Composite Structures 150 (2016) 28–40

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Finite element modelling of short steel tubes filled with rubberized

concrete
A.P.C. Duarte a, B.A. Silva b, N. Silvestre a,⇑, J. de Brito b, E. Júlio b, J.M. Castro c
a
IDMEC, LAETA, Department of Mechanical Engineering, Instituto Superior Técnico, Universidade de Lisboa, Portugal
b
CEris/ICIST, Department of Civil Engineering, Architecture and Georesources, Instituto Superior Técnico, Universidade de Lisboa, Portugal
c
CONSTRUCT-LESE, Department of Civil Engineering, Faculty of Engineering (FEUP), University of Porto, Portugal

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

Article history: This paper presents a numerical investigation on the ductility and strength of short steel tubes filled with
Received 8 January 2016 Rubberized Concrete (RuC), which is a composite material that mixes concrete with rubber particles. This
Revised 11 April 2016 research concerns the enhancement of both ductility and energy absorption of CFST by considering a core
Accepted 25 April 2016
of RuC instead of normal concrete (NC). First, a brief literature review on the topic is presented. Then,
Available online 27 April 2016
based on an experimental programme conducted by the authors, numerical models of CFST and
RuCFST columns are developed. The results of non-linear analyses (ultimate strengths, load-shortening
Keywords:
curves and failure modes) are validated using experimental data, and good agreement is shown.
Rubberized Concrete (RuC)
Concrete Filled Steel Tubes (CFST)
Finally, a numerical study on the properties of confined NC and RuC is conducted. It is concluded that
Finite elements the concrete damaged plasticity model can be used to simulate RuC. The dilation angle plays a key role
Ductility in RuC and its lower value (compared to that of NC) influences the concrete confinement. Taking into
Confinement account the RuC dilation angle, steel yield stress and tube local slenderness, a new formula is proposed
to predict the concrete core confinement of the studied CFST and RuCFST columns with circular sections.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction decreases the natural frequency of a structural element and leads

One of the urgent demands for today’s sustainability is recy-
cling scrap tyres. Many companies are working towards putting
together sustainable programs to create more opportunities for
tyres to be recycled instead of discarding them in landfills. One
possible solution is to use rubber tyre particles as aggregates in
concrete. In the resulting Rubberized Concrete (RuC), the natural
aggregates are partially replaced by rubber aggregates, which can
be fabricated from tyres via either a cryogenic process or a
mechanical process. Additionally, this replacement of natural
aggregates by rubber particles implies less extraction of natural
resources, thus reducing the environmental impact.
During the last three decades, several authors have investigated
RuC. During the 90’s, Topçu [1] and Li et al. [2] pioneered the study
of the mechanical properties of RuC. These authors concluded that
most mechanical and physical properties of RuC are worse than
those of normal concrete (NC), the exception being its improved
ductility, thus recommending its use in applications where energy
absorption capacity is required and high strength is unnecessary. Li
et al. [2], and later Zheng et al. [3], concluded that the use of RuC
⇑ Corresponding author.
E-mail address: nsilvestre@ist.utl.pt (N. Silvestre).
to an increase of its damping ratio value compared to those of NC.
Several factors, such as the rubber particle size, the process of
production (mechanical or cryogenic), the percentage of replace-
ment, and the replacement of fine, coarse or total natural aggre-
gates with rubber particles, may play a key role on the properties
of RuC. Recently, Valadares et al. [4] studied the mechanical prop-
erties of 12 RuC mixes by extensively varying the parameters pre-
viously referred to. In short, even though the aforementioned
studies present a widespread set of parameters regarding RuC
compositions, overall conclusions indicate that RuC has lower
mechanical properties than NC, namely lower Young’s modulus
and lower (compressive and tensile) strength. This is a natural con-
sequence of the lower strength and stiffness of rubber particles, in
comparison with natural aggregates (NA). On the other hand, RuC
is lighter than NC because rubber particles have lower density than
natural aggregates. Because the decrease of stiffness is higher than
the decrease of mass, RuC structures have vibration frequencies
slightly lower than NC ones. However, the higher damping ratio
of RuC systems can be an advantage in some structures subjected
to dynamic loads. Additionally, RuC also has ultimate strain (exten-
sion) higher than NC, which is a valuable property regarding duc-
tility and energy absorption requirements.

http://dx.doi.org/10.1016/j.compstruct.2016.04.048
0263-8223/Ó 2016 Elsevier Ltd. All rights reserved.
 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40
Concrete Filled Steel Tubes (CFST) are one of the most successful
composite structural solutions available in the construction indus-
try. In fact, in CFST columns, the steel tube acts as formwork and
provides confinement to the concrete core, improving its strength
and ductility, whereas the concrete core reduces the steel tube sen-
sitivity to local buckling. In case of thin (and very thin) steel tubes,
the susceptibility to local buckling increases and, consequently, the
ductility of the tube decreases and its ability to dissipate energy
from dynamic actions also decreases [5–7]. Therefore, if thin-
walled steel tubes are used in CFST solutions, particular attention
should be paid to their ductility, a mandatory requirement for
structures located in seismic areas. In this scope, to increase the
CFST column’s ductility, by increasing the ductility of its concrete
core, would have clear advantages, and this can be achieved by
replacing NC with RuC. In reality, this replacement has two major
benefits:

The drop of RuC strength, in comparison to NC, must be lower if
concrete is confined. Thus, the strength decrease of confined
RuC (in parallel to NC) shall not be as severe as in the uncon-
fined case.

The gain in energy absorption must be higher when using RuC
instead of NC. Thus, the ductility of Rubberized Concrete Filled
Steel Tubes (RuCFST) columns shall increase in comparison with
their CFST counterparts.
Taking advantage of both the structural performance of the
CFST columns and of the improved ductility and energy dissipation
capacities of RuC compared to those of NC [1–4], RuCFST were
recently studied [8,9]. Duarte et al. [8] conducted an experimental
investigation on short cold formed steel tubes, considering three
cross section geometries (circular, square and rectangular), filled
with three concrete mixes (one NC and two RuC mixes with 5%
and 15% replacement of total NA volume with coarse rubber aggre-
gates), subjected to monotonic concentric compression. The
authors’ main conclusion was that, even though the strength and
stiffness of RuCFST columns were lower than those of CFST
columns, the former presented higher ductility than the latter,
especially for columns with circular sections. As previously men-
tioned, this is a major benefit for structures in seismic areas, where
energy dissipation requirements are mandatory. Silva et al. [9]
experimentally studied the flexural behaviour of long RuCFST cir-
cular columns subjected to both monotonic and cyclic bending
testing. These authors observed that the RuCFST column (i) with
diameter of 219 mm and tube thickness of 5 mm and (ii) infilled
with RuC15, presented a maximum drop of lateral force ranging
between 6% (monotonic) and 8% (cyclic) of its CFST counterpart.
The drift for maximum lateral load of the RuCFST column
presented an increase between 6% (monotonic) and 22% (cyclic)
of that exhibited by the CFST column.
The main objective of the research presented herein was to
develop and calibrate numerical models of both CFST and RuCFST
short columns experimentally studied by Duarte et al. [8] and to
analyse and draw conclusions regarding their performance. During
the last decade, several numerical investigations [10–13] have
been carried out to study the behaviour of CFST columns with dif-
ferent types of concrete inside (e.g. high-strength concrete). This
paper presents, for the first time, the modelling and simulation
of the structural behaviour of RuCFST columns. Hence, the first
challenge of this study results from how to model the mechanical
behaviour of rubberized concrete. Since, as described before, most
studies on RuC properties have mainly an experimental nature
[1–4], Duarte et al. [14] decided to firstly develop a numerical
study on RuC at a macroscale material’s level. These authors
employed an Image Processing-Extended Finite Element Method
coupled procedure, allowing the separate modelling of rubber
29
particles and concrete matrix with the latter presenting crack ini-
tiation and propagation. Taking into account the previous asser-
tions, the study described in the present paper intends to provide
a numerically study of RuC properties, but now at a structural level
and within the context of CFST columns. In particular, one of the
objectives is to assess if the concrete damaged plasticity model,
usually employed for standard concrete simulations, can be
extended to the simulation of RuC. Firstly, the development of
the numerical models of the circular, square and rectangular CFST
and RuCFST tested columns [8] is described in detail, with special
attention given to the modelling aspects of the concrete mixes
(NC and RuCs). Then, the numerical models are validated by com-
paring the numerical and experimental collapse strengths, com-
pressive load-axial shortening curves and collapse configurations.
Finally, as the numerical models show good agreement with the
experimental tests, an analysis of the behaviour of confined stan-
dard and Rubberized Concrete is made and conclusions are drawn.
2. Description of numerical models
In this section, the numerical models of the short columns
tested by Duarte et al. [8] are described. Regarding the geometry
of the columns, all sections (circular, square and rectangular)
experimentally studied by Duarte et al. [8] were modelled. For each
cross-section geometry, the five (four in the case of rectangular
sections) configurations (variations of steel tube diameter/width
and thickness) and, for each of these, the three steel grades of
the tested columns [8] were investigated (S235, S275 and S355).
Regarding the concrete cores, the three concrete mixes previously
investigated [8] were considered: (i) a normal concrete mix (NC)
and two Rubberized Concrete (RuC) mixes, obtained by replacing
(ii) 5% (RuC5) and (iii) 15% (RuC15) of the total natural aggregates
volume of the NC composition with coarse (4–11.2 mm size) tyre
rubber particles, in the coarse fraction of aggregates.
Static monotonic non-linear geometrical analyses of the models
were performed using the finite element (FE) commercial package
ABAQUS [15] and an incremental-iterative scheme based on the
modified Riks method.
2.1. Geometry and FE mesh
The geometry, size, steel grade and concrete mixes of the col-
umns are shown in Table 1 [8]. Taking the specimen labelled
R120x2_235_0 (Table 1) as an example: (i) ‘‘R” stands for the rect-
angular cross-section shape (‘‘C” – circular and ‘‘S” – square), (ii)
‘‘120x2” are the nominal major exterior width and nominal tube
thickness in millimetres, respectively, (iii) ‘‘235” is the S235 steel
nominal yield stress in N/mm2 and (iv) ‘‘0” stands for a NC core
(‘‘5” – RuC5 and ‘‘15” – RuC15). Additionally, in Table 1: (i) B is
the nominal exterior and nominal major exterior width of the
square and rectangular columns, respectively, (ii) b is the nominal
exterior minor width of the rectangular columns, (iii) D is the nom-
inal exterior diameter of the circular columns, (iv) t is the measured
tube (wall) thickness, (v) re is the measured exterior corner radius of
square and rectangular columns and (vi) H is the height of the short
columns (i.e. specimen length between 300 mm and 500 mm).
Fig. 1 provides an overview of the circular, square, and rectan-
gular columns FE models’ geometry and mesh. Due to the double
symmetry of the columns, only one fourth of each specimen was
modelled with adequate boundary conditions provided [10,11].
Hence, nodes from both concrete and steel parts in the XZ symme-
try plane were restrained in the Y axis direction and nodes in the
symmetry plane YZ were restrained in the X axis direction.
This option allowed the use of a refined mesh, thus improving
the numerical solution without additional computational time
(see Appendix A).
30 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40

Table 1
Geometrical and material characterization of CFST and RuCFST [8].

Geometry Specimens B [mm] b [mm] D [mm] t [mm] re [mm] H [mm] D/t, B/t
Circular C114x3_235_0/5/15 – – 114 2.70 – 300 42.2
C114x3_275_0/5/15 – – 114 3.20 – 300 35.6
C152x3_275_0/5/15 – – 152 2.85 – 500 53.3
C219x4_235_0/5/15 – – 219 3.85 – 500 56.9
C219x4_355_0/5/15 – – 219 4.25 – 500 51.5
Square S100x3_235_0/5/15 100 – – 3.00 9.3 500 33.3
S100x3_355_0/5/15 100 – – 3.05 5.0 500 32.8
S150x3_235_0/5/15 150 – – 3.05 5.0 500 49.2
S220x4_235_0/5/15 220 – – 3.70 8.5 500 59.5
S220x4.5_355_0/5/15 220 – – 4.70 12.0 500 46.8
Rectangular R120x2_235_0/5/15 120 60 – 1.65 8.0 500 72.7
R120x3_275_0/5/15 120 60 – 2.75 6.0 500 43.6
R160x5_275_0/5/15 160 80 – 4.50 12.0 500 35.6
R200x4_235_0/5/15 200 100 – 3.90 12.0 400 51.3

S6 R6
C4
C3 S4 S5 R4 R5

H/2 H/2 H/2

S3 R3
C2
H/10 S2 H/10 R2
C1 S1 H/10 R1

C2=C4 S3=S6 S2=S5

R3=R6 R2=R5

C1=C3 S1=S4 R1=R4

(a) (b) (c)

Fig. 1. Geometry and mesh of the FE models of the columns with (a) circular, (b) square, and (c) rectangular sections.

For the modelling of both steel and concrete parts, 8 node solid
three-dimensional brick FEs with full integration, designated by
C3D8 in ABAQUS [15], were adopted. Whereas for the simulation
of the concrete core, solid elements are the most common option,
for the simulation of steel tube, in CFST models, the adoption of
both solid or shell finite elements can be found in the literature
[10–12]. Because both solid (3D) and shell finite elements led to
similar results (see the study in Appendix A), we opted to follow
Ellobody et al. [10] and Dai and Lam [12] and use solid (3D) finite
elements to discretize the steel tubes.
In the longitudinal direction, along the specimens’ height, H, the
length of the FEs was kept uniform: H/100 for the steel part, and
H/50 for the concrete part. Along the cross-section contour, the
width of the FEs was kept constant: D/25 and B/25 for the steel
part, and D/12.5 and B/12.5 for the concrete part, respectively for
the circular and the square/rectangular columns, having D and B
the meaning previously referred to. In short, the mesh of the
numerical models with circular, square, and rectangular cross-
sections has, on average, 3970, 4620 and 2960 FEs, respectively.
2.2. Material
Regarding the cold-formed steel tubes, the three steel grades
studied (S235, S275 and S355) in Duarte et al. [8] were considered.
Table 2 shows the measured [8] mechanical properties of the steel
(average values ± standard deviation), namely, (i) yield stress fy, (ii)
tensile stress fu, (iii) tensile strain eu and (iv) Young’s modulus Es,
determined by testing five coupons taken from the flat surfaces
of tubes with the three different steel grades. Additionally, in
Fig. 2 the measured [8] stress–strain curves of one (out of five) cou-
pon per steel grade are presented.
Regarding the modelling of steel, an elastic–plastic behaviour
with hardening, bi-linear stress–strain constitutive law was
adopted. In terms of steel elastic properties, the experimental
 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40 31

Table 2 Table 4
Measured mechanical properties of steel tubes [8]. Measured mechanical properties of the concrete mixes [8].

Steel Grade fy [MPa] fu [MPa] eu Es [GPa] Concrete mix fcm [MPa] Ecm [GPa] fctm [MPa] ecm [%] BI [–] m [–]
S235 284.0 ± 13.8 403.0 ± 11.0 0.33 ± 0.03 187.2 ± 16.6 NC 49.5 ± 3.7 37.6 ± 0.8 3.4 ± 0.1 0.20 1.31 0.20
S275 367.7 ± 8.8 463.1 ± 9.9 0.25 ± 0.02 RuC5 39.3 ± 4.5 33.4 ± 0.6 2.6 ± 0.0 0.42 1.11 0.21
S355 455.5 ± 10.6 570.9 ± 9.8 0.24 ± 0.01 RuC15 25.2 ± 2.4 26.5 ± 0.7 2.0 ± 0.1 0.54 0.75 0.23

800 60
NC
700 RuC5
50
600 RuC15

Compressive stress [MPa]

Stress [MPa]

500 40

400
30
300
S235
200 20
S275
100
S355 10 A1 A2
0
0 0.1 0.2 0.3 0.4
0
Strain 0 0.001 0.002 0.003 0.004 0.005 0.006
Strain
Fig. 2. Experimental stress–strain curves of steel grades [8].
Fig. 3. Compressive stress–strain curves of NC, RuC5 and RuC15 [8].

value of the Young’s modulus (Table 2) was adopted and the stan-
dard value ms = 0.3 was used for the Poisson’s ratio. The experimen-
tally measured stresses (rexp) and strains (eexp) (Table 2) were
converted into true stresses (rtrue) and logarithmic strains (eln)
according to [15],
rtrue ¼ rexp ð1 þ eexp Þ;
eln ¼ lnð1 þ eexp Þ:
Hence, the values of yield (fy) and ultimate (fu) stresses and ten-
sile strains (eu) adopted in the models were: (i) fy = 284.6 MPa,
fu = 536.0 MPa, and eu = 0.29 (steel S235), (ii) fy = 368.4 MPa,
fu = 578.9 MPa, and eu = 0.22 (steel S275), and (iii) fy = 456.4 MPa,
fu = 707.9 MPa, and eu = 0.22 (steel S355).
Regarding the concrete cores, the three experimentally studied
[8] concrete mixes were considered in this study, as shown in
Table 3: (i) normal concrete (NC) and two Rubberized Concrete
(RuC) mixes, (ii) RuC5 and (iii) RuC15.
The experimentally [8] determined concrete properties, namely
(i) the compressive strength (assessed with cubic specimens), fcm,
(ii) the Young’s modulus (assessed with cylindrical specimens),
Ecm, and (iii) the splitting tensile strength (assessed with cylindrical
specimens), fctm, are presented in Table 4. Additionally, one cubic
specimen of each concrete mix was monitored with a strain gauge
while tested in compression to assess the increased ductility of
RuCs (RuC5 and RuC15) compared to that of NC. As it can be
observed (Table 4), and in agreement with the conclusions drawn
by other authors [1–4], all the mechanical properties of RuC are
lower than those of NC, being the only exception the maximum
strain (ductility) of RuCs, which is higher than that of NC, as shown
in Fig. 3: ec.RuC5 = 2.1ec.NC and ec.RuC15 = 2.7ec.NC [8]. Moreover, to
confirm the increase in ductility of RuC mixes compared to NC,
the brittleness index (BI) of the three concrete mixes was also
ð1Þ assessed. The BI is defined as the ratio A2/A1, where A2 is the
recoverable (elastic) energy at fracture (see area A2 in Fig. 3) and
ð2Þ A1 is the irrecoverable (plastic or residual) energy (see area A1 in
Fig. 3). By definition, the smaller the BI value, the more ductile
the material is [3]. According to the measured BI values of RuC
and NC mixes presented in Table 3 it can be confirmed that RuC’s
ductility tends to increase with the increase of rubber content.
Modelling the concrete inside the steel tubes is always a chal-
lenging task due to the passive confinement effect provided by
the steel tube, delaying the propagation of cracking in the concrete
core, thus increasing its compressive strength, fcm, and strain, ecm.
Regarding the elastic properties of the concrete core (NC, RuC5
and RuC15), the measured values of the Young’s modulus, Ecm
(Table 4) [8], were adopted. For the Poisson’s ratio of NC was
assumed mNC = 0.20 [16], whereas for the RuCs an estimate was
obtained based on the ‘‘rule of mixtures” [17],
mRuC ¼ mNC V concrete þ mrubber V rubber ð3Þ
where mRuC is the Poisson’s ratio of RuC, Vconcrete and Vrubber are the
volumetric fraction of the concrete matrix and rubber particles in
the RuCs mixes (Table 3), respectively, and mrubber is the Poisson’s
ratio of the rubber particles (mrubber = 0.50) [14].

Table 3
NC, RuC5 and RuC15 mix proportions [8].

Sand (0.0–4.0 mm) Limestone aggregates Tyre rubber aggregates Water Cement
(4.0–22.4 mm) (4.0–11.2 mm)
Density [kg/m3] Content [kg/m3] Density [kg/m3] Content [kg/m3] Density [kg/m3] Content [kg/m3] Content [kg/m3] Content [kg/m3]
NC 2615.0 802.7 2655.0 1000.3 1150.0 0.0 185.0 350.0
RuC5 802.7 910.2 39.2 185.0
RuC15 797.8 727.1 117.1 189.0
32 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40

The plastic behaviour of concrete was modelled using the Con- 45

crete Damaged Plasticity (CDP) model in ABAQUS [15], which makes = 0.9fcm,cyl,NC
40

- compressive stress [MPa]

use of a non-associated plastic flow potential G, based on the
Drucker–Prager hyperbolic function, 35
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi = 0.9fcm,cyl,RuC5
2
30
G ¼ ðefctm tanwÞ þ q2  ptanw ð4Þ
25
where (i) e is the flow potential eccentricity of the hyperbolic func- 20 = 0.9fcm,cyl,RuC15
tion (taken herein as e = 0.1 by default), (ii) W is the dilation angle,
(iii) q is the von Mises equivalent effective stress and (iv) p is the 15
hydrostatic pressure. 10 NC
Besides the flow potential eccentricity e and the dilation angle RuC5
5
W, the CDP model considers two additional parameters, fb0/fc0 RuC15
(ratio of initial equibiaxial compressive yield stress to initial uniax- 0
ial compressive yield stress), and Kc (ratio of the second stress 0.000 0.005 0.010 0.015
invariant on the tensile meridian to that on the compressive - compressive strain
meridian). The default values of these two parameters, respectively
fb0/fc0 = 1.16 and Kc = 2/3 [15], were assumed for all concrete mixes Fig. 4. NC, RuC5 and RuC15 compressive stress–strain curves adopted in the
models of columns with circular sections.
(NC, RuC5 and RuC15), regardless of the cross-section shape
(circular, square and rectangular).
The dilation angle W controls the amount of plastic volumetric
45
strain developed during plastic shearing and is assumed constant NC
during plastic yielding [18,19]. Regarding the dilation angle W of

- compressive stress [MPa]

40
RuC5
the concrete mixes, different values were adopted, regardless of 35 RuC15
the cross-sections shape: WNC = 20o for NC, WRuC5 = 15o for RuC5,
30
and WRuC15 = 10o RuC15. The typical value of the NC dilation angle
adopted in the numerical modelling of CFST, W = 20° [10,11,20], 25
was also adopted in this work. However, the W value for RuCs 20 = 0.4fcm,cyl,NC
has been proposed to be lower than that of the NC. Youssf et al.
[21] concluded that the dilation angle W of RuCs is lower than that 15
of NC due to the ability of the rubber particles (given by their 10 = 0.4fcm,cyl,RuC5
hyperelasticity) to prevent crack propagation. Duarte et al. [14] = 0.4fcm,cyl,RuC15
5
corroborated both experimentally and numerically this conclusion
regarding a splitting tensile test of a RuC15 specimen. Additionally, 0
rubber particles present an incompressible behaviour [14] 0.000 0.005 0.010 0.015
(mrubber  0.5), which may also contribute to the lower value of - compressive strain
the dilation angle W of RuCs compared to that of NC. Hence, and
taking into account the previous assertions, the fitting of experi- Fig. 5. NC, RuC5 and RuC15 compressive stress–strain curves adopted in the
models of columns with square and rectangular sections.
mental and numerical curves resulted in the lower values of RuCs’
dilation angles W presented before.
e
Besides the definition of the aforementioned parameters, the g¼ ð6Þ
CDP model also requires the definition of both the concrete’s com- ec
pressive and tensile stress–strain curves through the specification
of tabular data (pairs of stresses-r, strains-e). This was performed
ec
k ¼ 1:05Ecm ð7Þ
using the appropriate commands1 in ABAQUS [15]. For the com- f cm;cyl
pressive stress–strain curves of the three concrete mixes (NC, RuC5 where (i) Ecm is the concrete’s Young’s modulus (Table 4), (ii) fcm,cyl
and RuC15), an unconfined stress–strain law was adopted compris- is the equivalent cylinder concrete compressive strength and (iii) ec
ing two parts: (i) a non-linear strain hardening (ascending) branch, is the strain for r = fcm,cyl, which depends on the concrete mix
until the concrete’s compressive strength, fcm, is reached, and (ii) a (ec = 0.0020 for NC, ec = 0.0042 for RuC5, ec = 0.0057 for RuC15)
strain softening (descending) branch, as presented in Figs. 4 and 5 (Fig. 3).
and described hereafter. It was decided to use an unconfined Additionally, the equivalent cylinder concrete compressive
stress–strain law since, in this way, the confinement effects in the strength fcm,cyl of the concrete mixes (NC, RuC5 and RuC15) was
concrete core are included through contact (pressure) between steel assessed by means of [23],
and concrete.  
Regardless of the cross-section shape (circular, square and rect- f
f cm;cyl ¼ 0:76 þ 0:2log 10 cm f cm ð8Þ
angular), the non-linear strain hardening branch of the concrete 19:6
was attained by taking into account the experimental [8] NC and
where fcm is the average value of concrete compressive strength,
RuCs properties (Table 4) and the expression for non-linear
experimentally determined using cubic specimens [8], presented
behaviour provided in Eurocode 2 part 1–1 [22], which read,
in Table 4.
kg  g2 In the strain softening branch of the concrete compressive
r ¼ f cm;cyl ð5Þ
1 þ ðk  2Þg stress–strain law, a first approach consisted of modelling the con-
crete behaviour, inside both circular and square/rectangular steel
tubes, with a linear descending stress–strain curve. Similarly to
the procedure proposed by other authors [10,11,20,24], the slope
1 ⁄
CONCRETE COMPRESSION HARDENING and ⁄
CONCRETE TENSION STIFFENING of the descending branch was defined by a reduction factor k3,
commands. applied to the concrete equivalent compressive strength fcm,cyl at
 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40 33

P - compressive load [kN]
Fig. 6. Comparison between concrete descending branch modelling options of the
S100x3_235_0 column.
800 two transversal displacements (along axes X, Y) and all three rota-
tions of the reference nodes of both the top and bottom rigid plates
700 were restrained (Fig. 1). The previous boundary conditions
600 intended to model the columns’ experimental loading conditions
which consisted of imposing an increasing displacement upwards
500 to the column’s bottom section while maintaining its vertical dis-
400 placement fixed in the top section by means of a reaction beam
fixed to a metallic frame [8].
300 Exp In the normal direction, the contact interface (i) between the
column end-sections and top (supporting) and bottom (load bear-
200 Num - concrete linear
ing) rigid plates, and (ii) between the steel tube walls and concrete
100 Num - concrete Desayi core, was defined as hard contact in ABAQUS [15]. This means that
and Krishnan the aforementioned surfaces may separate under tensile forces but
0 cannot penetrate each other. In the tangential direction, a friction
0 1 2 3 4 5 coefficient of 0.25 [10,11,20] was adopted in both situations (con-
u - axial shortening [mm] tact between column end-sections and rigid plates and between
steel tube and concrete core). Moreover, it should be mentioned
that the same friction coefficient value (0.25) was used for the

Table 5
a given strain value, herein taken as ed = 0.010. However, whereas Comparison between numerical (PNum) and experimental (PExp) ultimate strengths.
for circular columns this first modelling approach (Fig. 4) proved to
Geometry Concrete Column PNum PExp PNum/PExp
be accurate, for square and rectangular columns a rather modest
mix [kN] [kN] [8]
agreement was obtained between numerical and experimental
Circular NC C114x3_235_0 721.2 723.1 1.00
curves. In particular, Hu et al. [20] have shown that square and
C114x3_275_0 919.9 930.7 0.99
rectangular steel tubes with B/t > 29 (herein B/t > 32.8) provide C152x3_275_0 1310.4 1313.7 1.00
small confinement to the concrete core and, henceforth, a second C219x4_235_0 2348.9 2323.5 1.01
concrete modelling approach based on the unconfined stress– C219x4_355_0 3144.7 2887.7 1.09
strain curve proposed by Desayi and Krishnan [16] was adopted RuC5 C114x3_235_5 620.5 597.0 1.04
for square/rectangular columns (Fig. 5), i.e., different from that C114x3_275_5 813.9 826.3 0.98
adopted for their circular counterparts. Additionally, the different C152x3_275_5 1128.5 1107.8 1.02
C219x4_235_5 1980.5 2197.4 0.90
behaviour of concrete inside circular and square/rectangular steel C219x4_355_5 2737.8 2603.6 1.05
sections is a natural consequence of the difference between shell
RuC15 C114x3_235_15 497.6 483.7 1.03
buckling (circular tubes) and plate buckling (square and rectangu- C114x3_275_15 675.1 673.6 1.00
lar tubes), the former providing better confinement than the latter C152x3_275_15 898.0 939.7 0.96
[20]. Recently, Dai and Lam [12] have also proposed different mod- C219x4_235_15 1521.7 1814.7 0.84
elling approaches for concrete confined by circular and elliptical C219x4_355_15 2230.0 2244.3 0.99
steel tubes, with the latter displaying a concrete curve qualitatively
Square NC S100x3_235_0 684.3 724.1 0.95
similar to the one proposed herein for square/rectangular steel sec- S100x3_355_0 905.8 851.8 1.06
tions. This second modelling approach provided better agreement S150x3_235_0 1370.0 1367.5 1.00
between experimental and numerical curves for square and S220x4_235_0 2480.0 2520.5 0.98
rectangular columns, as illustrated in Fig. 6 for the S100x3_235_0 S220x4.5_355_0 3641.0 –* –*

column. RuC5 S100x3_235_5 607.5 692.1 0.88

Nevertheless, and taking into account the first modelling S100x3_355_5 830.9 805.7 1.03
S150x3_235_5 1183.3 1156.6 1.02
approach described, it should be mentioned that, for circular col- S220x4_235_5 2404.8 2261.6 1.06
umns, a reduction factor k3 = 0.9 computed at a strain ed = 0.010 S220x4.5_355_5 3292.2 2969.4 1.11
provided good agreement between numerical and experimental RuC15 S100x3_235_15 499.0 523.0 0.95
load–displacement curves, regardless of the tube slenderness S100x3_355_15 723.7 691.0 1.05
(D/t) and concrete mix (NC, RuC5 and RuC15), whereas for S150x3_235_15 930.5 824.4 1.13
square/rectangular columns the Desayi and Krishnan curve [16] S220x4_235_15 1835.3 1690.8 1.09
crosses ed = 0.010 at a lower value, k3 = 0.4. S220x4.5_355_15 2708.4 2393.3 1.13

Finally, as mentioned earlier, ABAQUS [15] also requires the Rectangular NC R120x2_235_0 425.9 349.7 1.22
definition of the tensile behaviour of concrete, in the CDP model, R120x3_275_0 611.6 553.6 1.10
to be simulated by a softening law using the ⁄CONCRETE TENSION R160x5_275_0 1170.4 1221.5 0.96
STIFFENING command [15]. To do so, the experimentally deter- R200x4_235_0 1336.5 1425.1 0.94

mined splitting tensile strengths, fctm (Table 4) of the concrete RuC5 R120x2_235_5 370.1 349.1 1.06
mixes were adopted for the onset of cracking, and the tensile stress R120x3_275_5 556.4 500.6 1.11
R160x5_275_5 1093.3 1133.8 0.96
was linearly varied to zero for a corresponding ultimate cracking
R200x4_235_5 1192.2 1263.6 0.94
displacement of 0.08 mm, i.e. a fully opened crack [15].
RuC15 R120x2_235_15 290.1 262.3 1.11
R120x3_275_15 476.3 484.5 0.98
2.3. Loading and boundary conditions R160x5_275_15 960.6 1054.1 0.91
R200x4_235_15 972.6 1157.2 0.84
Loading was applied to the columns by continuously imposing Average 1.01
Standard deviation 0.08
an upward longitudinal displacement (along Z axis) to the bottom
*
rigid plate (Fig. 1) while maintaining the upper rigid plate fully PExp not available because the load cell capacity (3000 kN) was exceeded in the
restrained in terms of this degree-of-freedom. Additionally, the experimental test [8].
34 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40

1000
C
900

P - compressive load [kN]

800 H D
700 C
Y H D
600
500
400 Y
300 C114x3_275_0_Exp
C114x3_275_0_Num
200 LM
C114x3_275_15_Exp
100
C114x3_275_15_Num
O
0
0 5 10 15 LM
(a) u - axial shortening [mm] (b)
Fig. 7. Numerical vs. experimental results of columns with circular sections: (a) P–u curves of C114x3_275_0 and C114x3_275_15, and (b) failure mode of C114x3_235_0.

800

700
P - compressive load [kN]

C
600

500 D
Y C
400 D
Y
300
S100x3_235_5_Exp
200 S100x3_235_5_Num

100 S100x3_235_15_Exp

O S100x3_235_15_Num LM LM
0
LM
0 1 2 3 4 5
(a) u - axial shortening (b)
Fig. 8. Numerical vs. experimental results of columns with square sections: (a) P–u curves of S100x3_235_5 and S100x3_235_15, and (b) failure mode of S100x3_235_15.

modelling of the tangential interaction between the steel tube and
Regarding the use of RuCs in replacement of NC, the numerical
the three concrete mixtures (NC, RuC5 and RuC15), as the influence results provide the same trend observed in the experimental
of this parameter on the numerical results was found to be tests [8]: the axial strength and stiffness of RuCFST (using
negligible (see Appendix A). RuC5 and RuC15) are lower than those of CFST (using NC) due
to the lower compressive strength (fcm,cube) and Young’s modu-
3. Numerical results and discussion lus (Ecm) of RuCs.

The numerical (PNum) and experimental (PExp) ultimate
In this section, the results of non-linear FE analyses of the short strengths show good agreement for the CFST and RuCFST
columns are presented (Table 5 and Figs. 7–9) and validated by columns studied – the average value of PNum/PExp is 1.01 and
comparison with the experimental results [8]: (i) collapse standard deviation value is 0.08 (Table 5).
strengths (PNum and PExp), (ii) compressive load (P) vs. axial
Columns with circular sections (Fig. 7(a)) display numerical P-u
shortening curves (u) and (iii) failure modes. curves with four different parts, similarly observed in experi-
Table 5 presents the ultimate strengths, PNum, obtained from the mental curves [8]. They comprise (i) an elastic path (O–Y), (ii)
non-linear analyses, and a comparison between numerical (PNum) an initial ascending non-linear branch (Y–H), in which the
and experimental (PExp) ultimate strengths, as well as the ratio non-linear behaviour of concrete develops and steel yielding
PNum/PExp. Fig. 7(a) (C114x3_275_0 and C114x3_275_15), Fig. 8(a) spreads, (iii) a final ascending non-linear branch (H–C), in which
(S100x3_235_5 and S100x3_235_15), and Fig. 9(a) the concrete core tends to expand further but the circular tube
(R200x4_235_0 and R200x4_235_5) illustrate the variation of delays it (concrete confinement explains the inflection point H),
compressive load (P) with axial shortening (u), i.e. P–u curves, and (iv) a descending path (C–D), in which concrete crushes and
obtained numerically and experimentally. Additionally, Fig. 7(b) a local mechanism (LM) develops in the steel tube, associated to
(C114x3_235_0), Fig. 8(b) (S100x3_235_15) and Fig. 9(b) local buckling and steel yielding (Fig. 7(b)).
(R200x4_235_5) depict the configuration of column failure modes
For columns with circular sections (Fig. 7(a)), and comparing
obtained from the numerical analyses and experimental tests. the NC curve (C114x3_275_0) with the RuC15 curve
From Table 5 and Fig. 7 (circular columns), Fig. 8 (square col- (C114x3_235_15), it is observed that the H–C–D branches of
umns), and Fig. 9 (rectangular columns), it may be concluded that: these curves shift from a well-defined peak to a horizontal

1600
1400 C
P - compressive load [kN]
1200
Y C
1000
Y
800
600
400
200
O
0
0 1
(a) u - axial shortening [mm]
Fig. 9. Numerical vs. experimental results of columns with rectangular sections: (a) P–u curves of R200x4_235_0 and R200x4_235_5, and (b) failure mode of R200x4_235_5.
plateau. In fact, the axial shortening at which the ultimate load
occurs (maximum point C) increases with the rubber content:
uC = 5.60 mm (C114x3_275_0) and uC = 10.6 mm (C114x3_
275_15). Hence, the ultimate axial shortening increases
5.0 mm when the core of C114x3_275 column shifts from NC
to RuC15. In particular, a higher shortening increase of about
7.8 mm was experimentally observed [8] for C114x3_235 col-
umns when replacing the NC core by the RuC15 one2. In conclu-
sion, it is undeniable that the increase of rubber content in the
concrete core increases the axial shortening at which the ultimate
load is reached and this is beneficial since it improves the column
ductility.

The P–u curves of columns with square and rectangular sections
(Figs. 8(a) and 9(a)) also present the same three distinct parts
observed in the experimental curves [8]: (i) an elastic path
(O–Y) up to the onset of steel yielding and initiation of concrete
crushing, (ii) a non-linear ascending branch (Y–C), in which
plasticity spreads due to concrete non-linear behaviour and
strain hardening develops in steel, and (iii) a nonlinear descend-
ing path (C–D), in which a local mechanism (LM) develops in the
tube, leading to crushing of the concrete core inside the tube, as
shown in Figs. 8(b) and 9(b).

It may also be concluded from the observation of P–u curves
that the numerical models exhibit an initial stiffness higher
than the tested columns. Similar differences in elastic stiffness
were also observed by other authors when comparing numeri-
cal and experimental results [11,20,25]. The main causes for
these differences may be attributed to (i) small gaps on the
experimental test rig, (ii) existence of initial imperfections and
residual stresses in the steel tubes and (iii) small eccentricity
in the loading and out of ‘‘straightness” of the columns, which
were not considered in the models. Additionally, the fact that
the stress–strain curve for NC proposed in Part 1–1 of Eurocode
2 [22] (Eqs. (5)–(7)) was adopted for modelling the behaviour of
RuCs in compression may also contribute for a further stiffness
overestimation of the RuCFST models. This is due to the fact that
crack initiation in RuC occurs for a lower percentage of its
uniaxial compressive strength (fcm) than that of NC. In the for-
mer, the onset of cracking occurs for compressive stresses circa
30% of fcm, due to the stress concentration near the rubber
particles, though their propagation is more gradual than that
2
The higher steel yield stress of C114x3_275 column (fy = 368 MPa), compared to
that of C114x3_235 column (fy = 284 MPa), might explain the lower shortening
increase observed (5.0 vs. 7.8 mm) – it is well known that higher strength of steel is
often associated to lower ductility [5].
A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40 35
D
D
LM
R200x4_235_0_Exp
R200x4_235_0_Num LM
R200x4_235_5_Exp
R200x4_235_5_Num
2 3 4 5
(b)
of NC, due to the ability of rubber particles to prevent crack
propagation [14,26]. Conversely, in NC cracking initiates for
compressive stresses of about 70% of fcm [16,27] and their prop-
agation is more abrupt, thus resulting in a more brittle failure
compared to that of RuC. Since cracking is responsible for the
stiffness degradation of concrete and considering that the
stress–strain constitutive law of NC was adopted for the mod-
elling of RuC, this may explain the overestimation of stiffness
of the latter which then resulted in an overestimation of the
global stiffness of the RuCFST modelled.

Regarding the ductility of square columns, it can be observed
that shifting from a NC core (S100x3_235_0 – Fig. 6) to a
RuC15 core (S100x3_235-15 – Fig. 8(a)) leads to an axial short-
ening increase from u = 1.0 mm to u = 2.1 mm. Hence, it can be
concluded that the increase of rubber content (from 0% to 15%)
in square columns leads to an increase of the ultimate shorten-
ing of 1.1 mm, which is lower than that observed for columns
with circular sections (5.0 mm). This trend was also observed
in the experimental results [8], in which an increase in ductility
of 150% was observed for columns with circular sections, while
for the case of columns with either square or rectangular sec-
tions the increase in ductility was of 40%.

Finally, it should be referred to that the increase of concrete
strength due to the confinement provided by the circular steel
tube3 to the concrete core decreases with the increase in rubber
content (see P–u curves in Fig. 7(a)): between points H and C, the
increase of load due to confinement effects is DPHC = 120 kN for
C114x3_275_0 but only DPHC = 55 kN for C114x3_275_15. This
is probably due to the lower value of the dilation angle W
adopted for the RuC15 (W = 10°), compared to that of NC
(W = 20°). This trend was also reported in [8], from the observa-
tion of C114x3_235 P-u curves obtained in tests, and indicated as
a possible explanation for Eurocode 4 part 1–1 [28], being more
unsafe in predicting the ultimate loads of RuCFST columns, as
the rubber content increases, due to the overestimation of the
concrete confinement effect.
Based on the good agreement observed between numerical and
experimental results, including (i) ultimate strengths (PNum and
PExp), (ii) load-shortening (P–u) curves and (iii) short column failure
modes, it can be concluded that the proposed FE models were
accurately validated.
3
This effect is almost absent in square and rectangular tubes.
36 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40

20 20
C1 C2 C1 C2
18 C3 C4 18 C3 C4

fl - confining pressure [MPa]

fl - confining pressure [MPa]
16 16
14 14
12 12
10 10
8 8
6 6
4 4
2 2
0 0
0 5 10 15 0 5 10 15
u - axial shortening [mm] u - axial shortening [mm]
(a) (b)
20 20
C1 C2 C1 C2
18 C3 C4 18 C3 C4
fl - confining pressure [MPa]

fl - confining pressure [MPa]

16 16
14 14
12 12
10 10
8 8
6 6
4 4
2 2
0 0
0 5 10 15 0 5 10 15
u - axial shortening [mm] u - axial shortening [mm]
(c) (d)
Fig. 10. Numerical confining pressure fl versus shortening u for circular sections: (a) C114x3_275_0, (b) C114x3_275_15 (c) C219x4_235_0 and (d) C219x4_235_15.

4. Numerical study on confining pressure
As stated earlier, modelling the concrete inside the steel tubes
was a challenging task due to its passive confinement and, addi-
tionally, the simulation of RuC was also complex because informa-
tion on RuC material models is very scarce. Hence, the objective of
the numerical study described in this section, based on the good
agreement between the numerical models and experimental
results, is to further analyse how the concrete core behaves. The
aim is to understand the influence of relevant parameters, such
as cross-section shape, cross-section slenderness (D/t, B/t), steel
yield stress and concrete mix properties, on the level of concrete
confinement. In particular, NC and RuC15 are compared, being
the two most different concrete mixes studied in this work. In
order to achieve this goal, the numerical lateral confining pressures
(fl) between the steel and concrete parts are first analysed in detail
for some columns modelled. Figs. 10 (circular sections), 11 (square
sections) and 12 (rectangular section) show the variation of
numerical lateral confining pressure (fl) between the steel tube
and the concrete core with the column’s axial shortening (u).
The data was organized according to the following procedure:

For the circular sections (Fig. 10), four points were chosen (C1–
C4, see Fig. 1(a)) and two geometries were selected:
C114x3_275 (the most compact one, with D/t = 35.6) and
C219x4_235 (the most slender one, with D/t = 56.9). For each
case, both NC and RuC15 were considered. This leads to four
columns with circular sections: C114x3_275_0 (Fig. 10(a)),
C114x3_275_15 (Fig. 10(b)), C219x4_235_0 (Fig. 10(c)), and
C219x4_235_15 (Fig. 10(d)).

For the square sections (Fig. 11), six points were chosen (S1–S6,
see Fig. 1(b)) and the most compact column (B/t = 32.8) was
selected, with both NC and RuC15. This leads to two columns:
S100x3_355_0 (Fig. 11(a)) and S100x3_355_15 (Fig. 11(b));

For the rectangular section (Fig. 12), six points were chosen
(R1–R6, see Fig. 1(c)) and the most slender column (B/t = 72.7)
was selected with NC. This leads to the single column
R120x2_235_0 (Fig. 12).4
Fig. 10 shows that, due to the symmetry of revolution of circular
sections, fl is similar in nodes C1 and C2, as well as in nodes C3 and
C4 – the fl vs. u curves are almost coincident. In fact, whereas in
columns with circular sections, fl is almost constant along the
cross-section contour, even though it varies along the column
height (see the difference between C1/C3 and C2/C4), in columns
with square and rectangular sections, fl is not constant along the
section contour and it only develops at the corners. As seen in
Figs. 11 and 12, corner points S2 and S5 (square columns –
Fig. 11) and R5 (rectangular columns – Fig. 12) are the only ones
that develop values of fl different from zero. Regarding circular col-
umns, the lower fl values registered in nodes C1 and C2, compared
to those in nodes C3 and C4, are a consequence of the failure mode
of circular columns, in which the local mechanism tends to develop
4
The fl–u curves of column R120x2_235_15 were not depicted in Fig. 12 because
the confining pressure fI was found to be null in every node.
 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40 37

100 100
S1 S2 S3 S1 S2 S3
90 S4 S5 S6 90 S4 S5 S6

fl - confining pressure [MPa]

fl - confining pressure [MPa]

80 80
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
0 1 2 3 0 1 2 3
u - axial shortening [mm] u - axial shortening [mm]

(a) (b)
Fig. 11. Numerical confining pressure fl versus shortening u for square sections: (a) S100x3_355_0 and (b) S100x3_355_15.

100 marginal for square/rectangular columns (Figs. 11 and 12). In par-

R1 R2 R3 ticular, Fig. 13 shows the compressive stress (Z-axis direction) pat-
90 R4 R5 R6
tern in the concrete cores of C114x3_275_0 and S100x3_235_0
fl -confining pressure [MPa]

80 columns at their mid-height section (Z = H/2), when maximum

70 load occurs. It is observed that, even though the adopted uncon-
60 fined concrete properties in the non-linear strain hardening branch
are the same for circular and square columns (Figs. 4 and 5), con-
50
crete encased by the circular section reaches higher compressive
40 stress (55.0 < r < 66.0 MPa) than that observed in the square sec-
30 tion (41.3 < r < 42.8 MPa). Moreover, the stresses developed in
the circular section are roughly 30% to 60% higher than the adopted
20
unconfined NC strength (fcm,cyl,NC = 41.6 MPa – see Eq. (8) and
10 Table 4), whereas the stresses developed in the square section
0 are closer to the fcm,cyl,NC value.
0 1 2 3 A second remark concerning Figs. 10–12 is that, regardless of
u - axial shortening [mm] the section shape (circular – Fig. 10, square – Fig. 11, and rectangu-
lar – Fig. 12), the use of a NC core results in higher confining pres-
Fig. 12. Numerical confining pressure fl versus shortening u for rectangular section
sures, fl, than those developed between a RuC15 core and the steel
R120x2_235_0.
tube. This is due to the lower value of the dilation angle W adopted
for the RuCs (WRuC5 = 15° and WRuC15 = 10°) compared to that of NC
closer to the end section (Z = H/10 – Fig. 7(b)) rather than at the (WNC = 20°). As a consequence of the lower dilation angle of RuC15
column mid-height (Z = H/2). compared to NC, the concrete hardening effect DPHC (evidenced in
The tri-axial stress state to which the concrete core is subjected Fig. 7(a)) is also lower in the former than in the latter. The differ-
inside circular tubes, where fl develops along the whole section ence in the concrete mixes (NC and RuC15) dilation angles, W,
contour (Fig. 10), results in confining effects (increasing both the can also be observed by comparing the concrete volumetric strain,
concrete strength and ductility) in circular columns, these being ev, of columns C114x3_275_0 and C114x3_275_15, presented in

C3 S6
S5

C4

S4

(a) (b)
Fig. 13. Patterns of concrete compressive stresses at maximum compressive load: (a) C114x3_275_0 and (b) S100x3_235_0.
38 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40
5 0.032
u - axial shortening [mm]
NC
RuC15
4 RuC15
2
1 0.008
0
-0.002 -0.001 0.000 0.001
v - volumetric strain
Fig. 14. Numerical volumetric strain ev versus shortening u of C114x3_275 column
with of NC and RuC15.
Fig. 14. Being defined by ev = ea + 2 eh, where ea is the axial strain
and eh is the hoop strain, the volumetric strain ev of the NC and
RuC15 cores inside C114x3_235 tubes is plotted against the axial
shortening (u) of the columns. Fig. 14 shows that both concrete
cores present negative volumetric strains (ev) for low values of
column shortening (u < 1 mm). This evidence means that initially
the volume of the concrete cores (NC and RuC) decreases: the
lower Poisson ratios of NC (mNC = 0.20) and RuC15 (mRuC15 = 0.23 –
Eq. (3)) in comparison to that of the steel (ms = 0.30) also explains
the initial absence of contact pressure, fl, observed in Figs. 10–12
for small deformation values (u < 1 mm to 2 mm). It is also noted
that NC initially compacts further (ev = 0.040%) than RuC15
(ev = 0.015%), as the former is characterized by a lower Poisson
ratio (mNC = 0.20) than the latter (mRuC15 = 0.23 – Eq. (3)) – recall that
a maximum value of m = 0.50 in elasticity leads to ev = 0. However,
for increasing axial shortening (u), NC and RuC15 cores start to
behave inelastically due to cracking and become dilatant. At this
stage, the dilation angle W definitely influences the lateral expan-
sion of the NC and RuC cores, the contact between the steel tube
and the concrete core is initiated, and contact pressures emerge
(see Figs. 10–12). In the inelastic phase, Fig. 14 shows that since
the adopted dilation angle of RuC15 (WRuC15 = 10°) is lower than
that of NC (WNC = 20o), for a given axial deformation level, NC nat-
urally presents a higher positive volumetric strain, ev, than RuC15.
Regarding the influence of the slenderness of circular tubes
(D/t) on fl, Fig. 10 shows that fl values associated to the slenderest
column (C219x4_235, D/t = 56.9) are about half of the fl values for
the most compact section (C114x3_275, D/t = 35.6), for the same
concrete core. It should be recalled that, regardless of the steel tube
slenderness (D/t), the adopted concrete compressive stress–strain
curve was always the same. It should be also noted, on the other
hand, that C114x3_275 columns have a higher steel yield stress
(fy = 368 MPa) than the C219x4_235 columns (fy = 284 MPa), which
also contributes for the developed fl values: steel tubes with lower
yield stress, fy, provide lower confinement to the concrete core
[10,11,20].
The influence of the circular tubes’ slenderness (D/t) and yield
stress (fy) on the numerical confining pressures (fl) developed in
the three concrete core mixes (NC, RuC5 and RuC15) is illustrated
in Fig. 15, which presents the variation of ‘‘confining pressure”-to-
‘‘steel yield stress” ratio (fl/fy) as a function of the circular tube
slenderness (D/t) for different rubber particle contents (NC, RuC5,
RuC15). Fig. 15 confirms that, besides local slenderness (D/t) and
steel yield stress (fy) [10,11,20], the confining pressure fl also
depends on the concrete core mix (NC, RuC5 and RuC15) itself.
Hence, for the same cross-section slenderness (D/t), it can be stated
NC
RuC5
0.024
fl/fy
0.016
0.002 0.000
20 30 40 50 60
D/t
Fig. 15. Influence of the cross-section slenderness, steel yield stress and concrete
mix on the numerical concrete confining pressure.
that the confinement provided by the steel tube (fl/fy) decreases
with the increase in rubber content (from NC to RuC15).
Unlike the two-part curves relating fl/fy with D/t proposed by
other authors [20,29], a single expression is proposed herein to
take these effects into account. Given the fact that the confining
pressures (fl) vary with (i) the cross-section slenderness of the steel
tubes (D/t), (ii) the tube’s yield stress (fy), and (iii) the concrete core
mix (NC, RuC5 and RuC15), a single expression was fitted to the
existing data by minimizing the sum of the square of the residuals.
The general expression to predict the concrete confining pressure,
i.e. fl(fy, D/t, p), reads
" #
2:95  p 17:1  100p
f l ¼ 1:5f y þ ð9Þ
750 ðD=tÞ2
where fl, fy, D and t have the same meaning as before and p is the
rubber particle content of the concrete mix (pNC = 0, pRuC5 = 0.05
and pRuC15 = 0.15). Fig. 15 presents three curves obtained from Eq.
(9) for these three p values (NC, RuC5 and RuC15). The coefficient
of determination values (R2 – see Fig. 15) allow concluding that
these curves seem to be appropriate for the estimation of fl in circu-
lar CFST and RuCFST columns. Nevertheless, this formula should be
further calibrated against a larger number of experimental and
numerical results.
5. Conclusions
A numerical study on the structural behaviour of short steel
tubes filled with Rubberized Concrete (RuCFST) and different
cross-section shapes (circular, square and rectangular), subjected
to monotonic concentric compression, was described. Firstly, a
brief introduction and short literature review on the topic were
presented. Then, non-linear finite element (FE) models of cold-
formed steel tubes displaying circular, square and rectangular sec-
tions and filled with three concrete mixes (NC, RuC5 and RuC15)
were characterized and then simulated within the framework of
commercial FE package ABAQUS [15]. The numerical models were
based on an experimental programme conducted by the authors
[8]. The main conclusions of this study are:

The adoption in the models of the measured elastic and plastic
steel and concrete properties provided good agreement
between numerical and experimental results. Besides the
capability of the Concrete Damaged Plasticity to model the
inelastic behaviour of standard concrete (NC), it was con-
firmed that this model can also be extended to the simulation
of RuC;
 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40

The models were validated through the comparison between the
numerical and experimental results, including the columns’ ulti-
mate strength, compressive load vs. axial shortening curve and
failure mode configuration. Lower strength and stiffness, but
higher ductility, of RuCFST, compared to CFST columns, were
numerically observed, in agreement with the experimental data;

Compressive stresses developed in the concrete core of columns
with circular sections were found to be up to 60% higher than
the adopted unconfined concrete strength in the case of the
stockiest circular column filled with standard concrete, due to
confinement effects. In the columns with square and rectangu-
lar sections, concrete compressive stresses were of the magni-
tude of the unconfined concrete compressive strength;

As stressed out by other authors, the dilation angle of RuC was
shown to be lower than that of NC: for 5% and 15% of rubber
particles content, the dilation angle is 75% and 50% of that of
the NC, respectively. The lower dilation angles of RuCs resulted
in lower confining pressures developed between the steel tube
and the concrete cores, which also contributed for a lower con-
crete hardening effect observed in RuCFST rather than CFST: the
compressive strength increase in columns filled with RuC15 due
to the concrete hardening effect was found out to be 54% lower
than that observed in columns filled with NC;

Confining pressures in columns with circular sections were
analysed in detail and confirmed to be dependent not only on
the tube slenderness and steel yield stress, but also on the con-
crete mix (i.e. rubber particle content). Approximated curves
correlating the aforementioned parameters were fitted to exist-
ing data. A good agreement was shown and new expressions to
predict the confining pressure in the studied circular RuCFST
columns were put forward.
Acknowledgements
The first author gratefully acknowledges the financial support
provided by FCT (Fundação para a Ciência e a Tecnologia, Portugal),
(a)
Fig. A1. Numerical models of column S100x3_235_15 and corresponding collapse modes: (a) RM and FM, (b) WM and (c) SM.
39
through doctoral scholarship SFRH/BD/97176/2013. All the authors
would like also to acknowledge FCT for the financial support
through the research project ‘‘Recycling & Seismic Protection:
Sustainable High-Performance CFST Columns for Seismic Areas”
(PTDC/ECM/117774/2010). The corresponding author also
acknowledges the financial support provided by FCT, through
IDMEC, under LAETA, project UID/EMS/50022/2013. All the authors
would like to thank Ferpinta (Eng. Bruno Marques), Biosafe,
Biogoma, Recipneu and SECIL for supplying the materials needed
for the development of the experimental programme.
Appendix A.
In this appendix, some basic modelling aspects discussed earlier
in the paper are analysed in more detail. For this purpose, four
models of column S100x3_235_15 were considered:

Reference model (RM), i.e. The model described in the paper
(Fig. A1(a)): one quarter of the whole column + steel tube with
solid (3D) finite elements + steel–concrete friction coefficient of
0.25.

Model without symmetry simplifications (WM), i.e. the model
with the whole column (Fig. A1(b)). The difference between
RM and WM is only the geometry.

Model with shell finite elements (SM) to discretize the steel
tube (Fig. A1(c)). The sole difference between RM and SM is
the type of finite element used for the tube.

Model with a higher friction coefficient (FM) (Fig. A1(a)). The
unique difference between RM and FM is the value of friction
coefficient between the steel tube and the concrete core: in this
case a value of 0.50 was adopted, twice that of RM.
Using these models, four numerical compressive load (P) vs.
axial shortening (u) curves (Fig. A2) are presented and compared.
The observation of Figs. A1 and A2 prompt the following remarks:
(b) (c)
40 A.P.C. Duarte et al. / Composite Structures 150 (2016) 28–40
600
500
P - compressive load [kN]
400
300
200
RM
WM
100 SM
FM
0
0 1 2 3 4 5
u - axial shortening [mm]
Fig. A2. Numerical compressive load (P) vs. axial shorting (u) curves of column
S100x3_235_15 model with different modelling options.

Both RM and WM led to indistinguishable P–u curves and
collapse modes. The analysis of WM (model with the whole
column) took twenty times the computational time of RM (only
one fourth of the column). Hence the adoption of one quarter of
the columns geometry with adequate boundary conditions is
the most efficient modelling approach.

Both RM and SM led to very similar P–u curves and collapse
modes. A maximum difference of 5% was found only in
the post-peak descending branch of P–u curves. Because the
computational effort of RM and SM is also similar, it can be
concluded that RM can be used without prejudice.

Both RM and FM led to equal P–u curves and collapse modes.
The adoption of different friction coefficients for the tangential
interaction between the steel tube and the concrete core has a
negligible impact in the results.
Regarding these basic modelling issues (symmetry simplifica-
tions, type of finite element, friction coefficient), we conclude that
our reference model (RM) can be used with confidence.
References
[1] Topçu IB. The properties of rubberized concretes. Cem Concr Res 1995;25
(2):304–10.
[2] Li Z, Li F, Li JSL. Properties of concrete incorporating rubber tyre particles. Magn
Concr Res 1998;50(4):297–304.
[3] Zheng L, Huo XS, Yuan Y. Experimental investigation on dynamic properties of
rubberized concrete. Constr Build Mater 2008;22(5):939–47.
[4] Valadares F, Bravo M, de Brito J. Concrete with used tire rubber aggregates:
mechanical performance. ACI Mater J 2012;109(3):283–92.
[5] Elremaily A, Azizinamini A. Behavior and strength of circular concrete-filled
tube columns. J Constr Steel Res 2002;58(12):1567–91.
[6] Han L-H, Yang Y-F, Tao Z. Concrete-filled thin-walled steel SHS and RHS beam-
columns subjected to cyclic loading. Thin-Walled Struct 2003;41(9):801–33.
[7] Liang QQ. Strength and ductility of high strength concrete-filled steel tubular
beam-columns. J Constr Steel Res 2009;65(3):687–98.
[8] Duarte APC, Silva BA, Silvestre N, de Brito J, Júlio E, Castro JMC. Tests and
design of short steel tubes filled with rubberized concrete. Eng Struct
2016;112:274–86.
[9] Silva A, Jiang Y, Castro JM, Silvestre N, Monteiro R. Experimental assessment of
the flexural behaviour of circular rubberized concrete-filled steel tubes. J
Constr Steel Res 2016. http://dx.doi.org/10.1016/j.jcsr.2016.04.016. accepted
for publication.
[10] Ellobody E, Young B, Lam D. Behaviour of normal and high strength concrete-
filled compact steel tube circular stub columns. J Constr Steel Res 2006;62
(7):706–15.
[11] Ellobody E, Young B. Nonlinear analysis of concrete-filled steel SHS and RHS
columns. Thin-Walled Struct 2006;44(8):919–30.
[12] Dai X, Lam D. Numerical modelling of the axial compressive behaviour of short
concrete-filled elliptical steel columns. J Constr Steel Res 2010;66(7):931–42.
[13] Zhongqiu F, Ji B, Hirofumi M, Eizien A, Chen JS. Flexural behaviour of
lightweight aggregate concrete filled steel tube. Adv Steel Constr 2014;10
(4):385–403.
[14] Duarte APC, Silva BA, Silvestre N, de Brito J, Júlio E. Mechanical
characterization of rubberized concrete using an image-processing/XFEM
coupled procedure. Compos B Eng 2015;78:214–26.
[15] Simulia. ABAQUS Standard, User’s Manual, Version 6.10, Rhode Island, USA;
2010.
[16] Neville AM. Properties of concrete. 5th ed. New Jersey, USA: Prentice Hall;
2012.
[17] Silvestre N, Faria B, Lopes JNC. Compressive behavior of CNT-reinforced
aluminum composites using molecular dynamics. Compos Sci Technol
2014;90:16–24.
[18] Vermeer PA, de Borst R. Non-associated plasticity for soils, concrete and rock.
Heron 1984;29(3):3–64.
[19] Kwasniewski M, Rodríguez-Oitabén P. Study on the dilatancy angle of rocks in
the pre-failure domain. In: 12th ISRM international congress on rock
mechanics, Harmonising Rock Engineering and the Environment, Beijing,
China; 2011.
[20] Hu HT, Huang CS, Wu MH, Wu YM. Nonlinear analysis of axially loaded
concrete-filled tube columns with confinement effect. J Struct Eng, ASCE
2003;129(10):1322–9.
[21] Youssf O, ElGawady MA, Mills JE, Ma X. An experimental investigation of
crumb rubber concrete confined by fibre reinforced polymer tubes. Constr
Build Mater 2014;53:522–32.
[22] CEN. EN 1992-1-1, Eurocode 2: design of concrete structures – Part 1–1:
general rules and rules for buildings. Brussels, Belgium: European Committee
for Standardization; 2010.
[23] Lacroix EA. Comparative study of strength design methods for rectangular
reinforced concrete and composite steel-concrete columns [MSc Thesis in Civil
Engineering]. Canada: University of Manitoba; 1997.
[24] Mander JB, Priestley MJM, Park R. Theoretical stress–strain model for confined
concrete. J Struct Eng, ASCE 1988;114(8):1804–26.
[25] Natário P, Silvestre N, Camotim D. Web crippling failure using quasi-static FE
models. Thin-Walled Struct 2014;84:34–49.
[26] Khaloo AR, Dehestani M, Rahmatabadi P. Mechanical properties of concrete
containing a high volume of tire–rubber particles. Waste Manage 2008;28
(12):2472–82.
[27] Castro-Montero A, Jia Z, Shah SP. Evaluation of damage in Brazilian test using
holographic interferometry. ACI Mater J 1995;92(3):268–75.
[28] CEN. EN 1994-1-1, Eurocode 4: design of composite steel and concrete
structures – Part 1–1: general rules and rules for buildings. Brussels,
Belgium: European Committee for Standardization; 2004.
[29] Liang QQ, Fragomeni S. Nonlinear analysis of circular concrete-filled steel
tubular short columns under axial loading. J Constr Steel Res 2009;65
(12):2186–96.