Journal of Constructional Steel Research 148 (2018) 589–610

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Permanent seismic drifts in steel moment frames

M.A. Bravo-Haro, A.Y. Elghazouli ⁎
Department of Civil and Environmental Engineering, Imperial College London, United Kingdom

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

Article history: This paper examines residual drift demands in steel moment-resisting frames incorporating the influence of
Received 24 January 2018 degradation and ground motion frequency content. Detailed assessments are carried out using 54 multi-
Received in revised form 16 April 2018 storey framed buildings, with a wide range of structural characteristics, which are designed according to the
Accepted 3 June 2018
provisions of Eurocode 8. In order to identify the influence of cyclic and in-cycle degradation effects, the analysis
Available online 28 June 2018
is carried out with and without degradation modelling. Incremental dynamic analysis is employed in order to
Keywords:
achieve various limits of lateral strength demand, using a suite of 56 ground motion records. It is shown that
Steel structures residual drifts are markedly higher in degrading models in comparison with non-degrading models, with the
Moment frames differences being more pronounced in relatively short period ranges, when higher rates of cyclic deterioration
Cyclic deterioration are employed, and for comparatively high lateral strength demand levels. The residual drift demand is also
Seismic demands shown to increase with the increase in number of stories, and is often concentrated in the lower levels when
Residual drifts degrading models are used. Overall, significant residual drift demands are observed in the structural systems
considered, with a high likelihood of exceeding a 0.5% residual drift limit in most cases. Based on the results,
two simplified prediction relationships are proposed to estimate the permanent drifts of multi-storey steel
moment framed systems. The first is concerned with the design stage based on the results of elastic analysis,
whilst the second is associated with post-earthquake structural assessment based on actual measurements of
residual drifts.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction
The prediction of seismic residual drifts, representing the permanent
deformations of a structural system caused by inelastic deformations
that remain after the ground shaking has ended, is important for the
design of new structures as well as for the assessment of post-
earthquake conditions. Studies available in the literature that highlight
the significance of residual drifts as a structural index can be classified
into two main groups. Firstly, studies that use residual drifts as an engi-
neering demand parameter to perform seismic structural assessment of
existing buildings (e.g., [1–5]) which, as expected, find a strong correla-
tion between the damaged state and the residual drifts. Secondly, stud-
ies that use residual drifts in the design of new buildings according to
current seismic codes, and which indicate apparent inconsistencies
between the expected performance of the building and that observed
or computed at the end of the seismic excitation (e.g., [5–10]). However,
in contrast with these findings, most recent seismic standards for the
design of new buildings only focus on the utilisation of maximum
peak drifts as a structural index, whilst completely disregarding residual
drifts. This is the case in Eurocode 8 (EC8) [11] as well as in ASCE 7–10
[12]. In the case of seismic assessment and rehabilitation of existing
⁎ Corresponding author.
E-mail address: a.elghazouli@imperial.ac.uk (A.Y. Elghazouli).
structures, most current codes do not explicitly consider residual drifts
in the analysis. Instead, they account for the post-earthquake damaged
condition through the levels of inelastic rotation of yielding elements
(i.e., beams and columns), as is the case in Part 3 of Eurocode 8 [13] as
well as in ASCE/SEI 41-13 [14]. In contrast, recent guidance documents
such as FEMA P-58 [15] highlights the importance of evaluating residual
drifts for the seismic assessment of buildings. In this case, the use of
detailed numerical models that characterise structural performance
more precisely at large nonlinear deformations, including representa-
tion of in-cycle and cyclic deterioration of strength and stiffness of struc-
tural members, is essential.
Research undertaken on the main parameters that influence the
behaviour of residual drifts in different structural configurations can
be categorised by the study of single-degree-of-freedom (SDOF) and
multi-degree-of-freedom (MDOF) systems. Likewise, two subcategories
are typically utilised for the determination of absolute values of residual
drifts, namely: the direct approach where residual drifts are directly
obtained from dynamic analysis (i.e., ultimate recorded deformation),
and the indirect approach where the residual drift is determined as a
percentage of the maximum transient peak drift. In one of the earliest
attempts to understand the behaviour of residual drifts for a wide
range of fundamental periods, Kawashima et al. [16] developed a resid-
ual displacement response spectrum, where the hysteretic properties of
SDOF oscillators were characterised by a simple bilinear hysteresis

https://doi.org/10.1016/j.jcsr.2018.06.006
0143-974X/© 2018 Elsevier Ltd. All rights reserved.
590 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
model (i.e., elasto-plastic with hardening). In this indirect approach, the
residual drift ratio was found to be strongly dependent on the post-yield
response, whilst the influence of fundamental period, soil condition, and
ductility levels, were reported to be relatively insignificant. Neverthe-
less, Ruiz-García et al. [17] indicated a strong influence from the ductil-
ity demand level on the residual drifts, consistently using the indirect
approach of assessment and hysteretic elasto-plastic systems.
Studies on MDOF systems typically focused on examining the ampli-
tude and distribution of residual drifts over the height in multi-storey
buildings, and emphasised its importance for existing structures
[3, 10, 18, 19] as well as in the design of new structural systems [7, 8,
20, 21]. There is general agreement on the key characteristics that define
the amplitude and distribution of residual drifts, namely: the hysteretic
behaviour of the main dissipative structural components, ground
motion intensity (i.e., ductility demand), typology of plastic mechanism,
and the structural over-strength (e.g., [3, 4, 7, 16, 22]). However, in
comparison, limited studies are available on the correlation between
measurement of actual residual drift and levels of observed post-
earthquake damage, which is also related to the lack of specific residual
drift or reparability limits in most current seismic codes. After the 1995
Hyogoken-Nabu earthquake in Japan, Iwata et al. [23] assessed 12 dam-
aged steel buildings considering technical and economic aspects in
order to define clear criteria for potential reparability. The proposed
limits included a maximum residual inter-storey drift of 1.4% and a
maximum roof drift of 0.9%. Furthermore, research conducted in Japan
by McCormick et al. [24] considered the comfort living conditions of res-
idents in a building with permanent damage. A residual inter-storey
drift of 0.5% was indicated, which was shown to be a suitable limit as
a human comfort threshold as well as being associated with the require-
ments for extensive repair. Since there is a relationship between resid-
ual and peak drifts, and considering the significant available research
on the latter (e.g., [25–29]), maximum inelastic displacement ratios
are addressed herein as well.
The first objective of this paper is to provide a detailed insight into
the salient parameters that influence the magnitude and distribution
of seismic residual drift demands in steel moment frames designed
according to the provisions of Eurocode 8, using extensive dynamic
analysis on SDOF and MDOF systems. Secondly, the investigation
aims to provide detailed quantification of the influence of cyclic and
in-cycle degradation on residual drift demands, with direct compari-
son between the results of degrading and non-degrading systems.
Thirdly, based on the results, predictive models are proposed in
order to determine the maximum residual drift demands directly, as
well as for estimating maximum drift demands indirectly when resid-
ual drifts are known. Finally, the results are analysed collectively in
the light of provisions of current seismic codes and guidelines, for
which permanent drift demand limitations are either non-existent or
only indirectly considered by proxies such as through maximum
deformation limits.
Fig. 1. (a) Mechanical model of the ESDOF, cyclic normalised pushover and pushover/backbone of both hysteretic behaviour models assigned to the base of the mechanical model for non-
degrading (b) and degrading (c) systems.
2. Structural systems and ground motions
In this section, a detailed description of the SDOF and MDOF struc-
tural systems used in this study is presented, along with the modelling
details. Subsequently, the selection criteria for the earthquake records
and the final ensemble of ground motions are provided.
2.1. Mechanical model of SDOF systems
Preliminary assessment of residual drifts is carried out using equiva-
lent single-degree-of-freedom systems (ESDOF) instead of simple oscil-
lators. The term ‘equivalent’ stands for systems that represent globally
the behaviour of multi-storey systems and, to serve this purpose, they
have been calibrated as discussed below. The mechanical model shown
in Fig. 1(a) corresponds to an inverted pendulum with mass (m) at its
tip, and the rod is a massless element pinned at the base. The properties
of the moment-rotation hysteretic model are assigned to a rotational
spring at the base. Viscous damping is considered using a rotational
dashpot damper acting in parallel with the main rotational spring. Unlike
conventional simplified SDOF idealisations (e.g., oscillating mass
attached to a damped spring), this represents a more versatile model
that enables the inclusion of equivalent gravity loads and therefore is
capable of capturing lateral instability effects (i.e., P − Δ). The model
also readily allows for a change of the structural properties of the system,
either by tuning the mass or the initial elastic stiffness of the spring at the
base to represent different fundamental periods.
To provide a basis for comparison, the parametric study was under-
taken for two constitutive hysteretic models: a non-degrading model,
and a stiffness- and strength-degrading model. The non-degrading
model includes a post-yield isotropic strain-hardening branch without
in-cycle strength and stiffness degradation. This model is adopted over
simpler non-degrading bilinear hysteretic models since it is compara-
tively more realistic as it incorporates a curved yield transition zone.
The outcome of a cyclic pushover analysis of the non-degrading
model is shown in Fig. 1(b). Likewise, the Modified Ibarra-Medina-
Krawinkler (IMK) bilinear model [30] is adopted to represent the
behaviour of the degrading systems. This model has been calibrated
by Lignos and Krawinkler [31] and shown that it is capable of capturing
cyclic and in-cycle deterioration based on the dissipation of cumulative
hysteretic energy, for reproducing the moment-rotation behaviour of
steel components with local buckling phenomena being the main
source of degradation.
The N2 method adopted in EC8 [11] was partially followed to obtain
the equivalent structural properties of the SDOF. Given that the EC8 ver-
sion of the method is limited to perfectly plastic systems, an updated
multi-linear fit of the MDOF static pushover capacity curve to obtain
the equivalent SDOF curve, as proposed by De Luca et al. [32] was
used herein. In the case of the IMK degrading model, to calibrate the
equivalent rate of cycling degradation (Λ), a cyclic pushover analysis
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
of the multi-degree-of-freedom (MDOF) was carried out and used as a
target (i.e. force and displacement) response for the subsequent cyclic
pushover undertaken on the SDOF system. The mean squared error of
the cyclic force-deformation response was minimised in order to find
the near-optimal values of the equivalent rate of cyclic deterioration.
These values are shown in Table 3 along with the rest of the structural
characteristics of the multi-storey frames. The results of a cyclic push-
over for a reference equivalent degrading hysteretic model are also
shown in Fig. 1 (c). Finally, the pushover/backbone curves of both struc-
tural models can be seen in Fig. 1 (b) and (c) respectively along with αh
corresponding to the post yield-to-elastic stiffness ratio.
2.2. Design of multi-storey structures
To assess the influence of degradation modelling on the response
and distribution of residual drifts during nonlinear response history
analysis (NRHA), a large set of steel moment frames that comply with
the design provisions of Eurocode 3 (EC3) [33] and EC8 [11] was consid-
ered. Fig. 2 shows a plan and elevation of the typical structural system of
the buildings, which encompasses three lateral resisting moment
frames in the in-plane direction, whilst braced frames provide lateral
load resistance in the orthogonal direction. This lateral resisting config-
uration is common in European design practice, and allows the study of
the structure as plane frames (2-D) acting independently. Herein, the
interior moment resisting frame was considered and designed accord-
ing to the lateral force method based on an idealised first mode-
response pattern. Each frame consists of 3 bays of 6.0 m span, a constant
storey height of 3.5 m except for the first storey which has a height of
4.5 m. The frames are grouped in four different overall heights defined
Fig. 2. Structural configurations: (a) typical plan view, (b) – (e) schematic elevation of the 4 frame sets analysed (standard notation for structural beams and columns is illustrated in
(c) and (d), respectively).
591
by the number of stories, namely 3, 5, 7 and 9 stories. These four groups
are henceforth referred to as Sets A, B, C and D respectively. This ensem-
ble of frames offers a wide range of fundamental periods and structural
characteristics, which are used through this study to enhance the
understanding of the residual drifts behaviour and to underpin the pre-
dictive models. The selected frames are the same as those considered in
a recent investigation [25] where the focus was on the study of peak
inelastic roof and inter-storey displacement demands.
The design gravity and seismic loads were defined according to EC3
and EC8, considering permanent dead loads (DL) as 5.8 kN/m2 as for the
typical floors and 5.0 kN/m2 for the roof. The imposed live loads (LL)
consist of 3.0 kN/m2 in all floors. The standard load combination was
used to calculate the seismic masses and action effects (1.0DL +
0.3LL). Regular European steel profiles were used for the beams (IPE)
and the columns (HE). For simplicity and practicality, the same sections
were used for internal and external columns at a given storey, and the
same approach was also used for the beam profiles at each storey
level. To enable a wide range of design scenarios, the seismic design
was carried out assuming different seismic hazard conditions through
various combinations of peak ground acceleration (PGA), soil conditions
and serviceability drift limits. A common assumption for all the frames
was that a behaviour factor (q) equal to 4.0 was considered in the design
process. A lower value than the upper bound allowed by the code was
considered in order to facilitate the control of significant levels of
over-strength typically obtained in frames designed to EC8 [34]. Those
over-strength values can often reach magnitudes higher than the
assumed behaviour factor. It should be noted that constant relative
strength levels of demand were used for the assessment of all the struc-
tures in this study, anchored to the actual yield capacity of the buildings
592 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610

Table 1
Design details and structural characteristics of the study frames.

Frame Structural steel profiles Structural properties

ID Beams (first row) and Columns (second row) from first storey upwards T1, s T2, s H, m α ρ β γ αo Λ αh

A01 IPE500, IPE500, IPE450 0.53 0.15 11.50 1.47 0.20 0.73 0.89 2.41 14.10 0.03
HEB500, HEB500, HEB450
A02 IPE550, IPE550, IPE500 0.40 0.11 11.50 1.42 0.12 0.78 0.88 3.46 11.00 0.03
HEM600, HEM600, HEM600
A03 IPE400, IPE400, IPE360 0.74 0.21 11.50 1.90 0.18 0.74 0.89 2.71 19.88 0.04
HEB400, HEB400, HEB400
A04 IPE550, IPE550, IPE500 0.41 0.11 11.50 1.45 0.15 0.75 0.89 2.91 19.60 0.05
HEM550, HEM550, HEM550
A05 IPE330, IPE330, IPE300 1.00 0.28 11.50 2.42 0.14 0.74 0.88 3.28 15.67 0.03
HEB340, HEB340, HEB340
A06 IPE360, IPE360, IPE300 0.84 0.25 11.50 2.05 0.12 0.82 0.86 3.67 14.00 0.03
HEB400, HEB400, HEB300
A07 IPE450, IPE450, IPE400 0.62 0.18 11.50 1.57 0.18 0.73 0.89 2.59 9.64 0.04
HEB450, HEB450, HEB400
A08 IPE360, IPE360, IPE360 0.75 0.20 11.50 2.04 0.09 0.79 0.87 3.97 14.44 0.03
HEB450, HEB450, HEB400
B01 IPE750x161, IPE750x161, IPE750x161,IPE750x161, IPE450 0.40 0.13 18.50 1.43 0.14 0.74 0.85 2.96 8.50 0.04
HEM900, HEM900, HEM900, HEM900,HEM600
B02 IPE500, IPE500, IPE450, IPE450, IPE300 0.79 0.26 18.50 1.74 0.07 0.93 0.80 5.34 11.00 0.06
HEM600, HEM600, HEM550, HEM550, HEM400
B03 IPE450, IPE450, IPE450, IPE450, IPE300 0.97 0.31 18.50 1.69 0.14 0.73 0.84 3.31 14.40 0.04
HEM600, HEM600, HEM550, HEM550, HEB400
B04 IPE750x137, IPE750x137, IPE600, IPE600, IPE400 0.52 0.20 18.50 1.39 0.14 1.03 0.82 2.70 8.17 0.03
HEM650, HEM650, HEM650, HEM650, HEB400
B05 IPE450, IPE450, IPE400, IPE400, IPE300 0.93 0.29 18.50 1.94 0.07 0.93 0.79 6.16 11.40 0.08
HEM550, HEM550, HEB550, HEB550, HEB550
B06 IPE450, IPE450, IPE400, IPE400, IPE300 0.98 0.31 18.50 1.69 0.07 0.91 0.81 4.31 13.45 0.04
HEB550, HEB550, HEB550, HEB550, HEB400
B07 IPE400, IPE400, IPE400, IPE400, IPE300 1.10 0.34 18.50 1.99 0.10 0.79 0.83 4.21 16.95 0.04
HEB500, HEB500, HEB500, HEB500, HEB400
B08 IPE500, IPE500, IPE500, IPE500, IPE300 0.79 0.27 18.50 1.74 0.15 0.80 0.82 4.31 13.65 0.04
HEM550, HEM550, HEB550, HEB550, HEB400
B09 IPE750X137, IPE750X137, IPE600, IPE600, IPE400 0.52 0.19 18.50 1.40 0.17 1.00 0.81 2.95 8.25 0.04
HEM700, HEM700, HEM600, HEM600, HEB400
B10 IPE360, IPE360, IPE330, IPE330 IPE300 1.44 0.45 18.50 2.21 0.09 0.86 0.82 3.83 17.93 0.06
HEB400, HEB400, HEB400, HEB400, HEB300
B11 IPE400, IPE400, IPE400, IPE400, IPE300 1.13 0.37 18.50 2.06 0.18 0.79 0.82 4.21 15.90 0.04
HEB500, HEB500, HEB400, HEB400, HEB300
B12 IPE500, IPE500, IPE450, IPE450, IPE300 0.87 0.31 18.50 1.60 0.11 0.96 0.82 3.38 11.50 0.03
HEB550, HEB550, HEB550, HEB550, HEB300
B13 IPE400, IPE400, IPE360, IPE360, IPE300 1.16 0.38 18.50 2.01 0.09 0.88 0.80 4.66 14.63 0.05
HEB500, HEB500, HEB450, HEB450, HEB300
B14 IPE400, IPE400, IPE400, IPE400, IPE300 1.16 0.37 18.50 1.96 0.13 0.74 0.84 3.47 15.75 0.04
HEB450, HEB450, HEB450, HEB450, HEB300
B15 IPE400, IPE400, IPE400, IPE400, IPE300 1.11 0.35 18.50 1.98 0.12 0.80 0.83 4.21 14.80 0.04
HEB500, HEB500, HEB500, HEB400, HEB400
C01 IPE550, IPE550, IPE550, IPE550, IPE550, IPE550, IPE450 0.89 0.28 25.50 1.73 0.12 0.77 0.82 4.09 14.50 0.07
HEM700, HEM700, HEM700, HEM600, HEM600, HEM600, HEM550
C02 IPE500, IPE500, IPE500, IPE500, IPE500, IPE500, IPE300 1.05 0.34 25.50 1.82 0.09 0.76 0.82 4.55 15.10 0.05
HEM600, HEM600, HEM600, HEM600, HEM600, HEM600, HEB450
C03 IPE500, IPE500, IPE500, IPE500, IPE500, IPE500, IPE450 1.05 0.33 25.50 1.88 0.11 0.76 0.82 4.22 14.00 0.06
HEM600, HEM600, HEM600, HEM550, HEM550, HEM550, HEB450
C04 IPE550, IPE550, IPE550, IPE500, IPE500, IPE500, IPE450 0.97 0.32 25.50 1.56 0.11 0.89 0.81 3.77 13.44 0.06
HEM600, HEM600, HEM600, HEM550, HEM550, HEM550, HEB450
C05 IPE500, IPE500, IPE500, IPE450, IPE450, IPE450,IPE300 1.13 0.38 25.50 1.73 0.07 0.91 0.80 4.63 15.93 0.05
HEM550, HEM550, HEM550, HEM550, HEM550, HEM550, HEB450
C06 IPE450, IPE450, IPE450, IPE450, IPE400, IPE400,IPE300 1.35 0.46 25.50 1.72 0.14 0.85 0.81 3.99 16.00 0.05
HEB550, HEB550, HEB550, HEB500, HEB500, HEB500, HEB300
C07 IPE550, IPE550, IPE550, IPE550, IPE550, IPE550, IPE300 0.96 0.35 25.50 1.67 0.21 0.79 0.81 3.61 14.16 0.14
HEM600, HEM600, HEM600, HEB550, HEB550, HEB550, HEB300
C08 IPE400, IPE400, IPE400, IPE400, IPE400, IPE400, IPE300 1.66 0.54 25.50 1.92 0.18 0.75 0.83 3.36 15.56 0.06
HEB450, HEB450, HEB450, HEB400, HEB400, HEB400, HEB300
C09 IPE750x137, IPE750x137, IPE750x137, IPE600, IPE600, IPE600, IPE400 0.67 0.24 25.50 1.43 0.17 0.97 0.80 3.33 7.93 0.05
HEM800, HEM800, HEM800, HEM700, HEM700, HEM700, HEB450
C10 IPE450, IPE450, IPE450, IPE400, IPE400, IPE400, IPE300 1.50 0.51 25.50 1.64 0.13 0.89 0.82 2.94 13.07 0.05
HEB450, HEB450, HEB450, HEB450, HEB450, HEB450, HEB300
C11 IPE550, IPE550, IPE550, IPE550, IPE500, IPE500, IPE300 1.00 0.37 25.50 1.60 0.15 0.90 0.81 3.51 – 0.06
HEM550, HEM550, HEM550, HEB550, HEB550, HEB550, HEB300
C12 IPE450, IPE450, IPE450, IPE400, IPE400, IPE400, IPE300 1.52 0.52 25.50 1.60 0.18 0.90 0.81 2.94 12.53 0.05
HEB450, HEB450, HEB450, HEB400, HEB400, HEB400, HEB300
C13 IPE500, IPE500, IPE500, IPE450, IPE450, IPE450, IPE300 1.13 0.38 25.50 1.78 0.19 0.75 0.80 4.63 13.70 0.05
HEM550, HEM550, HEM550, HEM550, HEM550, HEM550, HEB450
C14 IPE450, IPE450, IPE450, IPE450, IPE400, IPE360, IPE300 1.37 0.50 25.50 1.70 0.14 0.91 0.79 4.10 13.60 0.05
HEB550, HEB550, HEB550, HEB500, HEB500, HEB500, HEB300
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610 593

Table 1 (continued)

Frame Structural steel profiles Structural properties

ID Beams (first row) and Columns (second row) from first storey upwards T1, s T2, s H, m α ρ β γ αo Λ αh

C15 IPE550, IPE550, IPE550, IPE500, IPE500, IPE360, IPE360 1.02 0.39 25.50 1.56 0.21 0.95 0.77 4.25 11.35 0.06
HEM600, HEM600, HEM600, HEB550, HEB550, HEB550, HEB300
C16 IPE550, IPE550, IPE550, IPE500, IPE500, IPE400, IPE300 1.04 0.39 25.50 1.57 0.15 0.94 0.79 3.87 12.35 0.05
HEM550, HEM550, HEM550, HEB550, HEB550, HEB450, HEB400
D01 IPE600, IPE600, IPE600, IPE600, IPE600, IPE600, IPE600, IPE550, IPE550 1.04 0.34 32.50 1.36 0.14 0.74 0.82 3.21 18.60 0.09
HEM700, HEM650, HEM650, HEM650, HEM650, HEM650, HEM600, HEM600, HEM600
D02 IPE550, IPE550, IPE550, IPE550, IPE550, IPE500, IPE500, IPE450, IPE450 1.25 0.42 32.50 1.34 0.15 0.85 0.81 3.72 14.92 0.11
HEM600, HEM600, HEM600, HEM600, HEM550, HEM550, HEM550, HEM500, HEM500
D03 IPE550, IPE550, IPE550, IPE500, IPE500, IPE500, IPE450, IPE450, IPE450 1.31 0.44 32.50 1.33 0.11 0.90 0.79 4.03 14.16 0.11
HEM600, HEM600, HEM550, HEM550, HEM550, HEM550, HEM550, HEM500, HEM450
D04 IPE550, IPE550, IPE550, IPE500, IPE500, IPE450, IPE450, IPE400, IPE360 1.39 0.49 32.50 1.27 0.13 1.00 0.78 3.96 14.36 0.09
HEM550, HEM550, HEM500, HEM500, HEM500, HEM450, HEM450, HEM400, HEM400
D05 IPE500, IPE500, IPE500, IPE500, IPE500, IPE450, IPE450, IPE400, IPE360 1.51 0.52 32.50 1.30 0.16 0.86 0.81 3.92 15.28 0.09
HEM500, HEM500, HEM500, HEM450, HEM450, HEM450, HEM400, HEM400, HEM400
D06 IPE450, IPE450, IPE450, IPE450, IPE450, IPE400, IPE400, IPE400, IPE360 1.77 0.59 32.50 1.42 0.14 0.85 0.81 4.37 13.84 0.13
HEM450, HEM450, HEM450, HEM400, HEM400, HEM400, HEM400, HEM400, HEM360
D07 IPE450, IPE450, IPE450, IPE450, IPE450, IPE400, IPE400, IPE400, IPE360 1.81 0.61 32.50 1.27 0.17 0.83 0.80 4.37 – 0.13
HEM450, HEM450, HEM360, HEM360, HEM360, HEM360, HEM360, HEM360, HEM360
D08 IPE750x137, IPE750x137, IPE750x137, IPE600, IPE600, IPE600, IPE500, IPE500, IPE360 0.92 0.34 32.50 1.24 0.12 0.97 0.76 3.71 13.76 0.10
HEM800, HEM800, HEM800, HEM700, HEM700, HEM700, HEM600, HEM600, HEM500
D09 IPE500, IPE500, IPE500, IPE500, IPE500, IPE500, IPE500, IPE500, IPE400 1.69 0.55 32.50 1.15 0.37 0.69 0.84 1.92 – 0.04
HEB450, HEB450, HEB450, HEB400, HEB400, HEB400, HEB400, HEB400, HEB300
D10 IPE550, IPE550, IPE550, IPE500, IPE500, IPE500, IPE500, IPE500, IPE450 1.28 0.42 32.50 1.35 0.09 0.89 0.80 3.84 18.80 0.21
HEM600, HEM600, HEM600, HEM600, HEM600, HEM550, HEM550, HEM500, HEM450
D11 IPE750x137, IPE600, IPE600, IPE600, IPE600, IPE600, IPE500, IPE500, IPE450 0.99 0.35 32.50 1.35 0.12 0.91 0.77 4.00 14.90 0.10
HEM800, HEM800, HEM800, HEM700, HEM700, HEM600, HEM550, HEM500, HEM450
D12 IPE600, IPE600, IPE600, IPE600, IPE600, IPE600, IPE500, IPE500, IPE400 1.05 0.37 32.50 1.33 0.12 0.78 0.81 3.62 16.60 0.08
HEM700, HEM700, HEM700, HEM700, HEM700, HEM700, HEM600, HEM600, HEM400
D13 IPE450, IPE450, IPE450, IPE450, IPE450, IPE450, IPE400, IPE400, IPE360 1.81 0.64 32.50 1.32 0.26 0.84 0.81 3.20 – 0.18
HEB500, HEB500, HEB500, HEB400, HEB400, HEB360, HEB360, HEB360, HEB360
D14 IPE450, IPE450, IPE450, IPE450, IPE450, IPE400, IPE400, IPE400, IPE360 1.91 0.66 32.50 1.27 0.34 0.88 0.79 3.30 13.64 0.10
HEB500, HEB500, HEB500, HEB360, HEB360, HEB360, HEB360, HEB360, HEB360
D15 IPE550, IPE550, IPE550, IPE500, IPE500, IPE500, IPE450, IPE400, IPE360 1.36 0.49 32.50 1.27 0.13 0.93 0.79 3.85 14.28 0.11
HEM550, HEM550, HEM550, HEM500, HEM500, HEM450, HEM450, HEM400, HEM400

as determined by pushover analysis. Therefore, the influence of the
design behaviour factor selection is attenuated by the structural capac-
ity assessment methodology that defines the levels of lateral demand
considered. Moreover, both the inter-storey serviceability limits and
the P-delta effect were limited according to the code requirements.
The latter effect was initially considered targeting a sensitivity coeffi-
cient (θ) lower than 0.1 in order to avoid second order effects. However,
for frames in which θ reached values between 0.1 and 0.2, seismic
actions were amplified by the factor 1/(1 − θ) as stipulated in EC8. As
noted before, no limitations or requirements are given in codified guid-
ance regarding the residual drifts.
Detailed description of the final design of the frames is given in
Table 1 together with the final steel member profiles. As shown in the
table, the fundamental period T1 obtained from eigenvalue analysis
ranges from 0.40 s to 1.91 s. The total height of the frames is 11.5, 18.5,
25.5 and 32.5 m, for the 3, 5, 7 and 9 storey frames, respectively. The
plasticity resistance ratio α (also referred to as αu/α1 in EC8) represents
the ratio of base shear (Vy) when the plastic mechanism has developed
in the system, to the base shear corresponding to first yield (V1), both as
obtained from pushover analysis. The values of plasticity resistance
ratios range from 1.15 to 2.42. The beam-to-column ratio (ρ) was calcu-
lated for the storey closest to mid-height of the frame and is defined by
the ratio of summation of second moment of area (I) to the length of the
member (l) (i. e. ρ = ∑(I/l)b/ ∑(I/l)c), where the subscripts b and c
represent beams and columns, respectively. The distribution of ρ ranges
from 0.07 to 0.37. The relative storey stiffness parameter (β) is estimated
as the ratio of the maximum inter-storey drift for the upper half of the
system to the lower half of the frame. The inter-storey drift profile
obtained from Eigenvalue analysis was used for its computation and
the distribution of β sample ranges from 0.69 to 1.03. This parameter
was initially introduced by Kumar et al. [35] (therein referred to as β3)
and proved to be a simple and useful measure for capturing the flexibil-
ity of upper stories that might experience earlier yielding and hence rel-
atively higher drift demands due to higher mode effects. The first mode
participation factor is denoted as γ and computed using eigenvalue anal-
ysis. The greater the value of γ the more first-mode dominated the sys-
tem is, and therefore the lower the contribution of higher modes effects.
The distribution of the first mode participation factor ranges from 0.76 to
0.90 where, as expected, shorter frames are generally more first-mode
controlled. The equivalent post-yield stiffness ratio, denoted αh, is esti-
mated from the pushover capacity curve of the MDOF system, and corre-
sponds to the piecewise linear fit used to define the equivalent SDOF
backbone curve (e.g., Fig. 1 c). Finally, αo, a parameter introduced by
Karavasilis et al. [36], is utilised to account for the formation of plastic
mechanisms. It corresponds to the ratio between the average of the plas-
tic moments of resistance of the columns of the first storey, and the
average of the plastic moments of resistance of all beams in the frame.
Higher values of αo represent a delayed plastic mechanism formation.
Finally, Λ corresponds to the equivalent cyclic deterioration rate for an
equivalent SDOF system, using the hysteretic model proposed by Ibarra
et al. [30]. The above parameters are discussed in more detail in subse-
quent sections of this paper.
2.3. Modelling of multi-storey frames
The two-dimensional frames considered in this study were
modelled using the open-source software OpenSEES [37]. Two
modelling approaches were followed throughout the study, namely:
(i) fibre-based distributed plasticity without degradation and, (ii)
lumped plasticity considering degradation effects for steel sections.
594 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
Although the distributed plasticity model does not account for degrada-
tion, it was used as a benchmark since this approach has been widely
employed and validated in seismic analysis studies and, additionally, it
is relatively unaffected by viscous damping modelling [38]. Initially, a
lumped plasticity model without deterioration was considered as well,
which provided very similar results to those from the distributed plas-
ticity model and is therefore not presented herein.
The first approach based on distributed plasticity, henceforth referred
to as “non-degrading” models, consists of force-based non-linear beam-
column elements for representing beams and columns, as shown in
Fig. 3(a). The constituent fibres are defined by an elasto-plastic material
with 0.5% strain hardening. The second approach, henceforth referred to
as “degrading” models, is made up of rotational springs used to idealise
the plastic hinges in beams and columns, include a hysteretic behaviour
described by the Ibarra-Medina-Krawinkler model [30] with parameters
obtained from the empirical relationships for other-than-RBS
(i.e., reduced beam section) introduced by Lignos and Krawinkler [31],
and subsequently adopted in the PEER/ATC-72 report [39]. These set of
parameters are representative of the expected behaviour of steel build-
ings designed to EC8. Column web panel zones were considered in
both structural models and shear-distortion behaviour, using a zero-
length element, was characterised according to the tri-linear hysteretic
model proposed by Gupta and Krawinkler [40]. Further details of the
above modelling approaches can be found elsewhere [25, 41].
Although the main aim of the study is to focus on the influence of deg-
radation modelling and ground motion frequency content, particular
attention was given to viscous damping in order to define a consistent
representation of inherent damping for both modelling approaches
used. The sensitivity of MDOF structural systems to different damping
assumptions has been examined in previous studies and various issues
have been raised, particularly when the Rayleigh damping approach is
used [38, 42, 43]. Distributed plasticity was found to be relatively insen-
sitive to viscous damping approaches, whilst lumped plasticity models
can develop spurious dynamic response depending on the viscous
damping model adopted, especially when the initial stiffness propor-
tional Rayleigh damping is assigned to zero-length elements [38]. In con-
trast, other researchers [44, 45] were able to overcome this shortcoming
by avoiding stiffness-proportional damping in plastic hinges alongside
modification to the local stiffness matrix of the elastic beam-column ele-
ments. In the current study, the latter approach was used; therefore, the
zero-length elements in the concentrated plasticity models were ex-
cluded from the assignment of damping properties, and the stiffness ma-
trices of the elastic elements between plastic hinges were also modified,
as recommended by Zareian and Medina [44]. This enabled the utilisation
of initial stiffness proportional damping for both modelling approaches.
2.4. Earthquake records
To carry out incremental dynamic analysis (IDA) on the study
frames, a suite of 56 far-field ground motions records was used,
comprising both horizontal components from 28 strong-motion record-
ings from the Pacific Earthquake Engineering Research (PEER) database
NGA-West2 [46]. The set comprise large moment magnitude
Fig. 3. Elevation of frame models (illustrated only for a 3-storey building): (a) distributed plasticity model, and (b) lumped plasticity model (all dimensions in meters).
earthquakes (between 6.50 and 7.62) and moderate distances of the re-
cording station site from the source (8.7–97 km), all recorded on firm
soils according to the NEHRP soil classification. Table 2 describes each
individual record alongside the main characteristics: moment magni-
tude (Mw), fault mechanism, mean period (Tm) and distance to the
source (Rs). Note that the Tm column contains the mean period for
both horizontal directional components.
Several frequency content parameters have been proposed in previ-
ous studies [47, 48]. For instance, a common and simple frequency con-
tent quantity is the predominant period (Tp), that defines the period of
the record based on the maximum peak of the acceleration response
spectrum; this was used by Miranda and Ruiz-García [47] where the
influence of stiffness degradation was assessed under soft soil condi-
tions. Another possible parameter is the smoothed spectral predomi-
nant period (To), which offers an average of the periods in the
acceleration response spectrum normalised by a weighting factor corre-
sponding to the natural logarithm of the spectral acceleration at the
same period (e.g. [48]). However, the mean period (Tm) has been
selected herein because of its simplicity and reliable estimation in com-
parison with other alternative parameters considered, as was shown by
Rathje et al. [49]. The mean period is determined as follows:
 
1
∑i C2i
fi
Tm ¼ ½for 0:25 Hz ≤ f i ≤20 Hz with Δ f ≤0:05 Hz ð1Þ
∑i C2i
where Ci is the Fourier amplitude at frequency fi obtained by applying a
discrete Fast Fourier Transform (FFT) for frequencies between 0.25 and
20 Hz, whereas Δf corresponds to the frequency step used in the FFT
computation. As can be observed, the mean period uses the full range
of frequencies of interest, utilising weighted average of the periods in
the frequency domain over a user-defined frequency range, for which
Fourier amplitudes are used herein as proportional weighting. This
method, which does not require the computation of acceleration
response spectra, has been consistently used in previous studies to char-
acterise the ground motion frequency content [41, 50, 51].
3. Residual drift demand in SDOF systems
This section focuses on determining the residual deformation
demands in equivalent single-degree-of-freedom (ESDOF) systems in
order to provide a fundamental insight that can then be extended into
more realistic multi-storey frames.
As discussed before, there are typically two methodologies that are
used for the estimation of residual drifts. Firstly, a so-called ‘direct’
approach, where the estimates are determined directly from the
residual deformation recorded at the end of the dynamic excitation.
Secondly, the so-called ‘indirect’ approaches where the residual drift
estimates are obtained as a proportion of the transient peak deforma-
tion experienced by the system through the dynamic excitation. Both
strategies are presented in this section as well as in the subsequent
statistical study on multi-storey frames (in Section 4) due to their com-
plementary purposes. In the former approach, the statistical analysis
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610 595

Table 2
Catalogue of the earthquakes used in the study and their seismological characteristics (fault mechanism: RV = reverse, SS = strike-slip, and RO = reverse oblique).

Earthquake name Date M Station Mech. Tm, s Rs, km

Chi-Chi, Taiwan 1999-09-20 7.62 CHY101 RO 0.96, 1.05 15.44

Chi-Chi, Taiwan 1999-09-20 7.62 TCU045 RO 0.53, 0.47 26.81
St Elias, Alaska 1979-02-28 7.54 Yakutat RV 2.02, 1.91 97.05
St Elias, Alaska 1979-02-28 7.54 Icy Bay RV 1.44, 1.10 31.56
Kocaeli, Turkey 1999-08-17 7.51 Duzce SS 0.89, 1.05 15.43
Kocaeli, Turkey 1999-08-17 7.51 Arcelik SS 0.63, 0.31 13.52
Manjil, Iran 1990-06-20 7.37 Abbar SS 0.32, 0.32 12.97
Kern County 1952-07-21 7.36 Taft Lincoln School RV 0.54, 0.55 38.89
Landers 1992-06-28 7.28 Yermo Fire Station SS 0.66, 0.91 23.80
Landers 1992-06-28 7.28 Coolwater SS 0.56, 0.42 19.97
Duzce, Turkey 1999-11-12 7.14 Turkey Bolu SS 0.78, 0.55 12.41
Hector Mine 1999-10-16 7.13 Amboy SS 0.64, 0.79 43.05
Hector Mine 1999-10-16 7.13 Mine Hector SS 0.63, 0.61 12.02
Cape Mendocino 1992-04-25 7.01 Eureka - Myrtle & West RV 1.09, 0.75 41.97
Cape Mendocino 1992-04-25 7.01 Rio Dell Overpass RV 0.44, 0.54 14.33
Loma Prieta 1989-10-18 6.93 Capitola RO 0.49, 0.49 15.23
Loma Prieta 1989-10-18 6.93 Gilroy Array #3 RO 0.62, 0.37 12.82
Kobe, Japan 1995-01-16 6.90 Nishi-Akashi SS 0.53, 0.49 8.12
Kobe, Japan 1995-01-16 6.90 Shin-Osaka SS 0.73, 0.76 19.62
Northridge-01 1994-01-17 6.69 Canyon Country-WLC RV 0.33, 0.32 18.36
Northridge-01 1994-01-17 6.69 Canyon Country-WLC RV 0.56, 0.60 12.44
San Fernando 1971-02-09 6.61 LA - Hollywood Stor RV 0.36, 0.56 25.89
Superstition Hills-02 1987-11-24 6.54 El Centro Imp. Co. Cent SS 0.94, 0.65 18.52
Superstition Hills-02 1987-11-24 6.54 Poe Road (temp) SS 0.48, 0.49 11.67
Imperial Valley-06 1979-10-15 6.53 El Centro Array #6 SS 0.68, 0.62 22.54
Imperial Valley-06 1979-10-15 6.53 El Centro Array #7 SS 0.42, 0.45 13.48
Borrego 1942-10-21 6.50 El Centro Array #9 SS 0.58, 0.54 57.7a
Friuli, Italy-01 1976-05-06 6.50 Tolmezzo RV 0.51, 0.40 15.82

Rs is Campbell distance (but one).

a
Epicentral distance.

relies only on the last recorded deformation of the system, and is usually
expressed in terms of the yield strength (Fy) and the level of given lat-
eral strength demand (q′), as described below. This makes the proce-
dure more relevant to the design stage. In contrast, the second
approach has a dual purpose since it comprises a direct ratio of the tran-
sient peak deformation and the residual drift. On the one hand, to esti-
mate the residual drift, an estimate of the transient maximum
displacement needs to be carried out beforehand. Therefore, this strat-
egy builds the estimate of one engineering demand parameter (EDP)
on top of another. However, the duality of this approach enables an es-
timation of the maximum displacement based on the residual drifts,
making this strategy more suitable in the case of post-earthquake as-
sessment where the inelastic residual drifts of the system are known
since they are permanent and directly measurable. Both approaches
are examined herein.
The following subsections explore the effects of degrading systems
on the maximum inelastic response and the inelastic residual response
of equivalent SDOF systems. Measures of central tendency and disper-
sion of maximum and residual drifts are presented jointly with assess-
ments of the effects of the lateral strength demand (q′), fundamental
period (T1), normalised period (T1/Tm), cyclic deterioration rate (Λ)
and post-yield stiffness ratio (αh). The response of residual drifts is ex-
amined using the ESDOF introduced previously. For this purpose, 60
mechanical systems are devised considering a wide range of fundamen-
tal periods from T1 = 0.05s to T1 = 3.0s, and subjected to the ground
motion suite presented in Section 2.4, under seven different levels of
q′. These levels are reached by scaling each individual ground motion
record several times to represent distinct levels of q′, which along
with the scaling factor (SF), are defined as follows:
mSa
0
q ¼
Fy
where m is the mass of the system, Sa is the spectral acceleration for an
individual record at the fundamental period of vibration of the system
with 5% damping, and Fy is the lateral yield strength of the system.
For q′ = 1 in Eq. (3), the maximum displacement is in effect tuned to
the yield displacement, causing the response to remain approximately
linear irrespective of the ground motion applied. Each iterative combi-
nation of parameters is subjected to nonlinear response history analysis,
where both the absolute peak and residual displacement are recorded
to compute their respective inelastic displacement ratios, and therefore
estimate the residual deformation demand using the direct and indirect
approaches.
3.1. Direct approach for design stage evaluation
In this investigation, the maximum inelastic displacement ratio, Cr, is
defined as the maximum inelastic displacement demand, Δi,max, divided
by the product of the given inelasticity level, q′, and the maximum
elastic displacement demand of an equivalent elastic SDOF system.
The value of Δi,max is found as the maximum absolute displacement of
the mass (Fig. 1 (a)) during each response time-history analysis
(RTHA). Likewise, the residual displacement ratio, Cres, is expressed as
the residual inelastic displacement demand, Δi,res, divided by the
product of the inelasticity level of demand and the maximum elastic
displacement demand of an equivalent elastic SDOF system. These
displacement ratios are conceptually equivalent to the ones used
by Ruiz-Garcia and Miranda [4, 17]. Both ratios are expressed as
follows:
C r ¼ Δi;max
ð4Þ
q0 Δyield
ð2Þ

Fy Δi;res
SF ¼ q0 ð3Þ C res ¼ ð5Þ
mSa q0 Δyield
596 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610

Fig. 4. Mean maximum inelastic displacement ratio Cr as a function of the fundamental period T1 for non-degrading (a) and degrading systems (b). Mean residual inelastic displacement
ratio Cres as a function of the fundamental period T1 for non-degrading (c) and degrading systems (d).

3.1.1. Influence of fundamental period and lateral strength demand
To determine both Cr and Cres ratios, a total of 94,080 RTHA were per-
formed. The results are presented below for each hysteretic mechanical
model (i.e., non-degrading and degrading) through the 60 different fun-
damental periods. In order to obtain the central tendency behaviour, the
mean maximum inelastic displacement ratio is obtained using a moving
sample mean to help filter the variability and identify only the mean
trend. Fig. 4 (a) and (b) show the maximum residual displacement
ratio, Cr, for both hysteretic models considered herein. Upon direct com-
parison of the responses, it can be seen that for the same fundamental
period (T1) and the same level of given q′, higher absolute values of Cr
are exhibited when degrading models are used. Similar trends are
observed for both models, where two spectral regions are evident. In
the short period region, the maximum inelastic displacement exhibits
an exponential behaviour that is accentuated for higher levels of q′.
For non-degrading systems, the exponential behaviour starts from
T1 ≈ 0.5s, whereas for degrading systems it starts from around T1
≈ 1.0s. The second region of medium and longer periods tends to con-
verge around unity (i.e. equal displacement rule) for all level of lateral
strength demand. It can be seen that, for non-degrading systems, the
range of periods where Cr = 1 is wider, starting from T1 ≈ 1.5s. For
the case of degrading systems, this only applies to much higher values
of T1.
Similar observations can be made regarding the residual displace-
ment ratios, for non-degrading and degrading systems, in terms of
global trends, as shown in Fig. 4 (c) and (d). For the entire range of
periods, and for both approaches, Cres tends to converge to the same
absolute value for higher levels of q′. As before, two regions of periods
are shown, and similar observations as for Cr occur. For medium and
long periods, although the values of Cres are period-independent
(i.e. flat line), they do not converge to the same values as was observed
for the case of Cr, for different levels of q′.

Fig. 5. Mean maximum inelastic displacement ratio Cr as a function of the tuning ratio T1/Tm for non-degrading (a) and degrading systems (b). Mean residual inelastic displacement ratio
Cres as a function of T1/Tm for non-degrading (c) and degrading systems (d).
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
3.1.2. Influence of ground motion frequency content
In the results presented in this section, the fundamental period is
divided by the mean period of the respective ground motion record
used to perform the nonlinear dynamic analysis (see Section 2.4). For
the case of the maximum displacement ratio shown in Fig. 5 (a) and
(b), the two normalised spectral regions, T1/Tm, are better defined in
comparison with Fig. 4 (a) and (b). Again, comparing the two models,
higher absolute values of Cr are found in the case of degrading systems,
throughout similar levels of q′ and T1/Tm. Comparing the results of Cr,
there is a noticeable point of convergence at which both structural
models produce values of Cr close to unity. However, whereas this
point occurs at T1/Tm ≈ 1.5s for the non-degrading system, it is now
approximately 2.5 for degrading models. For the case of residual dis-
placement ratios, Cres, presented in Fig. 5 (c) and (d) similar observa-
tions regarding the influence of fundamental period remain valid as
discussed below.
In addition to the influence of ground motion frequency content on
peak and residual inelastic drift, it is worth observing the dispersion of
Cr and Cres for both non-degrading and degrading cases. The assessment
of the dispersion corresponds to the record-to-record variability quanti-
fication. To assess the dispersion, coefficients of variation (CoV) of Cr and
Cres are determined considering normalised periods and levels of q′. Fig. 6
(a) and (b) show coefficients of variation of maximum inelastic displace-
ment ratio for non-degrading and degrading systems, respectively. The
same trend observed in the computation of Cr is observed herein in
both cases. In general, the record-to-record variability increases for
higher levels of q′ in both structural systems, exhibiting comparable
levels of dispersion at low levels of q′. However, as q′ increases, the dis-
persion in the case of degrading systems grows more significantly. This
behaviour is more pronounced in the shorter region of T1/Tm lower
than 1.0 (where higher levels of Cr were computed), where the CoV
values reach approximately twice the level of their non-degrading coun-
terparts, particularly at higher levels of lateral strength demand.
The record-to-record variability is subsequently computed for the
case of residual drift ratios, Cres. The CoV values obtained for non-
degrading and degrading cases are shown in Fig. 6 (c) and (d), respec-
tively. It is clear that higher dispersion is displayed by Cres compared
to Cr in Fig. 6 (a) and (b). This observation is in agreement with former
studies performed on SDOF systems (e.g., [4]). In particular, despite a
higher absolute dispersion of Cres when compared to Cr, the variation
in CoV for different levels of q′ is less clear. Similar levels of disperssion
Fig. 6. Coefficient of Variation (CoV) of maximum inelastic displacement ratio Cr with respect to T1/Tm for non-degrading (a) and degrading systems (b); CoV of residual modification factor
Cres with respect to T1/Tm for non-degrading (c) and degrading systems (d).
597
are observed for all inelasticity levels, with higher record-to-record var-
iability obtained when degrading structural models are used.
3.1.3. Influence of cyclic degradation rate
To assess the influence of cyclic deterioration directly, the relative
inelastic ratios computed from degrading and non-degrading systems
are determined. Firstly, the maximum inelastic displacement ratio of
degrading systems (Cr,deg) is normalised by the maximum inelastic dis-
placement ratio computed from non-degrading systems (Cr,non−deg), as
shown in Fig. 7. For this purpose, three different rates of cyclic degrada-
tion are used. Results are only reported in terms of T1/Tm based on
observations made above. The degrading systems in Fig. 7
(a) corresponds to the reference model SDOF system used above,
which has a moderate rate of cyclic deterioration (i.e., ∧ = 12.53). The
second degrading range of systems in Fig. 7 (b) accounts for a low rate
of cycling deterioration (i.e., ∧ = 20), and the third range of systems
in Fig. 7 (c) represents a high rate of cycling deterioration (i.e., ∧ =
7.93). These three selected rates are tuned with the equivalent rates of
cyclic deterioration estimated from the 54 steel multi-storey frames,
based on the calibration procedure described on Section 2.1. It can be
seen from the figures that the rate of cyclic deterioration plays a signif-
icant role in the maximum inelastic drift demands, particularly for short
period ratios. An exponential increase of the rate is observed in all cases
at short period ratios, added to the common trend of higher rates for in-
creasing lateral strength demand. For a normalised period T1/Tm = 1.0,
Cr is on average 12% higher for low rates of cyclic deterioration assuming
q′ = 6. The same case can be up to 17% higher for a medium deteriora-
tion rate, and achieve a rate of up to 50% when high rates of deteriora-
tion are used.
The same ratios discussed above for Cr are also presented for Cres in
Fig. 8. In this case, the overall trends are similar, but the dispersion of
the results for Cres is higher in comparison with Cr. Therefore, the differ-
ence between the different levels of inelasticity is less clear, and in some
regions of T1/Tm even exhibiting a reversal in trends. For the three
degrading models, the increase in the response of residual drifts is
more significant than that observed above in the case of maximum
drift demand, especially in the shorter range of the normalised spectral
region. For the low and moderate rate of deterioration and at lower
levels of q′, some reversal in behaviour is observed since the converging
flat lines are marginally below unity. Nonetheless, in this same spectral
region, an amplification of residual drifts can be seen for relatively high
598 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610

Fig. 7. Maximum inelastic displacement ratio of degrading (Cr,deg) to non-degrading (Cr,non−deg) systems, for three levels of cyclic deterioration: (a) moderate, (b) low, and (c) high.

levels of inelasticity. The latter observation remains valid for the case of
high rate of cyclic deterioration as shown in Fig. 8 (c).
3.1.4. Influence of post-yield stiffness ratio
In order to illustrate the influence of the post-yield stiffness ratio, αh,
two distinctive levels are considered herein for the SDOF systems. This
value is computed as the ratio between the post-yield and the elastic
stiffness, and it has been shown in previous studies to have an influence
on residual drift demands [3, 20, 52]. The response is obtained using
both non-degrading and degrading hysteretic models, in order to main-
tain the basis for comparison. It is important to note that this can be
readily determined in a study on SDOF systems, which is not the case
for a more realistic MDOF situation. However, an estimation may be
established from an idealised equivalent level of the post-yield stiffness
ratio, obtained from a nonlinear static (pushover) analysis. In the case of
the family of 54 multi-storey buildings considered in this study, push-
over analysis has shown a median value of αh of 5%, with a minimum
ratio of 3% and a maximum ratio of 21%. The value of αh for each individ-
ual frame is presented in the rightmost column of Table 1. Fig. 9 shows
the influence of αh on the computation of Cres, when a relatively high
ratio (i.e., αh = 10%) and a comparatively low ratio (i.e., αh = 4%)
are considered. The first observation on αh = 10% for non-
degrading systems, depicted in Fig. 9 (a) and (b), is that these sys-
tems do not show the same increase consistently noted before in Cr
for the low region of T1/Tm. In contrast, a similar behaviour of Cr is ob-
served in the case of T1/Tm ≥ 1.0, where Cr seems also to be unaffected
by the level of q′.
In the case of αh = 4%, less differences are shown by considering
degradation, as observed in Fig. 9 (c) and (d). This level of αh is closer
to a perfect elasto-plastic case (i.e., αh = 0.0%) which corresponds, for
example, to the case of the equivalent SDOF procedure considered in
EC8 (e.g., N2 method). Both non-degrading and degrading systems
show a more clear influence of q′ on the response of Cres throughout
the ranges of T1/Tm when compared to the 10% cases. The influence of
degradation increases with q′, and both systems exhibit a saturation
trend for Cres as q′ increases. Finally, comparing degrading systems
with different αh, as is the case of Fig. 9 (b) and (d), the effect of post-
yield stiffness ratio is evident in terms of the magnitude of Cres as well
as the sensitivity due to the levels of q′, which exhibit a more clear influ-
ence in the case of αh = 4%.

Fig. 8. Residual inelastic displacement ratio of degrading (Cr,deg) to non-degrading (Cr.non−deg) systems, for three levels of cyclic deterioration: (a) moderate, (b) low, and (c) high.
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610 599

Fig. 9. Influence of post-yield stiffness ratio on Cres: (a) and (b) αh = 10% for non-degrading and degrading systems respectively; (c) and (d) αh = 2.0% for non-degrading and degrading
systems respectively.

3.2. Indirect approach for post-earthquake evaluation
The residual drift (Δi,res) or maximum drift (Δi,max) can be estimated
based on the ratio between the two, which is referred to as the residual
ratio SRDR, given in Eq. (6) and following the notation firstly introduced
by Kawashima et al. [16]. It can be seen that SRDR is equivalent to the
ratio between Cres and Cr based on the notations used previously, for
the corresponding direct approach. However, the notation SRDR notation
is also used herein to help differentiate between the direct and indirect
methodologies.
Δi;res
SRDR ¼
Δi; max
The mean residual ratios, SRDR, are computed for the same four hyster-
etic models: degrading and non-degrading reference cases, in addition to
two degrading models accounting for relatively high and low rates of
cyclic deterioration. The residual ratios are presented only in terms of
T1/Tm. Results in terms of the influence of the fundamental period are
not shown here for brevity. As shown in Fig. 10, all systems exhibit a
very similar response where, for the same level of q′, the residual drifts
are largely unaffected by the period ratio, except for very short periods
ratios. In general, higher residual ratios are observed for higher levels of
q′, although the variability tends to be less significant for some periods
where the ratios are mainly overlapping. The residual ratio also tends
to reach an upper limit for all given lateral strength demand levels.
This uppermost limit is slightly higher for the case of non-degrading sys-
tems, and oscillates around 40% of the transient peak deformation. This is
consistent with the observed behaviour for the low deterioration rate in
Fig. 10 (c). In the case of moderate and high rates of cyclic deterioration,
ð6Þ
this uppermost limit tends to be lower, oscillating on average around 35%
to 30%, respectively. It is also important to note that the latter saturation
levels of SRDR may vary as the post-yield stiffness ratio (αh) changes. For
instance, for a higher rate of αh = 10%, a distinctive SRDR saturation oscil-
lating around 0.1 − 0.2 would be obtained, in line with the results shown
in the previous section, although this is not included here for brevity.

Fig. 10. Mean residual ratios, SRDR, as function of T1/Tm. (a) non-degrading systems, and three degrading systems with different rates of cyclic deterioration: (b) moderate, (c) low, and
(d) high.
600 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
Except for very low values of T1/Tm, it is shown that the structural model
used to estimate SRDR does not play a notable role, whenever the maxi-
mum drift is already given. This indicates that the key factors that influ-
ence the global inelastic demand primarily affect the maximum drift
rather than the relationship between the residual and maximum drifts.
This is particularly relevant when using known inelastic residual drifts
to predict maximum drifts, as is the case in post-earthquake structural
assessments.
Following on from the above, a direct comparison of residual ratios
between degrading and non-degrading cases is presented in terms of
normalised period to include the ground motion frequency content, as
shown in Fig. 11. On average, the residual ratio is higher when non-
degrading models are used, since the ratios are below unity. This ratio
is largely unaffected by the period. For the three considered degrading
systems, the difference between the residual ratios is higher
(i.e., higher SRDR computed for non-degrading models) as the lateral
strength demand increases albeit slightly. The stability of the ratio for
both models confirms the correlation between residual and maximum
drift demands. In addition, although the SRDR seems to be consistently
lower for the case of degrading systems, as observed above, the absolute
magnitudes of maximum and residual drift demands are systematically
higher when degrading models are used.
4. Residual drift demands in multi-Storey frames
In order to evaluate the magnitude and distribution of residual drift
demands in realistic MDOF systems, the response of the 54 steel moment
resisting frames described in Section 2.2 is examined through extensive
NRHA. As in the case of the SDOF systems, the full suite of 56 ground mo-
tion records described in Section 2.4 is used in the analysis.
4.1. Methodology and scaling
The methodology followed to investigate the residual drifts demand
in MDOF is essentially based on Incremental dynamic analysis (IDA)
[53], as in the initial SDOF study described above, with the objective of
assessing the residual drift demand for specific levels of seismic lateral
strength demand (q′). To this end, the spectral acceleration at the fun-
damental period of the system is selected as the intensity measure
(IM) to define the scaling of the original records. Subsequently, the scal-
ing factor, SF, used for each record to attain the targeted inelasticity
Fig. 11. Degrading residual ratio, SRDR, deg, normalised by non-degrading residual ratio, SRDR, non−deg, as a function of T1/Tm, for three levels of cyclic deterioration: (a) moderate, (b) low, and
(c) high.
level, q′, is defined as follows:
V1
SF ¼ q0 ð7Þ
Sa ðT 1 Þmγ 1
In the above equation, Sa(T1) is the spectral acceleration for an indi-
vidual record at the fundamental period of the frame (T1), and V1 is the
base shear corresponding to the formation of the first plastic hinge in
the frame when subjected to a first-mode force profile from nonlinear
static (pushover) analysis. For brevity, the outcome of the pushover
analysis for the full family of frames is not presented here, but can be
found elsewhere [25]. In the above relationship, m is the seismic mass
and γ1 represents the effective mass participation ratio of the first
mode. Finally, q′ as stated above corresponds to the given lateral
strength demand, that in this study is assumed to have 4 different levels,
namely q′ = 3, 4, 5, and 6. This method of scaling the acceleration
time histories, related to the lateral yield capacity of the structure
(i.e., anchored to V1), enables assessment of the seismic behaviour inde-
pendently of the seismic hazard scenario. It is important to recall that
the 54 frames were designed considering different seismic hazard sce-
narios, among other parameters, as described in Section 2.2.
The results of nearly 25,000 individual NRHA cases, which were
carried out within this study, are presented and discussed below. As
the residual drifts throughout all the stories were recorded, they are
used to examine the residual inter-storey and roof drifts. The central
tendency of the residual drift demands in the frames is characterised
by means of the counted sample median. As the residual drift samples
have an approximate lognormal distribution [17, 18], as shown below,
the sample counted dispersion corresponds to the standard deviation
of the natural logarithm of the raw data. To provide a basis for compar-
ison, all the results are presented considering both degrading and non-
degrading models and, where required, direct comparison is performed
to quantify the influence of cyclic and in-cycle deterioration.
4.2. Residual drifts for frame sets
The distributions of residual drifts ratios (RIDR) over the height are
shown in Fig. 12, arranged by groups of frames, thus by number of
stories. Initially, it can be seen that both structural models exhibit simi-
lar patterns of residual drift demands, with the magnitude of residual
drifts for degrading models being higher than the degrading counter-
part when compared by storey or inelasticity level. Furthermore,
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610 601

Fig. 12. Mean residual drift distribution for all lateral strength demand levels and ground motion records. Set A to Set D in (a)–(d) respectively, for both non-degrading and degrading cases.

residual drift demands are consistently higher as the inelasticity levels
increases and, as discussed below, the influence of deterioration
increases with the increase in the level of q′. Secondly, it can be observed
that, for low-rise frames (i.e., Sets A and B with 3 and 5 stories, respec-
tively), the pattern of residual drift demand over the height tends to be
more uniform, especially at the lowest levels of inelasticity. In the case
of mid-rise frames (i.e., Sets C and D with 7 and 9 stories, respectively)
the residual drift demand distribution is shown to be less uniform and
the highest residual drift amplitudes are concentrated at the base. As
the number of stories increases, the difference between the residual
drift demands computed for both structural models is higher in terms
of absolute drifts as well as in terms of the deformation profile, as the
degrading models exhibit a more rapid concentration of the demand
at the lowermost level of the frames. The effects due to higher modes
of vibration are not clearly perceptible in the residual drift demand at
the uppermost level of taller buildings. At this stage, it is worth noting
that the computed residual drift demand profiles are consistent with
the maximum inter-storey drift demand distribution over the height
exhibited by the frames [25]. This is consistent with the results obtained
in the initial study on SDOF (e.g., Fig. 4 and Fig. 7), as well as with previ-
ous studies [1, 3, 25]. The correlation between peak deformation and re-
sidual deformation demands is discussed further in subsequent parts of
this paper.
With the aim of comparing the residual drift demands of frames
with different heights, two demand parameters of particular interest
are the maximum residual drift at any storey (Max. RISD) and the max-
imum roof residual drift demand (Max. RRD), are assessed. Fig. 13
(a) and (b) show the distribution of the maximum residual drift at

Fig. 13. Mean maximum residual drift (all stories) for various frame sets, for non-degrading (a) and degrading (b) cases.
602 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610

Fig. 14. Mean maximum roof residual drift for various frame sets, for non-degrading (a) and degrading (b) cases.

any storey of the structural system, grouped by height of frames accord-
ing to the sets, with respect to q′. The first significant observation is that
the number of stories does not play a significant role in the maximum
computed residual drift demand. For non-degrading systems, a largely
constant increase of maximum RISD is observed as the lateral strength
demand increases, for all frame sets. For degrading systems, a larger
yet approximately constant increment of maximum RISD is observed
as q′ increases. Nevertheless, in the case of degrading systems, when
the lateral strength demand reaches the highest levels (i.e., q′ = 5 and
6), differences due to the number of stories are shown, although a
clear trend among the sets of frames is present.
In addition to the response of maximum RISD with respect to
the lateral strength demand, the central tendency of the roof resid-
ual drift demand is also extracted and arranged for each frame set.
Fig. 14 (a) and (b) show the case of degrading and non-degrading
systems, respectively. In this case, unlike the maximum RISD, there
is a clear difference in the trends of roof residual drift demands
due to the number of stories (i.e., arranged by sets). A more signif-
icant increase in demand is observed in the case of lower-rise sys-
tems, consistently for both structural models. A faster increase in
demand occurs for degrading systems. This is consistent with the
residual drift demand distribution over the height as discussed
above, where despite the increase of demand with the increase of
lateral strength demand, residual drift deformations tend to be con-
centrated at the lower levels of the frames.
Together with the distribution of residual drift demands, the vari-
ability of the dispersion of the results is statistically important, in
order to account for record-to-record variability, particularly to assess
the differences between both modelling approaches. Fig. 15 (a) to
(d) show the computed dispersion, σlnRISD, arranged by frame set. In

Fig. 15. Mean dispersion of residual drifts over height for frame Sets A-D ((a)-(d)), for non-degrading and degrading cases and different values of q′.
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610 603

Fig. 16. Relationship between mean maximum residual drift, fundamental period and inelasticity level, for non-degrading (a) and degrading (b) systems for all frames.

general, the distribution of the dispersion over the height is significant
for each individual set of frames at every level of lateral strength
demand. Although no trend in the distribution of dispersion is observed
with respect to the number of stories or the level of lateral strength
demand, degrading models seem to exhibit consistently higher disper-
sion. This is more significant for Set D frames, as shown in Fig. 15 (d).
The dispersion of residual drift demand is higher than the dispersion
of transient peak deformations computed for the same systems and
ground motion records.
Besides the above assessment of residual drift demand organised by
set of frames, the influence of fundamental period is examined with
respect to the lateral strength demand level and the maximum residual
drift demand computed at any level. Fig. 16 (a) and (b) depict the vari-
ation of Max. RISD for various inelasticity levels for all 54 frames, and
using both modelling approaches. Both models exhibit a consistent in-
crease in the Max. RISD when the lateral strength demand increases.
In the case of non-degrading systems, this rate increase is more moder-
ate (i.e., flatter surface), whereas in the case of degrading models, the

Fig. 17. Mean correlation over the height between the maximum transient drift and residual drift, for frame Sets A-D ((a)–(d)), for non-degrading and degrading cases.
604 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610

Fig. 18. Height-wise distribution of residual ratios for frame Sets A–D ((a)–(d)), for non-degrading and degrading cases.

rate increases with q′. On average, the Max. RISD increases as the funda-
mental period increases, especially in the case of degrading systems
where this is accentuated. At high levels of lateral strength demand
(i.e., q′ = 5 and 6), the magnitude of Max. RISD demand is significantly
higher for the case of degrading systems, particularly for relatively high
values of fundamental periods.
4.3. Residual drift ratios for frame sets
The statistical results of residual ratios, SRDR, are presented and
discussed in this section. The residual drift ratio is defined by Eq. 6, as
in the initial study on SDOF described in Section 3. The same measures
of central tendency and dispersion, as considered above for residual
drift demands, are used herein. As discussed in detail at SDOF level,
this indirect method can be used to either estimate residual drifts or
transient peak deformations. Hence, to illustrate the efficiency of SRDR
in the case of MDOF systems, the correlation between its components
is presented below, namely residual drift demand (RISD) and maximum
drift demand (Max. ISD), computed at the same storey. The correlation
coefficient ρ is used to express, through a linear relationship, the rela-
tionship between the variables. This is defined as the covariance of
RISD and Max. ISD, normalised by the product of their standard devia-
tions [54]. This coefficient only results in values within the range
[−1,1] and indicates the degree of linear dependence; i.e. there is a lin-
ear correlation when ρ is close to unity.
Fig. 17 (a) to (d) show the median correlation coefficient over the
height arranged by frame sets, and for both structural models. In gen-
eral, high absolute values of ρ are observed for both structural ap-
proaches. Although it seems to depend on the lateral strength demand
level, this relation is not apparent, since non-uniform profiles are

Fig. 19. Mean maximum residual ratio (all stories) per set of frames, for non-degrading (a) and degrading (b) cases.
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
Fig. 20. Relationship between residual ratio, fundamental period and inelasticity level of demand, for non-degrading (a) and degrading (b) models, for all frames.
observed for different lateral strength demand levels. Moreover, a mag-
nitude close to unity is observed at the top storey levels of the low-rise
sets (i.e., Sets A and B) for the case of degrading systems. For the other
two mid-rise sets, there is a reversal of this behaviour, since the higher
linear correlation is observed at lower levels of the frames, where the
transient maximum drifts and residual drifts exhibit higher values, as
noted before.
With the high linear correlation between residual and maximum
drift demands [1, 2], it is interesting to observe the height-wise distribu-
tion of residual drift ratio, SRDR. Fig. 18 (a) to (d) present SRDR distribu-
tions per various frame sets considering both structural models. In
general, similar behaviour is observed irrespective of the number of
stories, for both structural models. The absolute magnitude of SRDR
tends to increase when the lateral strength demand increases, although
the differences from one inelastic demand level to another are relatively
small. The differences between structural models are not clear upon di-
rect comparison for the same storey and q′ level. The only exception is
for the case of the highest city level (i.e., q′ = 6) which shows a higher
rate for degrading models. Moreover, highly uniform SRDR profiles over
the height are observed irrespective of the set of frames or inelasticity
level, irrespective of the magnitude and location of maximum drift de-
mands concentration. The highest computed values of SRDR tend to sat-
urate when they reach values around 0.4, as observed before in the
initial SDOF study corresponding to a post-yield stiffness ratio of about
5%, used for the reference SDOF case.
Besides the distribution of SRDR over the height, the relationship be-
tween the median maximum computed value of SRDR (at any storey
level) and the lateral strength demand, when grouped by frame set, is
also examined. Fig. 19 (a) and (b) depict this relationship for degrading
and non-degrading systems, respectively, and shows nearly identical
trends of maximum values of SRDR. A slight increase in maximum SRDR
as q′ increases, as well as the saturation limit of SRDR corresponding to
the 40% of the maximum drift demand, can be observed, irrespective
of the set of frames. This latter observation, is in agreement with the
initial study conducted on equivalent SDOF systems, as illustrated for
example before in Fig. 10.
When the influence of the period is included, together with the
maximum residual ratio, SRDR, and q′, similar behaviour is observed
for both models as shown in Fig. 20. The median of maximum resid-
ual drift is unaffected by the period, and it is not directly influenced
by structural deterioration. It also remains largely constant irrespec-
tive of the lateral strength demand imposed on the structure. As ob-
served before for SDOF systems, and underpinned by the high linear
correlation found between residual and maximum drift demands,
the residual ratio is an efficient indirect predictor. As noted before,
if appropriately calibrated it can be used to estimate either residual
drifts or especially maximum drifts starting from known permanent
inelastic drifts measured in post-earthquake assessment. In this
605
respect, the following section presents a calibrated predictive
model based on the findings discussed above.
5. Predictive residual drift models for multi-Storey frames
Based on the results discussed above, this section firstly describes
models for predicting residual drift following the direct approach.
Predictive models for estimating the maximum drift based on residual
drift demand are then also presented for use within an indirect
approach, as discussed before. The details and limitations of the models
are described below.
5.1. Direct approach
Supported by the findings regarding the strong linear correlation
between the maximum and residual drift demands, besides the obser-
vations on the dependency of residual drift demands on the fundamen-
tal period of the structure (T1), the ground-motion frequency content
(Tm), and the lateral strength demand (q′), a nonlinear model devel-
oped previously by the authors [25] is extended to account for residual
drifts. The original model was developed based on the same sets of
frames and records, but using the data for maximum drifts, enabling
the prediction of maximum roof and maximum expected storey dis-
placements based on elastic analysis (i.e., direct approach). In the fol-
lowing, the predictive model for the maximum residual drift demand
is introduced.
Similar to the Cres modification factor considered in Section 3.1 for
SDOF systems, a residual drift modification factor, Δresmod, for MDOF sys-
tems, is described herein. This is derived using the recorded permanent
roof drifts from the extensive NRHA presented in Section 4, and is
expressed as follows:
Δres
Δres
mod ¼ ð8Þ
q0 Δ1;roof
where q′ is the given level of lateral strength demand, Δ1,roof is the
roof displacement at first yield obtained from pushover analysis
based on first-mode lateral load pattern, and Δ res corresponds to
the maximum residual drift recorded from the NRHA response. By
definition, this factor enables the prediction of maximum expected
residual drift demand at the design stage by utilising simplified and
Table 3
Nonlinear regression coefficients for the residual drift modification factor [ Δres
mod ].
a0 a1 a2 a3 a4
−2.072 0.201 −1.584 −0.101 −1.779
606 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610

Fig. 21. Maximum residual drift modification factor prediction model: (a) non-degrading model, (b) degrading model, (c) degrading-to-non-degrading ratio, and (d) confidence interval
and dispersion of the model for q′ = 3.

widely used elastic analysis procedures. Following the format of to estimate the regression coefficients were conducted in Matlab and
previous sections, both structural modelling approaches are used the coefficients are presented in Table 3.
for the calibration of Δres
mod to provide a basis for comparison of the in-

    
fluence of structural deterioration on the estimation of residual drift 0 0 T1
ln Δres
mod ¼ a0 þ a1 q þ a2 þ a3 q ln min 1
demand.    Tm
The obtained sample for Δres
mod, accounting for the four levels of lateral
T1
þ a4 ln max 1 ð9Þ
strength demand and all frames sets, comprises 12,096 values (corre- 2:818T m
sponding to 54 frames, 56 ground motion records, and 4 lateral strength
demand levels), for each structural model (i.e., 24,192 data points). The Absolute values of the fitted relationship of Δres mod is given in Fig. 21
statistical sample of Δres
mod are assumed to be log-normally distributed (a) and (b) for the case of the non-degrading model and its degrading
[55], and therefore the nonlinear regression of Δres mod is performed counterpart, respectively. The overall trend of the nonlinear model at dif-
on the log-transformed residual drift modification factor directly ferent levels of inelasticity exhibits a similar behaviour to that observed in
(log(Δres
mod)). A detailed discussion regarding the latter assumption and maximum drift modification factors in previous studies [25, 35, 56], but
its merits is given elsewhere [25] for the equivalent case of the maxi- with a lower magnitude. This is consistent with observations made
mum drift modification factor. above on the strong linear relationship between maximum and residual
The resulting nonlinear prediction model for the residual drift drift demands when assessed over the height of the frames. Three discrete
demand is given in Eq. (9) which, in line with the results and observa- regions can be observed for the response of both degrading and non-
tions discussed above, is dependant on the fundamental period (T1), degrading cases. Within the middle (and wider) region, it is clear that
the frequency content of the motion record (Tm), and to a lesser extent the residual modification factor is relatively unaffected by period ratios
the lateral strength demand (q′). The nonlinear regressions performed ranging from 1.0 to 2.8. For longer periods, the modification factor does

Fig. 22. Residual drift versus maximum inter-storey drift, using all frames based on degrading systems.
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
Fig. 23. Residual roof drift versus maximum roof drift, using all frames based on degrading systems.
not exhibit similar trends for both degrading and non-degrading cases. In
the case of non-degrading models, it increases linearly with T1/Tm,
whereas in the case of degrading systems the trend is largely flat for all
T1/Tm. For short period ratios below 1.0, both cases exhibit nonlinear
trends that increase as T1/Tm decreases. Overall, within the three regions,
the higher the lateral strength demand, the higher the difference between
the models, being more pronounced for shorter period ranges. The latter
is directly quantified in Fig. 21 (c) where the ratio of residual roof modifi-
cation factors of degrading to non-degrading models is presented.
In the first region, where T1/Tm is below 1.0, Δres mod increases as T1/Tm
decreases; the elongated period of the structure approaches the main
period of the ground motion, hence amplifying the peak drift demand
and in turn Δresmod. In the short region of T1/Tm, the latter also explains
the higher demand in the case of degrading models since deterioration
phenomena by definition (i.e. strength and stiffness loss) influence the
period elongation rate. For the intermediate range of T1/Tm, as the
period is elongated further, T1 becomes higher than Tm (i.e. moving
away from first-mode resonance), hence the influence of T1/Tm in this
middle range becomes less significant. For the third region (i.e. period
ratio ranges above 2.8), Δres mod shows increasing trends with T1/Tm as
the mean period of the ground motion approaches the second-mode
of the structure. The clear difference in the slope of Δres mod trends
between the non-degrading and degrading cases can be attributed in
part to the concentration of inelastic demand in lower stories in the
case of degrading models, leading to a reduced influence from higher
modes.
With respect to the influence of q′, in general, the higher the inelas-
ticity the higher Δres
mod throughout all T1/Tm ranges. However, in the case
of degrading models, the change in Δres mod is more significant when q′
increases, with the behaviour being largely unaffected by degradation
for values of q′=3. This is illustrated further in Fig. 21 (c) which depicts
the ratio between Δres res
mod,deg and Δmod,non−deg. It is evident that the influ-
ence of deterioration modelling is most significant in the short region
of T1/Tm, and increases with higher q′. For values of T1/Tm N 1.0, the
influence of deterioration modelling becomes largely unaffected by
the T1/Tm, but remains sensitive to q′. Finally, Fig. 21 (d) depicts the
proposed relationship corresponding to the non-degrading case when
the lateral strength demand is 3, along with the 95% confidence interval
in the estimate of the mean logarithmic residual drift modification
factor, shown in the grey region. The dotted lines designate the
plus/minus one standard deviation of the ln(Δres mod) of the nonlinear
relationship indicated by the central solid line.
5.2. Indirect approach
A predictive model to estimate the maximum deformations, based
on residual drifts recorded from the NRHA, is presented and discussed
in this section. This corresponds to a nonlinear relationship for use in
the context of the indirect approach, where maximum expected drift
demands can be predicted based on known residual drift demands,
which is usually the case in post-earthquake assessment, as discussed
607
before. Therefore, the samples of recorded permanent drifts and
recorded maximum drifts are used, in conjunction with the structural
characteristics of the frames, to find the best nonlinear relationship
between Max. RISD and Max. ISD.
The assembled samples of residual and peak drift demands recorded
from the NRHA results, are initially presented together to illustrate the
close linear relationship between them, as observed above (e.g., see
Fig. 17). A relationship between the residual drifts and maximum
inter-storey drifts, across all stories, is firstly observed, as shown in
Fig. 22. Secondly, only the residual roof drift versus maximum transient
drift demand is depicted in Fig. 23. This disaggregation of the data is
interesting since for such indirect approach it is usually assumed that
the maximum transient drift demand is only likely to occur in the
uppermost levels of an individual structure (e.g., [1]). This may not
however always be the case and, therefore, the nonlinear relationship
to estimate the maximum drift demands based on residual drift
demands is built on top of the data computed for all stories.
Despite the high linear correlations between the variables of inter-
est, it is important to recall that the above relationships include all
sets of frames, hence the individual structural characteristics of each
set cannot be inferred. However, as shown before, several structural
characteristics play a significant role in the distribution and magnitude
of residual and peak drift demands. Therefore, key structural variables
were considered in order to find an optimal nonlinear relationship, by
testing multiple equations. Two predictive models are introduced: the
first can be used with minimal structural information and does not
require static pushover analysis and, the second, requires a static push-
over analysis to determine αh. Eqs. (10) and (11) presented below show
the proposed relationship to estimate the maximum drift demand
(Δmax) as a function of residual drift demand (Δres), for both predictive
models respectively.
T1
Δmax ¼ ao þ a1 Δres þ a2 Δ2res þ a3 Δres T 1 ð10Þ
Ns
 
T1
Δmax ¼ bo þ b1 Δres þ b2 Δ2res þ b3 Δres T 1 ½1 þ b4 α h  ð11Þ
Ns
In addition to Δmax and Δres defined previously, T1 corresponds to the
fundamental period of vibration of the structure and αh is the post-yield
stiffness ratio obtained from pushover analysis using lateral loading pat-
tern complying with the first modal shape. The parameters of the equa-
tion were determined using Matlab, employing least square curve
fitting based on the widely-used Levenberg-Marquardt algorithm. The
Table 4
Nonlinear regression coefficients (Eq. (10))
ao a1 a2 a3
0.122 1.450 −0.582 −0.130
608 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
Table 5
Nonlinear regression coefficients (Eq. (11))
bo b1 b2 b3
0.101 1.384 0.478 −0.418
nonlinear regression coefficients are presented below in Tables 4 and 5,
for both predictive models respectively.
A graphic representation of the goodness of the presented nonlinear
regression can be seen in Fig. 24. Here, the exact values obtained from
the extensive IDA response and used to calibrate the model are plotted
against the resulting predicted values using Eq. (10). The correlation
coefficient when the 90% of the counted sample of data is considered
is R2 = 0.95 and, when the full set of counted points of the sample are
used, is R2 = 0.87.
5.3. Overall fragility assessment
This study has provided an insight into the residual displacement
demands in a large set of steel moment resisting frames designed ac-
cording to the provisions of EC3 [33] and EC8 [11]. The NRHA were
not conducted in order to assess the performance under specific seismic
hazard levels (e.g., damage limitation or significant damage), but
instead they were examined for a range of given constant relative
strength demand, reached by scaling of ground motion records with
respect to the lateral yield capacity of the structure. Those levels of
strength demand induced levels of inelasticity between q′ = 3 and
q′ = 6. The set of 54 frames were designed in order to represent differ-
ent seismic scenarios as described above, in addition to offering a wide
range of periods and structural properties. Consequently, they do not
necessarily correspond to an optimal design, where for example over-
strength due to steel section selection may be minimised. Notwith-
standing this, the levels of seismic performance obtained through
NRHA correspond to those commonly used in practice for designing
moment frames, and various relevant observations can be drawn
based on the overall behaviour.
The median maximum residual drift demand per set of frames is
presented in Table 6, populated from the results discussed in
Section 4.2 (e.g., see Fig. 12). Overall, for the various sets and inelasticity
levels, mean average values between 0.75 and 2.47% are experienced by
the structures. Based on the relevant residual drift limits reported in
previous studies (e.g., Iwata, Sugimoto [23], McCormick, Aburano
[24]), where a residual drift of 0.5% was considered represent total eco-
nomic loss of the structure, the results of this study reveal that the aver-
age of the frames designed according to EC8, have a high potential of
exceeding this limit. These results are comparable to those reported
by Erochko et al. [7] where special moment resisting frames designed
to comply with ASCE 7-05, exhibited average residual drifts between
0.5% and 4.0%, and were considered to have a high potential of suffering
total economic loss, using the same permissible drift of 0.5%.
Fig. 24. Direct comparison of maximum drifts from RTHA versus predicted values based on residual drifts.
Table 6
Median maximum residual drifts per set of frames
b4 Median maximum residual drift, %
4.561 q′ = 3 q′ = 4 q′ = 5 q′ = 6
Set A 0.80 1.32 1.97 2.43
Set B 0.71 1.12 1.54 1.92
Set C 0.75 1.07 1.46 1.67
Set D 0.82 1.17 1.85 2.47
When assessed probabilistically, because of the variability of the
results and to appraise the results beyond average considerations, the
cumulative distribution functions of residual drifts are presented in
Fig. 25. The curves represent the fragility curves of each individual
level of lateral strength demand (q′), taking into consideration both
modelling approaches. Two residual drift demand limits are considered
to assess the probability of exceedance, for each set of frames, as listed in
Table 7. From the results it can be seen that at the lowest level of lateral
strength demand (q′ = 3), 60% and 66% of earthquakes caused a resid-
ual drift demand higher than 0.5%, considering non-degrading and
degrading models, respectively. As shown before, probabilities of
exceedance, for both considered drift limits, are consistently higher in
the case of degrading systems with respect to their non-degrading
counterparts. For the highest level of lateral strength demand, around
90% of the considered ground motions records, induced a permanent
drift higher than 0.5%.
6. Conclusions
This paper has examined residual drift demands in steel moment-
resisting frames designed according to the provisions of EC3 and EC8.
Particular attention has been given to the influence of degradation
modelling as well as that of the ground motion frequency content.
Detailed studies carried out on a set of 54 structural systems and using
a suite of 56 ground motion records provided insights into the influence
of the period of vibration, levels of lateral strength demand, post-yield
stiffness ratio and various structural properties, on the extent and distri-
bution of residual drift demands. Based on the results obtained, the
main conclusions are summarised below.
In the initial study conducted on equivalent SDOF systems including
P-delta effects, residual drift demands were shown to be primarily
governed by the period ratios, post-yield stiffness ratio and to a lesser
extent on the level of lateral strength demand. Systems with shorter pe-
riods of vibration typically exhibited higher permanent drifts, which
increased with the increase in lateral strength demands. Higher ratios
of post-yield stiffness led to lower residual drift demands, with the
higher the post-yield stiffness ratio the lower the influence of lateral
strength demand on residual drift demands is. When the indirect
approach is used, SRDR is notably affected by αh, leading to a residual
drift demand that is approximately 40% of the maximum drift demand
 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610 609

Fig. 25. Fragility curves of median maximum residual drifts (all stories) for each individual level of inelastic demand, for non-degrading and degrading cases.

when a representative αh of 5% is considered as in the SDOF reference maximum residual drift at any storey, Δres, the number of stories,
case. The residual drifts are markedly higher in degrading models in Ns, and the post-yield stiffness ratio, αh.
comparison with non-degrading models, with the differences being Overall, the entire set of buildings experienced significant perma-
more pronounced in relatively short period ranges, when higher rates nent residual drift demands, for all levels of lateral strength demand,
of cyclic deterioration are employed, and for comparatively high lateral with an average magnitude between 0.7% and 2.5% for lateral strength
strength demand levels. Higher levels of record-to-record variability are demands of q′ of 3 and 6, respectively. These values are both perceptible
also consistently observed in residual drift demands when directly com- to occupants and would involve a high repair cost. If a residual drift
pared to peak drift demands. demand threshold of 0.5% is assumed to represent a total loss, the
The 54 multi-storey frames subjected to 4 different levels of lateral results of this study would indicate that moment-resisting frames
strength demand displayed similar general behaviour to that observed designed according to EC3 and EC8 have a high likelihood of incurring
in the SDOF analysis. The maximum residual drift demand, determined irreparable damage from an economic perspective. Probabilities of
at any storey, is typically affected by the number of stories, the level exceeding this limit are on average between 66% and 91% for lateral
of lateral strength demand, the degradation effects and the post-yield strength demand levels of 3 and 6, respectively, in the case of degrading
stiffness ratio. The magnification in residual drift due to degradation structural systems. However, it is important to note that these findings
increases significantly with the number of stories and lateral strength need to be considered alongside the high record-to-record variability
demand level. This observation also applies for the case of maximum obtained in the results, and the lateral frame over-strength which exists
roof residual drift demand. Both structural models indicate higher when non-optimised design procedures are followed. Importantly, this
sensitivity to record-to-record variability in the case of residual drift study points to the need for specific guidelines in seismic codes such as
compared to peak drift demand, with this variability being higher for EC8 with respect to residual drift criteria. The investigation also rein-
degrading models. It is also important to note that the highest demand forces the importance of employing refined models which are capable
is not necessarily located at the uppermost storeys since the maximum of capturing degradation phenomena in order to obtain reliable predic-
residual demands may be concentrated in the lower levels, particularly tions of both peak and residual drift demands.
when degrading models are used. Regarding the indirect approach,
the most relevant observation is that on average, residual drifts Acknowledgments
demands are typically about 30–40% of the maximum drift demand, for
levels of lateral strength demands of 3–6 respectively, for all set of frames. The first author would like to express his sincere gratitude to the
Based on the results obtained, two simplified models to estimate “Consejo Nacional de Ciencia y Tecnología” (CONICYT, Chile) for the
the maximum residual drift and maximum peak displacement funding of his doctoral studies at Imperial College London.
demands were proposed. The first is for a residual drift modification
factor, Δres
mod, which can be applied in conjunction with the maxi- Notation
mum computed displacement at first yielding, as obtained from
elastic analysis, to predict the maximum expected residual drift.
This modification factor is defined as a function of the period ratio The following symbols are used in this paper:
and the level of lateral strength demand. The second is a model
for the prediction of maximum displacement demand (Δmax) once T1 Fundamental period of vibration
the permanent drifts (Δres) are known, as is the case in post- T2 Second period of vibration
earthquake structural assessment. In the latter case, two predictive H Total height of the building
models were suggested, depending on whether a pushover analysis α Plasticity resistance ratio (EC8)
is undertaken or not. The maximum demand Δmax was found to be ρ Beam-to-column ratio
primarily dependent on the period of vibration, the known β Relative storey stiffness
γ, γ1 First mode participation factor
αo Plastic mechanism ratio
αh Post-yield stiffness ratio
Table 7
Probability of exceedance for mean maximum residual drift, for 0.5% and 1%. Tm Mean period of ground motion
Ci Fourier amplitude of an specific frequency
Non-degrading, % Degrading, %
q′ Lateral inelastic demand
Demand Level P(RISD N 0.5%) P(RISD N 1.0%) P(RISD N 0.5%) P(RISD N 1.0%) SF Scaling factor of lateral inelastic demand
q′ = 3 60.51 28.03 66.52 36.51 V1 Base shear (from pushover) at first yield
q′ = 4 75.35 45.43 80.61 57.32 m Total seismic mass
q′ = 5 83.34 57.90 87.23 70.39 Fy Lateral yield strength
q′ = 6 88.35 67.13 91.11 79.01
Sa(T1) Spectral acceleration at T1 – Intensity measure
610 M.A. Bravo-Haro, A.Y. Elghazouli / Journal of Constructional Steel Research 148 (2018) 589–610
Cr Maximum inelastic displacement ratio
Δi,max Maximum inelastic displacement
Δyield Yield displacement for equivalent elastic system
Cres Residual displacement ratio
Δi,res Residual inelastic displacement demand
SRDR Residual ratio
RISD Residual inter-storey drift
RRD Roof residual drift
Δres
mod Residual drift modification factor (predictive model)
Δres Maximum residual drift (NRHA)
Δ1,roof Roof displacement at first yield (from pushover)
Δmax Maximum drift demand (predictive model)
References
[1] A. Christidis, E. Dimitroudi, G. Hatzigeorgiou, D. Beskos, Maximum seismic displace-
ments evaluation of steel frames from their post-earthquake residual deformation,
Bull. Earthq. Eng. 11 (6) (2013) 2233–2248.
[2] G.D. Hatzigeorgiou, G.A. Papagiannopoulos, D.E. Beskos, Evaluation of maximum
seismic displacements of SDOF systems from their residual deformation, Eng. Struct.
33 (12) (2011) 3422–3431.
[3] J. Ruiz-García, E. Miranda, Evaluation of residual drift demands in regular multi-
storey frames for performance-based seismic assessment, Earthq. Eng. Struct. Dyn.
35 (13) (2006) 1609–1629.
[4] J. Ruiz-García, E. Miranda, Residual displacement ratios for assessment of existing
structures, Earthq. Eng. Struct. Dyn. 35 (3) (2006) 315–336.
[5] S. Uma, S. Pampanin, C. Christopoulos, Development of probabilistic framework for
performance-based seismic assessment of structures considering residual deforma-
tions, J. Earthq. Eng. 14 (7) (2010) 1092–1111.
[6] E. Bojórquez, J. Ruiz-García, Residual drift demands in moment-resisting steel
frames subjected to narrow-band earthquake ground motions, Earthq. Eng. Struct.
Dyn. 42 (11) (2013) 1583–1598.
[7] J. Erochko, C. Christopoulos, R. Tremblay, H. Choi, Residual drift response of SMRFs
and BRB frames in steel buildings designed according to ASCE 7-05, J. Struct. Eng.
137 (5) (2010) 589–599.
[8] D. Pettinga, C. Christopoulos, S. Pampanin, M.J.N. Priestley, Effectiveness of simple
approaches in mitigating residual deformations in buildings, Earthq. Eng. Struct.
Dyn. 36 (12) (2007) 1763–1783.
[9] J. Ruiz-García, J.D. Aguilar, Aftershock seismic assessment taking into account
postmainshock residual drifts, Earthq. Eng. Struct. Dyn. 44 (9) (2015) 1391–1407.
[10] S. Uma, S. Pampanin, C. Christopoulos, A probabilistic framework for performance-
based seismic assessment of structures considering residual deformations, Proceed-
ings of the 1st ECEES, Geneva, Switzerland 2006, p. 731.
[11] CEN, EN-1998-1:2004, Eurocode 8: Design of Structures for Earthquake Resistance –
Part 1: General Rules, Seismic Actions and Rules for Buildings, European Committee
for Standardization, Brussels, 2004.
[12] ASCE/SEI, Minimum Design Loads for Buildings and Other Structures, American
Society of Civil Engineers/Structural Engineering Institute, Reston, VA, 2010.
[13] CEN, European standard EN 1998-3:2005, Eurocode 8 – Design of Structures for
Earthquake Resistance – Part 3: Assessment and Retrofitting of Buildings, Comite
Europeen de Normalisation, Brussels, 2005.
[14] Seismic Evaluation and Retrofit of Existing Buildings: ASCE Standard ASCE/SEI 41-
13, American Society of Civil Engineers, 2014.
[15] P58 FEMA, Seismic performance assessment of buildings, Prepared by Applied
Technology Council as ATC, 2012, 2012, p. 58.
[16] K. Kawashima, G.A. MacRae, J.I. Hoshikuma, K. Nagaya, Residual displacement
response spectrum, J. Struct. Eng. 124 (5) (1998) 523–530.
[17] J. Ruiz-García, E. Miranda, in: John A. Blume (Ed.), Performance-based Assessment
of Existing Structures Accounting for Residual Displacement. Report No-153,
Stanford University, Stanford, CA, 2005 , Earthquake Engineering Center.
[18] J. Ruiz-García, E. Miranda, Probabilistic estimation of residual drift demands for seis-
mic assessment of multi-story framed buildings, Eng. Struct. 32 (1) (2010) 11–20.
[19] U. Yazgan, A. Dazio, Post-earthquake damage assessment using residual displace-
ments, Earthq. Eng. Struct. Dyn. 41 (8) (2012) 1257–1276.
[20] S. Pampanin, C. Christopoulos, M.J.N. Priestley, Performance-based seismic response
of frame structures including residual deformations part II: multi-degree of freedom
systems, J. Earthq. Eng. 7 (01) (2003) 119–147.
[21] J. Ruiz-García, C. Chora, Evaluation of approximate methods to estimate residual drift de-
mands in steel framed buildings, Earthq. Eng. Struct. Dyn. 44 (15) (2015) 2837–2854.
[22] C. Christopoulos, S. Pampanin, M. Nigel Priestley, Performance-based seismic
response of frame structures including residual deformations part I: single-degree
of freedom systems, J. Earthq. Eng. 7 (01) (2003) 97–118.
[23] Y. Iwata, H. Sugimoto, H. Kuguamura, Reparability limit of steel structural buildings
based on the actual data of the Hyogoken-Nanbu earthquake, Proceedings of the
38th Joint Panel. Wind and Seismic Effects. NIST Special Publication, 1057, 2006,
pp. 23–32.
[24] J. McCormick, H. Aburano, M. Ikenaga, M. Nakashima, Permissible residual deforma-
tion levels for building structures considering both safety and human elements, Pro-
ceedings of the 14th World Conference on Earthquake Engineering, 2008.
[25] M. Bravo-Haro, A. Tsitos, A.Y. Elghazouli, Drift and rotation demands in steel frames
incorporationg degradation effects, Bull. Earthq. Eng. (2018) , In press - available
online https://doi.org/10.1007/s10518-018-0389-6.
[26] M. Chenouda, A. Ayoub, Inelastic displacement ratios of degrading systems, J. Struct.
Eng. 134 (6) (2008) 1030–1045.
[27] A.K. Chopra, C. Chintanapakdee, Inelastic deformation ratios for design and evalua-
tion of structures: single-degree-of-freedom bilinear systems, J. Struct. Eng. 130 (9)
(2004) 1309–1319.
[28] A. Nassar, J. Osteraas, H. Krawinkler, Seismic design based on strength and ductility
demands, 10th World Conference on Earthquake Engineering, 1992.
[29] J. Ruiz-García, E. Miranda, Inelastic displacement ratios for evaluation of structures
built on soft soil sites, Earthq. Eng. Struct. Dyn. 35 (6) (2006) 679–694.
[30] L.F. Ibarra, R.A. Medina, H. Krawinkler, Hysteretic models that incorporate strength
and stiffness deterioration, Earthq. Eng. Struct. Dyn. 34 (12) (2005) 1489–1511.
[31] D.G. Lignos, H. Krawinkler, Deterioration modeling of steel components in support
of collapse prediction of steel moment frames under earthquake loading, J. Struct.
Eng. 137 (11) (2010) 1291–1302.
[32] F. De Luca, D. Vamvatsikos, I. Iervolino, Near-optimal piecewise linear fits of static
pushover capacity curves for equivalent SDOF analysis, Earthq. Eng. Struct. Dyn.
42 (4) (2013) 523–543.
[33] CEN, EN-1993-1-1:2005, Eurocode 3: Design of Steel Structures, Part 1–1: General
rules and rules for buildings, European Committee for Standardization, Brussels,
2005.
[34] A.Y. Elghazouli, Assessment of European seismic design procedures for steel framed
structures, Bull. Earthq. Eng. 8 (1) (2010) 65–89.
[35] M. Kumar, P. Stafford, A.Y. Elghazouli, Influence of ground motion characteristics on
drift demands in steel moment frames designed to Eurocode 8, Eng. Struct. 52
(2013) 502–517.
[36] T.L. Karavasilis, N. Bazeos, D.E. Beskos, Drift and ductility estimates in regular steel
MRF subjected to ordinary ground motions: a design-oriented approach, Earth-
quake Spectra 24 (2) (2008) 431–451.
[37] F. McKenna, OpenSees: a framework for earthquake engineering simulation,
Comput. Sci. Eng. 13 (4) (2011) 58–66.
[38] A.K. Chopra, F. McKenna, Modeling viscous damping in nonlinear response history
analysis of buildings for earthquake excitation, Earthq. Eng. Struct. Dyn. 45 (2)
(2016) 193–211.
[39] ATC, Interim guidelines on modeling and acceptance criteria for seismic design and
analysis of tall buildings, PEER/ATC, Redwood City, California, 2010.
[40] A. Gupta, H. Krawinkler, Seismic demands for the performance evaluation of steel
moment resisting frame structures, Stanford University, 1999.
[41] A. Tsitos, M.A. Bravo-Haro, A.Y. Elghazouli, Influence of deterioration modelling on
the seismic response of steel moment frames designed to Eurocode 8, Earthq. Eng.
Struct. Dyn. 47 (2) (2018) 356–376.
[42] F.A. Charney, Unintended consequences of modeling damping in structures, J. Struct.
Eng. 134 (4) (2008) 581–592.
[43] A. Hardyniec, F. Charney, An investigation into the effects of damping and nonlinear
geometry models in earthquake engineering analysis, Earthq. Eng. Struct. Dyn. 44
(15) (2015) 2695–2715.
[44] F. Zareian, R.A. Medina, A practical method for proper modeling of structural
damping in inelastic plane structural systems, Comput. Struct. 88 (1) (2010)
45–53.
[45] D.G. Lignos, T. Hikino, Y. Matsuoka, M. Nakashima, Collapse assessment of steel
moment frames based on E-Defense full-scale shake table collapse tests, J. Struct.
Eng. 139 (1) (2012) 120–132.
[46] T.D. Ancheta, R.B. Darragh, J.P. Stewart, E. Seyhan, W.J. Silva, B.S.J. Chiou, K.E.
Wooddell, R.W. Graves, A.R. Kottke, D.M. Boore, NGA-West2 database, Earthquake
Spectra 30 (3) (2014) 989–1005.
[47] E. Miranda, J. Ruiz-Garcia, Influence of stiffness degradation on strength demands of
structures built on soft soil sites, Eng. Struct. 24 (10) (2002) 1271–1281.
[48] E.M. Rathje, F. Faraj, S. Russell, J.D. Bray, Empirical relationships for frequency con-
tent parameters of earthquake ground motions, Earthquake Spectra 20 (1) (2004)
119–144.
[49] E.M. Rathje, N.A. Abrahamson, J.D. Bray, Simplified frequency content estimates of
earthquake ground motions, J. Geotech. Geoenviron. 124 (2) (1998) 150–159.
[50] A.Y. Elghazouli, M. Kumar, P. Stafford, Prediction and optimisation of seismic drift
demands incorporating ground motion frequency content, Bull. Earthq. Eng. 12
(1) (2014) 255–276.
[51] M. Kumar, J. Castro, P. Stafford, A.Y. Elghazouli, Influence of the mean period of
ground motion on the inelastic dynamic response of single and multi degree of free-
dom systems, Earthq. Eng. Struct. Dyn. 40 (3) (2011) 237–256.
[52] G.A. Macrae, K. Kawashima, Post-earthquake residual displacements of bilinear
oscillators, Earthq. Eng. Struct. Dyn. 26 (7) (1997) 701–716.
[53] D. Vamvatsikos, C.A. Cornell, Incremental dynamic analysis, Earthq. Eng. Struct. Dyn.
31 (3) (2002) 491–514.
[54] J.R. Benjamin, C.A. Cornell, Probability, statistics, and decision for civil engineers,
Courier Corporation, 2014.
[55] E. Limpert, W.A. Stahel, M. Abbt, Log-normal distributions across the sciences: keys
and clues: on the charms of statistics, and how mechanical models resembling gam-
bling machines offer a link to a handy way to characterize log-normal distributions,
which can provide deeper insight into variability and probability—normal or log-
normal: that is the question, AIBS Bull. 51 (5) (2001) 341–352.
[56] T. Karavasilis, N. Bazeos, D. Beskos, Maximum displacement profiles for the perfor-
mance based seismic design of plane steel moment resisting frames, Eng. Struct.
28 (1) (2006) 9–22.