Conceptual Design of Modular Bridges Including Layout

Optimization and Component Reusability

Alexis Tugilimana1; Ashley P. Thrall, A.M.ASCE2; and Rajan Filomeno Coelho3

Abstract: Modular or panelized bridges, comprised of stacked rectangular panels forming girder-type bridges, are the most widely used
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

approach for rapidly erectable bridging. Limitations of these systems, however, include inefficient material distribution of both panels and lat-
eral bracing. This research addresses this challenge by implementing structural optimization for the conceptual design of modular bridges,
including module topology and spatial orientation optimization. This contribution generalizes an existing formulation for optimizing modular
trusses to include (1) reusability of modules among multiple structures and (2) practical considerations in design, such as multiple types of
modules, multiple load cases, the capability to compute the displacements in an elastic design formulation, and limitation on stresses. This
methodology is demonstrated for the conceptual design of single- and double-story simply supported bridges. Further studies find that (1)
incorporating module reusability results in a trade-off between constructability and material efficiency and (2) better designs can be obtained
by modifying the module configuration. This research culminates in guidelines to assist designers during preliminary design phases. DOI:
10.1061/(ASCE)BE.1943-5592.0001138. © 2017 American Society of Civil Engineers.
Author keywords: Modular design; Structural optimization; Topology optimization.

Introduction structural optimization method to modular structures, in which the

module topology is optimized for minimum compliance. Chen et al.
The state of the practice in rapidly erectable bridging is to assemble (2010) adapted this strategy to the multiobjective optimization of
rigid rectangular modules or panels [Fig. 1(a)] in a longitudinal con- continuous structures for maximum stiffness and thermal conduc-
figuration to form a girder-type bridge. Panels can be stacked verti- tivity. Ryvkin and Fuchs (1999) and Moses et al. (2003) applied a
cally or transversely to achieve longer spans and/or carry higher Fourier transform to reduce the periodic problem into the optimiza-
loads [Fig. 1(b)]. This is the predominant approach for rapidly erect- tion of one representative unit cell. Although these strategies
able gap-crossing solutions. The universality of this approach has improve design, they only consider modularity as a simple topology
been demonstrated through the Bailey, Mabey-Johnson, and Acrow repetition for a module arrangement a priori defined. These
systems, which have been widely used for military operations, tem- approaches are known to decrease the performance of the optimized
porary civil infrastructure, and disaster relief since World War II design (e.g., higher optimized compliance) when numerous mod-
(Joiner 2001; Russell and Thrall 2013; Acrow Bridge 2016). ules are considered (Huang and Xie 2008; Moses et al. 2003).
Limitations of the efficiency of these systems include that (1) the In a recent contribution, Tugilimana et al. (2017) demonstrated
cross section is not varied with changing flexural and shear that this limitation can be reduced when the spatial orientation of
demands along the span and (2) when stacked the material is con- each module is considered as an additional design variable. This
centrated at the neutral axis in which it contributes little to the load- strategy applies a continuum-based formulation for the lower bound
carrying capacity (Gerbo et al. 2016; Wang et al. 2016). This paper plastic design optimization of modular trusses. However, this
addresses these limitations by implementing structural optimization method is limited to the optimization of only one type of module
to achieve more efficient panelized bridging systems. under one load case (neglecting self-weight). Additionally, the use
Existing research in structural optimization of modular struc- of plastic design only focuses on the equilibrium between internal
tures has focused on module topology periodicity, in which the and external forces, disregarding the evaluation of the displace-
design domain is divided into unit cells that share the same topol- ments that allow elastic equilibrium.
ogy. Huang and Xie (2008) extended the bidirectional evolutionary
Despite the design improvements using the strategies dis-
cussed previously, these methods do not yet exploit one of the
1 main advantages of modular design: the capability of reusing
Ph.D. Student, BATir Dept., Univ. Libre de Bruxelles, Avenue F.
Roosevelt, 50, CP 194/2, 1050 Brussels, Belgium (corresponding author). components in different projects. This results in a reduction of the
E-mail: [email protected] effort and investment dedicated to the design of all of the struc-
2
Myron and Rosemary Noble Assistant Professor, Structural Engineer- tures. Further, components can be prefabricated and mass pro-
ing, Kinetic Structures Laboratory, Dept. of Civil & Environmental duced, resulting in a reduction in fabrication costs and higher
Engineering & Earth Sciences, Univ. of Notre Dame, 159 Fitzpatrick quality products. Finally, ensuring component reusability enables
Hall, Notre Dame, IN, 46566. stakeholders to stockpile modules and use them for military oper-
3
Industrial Collaborator, BATir Dept., Univ. Libre de Bruxelles, ations and disaster relief, situations for which the site conditions
Avenue F. Roosevelt, 50, CP 194/2, 1050 Brussels, Belgium.
(i.e., span) and loading may not be known a priori. However,
Note. This manuscript was submitted on December 16, 2016; approved
on May 31, 2017; published online on September 1, 2017. Discussion pe- there is limited existing research investigating optimization of
riod open until February 1, 2018; separate discussions must be submitted modular structures incorporating reusability. Fellini et al. (2004)
for individual papers. This paper is part of the Journal of Bridge and Torstenfelt and Klarbring (2006) integrated the capability of
Engineering, © ASCE, ISSN 1084-0702. using car components in multiple automotive body structures by

© ASCE 04017094-1 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 a5 a5 a2

a1 a3 a4 a4 a3 a1 a4 a3 a1

a2 a2 a5
(a)

a1 a2 a3
(b)

(s) (s) (s)

λ m,1 λ m,2 λ m,3
Fig. 1. (a) Bailey Bridge panel (b) in a double-triple girder-type con-
(c)
figuration (reprinted from Department of the Army 1986)
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

am = λ m,1 a1 λ m,2 a2 λ m,3 a3

(s) (s) (s)
+ +
minimizing the total mass of the vehicles involved in the design, (d)
under all of the constraints occurring in each scenario. Kalyanmoy
et al. (2015) extended the latter methodology to multiobjective Fig. 2. Relationship between the module orientation and the
optimization of small-scale applications including a cantilever module topology denoted am: (a) module rotation; (b) correspond-
beam and a three-bar truss problem. The efficiency of these strat- ing topologies; (c) associated design variables; and (d) module
egies lies in the fact that they aggregate multiple scenarios into a topology
unique optimization problem that can be solved by standard opti-
mization strategies. However, most of the applications focus on
automotive engineering and do not consider large-scale civil Design Variables
applications. Conventional methods for optimizing material distributions for
bridge design use truss topology optimization with a ground struc-
ture approach (Descamps and Filomeno Coelho 2013). This method
Objectives and Scope
discretizes the design domain into a grid of interconnected truss ele-
ments (a so-called a ground structure), in which the cross-sectional
The objective of this paper is to develop a conceptual design strat-
areas are continuous variables (Dorn et al. 1964). This assumption
egy using module topology and spatial orientation optimization for
follows from the observation that most material distribution optimi-
modular bridges including (1) reusability of modules among multi-
zation procedures for continuum domains converge toward trusslike
ple structures and (2) practical considerations, such as multiple
configurations exhibiting low bending and shear (Bendsøe and
types of modules, multiple load cases, the capability to compute the
Sigmund 2003).
displacements in an elastic design formulation, and a limitation on
Following this approach, this paper considers the optimization
the stresses. This contribution is applied to the conceptual design of
of multiple types of modules by focusing on their cross-sectional
single- and a double-story simply supported bridges. First, the capa- ð Þ
vectors, denoted by ai 0 for the ith module type and concatenated in
bility of the methodology for designing lightweight structures is a global topology design variable as follows:
demonstrated by optimizing each bridge considered separately.
n o
Second, the impact of integrating component reusability is investi- ð Þ ð Þ ð Þ
að0Þ ¼ a10 ; …; ai 0 ; …; aN0T (1)
gated by considering the two structures simultaneously. Third, the
impact of module configuration (i.e., module type definition and
position in the structure) is studied, incorporating also the influence Including spatial orientation to module topology optimization
of module reusability. This contribution culminates in guidelines follows the strategy introduced in Tugilimana et al. (2017). This
that can assist designers in defining modular structure properties prior work emphasized that a continuous optimization formulation
(e.g., module definition, number of different modules) during pre- cannot be derived if rotations are modeled as rigid body displace-
liminary design phases. ments. Continuously rotating the modules inherently results in their
disconnection, making the application of the supports and the loads
impossible and, therefore, the analysis of the structure. To circum-
Optimization Problem vent this, Tugilimana et al. (2017) focused only on rotations ensur-
ing the module assembly and demonstrated that these rotations con-
This approach generalizes the formulation introduced in Tugilimana sist of a finite set of possible module orientations that can be
ð Þ
et al. (2017) to include (1) reusability of modules and (2) practical modeled by a permutation of the cross-sectional vector ai 0 [Figs.
design considerations. 2(a and 2b)]. The problem was therefore transformed into a discrete
First, some notations are presented. Multiple structures, indi- optimization problem, aiming to select the best module orientation
cated by the index s [ {1,…,smax} (where smax = total number of among this discrete set of rotations.
structures), are considered for the reusability of the modules. The To derive a continuous equivalent formulation, each module to-
different types of modules are denoted by the index i [ {1,…, NT}, pology has been expressed as a linear combination of all of the top-
where NT corresponds to the total number of different types of mod- ologies ensuring the module assembly, in which the variables of
ðs Þ
ules in each structure. Each structure s is subjected to a series of Nc this combination, denoted λm;p for the pth rotation of the mth mod-
load cases, indicated by the index k [ {1,…, Nc}. Each load case k ule in structure s, are considered as additional design variables
ðsÞ
in the structure s is represented by a force vector f k and is associ- [Figs. 2(c and d)]. The convergence toward only one of these con-
ðs Þ
ated with a displacement vector uk . The total number of modules figurations is ensured by the introduction of complementary con-
in the structure s is denoted by M(s). straints for each module m as follows:

© ASCE 04017094-2 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 X
Np suffers from the well-known stress singularity phenomenon, which
sÞ
λðm;p  1 ¼ 0 8m ¼ 1; …; MðsÞ ; 8s ¼ 1; …; smax (2) means that the optimality conditions cannot be satisfied when stress
p¼1 constraints are integrated (Achtziger and Kanzow 2008). To over-
come this issue, Stolpe and Svanberg (2003) considered the internal
ðsÞ
X
Np forces te;k as additional design variables (for the kth load case and in
sÞ
 1 ¼ 0 8m ¼ 1; …; MðsÞ ; 8s ¼ 1; …; smax
2
λðm;p (3) the eth element in structure s) and disaggregated the equilibrium
p¼1 equations as follows:

ðs Þ
where Np = number of coefficients λm;p in the mth module of struc- X
Ne;s
ðsÞ ðsÞ
X
Ne;s  
te;k cðesÞ ¼ f k þ r ae að0Þ ; kðsÞ lðesÞ gðesÞ
ture s, which depends on the ground structure node distribution and
ðs Þ e¼1 e¼1
connectivities of the module. For clarity, the coefficients λm;p are
concatenated in a spatial orientation design variable vector kðsÞ for 8k ¼ 1; …; Nc ; 8s ¼ 1; …; smax (5a)
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

structure s. This emphasizes an important characteristic of this

novel formulation: although the module topology variable a(0) is  
common for the different structures, the spatial orientation of the E ae að0Þ ; kðsÞ ðsÞ ðsÞ
ðsÞ
cðesÞ> uk ¼ te;k 8e ¼ 1; …; Ne;s ;
modules remains dependent on the considered structure, justifying le
the presence of the index s on the design variables kðsÞ .
In the remainder of the paper, any truss element e contained in 8k ¼ 1; …; Nc ; 8s ¼ 1; …; smax (5b)
 be written
structure s will  implicitly with respect to the design varia- ðs Þ ðsÞ
bles, i.e., ae að0Þ ; kðsÞ 8e ¼ 1; …; Ne;s , where Ne,s = total number of where ce and ge = direction cosine vector and the gravitational
bars in the structure. load vector of the eth element in the structure s; and r = material
density. The second term of Eq. (5a) represents the self-weight of
Objective Function and Constraints the structure and generalizes the formulation introduced in
Tugilimana et al. (2017) to account for the member self-weight.
To include the reusability of the modules between the smax numbers This research evaluates yielding of elements in tension and com-
of structures, this paper follows the formulation introduced in pression, neglecting global and member-level stability, which are
Torstenfelt and Klarbring (2006). This prior work minimizes the usually considered in later design stages (Briseghella et al. 2013;
sum of the objective functions corresponding to each structure Krog et al. 2002). Following the AASHTO LRFD bridge design
under all of the constraints involved in the multiple scenarios. On specifications (AASHTO 2012), the stresses in compression and
this basis, an objective function ensuring the design of cost-effective tension are limited to Uc Fy and Uy Fy, respectively, where Fy =
trusses can be formulated by focusing on the volume of the smax yield strength; and U = resistance factor (Uc = 0.9, Uy = 0.95).
structures, denoted vtotal, as follows: ðs Þ
Considering the internal forces te;k as additional design variables
leads to a formulation for the stress limitation in compression and
smax X
X Ne;s  
vtotal ¼ lðesÞ ae að0Þ ; kðsÞ (4) tension as follows:
s¼1 e¼1 ðs Þ
 
te;k  ae að0Þ ; kðsÞ Uc Fy 8e ¼ 1; …; Ne;s ;
ðsÞ
where le = length of the eth truss element in structure s. Eq. (4) can 8k ¼ 1; …; Nc 8s ¼ 1; …; smax (6)
be reformulated into a weighted sum of the volumes of the different
module types, with weights that correspond to the number of mod-
ðs Þ
 
ules in each type in the structures. Therefore, this objective function te;k  ae að0Þ ; kðsÞ Uy Fy 8e ¼ 1; …; Ne;s ;
is well suited for the present application because it will minimize
the volume of each module type, accounting for their repetition in 8k ¼ 1; …; Nc 8s ¼ 1; …; smax (7)
multiple structures.
Regarding practical design considerations, this paper first
extends the formulation introduced in Tugilimana et al. (2017) to Problem Formulation
consider multiple load cases. This is achieved by aggregating all of
the constraints (i.e., stress limitation and equilibrium) related to The novel formulation for the module topology and spatial orienta-
each load case into a unique formulation, ensuring that the opti- tion optimization, incorporating module reusability, multiple types
mized design is feasible for all load cases. of modules, multiple load cases with the self-weight, and stress lim-
The optimization formulation is then extended to ensure elastic itations, can be stated as follows:
equilibrium under multiple load cases. Focusing on the kth load
case, conventional truss topology optimization problems integrate X
smax X
Ne;s  
Minimize lðesÞ ae að0Þ ; kðsÞ (8a)
the equilibrium equations using a displacement analysis model, ðsÞ ðsÞ
að0Þ ;uk ;tk ;kðsÞ s¼1 e¼1
ðs Þ
considering the displacement field vector uk as additional design
variables (Bendsøe and Sigmund 2003). Such an approach (a so-
called simultaneous analysis and design) includes a series of addi- X
Ne;s
ðs Þ ðsÞ
X
Ne;s  
tional equality constraints ð0Þ ðsÞ ðs Þ
 of theform K a ; k uk  f k ¼ 0
ðsÞ
subject to te;k cðesÞ  f k  r ae að0Þ ; kðsÞ lðesÞ gðesÞ ¼ 0
ð0Þ ðsÞ
for structure s, where K a ; k = stiffness matrix parameterized e¼1 e¼1
by the topological and rotational design variables. Therefore, this (8b)
formulation searches the cross-sectional areas a(0) and the displace-
ðs Þ
ment vectors uk , such that at the solution the previously mentioned  
equality constraints are enforced, ensuring elastic equilibrium of all E ae að0Þ ; kðsÞ ðsÞ ðsÞ
ðsÞ
cðesÞ> uk  te;k ¼ 0 8e ¼ 1; …; Ne;s (8c)
the structures considered in the process. However, this formulation le

© ASCE 04017094-3 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

   ðsÞ
ae að0Þ ; kðsÞ Uc Fy  te;k  0 8e ¼ 1; …; Ne;s (8d) Initialization
a(0), uk(s), t(s)
k ,λ
(s)

ðsÞ
 
te;k  ae að0Þ ; kðsÞ Uy Fy  0 8e ¼ 1; …; Ne;s (8e)
Compute v totalwith Eq. (8a)
and constraints with Eq. 8(b-g)
8k ¼ 1; …; Nc ; 8s ¼ 1; …; smax
Update
Check optimality using a(0), uk(s), t(s)
k ,λ
(s)
X
Np
sÞ
λðm;p  1 ¼ 0 8m ¼ 1; …; MðsÞ ; 8s ¼ 1; …; smax (8f ) a Lagrangian formulation using a line-search
p¼1 or a trust-region
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

X
Np
Convergence?
No
sÞ
 1 ¼ 0 8m ¼ 1; …; MðsÞ ; 8s ¼ 1; …; smax
2
λðm;p (8g)
p¼1
yes

in which Eqs. (8b) and (8c) ensure equilibrium, Eqs. (8d) and (8e) End
limit the stresses in compression and tension respectively, and Eqs.
(8f) and (8g) allow the algorithm to converge toward a configuration Fig. 3. Flowchart summarizing the iterative resolution of Eqs. (8a)–(8g)
in which all of the modules are correctly assembled.
Eqs. (8a)–(8g) present attractive outcomes compared with the
state of the art in optimization of modular structures. First, it inte-
grates the simultaneous optimization of the module rotation and
spatial orientation, making the strategy competitive with respect to
any other approaches neglecting the module rotation (Huang and
Xie 2008; Chen et al. 2010). Second, it generalizes the existing plas- Deck Abutment
tic design formulation already reported in Tugilimana et al. (2017)
to a more general framework corresponding to the preliminary D Girders
design of truss systems. Finally, this methodology includes the reus- L
ability of the modules among multiple structures, with each one
4.572
subjected to different load cases. m
From a practical viewpoint, Eqs. (8a)–(8g) can be efficiently
solved by an interior-point method coupling line-search and trust- Fig. 4. Simply supported bridge case studies (Note: Single-story struc-
region methods. This strategy iteratively updates the design variables ture: span L = 24.5 m, depth D = 1.52; double-story structure: span L =
ðsÞ ðsÞ
a(0), uk , tk , and kðsÞ using a gradient-based method. At each itera- 48.7 m; depth D = 3.05 m)
tion, the objective function vtotal and constraints [Eqs. (8b)–(8g)] are
evaluated. The convergence criteria are based on a Lagrangian for-
mulation, which exploits the stationarity of the objective function Conventional Bailey bridges use a through-type deck. In this
and the feasibility of the constraints [Eqs. (8b)–(8g)] with respect to paper, a configuration involving a deck on the top of the girders
the design variables. The update scheme uses a combination of a line (Fig. 4) is investigated as an improvement on the state of the prac-
search and a trust region, depending on the positive semidefiniteness tice because it (1) prevents vehicular collisions with the load
of the Hessian of the Lagrangian. A flowchart summarizing the nu- bearing structure and (2) enables more efficient lateral bracing
merical procedure is depicted in Fig. (3). More details can be found between the girders (because these no longer conflict with vehicu-
in Byrd et al. (1999) and Byrd et al. (2000). Note that the simultane- lar traffic). Each bridge carries one lane of vehicular traffic on a
ous analysis and design in Eqs. (8a)–(8g) does not require any deck supported by two girders spaced at 4.572 m (15 ft), to allow
 analysis
explicit structural  ðsof the truss structures [i.e., resolution of a 3.658-m (12-ft) design lane as prescribed by AASHTO (2012).
Þ ðsÞ
the system K að0Þ ; kðsÞ uk ¼ f k ]. The algorithm iteratively con- As a simply supported bridge, one extremity of each girder is
verges toward a solution respecting Eqs. (8b) and (8c). At this point, restrained in translation in all directions, whereas the other one is
for a cross-sectional vector a(0) and module rotations kðsÞ , the struc- free to translate along the longitudinal axis.
ðs Þ All elements are assumed to be ASTM A36 steel (AASHTO
ture is at equilibrium under the applied loads f k .
2012), characterized by a yield strength Fy of 248 MPa (36 ksi), a
Young’s modulus E of 200E3 GPa (29,000 ksi), and a density r of
Case Studies 7,800 kg/m3 (0.28 lb/in3).The dead load (DL) includes the self-weight
of the structure and the deck [a lightweight deck is assumed with a
The conceptual design strategy proposed in this paper is demon- self-weight of 13.1 kN/m (900 lb/ft)]. For the live load, the optimiza-
strated for two simply supported bridge case studies spanning 24.38 tion procedure is implemented for the worst loading on a simply sup-
m (80 ft) and 48.77 m (160 ft). These spans are representative of ported span. The vehicular loads are those prescribed by AASHTO
single- and double-story Bailey systems, respectively (Department (2012) and correspond to a uniformly distributed lane load (LA) of
of the Army 1986), offering the opportunity for comparison with a 9.34 kN/m (640 lb/ft) on the full span, and a design truck load (TL) of
state-of-the-practice system. A depth of 1.524 m (5 ft) and 3.048 m 35.58/142.3/142.3 kN (8/32/32 kip) positioned at midspan [Fig. 5(a)].
(10 ft), respectively, were selected to represent the depth of these The tandem loads are not considered because the design TL governs
Bailey systems and to ensure that both structures exhibit the same for this simply supported system. Wind loads (WLs) of 2.39 kPa (50
span-to-depth ratio (Fig. 4). psf) are applied separately in two opposite directions orthogonal to

© ASCE 04017094-4 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 the bridge length [Fig. 5(b)], leading to two different load cases Type 1 has a length of 1.524 m (5 ft), a height of 1.524 m (5 ft), and
(WL1 and WL2) associated with each wind direction. These loads a width 0.6 m (2 ft). These dimensions are selected because they
are combined according to the Strength V combination prescribed by have the same height but half the length of conventional panels to
AASHTO (2012). A multiple presence factor of 1.2 is applied to the take advantage of module rotation. Based on the precedent of exist-
live load because only one vehicular lane is included. A dynamic ing panelized bridges, these dimensions ensure that the module is
impact factor of 1.33 is associated with the TLs. The two load combi- easily transportable and manageable for erection on site. Module
nations (LCs) considered in the analysis are Type 1 consists of a ground structure with 120 truss elements [Figs.
6(a–d)]. The girder of the single-story structure is consequently di-
LC1 : 1:25 DL þ 1:35  1:2 LA þ 1:35  1:2  1:33 TL vided into 16 modules, and the girder of the double-story structure
is 64 modules stacked in two layers [Figs. 6(e and f)]. The lateral
þ 0:4 WL1 (9)
bracing consists of three different types of modules, Module Types
2 and 3 [Figs. 7(a–e)], which consist of diagonal elements and
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

LC2 : 1:25 DL þ 1:35  1:2 LA þ 1:35  1:2  1:33 TL Module Type 4 [Fig. 7(f)], which is a single transverse element.
Module Type 4 is inspired by conventional panelized systems, in
þ 0:4 WL2 (10) which the deck is supported by transverse beams (transoms) every
3.048 m (10 ft). Module Types 2 and 3 offer the opportunity for
ðsÞ
where DLs, TLs, and WLs are concatenated in the force vectors f k ; improved lateral bracing strategies for these systems. Figs. 7(g and h)
and the load factor of 1.25 is applied to the self-weight in Eq. (8b). depict the position of each type of module in the single- and double-
story bridges, respectively. The corresponding number of modules
in each structure and their number of elements in each type are
Definition of Module Configurations indicated in Table 1.
Although Configuration I follows a module distribution similar
This paper investigates the impact of the module definition (i.e., to the state-of-the-practice systems, Configuration II (Figs. 8 and 9)
ground structure) and distribution (i.e., position in the structure) on aims to improve a deficiency of the existing systems: a concentra-
the optimized design. This is performed by considering the follow- tion of material along the neutral axis of the girders. The module di-
ing two module configurations. vision for the lateral bracing remains the same as Configuration I
In Configuration I, four types of modules are considered (Figs. 6 (Fig. 7). The difference lies in the module girders, which are now di-
and 7). Each girder is comprised of rectangular modules: Module vided into a rectangular box [denoted Module Type 5, Figs. 8(a–d)]

35.58kN 142.3kN 142.3kN

9.340 kN/m

4.267m 4.267m WL1 WL2

(a) (b)

Fig. 5. Live load distribution on the simply supported bridge shown in (a) elevation and (b) cross section

Module Type 1

1.524 m

1.524 m
1.524 m 0.600 m
(a) (b)
0.600 m 1.524 m
(d)
0.600 m

(c)

(e)

(f)

Fig. 6. Module Type 1 shown in (a) elevation, (b) cross section, (c) plan, and (d) isometric views; these modules form the girders for Configuration I
in the (e) single-story and (f) double-story structures

© ASCE 04017094-5 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 Module Type 2 Module Type 3

1.524 m

4.5
3.05 m 4.572 m 72
m
(a) (b) 3.048 m
(e)

Module Type 4
4.572 m 1.524 m
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

4.5
72 4.5
m 72
(c) 3.048 m m
(d) (f)

1.524 m Type 1
1.524 m
1.524 m Type 2
8 m
4.572 m 3.04 Type 3
m
48 Type 4
(g)
4.572 m 3.0
(h)

Fig. 7. Module Type 2 shown in (a) elevation, (b) cross section, (c) plan, and (d) isometric views; Module Type 3 shown in (e) isometric view;
Module Type 4 shown in (f) isometric view; module positions for Configuration I in the (g) single-story and (h) double-story structures

Table 1. Module Types for Configuration I A more advanced design step would require the consideration of
WL on live loads and WLs applied only on the surface of the truss
Number of modules elements.
Module type Number of bars Single-story Double-story
1 120 32 128 Results
2 4 8 32
3 2 16 48 Implementing the new optimization methodology [Eqs. (8a)–(8g)]
4 1 18 51 on the two case studies using Configurations I and II leads to opti-
mized designs that demonstrate (1) the capability of the methodol-
ogy to efficiently optimize module topology and spatial orientation
and a planar rectangular component [denoted Module Type 6, Fig.
including different types of module and practical design constraints,
8(e)]. Module Type 6 is only positioned at the bottom and the top of
(2) the impact of module reusability, and (3) the impact of the mod-
the girders and is not placed at midheight for the double-story struc-
ule configuration on the optimized designs. Results are summarized
ture (Fig. 9). A summary of the number of modules in each structure
and compared in Table 3.
is given in Table 2 for Configuration II.
Regarding the load implementation on the modules of
Configurations I and II, the self-weight of the deck and the live Demonstration of Methodology
loads are equally distributed on each girder. Because Module Type This paper first demonstrates the optimization strategy for the
4 mimics the transverse elements supporting the deck, these loads single- and double-story Configuration I structures because they
are applied every 3.048 m (10 ft) at the top chord, at the module most closely represent the state of the practice. Results do not
joints. This corresponds to the extremities of Module Type 1 for include component reusability.
Configuration I and Module Types 5 and 6 for Configuration II. Optimized Module Type I in single-story Configuration I [Figs.
Their magnitudes are taken proportionally to the area of the deck 10(a–d)] is composed of large horizontal elements on the bottom
supported by each transverse element. For the WL, the horizontal and top, forming the chords that resist flexure in the global bridge
pressure of 2.39 kPa (50 psf) is projected at the four corners of each structure. These are joined by diagonal elements for shear resist-
Module Type 1 for Configuration I and Module Type 5 for ance. The optimized spatial orientation of the modules along the
Configuration II, assuming the entire plain vertical square surface girder is such that there is a 180° rotation about the axis perpendicu-
of the 1.524-m (5-ft) side is exposed to the wind. This conservative lar to the span [Fig. 11(a)] for one of every two modules. This spa-
application of the WL enables its easier implementation for the pre- tial orientation ultimately results in a Warren truss type system [Fig.
liminary design stage. This implementation procedure, coupled 10(j)], which is conventionally known for its economy because it
with the definition of load cases LC1 and LC2 [Eqs. (9) and (10), does not include vertical members. For lateral bracing, the optimi-
ðsÞ
respectively], enables the definition of the external force vector f k . zation procedure has removed Module Type 4 (i.e., the transverse

© ASCE 04017094-6 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 Module Type 5 Module Type 6

1.524 m

1.524 m
1.524 m 0.600 m
(a) (b) 1.524 m
0.600 m
0.600 m 1.524 m (e)
0.600 m (d)
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

(c)

Type 5 Type 6

(f)

(g)

Fig. 8. Module Type 5 shown in (a) elevation, (b) cross section, (c) plan, and (d) isometric views; Module Type 6 shown in (e) isometric view; these
modules form the girders for Configuration II in the (f) single-story and (g) double-story structures

Type 2
Type 5
Type 3
Type 6
Type 4
(a) (b)

Fig. 9. Module positions for Configuration II in the (a) single-story and (b) double-story structures

Table 2. Module Types for Configuration II topology and spatial orientation optimization. The optimized
rotations of Module Type 1 in the bottom layer of the girder are
Number of modules
identical to that in the single-story structure, whereas the compo-
Module type Number of elements Single-story Double-story nent in the top layer has rotated 180°[(Fig. 11(b)]. This orienta-
1 — — —
tion, coupled with the optimized Module Type 1 in Fig. 12(d),
2 4 8 32 enables the elimination of material at the neutral axis, solving the
3 2 16 48 issue of material inefficiency imposed by module topology perio-
4 1 18 51 dicity. Also, in Fig. 12(i) the optimized material distribution in
5 64 32 128 the girders exhibits strong similarities with the Howe truss sys-
6 28 64 128 tem, in which the diagonal members are subjected to compression
and the vertical elements are in tension (under the loads consid-
ered here). The large top and the bottom chord ensure sufficient
elements) by enforcing its cross section to be zero. Module Type 2 bending resistance. The simultaneous spatial orientation and
[Figs. 10(e–h)] only includes two diagonal members with the same module topology optimization enables the definition of a Module
orientation in half of the structure. The symmetry with respect to Type 1 that is lighter for the double-story structure than for the
midspan is ensured by the spatial orientation optimization. Module single-story system (Table 3). For lateral bracing, Module Types
Type 3 is composed of two crossing members [Fig. 10(i)]. Coupled 3 and 4 have been removed. Module Type 2 is composed of four
with Module Type 2 and the crossing elements in the diagonal of crossing members with two thicker elements in the diagonal plane
Module Type 1, these modules enable the transmission of the WL of the rectangular prism defining the module [Figs. 12(e–h)]. The
to the supports and maintain the internal forces below the maximum optimized orientation of the lateral bracing modules follows that
allowable value. of Module Type 1: one of every two modules along the bottom
The optimized Module Type 1 in double-story Configuration layer is rotated 180° around the vertical axis, and the modules on
II [Figs. 12(a–d)] exhibits topological characteristics similar to the top layer are also rotated 180° around the longitudinal axis.
that of the single-story structure, except that no large elements are This enables the generation of a bracing system linking the top
generated at the top side of the component. This topological dif- and the bottom chords, as well as the neutral axis, for the trans-
ference results from the simultaneous optimization of module mission of the WL to the supports.

© ASCE 04017094-7 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 Table 3. Detailed Comparison of the Optimized Module Type Volumes, Optimized Structure Volumes, and Material Efficiencies for the Single- and
Double-Story Structures Including or Not Module Reusability

Mod. Type Vol. (10−3 m3)

Struct. 1 2 3 4 5 6 Struct. Vol. (10−3 m3) Mat. Eff. (m2/kN)
Single-story: Conf. I 29.09 0.364 0.364 0 — — 939.7 8.269
Double-story: Conf. I 26.96 0 3.497 0 — — 3.570 8.706
Reus.: Conf. I 39.00 0 4.312 0 — — 1.284 (S) 5.144 (D) 6.05 (S) 6.043 (D)
Reus.: Conf. II — 0.560 2.713 0 8.506 18.62 1.495 (S) 3.587 (D) 5.2 (S) 8.666 (D)
Note: Struct. = structure; Mod. = module; Vol. = volume; Mat. = material; Eff. = efficiency; S = single-story; D = double-story; Conf. = configuration;
Reus. = reusability.
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

Module Type 1 Module Type 2

1.524 m 1.524 m

3.048 m 4.572 m
1.524 m 0.600 m (f)
(e)
(a) (b)

0.600 m (d)

(c)
4.572 m
Module Type 3

4.57 (h)
2m 8m
3.04 (g)
(i)

(j)

(k)

(l)

Fig. 10. Single-story structure without module reusability for Configuration I; optimized Module Type 1 shown in (a) elevation, (b) cross section,
(c) plan, and (d) isometric view; optimized Module Type 2 shown in (e) elevation, (f) cross section, (g) plan, and (h) isometric view; isometric view
of the optimized Module Type 3 in (i); the optimized single-story structure is shown in (j) elevation, (k) plan view, and (l) isometric view

The efficiency of these two systems is demonstrated through a the self-weight because the bending moment of a simply supported
comparison with the state-of-the-practice Bailey bridges. This pa- bridge is proportional to the span squared. The self-weight of the
per uses the material efficiency metric introduced in Gerbo et al. Bailey bridge is based on data available from the Department of the
(2016), which is defined to be the span length squared divided by Army (1986) and includes the panels [2.566 kN (577 lb)], the end

© ASCE 04017094-8 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 180° Impact of Module Configuration
Because module reusability negatively impacts the material effi-
ciency of the double-story structure, an alternative design using
180° 180° Configuration II is investigated.
The results (Fig. 14) show that optimized Module Type 5 [Figs.
14(a–d)] consists of the diagonal elements required for carrying
shear, whereas Module Type 6 [Fig. 14(e)] ensures the bending re-
180° 180° sistance of the global bridge structure. Using Module Type 5 with
(a) (b) Module Type 6 on its bottom and top sides enables the definition of
a component similar to Module Type 1 in the Configuration I
Fig. 11. Optimized rotation of Module Type 1 in the (a) single-story
single-story structure, whereas only using Module Type 6 on the
structure and the (b) double-story structure
bottom side mimics Module Type 1 in the double-story structure
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

without reusability. In the single-story structure, the use of Module

posts [0.56 kN (126 lb)], the transoms [2.749 kN (618 lb)], the sway Types 5 and 6 to form the girders results in a slightly heavier struc-
bracing [0.302 kN (68 lb)], and the vertical and horizontal bracing ture than the one obtained with Configuration I using reusability
of the panels [0.196 kN (44 lb)]. Because this work neglects global because the topology periodicity of Module Type 6 requires that the
stability and the presence of the deck, the stringers and other struc- bottom and the top sides of the girders be identical. However, the
tural elements contributing to the lateral stability are not included. use of two different module types in the girders enables the genera-
According to the dimensions of those structural elements, a double- tion of a double-story structure free of material at the neutral axis,
single structure (analogous to the single-story structure investigated reducing material inefficiency obtained by topology periodicity
here) and a double-double structure (analogous to the double-story when reusability is included. This improvement is followed by the
structure) have material efficiencies of 4.16 and 5.127, respectively. generation of small inclined and vertical elements in Module Type 5
The material efficiencies shown in Table 3 demonstrate the that are not present in Module Type 1 of Configuration I. These are
improvement of the proposed systems compared with the state-of- required to ensure elastic equilibrium at the neutral axis because of
the-practice Bailey bridges. However, conventional panelized sys- the absence of Module Type 6 at this position.
tems are usually designed to be used in multiple applications. A material efficiency of 5.199 is obtained with Configuration
Therefore, a more relevant comparison with existing modular II single-story structure including reusability (Table 3). The
bridges requires the integration of module reusability. This is decrease of efficiency compared with the solution involving
addressed by the next subsection. Configuration I results from the topology periodicity of Module
Type 6 and the generation of the small inclined and vertical ele-
ments in Module Type 5. However, a significant improvement is
Impact of Module Reusability
brought to the double-story structure by Configuration II, in
To demonstrate the impact of module reusability, the single- which a material efficiency of 8.666 is obtained. Despite the con-
and double-story structures of Configuration I are optimized sideration of reusability, this value stays close to that obtained in
simultaneously. the double-story structure without reusability in Configuration I,
Optimized Module Type 1 in Configuration I [Figs. 13(a–d)] demonstrating the capability of the present methodology with
exhibits strong topological similarities with the single-story solu- Configuration II to generate lightweight modular structures.
tion without reusability, but differs from the double-story solution
by the presence of large horizontal elements at the top side of the
module. Indeed, the large horizontal elements at the bottom of Discussion
the module are generated by the double-story structure, whereas
the elements at the top of the module result from the considera- As indicated in Table 3, the proposed approach is able to design
tion of the single-story structure. Both top and bottom sides of the lightweight modular bridges exhibiting competitive material effi-
modules are linked by diagonal members with a cross-sectional ciencies compared with the existing panelized systems. This
area close to that from the solution without module reusability in improvement is explained by the simultaneous consideration of the
the double-story structure. Integrating reusability leads to a sin- module topology and spatial orientation optimization, limiting the
gle-story structure exhibiting strong similarities with a Warren negative impact of topology periodicity imposed by modularity.
truss system [Figs. 13(e and f)], whereas the double-story struc- This is clearly illustrated in Fig. 12(i), in which the girders of the
ture now possesses large horizontal elements at the neutral axis double-story structure seem to be composed of square modules with
because of topology periodicity and the interaction with the sin- a length of 3.048 m (10 ft), whereas Module Type 1 is only a length
gle-story system. For lateral bracing, the solution stays close to of 1.524 m (5 ft). Simultaneously optimizing module topology and
that in the double-story structure without reusability, because the module spatial orientation, therefore, enables the grouping of the
wind effects are dominant when the surface exposed to the wind modules into larger components, which is equivalent to a reduction
is greatest, corresponding to the larger span and the greater depth of the total number of modules and a decrease of the negative
of the structure. impact of topology periodicity.
The impact of module reusability is clearly illustrated by focus- The impact of module reusability on the optimized topology and
ing again on the material efficiency (Table 3). The simultaneous optimized material efficiency has been clearly demonstrated by the
consideration of the two structures forces the optimization proce- simultaneous consideration of the single- and double-story struc-
dure to merge the topological characteristics of each optimized tures. The optimized spatial orientations remain unchanged whether
design when it is considered separately. Compared with the opti- module reusability is included or not, but the optimized module top-
mized designs without reusability, the material efficiencies have ologies clearly gather the topological characteristics of the solutions
decreased but remain higher than that of the state-of-the-practice of each structure considered separately. From a material distribution
Bailey bridges. viewpoint, Configuration I single-story structure with reusability

© ASCE 04017094-9 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 Module Type 2

Module Type 1 1.524 m

3.048 m 4.572 m
1.524 m (e) (f)

1.524 m 0.600 m
(a) (b)

4.572 m
0.600 m (d)
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

(c)

(h)
(g)

(i)

(j)

(k)

Fig. 12. Double-story structure without module reusability for Configuration I; optimized Module Type 1 shown in (a) elevation, (b) cross section,
(c) plan, and (d) isometric view; optimized Module Type 2 shown in (e) elevation, (f) cross section, (g) plan, and (h) isometric view; the optimized dou-
ble-story structure shown in (i) elevation, (j) plan view, and (k) isometric view

stays close to the solution without reusability. However, the mate- II strongly limits the deterioration of the material efficiency for the
rial efficiency of the double-story structure significantly decreases double-story structure (0.46% of decrease compared with the solu-
because of topology periodicity of Module Type 1 coupled with the tion without reusability), whereas the single-story sees its material
consideration of reusability, leading to the generation of material at efficiency decreasing a little. This evolution is followed by addi-
the neutral axis. However, measuring the impact of reusability on tional small elements in the optimized topology of the girder mod-
the design cannot only rely on the material efficiency. Indeed, ules. Therefore, selecting an appropriate module configuration for a
allowing structures to share a common set of modules (1) decreases given problem is a challenge for designers. Indeed, finding a module
the time required for designing multiple structures because only one definition that ensures material efficiency requires large modifica-
optimization process has to be performed; (2) decreases fabrication tions to the project, which can only be applied at preliminary design
cost because mass production can be implemented, which also stages. Moreover, module definition involves a trade-off between ef-
results in a higher quality product; and (3) improves the on-site ficiency (i.e., minimizing the impact of topology periodicity), low
erection because the modules can be stockpiled for future use, are manufacturing cost of the modules, and simple assembly of the com-
transportable, and are easily assembled because few connections ponents (through a small number of connections and a simple mod-
need to be made. ule topology). In the case studies presented in this research, Module
The comparison between Configurations I and II emphasizes the Configuration I uses only four different types of modules, limiting
sensitivity of the optimized design to the module definition and dis- the manufacturing cost and the number of connections required for
tribution. Including module reusability in Configuration I results in erection. However, these advantages result in a low material effi-
a decrease of the material efficiency of 27 and 31% for the single- ciency when module reusability is introduced. On the other hand,
story and double-story structures, respectively. Using Configuration Module Configuration II improves the efficiency by considering five

© ASCE 04017094-10 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 Module Type 1

1.524 m
0.600 m

1.524 m 0.600 m
(a) (b) (c) (d)
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

(e)

(f)

(g)

(h)

Fig. 13. Optimized Module Type 1 in Configuration I with reusability shown in (a) elevation, (b) cross section, (c) plan, and (d) isometric view; opti-
mized single-story structure shown in (e) elevation and (f) plan view; optimized double-story structure shown in (g) elevation and (h) plan view; only
Module Type 1 is depicted in detail (a–d) because most topological modifications occur in the girders and not in the lateral bracing

types of modules, but it exhibits a more complex topology. This neg- Guidelines for the Design of Modular Structures
atively impacts the erection cost and time because more connections
are required between the modules. This discussion highlights a challenge for designers: selecting a
Finally, these lightweight modular structures are obtained due to modular configuration at early design stages. However, this paper
the well-adapted mathematical structure of the optimization problem shows that preliminary decisions can help isolate module defini-
formulation [Eqs. (8a)–(8g)]. The relevant assumptions brought to tions and distributions that maintain a good compromise between
the initial design problem enable the derivation of the objective func- efficiency and constructability. These decisions are summarized by
tion and constraints written explicitly in terms of all of the design the following guidelines:
variables. This allows the application of sparse interior-point algo- • Defining module types: The module types should follow the
rithms coupling Newton step and trust-region methods, which are division of the structure into its different subsystems (e.g.,
well-known for the handling of large-scale optimization problems. girders that carry gravity loads in bending and lateral bracing
In the present application, Configuration I double-story structure that transmits WLs to the supports), and their ground structures
without reusability involves 39,077 design variables and 38,360 should be adapted to the load path they have to carry.
equality and 76,428 inequality constraints. When module reusability Following these rules in the present application, Module Type
is included, these values increase up to 49,425, 48,898, and 97,740, 1 (and Module Types 5 and 6 in Configuration II) have been
respectively. To illustrate the processing of Eqs. (8a)–(8g), chosen to be able to generate rectangular prism girders, with a
Configuration I single- and double-story structures without reusabil- sufficiently dense ground structure for light optimized designs.
ity converge after 9,981 s (99 iterations) and 22,431 s (165 iterations) Module Types 2–4 have been selected to mimic existing brac-
CPU on a 2  12 core Intel Haswell processor with 256 GB RAM. ing systems used in the bridge industry, and they are associated
With reusability, 37,896 (131 iterations) is required on the same with less dense ground structures to avoid convergence toward
computer. Note that these performances are purely illustrative complex topologies between the girders.
because the convergence speed and the number of iterations depend • Number of module types: Selecting the number of different
on the starting point. types of modules is also a trade-off between efficiency and

© ASCE 04017094-11 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 Module Types 5 and 6

1.524 m 0.600 m
0.60 4m
0m 1.52
1.524 m 0.600 m
(a) (b) (c) (d) (e)
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

(f)

(g)

(h)

(i)

Fig. 14. Optimized Module Type 5 in Configuration II with reusability shown in (a) elevation, (b) cross section, (c) plan, and (d) isometric view; opti-
mized Module Type 6 shown in isometric view in (e); optimized single-story structure shown in (f) elevation and (g) plan view; optimized double-
story structure shown in (h) elevation and (i) plan view; only Module Types 5 and 6 are depicted in detail (a–e) because most topological modifications
occur in the girders and not in the lateral bracing

ease of fabrication of components and erection. As shown in improve the total weight when module reusability is included.
the results, using few different types of modules (in Confi- However, this also results in more complex topologies with
guration I) can improve fabrication and erection, despite a numerous connections required between modules. Therefore,
decrease of material efficiency because of topology perio- module reusability should be used primarily for temporary
dicity when module reusability is included. At early design structures, in which the economy brought by reusability of the
stages, a relevant choice in the number of module types components can counterbalance the loss in material efficiency
could include the comparison of multiple possibilities as and the increase in fabrication complexity.
performed in this paper, with an ultimate decision according
to the technical specifications and particular demands of the
stakeholders. Conclusions
• Module dimensions: Dimensions of the ground structure of the
different module types should be mostly driven by constraints In this paper, a novel strategy for the design optimization of modular
related to transportation and on-site erection of the compo- structures is proposed, generalizing the existing formulation for mod-
nents. For structures composed of modules carried by individ- ule topology and spatial orientation optimization to (1) reusability of
uals (as opposed to using heavy lifting equipment), modules among multiple structures and (2) practical design consider-
conventional dimensions of 3.05 m (10 ft), used in precedents ations, such as multiple types of modules, multiple load cases, the
such as Bailey, Acrow, or Mabey-Johnson (Joiner 2001; capability of computing the displacements in an elastic design formu-
Russell and Thrall 2013), should not be exceeded. However, lation, and a limitation on the member stress. This contribution is
for projects using heavy lifting equipment, larger dimensions aimed at assisting engineers at early design stages, in which crucial
can be considered still within the constraints of transportabil- but complicated decisions have to be made about the module defini-
ity, and should be preferably transportable in ISO containers. tion, their distribution in the structures, and the possibility of includ-
• Module reusability: The effects of module reusability mostly ing the reusability of the modules. To address this challenge, the exist-
depend on the number of module types and their definition. ing plastic design formulation for optimizing module topology and
Increasing the number of module types can considerably module spatial orientation has been extended to include elastic

© ASCE 04017094-12 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094

 equilibrium in a disaggregated form, which is well suited to include Byrd, R. H., Gilbert, J. C., and Nocedal, J. (2000). “A trust region method
stress limitations without numerical singularity. The handling of reus- based on interior point techniques for nonlinear programming.” Math.
ability has been performed by considering the module topology as Program., 89(1), 149–185.
Byrd, R. H., Hribar, M. E., and Nocedal, J. (1999). “An interior point algo-
common design variables for the different structures, whereas the
rithm for large scale nonlinear programming.” SIAM J. Optim., 9(4),
module spatial orientations have been associated with each structure 877–900.
separately. The resulting large-scale optimization problem is well Chen, Y., Zhou, S., and Li, Q. (2010). “Multiobjective topology optimiza-
suited for the application of mathematical programming using sparse tion for finite periodic structures.” Comput. Struct., 88(11–12),
interior-point method and sequential quadratic programming. 806–811.
This methodology has been demonstrated for case studies of two Department of the Army. (1986). “Bailey bridge.” Field manual no. 5-277,
simply supported bridges. The results have demonstrated the practical Headquarters, Dept. of the Army, Washington, DC.
ability of the proposed methodology to define lightweight designs at Descamps, B., and Filomeno Coelho, R. (2013). “A lower-bound for-
mulation for the geometry and topology optimization of truss
preliminary stages. Additionally, the impact of module reusability has
Downloaded from ascelibrary.org by Technische Universitat Munchen on 05/08/23. Copyright ASCE. For personal use only; all rights reserved.

structures under multiple loading.” Struct. Multidiscip. Optim.,

been illustrated, in which it has been shown that there is a trade-off 48(1), 49–58.
between material efficiency and constructability. This work has also Dorn, W. S., Gomory, R. E., and Greenberg, H. J. (1964). “Automatic
emphasized that better designs can be obtained by modifying the design of optimal structures.” J. de Mecanique, 3, 25–52.
module configuration. This paper culminates in guidelines for design- Fellini, R., et al. (2004). “A sensitivity-based commonality strategy for fam-
ers at preliminary design stages, related to the module definition, ily products of mild variation, with application to automotive body
dimensions, and distribution within the structures. structures.” Struct. Multidiscip. Optim., 27(1–2), 89–96.
Future areas of research could investigate extensions of the pro- Gerbo, E., Casias, C., Thrall, A., and Zoli, T. (2016). “New bridge forms
posed methodology to meet requirements of further design stages. composed of modular bridge panels.” J. Bridge Eng., 10.1061
/(ASCE)BE.1943-5592.0000871, 04015084.
Most panelized bridge systems are composed of beam elements. Huang, X., and Xie, Y. M. (2008). “Optimal design of periodic structures
Therefore, one improvement could consist of integrating structural using evolutionary topology optimization.” Struct. Multidiscip. Optim.,
elements subjected to bending and shear, under additional constraints 36(6), 597–606.
well known to be dominant in a system such as member-level stability. Joiner, J. (2001). One more river to cross: The story of British military
Further, the lateral bracing modules should be investigated for their bridging, Pen and Sword Books, South Yorkshire, U.K.
ability to contribute to global stability of the system. Additional load Kalyanmoy, D., Zhu, L., and Kulkarni, S. (2015). “Multi-scenario, multi-
cases should be considered consistent with design code. Importantly, objective optimization using evolutionary algorithms: Initial results.”
earthquake loads should be included for design in seismic zones. Proc., 2015 IEEE Congress on Evolutionary Computation (CEC),
IEEE, New York, 1877–1884.
Krog, L., Tucker, A., and Rollema, G. (2002). “Application of topology, siz-
Acknowledgment ing and shape optimization method to optimal design of aircraft compo-
nents.” Proc., Altair Hyperworks 3rd UK Conf., Altair, Troy, MI, 1–12.
Moses, E., Fuchs, M. B., and Ryvkin, M. (2003). “Topological design of
The authors would like to thank the Fond National de la modular structures under arbitrary loading.” Struct. Multidiscip. Optim.,
Recherche Scientifique (FNRS, Belgium) for its financial support. 24, 407–417.
Russell, B., and Thrall, A. (2013). “Portable and rapidly deployable
bridges: Historical perspective and recent technology develop-
References ments.” J. Bridge Eng., 10.1061/(ASCE)BE.1943-5592.0000454,
1074–1085.
AASHTO. (2012). AASHTO LRFD bridge design specifications, customary Ryvkin, M., and Fuchs, M. B. (1999). “Optimal design of infinite repetitive
U.S. units, Washington, DC. structures.” Struct. Optim., 18(2–3), 202–209.
Achtziger, W., and Kanzow, C. (2008). “Mathematical programs with van- Stolpe, M., and Svanberg, K. (2003). “A note on stress-constrained truss to-
ishing constraints: Optimality conditions and constraint qualifications.” pology optimization.” Struct. Multidiscip. Optim., 25(1), 62–64.
Math. Program., 114(1), 69–99. Torstenfelt, B., and Klarbring, A. (2006). “Structural optimization of modu-
Acrow Bridge. (2016). “Acrow 300 series vs. 700XS series.” Proc., Florida lar product families with application to car space frame structures.”
DOT Design Conf., Florida Department of Transportation (FDOT), Struct. Multidiscip. Optim., 32(2), 133–140.
Tallahassee, FL. Tugilimana, A., Thrall, A. P., Descamps, B., and Filomeno Coelho, R.
Bendsøe, M. P., and Sigmund, O. (2003). Topology optimization: Theory, (2017). “Spatial orientation and topology optimization of modular
methods, and applications, Springer, Berlin. trusses.” Struct. Multidiscip. Optim., 55(2), 459–476.
Briseghella, B., Fenu, L., Lan, C., Mazzarolo, E., and Zordan, T. (2013). Wang, Y., Thrall, A. P., and Zoli, T. P. (2016). “Adjustable module for
“Application of topological optimization to bridge design.” J. Bridge variable depth steel arch bridges.” J. Constr. Steel Res., 126(Nov),
Eng., 10.1061/(ASCE)BE.1943-5592.0000416, 790–800. 163–173.

© ASCE 04017094-13 J. Bridge Eng.

J. Bridge Eng., 2017, 22(11): 04017094