublesome problem of bridge retrofitting to meet Ministero dei Lavori Pubblici. (1990). “Aggiornamento delle Norme
the code requirements resulted in a more aesthetic bridge. Tecniche per la progettazione, la esecuzione e il collauso dei ponti
Furthermore, lightening the bridge through the insertion of two stradali.” Rep. D.M. 4.5.90, Ministero dei Lavori, Pubblici, Rome
large cavities in the steel superstructure jointed to the abutments was (in Italian).
Neves, M., Rodrigues, H., and Guedes, J. (1995). “Generalized topology
actually less expensive than reinforcing the existing foundations and
design of structures with a buckling load criterion.” Struct. Multidiscip.
completing the bridge with prestressed concrete. Optim., 10(2), 71–78.
Topology optimization was used as a powerful tool for this Powell, M. J. D. (1964). “An efficient method for finding the minimum of
conceptual approach to the retrofitting of the bridge. Topology op- a function of several variables without calculating derivatives.” Comput.
timization was implemented with the SIMP method, a very robust, J., 7(2), 155–162.
efficient, and suitable design tool. A set of design solutions with Presidenza del Consiglio dei Ministri. (2003). “Primi elementi in materia di
elliptical holes drawn by regularizing the almost elliptical forms of criteri generali per la classificazione sismica del territorio nazionale e di
the cavities achieved by means of topology optimization was then normative tecniche per le costruzioni in zona sismica.” OPCM 3274
obtained. Among these design solutions, a good compromise be- Presidenza del Consiglio dei Ministri, Rome (in Italian).
tween the competing requirements of VR and structural perfor- Rozvany, G. I. N. (2009). “A critical review of established methods of
structural topology optimization.” Struct. Multidiscip. Optim., 37(3),
mance was identified by defining an appropriate OI that assigned
217–237.
a suitable score to each design solution. The optimization process Samartin, A. (1995). “Application of optimization techniques to struc-
led to the best score being assigned to a design solution with two tural design.” Proc., Int. Symp. Lightweight Structures in Civil Engi-
elliptical holes. neering, Magat-Magdalena Burska, Warsaw, Poland 795–804.
A weight reduction of 40% with respect to the design solution Sigmund, O. (2001). “A 99 line topology optimization code written in
in reinforced concrete was hence obtained, thus avoiding reinforcement Matlab.” Struct. Multidiscip. Optim., 21(2), 120–127.
of the already existing foundations. The updated rules of the new Stromberg, L. L., Beghini, A., Baker, W. F., and Paulino, G. H. (2010).
seismic code were satisfied, however, with the additional benefits of “Application of layout and topology optimization using pattern grada-
improving the architectural form and aesthetics of the bridge. tion for the conceptual design of buildings.” Struct. Multidiscip. Optim.,
43(2), 165–180.
Xie, Y., and Steven, G. P. (1992). “Shape and layout optimization via an
References evolutionary procedure.” Proc., Int. Conf. on Computational Engi-
neering Science, Hong Kong Univ., Hong Kong.
Allahdadian, S., and Boroomand, B. (2010). “Design and retrofitting Zordan, T., Briseghella, B., and Mazzarolo, E. (2010). “Structural opti-
of structures under transient dynamic loads by a topology optimization mization through step-by-step evolutionary process.” Struct. Eng. Int.,
scheme.” Proc., 3rd Int. Conf. on Seismic Retrofitting, Iranian North- 20(1), 72–78.
West Retrofitting Center (Iranian Retrofitting Researchers Ins.), Tabriz, Zordan, T., Briseghella, B., and Siviero, E. (2006). “Recent trends in the
Iran, 1–9. structural refurbishment of existing bridges.” Studi Ric.—Politec.
Ansys 11.0 [Computer software]. Canonsburg, PA, Ansys. Milano, 26, 41–73.

800 / JOURNAL OF BRIDGE ENGINEERING © ASCE / AUGUST 2013

J. Bridge Eng. 2013.18:790-800.

You might also like