Vous êtes sur la page 1sur 15

Engineering Structures 45 (2012) 177–191

Contents lists available at SciVerse ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Genetic algorithm optimization of composite and steel endplate semi-rigid joints


Fernando Busato Ramires a, Sebastião Arthur Lopes de Andrade a,
Pedro Colmar Gonçalves da Silva Vellasco b,⇑, Luciano Rodrigues Ornelas de Lima b
a
Civil Engineering Department, Pontifical Catholic University of Rio de Janeiro, PUC-Rio, Brazil
b
Structural Engineering Department, State University of Rio de Janeiro, UERJ, Brazil

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

Article history: The main goal of this paper is to describe a process developed to reduce the cost of composite and steel
Received 6 November 2009 semi-rigid endplate joints. The developed process is based on an adequate selection of the most signifi-
Revised 25 May 2012 cant variables that controls the joints structural performance to obtain effective cost solutions. This pro-
Accepted 27 May 2012
cedure was based on the use of a genetic algorithms producing trustable results in terms of economy and
Available online 25 July 2012
efficiency, with a reduced computational effort. Four set of examples involving flush and extended end-
plate joints are presented to illustrate the developed optimization procedure. These joints were initially
Keywords:
considered to be made only with steel members. In a subsequent phase equivalent composite joints were
Steel structures
Composite and steel semi-rigid joints
also considered to enable a direct comparison of the these alternative typologies. The global costs asso-
Composite construction ciated to the adopted joints were evaluated taking into account parameters like: endplate dimensions,
Genetic algorithms number of adopted bolts, bolt diameter among others. The performed examples validated and proved
Structural optimization the viability of using GA to optimize composite and steel semi-rigid joints.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction strength bolts. Header plate joints were initially investigated by


Sommer [7], that later performed four tests using bolted angle
One of the most desirable aims of the Engineering Science is the joints. Extended and flush endplate joints tests are also easily
development of cost effective structural elements usually achieved found in the literature like: Ostrander [8], Bailey [9] and Surtees
with the implementation of new design philosophies and proce- and Mann [10], Packer and Morris [11] and Johnson and Walpole
dures. A typical example of these new trends can be identified in [12].
the composite and steel portal frame design that was traditionally From the eighties, various semi-rigid tests were performed by
based on rigid or pinned beam to column joint assumptions. De- authors like: Jaspart [13], Cruz et al. [14] on extended endplate,
spite this fact, it is largely recognized that the majority of the struc- web cleats, and seat angle semi-rigid joints. Cruz et al. [15,16] re-
tural joints do not present this behaviour. Numerous investigations viewed several experimental investigations highlighting the stud-
were conducted over the last few years to assess the actual re- ies of: Brozetti on flush and extended endplate joint tests;
sponse of semi-rigid joints and led to various formulations that en- Zoetemeijer on additional flush endplate joint tests; Klein, Braun
able the evaluation of the joint main characteristics in terms of and Ellmerer on internal and external welded joints tests and final-
stiffness, strength and rotation capacity like the recommendations ly by Humer on internal endplate joint tests using stiffeners
present in the current Eurocodes 3 and 4 [1,2]. welded to the column flange and web. Azizinamini and co-workers
Since the first studies that evaluated the rotational stiffness of [17,18] conducted an experimental investigation on beam to col-
riveted beam to column joints performed by Wilson and Moore umn joints using top, bottom and web angles subjected to static
in 1917 [3], various tests have been conducted to determine the ac- loads.
tual response of beam to column joints. Riveted and welded joints From the nineties, tests performed on structural joints by Brazil-
using angles were tested by Young and Jackson [4], by Rathbun [5], ian authors are also reported. Queiroz [19] conducted, in Innsbruck,
while Bell et al. [6], performed tests on similar joints using high a series of tests using welded joints. Carvalho et al. [20], reported an
experimental investigation on beam to column joints using top,
bottom and web angles on the column major axis while Lima et
⇑ Corresponding author. Address: Structural Engineering Department, State al. [21,22] performed similar tests on the column minor axis. Ribe-
University of Rio de Janeiro, UERJ, rua São Francisco Xavier 524, s.5018-b, Maracanã, iro [23] performed a series of endplate joint tests in order to deter-
Rio de Janeiro, RJ, 20550-900, Brazil. Tel./fax: +55 21 25877537.
mine the influence of parameters like the endplate thickness and
E-mail addresses: ramires@aluno.puc-rio.br (F.B. Ramires), andrade@civ.puc-rio.br
(S.A.L.de Andrade), vellasco@eng.uerj.br (P.C.G. da Silva Vellasco), luciano@eng.uerj.br
the bolt diameter on the semi-rigid joint structural response.
(L.R.O. de Lima). Various other investigations, not only reporting on experiments

0141-0296/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.engstruct.2012.05.051
178 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

Fig. 1. Beam to column joint mechanical model.

but also on numerical and analytical studies can be easily found in The GA technique has been extensively used over the last few years
the literature [13,15,16,24]. in optimization processes, not only in engineering related applica-
Despite the significant knowledge growth the composite semi- tions but also in cases where it is aimed to simplify, or even orga-
rigid design is still facing some resistance from structural engineers. nize, complex work routines. Focusing on the objective of
This can be partly explained by the fact that few design standards minimizing the effort spent in the structural design, the present
allow or even encourage the use of this design philosophy. On the paper presents an automatic optimization procedure implemented
other hand, the semi-rigid design procedure involves repetitive to determine the optimum joint design. This optimum joint has the
and exhaustive tasks due to the fact that numerous geometric required flexural strength and initial stiffness aiming at the mini-
and mechanical properties have to be considered. All these aspects, mum joint fabrication cost. This procedure was implemented for
added to the various interconnected relations of the involved vari- joints made with beams produced of rolled W and IPE profiles
ables, inhibit the optimum joint configuration to be easily obtained. and was focused on the transmission of a number of percentage
When the search for optimal structural joints are considered levels of the beam plastic moment capacity.
within this context the Genetic Algorithms (GAs) proves to be a The optimization procedure that uses genetic algorithms is in-
valuable tool for optimizing steel and composite semi-rigid joints. spired in the Darwin principle of the species genetic evolution.
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 179

Δ Δ
Rd
K F ke kp ; F F

F F

ideal
behaviour
kp
FRd EC3

Δ ke Δ
Δf

Fig. 2. Typical load versus displacement (F  D curve for a generic component.

Fig. 3. Genetic algorithm structure.

Fig. 4. Genetic operator – crossover operation.

The process is based on probabilistic algorithms, considering produce an efficient hybrid algorithm. They showed that introduc-
survival and reproduction principles of the best fit individuals. ing SA strategically in the GP process its productivity would be in-
The Genetic Algorithm (GA) is a search mechanism used in the creased. The hybrid Folino et al. algorithm [31], named GP/SA, was
computer science to determine approximated solutions in optimi- used to determine the flexural capacity and rotation stiffness of
zation problems. The genetic algorithms are a particular class of steel beam to column joints leading to interesting results.
evolutionary algorithms that use techniques inspired on the evolu- Over the last few years a particular subarea of the GP, with a lin-
tionary biology like: inheritance, mutation and natural selection ear structure similar to a DNA molecule, named Linear Genetic Pro-
[25]. gramming (LGP) was developed [29]. The LGP is also a machine
The genetic programming is an automatic programming tech- learning method that involves programs through low level lan-
nique that enables the computer programming evolution to solve guages instead of the traditional Koza tree-based GP [26] that is
(or approximately solve) problems. It manipulates correct and represented in the form of a functional in the programming lan-
incorrect solutions, encourage inconsistent and contradictory ap- guage. Despite of the significant advantages of the LGP over the
proaches, do not present a dynamic logic variability, it is mainly other approximation models [29,32], very few works have been
probabilistic, produces non-conform solutions and do not present developed involving Civil Engineering applications.
a clearly defined termination criteria [26]. The Genetic Program- The software developed in the present study starts the design
ming (GP) [26,27] is a subarea of the evolutionary algorithms process executing an elastic analysis to determine the first estima-
[28,29] that follows the Darwin evolution theory. The GP can be tive of the structural response. This preliminary data is used in the
usually treated as a machine learning technique that searches a genetic algorithm system with its basic operations (crossover,
program space instead of data space [27] and have been used with mutation, etc.) to obtain a new parameter value that is subse-
success in the solution of Civil Engineering problems [25,30]. quently compared to the desired objectives. This iterative proce-
Folino et al. [31], combined GP and Simulated Annealing (SA) to dure is repeated until the optimum values are reached, i.e.,
180 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

Fig. 5. SRJ Tool – Semi-Rigid Joint Tool.

Fig. 6. SRJTool – Semi-Rigid Joint Tool graphic interfaces.

minimizing the joint cost for a required flexural capacity and/or rithm software, Evolver [33], producing trustable results in terms of
initial stiffness. This procedure is based on the use of a genetic algo- economy and efficiency, with a reduced computational effort.
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 181

Fig. 7. Joint geometrical properties manipulated by the genetic algorithms.

Fig. 8. Excel spreadsheet used to link the SRJTool to the Evolver software.
182 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

Fig. 9. Interactive procedure using SRJTool and Evolver softwares.

2. The semi-rigid joint design procedure netic’’ operators that are inspired on the Darwin’s principle of the
species evolution and genetics. In 1990, Koza [26] proposed an
This paper used the procedures for the endplate beam to col- evolution-based system, genetic programming, to search for the
umn steel and composite joint design according to the component most fit computer program to solve a particular problem. This
method proposed in the Eurocodes 3 and 4, [1,2], as well as some computational intelligence technique was used by the authors to
recommendations proposed by Ahmed and Nethercot [24] and Ra- calibrate and forecast the post-limit stiffness of the column web
mires [34]. In this procedure the joint global response can be deter- out of plane component of minor axis beam-to-column joints [22].
mined based on the mechanical properties of its parts Genetic Algorithms (GAs) use a vocabulary borrowed from nat-
(components). The components are assembled in a mechanical ural genetics. It is common to talk about individuals (or genotypes,
model as illustrated in Fig. 1. The adopted components are: column structures) in a population; quite often these individuals are called
web panel in shear (cws), column web panel in compression (cwc), also strings or chromosomes. Chromosomes are made of units,
column web panel in tension (cwt), column flange in bending genes (also features, characters, or decoders) arranged in linear
(cfb), endplate in bending (epb), beam flange in compression (bfc), succession where every gene controls the inheritance of one or
beam web in tension (bwt), longitudinal reinforcing bars in tension several characters.
(rbt) and bolts in tension (bt). Fig. 3 presents the structure of a genetic algorithm where each
In general, each component is characterized by a non-linear individual represents a potential solution to the problem in hand.
force–displacement (F–D) curve, adequately represented by a bi- The first population is randomly generated. After creating it, each
linear approximation, depicted in Fig. 2, when only the initial stiff- member is evaluated by computing the representative objective
ness or joint bending capacity is required [35]. In this procedure, and constraint functions, and are later compared to other mem-
the post-limit stiffness of the components is disregarded since a bers. Each of these solutions is evaluated to determine a measure
perfect elastic–plastic behaviour is assumed. However, this meth- of its fitness. Subsequently, a new population (iteration t + 1) is
od was shown to be unconservative, whenever ductility is a critical formed by selecting the most fit individuals. Some members of
issue. the new population undergo transformations by means of genetic
The component method application to steel joints requires the operators like: crossover and mutation. Crossover is a genetic oper-
following steps: ator which forms a new chromosome by combining parts of each of
two parental chromosomes, Fig. 4. Mutation is a genetic operator
 Selection of the relevant (active) components from a global that forms a new chromosome by making (usually small) altera-
components list (20 different components are currently tions to the values of genes in a copy of a single parent chromo-
codified in Eurocodes 3 and 4 [1,2]); some. This process of going from the current population to the
 Evaluation of each component force–deformation next population constitutes one generation in the genetic algo-
response; rithm evolution process. After some generations the program con-
 Assembly of the active components for the evaluation of verges to a feasible solution where the best individual represents a
the joint moment versus rotation curve using a representa- near-to-optimum solution.
tive mechanical model.
4. SRJTool developed software and the proposed GA modules
3. Genetic algorithms
The SRJTool (Semi-Rigid Joint Tool) software was conceived
During the last thirty years there has been a growing interest in using a Borland Delphi 7 Development Tool (Object Pascal) [32]
problem solving algorithms based on principles of evolution. They to facilitate the joint semi-rigid design and it is illustrated in
have selection processes based on individual fitness, and some ‘‘ge- Fig. 5. It was developed to provide a computer user-friendly tool
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 183

(a)

(b)

Fig. 10. Adopted optimization joint models: (a) Extended steel endplate, (b) flush steel endplate, (c) composite extended endplate (d) composite flush endplate.

that could help researchers and structural engineers to easily zation process that are manipulated by the genetic algorithms.
implement the component method procedures (proposed in Euro- These variables are: number of bolt rows (nr), endplate thickness
code 3 [1]) without the need of using time-consuming finite (tp), bolt diameter (bd), vertical distance from the endplate top edge
elements packages. Additionally, since the program was developed to the top beam flange external face (dsb), horizontal distance from
for a specific use, it is possible to control every step of the design the bolt first column to the endplate edge (e2), horizontal pitch be-
procedure and add extra functions as the Genetic Modules. tween bolts (e3), vertical distance from the first bolt row to the
Steel joints may present a wide range of geometries, with differ- endplate top edge (e1), vertical distance between first and second
ent numbers of bolt rows and connecting parts. Due to this wide bolt rows (p1), vertical distance between second and third bolt
range it was necessary to create a standard way of modelling the rows (p2), vertical distance between third and fourth bolt rows
joint so that SRJTool could evaluate the component method model. (p3), vertical distance between forth and fifth bolt rows (p4), verti-
For this reason, a user-friendly interface was used to input the joint cal distance from the fifth bolt row to the endplate bottom edge
parameters. The first window is used to input the joint details fol- (en), reinforcing steel bar number (nb) and reinforcing steel bar
lowed by the structure data like: bracing, analyses type, etc. The diameter (rbd), For cases where more than five bolt rows are used,
next step consists in choosing the joint type, location, design codes other vertical distances should also be considered.
and the presence of stiffeners. Finally, the profile properties and As the Evolver software is a Microsoft Excel plug-in [33] aiming
endplate characteristics are evaluated. Some of these procedures to facilitate the work, a spreadsheet was used to make the data
may be observed in Fig. 6. transfer between the SRJTool data and the Evolver data. This
In order to evaluate the joint global response their full geomet- spreadsheet may be observed in Fig. 8 where the adopted chromo-
rical and mechanical properties should be considered. With these some is represented.
results in hand, the joint mechanical model can be characterized In the optimization process, the first step consists of using the
according to the type of connection. Finally, the component resis- SRJTool software to obtain the first joint configuration just consid-
tance is evaluated and the joint moment versus rotation curve ering the beam and column parameters, [36]. Subsequently, using
may be obtained. Fig. 7 presents the variables used in the optimi- the above mentioned spreadsheet, the optimization process begins
184 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

(c)

(d)

Fig. 10 (continued)

with the genetic algorithm operating the changes in the chromo- 5. Composite and steel joint optimization examples
some (crossover and mutation) that it created after the first
iteration. A new geometric joint configuration is generated with a This paper uses four sets of examples to illustrate the developed
different bending resistance, initial stiffness and rotation capacity. optimization procedure, Fig. 10. The first example considers com-
While the target value is not reached, the optimization process con- posite and steel flush endplate joints where the optimization aim
tinues. This interactive procedure can be observed in Fig. 9. was the required joint flexural capacity Mj,Rd, The second example
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 185

Table 1 650
Summary of the performed optimization examples.
603,31
Stage I (steel and composite) ? bending capacity [Mj.Rd] 600
Flush end plate Mj.Rd = 120 kN m
Extended end plate Mj.Rd = 423 kN m

Mj,Rd [kNm]
550
Stage II (steel and composite) ? bending capacity [Mj.Rd] + cost minimization [$]
Flush end plate Mj.Rd = 120 kN m + reduce cost
500
Extended end plate Mj.Rd = 423 kN m + reduce cost
453,31
452,31
Stage III (steel and composite) ? beam plastic moment capacity [Mpl] 450
Flush end plate Mj.Rd = 200 kN m 436,57 427,29
(steel and composite) 40.39% of Mpl 416,48
Flush end plate Mj.Rd = 300 kN m 400
(composite) 60.58% of Mpl
first joint 1 2 3 4 5
Flush end plate Mj.Rd = 400 kN m Optimization number
(composite) 88.77% of Mpl
Extended end plate Mj.Rd = 700 kN m Fig. 11. Extended steel endplate model results – first example.
(steel and composite) 88.23% of Mpl
Extended end plate Mj.Rd = 800 kN m
(composite) 100.83% of Mpl
Extended end plate Mj.Rd = 900 kN m ity Mj,Rd. The third optimization example aimed to determine joints
(composite) 113.448% of Mpl able to transmit bending moments levels expressed in terms of a
percentage of the plastic bending moment capacity of the adopted
Stage IV (steel and composite) ? beam plastic moment capacity [Mpl] beam IPE steel profiles. The fourth example considers composite
Profile Mpl Joint configuration Parameter Results
flush endplate joints and steel extended endplate joints able to
Beam kN m Comparison transmit bending moment levels corresponding to 60%, 80% and
W 60% Flush composite Cost Mj,Rd
100% of the plastic bending moment capacity of the adopted beam
and 80%  &
IPE 100% Extended Steel Minimization Sj,ini
IPE and W steel profiles, aiming to minimize their global fabrica-
tion costs. At this point it is important to clarify the stop criteria
used in the optimization process: i.e. a 2% deviation from the
objective. Due to the large number of adopted restrictions they will
was performed considering the same joint typology to minimize not be listed in the present paper. The adopted constraints were all
the joint cost (€) while observing the required joint flexural capac- based on the Eurocode 3 [1] design and detailing limitations like:

Table 2
Extended endplate steel joint model results (objective: Mj.Rd = 423 kN m).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First joint 603.31 105290.43 113.59
1 Final joint (first/final) 436.57 1.38 84675.45 1.24 79.34 1.43
2 453.31 1.33 91317.91 1.15 78.96 1.44
3 452.31 1.33 98108.85 1.07 108.09 1.05
4 416.48 1.45 88888.92 1.18 59.17 1.92
5 427.29 1.41 89901.43 1.17 68.74 1.65

Table 3
Flush endplate steel joint model results (objective: Mj.Rd = 120 kN m).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First joint 199.07 26625.43 69.11
1 Final joint (first/final) 115.82 1.72 26013.58 1.02 43.11 1.60
2 120.37 1.65 39962.47 0.67 44.21 1.56
3 119.86 1.66 40582.04 0.66 51.70 1.34
4 119.13 1.67 41126.51 0.65 44.74 1.54
5 137.14 1.45 43073.46 0.62 44.00 1.57

Table 4
Composite extended endplate steel joint model results (objective: Mj.Rd = 423 kN m).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First joint 603.31 105290.43 113.59
1 Final joint (first/final) 435.08 1.39 190889.86 0.55 60.79 1.87
2 431.13 1.40 189308.62 0.56 59.72 1.90
3 432.61 1.39 190879.71 0.55 60.27 1.88
4 433.31 1.39 185180.88 0.57 85.73 1.32
5 411.64 1.47 185682.97 0.57 78.62 1.44
186 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

210 All the examples presented in this paper were limited to two
199,07 types of structural joints, i.e., flush and extended endplate joints.
These joints were initially considered to be made only with steel
180
members. In a subsequent phase equivalent composite joints were
Mj,Rd [kNm]

also considered to enable a direct comparison of the these alterna-


150 tive typologies. Table 1 summarizes the main objectives of the per-
137,14 formed optimizations. In first and second examples, five
independent optimizations were made to determine the best re-
120
115,82 120,37 119,86 119,13 sults. In the other examples a single optimization was performed
aiming to transmit the required bending moment level. All the
90 examples started from an initial basic joint configuration, as can
1 2 3 4 5
First Joint be observed in the tables presented in this paper. These tables also
Optimization number
depict the ratio between the initial/final parameters that were
Fig. 12. Flush steel endplate model results – first example. optimized to determine the best alternative for each of the inves-
tigated joints. Fig. 10 presents typical examples of the joint dimen-
sions and parameters obtained with the optimization processes
being in line with the current design practice.
650
5.1. First optimization example
603,31
600
The first example involved flush and extended steel endplate
Mj,Rd [kNm]

550 joints, Tables 2 and 3, aiming to transmit bending moment values


close to 120 kN m and 423 kN m. The bending moment optimiza-
500 tion results are illustrated in Figs. 11 and 12, showing, as expected,
a large dispersion of the initial stiffness, Sj,ini and the joint global
cost values since no restriction was applied to these variables dur-
450 435,08 431,13 432,61 433,31
ing the optimization process. A large reduction of the initial stiff-
411,64 ness, Sj,ini can be observed when the extended endplate model
400
final results are compared to the initial steel joint model. A differ-
first joint 1 2 3 4 5
ent trend can be observed with the flush endplate joint final results
Optimization number
that showed a significant stiffness increase when compared to the
Fig. 13. Composite extended endplate model results – first example. initial steel joint model.
Composite flush and extended endplate joints were also opti-
mized to transmit bending moment values close to 120 kN m and
423 kN m, Tables 4 and 5 where, in both cases, an abrupt increase
210
199,07
of the initial stiffness, Sj,ini can be observed when compared to their
190 equivalent steel joint models, demonstrating one of the advantages
of using composite joints. The bending moment results optimiza-
170 tion results can also be visualized in Figs. 13 and 14.
M j,Rd [kNm]

150
144,24 5.2. Second optimization example
131,37 133,45
130
120,01 The second example involves flush and extended steel and com-
110 116,36 posite endplate joints, Tables 6–9, aiming to minimize their global
cost and to transmit bending moment values close to 120 kN m
90 and 423 kN m. At this point, it is necessary to clarify that, in the
first joint 1 2 3 4 5 present example, the additional cost related to the adoption of lon-
Optimization number gitudinal reinforcing bars in the composite models was not in-
cluded in the joint global cost evaluation procedure. One more
Fig. 14. Composite extended flush model results – first example.
time, a large dispersion of the initial stiffness, Sj,ini can be observed
since no restriction was applied to this variable during the optimi-
minimum and maximum bolt to bolt distance among many others. zation process. The steel endplate joint results are also depicted in
Further details should be referenced in the Eurocode 3 [1]. Figs. 15 and 16 and the composite endplate joints results can be
visualized in Figs. 17 and 18. Once again an abrupt increase of

Table 5
Composite flush endplate steel joint model results (objective: Mj.Rd = 120 kN m).

Optimization Configure Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First joint 199.07 26625.43 69.11
1 Final joint (first/final) 144.24 1.38 45696.35 0.58 46.64 1.48
2 120.01 1.66 43101.19 0.62 47.23 1.46
3 116.36 1.71 42273.29 0.63 45.28 1.53
4 131.37 1.52 46263.31 0.58 47.43 1.46
5 133.45 1.49 45256.15 0.59 50.82 1.36
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 187

Table 6
Extended endplate steel joint model results (objective: Mj.Rd = 423 kN m + minimum cost).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First Joint 603.31 105290.43 113.59
1 Final joint (first/final) 428.55 1.41 93939.45 1.12 57.83 1.96
2 446.77 1.35 88288.55 1.19 73.23 1.55
3 442.66 1.36 102563.99 1.03 64.98 1.75
4 436.78 1.38 78304.26 1.34 75.49 1.50
5 440.91 1.37 82630.96 1.27 82.70 1.37

Table 7
Flush endplate steel joint model results (objective: Mj.Rd = 120 kN m + minimum cost).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First Joint 199.07 26625.43 69.11
1 Final joint (first/final) 128.30 1.55 35112.41 0.76 26.17 2.64
2 131.26 1.52 37684.87 0.71 40.39 1.71
3 127.23 1.56 44418.01 0.60 48.52 1.42
4 177.74 1.12 42149.20 0.63 30.42 2.27
5 133.30 1.49 39389.86 0.68 47.39 1.46

Table 8
Composite extended endplate steel joint model results (objective: Mj.Rd = 423 kN m + minimum cost).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First Joint 603.31 105290.43 113.59
1 Final joint (first/final) 430.76 1.40 189428.81 0.56 60.87 1.87
2 431.30 1.40 189186.76 0.56 60.96 1.86
3 433.59 1.39 189902.87 0.55 61.51 1.85
4 430.89 1.40 190646.30 0.55 60.54 1.88
5 431.47 1.40 190702.13 0.55 58.65 1.94

Table 9
Composite flush endplate steel joint model results (objective: Mj.Rd = 120 kN m + minimum cost).

Optimization Configuration Mj.Rd (kN m) Sj.ini (kN m/rad) Total cost ($)
First Joint 199.07 26625.43 69.11
1 Final joint (first/final) 115.28 1.73 42069.40 0.63 45.13 1.53
2 113.50 1.75 41898.21 0.64 44.97 1.54
3 116.36 1.71 42273.29 0.63 45.28 1.53
4 101.97 1.95 38538.18 0.69 45.13 1.53
5 128.05 1.55 47407.21 0.56 45.28 1.53

650 210
199,07

603,31 190
600
177,74
170
Mj,Rd [kNm]
Mj,Rd [kNm]

550
150
500 131,26 133,30
130 127,23
128,30
446,77
450 110
440,91
442,66 436,78
428,55
400 90
First Joint 1 2 3 4 5 First Joint 1 2 3 4 5
Optimization number Optimization number

Fig. 15. Extended steel endplate model results – second example. Fig. 16. Flush steel endplate model results – second example.

5.3. Third optimization example


the initial stiffness, Sj,ini can be one more time observed when com-
pared to their equivalent steel joint models, demonstrating again The third optimization example aimed minimize the joint costs
one advantage of the using composite joints. while the joints were still capable to transmit bending moments
188 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

650 Table 13
Composite and steel models cost comparisons and structural performance.

600 603,31 $steel/$comp. Mm/Ma Sm/Sa


1.269 0.986 0.971
Mj,Rd [kNm]

550 1.260 1.200 1.056


1.126 1.312 1.057

500

levels expressed in terms of a percentage of the plastic bending


450 430,76 431,30 433,59 430,89
431,47
moment capacity of the adopted beam IPE steel profiles. In contrast
to the previous example, the performed optimization included the
400 additional cost of adopting longitudinal reinforcing bars in the
First Joint 1 2 3 4 5
composite joint global cost evaluation procedure.
Optimization number The flush steel and composite endplate joints were optimized to
Fig. 17. Composite extended endplate model results – second example.
minimize their final global cost while transmitting bending mo-
ment levels close to 40%, 50% and 60% of the plastic bending mo-
ment capacity of the adopted IPE500 steel profile, Tables 10 and
210 11. The composite joint advantages can be observed when the
199,07 models are compared. The results indicated that the composite
190 joints performed better when higher bending transmission levels
are required while their equivalent steel joints are restricted to a
170 limited bending moment transmission capacity. Table 11 also indi-
Mj,Rd [kNm]

cated that a higher initial stiffness, with a negligible cost increase,


150
was also obtained with the composite joints.
The extended steel and composite endplate joints were opti-
130
128,05 mized to minimize their final global costs and to transmit bending
115,28 113,50 116,36
110 moment levels around to 88%, 100% and 113% of the plastic bend-
101,97 ing moment capacity of the adopted IPE600 steel profile, Tables 12
90 and 13. Similar conclusions to the flush endplate joints can be
First Joint 1 2 3 4 5 made confirming the better performance of the composite joints
Optimization number without affecting the joint global cost. Table 13 indicated again
that a substantial increase of the joint initial stiffness and flexural
Fig. 18. Composite flush endplate model results – second example.
capacity, with an insignificant cost increase, was also obtained
with the composite joints.

Table 10
Flush endplate steel joint model results.
5.4. Fourth optimization example
First joint configuration Flush steel joint
Mpl % Mpl Mj.Sd $ Mj.Rd Mj.Sd/ $ $first/ Sj.ini The fourth example considers composite flush endplate joints
Mj.Rd $final and steel extended endplate joints able to transmit bending mo-
500 40.00 200 53.00 204.81 0.977 52.99 1.000 57915.86 ment levels corresponding to 60%, 80% and 100% of the plastic
Flush composite joint bending moment capacity of the adopted beam IPE and W steel
500 40.00 200 53.00 197.94 1.010 52.32 1.013 58639.62
profiles, aiming to minimize their global fabrication costs. Fig. 19
500 60.00 300 53.00 297.92 1.007 52.70 1.006 72201.19 present the optimization results of composite flush endplate joints
500 80.00 400 53.00 407.48 0.982 56.81 0.933 87735.74 and steel extended endplate joints using, for the beams, four rolled
W steel profiles: W610  101, W530  92, W460  74 and
W410  67. In this figure results and parameters obtained in all
the optimizations performed, i.e., the joint cost expressed in Euros,
Table 11 the joint initial stiffness divided by 1000 and the joint flexural
Composite and steel models cost comparisons and structural performance.
capacity Mj,Rd. Every graph depict the individual results for every
$steel/$comp. Mm/Ma Sm/Sa investigated steel profile comparing the two investigated joints,
1.013 0.966 1.012 i.e. the composite flush endplate joint and the steel extended end-
1.006 1.455 1.247 plate joint. It can be observed that the first two graph groups are
0.933 1.990 1.515 related to optimization procedures associated to a bending

Table 12
Extended endplate steel joint model results.

First joint configuration Extended steel joint


Mpl % Mpl Mj.Sd $ Mj.Rd Mj.Sd/Mj.Rd $ $first/$final Sj.ini
793 88.30 700 125.19 670.34 1.045 125.19 1.000 228007.50
Extended composite joint
793 88.30 700 125.19 661.25 1.059 98.62 1.269 221338.84
793 100.90 800 125.19 804.10 0.995 99.39 1.260 240790.06
793 113.50 900 125.19 883.36 1.019 88.06 1.422 243371.73
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 189

Extended Steel Joint X Flush Composite Joint moment level equal to 60% of the adopted beam profile plastic
W 610x101 moment resistance while the other graphs are related to bending
800 moment levels equal to 80% and 100% of the adopted beam profile
Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] 676,00
700 plastic moment resistance.
Cost / sj ,ini/ mj,rd

(s) _ steel (c)


600 (c) - composite 537,57 541,55 It can also be noted that not all the graphs presented values ex-
500
413,31 (s) (c) actly at the target moments. A ±2% difference from the optimiza-
403,12
400 tion value was considered accurate enough due to the use of
(s) (c)
300 discrete variables, like the bolt number that has to be equal to an
200,22 222,35
(s)
200 133,02
even integer value. These results indicated that the extended steel
106,17 118,98
100 52,38 61,67 40,92 46,49
endplate joint could not produce results for bending moment lev-
36,53
0,00
0,00 0,00 els equal to 100% of the adopted beam profile plastic moment
0
58,85% 60,34% 78,48% 79,06% - 98,69% resistance. The flexural resistance of the composite flush endplate
joints is, in the majority of cases, higher than their extended steel
Target Mpl [kN.m]
Steel Profile Bending Plastic Capacity (%) endplate joint counterparts for the heavier steel profiles, i.e. W610,
W530 and W460. This fact did not happen for the W410 lighter
Extended Steel Joint X Flush Composite Joint steel profile where this difference was reduced as the steel
W 530x92 endplate joint becomes more efficient.
600 558,81 The steel endplate joint initial stiffness was usually higher than
(s) _ steel
Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] (c) - composite (c) the composite flush endplate joint counterparts due to the larger
500 444,62
441,07 stiffness associated to the extended endplate joint components.
Cost / sj,ini / mj,rd

(s) (c)
400 Alternatively the composite flush endplate joint costs were smaller
344,13
326,71
(c) than their extended steel endplate joint counterparts, and in this
300 (s)
case, a similar conclusion to the one associated to the flexural
200 182,18 166,87 capacity can be reached as the composite joints were cheaper for
153,83 147,00
131,73 (s) heavier steel profiles while the steel joints were more economical
100 43,64
36,36 52,92 38,84 43,54 for the lighter steel sections.
0,00 0,00
0,00 Fig. 20 depicts the optimization results of composite flush end-
0
plate joints and steel extended endplate joints using, for the beams,
58,97% 62,12% 79,62% 80,26% - 100,87%
four rolled IPE steel profiles: IPE600, IPE550, IPE450, and IPE400.
Target Mpl [kN.m]
Steel Profile Bending Plastic Capacity (%)
The same parameters previously mentioned were used in Fig. 19.
One more time it could be noted that not all the graphs presented
Extended Steel Joint X Flush Composite Joint
values exactly at the target moments presenting a 2–3% difference.
W 460x74
450 A close inspection of the graphs present in Fig. 20 indicates that the
400 Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] 385,90 IPE600 steel profile could not produce any composite or steel joint
(s) _ steel (c) solution for bending moment levels equal to 100% of the beam
Cost / sj,ini / mj,rd

350 328,43
(c) - composite 307,31
300 (s)
(c) plastic moment resistance. The conclusions associated to the com-
250 238,12 238,27 posite models are the opposite as it happened with the W steel
200
(s) (c) profiles, Fig. 16, i.e., in this case the steel joints prevail for heavier
130,33 sections while the lighter sections were better suited to the com-
150 118,73 121,96
105,58 (s) posite joints. This apparent contradiction can be explained by the
89,70
100
30,85 36,49 32,68 38,98 discrete nature in which the W and the IPE rolled steel section ser-
50 29,21
0,00
0,00
0,00 ies were conceived.
0
The steel endplate joint initial stiffness was higher than the
60,74% 60,78% 78,40% 83,78% - 98,44%
composite flush endplate joint counterparts due to the larger stiff-
Target Mpl [kN.m]
ness associated to the extended endplate joint components. The
Steel Profile Bending Plastic Capacity (%)
difference between stiffness of the steel and composite joints in-
Extended Steel Joint X Flush Composite Joint
crease as lighter steel sections are considered. All the composite
W 410x67 flush endplate joint costs were smaller than their extended steel
350 320,17 endplate joint counterparts. It could also be observed that there
300 Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] (c) is a proportionality between the required bending moment capac-
(s) _ steel
ity and the final joint costs.
Cost / sj,ini / mj,rd

(c) - composite 244,08 242,46


250
(s) (c) The results demonstrated that the IPE sections were associated
183,88
200 181,93 to a better performances in terms of flexural capacity and initial
(s) (c)
150 stiffness than the W profiles for the two heaviest sections. This
108,24 conclusion was achieved despite the fact that the W sections were
91,21 94,76
100 78,37 (s)
66,64 associated to larger geometrical properties (for example the beam
50 27,39 24,94 33,08 29,34 32,53 plastic module). Alternatively for the two lighter W sections the
0,00
0,00
0,00
0 performance in terms of flexural capacity, cost and initial stiffness
59,51% 58,88% 78,99% 78,47% - 103,61% was better than the IPE profile counterparts. It is important to ob-
Target Mpl [kN.m] serve that the W sections were associated to larger geometrical
Steel Profile Bending Plastic Capacity (%) proper. The same conclusions drawn for the steel joints can be ap-
plied to the composite joints, i.e. heavier sections performed better
Fig. 19. Cost and structural performance of the composite and steel models with W with the W sections while lighter sections with the best perfor-
steel profiles. mance were associated to the IPE profiles.
190 F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191

Extended Steel Joint X Flush Composite Joint


IPE 600
700 635,17 622,70
600 (s) (c)

Cost / sj,ini / mj,rd


499,67
500 471,93
Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm]
(s)
(c)
400
(s) _ steel
300 (c) - composite
197,14 216,81
173,49 192,93
200
83,50
100 56,07 46,07 54,36
0,00
0,00
0,00 0,00
0,00
0,00
0
63,01% 59,51% 80,10% 78,52% - -
Target Mpl [kN.m]
Steel Profile Bending Plastic Capacity (%)

Extended Steel Joint X Flush Composite Joint


IPE 500
700
Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] 613,75 616,04
600 (c)
(s)
Cost / sj,ini / mj,rd

(s) _ steel 498,08 491,74


500
(c) - composite
(s) (c)
400 368,55 379,32

(s) (c)
300
205,41
200 168,61 181,40 173,55
161,35
139,85
83,34
100 50,78 43,11 53,44 45,98 60,17

0
58,87% 60,59% 79,57% 78,55% 98,04% 98,41%
Target Mpl [kN.m]
Steel Profile Bending Plastic Capacity (%)

Extended Steel Joint X Flush Composite Joint


IPE 450
400 376,20 378,29
Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] (s) (c)
350
310,93
(s) _ steel 300,23
Cost / sj,ini / mj,rd

300 (c) - composite (c)


(s)
250 224,79 224,87

200 (s) (c)


148,42
150 125,97 119,52
106,15 98,70
100 83,49
60,56
36,93 35,27 45,06 36,42
50 31,22

0
60,74% 60,78% 78,40% 83,78% - 98,44%
Target Mpl [kN.m]
Steel Profile Bending Plastic Capacity (%)

Extended Steel Joint X Flush Composite Joint


IPE 400
300 284,50 285,51
Cost $ sj,ini [kNm/rad.1000] mj,rd [kNm] (s) (c)
250 229,60
(s) _ steel 228,46
Cost / sj,ini / mj,rd

(c) - composite (s) (c)


200 173,29
170,87
(s) (c)
150
117,39
87,46 94,32
100 83,11
69,36
57,28
40,72 43,75
50 28,82 25,56
34,70 33,11

0
59,51% 58,88% 78,99% 78,47% - 103,61%
Target Mpl [kN.m]
Steel Profile Bending Plastic Capacity (%)

Fig. 20. Cost and structural performance of the composite and steel models with IPE steel profiles.
F.B. Ramires et al. / Engineering Structures 45 (2012) 177–191 191

6. Conclusion [8] Ostrander JR. An experimental investigation of end-plate connections. MSc


dissertation, University of Saskatchewan, Canada, 1970.
[9] Bailey JR. Strength and rigidity of bolted beam-to-column connections.
This paper main aim was to minimize the effort associated to Conference on joints structures, University of Sheffield, 1970.
the composite and steel semi-rigid joint design describing an auto- [10] Surtees JO, Mann AP. End-plate connections in plastically designed structures.
Conference on joints in structures, University of Sheffield, 1970.
mated optimization process developed specifically for this pur-
[11] Packer JA, Morris LJ. A limit state design method for the tension region of
pose. The main objective of the optimization was to reduce the bolted beam-to-column connections. Struct Eng 1977;55(1):446–58.
global joint cost ensuring that the associated joint flexural capacity [12] Johnson ND, Walpole WR. Bolted end-plate beam-to-column connections
under earthquake type loading. Rep. 81–7, University of Canterbury,
and initial stiffness were limited to their associated ultimate and
Christchurch, New Zealand, 1981.
serviceability limit states. Combining all the involved adjustable [13] Jaspart JP. Etude de la Semi-rigidité des Noeusds Poutre-colonne et Son
variables the process demonstrated its potentiality leading to more Influence Sur la Résistance et la Stabilité des Structures en Acier. Phd thesis,
efficient design solutions. Université de Liège, Belgium, 1991 [in French].
[14] Cruz PJS, da Silva LAPS, Rodrigues DS, Simões RAD. Database for the semi-rigid
When the results of the flush endplate joints it WAs possible to behaviour of beam-to-column connections in seismic regions. J Construct Steel
observe that the optimum joints presented a 10% cost reduction Res 1998;46(120):1–3.
while the initial stiffness was increased up to 61%. This stiffness gain [15] Cruz PJS, Silva LAPS, Rodrigues DS. Sericon II: a global database for tests on
structural connections. In: Proceedings of cost C1 international conference on
leads to smaller deflections optimizing one of the serviceability limit the control of semi-rigid behaviour of civil engineering structural connections,
states that usually controls the composite beam design process. The Liege, Belgium, 1998.
fourth example results indicated that the composite flush endplate [16] Cruz PJS, da Silva LAPS, Rodrigues DS, Simões RAD. Database for the semi-rigid
behaviour of beam-to-column connections in seismic regions. J Construct Steel
joints performed better, in terms of flexural capacity confirming that Res 1998;46(120):1–3.
the incorporation of the longitudinal bars and the reinforced [17] Azizinamini A, Bradburn JH, Radziminski LB. Initial stiffness of semi-rigid steel
concrete slab in the composite joint models led to a substantially beam-to-column connections. J Construct Steel Res 1987;8:71–90.
[18] Azizinamini A, Radziminski LB. Static and cyclic performance of semi-rigid
higher flexural capacity than the non-composite joint models.
steel beam-to-column connections. J Struct Eng 1989;115(12):2979–99.
In the developed optimization process various ideas for future [19] Queiroz G. Análise Experimental e Análise Não-Linear de Nós Soldados Semi-
improvements are possible like: standardizing the structural ele- Rígidos. PhD thesis, UFMG, Belo Horizonte, Brazil, 1995 [in Portuguese].
[20] Carvalho LCV, Andrade SAL, Vellasco PCGS. Experimental analysis of bolted
ments, using readily available commercial products with lower
semi-rigid connections. J Construct Steel Res 1998;46(1–3):1–20.
associated costs like some specific bolt diameters or inhibiting [21] Lima LRO, Vellasco PCGS, Andrade SAL. Bolted semi-rigid connections in the
the use of some unavailable plate thicknesses. A broader investiga- column’s minor axis. In: 2nd European conf. steel structures – Eurosteel,
tion could be performed envisaging a more comprehensive insight Prague, 1999. p. 1–14.
[22] Lima LRO, Vellasco PCGS, Andrade SAL, Silva LAPS. Experimental and
of the computational intelligence techniques to improve the devel- mechanical model for predicting the behaviour of minor axis beam-to-
oped processes in particular the genetic algorithm method in order column semi-rigid joints. Int J Mech Sci 2002;44(6):1047–65.
to better control the variable constrains and maximum number of [23] Ribeiro LFL. Comportamento Estrutural de Ligações Viga-Coluna com Chapa de
Topo: Análise Teórico-Experimental. PhD thesis, USP, São Paulo, 1998 [in
performed optimizations enabling with these approaches, better Portuguese].
data comparisons and crosschecks. [24] Ahmed B, Nethercot DA. Prediction of initial stiffness and available rotation
capacity of major axis composite flush end plate connections. UK: University
of Nottingham; 1996.
Acknowledgements [25] Javadi AA, Rezani M, Mousavi Nezhad M. Evaluation of liquefaction induced
lateral displacements using genetic programming. Comput Geotech
2006;33(4–5):222–33.
The authors gratefully acknowledge the financial support pro- [26] Koza JR. Genetic programming on the programming of computers by means of
vided by the Brazilian National and State Scientific and Technolog- natural selection. Cambridge (MA): The MIT Press; 1992.
ical Agencies: CNPq, CAPES and FAPERJ. [27] Banzhaf W, Nordin P, Keller R, Francone F. Genetic programming – an
introduction. on the automatic evolution of computer programs and its
application. Heidelberg/San Francisco: Dpunkt/Morgan Kaufmann; 1998.
[28] Langdon W, Poli R. Foundations of genetic programming. Springer; 2002.
References
[29] Norvig P, Russel S. Artificial intelligence. a modern approach. Prentice Hall;
1995.
[1] Eurocode 3, EN1993-1-1, previous term Eurocode 3. Design of steel structures, [30] Ashour AF, Alvarez LF, Toropov VV. Empirical modeling of shear strength of RC
general rules and rules for buildings, European Com. for Standardization, 2005. deep beams by genetic programming. Comput Struct 2003;81(55):331–8.
[2] Eurocode 4, EN – 1994: design composite steel and concrete structures. Part [31] Folino G, Pizzuti G. Spezzano. Genetic programming and simulated annealing:
1.1 General rules and rules for buildings, European Comission for a hybrid method to evolve decision trees. Genetic programming, EuroGP’2000,
Standardization, 2002. Springer-Verlag, 2000. p. 294–303.
[3] Wilson WM, Moore HF. Tests to determine the rigidity of riveted joints in steel [32] Borland. Borland Delphi 7 Development Tool (Object Pascal): reference library
structures. Bulletin no. 104. Engineering Experiment Station, University of guide. Scotts Valley (California): Borland International, Inc.; 1996.
Illinois, 1917. [33] Evolver 4.0 – Genetic optimization add-in for Microsoft Excel, Palisade
[4] Young CR, Jackson KB. The relative rigidity of welded and riveted connections. Decision Tools, 1999.
Can J Res 1934;11(1–2):62–134. [34] Ramires FB. Avaliação Estrutural de Ligações Semi-rígidas em Aço com Placa
[5] Rathbun JC. Elastic properties of riveted connections. Trans ASCE 1936;101: de Extremidade, MSc dissertation, PUC-Rio, 2004 [in Portuguese].
524–63. [35] Faella C, Piluso C, Rizzano G. Structural steel semi-rigid connections – theory.
[6] Bell WG, Chesson EJ, Munse WH. Static tests of standard riveted and bolted Design and software. CRC Press; 1999.
beam-to-column connections. University Illinois Engineering Experiment [36] Ramires SFB, de Andrade SAL, da Vellasco PCGS. Experimental analysis of
Station, 1959. composite semi-rigid beam to colum joints. Eurosteel 2008 – fifth european
[7] Sommer WH. Behaviour of welded header plate connections. MSc dissertation, conference on steel and composite structures, Graz, v.A, 2008. p. 471–6.
University of Toronto, Canada, 1969.

Vous aimerez peut-être aussi