Advances in Engineering Software 41 (2010) 286–294

Contents lists available at ScienceDirect

Advances in Engineering Software

journal homepage: www.elsevier.com/locate/advengsoft

Shape and discrete sizing optimization of timber trusses by considering

of joint flexibility
S. Šilih *, S. Kravanja, M. Premrov
Faculty of Civil Engineering, University of Maribor, Smetanova 17, 2000 Maribor, Slovenia

a r t i c l e i n f o a b s t r a c t

Article history: The paper presents the shape and discrete sizing optimization of timber trusses with the consideration of
Received 16 May 2008 joint flexibility. The optimization was performed by the Mixed-Integer Non-linear Programming (MINLP)
Received in revised form 31 January 2009 approach. In the optimization model an economic objective function for minimizing the structure’s self-
Accepted 17 July 2009
manufacturing costs was defined. The design conditions in accordance with Eurocode 5 were considered
Available online 13 August 2009
as optimization constraints. The internal forces and deflections were calculated by finite element analy-
sis. The structural stiffness matrix was composed by considering fictiously decreased cross-sectional
Keywords:
areas of all the flexibly connected elements. The cross-section dimensions and the number of fasteners
Discrete sizing optimization
Joint flexibility
were defined as discrete sizing variables, while the joint coordinates were considered as shape variables.
Mixed-Integer Non-linear Programming The applicability of the proposed approach is demonstrated through some numerical examples, pre-
Shape optimization sented at the end of the paper.
Timber Ó 2009 Elsevier Ltd. All rights reserved.
Truss

1. Introduction small, it can cause higher costs as a consequence of the larger

In the past decades, optimization has become a notable part of
the structural design process, especially in the preliminary parts of
the procedure. From a number of different types of engineering
structures, trusses are probably the most frequent grid-like struc-
tures, which have been applied to test different optimization tech-
niques since the early 1960s, Schmidt [1]. Compared to other types
of structures, the analysis and design of trusses is relatively simple
and can be easily written as a mathematical model. On the other
hand, trusses are in general composed from a high number of ele-
ments (bars). The ‘‘classical” method of optimizing a truss, namely
by repeating the analysis of different structural alternatives, is thus
not a convenient option.
Moreover, a number of design parameters has to be considered
when a truss is designed. The most important design parameters
are the type of truss configuration, the span/depth ratio (S/D) and
the number of intermediate members (diagonals and verticals). The
basic configuration types are the flat, the pitched and the bow-
string type. Each type has different advantages, e.g. while the
pitched truss is in general a more efficient solution when com-
pared to the flat type, the flat type is often preferred for reasons
such as easier serviceability of roofs. The S/D ratio has a direct
influence on the cross-sectional (CS) dimensions of bars. Although
a low S/D ratio enables the CS dimensions of bars to be relatively
* Corresponding author. Tel.: +386 2 22 94 300; fax: +386 2 25 24 179.
E-mail address: simon.silih@uni-mb.si (S. Šilih).
depth of the structure. The number of intermediate members
(diagonals and verticals) also influences the CS dimensions, i.e. a
higher number of bars in general allows the CS dimensions to be
smaller. Also, the stiffness of the structure is increased, and the
deflections are reduced. With a higher number of bars, however,
the number of connections is also higher which causes additional
costs and erection difficulties. It is also important to note that
not only each of the mentioned parameters affects the final truss
design independently, but different design parameters also influ-
ence each other. For example, a truss with a lower S/D ratio re-
quires less intermediate members, a pitched truss enables a
higher S/D ratio (smaller depth of the truss) than a flat one, etc.
Therefore, finding an optimal truss with respect to all design
parameters is in general quite a comprehensive and time-consum-
ing process.
The task of the classical structural analysis is to obtain a re-
sponse of a structure which is completely defined (the geometry,
CS dimensions, etc. are known). When structural optimization is
used on the other hand, different design parameters can be defined
as variables, which are in the calculation process optimized simul-
taneously in order that the final structure satisfies all the design
constraints. In the past different design parameters of trusses have
been optimized in this way. Topology optimization of trusses,
which stands for finding the optimal number and configuration
of bars, has been performed since the 1960s, see Dorn et al. [2]
and is still one of the main spheres of activities of structural opti-
mization, e.g. Kaveh and Kalatjari [3]. Another task is the so-called

0965-9978/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.advengsoft.2009.07.002
 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294
shape optimization of trusses, where the coordinates of the joints
of a truss are optimized, see e.g. Šilih et al. [4], enabling the optimal
S/D ratio as well as, e.g. the optimal slope of a pitched truss to be
calculated. However, most of the past research work was per-
formed on metal trusses, with some exceptions e.g. on composite
trusses, see Kravanja and Šilih [5], while timber trusses have been
quite neglected.
One of the most important characteristics of timber trusses is
that the joint connections are not as rigid as they are in steel or
concrete structures. Undesirable slips in joints lead to reduction
in truss stiffness and consequently the final deflections are in-
creased. Therfore, the type of joint connections is an additional
important timber truss design parameter. The flexibility of the
joints in timber trusses depends greatly on the selected type of
connections. Not only the type of connectors (bolts, nails, dowels),
but also their dimensions (diameter, length) as well as their num-
ber and disposition can have considerable influence on the struc-
tural design. For this reason, the timber truss design represents a
more comprehensive process when compared to the design of steel
or concrete trusses.
Timber suffers for the unfavourable material characteristic,
which do exhibits only limited ductility in compression and prac-
tically no ductility in tension. However, due their flexibility, steel
dowel type connections provide the needed ductility and energy
dissipation.
Well-built timber structures thus maintain a good performance
under wind and earthquaque loads. Timber trusses in particular
have become known for their pleasing architectural appearance,
lightweight design and easy fabrication. The use of timber trusses
to bridge over large open areas with a few or no intermediate sup-
ports is still on the increase. Timber trusses with metal-plate con-
nections have been found to be favourable structures for roof
framings for spans greater than 20 m.
The present paper represents a continuation of the research
work introduced in Šilih et al. [6]. In the preliminary work, the
optimization of timber trusses with special attention to joint flex-
ibility was carried out by using the Non-linear Programming (NLP)
approach. Sizing optimization of pitched trusses with a fixed slope
of the top chord was performed for different S/D ratios and differ-
ent numbers of intermediate members. The design constraints
were defined in accordance with the European standard for the de-
sign of timber structures, Eurocode 5 [7]. The corresponding ulti-
mate limit state (ULS) and serviceability limit state (SLS)
conditions were considered. An economic objective function repre-
senting the manufacturing costs of the structure was introduced. It
has been proven that flexible joint connections exercise a signifi-
cant influence on the final structural design. A comparative numer-
ical study demonstrated that the manufacturing costs can increase
by over 10% and the timber mass by over 15% if the flexibility of the
joints is additionally considered in the design process. The influ-
ence becomes greater when the number of flexibly connected
members increases. Trusses with a lower number of diagonals
and verticals have thus proven to be more convenient. Besides,
lower S/D ratios have also proven to be preferable. At a larger
depth of the truss, the stress-controlled (ULS) conditions become
decisive instead of the deflection-controlled (SLS) ones.
The study presented additionally includes the shape optimiza-
tion of timber trusses. In this way the S/D ratio (the depth of the
truss girders) as well as the slope of the top chord are optimized
simultaneously with the CS dimensions and joint connections (re-
quired number of embedded fasteners).
The second important contribution of the paper is that discrete
sizing optimization is also performed. In the previous research the
application of the continuous NLP optimization approach was
demonstrated. In this approach, the sizing variables (CS dimen-
sions, number of fasteners in each joint) take some real value be-
287
tween the defined lower and upper bound. In order to obtain a
structure of practical applicability, these variables have to be
rounded to the nearest upper discrete value (e.g. 1 cm for CS
dimensions; 1 fastener). Subsequently, the structure has to be
reanalysed. Using discrete sizing optimization, on the other hand,
the discrete/standard values for sizing variables are calculated
simultaneously during the optimization process. Alongside contin-
uous variables (shape varibles, stresses, deflections, etc.), discrete
variables are also included. The problem thus corresponds to a
mixed discrete/continous type of optimization problems, which
can be solved with the use of the Mixed-Integer Non-Linear Pro-
gramming (MINLP) optimization approach, Duran and Grossmann
[8]. For the solution of the defined problem, the Modified Outer
Approximation/Equality Relaxation (OA/ER) algorithm, Kravanja
and Grossmann [9], was applied. The MINLP optimization ap-
proach and the Modified OA/ER algorithm are discussed in Section
2 of the paper, whilst the main design constraints for timber
trusses are given in Section 3. The applicability and efficiency of
the proposed approach is shown through some numerical exam-
ples presented at the end of this paper. All optimizations are per-
formed at fixed topology, i.e. fixed number and disposition of
truss elements.
2. The MINLP optimization approach
2.1. The general MINLP model formulation
The general non-linerar continuous/discrete optimization prob-
lem (MINLP-G) can be formulated as follows:
min z ¼ cT y þ fðxÞ
subjected to : hðxÞ ¼ 0
gðxÞ 6 0
ðMINLP-GÞ
By þ Cx 6 b
x 2 X ¼ fx 2 Rn : xLO 6 x 6 xUP g
y 2 Y ¼ f0; 1gm
where x is a vector of continuous variables specified in the compact
set X and y is a vector of binary 0–1 variables. Functions f(x), h(x)
and g(x) are continuous and differentiable non-linear functions in-
volved in the objective function z, equality and inequality con-
straints, respectively. Finally, By + Cx 6 b represents a subset of
mixed linear equality/inequality constraints.
It should be noted that in the context of the considered timber
truss optimization, continuous variables define continuous struc-
ture parameters (nodal coordinates, internal forces, deflections,
etc.) while binary variables are linked to the discrete/standard CS
dimensions. Equality and inequality constraints and the bounds
on continuous variables represent a rigorous system of functions
taken from structural analysis. Logical constraints that must be ful-
filled for discrete decisions and structure configurations are given
by By + Cx 6 b. In this paper, the economic objective function z is
proposed to minimize the structure’s self-manufacturing costs. It
involves fixed costs charges in the term cTy for manufacturing,
whilst the dimension dependant costs are included in the function
f(x).
2.2. The Modified OA/ER algorithm
The OA/ER algorithm comprises the solving of an alternative se-
quence of Non-linear Programmig (NLP) optimization subproblems
and Mixed-Integer Linear Programming (MILP) master problems,
see Fig. 1. The former corresponds to the continuous optimization
of parameters for a mechanical superstructure with fixed discrete/
binary variables and yields an upper bound to the objective to be
288 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294

constraints, considered in the discussed timber truss optimization,

MINLP
are captured in Table 1. The explanation of the equations is given
Superstructure below.
The tensioned bars are checked for tensile resistance using Eq.
Combined optimization
(1), where rt,0,d is the design tensile stress along the grain, calcu-
lated by considering the net cross-sectional area of the tensioned
Fixed binary variables
element; ft,0,d is the design tensile strength parallel to grain; ft,0,k
is the characteristic tensile strength parallel to grain; kmod is the
modification factor which takes into account the effect of the dura-
NLP tion of the load and moisture content; and cm is the partial safety
Subproblem factor for a material property. The values for kmod are taken from
Table 3.1 and the values for cm from Table 2.3 of Eurocode 5 [7].
Continuous optimization The compressed bars are checked for compressive resistance (2)
New binary as well as for buckling resistance (3–4), where rc,0,d is the design
variables compressive stress along the grain; fc,0,d is the design compressive
MILP
strength along the grain depending on the characteristic compres-
sive strength fc,0,k, kc,y and kc,z represent the coefficients, which
Master problem
approximate the buckling about the y and z axes in accordance
Discrete optimization with Eurocode 5. The buckling coefficient about the y-axis kc,y is

Convergence ?
Table 1
NO ULS design constraints for timber trusses.

YES Tensile resistance of bars:

STOP
rt;0;d 6 ft;0;d ¼ kmod  ft;0;k =cm ð1Þ
Fig. 1. Steps of the outer approximation/equality relaxation algorithm.
Compressive/buckling resistance of bars:

minimized. The latter involves a global approximation to the rc;0;d 6 fc;0;d ¼ kmod  fc;0;k =cm ð2Þ
superstructure of alternatives in which new values of discrete/bin-
ary variables (new standard sizes) are identified so that its lower rc;0;d 6 kc;y  fc;0;d ¼ kc;y  kmod  fc;0;k =cm ð3Þ
bound does not exceed the current best upper bound. A global lin-
ear approximation includes the linearizations of the non-linear
objective function and the non-linear (in)equality constraints accu- rc;0;d 6 kc;z  fc;0;d ¼ kc;z  kmod  fc;0;k =cm ð4Þ
mulated at each NLP subproblem solution as well as the linear con-
straints from the original MINLP problem.  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
The OA/ER algorithm, Duran and Grossmann [10], like all other kc;y ¼ 1 ky þ ky  k2rel;y ð5Þ
MINLP methods, does not generally guarantee that the obtained
solution is the global optimum. This is due to the presence of non-
h   i
convex functions that may cut off the global optimum. In order to ky ¼ 0:5  1 þ bc  krel;y  0:3 þ k2rel;y ð6Þ
reduce the undesirable effects of nonconvexities, a Modified OA/ER
algorithm was developed [9]. The Modified OA/ER algorithm
allows the following modifications to be applied to the master qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
krel;y ¼ ky =p  fc;0;k =E0;0:5 ð7Þ
problem: deactivation of linearizations, decomposition and deacti-
vation of the objective function linearization, use of the penalty
function, use of the upper bound on the objective function as well ky ¼ ly =iy ð8Þ
as a global convexity test and the validation of the outer
approximations. Lateral resistance of fasteners:
In the past the Modified OA/ER algorithm was successfully used
for the discrete/continuous optimization of different types of struc- F v ;Rd  N 1d ð9Þ
ture, e.g. topology, standard and rounded dimension optimization
of hydraulic steel gates, Kravanja et al. [11,12], topology and dis-
F v ;Rd ¼ kmod  F v ;Rk =cm ð10Þ
crete sizing optimization of steel trusses, Šilih and Kravanja [13],
Kravanja et al. [12], standard sizing optimization of composite
beams, Kravanja and Šilih [14], Kravanja et al. [12], etc. 8
> fh;1;k  t1  d ðaÞ
>
> rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 
< 4My;Rk
F v ;Rk ¼ min fh;1;k  t1  d  2 þ f dt 2  1 ðbÞ ð11Þ
>
> h;1;k 1
>
: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3. Timber truss design criteria 2:3  My;Rk  fh;1;k  d ðcÞ

The timber design constraints for both the ultimate limit state
fh;1;k ¼ 0:082  ð1  0:01dÞ  qk ð12Þ
(ULS) and the serviceability limit state (SLS) are taken from Euro-
code 5 [7]. At the ULS, the truss members are checked for the ten-
sile or compressive/buckling resistance. The corresponding load
carrying capacity of the fasteners is also checked. The ULS design
 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294
determined using Eqs. (5)–(8). The factor bc in Eq. (6) equals to 0.2
for solid timber and 0.1 for glued laminated timber. krel,y is the rel-
ative slenderness ratio, calculated by Eq. (7), where ky is the slen-
derness ratio corresponding to bending about the y-axis, ly is the
buckling length of the compressed element corresponding to bend-
ing about the y-axis and iy is the radius of gyration of the cross-sec-
tion about the y-axis. The buckling coefficient about the z-axis kc,z
is determined the same way using Eqs. (5)–(8), where the indexes
‘‘y” should be replaced by ‘‘z”.
The fasteners in all joint connections should satisfy criteria (9),
where Fv,Rd is the design lateral load carrying capacity per shear
plane per fastener; N1d is the ultimate force acting on a fastener
per shear plane and Fv,Rk is the characteristic lateral load carrying
capacity per shear plane per fastener. Fv,Rk is calculated by Eq.
(11), where the steel plate of thickness t1 is considered to be placed
as the central member of a double shear connection (see Fig. 2).
fh,1,k is the characteristic embedment strength in timber and My,Rk
denotes the characteristic fastener yield moment. These values de-
pend on the type of fastener. When bolts or dowels are used, they
are calculated by Eqs. (12) and (13), where d is the fastener diam-
eter (in mm), qk is the characteristic timber density (in kg/m3) and
fu is the characteristic tensile strength of the fastener material (in
N/mm2).
It should be noted that the connections are assumed as being
perfectly centric, i.e. no additional bending moment due to eccen-
tricity is accounted for in the calculation. Therefore the connec-
tions have to be designed adequately. This is important in
particular at tensioned connections, while at compressed ones
the majority of the force is assumed to be transmitted by contact
stresses between timber elements.
The serviceability limit states for deflection require that the
maximal initial and final deflection must be calculated within a
specified range. The main distinction in the calculation of the tim-
ber truss deflection, when compared to steel or concrete truss
m  ðSLSÞ ð18Þ
deflection, is that an additional deformability of the intermediate
elements (diagonals and verticals) to the upper and lower chord
members has to be considered. Mechanical fasteners in timber
are not rigid, as they are in concrete or steel, but rather flexible.
Truss deflection thus increases with the flexibility of the joints
and can be described as a sum of:
f ¼ fo þ f1
The deflection fo represents the vertical displacement of the
truss if the flexibility of the fasteners is not considered. If the influ-
ence of the local bending moments is neglected, it depends only on
the axial forces Ni and the axial stiffness (EA/L)i of all i, i 2 I, truss
members in the form of:
XI
Ni  N i  Li
f0 ¼
i¼1
E i;mean  Ai
steel plate
timber
dowel
Fig. 2. Double shear connection with a steel plate as the central member.
289
where Ni is the axial force in the element i caused by the
 i is the axial load caused by a virtual unit force
external load; N
in the direction of the desired deformation and Ei,mean stands
for the mean value of the modulus of elasticity. The second
term f1 represents the deflection as a consequence of the joint
(fastener) flexibility of all m, m 2 M, intermediate timber mem-
bers (diagonals and verticals). It can be accounted for in the
form of:
XM m  1
Nm  N 1
f1 ¼  þ ð16Þ
m¼1
K ser km;1 km;2
where M represents the total number of intermediate timber
members, flexibly connected to the truss chords, km,1 and km,2
are the numbers of fasteners at both ends of the considered
mth element, and Kser denotes the fasteners’ slip modulus taken
for different types of fasteners from Table 7.1 of Eurocode 5
[7].
According to Table 7.1 in Eurocode 5 [7], the slip modulus Kser
for bolts and dowels, when considering two shear planes, should
be calculated as:
K ser ¼ 2  q1:5
m  d=35 ð17Þ
where qm is the mean timber density (in kg/m3) and d is the fas-
tener diameter (in mm).
Since the intermediate members are flexibly connected, their
stiffness actually decreases. In finite element analysis the joint
flexibility is considered in such a way that the cross-section areas
Am of all the intermediate members are replaced by a fictitiously
decreased cross-section area Am , see e.g. Steck [15]:
A
Am ¼
Em;mean Am 1
1þ Lm
 K ser km;1
þ K ser1km;2
Eq. (18) for Am yields the exact solution for Eq. (14) by consid-
ering the flexibility of the joints in the form of Eq. (16). It should
be noted that form Am can also be used for the calculation of inter-
nal forces, but by respecting the reduced value for the stiffness of
the fasteners. In this case, Eq. (2.1) from Eurocode 5 [7], Ku =
ð14Þ 2/3  Kser should be considered instead of Kser.
The total deflection defined in Eq. (14) should not exceed
the range of limiting values for beam deflections, depending upon
the level of deformation deemed to be acceptable. In the
following optimization procedure, the total deflection f is limited
with the recommended values given in Table 7.2, of Eurocode 5.
The limiting value of the instantaneous deflection for the sim-
ply supported beam is recommended as ranging from L/300 to
ð15Þ
L/500.
4. MINLP optimization of timber trusses
The MINLP optimization model for shape and discrete sizing
optimization of timber trusses was developed according to the
general MINLP model formulation (MINLP-G) discussed in Section
2 of this paper. The GAMS (General Algebraic Modelling System),
Brooke et al. [16], was used as the interface for mathematical mod-
elling and data inputs/outputs.
An economic objective function is defined in the model to
minimize the structure’s self-manufacturing costs subjected to
design, stress and stability constraints known from structural
analysis. Internal forces are proposed to be determined by the fi-
nite element equations, whilst the dimensioning is performed in
accordance with Eurocode 5 [7]. The objective function is thus
defined:
290 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294

Y xi
upper chord
,l
i

vertical
Ai l
i, na hi
go
dia
X bi
lower chord A i = bi×hi
Fig. 3. A plane timber truss with its typical elements.

X
I X
M X
J one discrete value qi,l is assigned to the standard design variable hi.
min cost ¼ ct  bi hi li þ ðcfm þ cfl Þ  2  km þ ðcfm þ cfl Þ  kj Design variables bi are determined in the same manner as hi.
i¼1 m¼1 j¼1
X
ð19Þ hi ¼ qi;l  yst i2I ð20Þ
i;l
l2LðiÞ
X
where cost represents the self-manufacturing (material and labour) yst
i;l ¼ 1 i2I ð21Þ
costs of the structure; ct denotes the price of the manufactured and l2LðiÞ
built-in timber material per m3; the sum of the products between
widths bi, heights hi and lengths li of i, i 2 I, timber members, repre-
sents the volume of the truss in m3 (see Fig. 3); cfm is the material
5. Numerical example
cost of one fastener together with the adjoining steel plates, whilst
cfl denotes the manual labour costs for the handling, assembling,
5.1. Input data
drilling and bolting, defined per fastener. Since the required number
of fasteners is equal for both ends of each intermediate member, the
The paper shows the numerical example originally presented in
total number of fasteners in the mth member is 2km, where
Steck [15], where calculations for a pitched truss with the maximal
km = km,1 = km,2. The last term of the objective function represents
(mid-span) height of 300 cm and ten upper chord members were
the sum of all fasteners required in the joints of the chord members.
made using classical structural analysis. In Šilih et al. [6] structural
The variable kj stands for the number of fasteners of the jth joint,
optimization was applied. Continuous sizing optimization was per-
j 2 J. It is evaluated considering the resulting force on account of
formed for the trusses with different numbers of upper chord bars,
the axial forces of all intermediate elements connected to joint j.
whilst the height of the truss as well as the slope of the upper
Since the dimensions of the steel plates depend directly on the
chord were considered as fixed.
number of calculated fasteners, the costs of the steel plates are in-
In the present research, discrete sizing optimization was ap-
cluded in the values cfm and cfl.
plied. Alongside this, the height of the truss and the slope of the
The input data of the optimization model are the global geom-
top chord were considered as variables. In this way the shape of
etry of the truss (span of the truss girder), the supporting and load-
the truss is also optimized in addition to the CS dimensions. The
ing conditions, the diameter of the considered fasteners, the
thickness of the metal plates, as well as the material characteristics
of all the used components (timber, fasteners, plates). Table 2
The cross-section dimensions bi and hi of i, i 2 I, truss timber The input data for the optimization.
members and the number of fasteners km and kj are defined as Geometrical data
independent sizing variables, while the coordinates xj and yj of j, Span L = 22.5 m
j 2 J joints are considered as independent shape variables. Distance between lateral supports of the top chord: 380 cm
The finite element equations for the calculation of the internal Load
forces and deflections of the structure are defined as equality con- Permanent action: g = 2.0 kN/m (constant uniform load)
straints. The stiffness matrix of the structure is composed by con- Variable action (snow): s = 5.0 kN/m (constant uniform load)

sidering the fictitiously decreased cross-section areas of all the Timber characteristics: glued laminated timber GL32h, homogeneous glulam, EN
intermediate timber elements (diagonals and verticals) in accor- 1194 [17] classification
fc,0,k = 29 N/mm2
dance with Eqs. (17) and (18), where Ku and Kser are considered
ft,0,k = 22.5 N/mm2
for the calculation of the internal forces and the deflections, E0,mean = 13,700 N/mm2
respectively. It should be noted that during the optimization pro- E0,05 = 11,100 N/mm2
cess, two FEM analyses were performed simultaneously, one for qk = 430 kg/m3
the calculation of the internal forces at the ultimate limit state qm = 500 kg/m3
(using the slip modulus Ku) and one for the calculation of the Fasteners: dowels M14, steel S 235
d = 14 mm
deflections at the serviceability limit state (using the slip modulus
fu,k = 360 MPa
Kser). The ULS and SLS design conditions, described in Section 3, are Kser = 8944.27 N/mm
defined as inequality constraints.
Steel plates: steel S 235
The discrete sizing variables bi of i, i 2 I, truss members are de- Thickness ts = 8 mm
fined with the introduction of discrete binary 0–1 variables. A vec-
ULS and SLS criteria (Eurocode 5)
tor of l, l 2 L(i), discrete values of the standard height alternatives kmod = 0.9; short term load (snow), service class S2 (moisture content
qi,l = {qi,1, qi,2, . . ., qi,l} as well as an associated vector of l binary corresponding to a temperature of 20 °C and the relative humidity of the
0–1 variables yst st st st
i;l ¼ fyi;1 ; yi;2 ; . . . ; yi;l g are proposed to be defined
surrounding air only exceeding 85% for a few weeks per year)
for each variable hi. The variable hi is then expressed as a scalar cm = 1.25 (glued laminated timber)
cm = 1.3 (fasteners)
product between vectors qi,l and yi,l (Eq. (20)), where the sum of
Allowed vertical displacement: L/300 = 7.5 cm
the l binary variables yi,l is equal to 1 (Eq. (21)). Consequently, only
 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294
global geometry, loads and material properties were assumed to be
the same as those in Steck [15]. They are presented in Table 2.
A study was done for four different numbers n = 6, 10, 14 and 18
of the upper (compressed) chord members; see Fig. 4. For each de-
fined truss arrangement a simultaneous shape and discrete/stan-
dard sizing optimization was performed. For the purpose of an
adequate subsequent comparison of results to solutions previously
presented in Šilih et. al. [6] the following assumptions and simpli-
fications have been defined in the optimization model:
 the secondary structure and the lateral bracing system is not
included in the optimization, thus;
 a partial cost optimization of plane timber trusses is performed;
 the distances between the lateral supports, i.e. the out-of-plane
buckling lengths of the top chord members, are considered to be
equal for all proposed alternative truss configurations;
 the shape and discrete/standard sizing optimizations for differ-
ent truss arrangements (see Fig. 4) are performed at fixed topol-
ogies (i.e. the number and disposition of bars do not change
during the optimization process).
The self-weight of the truss members was automatically deter-
mined through the optimization process by considering the actual
calculated element dimensions and added to the permanent load g.
Both the permanent and the snow load (see Table 2) are in the cal-
culation approximated as nodal forces.
5.2. Optimization parameters
The considered economic parameters for the objective function
Eq. (19) were the costs ct = 900 EUR/m3 for the GL32 h timber
material, while cfl = 1.5 EUR and cfm = 1.0 EUR for one M14 S 235
dowel (including the corresponding part of the steel plates). The
(a) n = 6
Fi
(b) n = 10
Fi
(c) n = 14 Fi
(d) n = 18 Fi
Fig. 4. A plane timber truss with four different numbers of upper chord members.
291
lo lo
lower bounds on the CS dimensions bi =hi were taken as
12/12 cm for chord members and 8/8 cm for intermediate mem-
bers (diagonals, verticals). The upper bounds on the CS dimensions
up up
bi =hi for all members were taken as 20/20 cm. The vectors of dis-
crete values for both CS dimensions, bi and hi were defined as
qbl  qhl ¼ 8; 9; 10; 11; 12; 13; 14; 15; 16; 17; 18; 19; 20 (cm).
The bounds on shape variables (see Fig. 3) were taken as:
lo up
h0 =h0 ¼ 87:5=300 cm, hlo/hup = 187.5/487.5 cm and alo/aup =
0°/15°. In this way, a flat truss (a = 0°) is also considered a possible
solution.
5.3. Optimization
The optimization was carried out by a user-friendly version of
the MINLP computer package MIPSYN, Kravanja, Soršak and Kra-
vanja [18], the successor of PROSYN, Kravanja and Grossmann [9]
and TOP, Kravanja, Kravanja and Bedenik [19]. MIPSYN represents
the implementation of the Modified OA/ER algorithm, described in
Section 2 of the paper. Within the optimization process, the GAMS/
CONOPT2 (Generalized reduced gradient method), Drud [20] was
used to solve NLP subproblems and the GAMS/Cplex 7.0 (Branch
and Bound) [21] was used to solve MILP master problems.
Since a binary variable yst
i;l has to be defined for each discrete va-
lue l, l 2 L, that can be attributed to a CS dimension of any element
i, i 2 I, the discussed discrete sizing optimization problems can con-
tain a very high number of binary variables. The solution of such
extensive discrete/continuous optimization problems may become
difficult. For this reason, a special technique for the reduction of
the number of binary variables, the so-called prescreening proce-
dure was applied. The initial step of the OA/ER algorithm (the 1st
NLP subproblem) corresponds to continuous optimization, where
all discrete dimensions are temporarily relaxed into continuous
parameters. The intermediate solution of the 1st NLP subproblem
α
h ΣFi = q⋅L
h0
α
h ΣFi = q⋅L
h0
α
h ΣFi = q⋅L
h0
α
h ΣFi = q⋅L
h0
292 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294

is thus a continuous one, where the CS dimensions take some real Table 6
values between the defined lower/upper bounds. Using the presc- Convergence of the Modified OA/ER algorithm (numerical example 4, n = 18).

reening procedure after the 1st NLP, only those binary variables are MINLP Value of objective No. of active/all CPU time (s)
activated, which define the discrete/standard dimensions lying in a iteration function (N) binary variables
defined, reasonably small neighbourhood of the continuous 1 1. NLP 2128.9645 0/806 7.578
dimensions calculated by the initial subproblem. All other binary 2 1. MILP 2134.9750 68/806 0.461
variables are de-activated, i.e. their value is set to zero. In this 2. NLP 2180.9378 2.711
3 2. MILP 2153.4890 0.711
way, in the subsequent discrete sizing optimization the set of ac-
3. NLP 2199.9299 3.258
tive binary variables is considerably smaller than the entire set of P
= 14.719
binary variables.
In the present numerical examples, only one lower and one
upper discrete value for each of the calculated continuous sizing
variable from the 1st NLP were activated in the prescreening, while The OA/ER algorithm was terminated when the solutions
all the other values were de-activated. In addition, symmetry was stopped improving, i.e. when the next NLP subproblem yielded a
also applied. As the trusses have symmetric geometry, loading and worse solution than the previous one. In all cases only few MINLP
supporting, the intermediate elements (diagonals and verticals) iterations as well as low CPU times had been spent until the opti-
which lie symmetrically regarding the vertical mid-span axis are mal solution was achieved (see Tables 3–6). Using the prescreening
expected to have the same CS dimensions. By fixing this symmetry, procedure and applying symmetry resulted in a considerable
the number of independant sizing variables and thus the number reduction in the set of active binary variables. Regarding the whole
of binary variables was additionally reduced. sets of binary variables, less than 10% of binary variables were ac-
The convergence of the Modified OA/ER algorithm for the per- tive in all four cases of MINLP optimization.
formed MINLP optimizations is presented in Tables 3–6, where
the numbers of active binary variables are also shown. 5.4. Results

Table 3 The optimal results obtained for the discussed truss considering
Convergence of the Modified OA/ER algorithm (numerical example 1, n = 6). different topologies (defined by the number of upper chord ele-
ments n) are shown in Table 7. As can be seen, both the minimal
MINLP Value of objective No. of active/all CPU time (s)
iteration function (N) binary variables manufacturing costs as well as the minimal timber mass are ob-
tained at the topology with the lowest number of elements
1 1. NLP 1607.0777 0/338 2.410
2 1. MILP 1636.3700 32/338 0.781 (n = 6). The costs and mass increase monotonically when the num-
2. NLP 1683.0274 0.469 ber of truss elements increases, which is graphically presented in
3 2. MILP 1639.8390 0.220 Fig. 5.
3. NLP 1703.8376a 0.559
4 3. MILP 1641.2600 0.231
4. NLP 1707.7492a 0.590
Table 7
5 4. MILP 1674.9060 0.250
The obtained optimal results.
5. NLP 1685.2760 0.340
P
= 5.850 n=6 n = 10 n = 14 n = 18

a Total costs (EUR) 1683.03 1842.18 1987.01 2180.94

Locally infeasible solution.
Timber mass (kg) 700.29 723.44 771.95 851.35
Lower chord dimension b/h (cm) 12/13 12/13 12/13 12/13
Upper chord dimension b/h (cm) 14/14 16/13 16/13 18/13
Depth at mid-span (h) (cm) 465.05 319.55 334.91 350
Table 4 Depth at edges (h0) (cm) 212.98 120.81 97.03 94.54
Convergence of the Modified OA/ER algorithm (numerical example 2, n = 10). Slope of top chord (a) (°) 12.63 10.02 11.94 12.71
Total number of dowels 119 149 165 184
MINLP iteration Value of objective No. of active/all CPU time (s)
Max. deflection without 1.89 3.50 3.48 3.30
function (N) binary variables
joint flexibility (f0) (cm)
1 1. NLP 1764.2558 0/494 4.570 Max. deflection considering joint 2.85 4.85 5.16 5.10
2 1. MILP 1815.0570 44/494 0.310 flexibility (f = f0 + f1) (cm)
2. NLP 1842.1836 0.973 Ratio (f0 + f1)/f0 1.51 1.39 1.48 1.55
3 2. MILP 1849.7520 0.350
3. NLP 1884.5859 1.363
P
= 7.566

2300
2100
COSTS (EUR), MASS (kg)

Table 5 1900
Convergence of the Modified OA/ER algorithm (numerical example 3, n = 14). 1700
MINLP Value of objective No. of active/all CPU time (s) 1500 COSTS MASS
iteration function (N) binary variables 1300
1 1. NLP 1927.9943 0/650 3.648 1100
2 1. MILP 1960.4980 56/650 0.390 900
2. NLP 2001.3247 1.207
3 2. MILP 1974.0150 0.520
700
3. NLP 1987.0091 1.309 500
4 3. MILP 1989.2800 0.831 6 10 14 18
4. NLP 1989.2796 1.539 Topology (no. of upper chord members)
P
= 9.444
Fig. 5. The obtained costs and masses for trusses with different topologies.
 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294 293

6 design higher trusses (lower S/D ratios) with lower numbers of

intermediate members.
5
Displacement (cm)

However, the maximal height of a bearing structure is often

4 limited by architectural demands or other conditions. Considering
the remaining truss configurations (n = 10, 14 and 18) it can be ob-
3
served that the optimal heights (h, h0) do not vary significantly (see
2 Fig. 7). Generally, it is expected that when the height of a truss is
f0 f0+f1
approximately the same, increasing the number of intermediate
1 members would improve the structure’s stiffness, i.e. the deflec-
0
tions would be reduced. This assumption proves to be correct
6 10 14 18 when only the deformability of the truss members (bars), i.e. the
Topology (no. of upper chord members) deflection f0, is considered (see Fig. 6). When the flexibility of the
joints is additionally considered, however, the matter proves to
Fig. 6. The calculated displacements for trusses with different topologies.
be different. Fig. 6 shows, that the total deflection (f0 + f1) remains
approximately the same beyond some points. One can also see that
the absolute difference between the total deflection f0 + f1 and
500
deflection f0 increases proportionally with the number of flexibly
450 connected truss members (diagonals and verticals). Therefore it
400 is concluded that due to the flexibility of the joints the stiffness
350 of the truss can not be improved just by increasing the number
h, h0 (cm)

300 of intermediate bars. In order to achieve a better stiffness, the

250 joints would also have to be designed as very stiff, which would
h
200 of course increase the costs of the structure.
h0
150
Table 7 too, shows that the deflection resulting from the flexi-
bility of the joints (f1) represents a significant part of the total
100
50
0
6 10 14 18
Topology (no. of upper chord members) Table 8
Comparison of results.
Fig. 7. The optimal heights for trusses with different topologies. MINLP Present NLP Ref. [6] Improvement
result (%)
n=6 Costs (EUR) 1683.03 1865.35 9.77
The comparison of the deflections for different truss configura- Timber mass (kg) 700.29 766.86 8.68
tions (Fig. 6) shows that the deflections of the truss with the lowest
n = 10 Costs (EUR) 1842.18 2018.08 8.72
number of members (n = 6) are considerably smaller than the Timber mass (kg) 723.44 808.66 10.54
deflections of trusses with other configurations (n = 10, 14, 18).
n = 14 Costs (EUR) 1987.01 2239.66 11.28
This can be attributed to the considerably larger height of the truss Timber mass (kg) 771.95 906.76 14.87
with six upper chord members when compared to other configura-
tions. Therefore it can again be concluded that it is favourable to

Detail A

1 9 13
beff = b - t
heff = h - 2d

Aeff = beff ⋅ heff

2.0
13

3 .0
d = 1.4

12
8
6.

13
2 x 7.0

5x 12
4 .2

7.0
0

9
7.

h = 13
x
2

1 9 .8
39
0
9.8
9.

12
13

5.6 5.6
13

8.0 9 x 7.0 11.0 t = 0.8

b = 12
12
13 82

(a) Elevation (b) Section 1-1

Fig. 8. Example of a perfectly centric joint (all measures in cm).
294 S. Šilih et al. / Advances in Engineering Software 41 (2010) 286–294
deflection, as the ratio between the later and the deflection f0 amo-
uted between 1.4 and 1.55. In other words, the deflections due to
slips in connections represent over 40% of the total deflections.
The allowed displacement of 7.5 cm was not reached in all the
solutions, i.e. the ULS conditions were decisive at all optimal solu-
tions. Thus it can again be concluded that stress-controlled designs
are superior to deflection-controlled ones.
In order to estimate the obtained results, the optimal solutions
were compared to the results obtained in the previous research,
see Šilih et al. [6]. In the comparative reference the trusses were opti-
mized at a fixed shape (depth at mid-span h = 300 cm, slope of top
chord a = 10°). Also, the sizing optimization was performed by the
continuous NLP approach while the final results were obtained by
subsequent rounding of dimensions and re-analysis. A comparison
of the obtained results is presented in Table 8. The last column in Ta-
ble 8 represents the ratio between the difference in results (costs,
mass) and the results from [6] and thus shows the improvement ob-
tained by the presented MINLP optimization approach.
Data from Table 8 show that the presented MINLP optimization
approach yielded better results in all compared cases. The achieved
improvement in costs was between 9% and 11%, whilst the
improvement in mass was between 9% and 15%. The results thus
proved the superiority of the proposed discrete/continuous MINLP
shape and discrete sizing optimization when compared to the con-
tinuos NLP sizing optimization. It has also been shown that the
height of the truss and the slope of the top chord have a consider-
able influence on the final result.
As mentioned before, special care should be taken when the
connections are designed. The joints have to be perfectly centric
i.e. the appearance of additional bending moments due to eccen-
tricity has to be prevented. An example of such joint is shown in
Fig. 8 by considering the calculated required numbers of fasteners
in each connected element. The net cross-sectional area Anet of a
tensioned bar is also indicated.
6. Conclusions
The paper presents the shape and discrete sizing optimization
of metal-plate connected plane timber trusses by considering the
flexibility of the joints. The optimization was performed by the
Mixed-Integer Non-Linear Programming (MINLP) approach. The
main objective was to obtain a truss structure, which would be
optimal with respect to different design parameters such as span
to depth ratio, slope of top chord, etc. in a simultaneous optimiza-
tion process. The discrete sizing optimization enables the discrete/
standard values of sizing variables (CS dimensions, number of fas-
teners in each connection) to be calculated directly in the optimi-
zation process and that no re-analysis is required.
The optimization model for the cost optimization of timber
trusses was developed. An economic objective function was de-
fined in order to minimize the structure’s self-manufacturing costs,
subjected to the design, deflecion and stress (in)equality con-
straints. The finite element equations were as equality constraints
defined for the calculation of the internal forces and the deflections
of the structure. The stiffness matrix of the structure was com-
posed by considering fictiously decreased cross-sectional areas of
the flexibly connected elements. The design constraints were
determined in accordance with Eurocode 5 in order to satisfy both
the ultimate and the serviceability limit states.
The cross-sectional dimensions of the elements and the num-
bers of fasteners in the connections were defined as discrete/stan-
dard sizing variables, while the vertical coordinates of the joints
and the slope of the top chord were defined as continuous shape
variables. The discrete sizing variables were included through bin-
ary variables and the corresponding logical relations.
For the solution of the comprehesive mixed discrete/continuous
MINLP optimization problems the Modified Outer Approximation/
Equality Relaxation (OA/ER) algorithm was used. In order to reduce
the set of active binary variables in each optimization subphase a
special prescreening technique was additionally applied.
The efficiency and the the applicability of the proposed optimi-
zation approach is presented through a numerical example. The
obtained results proved the importance of consideration of joint
flexibility. Deflections due to slips in connections represent over
40% of the total deflections. In general it is preferable to design
higher timber trusses with a lower span/depth ratio and with a
smaller number of diagonal and vertical elements. By increasing
the number of elements the deflections cannot be reduced as effec-
tively as this is possible in the case of, e.g., steel trusses with rigid
connections. While, by adding flexibly connected intermediate ele-
ments, the deflections due to the deformability of the elements de-
crease, the deflections due to the flexibility of the joints increase
and the total deflections remain at the same level. The comparison
of the optimal results to previously obtained solutions has proved
the superiority of the proposed MINLP optimization method when
compared to the NLP approach.
References
[1] Schmidt LA. Structural design by systematic synthesis. In: Proceedings of 2nd
conference on electronic computations, NY: ASCE; 1960. p. 105–22.
[2] Dorn WS, Gomory RE, Greenberg HJ. Automatic design of optimal structures. J
Mécanique 1964;3:25–52.
[3] Kaveh A, Kalatjari V. Topology optimization of trusses using genetic algorithm,
force method and graph theory. Int J Numer Meth Eng 2003;58:771–91.
[4] Šilih S, Kravanja S, Bedenik BS. Shape optimization of plane trusses. In: Hendriks,
MAN, Rots JG, editors. Finite elements in civil engineering applications,
proceedings of the third DIANA world conference, Tokyo, 2002. p. 369–73.
[5] Kravanja S, Šilih S. Optimization based comparison between composite I
beams and composite trusses. J Construct Steel Res 2003;59:609–25.
[6] Šilih S, Premrov M, Kravanja S. Optimum design of plane timber trusses
considering joint flexibility. Eng Struct 2005;27:145–54.
[7] CEN/TC 250/SC5 N173. Eurocode 5: Design of Timber Structures, Part 1-1
General rules and rules for buildings, Final draft prEN 1995-1-1, Brussels,
European Comitee for Standardization; 2002.
[8] Duran MA, Grossmann IE. Mixed-integer nonlinear programming algorithm for
process synthesis. American institute of chemical engineers – annual meeting,
New York: AIChE; 1984. p. 23–30.
[9] Kravanja Z, Grossmann IE. New developments and capabilities in PROSYN – an
automated topology and parameter synthesizer. Comput Chem Eng
1994;18:1097–114.
[10] Duran MA, Grossmann IE. An outer-approximation algorithm for a class of
mixed-integer nonlinear programms. Math Program 1986;36:307–39.
[11] Kravanja S, Kravanja Z, Bedenik BS. The MINLP approach to structural
synthesis, Part III: synthesis of roller and sliding hydraulic steel gate
structures. Int J Numer Meth Eng 1998;43:329–64.
[12] Kravanja S, Šilih S, Kravanja Z. The multilevel MINLP optimization approach to
structural synthesis: the simultaneous topology, material, standard and
rounded dimension optimization. Int J Adv Eng Software 2005;36:568–83.
[13] Šilih S, Kravanja S. Synthesis of trusses using the MINLP optimization
approach. In: Hernandez S, Brebbia CA, El-Sayed MEM, editors. Computer
aided optimum design of structures IX. Southampton: WIT Press; 2005. p.
221–33.
[14] Kravanja S, Šilih S. The MINLP optimization of composite I-beams. In: Hernandes
S, Brebbia CA, editors. Sixth international conference on computer aided
optimum design of structures (OPTI 2001). Southampton: WIT Press; 2001. p.
401–7.
[15] Steck G. Fachwerbinde aus Brettschictholz un Vollholz, Holzbauwerke:
Bauteile Step 2, Konstruktionen, Details nach Eurocode 5. Düsseldorf:
Fachverlag Holz; 1995.
[16] Brooke A, Kendrick D, Meeraus A. GAMS – a user’s guide. Redwood City,
CA: Scientific Press; 1988.
[17] EN 1194. Timber structures – glued laminated timber – strength classes and
detemination of characteristic values, Brussels: European Comitee for
Standardization; 1999.
[18] Kravanja S, Soršak A, Kravanja Z. Efficient multilevel MINLP strategies for
solving large combinatorial problems in engineering. Opt Eng
2003;4:97–151.
[19] Kravanja S, Kravanja Z, Bedenik BS. The MINLP approach to structural
synthesis, Part I: A general view on simultaneous topology and parameter
optimization. Int J Numer Meth Eng 1998;43:263–92.
[20] Drud AS. CONOPT – a large-scale GRG code. ORSA J Comput 1994;6:207–16.
[21] CPLEX user notes, ILOG inc.