Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

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

Towards LES as a design tool: Wind loads assessment on a high-rise building

M. Ricci a, b, *, L. Patruno a, I. Kalkman b, S. de Miranda a, B. Blocken b, c
a
DICAM, University of Bologna, Viale Risorgimento 2, 40136, Bologna, Italy
b
Building Physics and Services, Department of the Built Environment, Eindhoven University of Technology, P.O. box 513, 5600, MB, Eindhoven, the Netherlands
c
Building Physics Section, Department of Civil Engineering, KU Leuven, Kasteelpark Arenberg 40 - bus 2447, 3001, Leuven, Belgium

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

Keywords: The accurate evaluation of wind loads on high-rise buildings represents a key point in their design process.
Wind loading Traditionally, atmospheric boundary layer wind tunnel tests, conceived in order to be representative of the wind
LES conditions expected on site, are used for this purpose. Recently, owing to the increase in computational power,
Turbulent inlet Computational Fluid Dynamics (CFD) techniques have gained interest as a complementary tool to wind tunnel
High-rise building testing. Unfortunately, wind flow around bluff bodies is often very complex and substantial research efforts are
Computational wind engineering
still needed in order to assess the accuracy and reliability of CFD results. In this paper, Large Eddy Simulations
(LES) are performed aimed at evaluating the wind loads on an isolated high-rise building. In order to assess the
capabilities of LES for adoption as a design tool, the results are analysed in terms of both pressure distributions
and internal forces on the structural elements. It is found that the accuracy of LES in reproducing the fluctuating
pressure field is not necessarily maintained when internal forces are taken into account. Nevertheless, the design
values predicted by LES can be still considered satisfactory, in particular when maximum and minimum values
over different angles of attack are considered.

1. Introduction approaches. It satisfied the need for a relatively simple technique able to
evaluate the structural response in a reduced amount of time, without
The new generation of high-rise buildings is tending towards taller employing expensive aeroelastic models and without introducing sim-
and more slender structures, while their shapes are becoming increas- plifications typical of analytical approaches that are not generally
ingly complex and often unconventional. Due to these characteristics, applicabile (Davenport, 1961; Kareem, 1984; Kareem and Zhou, 2003).
particular attention has to be paid to the design of these structures, which The main characteristic of this method is that the structural response can
have to face increasing environmental loads while still remaining as light be reconstructed by measuring only forces and moments at the building
as possible. In particular, wind loads on tall buildings can play a funda- base. In this experimental approach, a rigid model of the building is
mental role in the design process and have to be accurately assessed with mounted on a balance characterised by high stiffness and high sensitivity,
respect to both structural integrity and serviceability (Irwin, 2009). which records time histories of shear forces, torque and bending mo-
The turbulence naturally present in the Atmospheric Boundary Layer ments. The structural response can be evaluated by post-processing re-
(ABL), together with aerodynamic phenomena typical of bluff bodies like sults in the frequency domain (Chen and Kareem, 2005) or in the time
vortex shedding and intermittent reattachment of shear layers, causes the domain (Xie and Irwin, 1998).
structure to experience dynamic forces which lead to along-wind and It should be noted that methodologies based on the use of HFFB
across-wind vibrations. measurements always imply the introduction of assumptions with respect
The traditional approach for the assessment of wind loads on high- to the structural behaviour and/or the pressure distributions. For this
rise buildings strongly relies on atmospheric boundary layer wind tun- reason the HFFB method can perform well only for the fundamental
nel practice. The most commonly adopted experimental techniques can modes of the structure (Zhang et al., 2015), while high frequency mode
be subdivided into three main methodologies (Irwin, 2009). The effects can not be satisfactorily predicted and structures characterised by
High-Frequency Force Balance (HFFB) method, initially developed by unconventional geometries cannot be easily studied. Furthermore, in
Tschanz et al. (Tschanz and Davenport, 1983), represents one of the first order to avoid inertial effects the model should be as stiff as possible. This

* Corresponding author. DICAM, University of Bologna, Viale Risorgimento 2, 40136, Bologna, Italy.
E-mail address: [email protected] (M. Ricci).

https://doi.org/10.1016/j.jweia.2018.07.009
Received 14 June 2017; Received in revised form 21 May 2018; Accepted 12 July 2018
Available online 21 July 2018
0167-6105/© 2018 Published by Elsevier Ltd.
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

requirement might be hard to respect, in particular when dealing with scale model (Ricci et al., 2016). Additionally, it should be noted that the
very slender tall buildings. generation of realistic unsteady boundary conditions, able to reproduce
The second experimental methodology commonly adopted to assess the main features of the turbulence found in the ABL, is imperative in
wind loads on tall buildings is the High-Frequency Pressure Integration order to obtain accurate results in terms of pressure distributions and,
(HFPI) method. Developed by Irwin et al. (Irwin and Kochanski, 1995), thus, forces acting on the structure. Among other structures, this fact has
this approach consists of equipping the exposed surfaces of the building been observed for high-rise buildings and, in particular, in the framework
model with a number of pressure taps, closely spaced in order to accu- of the CAARC benchmark (Zhang et al., 2015; Elshaer et al., 2016; Braun
rately sample the fluctuating pressure field acting on it at each sampling and Awruch, 2009), since results of different turbulent inflow generation
time. Differently from the HFFB method, no assumptions on the modal techniques were analysed in this case (Aboshosha et al., 2015; Huang
shapes or on the spatial distribution of the pressure field are necessary, so et al., 2006).
that wind effects on structures with irregular distributions of stiffness and As a result, even if CFD nowadays represents a very attractive tool for
mass and complex modal shapes can be analysed. Indeed, modal forces the assessment of wind loads on structures, it appears that substantial
can be reconstructed directly from pressure measurements and the research efforts are still needed in order to evaluate its reliability and
structural response can be consequently assessed. Furthermore, again accuracy. In particular, the available literature is focussed mainly on the
differently from the HFFB method, the pressure measurements are not comparison of pressure distributions and their statistics, typically up to
affected by inertial forces caused by the model itself. On the other hand, second order. Even if such studies are definitely necessary, they do not
buildings with complex geometrical shapes would require a very large appear to be sufficient in order to assess the capabilities of the numerical
number of pressure taps acquiring data simultaneously and, sometimes, models in the prediction of wind loads. In fact, internal forces on struc-
the limited number of available taps might lead to inaccurate sampling of tural members are related to a number of factors which are often not
the pressure field. taken into account in such studies like, for example, the spatial coherence
The third experimental methodology for wind load assessment on of the pressure field and its frequency content, whose relevance and
high-rise structures involves the use of aeroelastic models. In this case, overall effect can be amplified or weakened depending on the mechanical
the model is intended to reproduce the dynamic properties of the full- transfer functions characterising the structural dynamic behaviour itself.
scale structure, approaching its modal shapes up to a certain natural A detailed evaluation of LES capabilities in the simulation of flow around
frequency as closely as possible. The adoption of aeroelastic models al- low-rise buildings with specific application to structural design has been
lows measurement of the full response of the structure, taking into ac- recently reported in (Ricci et al., 2017).
count aeroelastic effects like aerodynamic damping. By adopting such an In this context, the present paper represents an essential step towards
approach, complex fluid-structure interactions can be experimentally the assessment of LES as a design tool for high-rise buildings. In partic-
investigated. Nevertheless, due to its high costs, this technique is usually ular, the turbulent fluctuations which characterise the wind impinging on
limited to the study of structures for which aeroelastic effects are ex- the structure are generated by means of the Modified Discretizing and
pected to be of primary importance. Despite the advantages of aeroleastic Synthesizing Random Flow Generator (MDSRFG), which produces a so-
models, in many cases, the HFPI method represents a good alternative lenoidal fluctuation field and allows control of its spatial and temporal
and a satisfactory trade-off between accuracy and complexity of the correlations (Castro and Paz, 2013). First, results are analysed in terms of
experimental tests. statistics of pressure distributions on the building and systematically
In the last decades, thanks to the significant growth of available compared with wind tunnel measurements. Then, dynamic structural
computer power, numerical approaches based on Computational Fluid analyses are carried out for each considered angle of attack starting from
Dynamics (CFD) have become more and more adopted as a comple- both numerically predicted and experimentally obtained pressure data.
mentary tool to investigate wind flow around buildings (Blocken, 2014). Results are analysed in terms of envelopes of internal forces acting on
Numerical methods have several advantages compared to wind tunnel structural elements, allowing an analysis of the effectiveness of the
tests. In particular, the costs of CFD analyses are generally lower than adopted numerical strategy in predicting design wind loads on the
those of experiments and each quantity of interest can be measured considered high-rise structure. It should be noted that by adopting such a
everywhere in the computational domain rather than sampled at just a framework, all the elements which play an important role in the defini-
few points. Due to these characteristics, numerical simulations allow the tion of the structural response are automatically taken into account,
study of phenomena which might be difficult to analyse in wind tunnel allowing the study of the effectiveness of the proposed numerical strategy
tests. compared to the traditional experimental practice.
CFD simulations based on the Reynolds-Averaged Navier-Stokes The paper is organised as follows. In Section 2 the experimental setup
(RANS) models have been deeply investigated in the p5 : 2ast. However, adopted to obtain the data used for a comparison with numerical results
such models are often found to be inaccurate where wind loads on bluff is briefly described, while Section 3 reports the main features of the
bodies are concerned and their predictive capability is limited to mean adopted computational model together with a discussion on the inflow
flow properties, while the ability to accurately predict turbulent fluctu- condition adopted for the LES. Section 4 is devoted to the analysis of
ations is recognised to be of fundamental importance for the assessment pressure distributions for each considered angle of attack. Then, Section
of the dynamic response of structures (Huang et al., 2010). The need to 5 describes the results obtained from the dynamic structural analyses,
correctly take into account the flow dynamics led researchers to move focusing on internal forces on structural elements. Finally, in Section 6
towards scale-resolving turbulence approaches, such as Large Eddy some conclusions are drawn.
Simulations (LES), which are promising for the numerical prediction of
wind loads on buildings. Although such models are commonly consid- 2. Experimental setup
ered to be well suited for the analysis of flows around bluff bodies, it has
often been observed that results are considerably scattered when In this section the experimental setup adopted to obtain pressure data
compared to experimental measurements, even when simple geometries used for comparison with numerical results is described. Experiments
are considered (Bruno et al., 2014; Patruno et al., 2016a; Ricci et al., were carried out at the atmospheric boundary layer wind tunnel of the
2016). Indeed, the complex, instability-driven phenomena observed in Tokyo Polytechnic University (TPU) and results are collected in a public
the turbulent flows around bluff bodies, like shear layer detachmen- database organised according to the geometry and the layout charac-
t/reattachment and vortex shedding, render the simulation of this kind of teristics (Tokyo Polytechnic Univers, 2003). The present work focuses on
flow an extremely challenging task and results are often found to be an isolated high-rise building characterised by a height (H) to breadth (B)
dependent on the simulation setup and the adopted turbulence subgrid ratio equal to and a depth (D) to breadth (B) ratio equal to 1 : 2 (see Fig. 1

2
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

(a)). During wind tunnel tests, the length scale was equal to 1=400, 3.1. Domain, grid and solver settings
leading to a model with B ¼ 200 mm, D ¼ 100 mm and H ¼ 500 mm.
The wind tunnel section was 2:2 m wide and 1:8 m high, so that the The computational domain dimensions in the across-wind section are
maximum blockage ratio was less than 2:5%. The wind profile repro- close to those of the wind tunnel, although slightly reduced with respect
duced in the wind tunnel corresponded to that of terrain category IV to them. A three dimensional view of the adopted computational domain
according to the Architectural Institute of Japan (AIJ) standards is presented in Fig. 2, while Fig. 3 (a) and Fig. 3 (b) show lateral and top
(Architectural Institute of Japan, 2004): views, respectively. The across-wind section is 4:4H wide and 3:6H high

α while the distance of the building from the inlet boundary is equal to
UðzÞ ¼ 1:7
z
Uref ; Zb < z  ZG ; (1) 3:6H. The resulting blockage ratio at 0∘ angle of attack is equal to 2:5%,
ZG while at 90∘ it equals 1:25%; in both cases it is lower than the maximum
of 3:0% suggested by the COST guidelines (Franke et al., 2007). In

α
Zb accordance with the recommendation by Tominaga et al. (2008) the
UðzÞ ¼ 1:7 Uref ; z  Zb ; (2)
ZG distance of the high-rise building from the outlet boundary is set equal to
10H.
where the exponent α is equal to 0.25, ZG is the reference height of the Aiming to reduce as much as possible the along-wind deterioration of
ABL equal to 550 m, and Zb represents the characteristic dimension of the profiles imposed at the inlet (Blocken et al., 2007), five staggered rows of
surface roughness elements equal to 20 m. Uref is the reference wind square blocks with an edge length equal to 0:06H are placed upstream of
velocity at reference height. All the quantities relative to the AIJ stan- the model. The block distribution is the same as that adopted during the
dards are intended to be in full scale. In the experiments, the wind ve- wind tunnel tests. In order to check the aerodynamic roughness related to
locity at the height of the building was equal to UH ¼ 11:11 m=s, leading the adopted layout and the geometry of the blocks, the following equa-
to a Reynolds number equal to Re ¼ UHνH ¼ 3:8  105 . tion proposed by Lettau is used (Lettau, 1969):
The adopted turbulence intensity profile was in agreement with AIJ
Ar
standards for terrain category IV as reported in Eqs. (3) and (4): z0 ¼ 0:5h ; (5)
At

α0:05
z
IðzÞ ¼ 0:1 ; Zb < z  ZG ; (3) where z0 is the aerodynamic roughness length as defined by the Euro-
ZG
pean standard for wind actions on structures EN1991-1-4:2005 (EN

α0:05 1991-1-4, 2005), Ar is the area of the element normal to the wind di-
Zb rection and At is the ground area per roughness element, as reported in
IðzÞ ¼ 0:1 ; z  Zb : (4)
ZG Fig. 4 (a).
The mean velocity and turbulence intensity profiles were obtained by
means of spires and square blocks as roughness elements. Experiments
were conducted for 21 wind directions, from θ ¼ 0∘ to θ ¼ 100∘ with 5∘
increments. The angle of attack θ is defined in an anti-clockwise sense
around the vertical axis (z) with θ ¼ 0∘ corresponding to the direction
orthogonal to the edge D, as reported in Fig. 1 (b). The model was
equipped with 510 pressure taps that acquired data synchronously at a
sampling frequency of 1000 Hz for a duration of 32.8 s per wind
direction.

3. Computational model

In the present section the characteristics of the numerical model

adopted for LES are described. In particular, Section 3.1 describes the Fig. 2. Computational domain adopted for the numerical study: three-
computational domain characteristics and the numerical settings, while dimensional view.
Section 3.2 focuses on the generation of the turbulent inflow data.

Fig. 1. Geometry of the high-rise building: (a) three-dimensional and (b) Fig. 3. Computational domain adopted for the numerical study: (a) lateral and
top view. (b) top views.

3
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 4. Three-dimensional view of (a) the roughness block geometry and (b) boundary layer height development according to Elliot's equation (6) (Elliott, 1958).

The resulting roughness length is about 0:3 m in full scale and is in divergence-free velocity field at the inlet. Time discretisation is per-
good agreement with both the roughness length of terrain category IV in formed with the second order accurate backward differentiation scheme.
EN1991-1-4:2005 (EN 1991-1-4, 2005) and with that considered in the The adopted dimensionless time step value, based on H, is Δt  ¼ ΔtU H ¼
H

present study. 4:4  103 . The maximum Courant-Friedrichs-Lewy number in all sim-
When the roughness fetch ends a new boundary layer starts to ulations is 2.9, while it is lower than 1.0 in 97% of the domain and its
develop which corresponds to a lower roughness length. This new mean value in the whole domain equals 0.06. The maximum dimen-
boundary layer might affect results at least in the lower part of the high- sionless wall distance y þ is equal to 9.0, while it is lower than 5.0 on 94%
rise building. Its height can be estimated with Eq. (6) proposed by Elliot of the high-rise building surface.
(Elliott, 1958): Spatial discretisation of the advective terms is performed by means of



0:8 the LUST scheme, which is second-order accurate and has been shown to
z0;1 d
Hbl ¼ z0;2 0:75 þ 0:03log ; (6) perform well in particular for LES in complex geometries (Weller, 2012),
z0;2 z0;2 offering a good trade-off between low dissipative behaviour and nu-
merical stability. For all the other terms of the equations a centered
where Hbl is the height of the new boundary layer, z0;1 and z0;2 are the
second-order differentiation scheme is adopted.
aerodynamic roughness lengths characterising the two different zones
Regarding the LES subgrid scale model, the Smagorinsky-Lilly model
and d is the distance measured from the last roughness element. The
(Lilly, 1992) is adopted with an additional transport equation for the
distance between the roughness blocks and the model is chosen in order
subgrid turbulent kinetic energy. This model is able to adjust the tur-
to limit the new boundary layer height to H=10, which according to this
bulent eddy viscosity depending on the subgrid kinetic energy, showing a
equation gives d ¼ 1:4H (see Fig. 4 (b)).
less dissipative behaviour compared to the standard Smagorinsky-Lilly
With respect to the boundary conditions; the mean velocity profile is
model (Ricci et al., 2016).
prescribed at the inlet boundary, while Neumann conditions on the ve-
The high-rise building is equipped with 2844 pressure monitors and
locity field are imposed at the outlet. The turbulent part of the inflow is
data are acquired at each time step. Simulations are performed with the
generated by means of the MDSRFG method, discussed in more detail in
open source Finite Volume software OpenFOAM ® v. 2.3.0 using 96 cores
the next section. Bottom surface, building and roughness blocks surfaces
of the Galileo cluster at the Italian supercomputing institute CINECA (6
are modelled as smooth walls, while symmetry boundary conditions are
nodes with 2-eight core Intel® Xeon® 2:40 GHz processors with 128 GB
imposed on the other domain boundaries.
RAM per node). Each simulation required about 2:5  104 CPU hours.
A structured mesh is adopted near the high-rise building surfaces, as
shown in Fig. 5 (a), where cell dimensions in x, y and z directions are
respectively δx =H ¼ δy =H ¼ δz =H ¼ 3:1  103 , leading to a resolution 3.2. Turbulent inflow characteristics
higher than that suggested by Tominaga et al. (2008). At the wall a
structured mesh is adopted for the boundary layer with a first cell height The turbulent part of the simulated atmospheric boundary layer is
of δz =H ¼ 5:1  104 . Proceeding away from the building the mesh is generated by means of the MDSRFG method (Castro and Paz, 2013). In
slowly coarsened up to δx =H0 ¼ δy =H0 ¼ δz =H0 ¼ 5:0  102 . This this method a homogeneous and anisotropic velocity field is computed
sizing is kept constant until the inlet boundary is reached in order to as:
correctly propagate inflow fluctuations and to keep the numerical dissi-
XM X N 


pation caused by the mesh as low as possible. A view of the mesh of the m;n xj t m;n xj t
ui ðx; tÞ ¼ pm;n
i cos kj þ ωm;n þ qm;n
i sin kj þ ωm;n ;
roughness blocks and the building can be observed in Fig. 5 (b). The final Ls τ0 Ls τ0
m¼1 n¼1
mesh contains about 1:2  107 cells. (7)
Pressure-velocity coupling is performed by means of the PISO algo-
rithm, modified as proposed by Kim et al. (2013) in order to obtain a with

Fig. 5. Lateral views (xz plane through ð0; 0; 0Þ) of the adopted mesh: (a) detailed view of the mesh near the building and (b) view of the roughness blocks and
the building.

4
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 Iu2 ¼ 0:75Iu1 ¼ 8:7% and Iu3 ¼ 0:5Iu1 ¼ 5:8% (Dyrbye and Hansen,
2 ðrm;n
i Þ
pm;n
i ¼ signðri Þ
m;n
Si ðf m ÞΔf m;n 2
; (8) 1999).
N 1 þ ðr i Þ
The turbulent length scale is not reported in the database for isolated
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi high-rise buildings of TPU (Tokyo Polytechnic Univers, 2003), therefore
2 1 two different values have been investigated: Lu1 =H ¼ 0:6 and Lu1 =H ¼
qm;n
i ¼ signðrim;n Þ Si ðf m ÞΔf 2
; (9)
N 1 þ ðr m;n i Þ
0:8. These values were chosen to be close to those reported by Kim and
Tamura (2014), who analysed several inflow conditions in the same wind
8 m;n m;n tunnel facility referred to in the present work. It should be noted that the
<k p ¼ 0;
km;n  qm;n ¼ 0; (10) MDSRFG procedure does not allow the a priori prescription of the tur-
: bulence length scale and, therefore, an iterative procedure is necessary to
jkm;n j ¼ f m =UH ;
match the target values. The iterative procedure consists of adjusting the
where i ¼ 1; 2; 3 denotes vector components in x, y and z directions, parameter Ls in Eq. (7) until the resulting field matches the desired length
respectively, while N is the number of random samples generated for scale. At the end of this procedure two synthetic inlet conditions are
each of the M frequencies f m . ωm;n is a random angular frequency obtained, hereafter referred to as inflow #1 and inflow #2, that differ
extracted from a Gaussian distribution N ðf m ; Δf Þ, Δf is the frequency only in the target turbulent length scale.
increment adopted in the spectra sampling, UH is the time averaged ve- In order to check whether the turbulence introduced in the proximity
locity as defined in Section 2, Ls is a scale factor calibrated a posteriori of the inlet is correctly propagated downstream, two simulations are
related to the turbulent length scale, and τ0 is a dimensionless parameter, performed by adopting the two aforementioned inflow conditions. The
set equal to one in the present work, that allows control over the time LES are performed in an empty domain identical to that shown in Fig. 2
correlation of the series. Parameters rim;n are random numbers extracted but without the building. The numerical setup adopted for these analyses
from a Gaussian distribution N ð0; 1Þ, while Si ðf Þ is the target spectrum is the same as that described in Section 3.1. Within the computational
characterising the i  th velocity component whose total variance is equal domain, 21,560 velocity monitors sampling at each time step are ar-
to ui ' 2 . For a detailed review of the characteristics of MDSRFG and other ranged in a regular grid ranging from  2:0  x=H  5:5, from 
methods used for the generation of synthetic turbulence based on the 1:0  y=H  1:0 and from 0  z=H  0:7.
spectral approach the reader is referred to Patruno et al. (Patruno and In order to check the convergence of the LES performed in the empty
Ricci, 2017). domain the velocity signal recorded at a point located at building height
In this paper, the well-known von Karman spectra reported below are H and in the location where the building will be placed afterwards is
assumed for the velocity components (Zhang et al., 2015): considered. Fig. 6 (a) shows the time history of the velocity in the along-
wind direction for inflow #1. In the spirit of the procedure proposed by
4ðIu1 UH Þ2 ðLu1 =UH Þ Bruno et al. (2010) the time history of the along-wind velocity is
Su1 ðf Þ ¼  5=6 ; (11) expressed as a function of the dimensionless time t  ¼ tUH =H and sub-
1 þ 70:8ðfLu1 =UH Þ2
divided in N sampling windows which extend from t  ¼ 0 to nT, where T
  has been chosen equal to 10 times the signal dimensionless integral time
4ðIu2 UH Þ2 ðLu2 =UH Þ 1 þ 188:4ð2fLu2 =UH Þ2 scale (calculated from the time autocorrelation function) and n ¼ 0;::;N.
Su2 ðf Þ ¼  11=6 ; (12)
1 þ 70:8ðfLu2 =UH Þ2 For each sampling window first and second order statistics of the signal
are calculated. Then, for each of these statistics the percentage residual
  φres is computed as φnres ¼ φn φ n1
 100. Fig. 6 (b) reports the trend of φres
4ðIu3 UH Þ2 ðLu3 =UH Þ 1 þ 188:4ð2fLu3 =UH Þ2 φn
Su3 ðf Þ ¼  11=6
; (13) for both the time average and the root mean square (r.m.s.). After 170t 
1 þ 70:8ðfLu3 =UH Þ2 residuals of both average velocity and its r.m.s. are lower than 1%,
indicating that a satisfactory convergence of inflow statistics can be
where I and L are the turbulence intensity and the turbulence length scale
considered achieved at this time, at least up to the second order statistics.
at the reference height H, respectively, while the u1 ; u2 ; u3 subscripts
Results in terms of profiles of average velocity U, along-wind turbu-
indicate the components in x; y and z directions. In the present work, Iu1
lence intensity Iu1 , and turbulence length scale Lu1 , at the location where
is set equal to 11:6% according to experimental measurements, while
the building will be placed afterwards are reported in Fig. 7 (a), (b) and

Fig. 6. Convergence of inflow velocity statistics: (a) time history of the along-wind velocity and (b) percentage residuals of its average and r.m.s.

5
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 7. Profiles of (a) average velocity, (b) along-wind turbulence intensity and (c) turbulence length scale obtained from LES and experiments. A comparison with
profiles given by AIJ reference standard (Architectural Institute of Japan, 2004) for terrain category IV is also provided.

(c) for the two analysed inflow conditions. No remarkable differences can different points, located respectively at the end of the roughness block
be observed in terms of average velocity profiles between inflow #1 and fetch at ð1:4H; 0; HÞ and in the location where the building will be
#2, with both almost overlapping and very close to experimental mea- placed at ð0; 0; HÞ. It can be seen that turbulence energy decay between
surements. When turbulence intensity is considered differences between the two locations is small, since spectra are very similar to each other.
inflow #1 and #2 slightly increase, although they are still lower than 3% Nevertheless, the underestimation of the turbulence intensity Iu1 at the
(in terms of relative percentage difference) everywhere. When the along- building height shown previously in Fig. 7 (b) can be observed also in
wind turbulence intensity profile is compared to experimental data, it can Fig. 8 (a). The same consideration holds also for the across-wind and
be observed that at the reference height LES simulations underestimate vertical wind velocity components. Furthermore, for all the considered
Iu1 of about 17:8% (in relative terms), while numerical simulations spectra, a frequency cut-off due to the mesh becomes evident starting
overestimate the experimental results as the ground is approached, so for from approximately fH=UH ¼ 2. The underestimation of the turbulence
z=H < ¼ 0:5. Such mismatches in turbulence intensity might lead to intensities shown by numerical simulations when results are compared to
mismatches of the same order of magnitude in terms of fluctuating target spectra might be due to the turbulence generation method adop-
pressures as will be later further discussed. It should be noted that these ted, which is not able to control the turbulence length scale a priori as
effects might be due to discrepancies between the inflow turbulence well as the coherence functions. On this regard, the adoption of inflow
spectra adopted in simulation and wind tunnel experiments as well as to turbulence generation methods able to control the inflow coherence
the roughness blocks introduced in the computational domain. Indeed, functions might improve the agreement between the target and the
even if the considered pattern of blocks lead to the target aerodynamic adopted spectra (Aboshosha et al., 2015).
roughness, as reported in Sec. 3.1, some discrepancies might exist be- In all, the synthetic inlet conditions comply satisfactorily with the
tween the layout of the blocks adopted in the proposed simulations and target profiles, even if some discrepancies between numerical results and
wind tunnel experiments, which can lead to the aforementioned over- experimental data have been observed in particular in term of along-
estimation of Iu1 in the lower part of the high-rise building. wind turbulence intensity profiles.
The turbulent length scale has been computed using the Wiener-
Khinchin theorem together with the Taylor hypothesis; the resulting in- 4. Large Eddy Simulations results versus experiments
tegral turbulence length scale at each height has been averaged in the
across-wind direction over a length equal to B, which is the maximum In this section, first the analysis of calculation convergence together
projection on the across-wind section of the building area (occurring at 0∘ with the obtained integral forces for each angle of attack are reported in
angle of attack). In terms of turbulent length scale, the differences be- Section 4.1. Then, Section 4.2 shows some characteristics of the flow
tween inflow #1 and #2 remain almost constant until the height of z= topology, while Section 4.3 compares results of the LES in terms of sta-
H ¼ 0:5 is reached and above z=H ¼ 0:2 both curves are almost constant. tistics of the pressure distributions to experimental results for each
In this range, L=H for inflow #1 is equal to 0.77 while for inflow #2 it is considered angle of attack. Finally, in Section 4.4, scatter plots between
0.65, showing a satisfactory agreement with the target values. Proceed- experimental and numerical predictions of the pressure distribution
ing with increasing height the two curves become closer and differences statistics are shown for all building surfaces.
almost vanish at height of z=H ¼ 1:0. For each simulation, pressures are acquired at each time step and data
Fig. 8 shows the power spectral density of the three velocity com- are subsequently coarsened to match the sampling frequency adopted in
ponents, namely Su1 , Su2 and Su3 , in dimensionless form for inflow #1 as the wind tunnel tests. The positions of the pressure probes on the
well as the target von K arman spectra used as inputs for the inflow tur- building model are identical to those reported for the experimental setup
bulence generation. Aiming to analyse the dissipation of the turbulent and shown in Fig. 9 (a), while Fig. 9 (b) reports the paths s1 , s2 and s3
fluctuations in the along-wind direction, spectra are plotted for two subsequently adopted for plotting pressure statistics.

6
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 8. Spectra of the velocity time histories obtained from LES: (a) along-wind direction, (b) across-wind direction and (c) vertical direction.

residuals of its average Cp and r.m.s. C'p , obtained by following the same
method used for the inlet profiles in Section 3.2. It can be observed that
after 250t  the residuals in terms of both Cp and C'p are lower than 1%.
The same result is observed also for P2 and P3 .
Convergence is checked not only in terms of local pressures but also in
terms of integral forces acting on the high-rise building. The same pro-
cedure adopted for the inlet condition and for surface pressures is applied
to the integral force coefficients in x and y directions, defined respec-
F
tively as CFx ¼ 0:5ρFUx2 DH and CFy ¼ 0:5ρUy2 DH where Fx and Fy denote the
H H

integral forces, while their time-averaged values and their r.m.s are
0 0
referred as CFx , CFy and C Fx and C Fy , respectively. As previously observed
for the pressure coefficient, also in this case after 250t  the residuals of
the integral force coefficients are below the threshold of 1%. In all, the
duration of 250t  is considered to be sufficient in order to reach satis-
Fig. 9. (a) Positions of the pressure probes according to the experimental setup factory convergence of results at least up to second order statistics, so the
and (b) view of the sampling lines adopted for plotting pressure coeffi- simulations at all the investigated angles of attack are run until this
cient statistics. duration is reached.
In order to analyse the performance of inflow #1 and #2, the pres-
4.1. Convergence and integral forces sure spatial correlations between each pressure probe and all the others
are calculated and the resulting correlation fields are interpolated over
Simulation convergence is checked for each angle of attack by ana- the whole building surface. A comparison between inflow #1 and #2 at
lysing the pressure signals acquired at the three locations indicated in 0∘ is provided in Fig. 11, which shows the pressure coefficient spatial
Fig. 9 (a), namely P1 ¼ ð  0:1H;0; 0:5HÞ, P2 ¼ ð0:1H; 0; 0:5HÞ and P3 ¼ correlation Cp;corr between the pressure probe P1 and all the other probes
ð0;  0:2H; 0:5HÞ. These points are chosen because they are located in along the path s1 for the two considered inflow profiles, together with the
regions with different flow topology, that is respectively on the wind- corresponding experimental data. It can be observed that for inflow #2,
ward, leeward and the lateral surface. For the sake of brevity, only results characterised by the larger turbulent length scale, results are in good
in point P1 for the 0∘ angle of attack and inflow #1 are reported. In agreement with the experiments. On the windward surface the two
particular, Fig. 10 (a) shows the time history of the pressure coefficient curves almost overlap, while inflow #1 shows a more rapid decay of
Cp ¼ 0:5pρU 2 for the location P1 , while Fig. 10 (b) shows the percentage correlations proceeding away from probe P1 . Differences become even
H
more evident when the leeward surface is analysed. In this part, results

7
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 10. Convergence of the pressure signal at location P1 : (a) time history of the pressure coefficient and (b) percentage residuals of its average and r.m.s.

Fig. 11. Spatial correlation between the pressure probe P1 and all the other probes along the paths s1 (a) and s2 (b): comparison between inflow #1 and inflow #2 at
0∘ angle of attack.

for inflow #1 are almost uncorrelated with the signal recorded in P1 , as Also for these wind directions, Table 1 shows that a satisfactory
Cp;corr is almost zero along the whole path. In contrast, correlations ob- agreement is achieved between numerical and experiments both in terms
0 0
tained for inflow #2 are negative and equal to about 0:2 in average, in of CFx and C Fx . Nevertheless, when C Fy is analysed, LES simulations
rather close agreement with the experiments. show to significantly underestimate its value for all the considered angles
Differences observed in pressure correlations are also reflected in the of attack, being the relative difference compared to experimental data
flow bulk parameters, in particular in terms of the r.m.s. of the along- equal to 36%, 22% and 5% for 0∘ , 45∘ and 90∘ , respectively. This fact
wind force coefficient CFx . Table 1 reports the statistics of the force co- might be due to the underestimation observed in LES of the energy in the
efficients for all the considered angles of attack, while Fig. 12 shows the across-wind direction, in particular for fH=UH > 0:5 (see Fig. 8 (b)). The
0
spectra of the base integral moments Mx and My for the three angles of underestimation of C Fy at the considered angles of attack is also reflected
attack considered. As can be noted, at 0∘ the r.m.s. of CFx is equal to 0.153 in spectra of the base moments reported in Fig. 12. As discussed in Sec.
when inflow #1 is considered, while it equals 0.196 when inflow #2 is 3.2, the causes of this underestimation might be found in different fac-
adopted, resulting in this case very close to the experimentally observed tors, such as the target values used in the turbulence generation as well as
value of 0.195. The better performance for inflow #2 in terms of both the adopted turbulence generation method itself.
0
pressure correlations and C Fx led us to consider only this inflow condi- In all, the integral force coefficients are predicted by LES with satis-
tion for both other angles of attack considered here (i.e. 45∘ and 90∘ ). factory accuracy for all the considered angles of attack, in particular in
terms of average quantities, with a maximum absolute relative error
between numerical and experimental predictions equal to about 8%
Table 1 (excluding CFy at 0∘ and 90∘ , when it is expected to be zero) recorded at
Integral force coefficients.
the angle of attack of 45∘ . It is worth to stress that the difference observed
0 0
Angle Source Inflow CFx C Fx C Fy C Fy in the global forces spectra do not allow to characterise with good ac-
0∘
LES #1 1.13 0.153 0.01 0.12 curacy the mismatch in terms of structural response due to the presence
LES #2 1.14 0.196 0.01 0.13 of dynamic amplification. Such aspect will be considered in Sec. 5.
Exp.  1.13 0.195 0.002 0.19
45∘ LES #2 0.81 0.15 0.37 0.07
Exp.  0.88 0.17 0.34 0.09 4.2. Flow topology
90∘ LES #2 0.42 0.08 0.002 0.18
Exp.  0.43 0.09 0.02 0.19
In this section, the flow topology obtained with LES is described for

8
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 12. Spectra of the integral moments Mx and My for the 0∘ , 45∘ and 90∘ angles of attack.

each considered angle of attack. In particular, a qualitative view of the which significantly affect the pressure distribution and consequently the
vortical structures for the 0∘ angle of attack is reported in Fig. 13 by wind load on the building, it can be observed in Fig. 14 that, when the 0∘
means of isocontour of the invariant λ2 (Jeong and Hussain, 1995), col- angle of attack is considered, the average flow field is detached from both
oured with pressure. As can be observed, the boundary layer approaching the lateral and top surfaces. Conversely, the average flow reattaches on
the high-rise building separates leading to the development of the the top surface for the 45∘ angle of attack and on both top and along-wind
well-known horseshoe vortex. The vertical component of the velocity surfaces for 90∘ , as shown in Figs. 15 and 16, respectively. In particular,
gradient that characterises the approaching boundary layer causes at 90∘ (shown in Fig. 16 (b)) and for nominally smooth inflow conditions
pressures to be higher near the top part of the high-rise building, where the flow is known to be detached, since the aspect ratio of the considered
velocities are higher, and lower near the ground. This pressure gradient rectangular shape is D=B ¼ 2 which is less than the threshold value of 2.8
drives the flow downward close to the windward surface of the high-rise reported by Noda et al. (Noda and Nakayama, 2003a). Nevertheless, the
building and, as it approaches the ground surface, deviates the flow incoming turbulence and the enhanced turbulent mixing cause shear
upwind. This is considered to be the mechanism responsible for the layer instabilities to occur further upwind than for the smooth inflow
instability of the incoming boundary layer (Randerson, 1984) which also condition, leading the mean flow to reattach also for aspect ratios smaller
controls the position of the horseshoe vortex core. than 2.8 (Noda and Nakayama, 2003b). Furthermore, it is worth noting
These flow dynamics can be better appreciated if the average that, in the considered case, the flow topology is complicated by the fact
streamlines are observed. In particular, Figs. 14–16 show the average that the model is immersed in a turbulent shear flow, that is the simulated
streamlines obtained with the Line Integral Convolution (LIC) technique atmospheric boundary layer, so the reattachment length on the side walls
(Cabral and Leedom, 1993) for 0∘ , 45∘ and 90∘ angles of attack, respec- changes between the bottom and top of the high-rise building. In
tively. The streamlines are plotted for the xz plane through ð0; 0; 0Þ and particular, Fig. 17 (a) shows the change in the along-wind position of the
for the xy plane through ð0; 0; 0:5HÞ. As can be observed in Fig. 14, when reattachment point (indicated as xr ) with height at the 90∘ angle of
the 0∘ angle of attack is considered the core of the horseshoe vortex is attack. In the same figure the curve 1=Iu1 ðzÞ is drawn, where Iu1 ðzÞ is the
located at approximately x=D ¼ 1:3 while it moves closer to the turbulence intensity profile, shifted by an offset that minimizes the dif-
building's windward surface as the angle of attack increases, reaching x= ference between both curves calculated by means of a least squares
D ¼ 1:0 at 90∘ . procedure. Interestingly, the two curves show a very similar trend, at
Focusing on the average flow reattachments/detachment locations, least up to approximately z=D ¼ 2:0 where the reattachment bubble
reaches its maximum along-wind extension, suggesting a linear relation
between the reciprocal of the turbulence intensity and the reattachment
length in the considered range. Moreover, it is worth noting that the
height z=D ¼ 2:0 at which the reattachment length reaches its maximum
value represents the boundary between the regions where the flow is
deviated downwards or upwards, as can be observed in Fig. 16 (a). Above
this height the flow is deviated upwards by the high suctions occurring in
the top part of the high-rise building, as shown in Fig. 17 (b), so the
along-wind position of the reattachment point starts decreasing from a
maximum of xr =D ¼ 0:34 to xr =D ¼ 0:005.

4.3. Pressure distributions

Fig. 13. Three-dimensional view of the flow topology at 0∘ : isocontours of the In this section, results obtained from LES are systematically compared
invariant λ2 at the dimensionless value of λ2 U 2 =H 2 ¼ 30:5 coloured with the with experimental data in terms of pressure distributions. The distribu-
0
instantaneous pressure coefficient. tion of the average pressure coefficient Cp and of its r.m.s. C p along the

9
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 14. Average streamlines obtained with the LIC technique at 0∘ : (a) view of the xz plane through (0,0,0) and (b) of the xy plane through ð0; 0; 0:5HÞ.

Fig. 15. Average streamlines obtained with the LIC technique at 45∘ : (a) view of the xz plane through (0,0,0) and (b) of the xy plane through ð0; 0; 0:5HÞ.

Fig. 16. Average streamlines obtained with the LIC technique at 90∘ : (a) view of the xz plane through (0,0,0) and (b) of the xy plane through ð0; 0; 0:5HÞ.

path s1 for the 0∘ angle of attack is reported in Fig. 18. As can be observed case, in particular in the leeward part of the path. Also in this case, C'p
a good agreement is reached in terms of Cp on both windward and appears to be satisfactorily predicted by LES on the windward surface,
leeward sides. Focusing on the windward surface, both numerical and with a maximum relative difference between numerical and experi-
experimental data show a maximum of Cp at approximately s1 =H ¼ 0:85. mental data of less than 8%.
0
Regarding the numerical prediction of C p it can be observed that while In order to analyse the pressure distributions characteristics also on
on the windward surface LES tends to slightly overestimate, on the the windward surface as previously done for the 0∘ and 45∘ angles of
0

leeward surface the behaviour is opposite, with LES underestimating C p

0 attack, when the 90∘ angle of attack is considered the Cp and C p distri-
in particular near the bottom side where the average streamlines in butions are plotted along the path s2 in Fig. 20. This path has particular
Fig. 14 (a) show the presence of a recirculation vortex. interest also because it allows to observe the effects of flow reattachment
When the angle of attack is equal to 45∘ and path s1 is analysed on the pressure distributions on lateral surfaces. Looking at the Cp dis-
(Fig. 19) the average pressure coefficient predicted with LES almost tribution, the same observations which were made for 0∘ and 45∘ hold.
0
overlaps the experimental data, while discrepancies in terms of C p pre- The numerical and the experimental data are in good agreement, with a
∘ maximum relative difference of about 15%, recorded close to the leading
viously observed for the 0 angle of attack become more evident in this

10
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 17. (a) Average position of the reattachment point and (b) average pressure field on the along-wind surface at 90∘ .

Fig. 18. Pressure coefficient at 0∘ along the path s1 : (a) average and (b) r.m.s..

Fig. 19. Pressure coefficient at 45∘ along the path s1 : (a) average and (b) r.m.s.

0
edges where high suctions due to flow separation occur. In this respect it Fig. 20 (b) the numerical results underestimate the peak of C p , with a
should be noted that the accurate reproduction of flow dynamics near maximum relative difference between numerical and experimental data
such zones represent a very challenging task for numerical simulations, equal to 20%. This underestimation occurs in the along-wind sides of s2
since a significant reduction of relative differences would require and might be due to discrepancies between the across-wind turbulence
extremely high spatial and, consequently, temporal resolutions, intensity in experiments and numerical simulations. Nevertheless, also
rendering LES very time consuming and, indeed, compromising its use the body aerodynamics as well as the turbulence subgrid-scale modelling
for practical applications. Moving away from the locations where the can play an important role and it is really challenging to assess the
0
highest suctions are recorded, the pressure recovery starts and the slope relative effects of these factors on the C p distribution. Conversely, the
of the curve increases until the reattachment point is reached, at s2 =H ¼ 0
0
numerical predictions of C p values on the windward part of the path
0:32 and after s2 =H ¼ 0:68. Looking at the distribution of C p reported in (0:4  s2 =H  0:6) are in good agreement with experimental

11
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 20. Pressure coefficient at 90∘ along the path s2 : (a) average and (b) r.m.s.

measurements, indicating that the energy content in the incoming flow is average pressure coefficient is analysed the numerical results are quite
correctly reproduced. This observation also holds for the previously accurate over the whole building and for all angles of attack. Table 2
analysed angles of attack. shows that at 0∘ the percentage of points in the 30% range of tolerance is
93:3%, while at 45∘ and 90∘ the corresponding percentages are 82:0%
4.4. Overall pressure distributions and 99:0%, respectively. The MNB equals 1:5% at 0∘ and 1:33% at 45∘ ,
while at 90∘ it changes sign and its absolute value increases, becoming
In order to analyse the accuracy of the numerical prediction of the equal to  9:47%. This indicates that in this case the average tendency of
pressure coefficient over the whole building, in this section scatter plots LES is to underestimate the mean pressure coefficient.
between the experimental measurements and data obtained with LES for The scatter plots of Fig. 21 (b), 22 (b) and 23 (b) show that the r.m.s.
all the pressure probes and all the considered angles of attack are re- of the pressure coefficient is predicted with lower accuracy than the
0
ported. Fig. 21 shows the scatter plot for Cp and C p at 0∘ angle of attack, average, with a distribution of points which slightly deviates from the
while the same plots for 45∘ and 90∘ are reported in Figs. 22 and 23, bisector for all considered wind directions. These characteristics are also
respectively. Furthermore, Table 2 reports the percentage of points reflected in Table 2 which reports that, while at 0∘ 94:5% of the points is
whose difference between numerical and experimental predictions is less within the 30% range of tolerance, this percentage decreases to 69:4%
than 10%, 20% and 30% (hereafter referred to as ranges of tolerance) for and 88:8% for 45∘ and 90∘ , respectively. Also the MNB values are higher
each angle of attack and each analysed statistic. The same ranges are than those observed for the distributions of Cp , showing a maximum
highlighted in the scatter plots. In addition to these data, Table 2 also value of 19% for the 45∘ angle of attack. Nevertheless, it is worth noting
shows the Mean Normalized Bias (MNB) for each case, which is the that in this case the overestimation of the numerical data is mainly
0
average relative error between numerical and experimental predictions concentrated in points with relatively small values of C p (see Fig. 22 (b)).
defined as follows: In all, the maximum absolute MNB observed for the Cp distribution is
0
N
 less than 10%, while for the C p distributions, the maximum absolute
1 X Qi;num  Qi;exp
MNB ¼ ; (14) MNB is less than 20%.
N i¼1 Qi;exp

where N is the number of pressure probes, while Qi;num and Qi;exp are the 4.5. Pressure spatial correlations
considered quantities obtained from the numerical model and from the
experiments, respectively. The MNB is used as an index able to synthet- In order to accurately assess the structural response, not only the local
ically characterise the correlation plots. statistics of the pressure coefficient should be well predicted, but also the
As previously observed in Section 4.3 for paths s1 , s2 and s3 , when the spatial correlations need to be correctly reproduced. In order to analyse

Fig. 21. Cp scatter plots at 0∘ : (a) average and (b) r.m.s..

12
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 22. Cp scatter plots at 45∘ : (a) average and (b) r.m.s..

Fig. 23. Cp scatter plots at 90∘ : (a) average and (b) r.m.s..

and s2 are reported in Fig. 24, while when the 90∘ angle of attack is
Table 2 considered, the probe P3 is located on the windward surface, so in this
Performance metrics for the pressure coefficient at 0∘ , 45∘ and 90∘ .
case Fig. 25 shows the spatial correlation along the paths s3 and s2 to
Performance 0∘ 45∘ 90∘ which P3 belongs. For the sake of brevity spatial correlations at 0∘ angle
metrics 0 0 0 of attack reported in Fig. 11 are not here repeated.
Cp C p Cp C p Cp C p
Focusing on the windward part of the paths a good agreement be-
10% tolerance 45:1 % 27:8 % 59:8 % 31:6 % 30:6 % 52:9 %
20% tolerance 89:6 % 53:9 % 76:1 % 49:2 % 90:2 % 80:8 %
tween experimental and numerical data can be considered achieved for
30% tolerance 93:3 % 94:5 % 82:0 % 69:4 % 99:0 % 88:8 % all the considered wind directions. Conversely, despite results at 0∘ being
MNB 1:15 %  1:33 % 19:0 %  4:54 % satisfactory for both the leeward and along-wind surfaces, numerical
14:4 % 9:47 % predictions at 45∘ and 90∘ angles of attack are inaccurate in these zones,
even showing values of the spatial correlation opposite in sign with
respect to experiments. Although it is not so straightforward to identify
this aspect, for each angle of attack the spatial correlations between each
the cause of the discrepancies observed in the pressure correlations it is
probe and all the others are calculated. Then for each angle of attack the
worth pointing out that they are affected by both the aerodynamic
spatial correlations between the probe located on the windward surface,
characteristics of the body as well as by the turbulent characteristics of
Cp;corr , and all the others are analysed. The spatial correlation Cp;corr
the incoming flow, such as the spatial spectra (which can be obtained
calculated in this way is normalised with the autocorrelation function of
with a Fourier transformation performed in the space domain) and the
the probe located on the windward surface for each wind angle of attack.
turbulence length scales in the along-wind and across-wind directions,
At 45∘ angle of attack, spatial correlations of probe P1 along the paths s1

13
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 24. (a) Spatial correlation between the pressure probe P1 and all the other probes along the paths s1 and (b) s2 at 45∘ angle of attack.

Fig. 25. (a) Spatial correlation between the pressure probe P3 and all the other probes along the paths s3 and (b) s2 at 90∘ angle of attack.

which can not be directly controlled with the MDSRFG method. as inputs the pressure fields obtained from both experiments and nu-
Regarding this last aspect, the adoption of inflow turbulence generation merical simulations. Indeed, the pressure time histories are available in
methods able to accurately reproduce the inflow coherence functions the probe locations (see Fig. 9 (a)) from two different sources, which are
(Aboshosha et al., 2015; Castro et al., 2017) might improve the results in from LES simulations and from experimental measurements. Using once
terms of pressure spatial correlations. In addition to the aforementioned a at a time these two different datasets, the accuracy of LES in repro-
aspects, some parameters, like the turbulence intensities in the ducing internal forces on the structural elements is assessed. In accor-
across-wind and vertical directions, have been assumed since they are not dance with the experimental setup, also in LES simulations the model of
reported in the experimental data. All these factors can contribute to the the high-rise building is considered to be rigid, therefore aeroelastic ef-
discrepancies observed for the pressure spatial correlations plots and also fects are not taken into account in the following analyses.
to those observed for the r.m.s. of the integral force coefficients reported The characteristics of the adopted structural model are reported in
in Table 1, since these quantities are directly related to each other. Figs. 26 and 27. The structure is a steel-framed tube embedding two cores
and counts 50 floors with a regular distribution along the height. The
5. Assessment of wind loads on structural elements steel-framed tube structure represents a very efficient structural solution
for high-rise buildings since it tends to behave as an equivalent hollow
In this section linear structural dynamic analyses are performed using tube, leading to a considerable saving of material compared to classical

Fig. 26. (a) Three-dimensional view and (b) frontal view of the structural model.

14
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 27. (a) View of a characteristic floor of the structural model and (b) sectional dimensions of frame and core elements.

framed buildings (Taranath, 1988). Each floor is composed of a concrete building top for all the considered directions.
slab with a thickness of 0:45 m. At the ground floor, the end sections of Results are analysed in terms of axial forces in 12800 sections. Fig. 29
the beams are considered perfectly clamped. (a)-(c) show scatter plots for the r.m.s. of the axial force (referred to as N')
The dimensions of the cross-sections of the structural elements for the 0∘ , 45∘ and 90∘ angles of attack, respectively, while Fig. 29 (d)
constituting the framed structure and the core are varied with height shows the envelope of these three angles. In order to better analyse the
0
every 40 m according to the 5 levels shown in Fig. 26 (b). According to characteristics of the distributions of N , in accordance with what was
the nomenclature introduced in Fig. 27 (b) the dimensions of the cross- previously done for the r.m.s. of the pressure coefficient, Table 4 reports
sections for each level are reported in Table 3. for each graph of Fig. 29 the percentage of points falling within the three
The first three natural structural modes are reported in Fig. 28, while considered ranges of tolerance, together with the MNB. As can be
for the linear dynamic analyses discussed below 10 structural modes are observed, while for the 0∘ angle of attack 99% of points is within the 30%
considered. The first natural frequency is equal to 0:204 Hz and is in good range of tolerance, when 45∘ and 90∘ are considered the same metric
agreement with an empirical estimate for steel-framed structures which decreases to 22:5% and 37:3%, respectively. When these results are
can be calculated as fI ¼ 1:0=ð0:1Nf Þ ¼ 0:2 Hz (Taranath, 1988), where 0
compared to those reported for C p in Table 2, it is observed that the
Nf is the total number of floors. accuracy shown by LES in predicting the r.m.s. of the pressure coefficient
The linear structural dynamic analyses are performed by adopting the is not reflected in the r.m.s. of the axial forces. Focusing on the MNB
procedure described in Patruno et al. (2016b). Following this method- reported in Table 4 a negative value is observed for all considered angles,
0
ology the structural response is assessed by means of modal superposition indicating that LES tends to underestimate the value of N . This can be
while quasi-static corrections are introduced to take into account the noted also in Fig. 29 (a), (b) and (c), where points are concentrated near
effect of high-frequency modes. In this way static, quasi-static and reso- the lower 30% tolerance boundary. The best agreement between the
nant responses to the wind loads can be simply and efficiently assessed. 0
values of N obtained starting from experimental and numerical pressure
The design wind speed adopted for the analyses is equal to 30 m=s at the fields is observed at 0∘ , for which the absolute value of MNB is lowest and
equals  14:2%. When the envelope for the three angles of attack is
Table 3 considered differences decrease: in this case 98:2% of the points is within
Dimensions of the sections of the structural elements. the 30% tolerance and the MNB is equal to  23:9%, despite a maximum
of 33:7% observed at 45∘ . Furthermore, if the higher level of tolerance
Level Core elements Framed tube elements
is increased from 30% to 38% the percentage of points included in the
bc ½m hc ½m tc ½m bf ½m hf ½m tf ½m new tolerance range increases significantly, reaching 77% for 45∘ and
1 0.7 0.7 0.18 0.3 0.5 0.08 99:6% for 90∘ , indicating that the distribution of the normalized bias is
2 0.7 0.7 0.14 0.3 0.5 0.05 quite narrow around its mean value.
3 0.7 0.7 0.1 0.3 0.5 0.045
In all, it is worth noting that in order to accurately assess the wind
4 0.7 0.7 0.08 0.3 0.5 0.04
5 0.7 0.7 0.04 0.3 0.5 0.04
load effects on structures the correct reproduction of the local pressure

Fig. 28. First three structural modes and their natural frequencies (coloured by displacement magnitude).

15
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

0
Fig. 29. Scatter plots for the r.m.s. of the axial force (N ).

Table 4 experimental data report a negative correlation. This fact might be

0 0
Performance metrics for the r.m.s. of the axial force (N ). responsible for the observed almost systematic underestimation of N in
Performance metrics 0∘ 45∘ 90∘ Envelope these cases.
10% 23:3 % 0% 0% 0:1 %
In order to also provide quantities of interest for the design of
20% 82:9 % 0:43 % 0:5 % 8:86 % structural elements Fig. 30 (a), (b) and (c) show scatter plots for the peak
30% 99:0 % 22:5 % 37:3 % 98:2 % axial forces Npeak at 0∘ , 45∘ and 90∘ , respectively, while Fig. 30 (d) shows
MNB  14:2 %  33:7 %  30:5 %  23:9 % the envelope for the three considered wind directions. The peak axial
forces are defined as the mean values plus (and minus) 3.5 times the
0
r.m.s. (N ). It can be seen that a good agreement between experimental
coefficient statistics on the building surfaces is not sufficient, even if it and numerical data is achieved, even if some discrepancies are observed
represents a necessary first step. Indeed, the structural analysis is also at all angles of attack, in particular for lightly loaded elements.
affected by the spectral content of the pressure signals as well as their If elements loaded less than 10% of the maximum axial force in the
spatial correlations. Bearing in mind this consideration, in the analysed whole structural model are disregarded, data reported in Table 5 show
0
cases it can be observed that the numerical predictions of N show the that the number of points within the 30% range of tolerance significantly
∘ increases and the absolute values of the MNB decrease for all considered
highest accuracy for the 0 angle of attack, for which also the pressure
correlations are in very good agreement with experimental data, as wind directions. In particular, when the total envelope is considered, the
shown in Fig. 11. Conversely, Figs. 24 and 25 show that for 45∘ and 90∘ MNB for Npeak equals 8:28% which is significantly lower (in absolute
0
the pressure correlations are predicted with lower accuracy, and the terms) than the value of 23:9% obtained for N . Since the average axial
0 0
scatter plots of N are less accurate as well. In particular, as previously force is also predicted more accurately than its r.m.s. N , it appears that
observed for these wind directions, LES results show a positive pressure the agreement between numerical and experimental results improves
correlation between the windward and leeward surfaces, whereas significantly when global effects are considered.

16
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

Fig. 30. Scatter plots for the peak axial force (Npeak ).

Table 5 choice of the target turbulent length scale, and since the MDSRFG pro-
Performance metrics for the peak axial force (Npeak ) disregarding elements cedure does not allow to fully control it a priori an iterative procedure was
loaded less than 10% of the maximum load. followed until the desired turbulent length scale was obtained. Then, in
Performance metrics 0∘ 45∘ 90∘ Envelope agreement with experimental practice, LES were performed in an empty
10% 78:0 % 5:18 % 0% 54:8 %
domain, representing the wind tunnel in absence of the high-rise building
20% 89:6 % 47:1 % 5:26 % 66:0 % model. A good agreement between both inflow profiles and the experi-
30% 99:0 % 96:8 % 70:0 % 90:9 % mental measurements was obtained in terms of average velocity and
MNB 0:283 %  18:6 %  27:3 %  8:28 % turbulence intensity profiles. Then, the high-rise building was introduced
in the computational domain and two LES were performed at 0∘ angle of
attack using the two inflow conditions. Results in terms of statistics of the
6. Conclusions
pressure distributions on the high-rise building were systematically
compared with experimental measurements, and for the best performing
In the present paper the capabilities of LES as a structural design tool
inlet condition simulations at angles of attack equal to 45∘ and 90∘ were
were investigated. In particular, the turbulent flow around an isolated
also performed.
high-rise building was simulated and the numerically predicted pressure
First the local statistics of the pressure field for the considered angles
field was compared to experimental measurements, after which both
were used to assess the wind load effects on the structure. of attack were analysed. In terms of Cp a good agreement between nu-
A key point in the simulation process is represented by the inflow merical predictions and experimental measurements was achieved, with
boundary conditions generation. In order to obtain accurate results in a maximum MNB for all cases of less than 10%. With respect to the
terms of pressure distributions the characteristics of the incoming tur- fluctuating part of the pressure field LES results were shown to be slightly
bulent flow need to be represented as accurately as possible. In order to less accurate, with a maximum MNB equal to 19%. Nevertheless, in
do this, two synthetic fluctuation fields matching the target spectra were particular for 45∘ and 90∘ it was observed that the larger relative errors
0
generated by means of the MDSRFG method and introduced in the are concentrated in points with relatively small values of C p . This fact
computational domain. The thus generated fields differed only in the might suggest that secondary flow mechanisms were not predicted by

17
M. Ricci et al. Journal of Wind Engineering & Industrial Aerodynamics 180 (2018) 1–18

LES with the same accuracy as global more energetic flow mechanisms. Bruno, L., Salvetti, M.V., Ricciardelli, F., 2014. Benchmark on the aerodynamics of a
rectangular 5:1 cylinder: an overview after the first four years of activity. J. Wind
In order to study the effects of the wind loads also in terms of internal
Eng. Ind. Aerod. 126, 87–106.
forces, once the characteristics of the pressure field were analysed, a Cabral, B., Leedom, L.C., 1993. Imaging vector fields using line integral convolution. Proc.
steel-framed tube structure was considered for the building and linear SIGGRAPH 263–270.
dynamic structural analyses were performed for each angle of attack, Castro, H.G., Paz, R.R., 2013. A time and space correlated turbulence synthesis method
for Large Eddy Simulations. J. Comput. Phys. 235, 742–763.
starting from both experimental measurements and numerical pre- Castro, H., Paz, R., Mroginski, J., Sorti, M., 2017. Evaluation of the proper coherence
dictions of the pressure field. It was observed that the accuracy shown by representation in random flow generation based methods. J. Wind Eng. Ind. Aerod.
0
LES in reproducing the C p distribution was not reflected in the r.m.s. of 168, 211–227.
0 Chen, X., Kareem, A., 2005. Coupled dynamic analysis and equivalent static wind loads
the axial forces N , with a maximum MNB over all analyses equal to  on buildings with three-dimensional modes. J. Struct. Eng. 131 (7), 1071–1082.
0
33:5%. Furthermore, differently from the distributions of C p , an almost Davenport, A., 1961. The application of statistical concepts to the wind loading of
0 structures. In: ICE Proceedings, no. 6480.
systematic underestimation was obtained in terms of N . This fact might Dyrbye, C., Hansen, S., 1999. Wind Loads on Structures. Wiley, Chichester, West Sussex,
be related to the results obtained from the analysis of pressure spatial England.
correlation fields. While at 0∘ a good agreement in terms of pressure Elliott, W.P., 1958. The growth of the atmospheric internal boundary layer. Eos,
Transactions American Geophysical Union 39 (6), 1048–1054.
correlations was obtained, at 45∘ and 90∘ some discrepancies were Elshaer, A., Aboshosha, H., Dagnew, A., Bitsuamlak, G., El Damatty, A., 2016. LES
observed in particular on the leeward surfaces, where numerical and evaluation for wind-induced responses of a tall building with isolated and complex
experimental results were opposite in sign. This different flow dynamics surrounding. Eng. Struct. 115, 179–195.
0 EN 1991-1-4, 2005. Eurocode 1: Actions on Structures -Part 1-4: General Actions -Wind
might be responsible for the systematic underestimation of N , which also Actions, vol. 4. European Committee for Standardization, pp. 1–148, 2005.
becomes particularly pronounced at 45∘ and 90∘ . Nevertheless, when the Franke, J., Hellsten, A., Schlünzen, H., Carissimo, B., 2007. Best Practice Guideline for the
envelope of the three angles is considered results significantly improved, CFD Simulation of Flows in the Urban Environment, vol. 44.
Huang, S., Li, Q., Xu, S., 2006. Numerical evaluation of wind effects on a tall steel
since the effects of minor flow mechanisms were partially hidden by the building by CFD. J. Constr. Steel Res. 63, 612–627.
0
envelop of the results. In this case the MNB characterising the N distri- Huang, S.H., Li, Q.S., Wu, J.R., 2010. A general inflow turbulence generator for large
eddy simulation. J. Wind Eng. Ind. Aerod. 98 (10–11), 600–617.
bution was equal to  23:9%.
Irwin, P.A., 2009. Wind engineering challenges of the new generation of super-tall
Regarding the peak axial forces results were shown to be more ac- buildings. J. Wind Eng. Ind. Aerod. 97 (7–8), 328–334.
0
curate than those obtained for N for all considered angles of attack due Irwin, P.A., Kochanski, W.W., 1995. Measurements of structural wind loads using the high
to the fact that the mean values of the axial forces were predicted with frequency pressure integration method. In: Proceedings of the 13th Structures
Congress, pp. 1631–1634.
higher accuracy than their fluctuating parts. When the envelope of the Jeong, J., Hussain, F., 1995. On the identification of a vortex. J. Fluid Mech. 285, 69.
three angles was considered, the relative MNB was equal to 8:28% Kareem, A., 1984. Model for predicting the acrosswind response of buildings. Eng. Struct.
indicating that a satisfactory level of accuracy could be considered 6 (2), 136–141.
Kareem, A., Zhou, Y., 2003. Gust loading factor - past, present and future. J. Wind Eng.
achieved. Ind. Aerod. 91 (12–15), 1301–1328.
Summarizing it can be stated that, with a view of using LES as a design Kim, Y.C., Tamura, Y., 2014. Effects of incident flows on wind loads and their
tool for the sizing of structural elements, a comparison between experi- combinations on a low-rise building immersed in a long upstream fetch. J. Struct.
Eng. 140.
mental and numerical results in terms of pressure fields is necessary but Kim, Y., Castro, I.P., Xie, Z.T., 2013. Divergence-free turbulence inflow conditions for
not sufficient. The structural response is deeply affected also by the large-eddy simulations with incompressible flow solvers. Comput. Fluids 84, 56–68.
spectral content of the pressure field as well as by its temporal and spatial Lettau, H., 1969. Note on aerodynamic roughness-parameter estimation on the basis of
roughness-element Description.pdf. J. Appl. Meteorol. 8, 828–832.
correlations. These characteristics are directly related to both the body Lilly, D.K., 1992. A proposed modification of the germano-subgrid-scale closure method.
aerodynamics and the turbulent inflow conditions. Considering the Phys. Fluid. Fluid Dynam. 4 (3), 633–635.
complexity of the problem and bearing in mind that a large number of Noda, H., Nakayama, A., 2003. Free-stream turbulence effects on the instantaneous
pressure and forces on cylinders of rectangular cross section. Exp. Fluid 34 (3),
factors contribute to the definition of the structural response, the present
332–344.
study indicates that a satisfactory accuracy can be obtained with LES if Noda, H., Nakayama, A., 2003. Reproducibility of flow past two-dimensional rectangular
the envelope of several angles of attack is considered, in particular in cylinders in a homogeneous turbulent flow by LES. J. Wind Eng. Ind. Aerod. 91 (1–2),
terms of peak forces. In the awareness that wind tunnel tests still repre- 265–278.
Patruno, L., Ricci, M., 2017. On the generation of synthetic divergence-free homogeneous
sent the most reliable technique to assess the structural response to the anisotropic turbulence. Comput. Meth. Appl. Mech. Eng. 315, 396–417.
wind action, the obtained results are considered to be encouraging for Patruno, L., Ricci, M., De Miranda, S., Ubertini, F., 2016. Numerical simulation of a 5:1
pursuing further research which is still needed in order to develop the use rectangular cylinder at non-null angles of attack. J. Wind Eng. Ind. Aerod. 151,
146–157.
of LES as a reliable tool for structural design. Patruno, L., Ricci, M., de Miranda, S., Ubertini, F., 2016. An efficient approach to the
evaluation of wind effects on structures based on recorded pressure fields. Eng.
Acknowledgements Struct. 124, 207–220.
Randerson, D., 1984. “Atmospheric Science and Power Production,” DOE Report.
Ricci, M., Patruno, L., de Miranda, S., Ubertini, F., 2016. Effects of low incoming
The authors are thankful to CINECA for providing the HPC facilities turbulence on the flow around a 5 : 1 rectangular cylinder at non-null-attack angle.
needed to accomplish the present research work. Math. Probl Eng., 2302340, 12 pages. https://doi.org/10.1155/2016/2302340.
Ricci, M., Patruno, L., de Miranda, S., 2017. Wind loads and structural response:
benchmarking LES on a low-rise building. Eng. Struct. 144, 26–42.
References Taranath, B.S., 1988. Structural Analysis and Design of Tall Buildings. McGraw-Hill.
Tokyo Polytechnic University Aerodynamic Database, 2003. http://www.wind.arch.t-
Aboshosha, H., Elshaer, A., Bitsuamlak, G., El Damatty, A., 2015. Consistent inflow kougei.ac.jp/system/eng/contents/code/tpu.
turbulent generator for LES evaluation of wind-induced responses for tall buildings. Tominaga, Y., Mochida, A., Yoshie, R., Kataoka, H., Nozu, T., Yoshikawa, M.,
J. Wind Eng. Ind. Aerod. 99, 272–284. Shirasawa, T., 2008. AIJ guidelines for practical applications of CFD to pedestrian
Architectural Institute of Japan, 2004. AIJ Recommendations for Loads on Buildings. wind environment around buildings. J. Wind Eng. Ind. Aerod. 96 (10–11),
Blocken, B., 2014. 50 years of computational wind engineering: past, present and future. 1749–1761.
J. Wind Eng. Ind. Aerod. 129, 69–102. Tschanz, T., Davenport, A.G., 1983. The base balance technique for the determination of
Blocken, B., Stathopoulos, T., Carmeliet, J., 2007. CFD simulation of the atmospheric dynamic wind loads. J. Wind Eng. Ind. Aerod. 13 (1–3), 429–439.
boundary layer: wall function problems. Atmos. Environ. 41 (2), 238–252. Weller, H., 2012. Controlling the computational modes of the arbitrarily structured C
Braun, A., Awruch, A., 2009. Aerodynamic and aeroelastic analyses on the CAARC grid. Mon. Weather Rev. 140, 3220–3234.
standard tall building model using numerical simulation. Comput. Struct. 87 (9–10), Xie, J., Irwin, P.A., 1998. Application of the force balance technique to a building
564–581. complex. J. Wind Eng. Ind. Aerod. 77–78, 579–590.
Bruno, L., Fransos, D., Coste, N., Bosco, A., 2010. 3D flow around a rectangular cylinder: a Zhang, Y., Habashi, W.G., Khurram, R.A., 2015. Predicting wind-induced vibrations of
computational study. J. Wind Eng. Ind. Aerod. 98 (6–7), 263–276. high-rise buildings using unsteady CFD and modal analysis. J. Wind Eng. Ind. Aerod.
136, 165–179.

18