Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Proceedings of the 25th CANCAM

London, Ontario, Canada, May 31 – June 4, 2014

VIBRATION CONTROL OF TALL BUILDINGS USING AERODYNAMIC


OPTIMIZATION
Ahmed Elshaer Girma Bitsuamlak Ashraf El-Damatty
Civil and Environmental Department Civil and Environmental Department Civil and Environmental Department
Western University Western University Western University
London, Ontario, Canada London, Ontario, Canada London, Ontario, Canada
aelshae@uwo.ca gbitsuam@uwo.ca damatty@uwo.ca

results in generating high across-wind force variations for


ABSTRACT relatively small changes in wind angle of attack caused by
turbulence. Shape changes that make them less like an airfoil
Traditionally the shape and orientation of tall buildings are can be effective in this situation [2].
driven by architectural considerations, functional
requirements, and site limitations rather than aerodynamic LITERATURE REVIEW
considerations. As a result, they are bluff bodies characterized
by high wind-structure interaction induced motion. Vibration The aerodynamic performance improvement for tall
control is becoming an important aspect of tall building design buildings using shape modifications has prompted many
in order to keep the building’s motions within comfortable scholars to study the relationship between the aerodynamic
limits. In this study, an optimization procedure is presented for characteristics of a structure and the resulting wind–induced
minimizing the wind-induced vibrations of tall buildings by responses. These include Dagnew [3], Kawai [4], Irwin and
aerodynamic mitigation of corners. Computational Fluid Breukelman [5], Tamura and Miyagi [6], Bitsuamlak et al. [7],
dynamics (CFD) models are used to simulate the Elshaer et al. [8], and Kareem et al. [9]. Rather than the
aerodynamics of wind around a building. An optimization classical sharp edged corner of most buildings, building
problem is presented to demonstrate the developed procedure. corners can be chamfered, rounded, recessioned, fined, or any
other non-uniform shapes. Tse et al. [10] investigated the
KEYWORDS: Tall Buildings, vibration, Optimization, CFD, effect of chamfered and recessed corners on tall building
Wind-induced responses responses. Kwok et al. [11][12] studied the effects of
chamfered corners and slotted corners of prismatic tall
INTRODUCTION buildings on the wind-induced responses. The latter reported
that chamfering 10% of the building’s width may result in a
Tall buildings attract people and businesses because of their 30% to 40% reduction in both alongwind and crosswind
elegant shape and being considered landmarks. However, this induced motions. Also, the corners on “Taipei 101” were
appeal could be affected due to vibrations resulting from stepped in order to reduce across-wind responses and drag,
lateral loads such as wind and earthquake. For super-tall resulting in 25% reduction in base moment [2]. Most of these
buildings, wind-induced motion will be even more significant. studies primarily discuss the advantage of group of
This will be even more significant in the recent generation of modifications compared to others. However, they fall short in
tall buildings owing to being elaborate and slender. Tall advising what type of mitigations provide a “near optimal”
buildings can be susceptible to excessive motion during wind solution rather than an “ad-hoc” solution that may or may not
events that can cause occupant discomfort and reduce the be “optimal”.
overall appeal of the structure [1]. Furthermore, keeping the
motions of a tall building within comfortable limits could OPTIMIZATION PROCEDURE
sometimes govern the design rather than the strength
requirements. Improving the aerodynamics of a building The Aerodynamic Shape Optimization (ASO) for tall
through corner modification can result in a significant buildings’ corners procedure starts by choosing the objective
reduction in the lateral load and limit vibrations caused by the function and the design variables. The objective function is an
wind leading to a more economic and comfortable tall aerodynamic characteristic that is intended to be minimized or
building. maximized. The design variables are geometric parameters
One of the most significant vibrations that buildings controlling the shape of the corner. Different combinations of
experience is the across-wind motion resulting from wind design variables will form different candidates in the
vortex shedding. Some cross-sections such as lens shape are optimization problem. Those candidates were ranked
more prone to across-wind buffeting because their streamlined according to their fitness (achieving of the objective function).
shape causes them to act somewhat like a vertical airfoil. This Values of the objective function for different candidates
depend on the values of the design variables for each
candidate. In the current ASO, the values of the objective
function were evaluated either using Computational Fluid
dynamics (CFD) models or surrogate models such as neural
networks (NN) trained by the data coming from CFD
simulations. The values of the objective function were
evaluated for different wind angles of attack A(i) and return to
the optimization algorithm with the highest value. The
procedure of evaluating the objective function values is
summarized in Figure 1. The role of the optimization
algorithm is to search for the optimal solution through an Figure 2 Geometric design variables
iterative procedure.
OPTIMIZATION ALGORITHM

The Genetic Algorithm (GA) was adopted in the current


research because of its reliable ability to find the global
extreme value of the objective function without being trapped
in a local extreme value. In addition, GA was found to be
efficient in solving similar optimization problems [13][14]. In
GA, the optimization process starts by evaluating the objective
function value for the initial population candidates. Crossover
and mutation operators were applied on the current population
to produce new candidates to be tested. The role of those
operators is to produce new candidates of higher fitness
(minimizing the objective function) and investigating areas of
the search space that had not been explored. After that, the
value of the objective function will be evaluated for the new
generated candidates. The step of applying the operators and
evaluating the objective function value is iteratively repeated
until there is no significant improvement in the optimal
Figure 1 Objective function evaluation
solution. Each repetition of that step is considered a new
generation in the optimization procedure.
In the current optimization problem, v1 and v2 were selected to In the current problem, the initial population was formed
be the design variables. c and a are calculated according to Eq. of 70 candidates with different combinations of v1 and v2. The
(1). The objective function is the standard deviation of the lift optimization analysis was continued for 50 generations. Three
coefficient, where the lift coefficient is calculated according to types of crossover operators and four types of mutation
Eq. (2). operators were applied in each generation on each of 23

c  v 1L a  v2 c
candidates of the existing population. The objective function
, (1) values were evaluated using the NN surrogate model, which
where c is the modification length of the corner, L is the side was trained by the data resulting from the CFD numerical
length of the square section of the building, and a is the simulations. NN and CFD will be explained in the following
distance measured from the edge of the building to the corner sections.
tip (Figure 2).

CL 
CFD MODEL PROPERTIES
FL
1  2 A
(3)
2 R p Large eddy simulations (LES) for the flow around a tall
building were conducted to obtain the time history of the lift
where FL is the lift force acting laterally to the wind direction,
 is the air density,  R is the reference velocity, and A p is
coefficient for different corner shapes, as shown in Figure 3.
The simulations were conducted using a length scale of 1:500,
the projected area of the building section. time scale of 1:100, and a uniform inlet velocity 10 m/s.
Commercial CFD package (STAR CCM+) [15] was utilized
to solve the LES. The dynamic Sub-Grid Scale model by
Smagornisky [16] and Germano et al. [17] is used to account
for the turbulence. The total number of mesh cells in each
model was more than 200,000 cells. The polyhedral mesh size
was less than (D/20), where D is the width of the building. The
time step was set to be equal to 5E-4 sec to maintain Courant output data. As shown in Figure 6, the regression coefficient
Friedrichs-Lewy (CFL) lower than (1.0). The total physical of the generated NN model was 0.9964, which reflects the
time was 0.5 sec, which produces from 10 to 18 vortex- reliability of the NN as a surrogate model in this application.
shedding cycles. The dimensions of the employed : R=0.99644

computational domain followed the recommendation of COST 3 Data


Fit

[18] guidelines. Figure 4 shows the generated mesh and the 2.5
Y=T

Output ~= 0.99*Target + 0.026


velocity contour for different corner shapes and angles of
2
attack.
1.5
1.5 data1
1 data2
1
data3
0.5 data4
0.5
0
CL

-0.5 0.5 1 1.5 2 2.5 3


Target
-1

-1.5
Figure 6 Fitting model (NN) vs target data (CFD)
-2

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4


OPTIMIZATION PROBLEM RESULTS
time (sec.)
Figure 3 Lift coefficient time history for different corner shapes
After running the analysis, the best fitness-generations curve
and the optimal corner shape was acquired, as shown in Figure
7. The latter curve shows the aerodynamic improvement over
generations resulted from the optimization process. The
optimization analysis was repeated four times to ensure
reaching the optimal solution without being trapped in local
minima. Figure 7 shows that after running the analysis more
than once all optimal solutions converges to the same shape,
which indicates reaching the global minima of the search
space. Optimal solution acquired from the analyses was of
v1 = 0.127, v2 = - 0.94, and CL' = 0.503. The resulted optimal
solution was found to be of less fluctuating lift coefficient by
24% than that of square cross-section.

Figure 4 Mesh scene and Velocity contour plan view


0.7
Run 1
NEURAL NETWORK Run 2
0.65 Run 3
CL' of Best Fitness

NN was used as an alternative model to estimate the Run 4


objective function value as it is considered an efficient data-
fitting tool that is capable of estimating outputs values based 0.6
on input values after being trained using a database of different
v1 = 0.127
inputs and outputs. A database of 200 samples was selected
v2 = - 0.94
randomly for different combinations of v1, v2, and angle of 0.55
attack, as shown in Figure 5.

40 40 0.5
35 35
10 20 30 40 50
30 30
Generations
25 25
Angle

Angle

20 20
Figure 7 Best fitness curve
15 15

10 10 CONCLUSIONS
5 5

0
-1 -0.5 0 0.5 1 1.5
0
0.06 0.08 0.1 0.12 0.14 0.16 0.18 In the current study, the idea of controlling the vibration
v2 v1
of tall buildings by corner shape modifications has been
Figure 5 Selected candidates to train NN discussed. An optimization procedure was introduced by
adopting the Genetic Algorithm (GA) to minimize the
The objective function values for all samples were
standard deviation of the lift coefficient of tall buildings by
angle of attack as input data, and corresponding C L values as
evaluated using CFD models. NN was trained using v1, v2, and
mitigating their corner shape. A demonstration example was
presented to show the capability of the developed procedure.
NN was trained using the data resulted from CFD models, [9] T. Kijewski and Ahsan Kareem, “Mitigation of
which simulate the aerodynamics of the tall buildings. NN was motions of tall buildings with specific examples of
found to be a reliable surrogate model for objective function recent applications,” Wind Struct., vol. 2, no. 3, pp.
estimation. The fluctuating lift coefficient of the resulted 201–251, 1999.
optimal solution is 24% less than that of square cross-section.
[10] K. T. Tse, P. A. Hitchcock, K. C. S. Kwok, S.
ACKNOWLEDGMENTS Thepmongkorn, and C. M. Chan, “Economic
perspectives of aerodynamic treatments of square tall
The authors would like to acknowledge SharcNet [19] for buildings,” J. Wind Eng. Ind. Aerodyn., vol. 97, no.
providing access to their high performance computation 9, pp. 455–467, 2009.
facility and CD-adapco for their powerful numerical tool [15].
REFERENCES [11] K. C. S. Kwok and P. A. Bailey, “Aerodynamic
devices for tall buildings and structures,” J. Eng.
[1] A. Kareem, “Dynamic response of high-rise Mech., vol. 113, no. 3, pp. 349–365, 1987.
buildings to stochastic wind loads,” J. Wind Eng. Ind.
Aerodyn., vol. 42(1–3), pp. 1101–1112, 1992. [12] K. C. S. Kwok, P. A. Wilhelm, and B. G. Wilkie,
“Effect of edge configuration on wind-induced
[2] P. A. Irwin, “Wind engineering challenges of the response of tall buildings,” Eng. Struct., vol. 10, no.
new generation of super-tall buildings,” J. Wind Eng. 2, pp. 135–140, 1988.
Ind. Aerodyn., vol. 97, no. 7–8, pp. 328–334, Sep.
2009. [13] A. M. El Ansary, A. A. El Damatty, and A. O.
Nassef, “Optimum Shape and Design of Cooling
[3] A. K. Dagnew and G. T. Bitsuamlak, “Computational Towers,” World Acad. Sci. Eng. anf Technol., vol. 9,
evaluation of wind loads on a standard tall building pp. 4–13, 2011.
using LES,” Wind Struct., vol. 18, no. 5, pp. 567–
598, 2014. [14] A. Oyama, “Real-Coded Adaptive Range Genetic
Algorithm And Its Application to Aerodynamic
[4] H. Kawai, “Effect of corner modifications on Design,” JSME Int. J. Ser. A, vol. 4, 2002.
aeroelastic instabilities of tall buildings,” J. Wind
Eng. Ind. Aerodyn., vol. 74–76, pp. 719–729, Apr. [15] “STAR CCM+.” CD-ADAPCO Product, www.cd-
1998. adapco.com/products/star-ccm, 2015.

[5] P. A. Irwin, B. Breukelman, C. Williams, and M. [16] J. Smagorinsky, “General circulation experiments
Hunter, “Shaping and orienting tall buildings for with the primitive equations,” 1963.
wind,” in Proceedings of the Structural Engineers
World Congress, ASCE, San Francisco, 1998, pp. [17] M. Germano, U. Piomelli, P. Moin, and W. H. Cabot,
19–23. “A dynamic subgrid-scale eddy viscosity model,”
Phys. Fluids, vol. A 3(7), pp. 1760–1765, 1991.
[6] T. Tamura and T. Miyagi, “The effect of turbulence
on aerodynamic forces on a square cylinder with [18] J. Franke, A. Hellsten, H. Schlünzen, and B.
various corner shapes,” J. Wind Eng. Ind. Aerodyn., Carissimo, Eds., COST . Best practice guideline for
vol. 83, pp. 135–145, 1999. the CFD simulation of flows in the urban
environment, vol. Action 732. Hamburg: University
[7] G. Bitsuamlak, W. Warsido, E. Ledesma, and A. Gan of Hamburg, 2007.
Chowdhury, “Aerodynamic mitigation of roof and
wall corner suctions by using simple architectural [19] “SHARCNET is a consortium of colleges,
elements,” J. Eng. Mech., Spec. ISSUE Exp. Methods universities and research institutes operating a
Damage Detect. Wind Eng., pp. 396–408, 2013. network of high-performance computer clusters
across south western, central and northern Ontario,”
[8] A. Elshaer, G. Bitsuamlak, and A. El Damatty, 2015. [Online]. Available: www.sharcnet.ca.
“Wind Load Reductions due to Building Corner
Modifications,” in 22nd Annual Conference of the
CFD Society of Canada, 2014.

You might also like