J. Wind Eng. Ind. Aerodyn.

104–106 (2012) 88–97

Contents lists available at SciVerse ScienceDirect

Journal of Wind Engineering

and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

Full-scale performance evaluation of tall buildings under wind

Rachel Bashor, Sarah Bobby n, Tracy Kijewski-Correa, Ahsan Kareem
University of Notre Dame, Notre Dame, IN, USA

a r t i c l e i n f o a b s t r a c t

Available online 26 May 2012 Full-scale monitoring is an important tool to evaluate the performance of dynamically sensitive
Keywords: structures and assure that the design assumptions and methodologies used to predict their responses
Dynamic response yield accelerations that are consistent with in-situ behaviors. Unfortunately, such sustained monitoring
Damping is not a common practice for tall buildings, prompting the establishment of the Chicago Full-Scale
High-rise Monitoring Program (CFSMP), which has spent a decade monitoring three tall buildings embodying
Full-scale measurements structural systems common to high-rise design. This paper presents a cross section of this database for
the two-fold purpose of evaluating the in-situ dynamic properties and comparing them to design values
and then using these properties to predict RMS accelerations’ levels based on wind tunnel test data to
evaluate their consistency with full-scale observations. Although the responses observed are low
amplitude in nature, some important observations can be made, including the range of wind speeds for
which RMS accelerations are more accurately predicted and that conservatism in these predictions is
achieved more consistently for the concrete building in the program. This paper also provides further
evidence to demonstrate the dependence of critical damping ratios on the structural system employed,
the amplitude sensitivity of dynamic properties, and the types of systems for which dynamic properties
are more accurately predicted.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction Full-scale monitoring provides the most faithful means to validate

The existing design procedures for tall buildings rely exclu-
sively on computational and scaled experimental models tested in
wind tunnels. Although these models have been extensively
refined, there is still considerable uncertainty associated with
the actual performance of these structures under dynamic excita-
tion. As structures become increasingly taller and more complex,
it is imperative that engineers be able to accurately predict their
performance in order to create cost-effective designs satisfy-
ing survivability, serviceability, and habitability requirements.
However, in the case of tall buildings, habitability requirements
— accelerations adversely affecting the comfort of the building
occupants — often become the governing limit state for design. As
such there is considerable interest in better quantifying not only
the levels of acceleration that cause discomfort to occupants, but
also in enhancing the reliability of predicted accelerations in the
design process (Bentz and Kijewski-Correa 2009). The reliability
of these design predictions rests squarely on the procedures
to predict accelerations through wind tunnel testing and the
dynamic properties supplied by designers as part of this process.
n
Corresponding author. Tel.: þ1 574 631 2540.
E-mail addresses: rbashor@eagle.org (R. Bashor), sbobby@nd.edu (S. Bobby),
tkijewsk@nd.edu (T. Kijewski-Correa), kareem@nd.edu (A. Kareem).
both of these aspects. Unfortunately, access to buildings for full-
scale validation of design predicted accelerations has been quite
limited (e.g., Littler, 1991; Li et al., 2004), and while there have
been greater efforts to catalog in-situ dynamic properties (e.g.,
Lagomarsino, 1993; Satake et al., 2003), data from buildings of
significant height is generally lacking. As a result, there has been
growing attention in full-scale monitoring of tall buildings, with
the addition of several tall buildings in China, e.g., Di Wang Tower
and the Bank of China (Li et al., 2003a, 2003b, 2005; Xu et al.,
2003). Unfortunately, most of the full-scale monitoring efforts
that have been published in the literature have not been
sustained long enough to observe responses under a wide
spectrum of wind events necessary to fully validate design
practices. In response to this need, nearly a decade ago, the
University of Notre Dame, The Boundary Layer Wind Tunnel
Laboratory (BLWTL) at the University of Western Ontario
(UWO), and Skidmore, Owings, & Merrill LLP (SOM) in Chicago
established the Chicago Full-Scale Monitoring Program (CFSMP)
to monitor three tall buildings in Chicago with the goal of
correlating the measured responses and dynamic properties of
the buildings with their predicted behavior under a range of wind
environments. CFSMP is a unique effort as it has accumulated a
decade of data for three signature tall buildings whose structural
systems are representative of those common to high-rise design.
While estimates of the dynamic properties of these buildings

0167-6105/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.jweia.2012.04.007
 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97
under select wind events have been reported in previous publica-
tions (e.g., Kijewski-Correa et al., 2006; Kijewski-Correa and Pirnia,
2007; Pirnia et al., 2007; Bentz and Kijewski-Correa, 2009) and
select events have been used to validate wind tunnel predicted
accelerations (e.g., Kijewski-Correa et al., 2006; Kijewski-Correa and
Kochly, 2007), this paper explores a larger cross section of recorded
datasets to explore the variability and potential amplitude-depen-
dence in in-situ dynamic properties for the fundamental sway
modes of these buildings using established system identification
techniques and then, using these in-situ dynamic properties, com-
pares predicted wind tunnel RMS accelerations to the values
observed in full-scale for these events.
2. Description of instrumented buildings
To ensure that this program continues to collect valuable data,
at the request of the building ownership, the structures’ identities
must remain anonymous so they are referred to herein gener-
ically as Buildings 1–3. These buildings are all located in down-
town Chicago and are representative of structural systems
common to high-rise design, i.e., buildings standing 60 stories
or higher. As discussed in Kijewski-Correa et al. (2006), each
building utilizes straight shaft reinforced concrete caissons
extending to bedrock. The primary lateral load-resisting system
of Building 1 is comprised of a steel tube with exterior columns,
spandrel ties, and additional stiffening elements. This allows a
near uniform load distribution on the columns across the flange
face with very little shear lag so that the building is dominated by
cantilever action. Building 2 is a reinforced concrete building with
shear walls located near the core of the building to provide lateral
load resistance. At two levels the core is tied to the perimeter
columns via reinforced concrete outrigger walls to control the
wind drift and reduce the overturning moment in the core shear
walls. Finally, Building 3 consists of a steel, moment-connected,
framed tube comprised of closely spaced wide columns and deep
spandrel beams along multiple frame lines. A summary of the
design-specified fundamental frequencies (fx, fy) and critical
damping ratios (zx, zy) for the x and y axes of the buildings is
found in Table 1. Note that the each building’s x and y axes are
assumed to positively align with cardinal directions of East and
North, respectively. A more complete description of the finite
element models that generated these frequencies was provided
previously in Kijewski-Correa et al. (2006), while damping values
were assumed by the designers based on general rules of thumb.
3. Description of instrumentation
3.1. Instrumentation system for building response
As discussed in Kijewski-Correa et al. (2006), the primary
instrumentation systems were installed in Buildings 1, 2, and
3 on June 14, 2002, June 15, 2002, and April 30, 2003, respectively.
Each building is instrumented with four Columbia SA-107 LN
high-sensitivity force-balance accelerometers that are capable of
accurately measuring accelerations down to 0 Hz with a 15 V/g
Table 1
Design-predicted dynamic properties in fundamental sway modes.
Property Building 1 Building 2
fx (Hz) 0.201 0.148
fy (Hz) 0.142 0.156
zx, zy 1.0% 1.0%*
n
1.0% damping used for serviceability and 1.5% used for strength design.
89
sensitivity. The accelerometers were installed in orthogonal pairs
at opposite corners of the ceiling at the highest possible floor of
each building, noted schematically in Fig. 1. The data outputs of
the accelerometers are sampled every 0.12 s by Campbell CR23X
data loggers to yield an overall system resolution of approxi-
mately 0.001 milli-g. The data logger is programmed to capture
continuous 10 min statistics from the accelerometers, and when
motions exceed a user-defined threshold, the data logger auto-
matically begins to capture continuous hour-long time-histories
of data for as long as the threshold level is exceeded. Estimates of
the sway and torsional responses are obtained via algebraic
manipulation of the accelerometer outputs. While Building 1 also
employs a high-precision global positioning system to monitor its
displacements, this data will not be investigated in the present
study, though additional details can be found in Kijewski-Correa
and Kochly (2007).
3.2. Instrumentation system for wind field characteristics
Wind speeds referenced in this paper are taken from one of the
National Oceanic and Atmospheric Administration Great Lakes
Environmental Research Laboratory (NOAA GLERL) Young 5103 V,
propeller-type sensors. This sensor is located approximately
4.83 km offshore of Chicago on the Harrison–Dever crib and
samples wind velocity at 1 Hz and records 5 min averaged
statistics (NOAA GLERL, 2006). Gradient wind speeds referenced
in this study are estimated from these surface wind speeds using
a procedure described in Kijewski-Correa et al. (2006), which
assumes gradient height to be 300 m over open water and uses
methods to account for the influence of terrain roughness and
fetch (ESDU, 2001). A minor correction to the wind azimuth is
also applied to account for the rotation of the velocity vector with
height (Davenport, 1987).
4. System identification techniques
In order to maintain consistency with previous published
analyses of data from the CFSMP, as described in Kijewski-
Correa et al. (2006), frequency domain estimates will be executed
on low-bias (less than 2%) power spectra using the Half Power
Data Logger Accelerometers
GPS Receiver
Y
X
N
Building 3
0.131
0.131
1.0% GPS Data Acquisition Anemometers
Fig. 1. Generalized location of instrumentations on buildings in CFSMP.
90 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97

Band Width (HPBW) Technique, while the time domain analysis Table 3
employs Analytic Signal Theory using Hilbert Transforms Estimated dynamic properties for wind event 2.
applied to locally averaged Random Decrement Signatures (RDS)
Direction Test Building 1 Building 2 Building 3
(Kijewski and Kareem, 2003). This implementation of the Random
Decrement Technique (RDT) begins with pre-processing by fn (Hz) z (%) fn (Hz) z (%) fn (Hz) z (%)
Butterworth bandpass filters to isolate each mode of interest
and a positive point trigger value is enforced to select the x-Sway HPBW 0.204 1.37 0.178 1.66 0.117 1.59
RDT 0.204 0.89 0.178 1.52 0.117 1.01
segments of the response averaged to generate the RDS (Bashor (CoV, %) (0.13) (12.61) (0.31) (11.33) (0.15) (10.18)
et al., 2005). As RDT is inherently sensitive to the trigger condi-
y-Sway HPBW 0.141 0.88 0.176 2.53 0.117 2.01
tions, which directly influence the number of segments captured
RDT 0.141 1.00 0.176 2.95 0.116 1.44
and thereby the quality of the RDS, repeated triggering is (CoV, %) (0.13) (6.43) (0.96) (5.54) (0.34) (24.20)
implemented, as proposed by Kijewski-Correa (2003) and pre-
viously implemented in the context of the CFSMP by Bashor et al.
(2005) and Kijewski-Correa and Pirnia (2007). This is accom- 5.1. Performance of system identification techniques
plished by generating a suite of RDSs associated with a range of
positive point triggers that are within a few percent of the desired For both wind events the natural frequency estimates are
trigger Xp. The resulting RDSs are then processed using the Hilbert completely consistent between the time and frequency domain
Transform and the natural frequency and the critical damping techniques, with RDT CoVs less than 1%. When comparing values
ratio are determined from the phase and amplitude of the analytic between the wind events, the natural frequencies diminish
signal, respectively. The resulting vector of frequency and damp- slightly in the y-sway response of Building 2 in the second event.
ing estimates are then averaged to yield mean estimate and Note that for this event, the y-axis experiences acrosswind action
corresponding coefficient of variation (CoV). The reliability of and comparatively larger responses; therefore, the reduction in
these frequency and time domain approaches for system identi- frequency is consistent with the amplitude dependence noted in
fication from wind-induced vibration data has been previously previous studies (Kijewski-Correa and Pirnia 2007). On the other
evaluated by Kijewski and Kareem (2002). hand, critical damping ratios estimated by RDT have CoVs that are
Both of the aforementioned approaches assume stationarity of one to two orders of magnitude higher than those associated with
the data, to varying extents. Although in practice wind is often natural frequency estimates. In fact, the CoVs are larger for the
viewed as a stationary random process, transient or nonstationary steel buildings, particularly Building 3, which is the building with
features are generally present in most field data. Therefore, before the strongest degree of coupling between modes. Interestingly,
data is processed by any of the aforementioned techniques, its the time and frequency domain system identification approaches
stationarity is established using the Run and Reverse Arrange- are most consistent in their damping estimate for the concrete
ments Tests (Bendat and Piersol, 2000). In addition to these two structure (within 14%) and show the most significant deviation
tests, additional verifications are made using a method proposed for the steel structures whose power spectra are more narrow-
by Montpellier (1996). band and generated with fewer spectral averages for a fixed
duration wind event. Furthermore, given the low bias require-
ment placed on the estimation of the power spectra, it is not
5. In-depth study of two wind events surprising that the damping estimates in the frequency domain
are not consistently larger than the unbiased time domain
Before exploring the trends in dynamic properties over multi- estimates, affirming that the residual error source is indeed
ple wind events, two wind events are selected for in-depth random. When comparing damping values between the two
discussion to evaluate the performance of the system identifica- events, Building 2 yields the most consistent damping values,
tion techniques being employed. It should be noted that con- regardless of the method employed, with HPBW results being
firmation of stationarity by at least two of the three tests within 22% and RDT results being within 18%. For Building 1,
discussed in the previous section was executed to qualify the while RDT results are quite consistent (within 12% between
two wind events featured here as stationary; the same criteria events), HPBW results deviate by as much as 50%.
will be used for all data presented in later sections of the paper. The case is similar for Building 3, where RDT damping results
The identified natural frequencies and critical damping ratios are are within 19% of one another, while HPBW results deviate by as
respectively presented in Tables 2 and 3 for Wind Events 1 and much as 48%. This again can be credited to the fact that the two
2 with respective mean hourly gradient winds and average wind steel buildings are characterized by considerably more narrow-
directions of 20 m/s and 2251 (SW) and 24 m/s and 2881 (WNW). band spectra and are thereby more susceptible to variance errors
The dynamic properties estimated by the RDT with local trigger- in the presence of limited amounts of data. Interestingly, while
ing are accompanied by their CoV to provide an indicator of RDT proves to be the more consistent damping estimator, parti-
relative reliability of the estimate. cularly for the two steel buildings, when comparing results
between the two events, the consistency is generally an order of
Table 2 magnitude better in the x-axis than the y-axis, which again
Estimated dynamic properties for wind event 1. experiences higher amplitude acrosswind response in Event 2.
As a result the lack of ‘‘consistency’’ may not be the result of
Direction Test Building 1 Building 2 Building 3
errors inherent to the method but potentially due to the ampli-
fn (Hz) z (%) fn (Hz) z (%) fn (Hz) z (%) tude-dependence previously observed in damping values in these
buildings (Kijewski-Correa and Pirnia, 2007). In particular, in non-
x-Sway HPBW 0.204 0.65 0.178 1.62 0.116 1.46 symmetric systems, the axes of the buildings typified by greater
RDT 0.204 0.87 0.178 1.42 0.116 1.04
frame action tend to manifest more amplitude dependence in
(CoV, %) (0.10) (23.88) (0.22) (7.43) (0.25) (20.63)
their dynamic properties: Building 1’s x-axis as a result of
y-Sway HPBW 0.141 1.14 0.177 2.07 0.116 1.06 potential shear lag along the elongated floor plate and Building
RDT 0.141 0.88 0.177 2.41 0.116 1.21
(CoV, %) (0.19) (8.89) (0.68) (8.01) (0.14) (22.96)
2’s y-axis where primary lateral resistance is derived from
slab and frame elements. The potential effects of amplitude
 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97
dependence may be evidenced by the fact that RDT damping
values in the y-axis of all three buildings are larger in Event 2 than
Event 1.
In previous analysis of the data from these buildings, damping
values similar to the RDT results reported here were observed for
Building 1, which tends to manifest comparable levels of damping
on both axes. In the case of Building 2, y-sway damping values
were previously observed to be larger than x-sway (Kijewski-
Correa and Pirnia, 2007), and this was affirmed by the RDT results
in this study. While Building 2 would be generally expected to
have greater damping than Building 1, by virtue of its use of
concrete, what is more interesting is the stark difference in
damping values between the two axes of this concrete building.
Reasons for this variation in damping along the two axes of this
building were explored in Kijewski-Correa and Pirnia (2007) for
this building and for other buildings in Erwin et al. (2007) and
Bentz and Kijewski-Correa (2008). These studies have demon-
strated that structural systems with greater degrees of frame
action tend to dissipate more energy than systems that are
dominated by cantilever action (member level axial deforma-
tions). In Building 2, shear walls and outriggers engage the
exterior columns to achieve global cantilever action in the
x-direction, whereas the y-direction is dominated by compara-
tively more frame action as the beams and slabs are the primary
mechanisms to engage the lateral resistance of the building.
Therefore the in-situ observations reported herein are consistent
with the hypothesis that frame-dominated systems yield higher
levels of damping. Even when comparing the RDT damping levels
in the two steel buildings, one may hypothesize that Building 1,
which has been observed to be dominated by cantilever action as
an essentially pure tube (Bentz and Kijewski-Correa, 2008) would
have less damping than Building 3, whose panel zone shear
deformations have been extensively studied (Bentz et al., 2010).
In total these observations help to provide a more rational basis
for the levels of damping to be assumed in design, as opposed to
the crude respective assumptions of 1% or 2% critical for steel and
concrete.
5.2. Comparison of results with finite element models
One of the primary objectives of the CFSMP is to validate
assumptions made in the development of finite element models
used in design by comparing their predictions and in-situ values.
Comparisons of the results in Tables 2 and 3 to the design
predictions in Table 1 affirms that Building 1’s in-situ funda-
mental sway frequencies show excellent agreement with the
design predictions, which may be explained by the fact that, as
mentioned previously, the structure’s elements are engaged
primarily axially as a structural tube and may thereby be less
0.21
0.209
0.208
Frequency (Hz)
0.207
0.206
0.205
0.204
0.203
0.202
0 0.2 0.4 0.6
Normalized Acceleration
Fig. 2. Natural frequency estimates for Building 1 in fundamental x-sway mode (left) and y-sway mode (right).
91
susceptible to uncertainties in modeling material, section or
connection details. Building 2 is 20% stiffer in-situ in the x-
direction and 13% stiffer in-situ in the y-direction than originally
predicted by the finite element model, while the converse is true
for Building 3 (11% softer in-situ), reaffirming observations in
previous studies (Kijewski-Correa et al., 2006). Moreover, the in-
situ values for Building 2 suggest a greater consistency between
the fundamental mode frequencies in sway than predicted by
finite element models. A variety of sources have been and
continue to be explored to determine the causes of this discre-
pancy, from rigid off sets in connection modeling, to the influence
of panel zones, to the assumptions regarding in-situ material
properties and the degree of cracking in concrete elements
(Kijewski-Correa et al., 2006; Bentz et al., 2010).
6. Analysis of overall trends in dynamic properties
In the following, several hundred wind events were identified
as stationary, i.e., having 80% of triggered data in a given event
pass the aforementioned tests, and were analyzed using the same
system identification approaches discussed previously. The num-
ber of records available for analysis varies for each building,
depending on the number of times it triggers. In this study, 500
events will be analyzed for Building 1 and 200 for each of
Buildings 2 and 3—events are defined as those having at least
five triggered 1 h time histories.
6.1. Overall trends in frequency estimates
The natural frequency estimates for Buildings 1, 2, and 3 from
both HPBW and RDT are shown in Figs. 2, 3 and 4, respectively.
For reference, the natural frequency assumed for the finite
element model is indicated in the figures by a thick horizontal
line except in instances where discrepancies between in-situ data
and predictions are so great that they cannot be reasonably
shown on the same figure. It should be noted that the amplitude
of responses for each of the buildings varies widely, as does the
number of triggered events displayed. While more pronounced
scattering may be evident with increasing amplitude, particularly
with Building 2, this cannot be fully concluded given the limited
amount of data. Still, clear evidence of amplitude dependence and
softening with increased response is displayed for all three
buildings, consistent with initial observations from a narrower
subset of CFSMP data in Kijewski-Correa and Pirnia (2007).
Interestingly, in some cases, though recognizing the limited
extent of the amplitude ranges available for analysis here, the
frequency softening appears to plateau (see Building 1 x-axis and
both axes of Building 2), whereas in the case of Building 3, there is
0.146
HPBW
RDT
0.145 FEM
0.144
0.143
0.142
0.141
0.14
0.8 0 0.5 1 1.5 2 2.5
Normalized Acceleration
92 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97

0.185 0.19 HPBW

0.188 RDT
0.184 FEM
0.183 0.186
0.184
Frequency (Hz)
0.182
0.182
0.181
0.18
0.18
0.178
0.179 0.176
0.178 0.174
0.177 0.172
0.176 0.17
0 0.2 0.4 0.6 0.8 0 0.1 0.2 0.3 0.4 0.5
Normalized Acceleration Normalized Acceleration

Fig. 3. Natural frequency estimates for Building 2 in the in fundamental x-sway mode (left) and y-sway mode (right). The FEM estimated natural frequency for x-sway
(not shown) is 0.148 Hz.

0.122 0.124
HPBW
0.123 RDT
0.121 FEM
0.122
0.12
Frequency (Hz)

0.121
0.119 0.12
0.118 0.119
0.118
0.117
0.117
0.116 0.116
0.115 0.115
0 1 2 3 4 5 0 1 2 3 4 5
Normalized Acceleration Normalized Acceleration
Fig. 4. Natural frequency estimates for Building 3 in the fundamental x-sway mode (left) and y-sway mode (right).

0.185 0.19
HPBW HPBW
0.184 RDT 0.188 RDT
0.183 0.186
0.184
0.182
Frequency (Hz)

0.182
0.181
0.18
0.18
0.178
0.179
0.176
0.178 0.174
0.177 0.172
0.176 0.17
10/2002 6/2003 3/2004 11/2004 7/2005 3/2006 11/2006 8/2007 10/2002 6/2003 3/2004 11/2004 7/2005 3/2006 11/2006 8/2007
Date of Event Date of Event

Fig. 5. Fundamental frequency estimates for Building 2 plotted against date of event for x-sway mode (left) and y-sway mode (right).

clear evidence of a strongly linear and decreasing trend with no
apparent plateau over the range of amplitudes considered.
As discussed in the previous section, the discrepancies between
predicted and in-situ natural frequencies for Building 2 could be
attributed to numerous modeling assumptions. For example, given
the age of the building, it is quite likely that the degree of cracking
included in finite element models intended to represent the
structure at key design limit states are not yet realized in the
structure at present. To observe if the structure’s frequencies have
reduced with time as the result of the natural process of cracking,
Building 2’s frequency estimates are presented with time in Fig. 5.
While there is potentially a slight softening evident from this
figure, given the comparatively limited time span of these observa-
tions in comparison with the expected life cycle of the building, a
definitive conclusion on the evolution of permanent softening
cannot be reached.
6.2. Overall trends in damping ratios
The resulting damping values for Buildings 1, 2 and 3 are
presented in Figs. 6, 7 and 8, respectively. As expected, the damping
estimates show significantly more scatter than the frequency
estimates for all three buildings and, consistent with the observa-
tions in Section 5, the scatter is more pronounced for HPBW. This of
course emphasizes the importance of evaluating damping over a
suite of events. To facilitate discussion, the average damping values
for each building are reported in Table 4. The average damping
estimate for Building 1 is within approximately 10% for the two
 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97 93

2.00% 2.00%
HPBW
1.80% 1.80% RDT
1.60% 1.60%
Damping (%) 1.40% 1.40%
1.20% 1.20%
1.00% 1.00%
0.80% 0.80%
0.60% 0.60%
0.40% 0.40%
0.20% 0.20%
0.00% 0.00%
0 0.2 0.4 0.6 0.8 0 0.5 1 1.5 2 2.5
Normalized Acceleration Normalized Acceleration

Fig. 6. Damping estimates for Building 1 for the fundamental sway mode in x-direction (left) and y-direction (right).

3.00% 4.50%
HPBW
4.00% RDT
2.50%
3.50%
2.00% 3.00%
Damping (%)

2.50%
1.50%
2.00%
1.00% 1.50%
1.00%
0.50%
0.50%
0.00% 0.00%
0 0.2 0.4 0.6 0.8 0 0.1 0.2 0.3 0.4 0.5
Normalized Acceleration Normalized Acceleration

Fig. 7. Damping estimates for Building 2 for the fundamental sway mode in x-direction (left) and y-direction (right).

3.00% 3.50%
HPBW
2.50% 3.00% RDT

2.50%
2.00%
Damping (%)

2.00%
1.50%
1.50%
1.00%
1.00%
0.50% 0.50%

0.00% 0.00%
0 1 2 3 4 5 0 1 2 3 4 5
Normalized Acceleration Normalized Acceleration

Fig. 8. Damping estimates for Building 3 for the fundamental sway mode in x-direction (left) and y-direction (right).

Table 4
isolation, which is consistent with the findings of Kijewski-Correa
Average in-situ critical damping ratios for fundamental sway modes.
and Pirnia (2007); though admittedly larger amplitude results
Building 1 Building 2 Building 3 would need to be observed to further confirm this trend.
In the case of Building 2, mean-sense agreement between the
HPBW (%) RDT (%) HPBW (%) RDT (%) HPBW (%) RDT (%)
two methods is quite satisfactory: x-sway results are within 10% of
zx 0.91 0.82 1.4 1.3 1.5 1.3 one another and y-sway results are identical. While HPBW again
zy 0.87 0.98 2.2 2.2 1.8 1.7 shows the greater degree of scatter, the mean-sense agreement of
the damping estimates suggests that both approaches were fairly
unbiased estimators. Rationale for the comparatively larger damp-
methods, despite the comparatively higher variance in HPBW ing value on the y-axis was previously offered in Section 5 and in
estimates, and remains slightly less than the 1% assumed in design. Kijewski-Correa and Pirnia (2007). Again this supports the obser-
Moreover for this building, which has the largest amount of data vation that systems governed by cantilever action have lower
available for analysis, there is some evidence of increasing damping damping values and highlights the importance of determining
with amplitude, particularly when viewing the RDT results in appropriate damping estimates based on structural system as well
94 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97

as material type. It is also important to note that the amplitudes of
response for the events analyzed for Building 2 are comparatively
smaller than those of the steel buildings. As discussed in Kijewski-
Correa and Pirnia (2007), at low amplitudes, RDT results can bias
toward higher damping estimates due to the ‘‘noise floor’’ in low
amplitude response data. Once the acceleration levels exceed this
‘‘noise floor,’’ the samples triggered in the RDT process more
rapidly converge to a stable Random Decrement Signature to
yield more reliable damping estimates. Therefore the degree of
scatter evidenced particularly in Building 2 may be driven by this
phenomena and with the availability of more data at RMS accel-
eration levels of 0.5 milli-g’s or greater, a more definitive estimate
of damping and any amplitude dependence can be ascertained.
Similarly in Building 3, the agreement between methods in a
mean-sense is quite reasonable, with average critical damping
ratios within 15% for x-sway and within 6% for the y-sway. Again,
although both Buildings 1 and 3 are steel structures, the esti-
mated in-situ damping values for Building 3 have been observed
to be higher than those estimated for Building 1 due to the
additional energy dissipation mechanisms afforded by systems
with greater panel zones and comparatively greater reliance on
distributed frame action discussed previously (Kijewski-Correa
and Pirnia, 2007; Bentz and Kijewski-Correa, 2008). Additionally,
there is substantial scatter in the low frequency range, even for
RDT, which may again be tied to some of the challenges men-
tioned previously for low amplitude responses.
6.3. Interpretation of results and implications for the practice
The significant scatter seen in the results can be attributed to
numerous factors, including temperature effects, features of the
wind induced response, and errors inherent in the analysis tech-
niques, among others. The determination of variability due to site
conditions (e.g. temperature effects) is beyond the scope of this
study. However, it is important to note that the established
that the results fail to benefit from analysis by a trained profes-
sional. Thus identification of outliers and intricacies in data
manipulation normally benefiting from computer analysis
coupled with engineering judgment may be overlooked, and
significant scatter can occur, in part due to the errors inherent
in the analysis techniques, as discussed previously. The results
presented must therefore be interpreted in a mean sense, and
ongoing work be performed to quantify the variability and to
explore additional sources of variability.
6.4. Comparison of predicted and measured response
One purpose of the Chicago Full-Scale Monitoring Program is to
compare wind tunnel response predictions with the responses
measured in full-scale on a continuous basis under a variety of
stationary wind conditions. High-frequency base balance (HFBB)
tests for the CFSMP were performed at the Boundary Layer Wind
Tunnel Laboratory (BLWTL) at the University of Western Ontario
(UWO) with the approaches employed described more fully in
Kijewski-Correa et al. (2006). These wind tunnel results were used
to estimate the response of each of the buildings for the gradient
wind speed and direction extrapolated from the NOAA sensor for a
given event that triggered full-scale data. An uncoupled analysis was
performed for Building 1, while a coupled analysis was required for
both Buildings 2 and 3. In both the uncoupled and coupled analyses
the Equivalent Static Wind Loads (ESWL) and corresponding build-
ing response are determined using the full-scale spectrum of
the base moment (SM) and the rms value of the spectrum (sM),
calculated as follows using scaling laws and wind tunnel data:
!  
f USM ðf Þ s2M
ðSM ðf ÞÞf s ¼ 2
ð1Þ
sM f fs
ms
s 
M 0
ðsM Þf s ¼ ðM Þf s ð2Þ
M0 ms

analysis techniques used in this paper admittedly have inherent where

errors that will cause variability in the estimated dynamic proper-    
fB U
ties, particularly in the damping estimates. While it is known that ðf Þf s ¼ ð3Þ
U ms B f s
high variability can be seen in the results from such analyses, the
0
methods used in this paper are widely used in practice. Therefore, and f is the frequency, M is the reference moment, B is the building
it is important to acknowledge the limitations of the techniques width, and U is the wind velocity. The resonant moment, resonant
and the associated impact on the estimates of the dynamic ESWL, and rms acceleration for the uncoupled analysis are then
properties particularly when a large number of events are analyzed determined using a mode shape correction factor of 0.7 and the
using bulk processing, as is the case in this paper. The levels of following equations, respectively:
variability have been quantified for both the frequency and time rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
p
domain approaches used in this study (Kijewski and Kareem, 2002; MR ¼ sCM M0f s ðSM ðf ÞÞf s ð4Þ
4z
Pirnia, 2009). The accuracy of the spectral analysis technique used
for this analysis is dependent on the smoothness of the PSD. 8
> mðiÞjj ðiÞ
Since HPBW is a point estimator, a jagged spectral peak is known > M R P mðiÞjj ðiÞzðiÞ
>
>
for sway
< i
to cause variability in the determined properties, and is particu- PR ðiÞ ¼ ð5Þ
> IðiÞj ðiÞ
larly notable in the damping estimates. While increasing the >
> M R P IðiÞjj ðiÞ for torsion
>
: j
number of segment averages does reduce the jaggedness, the bias i

associated with the fixed spectral resolution does not improve 8P

when more than 100 averages are used (Pirnia, 2009). RDT is also > P R ðiÞUjj ðiÞ
>
> P
known to give higher CoV values when limited amounts of data are >
>
>
i
2 jj ðiÞ for sway
< i mðiÞUjj ðiÞ
available. Averaging a large number of RDSs will reduce the sy€ ¼ P P ðiÞUj ðiÞ ð6Þ
deviation in the results, particularly in the damping estimates. >
> R j
>
> P 2 ðiÞ jj ðiÞ for torsion
i
>
> j
However, there are still variance issues for structures with long : IðiÞU j
i
periods and low damping such as the buildings monitored in this
study (Kijewski and Kareem, 2002; Pirnia, 2009). where sCM is the rms value of the non-dimensionalized moment
Typically full-scale data is presented on specific significant spectrum, z is the damping ratio, m(i) is the mass at floor i, I(i) is
events in publications. This paper also focuses on the presentation the moment of inertia at floor i, and fj(i) is the jth mode shape at
of data from many wind events; thus, it was necessary to perform floor i. The coupled analysis becomes more involved as it incorpo-
automated bulk processing on the data. While this provides the rates the loading cross-spectra determined between different
opportunity to present valuable data obtained over years of directions estimated from the coupled response of the building.
monitoring, the automated fashion of the analysis also indicates Details of this approach can be found in Chen and Kareem (2005)
 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97 95

1 1
Measured Measured

Normalized Acceleration
Low Prediction Low Prediction
High Prediction High Prediction
0.1 0.1

0.01 0.01

0.001 0.001
10 100 10 100
Wind Speed (mph) Wind Speed (mph)

Fig. 9. Comparison of full-scale measurements with wind tunnel predictions for Building 1 for y-sway (left) and x-sway (right) as a function of gradient wind speed.

1 1
Normalized Acceleration

Measured Measured
Low Prediction Low Prediction
High Prediction High Prediction
0.1
0.1
0.01

0.001 0.01
10 100 10 100
Wind Speed (mph) Wind Speed (mph)

Fig. 10. Comparison of full-scale measurements with wind tunnel predictions for Building 2 for x-sway (left) and y-sway (right) as a function of gradient wind speed.

1 1
Measured Measured
Normalized Acceleration

Low Prediction Low Prediction

High Prediction High Prediction

0.1 0.1

0.01 0.01
10 100 10 100
Wind Speed (mph) Wind Speed (mph)

Fig. 11. Comparison of full-scale measurements with wind tunnel predictions for Building 3 for x-sway (left) and y-sway (right) as a function of gradient wind speed.

and Bashor (2011). The generalized mass and stiffness for each Table 5
mode are then determined and used to calculate the rms deflection Range of gradient wind speeds for which wind tunnel predicted RMS accelerations
for each mode. The analysis techniques employed are described in correlate best with full-scale observations.
greater detail in Bashor (2011). While the in-situ frequency and
Lower wind speed (mph) Higher wind speed (mph)
damping values are used in this analysis in attempt to remove
errors due to inaccurate modeling assumptions, as the previous Building 1 40 (17.9 m/s) 50 (22.4 m/s)
section and previous studies (Bashor et al., 2005; Kijewski-Correa Building 2 26 (11.6 m/s) 35 (15.6 m/s)
et al., 2006) demonstrated, in-situ damping values are still an Building 3 (x-Sway) 41 (18.3 m/s) 45 (20.1 m/s)
Building 3 (y-Sway) 51 (22.8 m/s) 55 (24.6 m/s)
uncertain quantity. As a result, a range of damping values was
assumed when calculating the accelerations of the buildings from
the wind tunnel data. Therefore, upper and lower bounds for wind the same for both fundamental sway responses; however, Building
tunnel predictions are generated to envelope the possible range of 3 did show sensitivity to the mode in question. While the
predicted accelerations, as detailed in Bashor (2011). The results of responses were correctly predicted approximately 25% of the time
analyses for data recorded from Buildings 1, 2, and 3 are presented for all the buildings, in a mean sense, the wind tunnel predictions
in Figs. 9, 10 and 11, respectively, as a function of gradient wind follow the full-scale observed accelerations with a greater degree of
speed. As one would expect, there is scatter due to the wide- conservatism for the concrete building (Building 2). Furthermore,
ranging uncertainties associated with such response predictions, when reviewing Table 4, it becomes evident that the more dyna-
particularly given that on-site wind conditions are not monitored mically sensitive steel buildings showed greater correlation with
at each of the buildings but rather estimated from the NOAA wind tunnel predictions at higher gradient wind speeds than the
sensor. Table 5 lists, for each building, the range of gradient wind concrete building. Since all of the wind events considered in this
speeds for which the correlation between in-situ RMS accelerations study were well below the design gradient wind speed for Chicago
and wind tunnel predictions is greatest. In the case of Buildings (90 mph, 40.2 m/s), the extent to which HFFB measurements should
1 and 2, the wind speeds yielding the best correlation were be expected to accurately capture such low amplitude responses is
96 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97
somewhat questionable, as noted previously in Kijewski-Correa and
Kochly (2007). Ongoing monitoring will attempt to validate the
hypothesis that wind tunnel predictions will show greater degrees
of correlation for design level events.
7. Concluding remarks
As the design of tall buildings is often governed by the habit-
ability limit state — accelerations that adversely affect occupant
comfort — there is considerable need to validate the processes used
to predict RMS accelerations in the design stage. The Chicago Full-
Scale Monitoring Program has been focused on such validations by
observing the in-situ behaviors of three tall buildings whose systems
are common to high rise design for a decade. This paper presented a
cross section of that data to first extract estimates of dynamic
properties and then to use those to predict RMS acceleration levels.
The in-depth analysis of two selected wind events from the database
was first presented in order to evaluate the performance of the
system identification techniques used in this study. While the time
and frequency domain techniques produced consistent natural
frequency estimates, more considerable deviations were observed
in the damping values, though these were not due to bias but rather
random errors. In particular, variance errors in power spectral
estimates rendered the resulting critical damping ratios less reliable
than those from Random Decrement Signatures, particularly for the
more narrow-banded steel buildings. It was found that structural
systems with greater degrees of frame action tend to yield higher
levels of damping than systems that are dominated by cantilever
action, consistent with the findings of previous studies, and that
such systems tend to be more difficult to accurately model in
commercial finite element packages. The dynamic properties of
the three buildings were then determined for several hundred
storms to evaluate trends in the dynamic properties over a range
of amplitudes. It was found that the dynamic properties estimated
from an automated analysis suite for the bulk processing of several
hundreds of events showed significant scatter, partially due to the
inherent errors in the analysis techniques and the lack of identifica-
tion of outliers using human judgment, and must therefore be
considered in a mean sense. The discussion of additional sources of
error is beyond the scope of this paper and is being explored by the
authors. Clear evidence of amplitude dependence and softening with
increased response was displayed for all three buildings, consistent
with initial observations from a narrower subset of CFSMP data.
While damping values showed expected scatter, particularly for the
frequency domain identification approach, agreement in the mean
sense between methods was quite reasonable for all three buildings
and affirmed that the study’s purest steel tube had the lowest level
of damping while the study’s concrete building had the highest
damping, though less damping was observed on the axis where
outriggers are employed. In fact, the systems with panel zones and
comparatively greater reliance on distributed frame action tended to
produce higher damping values.
When the in-situ frequency and damping values were used to
predict the RMS accelerations of the buildings, considering the
potential range of these uncertain parameters, responses were
predicted accurately only 25% of the time for all three buildings.
While there is expected scatter, particularly given that on-site wind
conditions are not monitored at each of the buildings but rather
estimated from the NOAA sensor, the wind speeds over which the
predictions are most accurate were noted for each building and
tend to be higher for the steel buildings. However, since all of the
wind events considered in this study were well below the design
gradient wind speed for Chicago, the extent to which HFFB
measurements should be expected to accurately capture such low
amplitude responses is somewhat questionable. The program’s
continued monitoring of these buildings will enable the observation
of even higher amplitude events to further vet the hypotheses
posted here surrounding the amplitude-dependence of their
dynamic properties and the ranges of wind speed over which HFFB
methods prove most effective in predicting RMS accelerations to
inform habitability design of tall buildings.
Acknowledgments
The authors wish to gratefully acknowledge the support by the
National Science Foundation Grant no. CMMI 06-01143 and the
ongoing collaborations with the Boundary Layer Wind Tunnel and
Skidmore Owings and Merrill LLP. The authors must also sincerely
thank the building owners and management for their continuous
cooperation and enthusiasm, as well as students/engineers,
technicians, legal counsel and support staff of the collaboratory
for their efforts.
References
Bashor, R., 2011. Dynamics of Wind Sensitive Structures, A Dissertation, Department
of Civil Engineering and Geological Sciences, University of Notre Dame,
Notre Dame, IN.
Bashor, R., Kijewski-Correa, T., Kareem, A., 2005. On the wind-induced response of
tall buildings: the effect of uncertainties in dynamic properties and human
comfort thresholds. In: Proceedings of the 10th America’s Conference on Wind
Engineering, Baton Rouge, LA.
Bendat, J.A., Piersol, A.G., 2000. Random Data. John Wiley & Sons, Inc., New York.
Bentz, A., Kijewski-Correa, T., 2008. Predictive models for damping in buildings:
the role of structural system characteristics. In: Proceedings of the 2008
Structures Congress, 18th Analysis and Computation Specialty Conference,
Vancouver, Canada.
Bentz, A., Kijewski-Correa, T., 2009. Wind-induced vibrations of tall buildings: the
role of full-scale observations in better quantifying habitability. In: Proceed-
ings of the IMAC XXVII: A Conference and Exposition on Structural Dynamics,
Rosen Shingle Creek Resort and Golf Club, Orlando, FL.
Bentz, A., Young, B., Kijewski-Correa, T., Abdelrazaq, A., 2010. Finite element
modeling of common lateral systems in tall buildings: insights from full-scale
monitoring. In: Proceedings of the Structures Congress, Orlando, FL.
Chen, X., Kareem, A., 2005. Dynamic wind effects on buildings with 3D coupled
modes: application of high frequency force balance measurements. Journal of
Engineering Mechanics, ASCE 131 (11), 1115–1125.
Davenport, A.G., 1987. The structure of wind and climate. The application of wind
engineering principles to the design of structures. Lausanne, Switzerland
pp. 103–156.
Erwin, S., Kijewski-Correa, T., Yoon, S.-Y., 2007. Full-scale verification of dynamic
properties from short duration records. In: Proceedings of the Structures
Congress 2007, 16–19 May, Long Beach, CA.
ESDU, 2001. Computer Program for Wind Speeds and Turbulence Properties: Flat
of Hilly Sites in Terrain With Roughness Changes, Engineering Sciences Data
Unit-01008.
Kijewski-Correa, T., 2003. Full-Scale Measurements and System Identification:
A Time-Frequency Perspective, A Dissertation. Department of Civil Engineering
and Geological Sciences, University of Notre Dame, Notre Dame, IN.
Kijewski-Correa, T., Kochly, M., 2007. Monitoring the wind-induced response of
tall buildings: GPS performance and the issue of multipath effects. Journal of
Wind Engineering and Industrial Aerodynamics 95 (9–11), 1176–1198.
Kijewski-Correa, T., Pirnia, J.D., 2007. Dynamic behavior of tall buildings under
wind: Insights from full-scale monitoring. The Structural Design of Tall and
Special Buildings 16, 471–486.
Kijewski, T., Kareem, A., 2002. On the reliability of a class of system identification
techniques: insights from bootstrap theory. Structural Safety 24, 261–280.
Kijewski, T., Kareem, A., 2003. Wavelet transforms for system identification:
considerations for civil engineering applications. Computer-Aided Civil and
Infrastructure Engineering 18, 341–357.
Kijewski-Correa, T., Kilpatrick, J., Kareem, A., Kwon, D., Bashor, R., Kochly, M.,
Young, B.S., Abdelrazaq, A., Galsworthy, J., Isyumov, N., Morrish, D., 2006.
Validating the wind-induced response of tall buildings: a synopsis of
the Chicago Full-Scale Monitoring Program. Journal of Structural Engineering
132, 10.
Lagomarsino, S., 1993. Forecast models for damping and vibration periods of
buildings. Journal of Wind Engineering and Industrial Aerodynamics 48,
221–239.
Li, Q.S., Xiao, Y.Q., Wong, C.K., 2005. Full-scale monitoring of typhoon effects on
super tall buildings. Journal of Fluids and Structures 20, 697–717.
Li, Q.S., Xiao, Y.Q., Wong, C.K., Jeary, A.P., 2004. Field measurements of typhoon
effects on a super tall building. Engineering Structures 26, 233–244.
 R. Bashor et al. / J. Wind Eng. Ind. Aerodyn. 104–106 (2012) 88–97
Li, Q.S., Xiao, Y.Q., Wong, C.K., Jeary, A.P., 2003a. Field measurements of wind
effects on the tallest building in Hong Kong. The Structural Design of Tall and
Special Buildings 12, 67–82.
Li, Q.S., Yang, K., Wong, C.K., Jeary, A.P., 2003b. The effect of amplitude-dependent
damping on wind-induced vibrations of a super tall building. Journal of Wind
Engineering and Industrial Aerodynamics 91, 1175–1198.
Littler, J.D., 1991. The Response of a Tall Building to Wind Loading. Ph.D.
Dissertation. University of London, London.
Montpellier, P.R., 1996. The Maximum Likelihood Method of Estimating Dynamic
Properties of Structures. A Master’s Thesis. Department of Civil Engineering,
University of Western Ontario, London, Canada.
NOAA GLERL, 2006. Metadata for NOAA/GLERL Chicago, IL Met Station. National
Oceanic and Atmospheric Administration Great Lakes Environmental Research
Laboratory, 16 December 2011 /http://www.glerl.noaa.gov/metdata/chi/meta
data.htmlS.
97
Pirnia, J.D., 2009. Full-scale Dynamic Characteristics of Tall Buildings and Impacts
on occupant comfort. A Thesis. Department of Civil Engineering and Geological
Sciences, University of Notre Dame, Notre Dame, IN.
Pirnia, J.D., T. Kijewski-Correa, A. Abdelrazaq, J. Chung, and A. Kareem, 2007. Full
scale validation of wind-induced response of tall buildings: investigation of
amplitude-dependent dynamic properties. In: Proceedings of the Structures
Congress 2007, 16–19 May, Long Beach, CA.
Satake, N., Suda, K.-I., Arakawa, T., Sasaki, A., Tamura, Y., 2003. Damping evaluation
using full-scale data of buildings in Japan. Journal of Structural Engineering
129, 470–477.
Xu, Y.L., Chen, S.W., Zhang, R.C., 2003. Modal identification of Di Wang Building
under Typhoon York using the Hilbert-Huang Transform Method. Structural
Design of Tall and Special Buildings 12, 21–47.