Вы находитесь на странице: 1из 7

Journal of Constructional Steel Research 67 (2011) 1225–1231

Contents lists available at ScienceDirect

Journal of Constructional Steel Research


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

Optimization of steel floor systems using particle swarm optimization


G. Poitras a,∗ , G. Lefrançois a , G. Cormier b
a
Département de génie civil, Université de Moncton, Moncton, NB, Canada, E1A 3E9
b
Département de génie électrique, Université de Moncton, Moncton, NB, Canada, E1A 3E9

article info abstract


Article history: Particle swarm optimization (PSO) is used for the design of composite and non-composite steel floor
Received 29 September 2010 systems. The design problem is the minimization of the mass or the cost of a steel floor configuration
Accepted 26 February 2011 subject to constraints related to the Canadian S16 design standard. The PSO algorithm was applied to three
different steel floor bays. Outputs returned are the girder and beams sizes, steel deck profile, concrete slab
Keywords: thickness, number of interior beams and the number of steel studs needed per beam. Results show the
Particle swarm optimization
PSO can consistently find the optimum floor configuration by minimizing the total mass or cost while
Steel floor systems
Structural optimization
satisfying all design criteria.
Building design © 2011 Elsevier Ltd. All rights reserved.
Evolutionary algorithms
Composite floors

1. Introduction et al. [12] used three benchmark problems to demonstrate that


PSO is more computationally efficient than the GA. Similar results
Artificial intelligence methods, like the genetic algorithm (GA) were also found by Perez and Behdinan [4], using three test
and the particle swarm optimization algorithm (PSO), are stochas- cases (classical truss optimization examples). Furthermore, PSO
tic procedures that explore solutions to constrained problems. has fewer parameters than GA, is easier to implement and in many
In structural design problems, these optimization methods have optimization problems, shows faster convergence [13].
mostly been studied for pin-connected structures such as trusses Different approaches, all based on the standard PSO algorithm,
[1–8]. A highly constrained problem that should warrant atten- have been used. In particular, He et al. [14] implemented a passive
tion is the optimization of a steel–concrete composite or non- congregation (PSOPC) which improved the convergence rate and
composite beam steel floor system. When designing a commercial accuracy for some types of problems. Li et al. [1] presented a
building, floor systems, together with roof systems, are usually the heuristic PSO (HPSO) that combined PSOPC and a harmony search
first to be designed. They will ultimately be supporting all the grav- (HS) scheme to significantly improve the convergence rate.
ity loads that will, along with their self-weight, be subsequently The work presented here uses the standard PSO to optimize
transferred to the columns. Because of the nature of the design of the design of a composite and non-composite steel floor system,
floor systems, there are many girder, beam, deck and slab combina- subject to various constraints. Floor component design constraints
are handled by using penalty factors. Emphasis will be placed on
tions that will yield an acceptable system. The choice of this com-
the PSO parameter values giving reliable results.
bination is highly dependant on the experience of the engineer [9]
and without the use of these heuristic methods, it is difficult to de-
termine if a lighter floor system is possible. The parameters that 2. Particle swarm optimization
must be selected include all girder and beam sizes, the number of
beams, the depth and profile of the steel deck, the depth of the con- Particle swarm optimization (PSO) is an optimization algorithm
crete slab and, for composite floors, the number of studs required similar to genetic algorithms. It was first proposed by Kennedy
and Eberhart [15]. PSO is based on swarm intelligence, such as the
for desired composite action.
behavior of swarms of bees seeking out flowers to pollinate.
While the cost-optimization of a floor system has been studied
The swarm is composed of particles, where each particle rep-
with the use of a GA [10,11], the performance of optimization
resents a possible solution to an optimization problem. A particle
algorithms for this type of problem was never questioned. Hassan
will explore the search area based on two components: its personal
experience and the collective experience of the swarm. The swarm
tries to maximize the value of the objective function (which is de-
∗ Corresponding author. Tel.: +1 506 858 4566. fined in Section 3 for this application). The basic structure of the
E-mail address: gerard.poitras@umoncton.ca (G. Poitras). particle swarm algorithm is presented in the flow chart of Fig. 1.
0143-974X/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2011.02.016
1226 G. Poitras et al. / Journal of Constructional Steel Research 67 (2011) 1225–1231

function. The fitness of a particle is the value of the objective


function at the particle’s position. The objective function used in
this study is the total mass or cost of the floor.

2.3. Constraints

Once the fitness of a particle has been calculated, constraints


are verified for each particle, and a penalty is assigned for each
constraint that is not satisfied. The constraints used are explained
in Section 3.

2.4. Local and global bests

Each particle retains memory of its best position, called a local


best. This local best represents the best solution obtained (highest
fitness) by the particle. If the fitness at the particle’s current
position is higher than the fitness at the previous best-obtained
position, then this new position becomes the best position.
In a similar fashion, the swarm keeps track of its collective best
position, called the global best. The global best is the best position
of the local bests of the swarm (the position with the highest
fitness), which corresponds to the solution with the lowest total
mass or cost.

2.5. Position update

After the local and global bests are determined, the velocity of
each particle is calculated. The velocity of particle k, for time step
t + 1, is:
 
pg − ⃗
xk [t ]
v⃗k [t + 1] = w⃗vk [t ] + c1 ⃗r1
∆t
 
Fig. 1. Flow chart for the PSO. p̂k − ⃗
xk [t ]
+ c2 ⃗r2 (1)
∆t
The ‘‘particles’’ for the PSO are mathematical constructs,
having three main parameters: position, velocity and fitness. where v ⃗k is the velocity of the particle, p⃗k is the current position
Position represents the unknown variables of the problem, velocity of the particle, p̂k is the particle’s best position, pg is the best
determines the rate of change of the position, and fitness is a position of the swarm, w is the inertial weight, and ∆t is a
measure of how well the particle solves the optimization problem unit time step. The vectors ⃗ r1 and ⃗
r2 represent random numbers
(it is the value of the objective function at the particle’s current in the range [0, 1]; c1 and c2 are weights associated with the
position). These terms will be further defined in the following difference between the global best and the current position, and
sections. the difference between the local best and the current position. A
good starting point is c1 = c2 = 2 [13]. These parameters can also
2.1. Initial population be made to vary linearly at each iteration.
The inertial weight w balances the current velocity against the
The first step of the algorithm is creating the initial swarm of
local and global bests. For this problem, a variable inertial weight
particles. Each particle’s position consists of n variables, where
was used [16]. This prevents the particle from oscillating around
n is the number of unknowns in the optimization problem. The
the optimum point. The previous velocity is multiplied by the
n unknowns for the floor system are girder and beam sizes, the
number of interior beams, the thickness of the concrete slab, the inertial weight w , where w is defined as:
type of steel deck and for composite action only, the number of tmax − t
studs per beam. w = (w1 − w2 ) + w2 (2)
The variables are initialized to a random value within the search tmax
space. In addition to its position, each particle also has n velocities where t is the current iteration and tmax is the maximum number
(in a PSO algorithm context); one velocity for each unknown. All of iterations. A good starting point for these values is w1 = 0.9 and
particles in the initial population can optionally satisfy all design w2 = 0.4 [16].
criteria; a new particle would be generated if it does not satisfy the Finally, the particle’s new position is calculated according to:
design criteria. After many trials, a population size of 200 was used
to ensure consistent results. ⃗xk [t + 1] = ⃗xk [t ] + v⃗k [t + 1]∆t . (3)

2.2. Evaluation of the objective function An example is shown in Fig. 2. The particle’s new position is based
on three vectors: the previous velocity, the difference between the
The objective function is a measure of the quality of a solution. global best and the particle’s current position, and the difference
The swarm attempts to maximize (or minimize) the objective between the local best and the particle’s current position.
G. Poitras et al. / Journal of Constructional Steel Research 67 (2011) 1225–1231 1227

Table 1
User inputs and outputs of PSO algorithm.
User inputs Outputs

Bay width W North girder size GN


Bay length L South girder size GS
Additional dead surface load qD West edge beam size BW
Live surface load qL East edge beam size BE
Vibration live surface load qLv Interior beam size BI
Uniform dead loads on edge beams w Di Number of interior beams NB
Uniform live loads on edge beams w Li Type of steel deck PC
Deflexion limits Total height of concrete slab tc
Limits on number of interior beams Best total mass of floor system
High limits of beams and girders
Steel and concrete properties
Vibration control properties
Additional variables for composite algorithm
Steel stud diameter, DS Number of studs on north girder
Percentage of composite action Number of studs on south girder
Shoring Number of studs on west edge beam
Number of studs on east edge beam
Number of studs on interior beams
Height of studs in mm
PSO algorithm variables
Maximum number of generations MaxGen
Population size Pop
Global weight coefficient (initial and final) c1i , c1f
Local weight coefficient (initial and final) c2i , c2f
Initial inertial weight w1
Final inertial weight w2
Penalty factors for resistance, deflection and vibration (initial and final) pfri , pfdi , pf vi , pfrf , pfdf , pf vf

span of the edge beams is shorter than the span of the interior
beams (length L), the PSO code considers that they are the same.
Therefore, the shapes of the columns are ignored and are not part
of the design process.
Table 1 shows the variables used for PSO algorithm. It is impor-
tant to note that the algorithm can accommodate inside and cor-
ner bays since we can add uniform loads on all of the perimeter
beams and girders. Furthermore, all beams can be considered lat-
erally supported or not by the deck.
Fig. 2. Example of a particle’s new position. In order to consider loads from walls and adjacent bays being
transferred to edge members, input of uniform loads was included.
3. Design problem Inputs include the bay dimensions, W (mm) × L (mm), dead
surface loads superimposed to the mass of the deck and concrete
slab, qD (kPa), live surface load, qL (kPa), vibration live surface load,
3.1. Floor design problem qLv (kPa), uniform dead loads on girders and edge beams, w Di
(kN/m), and uniform live loads on girders and edge beams, w Li
The layout of the flooring system for the application of the (kN/m). Because vibrations are diminished when mass is added to
PSO algorithm is illustrated in Fig. 3. The two girders and the two the floor, only a portion of the live surface load is considered during
edge beams are pin-connected to the columns. Even though the vibration analysis [17].

Fig. 3. Steel floor system configuration and physical input parameters.


1228 G. Poitras et al. / Journal of Constructional Steel Research 67 (2011) 1225–1231

For composite design (see Table 1), the user must specify the Table 2
diameter of the steel studs to be used and the percentage of Prices for various floor components.

composite action that should be considered (40%–100%). However, Component Price ($) Unit
because of stud spacing requirements, the percentage of composite Non composite
action entered is sometimes not attainable. Steel 2.86 per kg
The algorithm will provide the north girder size, GN, south Steel deck 2.25 per kg
girder size, GS, west edge beam size, BW , east edge beam size, Steel deck (application) 5.40 per m2
Concrete 131.00 per m3
BE, and the interior beams size, BI. All beam and girder sizes will
Concrete (application) 5.40 per m2
be hot-rolled W shapes found in the Canadian Institute of Steel
Composite
Construction (CISC) Handbook excluding class 4 sections [18]. The Stud (interior beams) 5.00 per stud
steel deck choices were taken from the Canam⃝ r
Steel catalogue Stud (edge beams) 3.50 per stud
and the steel deck thickness considered was fixed as: 0.76, 0.91,
or 1.21 mm (18, 20 or 22 gauge). Each type of steel deck has 6
Table 3
possible slab thicknesses made of either lightweight or normal General design constraints.
density concrete. Slab thickness can be 90, 100, 115, 125, 140, 150,
Constraint
165, 190 or 200 mm, depending on the type of steel deck and
Mf
concrete slab configuration used. Moment cs1 = Mr
≤ 1.0
Furthermore, to make the algorithm usable for any design case Shear cs2 =
Vf
≤ 1.0
Vr
and to improve its performance, user options were added. They ∆L
Deflection cs3 = ∆adm
≤ 1.0
include: ap /g
Floor acceleration limit (walking) cs4 = a0 /g
≤ 1.0
• Number of beams: To reduce the number of possible combina-
tions and to have a realistic physical solution, the user may
modify the minimum and maximum number of interior beams or
(spacing) of the floor layout.
Fcost = cgirders + cedgebeams + cint .beams + cdeck
• Girder and beam depth: To reduce the number of possible
combinations of girders and beams, it is recommended to + cslab + cstuds (5)
specify limits on depth for girders and beams. This option is where mgirders is the combined mass of the two girders (GN + GS),
also helpful in the case where architectural limits have been medgebeams of the two edge beams (GW + GE), mint .beams is the com-
imposed. bined mass of interior beams, mdeck is the mass of the steel deck
• Concrete density: The user may instruct the algorithm to allow and mslab is the mass of the concrete slab. For the second objec-
for the use of only normal density concrete (2400 kg/m3 ), only tive function, c represents the cost of the floor components where
lightweight concrete (1840 kg/m3 ) or both. cstuds is the cost of all the studs used on the girders and beams of
• Stud location: For composite design, this option allows the user a composite floor only. The cost of each component is obtained by
to indicate if studs are to be welded to only the girders, only multiplying the mass with the prices given in Table 2. These prices
the beams or to both. The default setting assumes that they are were obtained from local engineering firms. Two PSO codes were
welded to all girders and beams. developed, one for non-composite design and another considering
• Type of composite construction: An option for shoring to support composite action.
the steel deck during the pouring and curing of the concrete The objective function is subjected to the constraints repre-
slab is given. Because the construction load is usually critical sented in Table 3 for all components of the non-composite and
when designing composite beams and girders, this could lead composite floors.
to reduced steel cross-sections. In constraint cs4 , a0 /g is the acceleration limit for human
• Floor bay location: For vibration consideration, parameters are comfort for vibrations due to human activities, taken as 0.5%, and
set by the user for the floor bay in consideration, an interior ap /g is the peak acceleration which is estimated by Eq. (6) [20]:
panel or edge panel and if there are additional bays along its
length or its width. ap P0 e−0.35fn
= (6)
Moreover, certain assumptions that are common in the industry g βW
have been made: where ap is the acceleration of the floor due to resonance with
• It is assumed that the girder and its carrying interior beams are harmonic loading, in m/s2 , g is the acceleration due to gravity, in
simply supported members without applied end moments; m/s2 , P0 = 0.29 kN is the amplitude of the force representing the
• The steel deck is designed for at least double span condition and excitation, fn is the natural frequency of the floor in Hz, β is the
can be shored were required. Steel decking ribs are parallel to modal damping ratio of the floor system taken as 0.05 and W is
the girders and perpendicular to the beams; the effective mass of the floor system.
• Deflection control for beams and girders is applied for live loads In order to incorporate constraints into the PSO algorithm, the
only for non-composite floors. Construction deflection criteria objective function is evaluated (fitness) for all the particles and
are also given for composite floors. then the constraints are calculated one-by-one for each component
of the floor configuration to see if they are violated. If none of
3.2. Optimization problem the constraints are violated, the fitness value of the solution is not
modified. However, when one or some of the constraint fails, the
The objective is to minimize the mass or the cost of the fitness value is multiplied for each failed constraint by a penalty
floor system whose components satisfy all S16 standards require- factor, thus making this solution less attractive. There are three
ments [19]. The fitness or objective functions to be optimized are possible penalty factors, one for resistance constraints (pfr), one for
defined as deflection constraints (pfd) and one for vibration constraints (pf v ).
Because constraints cs1 , cs2 and cs3 apply to six different floor
Fmass = mgirders + medgebeams + mint .beams components (5 beams and 1 deck), and cs4 is only checked once,
+ mdeck + mslab (4) there are actually 19 total constraints to verify. If a floor component
G. Poitras et al. / Journal of Constructional Steel Research 67 (2011) 1225–1231 1229

Table 4
Design parameters used for the corner bay non composite steel floor.
Bay size Dead load Live load Additional dead load Additional live load Live load used for vibration criteria
(mm) (kPa) (kPa) (kN/m) (kN/m) (kPa)

L = 6000 WDs = 1.6 WL = 4.8 wD = 10 (north) wL = 14 (north) wL = 0.5


W = 8000 +deck wD = 6 (south) wL = 4 (west)
+concrete wD = 4 (west)
+steel wD = 6 (east)

Table 5
PSO parameters used for the corner bay non composite steel floor.
Population Maximum c1 and c2 Penalty Linear inertial weight
size generation factor

100–1000 5000–30 000 0.25–2.5 1.2–10 w1 = 1.0


w2 = 0.05–0.4

fails only one constraint, its fitness value is multiplied by the


value of the corresponding penalty factor. In the worst case, if all
constraints would fail, the fitness value would be multiplied by
12 × pfr + 6 × pfd + 1 × pf v . For composite design, constraints
must also be checked under construction loads and for additional
loads to be considered for deflection limits.
Fig. 4. Convergence rates of the best global mass and the average mass (all
4. Numerical examples particles) for the 6 m × 8 m corner non-composite floor configuration.

To ensure the PSO algorithm performs well for any steel floor
system configuration, it is essential to find the PSO parameters
(w , c1 , c2 , pf ) for which the same optimal solution is constantly
obtained.
The preceding PSO algorithm was applied on three steel floor
bays. All design criteria used satisfy all requirements of the
Canadian building code, CSA S16 Limit States Design of Steel
Structures standard and the Steel Design Guide Series 11 for
floor vibrations control [20]. The steel beams are taken from the
Canadian Handbook of Steel Construction and steel decks are
taken from CANAM⃝ r
fabricator catalogues. The PSO algorithm was
written in Matlab and all test cases presented in this section were
executed on a Pentium 3.2 GHz computer. The three examples are,
• a corner bay non-composite 6 m × 8 m floor;
• an interior bay 10 m × 8 m composite or non-composite floor; Fig. 5. Convergence rates of the best global cost and the average cost (all particles)
• an interior bay 12 m × 10 m composite or non-composite floor. for the 6 m × 8 m corner non-composite floor configuration.

The first example was used to test the capability of the


seemed to work best with the combination of w1 = 0.6 and
PSO algorithm to find the best corner bay floor configuration
w2 = 0.05. The linear c1 and c2 parameters used varied from 2 to
that includes these restraints: 200–650 mm girders (2) and
0.5. Other possible configurations were obtained that were close
beams (3) with two to nine spaces, six different thickness of to the preceding configuration by no more than 50 kg. Fig. 4 shows
normal density concrete, nine type of steel decks, which gives the convergence rates of the best global mass and the average mass
1615 (girders and beams) × 9 (spaces) × 6 (concrete thickness) × (all particles) of this floor configuration.
9 (steel deck) = 52.6 × 1012 different possible configurations. The However, when the cost objective function is used, the best
design parameters used for this example are shown in Table 4. solution for the same example is a steel floor system consisting of
The additional dead and live loads take into account the load a W530 × 74 north girder, a W460 × 60 south girder, a W310 × 28
attributed to the adjacent bays and exterior walls. west beam, a W310 × 21 east beam, 3-W310 × 28 interior beams,
The calculations were carried out with different PSO parameters PC-3615 × 0.76 mm steel deck and a 90 mm concrete slab. The total
in order to find which ones give the best and more consistent cost of this floor configuration is $7195. The convergence rates for
results. The limits considered for these parameter studies are this configuration are shown in Fig. 5. A population size of 200 was
shown in Table 5. required in order to have consistent results. Since the cost of the
When using the mass objective function, the best solution for concrete slab is proportionally less important in the overall cost of
this example is a total mass of 9868 kg consisting of a steel floor the floor than the steel deck, it is expected that a thicker concrete
system having a W530 × 74 north girder, a W410 × 60 south floor combined with a lighter deck would be more economical.
girder, a W310 × 28 west edge beams, a W310 × 24 east edge However, when the mass objective function is used, the algorithm
beam, 3-W310 × 28 interior beams, PC-3615 × 0.91 mm steel will favor a thinner concrete slab combined with a thicker steel
deck and a 90 mm concrete slab. These results were constant for deck.
each run when using a population size of 200, 10 000 generations The second application consists of an interior 10 m × 8 m bay
and linear penalty coefficients of 2–1.2. The linear inertial weight subject to the same component restraints and PSO parameters
1230 G. Poitras et al. / Journal of Constructional Steel Research 67 (2011) 1225–1231

Table 6
Design parameters used for the 10 m × 8 m interior bay non composite steel floor.
Bay size Dead load Live load Additional dead load Additional live load Live load used for vibration criteria
(mm) (kPa) (kPa) (kN/m) (kN/m) (kPa)

L = 10 000 WDs = 1.6 WL = 4.8 wD = 10 (north) wL = 14 (north) wL = 0.5


W = 8000 +deck wD = 10 (south) wL = 14 (south)
+concrete wD = 4 (west) wL = 4 (west)
+steel wD = 4 (east) wL = 4 (east)

Table 7
Summary of results for the 10 m × 8 m steel floor configuration.
Composite action (%) Girders Edge beams Interior beams Concrete floor Cost ($)

1 0 W610 × 91 W460 × 60 2-W530 × 66 PC-3615 × 0.91 14 832


100 mm
2 50 W530 × 82 W460 × 60 2-W460 × 60 PC-2432 × 0.76 15 359
37-19 × 114 stud 31-19 × 114 stud 37-19 × 114 stud 165 mm
3 60 W530 × 82 W460 × 60 2-W460 × 60 PC-2432 × 0.76 15 224
40-19 × 114 stud 31-19 × 114 stud 37-19 × 114 stud 150 mm
4 75 W530 × 82 W460 × 60 2-W460 × 60 PC-2432 × 0.76 15 211
46-19 × 114 stud 34-19 × 114 stud 40-19 × 114 stud 140 mm

Table 8
Design parameters used for the 12 m × 10 m interior bay composite steel floor.
Bay size Dead load Live load Additional dead load Additional live load Live load used for vibration criteria
(mm) (kPa) (kPa) (kN/m) (kN/m) (kPa)

L = 12 000 WDs = 1.6 WL = 4.8 wD = 12 (north) wL = 20 (north) wL = 0.5


W = 10 000 + deck wD = 12 (south) wL = 20 (south)
+concrete wD = 4 (west) wL = 4 (west)
+steel wD = 4 (east) wL = 4 (east)

Table 9
Summary of results for the 12 m × 10 m steel floor configuration.
Composite action (%) Girders Edge beams Interior beams Concrete floor Cost ($)

1 0 W610 × 155 W530 × 74 3-W610 × 82 PC-3615 × 0.91 27 426


90 mm
2 50 W610 × 153 W610 × 82 2-W610 × 82 PC-2432 × 0.91 27 956
54-19 × 114 stud 46-19 × 114 stud 54-19 × 114 stud 190 mm
3 60 W610 × 140 W610 × 82 2-W610 × 82 PC-2432 × 0.91 27 445
64-19 × 114 stud 55-19 × 114 stud 64-19 × 114 stud 190 mm
4 75 W610 × 140 W610 × 82 2-W610 × 82 PC-2432 × 0.91 27 815
80-19 × 114 stud 69-19 × 114 stud 80-19 × 114 stud 190 mm

as the previous Example 1. The design parameters used for this


example are shown in Table 6.
When the mass objective function is used, the best solution for
this example is a total mass of 17 535 kg consisting of a steel floor
system having 2-W530 × 92 girders, 2-W460 × 60 edge beams,
2-W530 × 66 interior beams, a PC-3615 × 1.21 mm steel deck and
a 90 mm concrete slab. These results were constant for each run
when using at least 200 for the population size, 10 000 generations
and linear penalty coefficients of 2–1.2. The linear inertial weights
used are the combination of w1 = 1.0 and w2 = 0.05 and the
linear c1 and c2 parameters used are 2–0.5.
When the cost objective function is used, the most economical
floor configuration consists of 2-W610 × 91 girders, 2-W460 × 60
edge beams, 2-W530 × 66 interior beams, a PC-3615 × 0.91 mm
steel deck and a 100 mm concrete slab. The cost of this floor system Fig. 6. Convergence rates of the best global cost and the average cost (all particles)
is $14 832. The convergence of the algorithm is shown in Fig. 6. A for the interior non-composite 10 m × 8 m floor configuration.
composite floor was also considered, with the results summarized
in Table 7. amount of composite action considered for this case was 0%, 50%,
The third application consists of an 12 m × 10 m interior bay 60% and 75%.
subject to the same component restraints and PSO parameters The results of these four floor systems are presented in Table 9.
as the previous examples. The design parameters used for this The lowest cost is obtained with a floor system with no composite
example are shown in Table 8. The stud size was limited to 19 mm action. However, the cost of a composite floor system is only $19
in diameter and only class 1 and class 2 beams were selected. The more.
G. Poitras et al. / Journal of Constructional Steel Research 67 (2011) 1225–1231 1231

5. Conclusion References

For this work, a PSO algorithm was used for the design of com- [1] Li L, Huang Z, Liu F, Wu Q. A heuristic particle swarm optimizer for
optimization of pin connected structures. Computers & Structures 2007;85:
posite and non-composite steel floor systems. The optimization 340–9.
problem was to minimize the objective function representing the [2] Li LJ, Huang ZB, Liu F. A heuristic particle swarm optimization method for truss
mass or the cost of a steel floor configuration. The PSO algorithm structures with discrete variables. Computers & Structures 2009;87(7–8):
435–43.
handled the floor components specific design constraints by multi-
[3] Lee K-Y, Roh M-I, Jeong H-S. An improved genetic algorithm for multi-floor
plying the component’s mass by a user-defined penalty factor each facility layout problems having inner structure walls and passages. Computers
time a constraint was not satisfied. and Operations Research 2005;32:879–99.
The performance of this PSO algorithm was tested for three [4] Perez RE, Behdinan K. Particle swarm approach for structural design
optimization. Computers & Structures 2007;85:1579–88.
steel floor design configurations. For all test cases, the optimum [5] Kaveh A, Talatahari S. Particle swarm optimizer, ant colony strategy and
design was consistently found when suitable PSO parameters were harmony search scheme hybridized for optimization of truss structures.
used. A faster convergence rate depends heavily on the appropriate Computers & Structures 2009;87(5–6):267–83.
choice of these parameters. All components chosen by the PSO [6] Kaveh A, Talatahari S. A particle swarm ant colony optimization for truss
structures with discrete variables. Journal of Constructional Steel Research
algorithm always satisfied all design constraints, regardless of the 2009;65(8–9):1558–68.
PSO parameters used. [7] Schutte JF, Groenwold AA. Sizing design of truss structures using particle
The PSO algorithm consistently found the optimal mass or cost swarm. Structural and Multidisciplinary Optimization 2003;25(4):261–9.
[8] Fourie PC, Groenwold AA. The particle swarm optimization algorithm in size
of all floor components depending on which objective function and shape optimization. Structural and Multidisciplinary Optimization 2002;
used. With the cost objective function, it was found that composite 23(4):259–67.
action can be as economical as non-composite floors depending on [9] Nelson E. The progression of the structural engineer. In: Modern steel
the bay size and of the cost of the studs, because of the high cost of construction. American Institute of Steel Construction, Inc.; 2008. p. 93–5.
[10] Foley CM, Lucas WK. Optimal selection and design of composite steel
the studs. floor systems considering vibration. In: Proceedings of the 2004 structures
The results also showed that a small population can be used conference: building of the past, securing the future. 2004. p. 1445–59.
when dealing with an optimization problem having a large number [11] Platt BS, Mtenga PV. Parametric optimization of steel floor system cost using
evolver. Computer Aided Optimum Design in Engineering X 2007;119–28.
of combinations. For this work, having to deal with over a billion
[12] Hassan R, Cohanim B, de Weck O, Venter G. A comparison of particle swarm
combinations, a population of 200 particles is sufficient to find optimization and the genetic algorithm. In: 1st AIAA multidisciplinary design
the optimum floor configuration. The number of iterations and optimization specialist conference, no. AIAA-2005-1897. 2005.
convergence rate can be improved by making an initial run with [13] Kennedy J, Eberhart RC. Swarm optimization. San Francisco: Morgan
Kaufmann Publishers; 2001.
a large number of combinations (eg. all beam sizes are considered) [14] He S, Wu Q, Saunders J, Paton R. A particle swarm optimizer with passive
and then, once the preceding results are known, limiting the congregation. BioSystems 2004;78:135–47.
number of components available (eg. minimum and maximum [15] Kennedy J, Eberhart RC. Particle swarm optimization. In: Proc. of the IEEE int.
conf. on neural networks, IV. 1995. p. 1942–48.
size of beams available). Further studies are being conducted to
[16] Shi Y, Eberhart RC. Empirical study of particle swarm optimization. Proc. of the
improve the convergence rate and to apply this method to building IEEE int. congress on evolutionary computation, Vol. 3. 1999. p. 101–6.
bracing systems. [17] Murray TM. Building floor vibrations. Engineering Journal, AISC 1991;28(3):
102–9.
[18] Handbook of steel construction. 10th ed. Canadian Institute of Steel
Acknowledgements Construction; 2010.
[19] CSA (Ed.). CAN/CSA S16-09: limit states design of steel structures. Mississauga
(Ontario, Canada): Canadian Standards Association; 2003.
The authors would like to thank the Natural Sciences and
[20] Murray TM, Allen DE, Ungar EE. AISC steel design guide series 11 floor
Engineering Research Council (NSERC) of Canada, for the financial vibrations due to human activity. Chicago (IL): American Institute of Steel
support. Construction; 1997.

Вам также может понравиться