44

Rakenteiden Mekaniikka (Journal of Structural Mechanics)

Vol. 43, No 1, 2010, pp. 44-60
Advanced design and optimization of steel portal
frames
Petr Hradil, Matti Mielonen and Ludovic Flp
Summary Summary Summary Summary. This study brings new insights into the advantages of using more sophisticated
design methods for steel portal frames (e.g. geometrically and materially nonlinear analysis with
imperfections or GMNIA, and the general method introduced in Eurocode 3), compared to the
commonly used member checks with interaction formulae. The differences between the design
alternatives are discussed, focusing on assessing lateral-torsional stability, and the potential
benefits of using advanced shell model instead of widely used beam elements. In addition to the
advanced design methods, the topic of shape optimization of frames was explored using real-
coded genetic algorithm (GA).
The developed optimisation tools highlight the possibility of using GA in everyday design
in the future. The results of the study clearly point to the advantages of using advanced
modelling, e.g. GMNIA, instead of the classical member checks. While both methods are
accepted by the current steel design code EN 1993, using GMNIA can result in important
savings, because it eliminates some of the conservativeness brought in by the unavoidable
simplifications of the other methods. The experience shows that using complex 3D models is
possible with current computational capabilities.
Key words: portal frames, genetic algorithm, general method, lateral-torsional stability
Introduction
Portal frames for industrial buildings have been extensively studied because of their
widespread use. The improvement of the design methods for portal frames is one of the
recurring topics in the field of steel structures. Due to the large number of similar
framed structures, the desire to automate the design and manufacturing process was
popular from the very early stage [2]. As Dowling et al. [2] noted, there are two design
tendencies when trying to achieve more economical solutions: (a) to use compact hot-
rolled sections and exploit the advantages of plastic design and (b) to use slender built-
up sections with the most advantageous distribution of the material but keep the design
in the elastic range. The second option usually leads to slender structures, and therefore
stability becomes the main concern of the designer.
In case of side rails and purlins, the elastic design with very slender, Class 4 cross-
sections (i.e. cold-formed profiles) tends to be more economical solution compared to
the plastically designed hot-rolled continuous purlins. The better distribution of material
in cold-formed profiles is clearly offsetting the higher costs of fabrication. Moreover,
modern coating technologies also allow the use of material thicknesses in the range of
millimetres and below, without fears of corrosion.
45
Dowling et al. [2] predicted that an equivalent development towards slender
construction in main frame design will lead to similar economies to that achieved with
secondary elements. This conclusion has been supported by the experience in the US
[3], Canada and some European countries. However, the situation in Europe is not as
clear as in the US where the slender tapered solution almost completely replaced the
hot-rolled frames due to the mechanised fabrication of tapered elements. Several
European design documents [4], [5], [6], [7] including EN 1993 [1] are more focused on
plastic design of frames. In fact, both solutions coexist on the market in many countries.
The economical efficiency of the two solutions depends on several related factors,
such as fabrication cost, transport cost, labour cost. Dowling et al. [2] reported weight
saving in range of 30% in favour of slender tapered frames. However, it was debated
what this means in cost saving [8]. Other aspects of using the welded tapered solutions
were also discussed, with important focus on the question of lateral stability of beams
and columns [8].
The present study proposes to update the discussion on optimal frame design: (1) by
using design methods for stability from the EN 1993 [1], (2) by implementing advanced
analysis methods, which are not currently employed by design offices but have potential
for replacing current tools, and (3) by introducing optimization tools based on Genetic
Algorithms (GAs) to find optimised geometries.
Design methods
The study includes several methods for portal frames design to resist loads in
fundamental and seismic situations, using the limit states conditions from the EN 1993
[1] and EN 1998 [9]. To implement those design methods in optimization, Abaqus and
Excel based plug-ins were developed. Both plug-ins allow automatic generation of
computational models and automatic result evaluation. The results of the design checks
are expressed as a minimum load amplifier of the design loads to reach the ultimate
(ULS) or serviceability (SLS) limit state criteria. The vertical serviceability limit
corresponds to an apex deflection of span/200, whereas the horizontal serviceability
limit is height/100; derived in a simplified way from EN 1998 [9].

Method 1: Geometrically and materially nonlinear analysis on imperfect
structure (GMNIA)
The most computationally expensive method is the straightforward numerical
calculation of 3D shell model of the frame (Figure 1). No special checks for out-of-
plane stability are needed in such method, because the calculation is materially and
geometrically nonlinear and is taking into account appropriate initial imperfections.
Bow imperfections are inserted from a preliminary eigenvalue analysis, using the first
positive buckling shape of the model, scaled to the EN 1993 recommended amplitude
(i.e. using Table 5.1 from EN 1993). Sway imperfections are also applied to the model
according to EN 1993 rules. The material was modelled using a bilinear stress-strain
relationship, including a moderate strain hardening calculated for each used steel grade.

46

Figure 1 Global nonlinear analysis calculation steps: Undeformed model (left), First buckling
shape (middle), Deformation during nonlinear analysis with increasing vertical loads (right -
scale: 20)
Method 2: General method
This method takes into account out-of-plane stability with a global reduction factor.
According to 6.3.4 of EN 1993 [1], the resistance of the frame is checked using the
following condition:
0 . 1
1
,

M
k ult op

,
(1)
where
ult,k
is the minimum load amplifier of the design loads to reach the characteristic
resistance of the most critical cross-section without out-of-plane buckling effect. This
value is obtained from the nonlinear analysis with gradually increasing loads of a
simple 2D beam model of the frame. However, even the 2D model is materially and
geometrically non-linear and it includes in-plane bow and sway imperfections (but not
out-of-plane sway imperfections).
The out-of-plane buckling reduction factor
op
originates from the critical amplifier

cr,op
of the design loads to reach the elastic critical resistance with regards to lateral or
lateral torsional buckling. Because of the particularly difficult analytical expression of
critical load of frames which are made from eccentrically supported members with
variable cross-sections, the method is using the 3D shell model of the frame for the
evaluation of the critical amplifier.
Method 2 has the advantage of faster numerical calculation compared to Method 1.
The computational advantage results from the use of the simplified 2D beam model for
the nonlinear part of the analysis, and the use of the 3D shell model only for the
evaluation of the critical loads for out of plane buckling.

Method 3: EC3 interaction formulae
Linear structural analysis together with design using cross-sectional checks to express
limit states conditions are the commonly used methods well described in EN 1993 [1].
However, not all of the EN 1993 rules are applicable to elements with variable cross-
section and eccentric lateral restraints. Therefore, it was necessary to implement other
theories in the calculation. Even though the method is the most computationally
47
effective, it is more conservative, especially in its out-of-plane stability approach where
no lateral supports of compressed flanges are used by default.
The Method 3 uses the following modifications of the calculation described in EN
1993 [1]:

In-plane critical amplifier
While the critical load was a result of linear eigenvalue analysis in Method 2, the
analytical approach is adapted in this case. The calculations take into account the
presence of the axial force in the rafters and utilize the formulas proposed by Davies
[10].

Lateral torsional buckling of eccentrically restrained tapered member
The critical moment in our calculation is the smaller value of critical moment of
restrained member between fork supports according to the Appendix F of the SCI
Technical Report [7], and elastic critical moment of unrestrained part of member
between purlins or side rails.
(
(

|
|

\
|
=
min ,
2
min ,
2
min ,
min ,
2
min ,
2
1 0 2
,
1
min
z
t
z
w z
cr
t
cr
I E
I G a
I
I
a
I E
C M
c m
M

, (2)
TFB cr
z
cr
N
e
i
M
,
2
0
0
2
|
|

\
|
= .
(3)
Here m
t
is equivalent uniform moment factor adapted from [7], I
y,min
, I
z,min
, and I
w,min
are
sectional properties of the shallow end, a is the distance between lateral supports
(purlins, side rails), N
cr,TFB
stands for critical axial load of torsional flexural buckling of
eccentrically restrained member, C
1
is the moment gradient factor, e
z
is the distance
from shear centre to lateral support either at shallow end or at deep end of tapered beam.
When the relative slenderness is higher than 1, equivalent section factor c is also 1
and the minimum M
cr,0
was used, which comes from the deepest end. In all other cases
shallow end properties and following expression for equivalent section factor are used:
( ) L L c c
h
1 1
0
+ = , (4)
where L
h
and L are the lengths of haunch and the whole member respectively and the
basic equivalent section factor c
0
was adapted from [7].

Major axis flexural buckling of tapered member
It is very conservative to apply the standard formula for calculation of the critical load
of the tapered member using the shallow ends sectional properties. A more accurate
option is the approach proposed by apalas [11] used in Method 3, where critical loads
are based on the deep end cross-sectional properties and reduced by
n
factor.
48
Analytical results
To demonstrate the differences between the three design methods, the results of
calculation of one particular frame are presented in this chapter. A welded-tapered
frame with pinned supports was calculated with the parameters presented in Figure 2
and Table 1. The loads from the most critical fundamental combination of dead weight
and snow was redistributed into a set of concentrated forces at the purlin/rail connecting
points. The factored value of distributed load used for the ULS and SLS calculations
checks was 1638 N/m
2
and 1130 N/m
2
, respectively.


Figure 2. Geometry of the selected welded-tapered frame symmetrical part
Table 1. Cross-sections, material and loading of the selected frame

Flange thickness 10 mm Web thickness 8 mm
Flange width 260 mm Haunch length 3.6 m
Avg. purlin spacing 1.25 m Avg. side rail spacing 1.38 m
Purlins/rails eccentricity 120 mm Steel quality S275
Char. dead load 380 N/m
2
Char. snow load 750 N/m
2

Dead load safety factor 1.35 Snow load safety factor 1.5

The frame satisfies the ultimate and serviceability limit states conditions with all
three design methods, but with different levels of conservativeness. (Table 2). The
required computing times for carrying out the check, are also given in Table 2.
49
Table 2 Calculated load-bearing capacity of the frame at ultimate (ULS) and serviceability
(SLS) limit states and the calculation speed (tested on Pentium 4, 3 GHz with 1 GB RAM)

Method 1 Method 2 Method 3
ULS Capacity 2800 N/m
2
2272 N/m
2
1761 N/m
2

SLS Capacity 3088 N/m
2
3359 N/m
2
3410 N/m
2

Calculation speed 284 sec. 125 sec. 2 sec.

Figure 3 represents load-displacement curves produced by different methods. In case
of Method 1, the resistance of the frame can be estimated directly from the curve.
Because the model includes both in-plane and out-of-plane imperfections, and the
analysis method is materially and geometrically nonlinear, the effect of buckling is
implicitly taken into account. Therefore there is no need to affect the resistance by
buckling factors, while safety factors for material and loads are included in the model.
One question, when interpreting the curve, is what to consider on the curve as resistance
limit. In our case, the resistance limit is considered the reaching of the first yield,
because the frame is slender. However, this might supply conservative results if the
frame is made of compact members or if a stress concentration causes local yielding. In
these cases the resistance can be interpreted according to methods recommended by
ECCS [12].
In case of Method 2, the curve includes only in-plane imperfections, both sway and
bow, and the analysis is materially and geometrically nonlinear. The value of
ult,k
to be
used in equation 1, is estimated from this curve as corresponding to the point where
plastic strain reaches five times the yield strain in any cross-section. However,
ult,k
is
further reduced due to out of plane buckling with an extra reduction factor (
op
in
equation 1) calculated from the overall slenderness. The straight line corresponding to
Method 3 is the result of a linear analysis which includes sway imperfections.

1638
1130
0
1000
2000
3000
4000
5000
6000
0 50 100 150 200 250 300
apex vertical displacement (mm)
v
e
r
t
i
c
a
l

l
o
a
d

(
N
/
m
2
)
Method 1
Method 2
Method 3
SLS limit (100 mm)
design load
characteristic load


Figure 3. Load-displacement curves; comparison of different methods: Method 1 (solid line)
including all buckling and imperfections, Method 2 (dashed) including only in-plane buckling
and imperfections, Method 3 (dash-dot) including only initial sway imperfections
50
Optimization method
The steel portal frame optimization problem is formulated below:
{ }
{ }
Hh H B tf tw h
b h c b
SLS
ULS
w
x x x x x x x
x x x x x
x g
x g
x f
, , , , ,
or , , ,
, 0 ) (
, 0 ) ( subject to
) ( Minimise

(5)
Here, ) (x f
w
refers to optimization searching for a minimum weight. The possible
solutions are checked against two main constraints:
ULS
g , where ULS refers to ultimate
limit state check and
SLS
g , where SLS refers to serviceability limit state check. The
variable vector x comprises beam profile
b
x , column profile
c
x and haunch ratio
h
x variables, if the frame is assembled of hot-rolled sections. When welded-tapered
sections are optimized, number of possible variables increases. The variables available
for optimization are profile flange thickness
tf
x , profile web thickness
tw
x , profile
width
B
x , profile height
H
x and height of the profiles at the haunch
Hh
x . Theoretically,
these variables could be selected separately for the beam and column, however, it is
reasonable to limit the number of different plate sizes. Therefore, only height of the
lower end of profiles can be different in the beam and column, which leaves six
variables for optimization (
tf
x ,
tw
x ,
B
x ,
Hc
x ,
Hb
x ,
Hh
x ). Each variable has boundaries
coming from physical restrictions (e.g. haunch ratio 4-11) and readily available parts
(e.g. common plate sizes 4, 5, 6, 8, 10, 12, ).
The problem is both nonlinear and discrete, which causes difficulties for classical
direct and gradient-based optimization methods. On the other hand, genetic algorithms
(GAs) are successfully applied in the field of structural optimization [13], [14], [15] and
[16]. GAs have several advantages, such as possibility of parallel computing, easy
handling of multiple variables, and straightforward coding practice. Genetic algorithm is
a procedure which tries to mimic the natural evolution process. After an initial
population is created and analysed, the fitness of each individual is evaluated. Then a
new population is created by favouring the fittest individuals and by combining the
properties of the population members using genetic operators, such as crossover and
mutation. GA proceeds iteratively towards the optimal solution by creating a new
population using the properties of the previous one. Elitism ensures that the best
solution is kept during all of the genetic operations, and while no proof of convergence
in finite time exists, good results are found in a reasonable time.
A real-coded genetic algorithm (RCGA) can handle discrete and real variable types
easily and has been chosen for this problem. With RCGA, the coding-decoding
characterizing binary-coded GAs is also avoided.
Optimization literature provides a large catalogue of different selection, crossover
and mutation operators that can be combined to create a GA suited for the problem at
hand. In this paper, the well known simulated binary crossover (SBX) [17] and
parameter based polynomial mutation operator [13] are utilized. The crossover operator
has a self-adapting behaviour, which favours creating children near to parents, when the
parents are near to each other in the variable space. The basic behaviour of the genetic
51
algorithm is enhanced by two methods developed to improve the steel portal frame
optimization:
- a local search method is creating individuals very similar to the current elite
individual found, which ensures that the local optimum is found with very high
probability;
- the diversity of the population is maintained by using so called diversifying operator,
which introduces new genetic material to the population preventing premature
convergence to the local optimum [16].
In terms of optimization objective function, it has been opted to concentrate on
easily measurable performance parameters, such as weight, with the flexibility to
expand results to more financially oriented targets (e.g. price).
The algorithm can be used with all of the design methods and the flowchart of the
optimization combined with design methods is presented in Figure 4.



Figure 4 The basic diagram of two optimization tools developed at VTT.
AP-Frame contains the Abaqus based design methods and optimization, whereas
EV-Frame refers to the design and optimization developed in a Microsoft Excel
workbook. In both cases, the GA runs for predetermined number of generations, and the
best configuration found is given as an output. Figure 5 presents five example
optimisation runs of the same frame configuration, which shows how the weight of the
elite individual typically decreases during the genetic algorithm optimisation. In this
example, eight variables were optimized in a welded-tapered frame with population size
20 and maximum 50 generations.

52
2
2.5
3
3.5
4
4.5
5
0 10 20 30 40 50
F
r
a
m
e

m
a
s
s
[
t
o
n
s
]
Generation
1
2
3
4
5


Figure 5. Elite value fitness development in five optimisations of the same frame configuration
with the genetic algorithm.
Typically, the weight of the elite frame in the population decreases sharply during
the first generations, and then the search focuses around the elite individual refining the
solution. Sudden drop in fitness occurs usually when the algorithm finds a better
alternative shape or size of the frame elements, meaning that many variables can change
drastically.
Benchmarking the design and optimisation methods
In order to calibrate the tools, the results of the optimization study by Horridge et al.
[18] have been replicated. A couple of important points in this study must be taken into
consideration in order to understand the extent of calibration presented here:
- The configurations considered were based on the Universal Beam (UB) range of
class 1 profiles allowing for plastic analysis. In line with British practice, first order
plastic analysis was used for the design (i.e. no imperfections, no second order
effects and no buckling were considered).
- The frames were considered pinned. Therefore the structures cannot develop a plastic
mechanism with several plastic hinges.
- The eaves height was increasing together with the span and it should be noted that
the steel consumption (Figure 8) was considerably affected by this fact.
- No serviceability checks were reported in the original study [18].
- The results are not based on any formal optimisation method, rather it was attempted
to find optimal configurations based on engineering experience. The roof angle and
the haunch length were fixed in advance.
The cases reported by Horridge et al. were reproduced using the EV-frame and AP-
Frame tools. It was impossible to fully recover all original design assumptions because
53
some incompatibilities exist between the old British codes and the Eurocode methods.
Therefore, a few equivalent assumptions had to be used. Namely:
- The AP-Frame and EV-Frame tools were forced to neglect lateral torsional buckling.
This was achieved by setting buckling reduction factors to 1, and all out-of-plane
imperfections to 0.
- The safety factors of loading were modified to match those used by Horridge et al.
[18], i.e. = 1.7 for all loads.
- The basic wind speed of 46 m/s (3 seconds average) reported to be used with the
British code, was replaced with 15.5 m/s (10 minutes average) wind speed
compatible with EN 1991 and producing the same total horizontal load. It should be
noted that the factors controlling the wind pressure on the roof are also different in
the two codes, but this difference has not been eliminated.
- The steel grade 43 reported by Horridge et al. [18] has been replaced by S275, with
the same yield stress.
- Instead of the classic plastic method, allowing of the successive forming of plastic
hinges, the calculation uses linear and nonlinear elastic analysis with plastic sectional
properties, as described by EN 1993 [1]. This method could be adopted because
frames are pinned and they cannot create successive plastic hinges.



Figure 6. Geometry and loading of hot-rolled frames
The optimisation used by Horridge et al. [18] was emulated with the above
assumptions, and the original configurations were re-calculated using EV-Frame and
AP-Frame tools. These configurations had a ULS utilization factor in the range of 0.93-
1.22 with EV-Frame and 0.88-1.13 with AP-Frame (Figure 7). It can be concluded that
both software tools were able to recreate the initial set with good accuracy. EV-Frame is
slightly more conservative, predicting an average utilization factor of 1.06, suggesting
that the frame proposed by Horridge et al. [18] would fail using linear elastic analysis
with plastic sectional properties. AP-Frame is less conservative with nonlinear elastic
analysis based on equivalent plastic strain check to identify the plastic hinge. The
average utilization factor was 0.98 and the method predicted problems with only few of
original configurations.
54
0
1
1
4
.
4
x
5
_
4
.
5
1
4
.
4
x
5
_
6
1
4
.
4
x
5
_
7
.
5
1
4
.
4
x
5
_
9
1
8
x
5
.
2
5
_
4
.
5
1
8
x
5
.
2
5
_
6
1
8
x
5
.
2
5
_
7
.
5
1
8
x
5
.
2
5
_
9
2
1
.
6
x
5
.
5
_
4
.
5
2
1
.
6
x
5
.
5
_
6
2
1
.
6
x
5
.
5
_
7
.
5
2
1
.
6
x
5
.
5
_
9
2
5
.
2
x
5
.
7
5
_
4
.
5
2
5
.
2
x
5
.
7
5
_
6
2
5
.
2
x
5
.
7
5
_
7
.
5
2
5
.
2
x
5
.
7
5
_
9
2
8
.
8
x
6
_
4
.
5
2
8
.
8
x
6
_
6
2
8
.
8
x
6
_
7
.
5
2
8
.
8
x
6
_
9
3
2
.
4
x
6
.
2
5
_
4
.
5
3
2
.
4
x
6
.
2
5
_
6
3
2
.
4
x
6
.
2
5
_
7
.
5
3
2
.
4
x
6
.
2
5
_
9
3
6
x
6
.
5
_
4
.
5
3
6
x
6
.
5
_
6
3
6
x
6
.
5
_
7
.
5
3
6
x
6
.
5
_
9
3
9
.
6
x
6
.
7
5
_
4
.
5
3
9
.
6
x
6
.
7
5
_
6
3
9
.
6
x
6
.
7
5
_
7
.
5
3
9
.
6
x
6
.
7
5
_
9
4
3
.
2
x
7
_
4
.
5
4
3
.
2
x
7
_
6
4
3
.
2
x
7
_
7
.
5
4
3
.
2
x
7
_
9
Frame span x Eaves height x Centers distance
U
L
S

-

U
t
i
l
i
z
a
t
i
o
n

f
a
c
t
o
r
Re-calculated with EV-Frame Re-calculated with AP-Frame Optimised with EV-Frame

Figure 7. Recalculated and optimized cases utilization of ULS checks
The set of frame configurations used by Horridge et al. [18] have been optimized
using the EV-Frame, with exactly the same constraints (Figure 7) using beam and
column profiles as variables. While the original optimisation proved to be very good,
we found that marginal improvements are possible even in this very rigorously studied
set of frames.
For instance, the EV-Frame optimized frames had a slightly different balance
between beam and column compared to the original frames. On average, plastic
modulus of beams increased by 10% while plastic modulus of columns decreased by
7%, suggesting that weaker beams and stronger columns, compared to the usage of
Horrige et al. [18], are the optimal configuration.
In Figure 8, the steel consumptions are given for the frame distance 6 m, also
considered in the original study. These consumptions were calculated using a single
frame mass without any additional steel elements e.g. bracings, purlins, longitudinal
beams etc. It can be observed that the original configurations by Horridge et al. [18] and
the optimized ones by EV-Frame and AP-Frame have very similar weight.

5
10
15
20
25
30
35
14.4 18.0 21.6 25.2 28.8 32.4 36.0 39.6 43.2
Span (m)
S
t
e
e
l

c
o
n
s
u
m
p
t
i
o
n

o
f

a

f
r
a
m
e

(
k
g
/
m

2
)
Horridge (hot-rolled)
AP-Frame (hot-rolled)
EV-Frame (hot-rolled)
EV-Frame (welded-tapered)
Optimization methods
H
e
i
g
h
t

5
.
0
0

m
H
e
i
g
h
t

5
.
2
5

m
H
e
i
g
h
t

5
.
5
0

m
H
e
i
g
h
t

5
.
7
5

m
H
e
i
g
h
t

6
.
0
0

m
H
e
i
g
h
t

6
.
2
5

m
H
e
i
g
h
t

6
.
5
0

m
H
e
i
g
h
t

6
.
7
5

m
H
e
i
g
h
t

7
.
0
0

m
hot-rolled frames
(UB profiles)
welded-tapered
frames

Figure 8. Comparison of different design and optimisation methods
55
Optimization of welded- tapered frames
The optimization of hot-rolled frames, mentioned in previous chapter, was based on
two variables with limited amount of possible combinations. The result of this simple
task, where haunch ratio and roof pitch are fixed, could be found easily using other
methods, including calculation of all 5184 possibilities. The most important feature of
our optimization tools is the ability to find the lowest mass of welded-tapered frames
where millions of combinations are possible.




Figure 9. The geometry and loading of the studied welded-tapered frames
The first study based on our optimization tools aims to demonstrate the relation
between frame mass (or steel consumption per m
2
) and frame span (Figure 10) with
different snow load and eaves height. The variables optimized are haunch ratio, flange
thickness, web thickness, height of the profiles at the eaves, width of the profiles, height
of the beam at the top and height of the column at the base. Assuming that there is no
effective lateral support of compressed flanges, boundary conditions exists only at the
external flanges in 3D shell model. Also the wind load is considered as a secondary
variable action creating pressure on upwind side and uplift on downwind side according
to EN 1991[19] (Figure 9).

56
1
2
3
4
5
6
7
8
16 18 20 22 24 26 28 30 32
Span (m)
F
r
a
m
e

m
a
s
s

(
t
o
n
s
)
6 m, Snow 750 N/m2
6 m, Snow 1500 N/m2
6 m, Snow 2000 N/m2
8 m, Snow 750 N/m2
8 m, Snow 1500 N/m2
8 m, Snow 2000 N/m2
5
10
15
20
25
30
35
40
45
50
16 18 20 22 24 26 28 30 32
Span (m)
S
t
e
e
l

c
o
n
s
u
m
p
t
i
o
n

o
f

a

f
r
a
m
e

(
k
g
/
m
2
)
6 m, Snow 750 N/m2
6 m, Snow 1500 N/m2
6 m, Snow 2000 N/m2
8 m, Snow 750 N/m2
8 m, Snow 1500 N/m2
8 m, Snow 2000 N/m2


Figure 10. The effect of different snow loads on welded-tapered frames 6 m high (solid line)
and 8 m high (dashed). The load is represented as a characteristic value on flat roof (e.g. 2000
N/m
2
corresponds to basic snow load 2500 N/m
2
on the ground)
The lateral buckling of slender cross-sections is significantly reducing the load-
carrying capacity of the frame, and it is common practice to use lateral restraints
connected to the inner compressed flanges to suppress this negative effect. The lateral
support is usually considered to be strong enough to completely suppress out-of-plane
buckling.
However, the most frequently used type of lateral restraint, the diagonal stay,
transfers lateral loads by bending of purlins, and tends to be less effective when cold-
formed purlins are used. In order to evaluate the behaviour of such frames, the simple
2D beam model is not sufficient. In the following study, 3D model in AP-Frame
optimization tool expands by added purlins and lateral restraints (Figure 11). The
purlins are modelled as beam elements with 2 mm thick Z150 cross-section, and lateral
stays are simplified as rigid truss elements. The diagonal stays exist only in the corner
region of the frame supporting the whole length of the haunch and one third of the
column. It has been tested that adding more supports has no further effect on lateral
stability of the frame.
57

Figure 11. The lateral torsional buckling failure of 6 m high frame with 32 m span (5x scaled)
considering support of purlins.
The calculation covers 6 m high frames with basic snow load 2500 N/m
2
and basic
wind velocity 30 m/s according to EN 1991 [19]. The weight of welded-tapered frames
without lateral supports as well as fully restrained frames is optimized for comparison
(Figure 12). As it can be observed, the lateral supports in smaller frames restrains the
out-of-plane buckling very well while the effect of those supports rapidly decreases with
increasing span. This indicates that light gauge steel purlins can be relied upon for
stabilizing smaller span frames, but they are not effective when the span increases.
The graphs can also give an indication to the designer whether providing diagonal
stays is economical. Implementing diagonal stays in the design of structure brings an
extra expense: it requires connecting plates to be welded on the main frame and it very
often disrupts the internal sheeting. The gain in terms of material consumption on the
frame might not offset the expenses. In any case, the lateral stabilization of the frame
has to be modelled using 3D model with proper support stiffness to correctly account
for their effect.

58
1
2
3
4
5
6
7
8
16 18 20 22 24 26 28 30 32
Span (m)
F
r
a
m
e

m
a
s
s

(
t
o
n
s
)
EV-Frame without restraints
EV-Frame restrained
AP-Frame without restraints
AP-Frame restrained
AP-Frame with 3D lateral supports
5
10
15
20
25
30
35
40
45
50
16 18 20 22 24 26 28 30 32
Span (m)
S
t
e
e
l

c
o
n
s
u
m
p
t
i
o
n

o
f

a

f
r
a
m
e

(
k
g
/
m
2
)
EV-Frame without restraints
EV-Frame restrained
AP-Frame without restraints
AP-Frame restrained
AP-Frame with 3D lateral supports

Figure 12. The effect of out-of-plane buckling studied on 6 m high welded-tapered frames with
purlins and diagonal stays included in the model (solid fat line) compared to models without any
lateral restraints (solid thin line) and fully restrained (dashed line) using different design
methods.
Conclusions
The advanced 3D modelling methods explored in this paper are very effective, and offer
the designer the ability to model complex structural configurations and support
conditions. Because the simplified methods described in the design codes for special
configurations are naturally conservative, the use of advanced options can bring
substantial economical benefits.
On the other hand, due to advances in computing capabilities (both software and
hardware), the use of these 3D shell models is not any more out of reach of an average
design office. The calculation time in the range of a few minutes with the AP-frame tool
is still long compared to the runtime in range of few seconds with beam-based analysis
software, but both of them are much shorter compared to the time spent on creating the
model. If the model preparation is automated (like in case of AP-Frame) the advanced
tool is very competitive for single analysis.
In terms of optimization, the use of EV-Frame offers instant solution. It is clear that
similar tools should be, and for some cases are, used by the design offices. The GA
offers a strong and versatile option for optimization in the field of structural
engineering.
The optimization time of advanced modelling AP-Frame tool is still in the range of
many hours when used on a standard computer. However, the combination of 3D
modelling and GAs can take the advantage of parallel computing and is especially
suitable for server applications.
59
Concerning the specific results presented in this paper, it has been shown that using
slender welded-tapered frames instead of hot-rolled sections leads to decreased steel
consumption. Cost savings could be achieved by implementing modern fabrication
technologies. However, slender frames have to be properly designed especially
considering their lateral stability. In order to effectively design the lateral restraints,
calculation of 3D model is needed. The usual diagonal stay configuration of lateral
restraint, when it relies on light-gauge steel purlins, is not effective in preventing the
lateral buckling of the frame, especially in larger span frames. Further, implementing
shape optimization into the design process provides economical solution tailored to the
specific loading situation.
Acknowledgements
The material contained in this paper is based on the work supported by the European
Communitys RFS-PR-06054 project PRECASTEEL. This support is gratefully
acknowledged. All opinions, findings, conclusions or recommendations expressed in
this material are those of the writers. The European Community is not liable for any use
of the information contained in this paper.
References
[1] EN 1993:2005, Eurocode 3: Design of steel structures
[2] Dowling P.J., Mears T., Owens G., Raven K., A development in the automated
design and fabrication of portal framed industrial buildings, The Structural
Engineer, vol. 60A, Oct. 1982, pp. 311-319.
[3] Lee G.C., Ketter R.L., and Hsu T., The design of single storey rigid frames,
Cleveland, Ohio: Metal Building Manufacturers Association, 1981.
[4] Practical analysis of single-storey portal frames, European Convention for
Constructional Steelwork, 1991.
[5] King C. M.. Plastic design of single-storey pitched-roof portal frames to
Eurocode 3. Technical report. P147. UK: The Steel Construction Institute; 1995.
[6] Salter P., Malik A., King C., Design of Single-Span Steel Portal Frames to BS
5950-1:2000, Great Britain: The Steel Construction Institute, 2004.
[7] King, C. M., SCI P164 Technical Report: Design of steel portal frames for
Europe, The Steel Construction Institute, 2001
[8] Dowling P.J., Mears T., Owens G., Raven K., Discussion on: A development in
the automated design and fabrication of portal framed industrial buildings, The
Structural Engineer, vol. 61A, Dec. 1983, pp. 383-392.
[9] EN 1998-1:2004, Eurocode 8: Design of structures for earthquake resistance
[10] Davies J. M., Inplane stability in portal frames, The Structural Engineer, Vol 68,
No. 8, April 1990.
[11] apalas, V., Samofalov, M., Viaeslavas, , FEM stability analysis of tapered
beam-columns, Journal of Civil Engineering and Management, 2005, Vol XI., No.
3, 211-216.
60
[12] Recommended Testing Procedure for Assessing the Behaviour of Structural Steel
Elements under Cyclic Loads, Publication No. 45, ECCS Technical Committee 1 -
Structural Safety and Loadings, 1986.
[13] Deb, K. & Gulati, S, Design of truss-structures for minimum weight using genetic
algorithms, Finite Elements in Analysis and Design, 2001, vol. 37, no. 5, pp. 447-
465.
[14] Jalkanen, J. Tubular truss optimisation using heuristic algorithms, 2007, Doctoral
Dissertation, Tampere University of Technology.
[15] Saka, M.P., Optimum design of pitched roof steel frames with haunched rafters by
genetic algorithm, Computers & Structures, 2003, vol. 81, no. 18-19, pp. 1967-
1978.
[16] Mielonen, M-P. Steel portal frame optimization using genetic algorithms, 2010,
Masters Thesis, Aalto University School of Science and Technology
[17] Deb, K. & Agrawal, R.B., Simulated binary crossover for continuous search
space, Complex Systems, 1995, no. 9, pp. 115-148.
[18] Horridge, J.F., Morris, L.J., Comparative costs of single-storey framed structures,
The Structural Engineer, Vol. 64A, No. 7, July 1986
[19] EN 1991, Eurocode 1: Actions on structures


Petr Hradil, Matti Mielonen, Ludovic Flp
VTT Technical Research Centre of Finland
P.O.Box 1000
FIN-02044 VTT
FINLAND
petr.hradil@vtt.fi, matti.mielonen@vtt.fi, ludovic.fulop@vtt.fi