Journal of Building Engineering 31 (2020) 101379

Contents lists available at ScienceDirect

Journal of Building Engineering

journal homepage: http://www.elsevier.com/locate/jobe

Optimum location of outrigger in tall buildings using finite element analysis

and gradient-based optimization method
Han-Soo Kim *, You-Jin Lim , Hye-Lym Lee
Department of Architecture, Konkuk University, Seoul, Republic of Korea

A R T I C L E I N F O A B S T R A C T

Keywords: In this study, the optimum locations of outriggers minimizing the top drift of tall buildings are identified using a
Outrigger gradient-based nonlinear programming approach. Since the proposed optimization method uses finite element
Tall building analysis to evaluate the objective function for arbitrarily configured buildings, the discontinuity due to design
Optimization
variables (such as outrigger locations) is overcome by piecewise linear and quadratic interpolation functions. A
Lateral drift
Interpolation
series of optimum designs for three analysis models with different vertical profiles were performed to investigate
the relation between outrigger stiffness and optimum location by changing the cross-sectional area of outriggers.
As outrigger cross-sectional area increases, the optimum outrigger location moves down in the structure. The
design result demonstrates that correlation between outrigger stiffness in a practical range and its optimum
location is negligible; thus, for design purposes, variables related to outrigger optimum location and its stiffness
can be dealt with separately.

1. Introduction of tall building structures through the analysis of the structural perfor­
mance of outrigger-braced frame-core structures when the structures are
Outriggers are the most common system for controlling the lateral subjected to uniformly distributed and triangularly distributed loads
drift of high-rise buildings. Outriggers are horizontally oriented trusses, over the building height. Hoenderkamp et al. [7–9] analyzed the
beams, or wall members that connect a core to an external column, to outrigger in several structural types by considering the racking shear
reduce lateral displacement, and increase lateral stiffness. When lateral stiffness of the braced frame and outriggers, through modifying the
loads are applied, the outriggers reduce the overturning moment equations initially proposed by Smith and Salim [4]. Lee et al. [10]
developed in the core shear wall, and transmit the reduced moment to developed an analytical model, which considers the shear deformation
the outer columns. Because of these advantages, outrigger systems have of the core wall and the flexural deformation of the perimeter frame. Lee
come into widespread use in high-rise buildings since the 1980s [1]. and Tovar [11] presented a design method for optimal outrigger
Among the various studies carried out on the outrigger system, the placement using topology optimization. Most of the aforementioned
optimum location of outriggers for controlling the lateral drift of tall studies are based on maximum top drift. However, many design codes
buildings has been most active. Approximate methods were proposed by and engineering practices target control of the inter-story drift, such that
several authors. Taranath [2,3] studied the optimum location of Zhou et al. [12] studied the optimum locations of outriggers using
outrigger for a simplified model with outriggers, which were assumed to analytical method based on inter-story drift. Brunesi et al. (2016) pre­
have infinite rigidity to simplify the solution of the problem. However, sented the seismic response of high-rise mega-frame buildings with
the stiffness of outrigger influences the total drift and the core moment, outriggers and belt trusses by investigating inter-story drift, acceleration
as well as the optimum location of the outriggers for the maximum peak profiles, shear and bending moments demands. Zhou and Li [13]
reduction in drift. Smith et al. [4,5] proposed analytical equations for studied the seismic response of a high-rise steel structure with viscous
determining the optimum location of outriggers considering the stiffness damped outriggers and concluded that under minor earthquakes, ac­
of the outriggers. They analyzed the model having one to four out­ celerations, inter-story drifts and base shear force of the structure with
riggers, and presented the optimum location of outriggers as a graph damped outriggers were larger than those with fixed outriggers. Xing
based on the dimensionless coefficient [6]. studied the optimal designs et al. [14] presented the optimal vertical combination of outriggers with

* Corresponding author. Department of Architecture, Konkuk University, 120 Neungdong-ro, Gwangjin-gu, Seoul, 05029, Republic of Korea.
E-mail address: [email protected] (H.-S. Kim).

https://doi.org/10.1016/j.jobe.2020.101379
Received 14 August 2019; Received in revised form 24 March 2020; Accepted 25 March 2020
Available online 6 April 2020
2352-7102/© 2020 Elsevier Ltd. All rights reserved.
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

viscous dampers and buckling-restrained braces to control the seismic

responses of super tall model structures by examining all possible
scenarios.
Most early studies on the optimum location of outrigger have used a
constant cross-section for every column and core wall throughout the
building’s height. However, practical structures differ significantly from
structures with a uniform core and columns at each level. To reflect the
varying stiffness of core and columns by height, Kamgar and Rahgozar
[15] studied the optimum location of outriggers by using an energy
criterion for models having constant, non-uniform cross-section of shear
core subjected to uniformly distributed or triangularly distributed loads.
To simplify the solution of the problem, the shear core cross-section of
the non-uniform model is assumed as linearly varying along its height.
Park et al. [16] proposed an optimal design method for minimizing the
volume of core wall, outrigger, and external columns and calculating the
optimal locations of outriggers using a genetic algorithm. Chen et al.
[17] used a genetic algorithm for analyzing the effect of 8 parameters on
the optimum location of outrigger for models with the uniform and
non-uniform core and peripheral columns by using approximate equa­
tions. Chen and Zhang [18] applied the multi-objective genetic algo­
rithm to analyze the effect of outrigger numbers and locations in
outrigger-braced structures by using approximate equations.
The optimization on the locations of outriggers carried out in the
early studies have typically used analytic equations derived from the
simple assumptions on structural configuration and behavior. Finite
element analysis (FEA), rather than simple analytic equations, should be
used to reflect the various structural characteristics of tall buildings.
However, in order to use FEA for the optimization of the outrigger lo­
cations, it is necessary to overcome the discrete characteristics of the
design variables because the outrigger locations should be represented
by integers. In this study, piecewise interpolation methods were pro­
posed to solve the problem due to the discrete design variables.
The genetic algorithms, which do not require gradient information in
the search process can be used to determine the optimal locations of
outriggers. However, if the search space gets broader because of many
variables, such as the number of outriggers, these optimization methods
combined with FEA might not be efficient, requiring enormous times for
model evaluations [19]. The above mentioned studies with the genetic
algorithms used analytic equations not FEA. According to a literature
survey conducted by the authors, no paper on the optimal location of
outriggers has used FEA to evaluate the gradient of objective or con­
strained functions.
For this reason, this paper presents a gradient-based nonlinear
optimization program to determine the optimum location of outriggers
to control the lateral drift in an arbitrary shaped tall building by using
FEA and interpolation functions. By using this method, the relationship
between optimum locations and stiffness of outriggers is investigated for
three analysis models having multiple outriggers. If the degree of
coupling between these two important factors, which determines the
lateral drift at the top of the building is weak, the location and stiffness
of the outriggers can be determined separately.

2. Analysis of the outrigger system

2.1. Analysis model for finite element analysis

To demonstrate the effect of outriggers in a tall building by FEA, the

analysis models of a tall building structure with a reinforced concrete
core shear wall and an outrigger that span from the wall to the exterior
columns are analyzed. Fig. 1 shows the structure has a total 80 floors,
and the total height (H) is 280 m. The outrigger truss consists of diagonal
and horizontal members connecting the central core and perimeter
columns, and is installed over four floors. The moduli of elasticity of
Fig. 1. Analysis models and outrigger configuration. concrete and steel are (23 and 210) GPa, respectively. The Poisson’s
ratio for the core wall is 0.25. For FEA, which is performed by an in-
house program developed by the first author [20,21], the outrigger

2
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Table 1 the cross-sectional area. Con-Sec and Con-Str do not represent realistic
Vertical members in three analysis models. tall building structures but they are analyzed to examine the extreme
Analysis model Member Floor level Section size Width � depth cases in the vertical profile. The lateral load assumes that a uniformly
distributed wind load of 1.0 kPa is applied at each floor. The cross-
(m) pffiffiffi
sectional area of the diagonal member is 0.12 m2 and it is 2 times
Gen Column 1F–20F 1.5 � 1.5 the sectional area of the horizontal outrigger member so that the axial
21F–40F 1.5 � 1.2 stresses at the truss members are equal.
41F–60F 1.5 � 1.0
61F–80F 1.0 � 1.0
Shear wall 1F–30F 1.0 � 15.0 2.2. The effect of outrigger on lateral drift and base moment
31F–50F 0.8 � 15.0
51F–70F 0.7 � 15.0 By using FEA, the effect of outrigger on the lateral drift and the base
71F–80F 0.6 � 15.0 moment of the tall building described in the previous section can be
demonstrated as shown in Figs. 3–9. To begin with, the analysis is
Con-Sec Column 1F–80F 1.5 � 1.2
conducted for three models with an outrigger that is placed at the 40th
Shear wall 1F–80F 0.8 � 15.0
floor, and without an outrigger. The location of outrigger was deter­
Con-Str Column 1F 1.755 � 1.755 mined by the rule of thumb [3]. The bending moment of core wall was
2F–79F linearly varying calculated by multiplying the nodal forces of plane stress elements and
80F 0.196 � 0.196 the distance from the center of the core wall. Fig. 3 demonstrates that the
Shear wall 1F 1.205 � 15.0 maximum lateral drift is greatly reduced for all three models with an
2F–79F linearly varying outrigger. Fig. 4 shows that the bending moments in the core wall at the
80F 0.015 � 15.0
lower half are also significantly reduced for all three models with an
outrigger. It can be noted that the bending moment in the upper half is
not affected by the outrigger.
Figs. 5 and 6 indicate the relationship between the outrigger
sectional area, and the top drift of building and the bending moment at
the base of core wall, respectively. The analysis is executed for 50 values
of cross-sectional area of horizontal member of outrigger from (zero to
0.12) m2 when the outrigger is located at the 40th floor. Fig. 5 dem­
onstrates that if the outrigger sectional area increases, the maximum top
drift monotonically decreases. However, if the outrigger sectional area is
greater than a certain value, the reduction of top drift appears to be
negligible for all three models. Fig. 6 shows that the tendency of the base
moment is similar to that of the top drift.
To determine the relationship between outrigger location and lateral
displacement of the building and the base moment, finite element
analysis is conducted for the outrigger location of the 4th to 80th floor.
Fig. 7 shows that the maximum drift monotonically decreases until the
outrigger reaches its optimum location, and increases afterward; thus,
the overall shape of the graph looks like a parabolic function for every
model. The optimum location for minimizing the lateral displacement is
around the 45th floor. Fig. 8 shows that the location of outrigger for
minimizing the bending moment at the base of the core wall is different
from the location for the lateral displacement and is around the 15th
floor. Figs. 7 and 8 demonstrate that the two design objectives, which
Fig. 2. Schematic profiles of the three analysis models. are the lateral displacement and the base moment, cannot be met
simultaneously without compromise. In this study, the lateral
trusses are modelled with truss elements, and the perimeter columns are displacement is chosen as the primary design objective.
represented with beam elements. The core shear wall is modelled with Fig. 9 shows the correlation between maximum drift and outrigger
plane stress elements with incompatible mode [22]. The floor beams location of the Con-Sec model for several values of outrigger cross-
that connect the core shear wall and the perimeter columns are also sectional area. The optimum outrigger location is the location when
modelled with truss elements with sectional area of 0.03 m2 to represent the maximum lateral displacement is at its lowest value. If the outrigger
the shear connected floor steel beams. The outrigger trusses are pin sectional area gets smaller, the stiffness of the outrigger significantly
connected to the core wall and the perimeter columns. The base of affects the maximum top drift and optimum outrigger location, and if
structure is fixed to the ground. the area gets greater, the significance of the outrigger cross sectional
Three models are analyzed with different member sectional profile, area on its location and top drift becomes negligible.
as shown in Table 1 and Fig. 2. All three models are designed to have
similar top drift when they are subjected to uniform lateral load. The 3. Optimum design of the outrigger system
first model is the General (Gen) model having sectional profiles close to
the normal building, such that the four cross-sections of core and column 3.1. Formulation of the optimization problem
alternatively change along the building’s height. The second model is
the Constant-Section (Con-Sec) model having uniform core and column As described above, the maximum lateral displacement at the top of a
cross-section for overall height. The third model is the Constant-Stress tall building with outriggers depends on several factors. The location of
(Con-Str) model with a cross-section that is designed to receive nearly outriggers significantly affects the performance of the outrigger system.
constant axial stress on the core and the external columns from the The optimum location of outriggers can be determined using a mathe­
gravity loads. Therefore, the lower the floor of the building, the greater matical optimization method. However, a gradient-based optimization

3
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Fig. 3. Effect of outrigger installation at 40th floor on lateral displacement.

4th floor, respectively. The lateral displacement of the building with

integer design variables yi is calculated through FEA.

3.2. Piecewise interpolation

The problem given in Eq. (1) is a boundary constrained nonlinear

programming with integer variables, and it cannot be directly solved by
the gradient-based optimization methods, owing to the integrality
requirement. To minimize the objective function of the problem, the
function needs to be differentiable with respect to y. To obtain the
differentiable polynomial function gð~
yÞ, the relaxation can be done by an
interpolation method. Then, the integer nonlinear programming given
in Eq. (1) can be substituted with the following piecewise continuous
nonlinear programming with the interpolated polynomial function with
real variables.
Minimize gð~ yÞ
y 2 Rn
subject ​ to ~ (2)
Fig. 4. Effect of outrigger installation at 40th floor on bending moment in the yi min � ~yi � yimax i ¼ 1; :::; n
core wall.
where, ~ y is the interpolated variable vector of y, y~i is the interpolated
method, such as the steepest descent method [19], cannot be applied variable of yi , and Ris the set of real numbers.
without modification, because when FEA is used to evaluate the objec­ The piecewise quadratic interpolation (PQI, hereafter) method to
tive function, the location of the outrigger should be represented by an obtain gð~yÞ is proposed by Kim in the previous study [20]. In the PQI
integer. In this study, piecewise linear interpolation (PLI) and piecewise method, the integer design variable yi was obtained by rounding ~ yi to the
quadratic interpolation (PQI) were proposed to overcome the barrier nearest integer, as given in Eq. (5). The relaxation occurs using three
due to the discrete design variables. consecutive integer values and a quadratic polynomial function by
The optimization problem for the optimum location of outriggers can coordinating three values, as given in Eq. (4). The interpolation function
be formulated in the following standard form: can be seen as an extension of the univariate Lagrange interpolation
polynomial to the multivariate function. The relaxed objective function
Minimize gðyÞ gð~
yÞ is the arithmetic mean of each interpolation function gð~ yi Þ, as given
subject ​ to y 2 Zn (1) in Eq. (3). The original objective function gðyÞ is replaced with the
yimin � yi � yi max i ¼ 1; :::; n interpolated function gð~ yÞ. Fig. 10 shows the concept of the PQI with two
integer variables, y1 and y2 .
where, gðyÞ is the objective function representing the maximum value of
lateral displacement; y is a vector of the integer variables, and its 1X n
gð~
yÞ ¼ gð~
yi Þ (3)
element yi represents the location of the i-th outrigger; Z is the set of n i¼1
integers; and n is the number of outriggers.
The design variables are denoted with y rather than x, to emphasize
that the design variable should be an integer. The integer design variable
yi should not be greater than the highest floor yimax and not less than the
lowest feasible floor yimin , which in the analysis models are the 80th and

4
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Fig. 5. Effect of sectional area of outrigger installed at 40th floor on maximum lateral displacement.

Fig. 6. Effect of sectional area of outrigger installed at 40th floor on bending moment at the base of core wall.

Fig. 8. Effect of outrigger location on bending moment at the base of core wall.
Fig. 7. Effect of outrigger location on maximum lateral displacement.

5
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

To improve the efficiency of the optimization procedure, a simpler

technique of interpolation can be used. In this study, piecewise linear
interpolation (PLI, hereafter) was proposed. In PLI, gð~yi Þ given in Eq. (4),
and yi given in Eq. (5) are substituted by Eqs. (6) and (7), respectively. It
is worth noting that ceil function, rather than round function, is used to
get continuous interpolation functions in PLI.
gð~
yi Þ ¼ gðy y i Þðyi
b ~yi Þ gðyÞðyi 1 ~yi Þ (6)

yi ¼ d~yi e ¼ ceilð ~yi Þ (7)

Fig. 11 demonstrates the concept of PLI with two integer variables, y1
and y2 . As shown in Figs. 10 and 11 and Eqs. (4) and (6), n þ 1 times
FEAs are needed to evaluate gð~ yÞ for every design state in PLI, whereas
2n þ 1 times FEAs are needed in PQI. Meanwhile, the integer require­
Fig. 9. Lateral displacement of the Con-Sec model with varying outrigger cross- ment in gðyÞ can be utilized to reduce the computing time, by storing and
sectional area. reusing the objective function values with integer design variables.
Because of the arithmetic mean given in Eq. (3), the optimization
method that requires a Hessian matrix cannot be used. The steepest
descent method, which requires only a gradient vector, was used in this
study. The gradient of gð~yÞ in PLI can be obtained as Eqs. (8) and (9) by
differentiating Eqs. (3) and (6).
∂ gð~yÞ 1 ∂ gð~yi Þ
¼ (8)
∂ ~yi n ∂ ~yi

∂ gð~yi Þ
¼ gðy y i Þ þ gðyÞ
b (9)
∂ ~yi

3.3. Optimality conditions

The PLI function given in Eq. (6) is discontinuous and not differen­
tiable at y, and it is almost impossible for the gradient of the linear
interpolation function to yield zero. Therefore, the necessary conditions
Fig. 10. Piecewise quadratic interpolation (PQI) with two integer variables. for optimization with PLI should be modified as follows.
� �
g ~ yðk 1Þ
yðkÞ � g ~ (10)

1Þ
∂ gð~yðkÞ Þ ∂ gð~yðk Þ
⋅ 1Þ
�0 for i ¼ 1; 2; ⋯; n (11)
∂ ~yi ðkÞ ∂ ~yi ðk
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X n
ð1Þ 2
(12)
ðkÞ
ð~yi ~yi Þ < 1:0
i¼1

where, ~ yðkÞ is the current state of the relaxed design variable during the
steepest descent search.
Eq. (10) means that the steepest descent search keeps going down,
until the objective function does not yield lower value. The condition in
Eq. (11) substitutes the conventional optimality condition of rgðy* Þ ¼
0, and Eq. (12) is used to ensure the movement of the current iteration is
less than 1. When the conditions given in Eqs. (10)–(12) are met, the
Fig. 11. Piecewise linear interpolation (PLI) with two integer variables.
search is terminated, and the optimum ~ yi * is given as follows.
� �
y*i ¼ ~ (13)
ðkÞ ðk 1Þ
1 ~ yi þ ~ yi 2
gð~
yi Þ ¼ gðy y i Þðyi
b ~yi Þðyi þ 1 ~yi Þ
2
The design state given in Eq. (13) is the optimum solution of the
gðyÞðyi 1 ~yi Þðyi þ 1 ~yi Þþ (4)
relaxed problem, not the initially given problem with integer variables.
1
gðy þ b
y i Þðyi 1 ~yi Þðyi ~yi Þ The solution for the integer optimization problem may be obtained by
2 finding the nearest integers from the solution for the relaxed problem.
However, the solution with nearest integers may not be the correct
yi ¼ k~yi k ¼ roundð ~yi Þ (5)
optimum solution of the integer problem, because of the incomplete
where, ~yi is the same vector as y except the ith component, which is piecewise interpolation, and the nature of the integer nonlinear opti­
replaced with the relaxed variable ~yi ; gð~
yi Þ is the interpolated function by mization problem [23]. To supplement this incompleteness, a
branch-and-test module was added to find the correct optimum solution.
the Lagrange quadratic polynomial with successive three integers about
The branch-and-test module generates the neighboring design alterna­
the ith variable; and b
y i is the unit vector of the ith direction, such as
tives from the solution of the relaxed problem, and then tests the
f0; 1; 0; 0gT when i ¼ 2 and n ¼ 4.
objective function to find the correct optimum solution, which should be

6
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Fig. 12. Flow chart of steepest decent method with piecewise interpolation.

Table 2
Comparison of PLI and PQI.
Analysis model # of OR PLI PQI

Optimum location Max. lat. disp. # of FEA Optimum location Max. lat. disp. # of FEA

Gen 1 47 0.6505 14 47 0.6505 16

2 27,57 0.4336 22 27,57 0.4336 31
3 23,43,64 0.3557 58 23,43,64 0.3557 88
4 17,31,45,64 0.3252 110 16,28,44,64 0.3246 193
5 14,24,38,51,67 0.3043 184 14,24,38,51,67 0.3043 245

Con-Sec 1 47 0.7174 14 47 0.7174 16

2 28,57 0.4848 25 28,57 0.4848 31
3 20,39,62 0.4083 46 20,39,62 0.4083 56
4 16,30,46,65 0.3718 77 16,30,46,65 0.3718 126
5 13,25,37,51,68 0.3509 169 13,24,36,50,67 0.3509 256

Con-Str 1 50 0.7190 22 50 0.7190 22

2 33,61 0.4627 42 33,61 0.4627 63
3 25,45,66 0.3750 66 25,45,66 0.3750 122
4 20,36,52,69 0.3330 138 20,36,52,69 0.3330 199
5 16,29,42,56,71 0.3091 307 16,29,42,56,71 0.3091 338

pffiffiffi
represented by integers. During this study, it was noted that the correct proportioned to 2 times that of the horizontal members. Three initial
optimum can be found by moving up or down a group of outriggers. locations (IL-1, IL-2, IL-3) of the outriggers were applied, to verify the
Applying this heuristic branching, 2 � fn þðn 1Þ þðn 2Þ þ… þ1g ¼ robustness of the proposed optimization method.
nðn þ1Þ alternatives are generated and tested for the correct optimum of
the integer problem. IL 1: yi ¼ 4 þ i i ¼ 1; 2; :::; n (14)
Fig. 12 shows the flow chart of the search algorithm developed in this
IL 2: yi ¼ 79 nþi i ¼ 1; 2; :::; n (15)
study to find the optimum location of outrigger. The left side represents
the steepest descent search with a special algorithm for escaping the
IL 3: yi ¼ k 80 � i=ðn þ 1Þ k i ¼ 1; 2; :::; n (16)
oscillation that can be caused by rounding the design variables as given
in Eqs. (5) and (7). IL-1 and IL-2 place the outriggers at the lower and upper part,
respectively. IL-3 distributes the outrigger evenly along the height of
4. Optimum design of numerical examples building. It is worth noting that IL-3 is the recommended location by the
rule of thumb. The reliability of the proposed optimization method was
The optimization method proposed in Section 3 was applied to the verified by comparing the optimum locations of one and two outriggers
numerical examples described in Section 2. As mentioned above, the with the results by exhaustive search method, which evaluated 76 and
sectional area of the inclined members in the outrigger was 5700 cases, respectively.

7
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Table 3
Maximum lateral displacements subject to uniformly distributed lateral load before and after optimization started from three different initial locations.
Analysis model # of OR Optimum disp. Initial Location-1 Initial Location-2 Initial Location-3

Initial disp. Ratio Initial disp. Ratio Initial disp. Ratio

Gen 1 0.6505 2.0070 0.324 1.0799 0.602 0.6849 0.950

2 0.4336 1.8734 0.231 1.0432 0.416 0.4396 0.986
3 0.3557 1.7777 0.200 1.0186 0.349 0.3594 0.990
4 0.3246 1.6951 0.192 0.9958 0.326 0.3268 0.993
5 0.3043 1.6181 0.188 0.9731 0.313 0.3058 0.995

Con-Sec 1 0.7174 2.3598 0.304 1.1507 0.623 0.7475 0.960

2 0.4848 2.1979 0.221 1.1093 0.437 0.4903 0.989
3 0.4083 2.0827 0.196 1.0841 0.377 0.4097 0.997
4 0.3718 1.9824 0.188 1.0620 0.350 0.3731 0.997
5 0.3509 1.8883 0.186 1.0405 0.337 0.3524 0.996

Con-Str 1 0.7190 2.2049 0.326 1.5475 0.465 0.7932 0.906

2 0.4627 2.0860 0.222 1.4588 0.317 0.4931 0.938
3 0.3750 1.9987 0.188 1.3869 0.270 0.3875 0.968
4 0.3331 1.9224 0.173 1.3241 0.252 0.3399 0.980
5 0.3091 1.8505 0.167 1.2657 0.244 0.3148 0.982

Average 0.220 0.379 0.975

Table 4
Maximum lateral displacements subject to triangularly distributed lateral load before and after optimization started from three different initial locations.
Analysis model # of OR Optimum disp. Initial Location-1 Initial Location-2 Initial Location-3

Initial disp. Ratio Initial disp. Ratio Initial disp. Ratio

Gen 1 0.7088 2.2574 0.314 1.1582 0.612 0.7708 0.920

2 0.4764 2.1206 0.225 1.1164 0.427 0.4907 0.971
3 0.3905 2.0218 0.193 1.0884 0.359 0.3984 0.980
4 0.3595 1.9359 0.186 1.0626 0.338 0.3617 0.994
5 0.3376 1.8552 0.182 1.0369 0.326 0.3390 0.996

Con-Sec 1 0.7741 2.6408 0.293 1.2237 0.633 0.8317 0.931

2 0.5267 2.4749 0.213 1.1770 0.448 0.5405 0.975
3 0.4459 2.3558 0.189 1.1486 0.388 0.4493 0.992
4 0.4075 2.2513 0.181 1.1237 0.363 0.4090 0.996
5 0.3855 2.1525 0.179 1.0996 0.351 0.3869 0.996

Con-Str 1 0.7991 2.5179 0.317 1.7037 0.469 0.9239 0.865

2 0.5147 2.3963 0.215 1.6002 0.322 0.5693 0.904
3 0.4179 2.3062 0.181 1.5164 0.276 0.4422 0.945
4 0.3719 2.2267 0.167 1.4435 0.258 0.3856 0.965
5 0.3458 2.1514 0.161 1.3757 0.251 0.3566 0.970

Average 0.213 0.388 0.960

4.1. Comparison of PQI and PLI distributed load is adjusted to 112.5 kN, so that the overturning moment
at the bottom is the same as the moment by the uniformly distributed
Table 2 compares the optimum designs started at IL-3 with PLI and load. The maximum displacements at optimum state are shown once in
PQI. It can be observed that 2 cases (Gen-4OR and Con-Sec-5OR) among Tables 3 and 4, in which the objective function values after optimization
15 cases yield different locations. # of OR means the number of installed started from IL-1, IL-2, and IL-3 are the same. Therefore, it can be
outriggers and # of FEA is the number of FEAs performed while confirmed that the optimization method proposed in this study does not
searching the optimum solutions. In Gen-4OR case, PQI gives 0.2% less depend on the starting points. Table 3 shows that the average reduced
lateral displacement than PLI. In Con-Sec-5OR case, the lateral ratios of IL-1, IL-2, and IL-3 subjected to the uniform lateral loads are
displacement is identical up to 4 digits even though the locations are (22.0, 37.9, and 97.5) %, respectively. It can be noted that the lower
slightly different. It can be assumed when the analysis models have large floors represented by IL-1 are the worst location for the outriggers, and
discontinuity such as the Gen model, which has changes of sectional area the upper floors of IL-2 are not appropriate locations, either. In contrast,
at every 10 floors, linear interpolation is less adequate than quadratic the outrigger locations of IL-3, in which outriggers are located evenly
interpolation in finding optimum locations. However, the total number along the building height, are very effective locations for reducing the
of FEA performed with PLI is 71.8% (1294/1802) of PQI. Even though maximum lateral displacement. It is rather surprising that the average
PLI requires fewer number of FEA executions, PQI was chosen for the difference in maximum lateral displacement of IL-3 and the optimum
next studies, considering the quality of the optimum results. locations is only 2.5%. A similar tendency can be observed in the results
of models subjected to the triangularly distributed loads, as shown in
Table 4. The Con-Str model with one outrigger only shows rather large
4.2. Optimum designs started from different initial locations discrepancy with (9.4 and 13.5) % for the uniformly and triangularly
distributed loads, respectively.
Tables 3 and 4 show the maximum lateral displacements of the three
analysis models started from three different initial locations subjected to
uniformly distributed lateral load and triangularly distributed lateral
load, respectively. The lateral load at the top for the triangularly

8
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Fig. 13. Effect of outrigger stiffness on the optimum outrigger locationfor

Fig. 16. Effect of outrigger rigidity on optimum outrigger locationfor models
models with one outrigger.
with four outriggers.

4.3. Effect of outrigger stiffness on optimum locations

A series of optimum designs were performed, changing the cross-

sectional area of the outrigger from (0.006–0.24) m2 through the pro­
posed optimization method, searching for the optimum locations of
outriggers by interpolating the analysis results of FEA. The cross-
sectional areas of the outrigger were normalized with the reference
area of 0.12 m2. The initial locations were set at the IL-3, and the uni­
formly distributed loads were applied. For the range of outrigger
sectional area in this study, the results given by two interpolation
methods, PQI and PLI, were almost the same. For each analysis model,
the optimum location of each outrigger was derived, and Fig. 13–16
show the results by PQI for the analysis models with one, two, three, and
four outriggers, respectively. The marked solid lines represent the op­
timum locations of the outriggers to minimize the lateral displacement
at the top and the corresponding vertical axis is shown on the left side of
graph. It is noted that the locations of the outrigger should be defined in
integers. The maximum lateral displacements as a function of the
outrigger area are also presented in the figures with unmarked solid
Fig. 14. Effect of outrigger rigidity on optimum outrigger locationfor models lines. The axis for the lateral displacement is on the right side of the
with two outriggers.
graph. The allowable lateral displacement at the top of building, which
is assumed as 0.56 m, 1/500 of the total height of building [16,18], is
also indicated by dashed line to compare with the maximum lateral
displacement of each case.
The common features of all analysis results are summarized as fol­
lows. As the outrigger cross-sectional area increases, the optimum
location of the outrigger moves down in the structure. Regardless of the
number and sectional area of outriggers, the optimum location for the
Con-Str model is the highest. In the range of small cross-sectional area of
the outrigger, the optimum location drastically changes; however, in the
range of large cross-sectional area of the outrigger, the optimum loca­
tion changes insignificantly. The figures show that the latter is the
practical range of outrigger sectional area that enables the building to
satisfy the allowable maximum displacement. For this reason, it can be
confirmed that within the practical range of the outrigger stiffness, the
change of the optimum locations of outriggers is insignificant. There­
fore, it is possible to separate the stiffness, and the location for opti­
mizing the outriggers that minimize the lateral displacement of the top
of the building.
When the number of outriggers is 1 or 2, the optimum location of the
Con-Sec model and the Gen model tend to be similar. When the number
Fig. 15. Effect of outrigger rigidity on optimum outrigger locationfor models of outriggers is 3, the optimum location of the Con-Sec model is the
with three outriggers. lowest, followed by the Gen model, and then the Con-Str model. In the

9
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Fig. 17. Effect of number of outriggers of Con-Sec model on the maximum lateral displacement.

case of three and four outriggers, the optimum location of the highest numerical examples. The results using PQI and PLI for models having
outrigger barely changes. In the case of two or more outriggers, the one, two, three, and four outriggers were identical, except for 2 cases out
optimum location of each outrigger may be similar in some cases, but of the 15 cases, and the difference of the 2 cases in objective function
there is no clear tendency overall. values was negligible. By comparing the optimum locations started at
three different starting points, it is confirmed that the proposed opti­
4.4. Effect of the number of outriggers mization method yields an identical solution, regardless of the initial
locations of outriggers. Furthermore, it is noted that the initial locations
Fig. 17 shows the effect of the number of outriggers on the maximum given by the rule of thumb shows lateral displacements that are 97.5% of
lateral displacement. For comparison, the analysis is classified as the optimized values, except for the Con-Str model with one outrigger.
Constant-Area-Scheme and Divided-Area-Scheme for the Con-Sec model A series of optimum designs were performed to investigate the
subjected to the uniformly distributed load and the triangularly relation between the outrigger stiffness and the optimum locations by
distributed load. In the Constant-Area-Scheme, the outriggers of the changing the cross-sectional area of outriggers. As the outrigger cross-
model have a constant cross-sectional area of 0.12 m2; whereas in the sectional area increases, the optimum location of the outrigger moves
Divided-Area-Scheme, the given section area of outrigger is divided by down in the structure. Within the practical range of the outrigger cross-
the number of outriggers. Therefore, as the number of outriggers in­ sectional area, the change of the optimum location of the outrigger is
creases, the total area of outrigger in the Constant-Area-Scheme in­ within 2 stories. The result shows that it is possible to separate design
creases. As the number of outriggers increases, that in the Divided-Area- variables contributing to the outrigger stiffness and the optimum loca­
Scheme remains the same. Fig. 17 indicates that if the number of out­ tion for optimizing the outriggers that satisfy the design constraints,
riggers increases, the maximum drift of the structure decreases. How­ such as the allowable lateral displacement.
ever, if the number of outrigger is greater than 3 for the Divided-Area- It is expected that the proposed optimization method using PQI or
Scheme and 5 for the Constant-Area-Scheme, the effect of the number PLI can be applied to the elastoplastic problem such as seismic pushover
of outriggers on the maximum drift is almost negligible. Also, it can be analysis because it shows satisfactory results even for the analysis model
observed that the models show a similar tendency in decreasing lateral with sudden changes like the Gen model.
displacement as the number of outriggers increases, even though the
maximum lateral displacement of the model subjected to the triangular Funding information
load is greater than that of the uniform load.
National Research Foundation of Korea, Grant/Award Number: NRF-
5. Conclusions 2017R1A2B4010043.

In this study, the optimum locations of outriggers minimizing the top Declaration of competing interest
drift of the building are identified using a gradient-based nonlinear
programming. Since the proposed optimization method uses FEA to The authors declare that they have no known competing financial
evaluate the objective function for arbitrary configured buildings, the interests or personal relationships that could have appeared to influence
discontinuity due to the integer design variables, such as outrigger lo­ the work reported in this paper.
cations, is overcome by piecewise interpolation functions. Due to the
uncoupling among the design variables, the steepest descent method, CRediT authorship contribution statement
which does not require Hessian matrix, is used to find the optimum
solution. The gradients of objective function are obtained through Han-Soo Kim: Conceptualization, Methodology, Software, Re­
piecewise quadratic interpolation and linear interpolation from the sources, Writing - review & editing, Supervision. You-Jin Lim: Formal
discrete objective function values. analysis, Investigation, Data curation, Writing - original draft, Visuali­
Three analysis models having different cross-sectional profiles of zation. Hye-Lym Lee: Formal analysis, Investigation, Data curation,
vertical members with multiple outriggers were investigated as Writing - original draft, Visualization.

10
H.-S. Kim et al. Journal of Building Engineering 31 (2020) 101379

Acknowledgements shear core, belt truss and outrigger system using energy method, Int J Steel Struct.
17 (1) (2017) 1–8.
[9] H.S. Kim, Effect of horizontal members on column shortening of reinforced
This work was supported by the National Research Foundation of concrete building structures, Struct Des Tall Spec. 22 (5) (2013) 440–453.
Korea (NRF), with a grant funded by the Korea government (MSIP) [10] H.S. Kim, Optimum design of outriggers in a tall building by alternating nonlinear
(NRF-2017R1A2B4010043). programming, Eng. Struct. 150 (2017) 91–97.
[11] J. Lee, M. Bang, J. Kim, An analytical model for high-rise wall-frame structures
with outriggers, Struct Des Tall Spec. 17 (2) (2007) 839–851.
Appendix A. Supplementary data [12] S.B. Lee, A. Tovar, Outrigger placement in tall buildings using topology
optimization, Eng. Struct. 74 (2014) 122–129.
[13] H.S. Park, E. Lee, S.W. Choi, B.K. Oh, T. Cho, Y. Kim, Genetic-algorithm-based
Supplementary data to this article can be found online at https://doi. minimum weight design of an outrigger system for high-rise buildings, Eng. Struct.
org/10.1016/j.jobe.2020.101379. 117 (2016) 496–505.
[14] A.R. Parkinson, R.J. Balling, J.D. Hedengren, Optimization Methods for
Engineering Design, Brigham Young University, 2013.
References [15] B.S. Smith, I. Salim, Parameter study of outrigger braced tall building structures,
J. Struct. Div. 107 (10) (1981) 2001–2014.
[1] Y. Chen, K. Cai, X. Wang, Parameter study of framed-tube structures with [16] B.S. Smith, A. Coull, Tall Building Structures: Analysis and Design, John Wiley &
outriggers using genetic algorithm, Struct Des Tall Spec. 27 (14) (2017) e1499. Sons, New York, 1991.
[2] Y. Chen, Z. Zhang, Analysis of outrigger numbers and locations in outrigger braced [17] B.S. Taranath, Optimum belt truss location for high-rise structures, Struct. Eng. 53
structures using a multi-objective genetic algorithm, Struct Des Tall Spec. 27 (1) (8) (1975) 345–347.
(2018) e1408. [18] B.S. Taranath, Structural Analysis and Design of Tall Buildings, McGraw-Hill, New
[3] H.S. Choi, G. Ho, L. Joseph, N. Mathias, Outrigger Design for High-rise Buildings, York, 1988.
Council on Tall Buildings and Urban Habitat, Chicago, IL, 2012. [19] R.L. Taylor, P.J. Beresford, E.L. Wilson, A non-conforming elements for stress
[4] I.E. Grossmann, Review of nonlinear mixed-integer and disjunctive programming analysis, Int. J. Numer. Methods Eng. 10 (6) (1976) 1211–1219.
techniques, Opt. Eng. 3 (2002) 227–252. [20] J.R. Wu, Q.S. Li, Structural performance of multi-outrigger-braced tall buildings,
[5] J.C.D. Hoenderkamp, M.C.M. Bakker, Analysis of high-rise braced frames with Struct Des Tall Spec 12 (2) (2003) 155–176.
outriggers, Struct Des Tall Spec. 12 (4) (2003) 335–350. [21] L. Xing, Y. Zhou, M. Aguaguina, Optimal vertical configuration of combined energy
[6] J.C.D. Hoenderkamp, Shear wall with outrigger trusses on wall and column dissipation outriggers, Struct Des Tall Spec. 28 (2019) e1579.
foundations, Struct Des Tall Spec. 13 (1) (2004) 73–87. [22] Y. Zhou, H. Li, Analysis of a high-rise steel structure with viscous damped
[7] J.C.D. Hoenderkamp, Second outrigger at optimum location on high-rise shear outriggers, Struct Des Tall Spec. 23 (13) (2014) 963–979.
wall, Struct Des Tall Spec 17 (3) (2008) 619–634. [23] Y. Zhou, C. Zhang, X. Lu, An inter-story drift-based parameter analysis of the
[8] R. Kamgar, R. Rahgozar, Determination of optimum location for flexible outrigger optimal location of outriggers in tall buildings, Struct Des Tall Spec. 25 (5) (2016)
systems in tall buildings with constant cross section consisting of framed tube, 215–231.

11