You are on page 1of 10

Expert Systems With Applications 169 (2021) 114531

Contents lists available at ScienceDirect

Expert Systems With Applications


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

An efficient differential-evolution-based moving compensation


optimization approach for controlling differential column shortening in
tall buildings
Hoang-Anh Pham a, *, Duc-Xuan Nguyen b, Viet-Hung Truong c, *
a
National University of Civil Engineering, 55 GiaiPhong Road, Hanoi, Viet Nam
b
Vinh University, 182 Le Duan Road, Vinh, Viet Nam
c
Faculty of Civil Engineering, Thuyloi University, 175 Tay Son Street, Dong Da, Hanoi, Viet Nam

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

Keywords: In this paper, an efficient procedure integrating the moving compensation optimization (MCO) and a modified
Tall building differential evolution algorithm is presented for finding the optimal solution to compensate for the differential
Differential column shortening column shortening (DCS) in tall buildings. In the proposed compensation procedure, the number of compensa­
Moving compensation optimization
tion groups is minimized by maximizing the number of floors for each group stepwise with constraints placed on
Differential evolution
the compensation error at each floor level. Two optimal compensation problems are presented, including the
deterministic optimal compensation (DOC) and the reliability-based optimal compensation (ROC), which
correspond to ignoring or considering the uncertainties that are inherent in the predicted shortenings as well as
the correction amounts. A parameter-free, adaptive differential evolution algorithm is established to solve MCO.
Applications for a 70-story building and a 72-story building are examined to demonstrate the efficiency and
reliability of the presented compensation approach.

1. Introduction of tall buildings has been attracting significant interest in the literature.
For instance, Kim and Shin (2014a, 2014b), Kim (2017), Kim (2018),
Column shortening occurs in every structure, and it is particularly and Kim et al. (2019, 2020) proposed the outrigger system to reduce
troublesome in tall buildings because of the large cumulative axial DCS. The effect of the outrigger-belt system (OBS) on DCS in composite
deformation of vertical members like columns and walls due to huge tall buildings was also explored by Samarakkody et al. (2017), where it
vertical loading (Fintel et al., 1987). The excessive shortening in vertical was concluded that DCS can be minimized by suitable location of OBS.
members can cause either of both problems in the serviceability and For reinforced concrete (RC) tall buildings, Kim (2015), Kim and Shin
safety of the structure. Accordingly, the control of axial shortening in (2014a, 2012, 2014b) developed efficient methods by means of addi­
vertical members becomes an essential factor to secure the structural tional reinforcement to increase axial stiffness of vertical members
safety and serviceability of tall buildings. having a large magnitude of shortening.
In a tall building, vertical members can shorten by different amounts To evade issues related to DCS, DCS should be accurately predicted
because of the dissimilarity in loading conditions as well as design pa­ and properly compensated for (Fintel et al., 1987, Cargnino et al., 2012,
rameters like the reinforcement ration and the volume-surface ratio. The Fan et al., 2013). In a simple manner, the correction is conducted on
differential shortenings among vertical members, or differential column some chosen floors by the “lumped” amount that is equal to the sum of
shortenings (DCS), are of primary concern since they cause additional differential shortenings of a group of floors. This method, referred to as
stress in the vertical members and additional forces in connecting the lumped compensation method (LCM), cannot remove all the
girders and slabs (Yi and Tong, 2007). Besides, slab tilt due to DCS may different shortenings except the different shortening at the corrected
cause unfavorable effects on nonstructural parts like the interior parti­ level. The errors between the compensation amount and the predicted
tion, cladding system, and plumbing system (Carreira and Poulos, amount become larger as the number of floors lumped in the group in­
2007). Considering this, controlling DCS in the design and construction creases. The lumped correction amount can be averaged over the floors

* Corresponding authors.
E-mail addresses: anhph2@nuce.edu.vn (H.-A. Pham), truongviethung@tlu.edu.vn (V.-H. Truong).

https://doi.org/10.1016/j.eswa.2020.114531
Received 21 August 2020; Received in revised form 18 November 2020; Accepted 19 December 2020
Available online 24 December 2020
0957-4174/© 2020 Elsevier Ltd. All rights reserved.
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

to avoid excessive compensation errors. It means that the correction is 1997) is established to tackle the MCO problem. The proposed MCO
conducted on every floor by the same averaged amount. Obviously, the procedure is examined with two tall buildings, including a 70-story
number of floors in a group will determine the efficiency of the building in Fintel et al. (1987) and a practical 72-story building. The
compensation task. Increasing the number of floors will reduce the first example serves to show the advantage of MCO compared to OC and
number of compensation groups, and ease the compensation task; MAC in solving the DOC problem. The other example demonstrates the
however, it may increase the compensation error. Thus, a trade-off be­ applicability of MCO in the ROC problem, where uncertainties are of
tween construction simplicity and compensation efficiency needs to be concern.
taken into consideration when searching for a compensation solution.
Few systematic methodologies were established for searching the 2. Moving compensation optimization
compensation solution to control DCS, including the optimal compen­
sation approach (OC) (Park, 2003, Kim, 2008, Kim, 2011) and the 2.1. Formulation of problem objective
moving averaging correction approach (MAC) (Park et al., 2010, Park,
2003). In OC, the compensation is formulated as a constrained optimi­ Consider a tall building having NF floors. In order to compensate for
zation problem, and the objective of the compensation strategy is to the differential column shortenings, the floors are gathered into groups,
minimize the number of compensation groups (Park, 2003). However, and an equal correction amount is applied for every floor in a group
there are issues associated with the OC method proposed in Park (2003), (Park, 2003). Fig. 1 describes this grouping correction strategy, where xij
which are: 1) the optimization problem may have an unfixed number of is the anticipated differential shortening on the j-th floor in the i-th
variables since the number of lumped groups is unknown; 2) the group, and δi is the corresponding correction amount for the floors in the
objective function is often multi-modal, i.e., there are several possible i-th group.
compensation solutions. Thus, OC requires special optimization tech­ The proposed compensation strategy is to obtain a minimum number
niques to solve. In Park (2003), a parallel-simulated annealing (SA) al­ of groups to ease the construction task as suggested in Nguyen and Pham
gorithm (Leite and Topping, 1999, Park and Won Sung, 2002) was (2020). It is accomplished by maximizing the number of floors (Ni )
utilized for the optimal compensation of a 70-story building. In contrast,
gathered in each compensation group. Besides, the value of δi should be
MAC proposed in Park et al., 2010 and Park (2003) does not require to
chosen such that the magnitude of the sum compensation error is
solve an optimization problem. MAC is also an LCM using averaged
minimized.
correction values. The determination of compensation solution is per­
As aforementioned, there exists a difference between the compen­
formed stepwise by a rather simple computation routine. At each step,
sation amount δi and the anticipated differential shortening xij on each
MAC maximizes the number of floors for a lumped group by trials, where
the number of floors is increased until a constraint on the accumulated floor, which is determined as εij = xij − δi . The compensation error of the
compensation error at floor level is violated. In this way, MAC provides a i-th group is defined as
unique solution; however, the solution is not optimal under the given
Ni
∑ Ni (
∑ ) ∑Ni
constraint. A recently developed compensation method (Nguyen and ei = εij = xij − δi = xij − N i δi (1)
Pham, 2020), named MOC, takes advantage of both OC and MAC. j=1 j=1 j=1
Overall, a reliable compensation solution relies on the preciseness of
Hence, the sum compensation error over all floors up to the i-th
the anticipated DCS. However, the amount of column shortenings, both
group is determined by
elastic and inelastic, depends upon various factors such as the material
characteristics, loading history, and construction sequence, etc. The ∑ ∑i− Ni

(2)
1
evaluation of DCS is therefore complicated, especially for modern ei = k=1
ek + xji − N i δi
complex tall builds. Though numerous procedures have been developed j=1

to predict the column shortening, especially for RC tall buildings (e.g.,


procedures suggested in Fintel et al., 1987, Fan et al., 2013, American
Concrete Institute, 1992, Elnimeiri and Patel, 1997, Jeong and Kim
2004, Kim and Shin, 2011, Kim et al., 2012a, Kim et al., 2012b, Mor­
agaspitiya et al., 2010, Samarakkody et al., 2017), the analysis of DCS
still lacks accuracy compared to the on-site monitoring data due to the
inaccurate prediction of time-dependent deformation such as creep and
shrinkage (Zou et al., 2019). Various uncertainties are associated with
the column shortening of a building (Song and Yoon, 2006). For
example, numerical and experimental studies show that the variation of
ambient humidity has a significant influence on shrinkage and creep
(Zou et al., 2014, Wang et al., 2020, Gao et al., 2020). On the other hand,
the applied compensation amounts in practice may not be the same as
the computed values due to round-off and construction errors. The
existing compensation methods like OC, MAC, and MOC only deal with
deterministic values of predicted DCS as well as correction amount.
There is a dearth of work considering uncertainties for the compensation
of DCS.
This paper proposes an efficient procedure for obtaining the optimal
compensation of DCS in tall buildings. Firstly, the moving compensation
optimization (MCO) is introduced, in which the optimal compensation
solution is derived stepwise. The optimization objective is to minimize
the number of compensation groups while maintaining an allowable
compensation error at each floor level. Two optimal compensation
problems are presented, including the deterministic optimal compen­
sation (DOC) and the reliability-based optimal compensation (ROC). Fig. 1. Illustration of the grouping correction strategy.
Then, an optimizer based on differential evolution (DE) (Storn and Price,

2
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

∑ ⃒ ⃒
where ei is the sum compensation error; ek is the compensation error ⃒ ⃒
gj = ⃒dji ⃒ − θi (7b)
∑ k ( )
of the k-th group below the i-th group, i.e., ek = Nj=1 xkj − δk .
Therefore, the objective function of the compensation strategy is where Pj is the predefined safety probability for the slab tilt at the j-th
formulated in the following form: floor level; gj is the j-th constraint function.
( ) ⃒∑ ⃒
⃒ ⃒
Minimizef N i , δi = − N i + wi × ⃒ ei ⃒ (3) 3. Handling probabilistic constraints using first-order reliability
method
where wi (wi > 0) is a weighted factor. With the defined objective
function, the number of floors gathered in the i-th group is maximized, In this study, for simplicity, differential column shortenings, as well
while the sum compensation error is maintained under control. as correction amounts, are assumed as normally distributed and inde­
pendent uncertainties. Therefore, gj is also normally distributed. Using
2.2. Constraints on slab tilt the first-order reliability method (FORM), Eq. (7a) can be transformed
into the following forms:
As mentioned earlier, differential shortening can cause unwanted
slopes to floors. Therefore, constraints on the compensation error at gj + βj σgj ≤ 0, j = 1, ⋯, N i (8)
floor level need to be placed to control the slab tilt caused by DCS. The
where gj and σgj are the mean and standard deviation of gj , respec­
followings describe these constraints.
The compensation error at the j-th floor level of the i-th group, or the tively; βj is the corresponding reliability index. The value of βj is
( )
accumulated compensation error (ACE), is the cumulative difference calculated as βj = Φ− 1 Pj , where Φ− 1 is the inverse of the standard
between the correction values and the anticipated differential shorten­ normal density.
ings. This error is determined by Notable that gj is a linear function of the anticipated differential
j
shortenings and the correction amounts. Thus, we have
∑i− ∑ ( )
(4)
1 ⃒
dji = ek + xil − δi , j = 1, ⋯, N i ⃒∑i− 1 ∑j ( ) ⃒⃒
k=1 ⃒ i ⃒
l=1 gj = ⃒ k=1 ek + xil − δ ⃒ − θi (9)
⃒ l=1

where dij is the ACE at the j-th floor level of the i-th group. As suggested
( )1/2
by Park et al. (2013), the compensation error at a floor level should be ∑
i− 1 j (
∑ )
restricted by a tolerance to control the slab tilt, i.e. σ gj = σ 2ek + σ 2xi + σ2δi (10)
l
k=1 l=1
⃒ ⃒
⃒ i⃒
⃒dj ⃒ ≤ θi , j = 1, ⋯, N i (5) In Eq. (9), x and δ denote the mean values of the anticipated differ­
ential shortening and the correction amount, respectively; σ denotes
where θi is a predefined tolerance for compensation error. standard deviation; ek and σ 2ek are determined by
By adopting the constraints of Eq. (5), the optimal compensation
solution, including the floors for each compensation group (Ni ) together Nk (
∑ )
(11)
k
ek = xkj − δ
with the corresponding compensation value (δi ), will be determined by j=1
solving a series of constrained optimization problems stepwise, from
lower floors to upper floors. The procedure of this so-called moving Nk (
∑ )
compensation optimization (MCO) is: σ 2ek = σ 2xk + σ2δk (12)
j
j=1

• Obtain the differential shortening profile of vertical elements and set With predefined reliability indices βj , the optimal compensation so­
the error limit, θi ;
lution will be determined by solving a sequence of the following reli­
• At the i-th step (i starts from 1), solve the constrained optimization
ability constrained optimization problems:
problem, with the objective function defined in Eq. (3) and the
⃒ ⃒
constraints determined by Eq. (5), to obtain the optimal values Ni ( i)
⃒∑i− 1
⃒ ∑Ni ⃒
i⃒
and δi , for the i-th group; Minimize f N i , δ = − N i + wi × ⃒ k=1 ek + xil − N i δ ⃒ (13)
⃒ ⃒
• Set i = i + 1, and move to the next group.
j=1

s. t.
It is important to note that, at each step, the value of Ni cannot
cj ≤ 0, j = 1, ⋯, N i (14a)
exceed the number of the remaining floors, Nimax , that are not to be
grouped. It means that Ni must satisfy Eq. (6) ⃒
⃒∑i− 1 j ( ) ⃒⃒ (
j (
)
) 1/2
⃒ ∑ i ⃒

i− 1 ∑
i 2 2 2
∑i− 1 cj = ⃒ k=1 ek + xl − δ ⃒ + βj σ ek + σ xi + σ δi − θi ,
1 ≤ N i ≤ Nmax
i
= NF − k=1
Nk (6) ⃒ l=1
⃒ k=1 l=1
l

(14b)
2.3. Probabilistic constraints
where ek and σ 2ek are determined by Eq. (11) and Eq. (12), respectively. It
As aforementioned, the anticipated axial shortenings of vertical is notable that the established optimization problem has only two design
members are uncertain owing to various uncertainties associated with i
variables, Ni and δ , where Ni is an integer variable.
the input parameters used for calculating them. Moreover, the actual
correction amounts applied to the vertical members during the 4. Differential evolution-based optimizer
compensation process may not be the same as the computed values due
to round-off and construction errors. Therefore, the constraints for ACE The MCO problem presented in Section 3 possesses a multi-modal
presented in Section 2.2 are formulated in terms of probabilistic con­ objective function and non-linear constraints. Moreover, the number
ditions to account for these issues as of constraints is not predefined since the floors in the group are un­
[ ] known. Due to these reasons, MCO is best tackled by direct search
P gj ≤ 0 ≥ Pj , j = 1, ⋯, N i (7a)

3
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

optimization approaches. {
Among various direct search techniques, differential evolution (DE) cp = ( al + r∙(a)m − an ) if (rand[0, 1] ≤ s)
(17)
ap + r∙ abest − ap + r∙(am − an ) if (rand[0, 1] > s)
(Storn and Price, 1997) is one of the most popular population-based
optimization methods. Since DE has a simple structure, few control
where s is the parameter controlling the selection of mutation operator; r
parameters, and can be implemented easily, it has gained extensive
is a vector of D uniform random numbers in the interval [0,1].
applications in various structural optimization problems (Kitayama
The value of s is of vital importance to secure a balance between
et al., 2011, Das et al., 2016, Pham, 2016a, Pham, 2016b, Truong and
exploration ability and exploitation ability. Here, s is adapted with the
Kim, 2018, Anh and Duong, 2019, Ha et al., 2020). Therefore, in this
change of the population diversity during the evolution as follows.
study, DE is adopted to develop a practical optimizer for searching the
solution of the presented MCO problem. DI t
s= (18)
DI 0
4.1. Basic of DE
where DIt is a diversity index that measures the population distribution
In searching for the optimal solution of an optimization problem, DE at the t-th generation; DI0 is the diversity index of the initial population.
utilizes a population of P candidate solutions. Let ap = The diversity index is computed as (Pham, 2016a)
{ }
ap1 , ap2 , ⋯, apD ,p = 1, ⋯, P, is the p-th candidate solution (or indi­ ⎡ ⎛ ⎞2 ⎤1/2
vidual) in the population of the current generation, where ap1 , ap2 , ⋯, 1∑ ∑ ∑
P D P
a − a ⎠ ⎦ ; aj = 1 (19)
⎣ ⎝ pj j
DI t = apj
apD are the D design variables of the problem. P p=1 j=1 auj − alj P p=1
Corresponding to ap , a trial solution cp is created by the ‘mutation’
and ‘crossover’ operators as follows: where aj is the mean of the j-th design variable at the t-th generation;
alj and auj are the lower bound and the upper bound of the j-th design
• Mutation is applied to generate a mutant vector bp . Some conven­ variable, respectively. By using the proposed adjustment, larger values
tional mutation schemes are: of s are expected in the earlier generations, so that ‘DE/rand/1’ is used
more frequently to favor exploration, and smaller values of s are ob­
− DE/rand/1 : bp = al + F∙(am − an ) (15a)
tained in later generations to enhance exploitation by ‘DE/current-to-
best/1’.
− DE/best/1 : bp = abest + F∙(am − an ) (15b)
In Eq. (16), the scaling factor for creating each component of the
( )
- DE/current-to-best/1: bp = ap + F∙ abest − ap + F∙(am − an )(15c) mutant vector is generated randomly instead of a constant value pre­
where al , am and an are three randomly selected individuals in the defined as in the classical DE. Random scaling factors enable both
population; abest is the best solutions of the current population; F is a exploitation (by small value) and exploration (by large value)
positive real number, whose value is often predefined in the interval [0, throughout the search process. Moreover, the ‘crossover’ step is omitted
1]. in producing a trial solution. The proposed DE is, therefore, a parameter-
free algorithm, which may ease its application by structural engineers.
• Crossover is conducted to produce the trial solution cp as:
{ 4.3. Integer variable and bound constraint handling
bpj if(rand[0, 1] ≤ Cr )orIr = j
cpj = , j = 1, ⋯, D (16)
apj otherwise
The mutation operators in Eq. (17) generally produce rational
where Ir is a random integer in the interval [1, D]; Cr is a predefined numbers. Integer value of the variable Ni is obtained simply by the
crossover rate with its value in the interval [0,1]. rounding technique. Moreover, if the generated value of a design vari­
Then, cp is compared with ap , and the better one will become a new able violates its bound constraint, it will be set equal to the violated
candidate in the population of the next generation. The updated popu­ bound.
lation, therefore, consists of only good candidate solutions. The search
process stops when a termination criterion is satisfied. In this study, the 4.4. Comparison of solutions with constraints
searching process is terminated if the generations exceed a predefined
maximum number of generation Tmax , or if the objective value cannot be The presented compensation optimization problem consists Ni con­
improved in twenty subsequent generations. straints

cj ≤ 0, j = 1, ⋯, N i (20)
4.2. Modification in mutation
where cj is the j-th constraint function defined by Eq. (14b). In this study,
It is well known that DE’s efficiency is largely dependent on the
the measure of constraint violation of a solution is calculated as
mutation strategy. For instance, the ‘DE/rand/1’ scheme favors explo­
{ }
ration; however, it has weak exploitation ability and low convergence { }
C = max max 0, cj , j = 1, ⋯, N i (21)
speed. By contrast, the ‘DE/best/1’ or ‘DE/current-to-best/1’ scheme j
favors exploitation with fast convergence speed, but they can lose
Each of the candidate solutions need to be assessed in the selection
population diversity quickly and may be trapped in a local optimum in
process by its corresponding objective value and constraint violation.
multi-modal optimization problems. To achieve a good trade-off be­
Deb’s rules (Deb, 2000) are applied in the proposed DE to compare two
tween exploration and exploitation, using multiple mutation operators
solutions as follows:
is often recommended (Mallipeddi et al., 2011, Wang et al., 2011, Gong
et al., 2010, Takahama and Sakai, 2012, Xiang et al., 2015, Kushida
• If both solutions are feasible (C = 0) or have equal constraint
et al., 2015, Wu et al., 2016, Ho-Huu et al., 2016).
violation measure, the better solution is the one having smaller
In the present study, a simply adaptive mutation strategy is intro­
objective function value.
duced for generating new individuals. The proposed mutation strategy
• If both solutions are infeasible (C > 0), the better solution is the one
operates as
having smaller constraint violation measure.
• If one of the two solutions is feasible while the other is not, the
feasible one is the better.

4
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

5. Implementation of MCO Table 1


Control parameters for DE.
The stepwise procedure for the implementation of MCO is given in Parameter DE/rand/1 Modified DE
this section.
Population size P 50 50
Step 0: Define the compensation problem. Scaling factor F 0.8 rand[0,1]
Obtain the DCS data of the vertical elements of the building, Crossover rate Cr 0.9 N/A
including mean values and standard deviations. At the beginning, these Max. generation Tmax 100 100
data are predicted using probabilistic analysis of DCS (Song and Yoon,
2006). During the construction, accurate DCS should be updated with
best solution.
on-site monitoring data. If only the mean values are available, the
Repeat Step 3 until the number of generations exceeds Tmax , or if the
standard deviation values can be assumed based on engineer’s
smallest objective value cannot be improved in twenty subsequent
experiences.
generations.
Predefine the error limit (θi ) and the reliability index (β).
The best individual in the final population becomes the compensa­
Set the group indicator i = 1.
tion solution of the i-th group.
Step 1: Initialize the optimization parameters.
Step 4: Move to the next group.
Predefine the size of the candidate population (P), the maximum ∑i k
i Stop if no more floor remains, i.e. NF − k=1 N = 0; otherwise, in­
generation (Tmax ), and the bounds of the design variables, i.e. Nmax and
i crease i = i + 1 and repeat from Step 1.
δmax , where Nmax is the number of the remaining floors that are not to be
grouped, and δmax are the upper limit for the compensation amount.
6. Applications
Step 2: Generate the initial population.
{ }
The initial population of candidate solutions ap = Nip , δip ,p = 1,⋯, Two tall building examples, including a 70-story building in Fintel
P, is generated randomly as: et al. (1987) and a practical 72-story building in Hanoi, are considered
[ ( )] for illustrating the proposed MCO procedure. The first example serves to
Npi = round 1 + rand[0, 1] Nimax − 1 (22) show the advantage of MCO in solving the DOC problem. The other
example demonstrates the applicability of MCO in the ROC problem.
δip = rand[0, 1]δmax (23)

For each candidate solution ap , the corresponding objective function 6.1. DOC problem
value fap is calculated according to Eq. (13), and the corresponding
constraint violation Cap is calculated by Eq. (21). The 70-story building was formerly considered by Park (2003) and
Compute the diversity index of the initial population by Eq. (19). Park et al. (2013) to examine their compensation methods, including OC
The best candidate solution (abest ) is sought in the population by and MAC. Thus, in this investigation, MCO is performed to derive the
applying the comparison rules presented in Section 4.4. compensation solution for the DOC problem, and the results are
Step 3: Generate trial solutions. compared with those of OC and MAC. Fig. 2 illustrates half of the typical
Corresponding to each ap in the population of the current generation, layout of the building.
generate a trial solution cp by Eq. (17). Integer handling and bound The anticipated DCS after slab installation between the inner wall
constraint handling presented in Section 4.3 are applied to the generated and outer column are given in Park (2003). The predicted DCS,
trial solution. The objective value fcp and the constraint violation Ccp of including the elastic shortening and the creep and shrinkage shorten­
cp are then computed. ings, was based on the procedures proposed by Fintel et al. (1987) and
Compare cp with ap using the Deb’s rules presented in Section 4.4. an estimated construction sequence of 7 days per floor. These data are
Accept cp for the next generation if it is better than ap . assumed exact, and uncertainty is not considered. A tolerance of 0.2 (in.)
Compute the diversity index of the new population and update the for ACE is assumed (Park, 2003).
The proposed MCO procedure is applied to seek the optimal solution

Fig. 2. Half of the typical floor layout of the 70-story building (Fintel
et al. 1987).

Fig. 3. Convergence of the objective function value by DOC for the 1st group.

5
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

Table 2
Compensation solutions by OC, MAC, and DOC for the 70-story building with
compensation error limit of 0.2 (in.)
OC (Park, 2003) This study (MAC) This study (DOC)

i Ni δi (in.) i Ni δi (in.) i Ni δi (in.)

1 1 0.045 1 6 0.1293 1 8 0.1442


2 2 0.096 2 6 0.3336 2 10 0.4559
3 9 0.283 3 9 0.5484 3 8 0.7219
4 11 0.579 4 6 0.7954 4 16 1.0173
5 11 0.940 5 11 1.0207 5 7 1.0468
6 22 0.972 6 10 1.0300 6 12 0.7961
7 3 0.769 7 11 0.8319 7 6 0.4393
8 5 0.589 8 6 0.5703 8 3 0.1226
9 5 0.303 9 4 0.2607
10 1 0.021 10 1 0.0210

Fig. 5. The ACE by OC, MAC, and DOC for the 70-story building with
compensation error limit of 0.2 (in.)

Table 3
Compensation solutions by MAC and DOC for the 70-story building with
compensation error limit of 0.96 (in.)
This study (MAC) This study (DOC)

i Ni δi (in.) i Ni δi (in.)

1 14 0.2672 1 18 0.2829
2 14 0.6910 2 27 0.9276
3 26 0.9907 3 13 1.0156
4 13 0.6238 4 12 0.3591
5 3 0.1226
Fig. 4. Compensation curves by OC, MAC, and DOC for the 70-story building
with compensation error limit of 0.2 (in.)
ACE by MAC and DOC with the error limit of 0.96 (in.), and that by OC
with the error limit of 0.2 (in.). In summary, the proposed MCO pro­
cedure is beneficial in reducing the necessary groups for the compen­
sation of the building.
for the compensation of the building. Both the conventional DE using
‘DE/rand/1’ and the modified DE are utilized in this investigation. The
6.2. ROC problem
parameter setting for DE is given in Table 1. These parameter values are
chosen based on a preliminary investigation to obtain good performance
The second example is a ROC problem of the compensation of DCS
of DE. Each algorithm performs twenty independent runs to obtain
for the Lotte Center Hanoi building. This is one of the tallest RC build­
statistical results. It is found that both ‘DE/rand/1’ and the modified DE
ings in Vietnam, with a height of 285.25 m, and 72 floors (Fig. 8). The
end with the same result in all runs. Fig. 3 depicts the averaged
building was completed for hotel, residence, office, and headquarter
convergence curves of DOC performed by the two algorithms. It is
store. Fig. 9 displays the typical layout of the building given in MIDAS IT
shown that the modified DE converses faster than ‘DE/rand/1’ does. The
(2011).
time consumed by both algorithms is only some seconds on a personal
The controlling of DCS of this tall building was considered at the
computer with 2 GB RAM, and a dual-core CPU of 2.5 GHz.
design stage for the safety and serviceability of the structure. The pre­
Table 2, Figs. 4 and 5 compare the compensation solution of DOC
liminary analysis was carried out to grasp the size of future column
with that of OC using SA (Park, 2003) and MAC. From the results in
shortenings and set a management plan based on the anticipated
Table 2 and Fig. 4, it is shown that DOC can derive a fewer number of
shortenings (MIDAS IT, 2011). The shortening of the structure was
groups compared with OC and MAC. In this investigation, the number of
estimated using ACI 209 and PCA. The analysis was conducted under the
compensation groups by DOC is eight, whereas that by OC and MAC is
assumption of various environmental factors at the preliminary analysis
ten. It is seen that both MAC and DOC control the ACE at every floor
stage. However, material tests (compressive strength and elastic
level within the limit of 0.2 (in.). The largest ACE by MAC and DOC is
modulus tests, creep tests, and shrinkage tests) were conducted to obtain
0.19 (in.) and 0.20 (in.), respectively. OC, however, gives a maximum
specific data for the analysis to minimize the error.
error of 0.96 (in.) on the 47th floor, which exceeds the allowable
According to the preliminary analysis in MIDAS IT (2011), there
margin.
were excessive differential shortenings between the outer columns and
By adopting the limit of 0.96 (in.), MAC and DOC can further reduce
the inner core-wall at some locations. Therefore, compensation was
the compensation groups. As shown in Table 3 and Fig. 6, MAC and DOC
suggested to conduct during the construction period by extending the
obtain five and four compensation groups, respectively. Fig. 7 shows the
column length at some floor levels with an equal amount. The suggested

6
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

Fig. 6. Compensation curves by OC with 0.2 (in.) limit, and MAC and DOC with
0.96 (in.) limit.

Fig. 8. Perspective view of Lotte Ceter Hanoi (MIDAS IT, 2011).

probability of safety is taken as 0.99, i.e., the desirable reliability index


is 2.3263. Different uncertainty levels are considered in the analysis.
Table 5 presents the computed number of floors and the required
correction amounts by ROC corresponding to three different uncertainty
levels, with the coefficient of variation (CV) of 0.5%, 1%, and 2%. The
corresponding compensation curves are depicted in Fig. 10. The results
show that the compensation solution corresponding to the small level of
uncertainty (CV = 0.5%) is the same as that of the deterministic case.
When the CV value is larger, i.e., the level of uncertainty is higher, the
number of compensation groups increases. Fig. 11 shows the mean
values of accumulated compensation error at floor level controlled by
ROC. It is shown that when a higher uncertainty level is involved, ROC
Fig. 7. The ACE by OC with 0.2 (in.) limit, and MAC and DOC with 0.96
will maintain a lower magnitude of the difference between predicted
(in.) limit.
differential shortening and the correction amount to assure the desired
reliability.
Monte-Carlo simulations (MCS) with 100,000 samples are conducted
compensation solution is simple, but it is not optimal. Furthermore,
to verify the obtained compensation solutions. The safety probability of
there would be a possibility of changes in the column shortening
Eq. (5) is computed as:
occurring in the actual building due to many factors of environment and
materials. [⃒ ⃒ ] Ns
⃒ ⃒
P ⃒dji ⃒ ≤ θi = (22)
Considering the uncertainty of the shortening, ROC is employed to 100000
obtain the optimal compensation solution for the building. The location
to be examined is Line15, where large excessive deformation angles where Ns is the number of the samples that satisfy Eq. (5). Table 6 lists
occur, according to the preliminary analysis results in MIDAS IT (2011). the computed values of the safety probability and the corresponding
The amounts of DCS at Line15 are provided in the structural analysis reliability index. The obtained results are higher than the desired values.
report of the consultant (MIDAS IT, 2011). These data are considered as It means that the derived compensation solutions by ROC are reliable.
the mean values of the differential shortenings, and they are given in
Table 4. It is supposed that randomness is associated with both the 7. Conclusions
differential shortenings and the compensation amounts. All random
quantities are assumed normally distributed and uncorrelated. More­ A novel compensation optimization for controlling DCS, named
over, the calculated correction amounts will be rounded to millimeters. MCO, has been established in this study. By formulating the compen­
As required by the technical specification, the difference of short­ sation problem stepwise, MCO can derive the compensation solution,
ening between column C18 and the core-wall W12 of Line15 shall not including the number of compensation groups, the number of floors for
exceed an allowable value of L/240 (ca. 20 mm), i.e., θi = 20mm. The each group together with the corresponding correction amount in a

7
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

Fig. 9. Typical floor plan ofLotte Center Hanoi structure (MIDAS IT, 2011).

systematic way efficiently. Both the deterministic optimal compensation CRediT authorship contribution statement
(DOC) and the reliability-based optimal compensation (ROC) have been
considered. An efficient DE-based optimizer has been developed for Hoang-Anh Pham: Conceptualization, Methodology, Software,
MCO. The new method has been examined for the compensation of two Formal analysis, Investigation, Resources, Data curation, Writing -
tall buildings. In the case of DOC, MCO can find a smaller number of original draft, Writing - review & editing, Supervision, Project admin­
compensation groups, compared with two existing methods under the istration. Duc-Xuan Nguyen: Methodology, Software, Formal analysis,
same constraint on the slab tilt. When uncertainty is considered, i.e. Writing - original draft, Writing - review & editing. Viet-Hung Truong:
ROC, MCO can derive reliable compensation solutions. Another Software, Writing - original draft, Writing - review & editing, Formal
advantage of MCO is that it requires a very low computational cost. analysis, Visualization, Resources, Supervision.
Therefore, MCO provides a practical means to obtain the optimal solu­
tion for controlling DCS in tall buildings.

Table 4
Differential shortenings (DS) at Line15 of Lotte Center Hanoi building.
Floor DS (mm) Floor DS (mm) Floor DS (mm) Floor DS (mm) Floor DS (mm)

1 0 15 11.4 29 19.6 43 22.8 57 32.9


2 0.5 16 12.1 30 20.3 44 23.7 58 33.2
3 1.0 17 12.7 31 20.9 45 24.6 59 33.4
4 1.7 18 13.2 32 21.5 46 25.4 60 33.5
5 2.6 19 13.8 33 22.0 47 26.5 61 33.5
6 3.8 20 14.3 34 22.4 48 27.6 62 33.3
7 5.2 21 14.7 35 22.9 49 28.5 63 33.0
8 6.3 22 15.5 36 22.4 50 29.4 64 32.7
9 6.9 23 16.1 37 22.2 51 30.2 65 32.3
10 7.9 24 16.8 38 21.0 52 30.9 66 31.7
11 8.8 25 17.4 39 20.3 53 31.4 67 30.8
12 9.8 26 17.9 40 19.7 54 31.9 68 29.9
13 10.1 27 18.4 41 20.9 55 32.3 69 28.9
14 10.7 28 18.9 42 21.8 56 32.5 70 27.3

Table 5
Compensation solutions by ROC corresponding to different levels of uncertainty.
Deterministic CV = 0.5% CV = 1% CV = 2%

i Ni δi (mm) i Ni δi (mm) i Ni δi (mm) i Ni δi (mm)

1 11 4 1 11 4 1 11 4 1 11 4
2 16 14 2 16 14 2 16 14 2 12 13
3 20 22 3 20 22 3 20 22 3 18 20
4 25 31 4 25 31 4 19 32 4 9 26
5 5 28 5 17 32
6 1 24 6 3 31
7 2 24

8
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

References

American Concrete Institute. (1992). Prediction of Creep, Shrinkage, and Temperature


Effects in Concrete Structures, ACI 209 R-92. Detroit, MI: ACI.
Anh, P. H., & Duong, T. T. (2019). Weight optimisation of functionally graded beams
using modified differential evolution. Journal of Science and Technology in Civil
Engineering (STCE)-NUCE, 13(2), 48–63.
Cargnino, A., Debernardi, P. G., Guiglia, M., & Taliano, M. (2012). Axial shortening
compensation strategies in tall buildings. A case study: The new piedmont
government office tower. Structural Engineering International, 22(1), 121–129.
Carreira, D. J., & Poulos, T. D. (2007). Designing for effects of creep and shrinkage in
high-rise concrete buildings. Special Publication, 246, 107–132.
Das, S., Mullick, S. S., & Suganthan, P. N. (2016). Recent advances in differential
evolution – An updated survey. Swarm and Evolutionary Computation, 27, 1–30.
Deb, K. (2000). An efficient constraint handling method for genetic algorithms. Computer
Methods in Applied Mechanics and Engineering, 186(2-4), 311–338.
Elnimeiri, M., & Patel, D. (1997). Long-term vertical shortening of reinforced concrete
and composite high-rise structures. In Proceedings of 7th International Conference on
Computing in Civil and Building Engineering (pp. 2329–2353).
Fan, F., Wang, H., Zhi, X., Huang, G., Zhu, E., & Wang, H. (2013). Investigation of
construction vertical deformation and pre-deformation control for three super high-
rise buildings. Advances in Structural Engineering, 16(11), 1885–1897.
Fintel, M., Ghosh, S. K., & Iyengar, H. (1987). Column shortening in tall buildings-
Prediction and compensation. Publ. EB108 D. Portland Cement Association, Skokie, 3,
1–34.
Gao, F., Zhou, H., Liang, H., Weng, S., & Zhu, H. (2020). Structural deformation
monitoring and numerical simulation of a supertall building during construction
Fig. 10. Compensation curves by ROC for different levels of uncertainty. stage. Engineering Structures, 209, 110033.
Gong, W., Cai, Z., Ling, C. X., & Li, H. (2010). Enhanced differential evolution with
adaptive strategies for numerical optimization. IEEE Transactions on Systems, Man,
and Cybernetics, Part B (Cybernetics), 41(2), 397–413.
Ha, M.-H., Vu, Q.-V., & Truong, V.-H. (2020). Optimization of nonlinear inelastic steel
frames considering panel zones. Advances in Engineering Software, 142, 102771.
Ho-Huu, V., Vo-Duy, T., Luu-Van, T., Le-Anh, L., & Nguyen-Thoi, T. (2016). Optimal
design of truss structures with frequency constraints using improved differential
evolution algorithm based on an adaptive mutation scheme. Automation in
Construction, 68, 81–94.
Jeong, E.-H., & Kim, J.-H. (2004). A study on the prediction of shortening for steel-
reinforced concrete(SRC) column in the high-rise buildings. Journal of the Korea
Concrete Institute, 16(1), 36–42.
Kim, H.-S. (2015). Optimum distribution of additional reinforcement to reduce
differential column shortening: optimum distribution of additional reinforcement.
The Structural Design of Tall and Special Buildings, 24(10), 724–738.
Kim, H.-S. (2017). Effect of outriggers on differential column shortening in tall buildings.
International Journal of High-Rise Buildings, 6(1), 91–99.
Kim, H.-S. (2018). Optimum locations of outriggers in a concrete tall building to reduce
differential axial shortening. International Journal of Concrete Structures and Materials,
12(1).
Kim, H.-S., Jeong, S.-H., Shin, S.-H., & Park, J.-P. (2012a). Simplified column shortening
analysis of a multi-storey reinforced concrete frame. The Structural Design of Tall and
Special Buildings, 21(6), 405–415.
Kim, H.-S., Jeong, S.-H., & Shin, S.-H. (2012b). Column shortening analysis of tall
buildings with lumped construction sequences. The Structural Design of Tall and
Special Buildings, 21(10), 764–776.
Kim, H.-S., Lee, H.-L., & Lim, Y.-J. (2019). Multi-objective optimization of dual-purpose
outriggers in tall buildings to reduce lateral displacement and differential axial
shortening. Engineering Structures, 189, 296–308.
Kim, H.-S., Lim, Y.-J., & Lee, H.-L. (2020). Strength demand of dual-purpose outrigger
system for reducing lateral displacement and differential axial shortening in a tall
Fig. 11. Mean values of ACE by ROC for different levels of uncertainty.
building. The Structural Design of Tall and Special Buildings, 29(4).
Kim, H. S., & Shin, S. H. (2011). Column Shortening analysis with lumped construction
sequences. Procedia Engineering, 14, 1791–1798.
Kim, H. S., & Shin, S. H. (2012). Optimum distribution of reinforcement to reduce
differential column shortening. Journal of the Architectural Institute of Korea Structure
Table 6 & Construction, 28(9), 33–40.
Kim, H., & Shin, S. (2014a). Reduction of differential column shortening in tall buildings.
Reliability of the compensation solutions by MCS.
International Journal of Civil and Environmental Engineering, 8(2), 145–148.
[⃒ ⃒ ]
CV ⃒ ⃒ Kim, H., & Shin, S. (2014b). Reduction of differential column shortening by placing
P ⃒dij ⃒ ≤ θi β
additional reinforcement. Magazine of Concrete Research, 66(9), 456–464.
0.5% 0.9960 2.6529 Kim, Y. M. (2008). Optimal compensation of differential column shortening in tall
buildings for multi column groups. Journal of the Computational Structural Engineering
1% 0.9998 3.5678
Institute of Korea, 21(2), 189–197.
2% 0.9902 2.3332
Kim, Y. M. (2011). The optimal column grouping technique for the compensation of
column shortening. Journal of the Computational Structural Engineering Institute of
Korea, 24(2), 141–148.
Declaration of Competing Interest Kitayama, S., Arakawa, M., & Yamazaki, K. (2011). Differential evolution as the global
optimization technique and its application to structural optimization. Applied Soft
Computing, 11(4), 3792–3803.
The authors declare that they have no known competing financial Kushida, J. I., Hara, A., & Takahama, T. (2015). Rank-based differential evolution with
interests or personal relationships that could have appeared to influence multiple mutation strategies for large scale global optimization. In 2015 IEEE
the work reported in this paper. Congress on Evolutionary Computation (CEC) (pp. 353–360). IEEE.
Leite, J. P. B., & Topping, B. H. V. (1999). Parallel simulated annealing for structural
optimization. Computers & Structures, 73(1-5), 545–564.
Mallipeddi, R., Suganthan, P. N., Pan, Q. K., & Tasgetiren, M. F. (2011). Differential
evolution algorithm with ensemble of parameters and mutation strategies. Applied
Soft Computing, 11(2), 1679–1696.

9
H.-A. Pham et al. Expert Systems With Applications 169 (2021) 114531

MIDAS IT (2011). Lotte Center Hanoi Project: Column Shortening – Report for Storn, R., & Price, K. (1997). Differential evolution – A simple and efficient heuristic for
Preliminary Analysis (Hanoi). global optimization over continuous spaces. Journal of Global Optimization, 11(4),
Moragaspitiya, P., Thambiratnam, D., Perera, N., & Chan, T. (2010). A numerical method 341–359.
to quantify differential axial shortening in concrete buildings. Engineering Structures, Takahama, T., & Sakai, S. (2012, June). Differential evolution with dynamic strategy and
32(8), 2310–2317. parameter selection by detecting landscape modality. In 2012 IEEE Congress on
Nguyen, D. X., & Pham, H. A. (2020). Optimal Compensation of Axial Shortening in Tall Evolutionary Computation (pp. 1–8). IEEE.
Buildings by Differential Evolution. In ICSCEA 2019 (pp. 1137-1144). Springer, Truong, V.-H., & Kim, S.-E. (2018). Reliability-based design optimization of nonlinear
Singapore. inelastic trusses using improved differential evolution algorithm. Advances in
Park, H. S. (2003). Optimal compensation of differential column shortening in high-rise Engineering Software, 121, 59–74.
buildings. The Structural Design of Tall and Special Buildings, 12(1), 49–66. Wang, L., Zhao, X., & Yan, C. (2020). Time-dependent vertical shortening prediction for
Park, H. S., & Won Sung, C. (2002). Optimization of steel structures using distributed super-tall buildings by using a modified B3 model to consider moisture distribution.
simulated annealing algorithm on a cluster of personal computers. Computers & Engineering Structures, 209, 109994.
Structures, 80(14-15), 1305–1316. Wang, Y., Cai, Z., & Zhang, Q. (2011). Differential evolution with composite trial vector
Park, S. W., Choi, S. W., & Park, H. S. (2010). Average correction for compensation of generation strategies and control parameters. IEEE Transactions on Evolutionary
differential column shortening in high-rise buildings. Journal of the Computational Computation, 15(1), 55–66.
Structural Engineering Institute of Korea, 23(4), 395–401. Wu, G., Mallipeddi, R., Suganthan, P. N., Wang, R., & Chen, H. (2016). Differential
Woo Park, S., Woon Choi, S., & Seon Park, H. (2013). Moving average correction method evolution with multi-population based ensemble of mutation strategies. Information
for compensation of differential column shortenings in high-rise buildings: Moving Sciences, 329, 329–345.
average correction to compensate column shortenings. The Structural Design of Tall Xiang, W.-l., Meng, X.-L., An, M.-Q., Li, Y.-Z., & Gao, M.-X. (2015). An enhanced
and Special Buildings, 22(9), 718–728. differential evolution algorithm based on multiple mutation strategies.
Pham, A. H. (2016a). Discrete optimal sizing of truss using adaptive directional Computational Intelligence and Neuroscience, 2015, 1–15.
differential evolution. Advances in Computational Design, 1(3), 275–296. Yi, T., & Tong, X. (2007). Differential column shortening effects in typical medium-to
Pham, H. A. (2016b). Truss optimization with frequency constraints using enhanced high-rise buildings. In New Horizons and Better Practices (pp. 1–10).
differential evolution based on adaptive directional mutation and nearest neighbor Zou, D., Du, C., Liu, T., Teng, J., & Cheng, H. (2019). Time-dependent deformations of
comparison. Advances in Engineering Software, 102, 142–154. concrete columns under different construction load histories. Advances in Structural
Samarakkody, D. I., Thambiratnam, D. P., Chan, T. H. T., & Moragaspitiya, P. H. N. Engineering, 22(8), 1845–1854.
(2017). Differential axial shortening and its effects in high rise buildings with Zou, D., Liu, T., Teng, J., Du, C., & Li, B. (2014). Influence of creep and drying shrinkage
composite concrete filled tube columns. Construction and Building Materials, 143, of reinforced concrete shear walls on the axial shortening of high-rise buildings.
659–672. Construction and Building Materials, 55, 46–56.
Song, H.-C., & Yoon, K.-S. (2006). Probabilistic prediction and field measurement of
column shortening for tall building with bearing wall system. Journal of the Korea
Concrete Institute, 18(1), 101–108.

10

You might also like