Advanced Steel Construction Vol. 11, No. 2, pp.

165-184(2015) 165

DESIGN STRENGTH OF CONCRETE-FILLED STEEL COLUMNS

W.-H. Kang1,*, B. Uy2, Z. Tao1 and S. Hicks3

1
Institute for Infrastructure Engineering, University of Western Sydney, Penrith, NSW 2751, Australia
2
Centre for Infrastructure Engineering & Safety, The University of New South Wales, Sydney, NSW 2052, Australia
3
Heavy Engineering Research Association, Manukau, Auckland City, New Zealand

(Corresponding author: E-mail: w.kang@uws.edu.au)

Received: 18 March 2013; Revised: 17 June 2014; Accepted: 24 June 2014

ABSTRACT: The purpose of this paper is to recalibrate the capacity reduction factors, estimate the reliability of
current equations, and investigate the effect of these factors in AS 5100.6, the Australian Bridge Standard for
concrete-filled steel tubular columns. This work has important ramifications for other international codes of practice
as the Australian code has the identical or similar underlying design philosophy with Eurocode 4, AISC and the code
of practice in Hong Kong. The method developed by Johnson and Huang is extensively applied to the Australian
code format to recalibrate the capacity factors in AS 5100 for a target reliability of β = 3.04 based on an extensive
database of 1,583 test results covering a wide range of input parameter values. In addition, an inverse analysis
procedure based on Johnson and Huang’s method is proposed to estimate the reliability of design equations with
known capacity factors. The analysis results show that the interaction between the concrete and steel needs to be
considered for the current capacity factors in AS 5100. The results also show that the current capacity factors
provide greater reliability than the target reliability suggested in AS 5104:2005/ISO 2394:1998, but after considering
the additional uncertainties created due to the application of multiple capacity factors, the reliability was almost the
same as the recommended value. In conclusion, the current capacity factor values in AS 5100 are adequate with
regards to safety and can be maintained, but better optimised values would be preferable to improve the cost-safety
balance.
Keywords: Capacity factors, Composite structures, Concrete-filled steel columns, Design strength, Safety factor
calibration

1. INTRODUCTION

Concrete-filled steel tubular (CFST) members have been used in many structural applications
including high-rise buildings, bridges, and offshore structures, due to their outstanding mechanical
performance: such as high axial load capacity, good ductility performance, large energy absorption
capacity, and low strength degradation [1],[2]. In the structural design of these composite members,
it is desirable to achieve the most optimised balance between the cost and the safety of the
members in the context of the safety of the whole structure. The safety factors regarding the
ultimate structural member resistance in the existing design code provisions need to be promptly
checked as more experimental data becomes available and improved statistical techniques are
developed. In addition, these factors should be determined based on the statistical rationale
provided by a reliability analysis. In this context, this paper aims to recalibrate the capacity factors
for the design of CFST members, to estimate the reliability of the current bridge design Standard in
AS 5100 [3] and to investigate the optimal cost-safety balance of the current design codes. The
proposed work in this paper has important ramifications for other international codes of practice as
the Australian code has the identical or similar underlying design philosophy with Eurocode 4 (EC
4) [4], AISC [5], and the code of practice in Hong Kong [6].
166 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

In this paper, first a brief summary of the current code provisions of AS 5100: Australian Standard
for Bridges for CFST columns is provided, and the database employed for the reliability based
capacity recalibration analysis is introduced. An extensive database of failure tests of CFST
members developed by Tao et al. [7] is used, which merged two available test databases established
by Goode [8] and Wu [9]. Although this database includes the test results of six structural CFST
member types (circular and rectangular columns, beam-columns, and beams) with a total of 2,194
test results, we herein limit our attention to CFST columns and use a total of 1,583 test results.
Although more experimental studies have recently been carried on the axial load bearing capacity
of CFST columns such as Gupta et. al. [10], Lue et al. [11], Tao et al. [12], Yu et al. [13], Yang and
Han [14], and Chang et al. [15], they are not considered in this study as the number of test results in
the Tao et al.’s database is already statistically sufficient. Based on this information, we conduct the
following three types of reliability-based analyses: firstly, we recalibrate the capacity factors in AS
5100 using the statistical method proposed by Johnson and Huang [16]. This method is selected
because: it can calibrate multiple factors at once; it assumes a lognormal distribution which is
realistic for member strength with non-negative values; and it utilises the data only for member
resistance, neglecting the load effect and its uncertainties. Secondly, we estimate the reliability of
the design equations in AS 5100 with the existing specified capacity factors (0.9 for steel and 0.6
for concrete), by proposing a calculation procedure that is inverse of the method proposed by
Johnson and Huang. Thirdly, the reliability estimation for the design codes and the capacity factors
in AS 5100 is repeated, considering the effect of the applied capacity factors on the accuracy of the
equations. Based on these calibration and reliability estimation results, discussions and
recommendations are provided regarding useful in-sights into and supporting information for the
current design code provisions.

2. AUSTRALIAN CODE PROVISIONS FOR CFST MEMBERS

2.1 Ultimate Section Capacity

In the Australian bridge standard, AS 5100 [3], the ultimate section capacity (Nus) for a rectangular
CFST stub column under axial compression is calculated as follows:
Nus   As f y  c Ac f c' (1)

where As and Ac = the areas of the steel and concrete sections, respectively; fy = the nominal yield
strength of the steel; f’c = the characteristic compressive strength of the concrete; and ϕ and ϕc = the
capacity reduction factors for steel and concrete, respectively (with the existing values given as 0.9
and 0.6). In this equation, the contribution of the reinforcement is not considered, because no test
data in this study included this variable.

Eq. 1 can also be used for a circular CFST stub column. However, if the relative slenderness (λr) is
not greater than 0.5 and the eccentricity of loading under the greatest design bending moment is not
greater than do/10 (where do is the diameter of a circular section), the additional increase in
concrete strength due to the confinement effect should be considered. When the confinement effect
is considered, the ultimate section capacity is as follows:
  tf 
Nus   As2 f y  c Ac f c' 1  1 y'  (2)
 do fc 
where η1 and η2 = the coefficients for reflecting the confinement effect. η1 represents the concrete
strength increase, and η2 represents the steel strength reduction, due to the confinement effect. The
calculation procedure for these coefficients is given in Clause 10.6.2.2 of AS 5100. 6 [3].
 Design Strength of Concrete-Filled Steel Columns 167

2.2 Ultimate Member Capacity

To estimate the ultimate member capacity (Nuc), a slenderness effect represented by a slenderness
reduction factor,  c , should be considered. For a stub column defined by Le/do or Le/b ≤ 4 (where
Le = the effective length of a column, and b = the section width of a rectangular tube), this
slenderness effect can be ignored. Considering this slenderness reduction factor, the ultimate
member capacity of a CFST column is calculated as follows:
Nuc   c Nus  Nus (3)

where Nus is calculated either from Eq. 1 or 2. The slenderness reduction factor,  c , is calculated
using the following equation:
  90  
2

c   1  1    (4)

    

where ξ is the compression member factor and λ is a factor defined as a function of the relative
slenderness (λr). λr is defined by N us / N cr , where Nus is calculated from either Eq. (1) or (2) but
the capacity factors ϕ and ϕc are taken as unity; Ncr is calculated as follows:
 2  EI e
N cr  (5)
L2e

 EI e   EI s  c Ec I c (6)

where Is and Ic are the second moment of the areas of the steel sections and uncracked concrete
sections, respectively, and ϕ and ϕc are the capacity factors.

In this paper, we aim to calibrate the capacity reduction factors ϕ and ϕc in Eqs. 1 and 2 and
investigate their effect on the reliability of the design equations, based on the extensive database
developed by Tao et al. [7], introduced in the following section.

3. DATABASE FOR CFST STUB COLUMNS

Safety factors in the existing design code provisions may not provide the most optimal balance
between cost and safety if they were developed based only on a limited number of test results. In
this case, the factors need to be updated based on the latest collection from a database with a
sufficient number of test results covering a wide range of input parameter values. For this purpose,
we use the extensive database developed by Tao et al. [7], which covers the test results over the last
few decades by merging two available test databases established by Goode [8] and Wu [9]. Goode’s
database contains 1,792 test results from 92 references, and Wu’s database contains 1,514 test
results from 104 references. In Tao et al.’s database, 1,575 test results from Goode’s database were
included (918 for circular members and 657 for rectangular members), discarding irrelevant
datasets, and 619 test results from Wu’s database were included that had not been included in
Goode’s database. A total of 2,194 test results (1,232 for circular members and 962 for rectangular
members) from 130 references are included in Tao et al.’s database. In this paper, we limit our
attention to column members and use a reduced database with 1,583 test results (445 for
rectangular stub columns, 234 for long rectangular columns, 484 for circular stub columns, and 420
for circular long columns). Here, a stub column is defined as a short column member with Le/do ≤ 4
(for circular members) or Le/b ≤ 4 (for rectangular members), where Le is the effective length of a
168 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

column, do is the diameter of a circular section, and b is the section width of a rectangular section.
For a stub column, the slenderness effect is not considered.

In Tao et al.’s database, some references do not provide the mean measured compressive strength of
concrete (fcm) values. They report the compressive strength values of 150 mm cubes (fcu) instead. In
this case, the conversion table given by Yu et al. [17] representing the approximate relationship
between cylinder strength (fcm) and cube strength (fcu) was used to obtain the equivalent
compressive strength. This table was developed based on Chen et al.’s work [18], which determined
the equivalent compressive strength of high-strength concrete.

Figure 1 shows an overview of the distributions of the key parameters in Tao et al.’s [7] database:
the diameters (do) or the longer side lengths (B) of the steel tubes, the tube thicknesses (t), the mean
measured yield strength of the steel (fym), the mean measured compressive strength of the concrete
(fcm), the column lengths for long columns, and the width-to-thickness ratio or
diameter-to-thickness ratio with the effect of yield strength ((B/t)/√(250/fy)) or ((D/t)/(250/fy)). In
this figure, the parameter ranges for the four section/member types (rectangular stub columns,
circular stub columns, long rectangular columns, and long circular columns) are separately
presented in Figures 1 (a), (b), (c), and (d).

In Figure 1 (a), for rectangular stub columns, most members have longer side lengths between 100
and 200 mm, while some members have values outside of this range. Note that the longer to shorter
side length ratio (B/Bshort) is between 1.00 and 2.00. The thickness is mostly distributed between
1.47 and 8 mm, while some members have values outside of this range. The mean measured yield
strength of the steel (fym) is uniformly distributed between 192 and 400 MPa (N/mm2), while a few
members’ yield strengths is 835 MPa (N/mm2). The mean measured compressive strength of the
concrete (fcm) is almost uniformly distributed between 12 and 103 MPa (N/mm2). The
width-to-thickness ratio with the effect of yield strength ((B/t)/√(250/fy)) is mostly distributed
between 12 and 100, while some members’ values are outside of this range.

For circular stub columns, a similar distribution is observed in Figure 1 (b). Most members have
diameters between 100 and 200 mm, while several members have diameters up to 1,020 mm. The
thickness is mostly distributed between 0.52 and 8 mm, while some members have values outside
of this range. The mean measured yield strength of the steel (fym) is uniformly distributed between
186 and 400 MPa (N/mm2), while a few members have a yield strength of 853 MPa (N/mm2). The
mean measured compressive strength of the concrete (fcm) is uniformly distributed between 10 and
110 MPa (N/mm2). The diameter-to-thickness ratio with the effect of yield strength ((D/t)/(250/fy))
is mostly distributed between 12 and 150, but still many members’ have their values between 150
and 223.

For long rectangular columns, as shown in Figure 1 (c), most members have longer side lengths
between 100 and 200 mm. Note that the longer to shorter side length ratio (B/Bshort) is between 1.00
and 2.02. The thickness ranges between 1.5 and 6 mm, while some members have values outside of
this range. The mean measured yield strength of the steel (fym) is uniformly distributed between 217
and 500 MPa (N/mm2), while a few members’ yield strength is 550 MPa (N/mm2). The mean
measured compressive strength of the concrete (fcm) is almost uniformly distributed between 10 and
94 MPa (N/mm2). The length of the columns is also uniformly distributed between 482 and 4,000
mm, while some members’ column lengths are up to 4,494 mm. The width-to-thickness ratio with
the effect of yield strength ((B/t)/√(250/fy)) is mostly distributed between 18 and 100, and only few
members’ values are outside of this range.
 Design Strength of Concrete-Filled Steel Columns 169

B (mm): 60 100 200 300 400

# of Tests 19 334 73 19

t (mm): 1.47 2 4 6 8 11.89

# of Tests 23 122 215 51 34

fym (MPa): 192 300 400 500 600 700 835

# of Tests 142 202 36 24 13 28

fcm (MPa): 12 20 30 40 50 6070 80 90 103

# of Tests 17 75 168 69 45 11 30 15 15

(B/t)/√(250/fy): 12 50 100 175

# of Tests 292 133 20

(a) Rectangular stub columns

D (mm): 48 100 200 300 400 1020

# of Tests 30 371 44 26 13

t (mm): 0.52 2 4 6 8 10 13.3

# of Tests 75 202 144 39 16 8

fym (MPa): 186 300 400 500 600 853

# of Tests 150 233 65 19 17

fcm (MPa): 10 20 30 40 50 60 70 80 110

# of Tests 130 86 77 23 52 54 30 32

(D/t)/(250/fy): 12 50 100 150 223

# of Tests 261 148 42 33

(b) Circular stub columns

B (mm): 60 100 200 360
# of Tests 15 201 18

t (mm): 1.5 2 4 6 10
# of Tests 7 105 99 23

fym (MPa): 217 300 400 500 550

# of Tests 85 99 33 17

fcm (MPa): 10 20 30 40 50 60 70 94
# of Tests 26 44 55 37 32 19 21

L (mm): 482 1000 2000 3000 4000 4494

# of Tests 75 66 49 35 9

(B/t)/√(250/fy): 18 50 100 158

# of Tests 133 97 4

(c) Long rectangular columns

D (mm): 25 100 200 500

# of Tests 69 291 60

t (mm): 0.95 2 4 6 8 12.8

# of Tests 49 122 180 43 26

fym (MPa): 178 300 400 500 682

# of Tests 116 220 64 20

fcm (MPa): 10 30 40 50 60 96
# of Tests 191 119 63 31 16

L (mm): 450 1000 2000 3000 4000 5560

# of Tests 116 165 97 23 19
(D/t)/(250/fy): 8 50 100 180
# of Tests 314 95 11

(d) Long circular columns

Figure 1. Distribution of Parameter Values in Tao et al.’s Database [7]

170 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

For long circular columns, as shown in Figure 1 (d), most members have longer side lengths
between 100 and 200 mm, and the thickness ranging between 0.95 and 8 mm, while some members
have values outside of this range. The mean measured yield strength of the steel (fym) is uniformly
distributed between 178 and 500 MPa (N/mm2), while a few members’ yield strength is 682 MPa
(N/mm2). The mean measured compressive strength of the concrete (fcm) is almost uniformly
distributed between 10 and 96 MPa (N/mm2). The length of the columns is also uniformly
distributed between 450 and 4,000 mm, while some members’ column length is up to 5,560 mm.
The diameter-to-thickness ratio with the effect of yield strength ((D/t)/(250/fy)) is mostly distributed
between 8 and 100, and some values are outside of this range.

Because the experiments were conducted for research purposes, some cases may not represent the
realistic conditions of existing structures. However, the comprehensiveness of the database enables
us to cover a wide range of parameter values. The database lacks uniformity regarding some of the
key parameters because it was randomly assembled from numerous test results in the literature.
This should be carefully considered when the capacity factor calibration and reliability analysis
results are used.

4. CAPACITY FACTOR CALIBRATION METHOD

Based on the extensive database introduced in the previous section, we first adopt the statistical
method proposed by Johnson and Huang [16] to calibrate the capacity factors in the resistance
functions of CFST stub column members in AS 5100 [3]. Although this method was originally
developed for the calibration of the partial safety factors of short and slender concrete-encased
composite columns under combined axial compression and uni-axial bending, this method can be
extensively applied to the calibration of the partial safety factors or capacity factors of any
resistance function with multiple factors. In this study, this method is selected for the following
three reasons: 1) This method can calibrate multiple capacity factors at the same time, which is
required for the calibration of the two capacity factors for steel and concrete in CFST columns. 2)
This method assumes that the CFST columns’ strength follows a lognormal distribution with the
lower limit at zero, instead of a normal distribution, and this corresponds to reality [19]. 3) This
method considers the resistance of the CFST columns and neglects the load effect by using the First
Order Reliability Method (FORM) sensitivity factor for resistance. Therefore, we can fully utilise
the database introduced in the previous section, which does not have information about loading
uncertainties. In this section, a summary of Johnson and Huang’s method is provided. It includes a
modification to allow its application to the Australian design code format, which has capacity
factors  instead of partial safety factors M (where  = 1/M).

4.1 Capacity Factor Calibration Method

Let gR(x) be a resistance function in an existing code provision, where x is a vector of the input
parameters. In this study, gR(x) = the resistance functions in Eqs. 1 and 2, and x = {fy, f’c, do, B, L,
…}, where the values of fy and f’c are taken as those of fym and fcm (mean-measured values).

The first step in capacity factor calibration is to make the resistance function unbiased. All the
capacity factors inside the function are taken to be unity, and the intrinsic bias is corrected using the
following bias correction term:
1 N  rei 
b  
N i 1  rti 
(7)
 Design Strength of Concrete-Filled Steel Columns 171

where N = the number of test results in the database, e.g., in this study, N = 445, 234, 484, and 420
for rectangular stub columns, long rectangular columns, circular stub columns, and long circular
columns, respectively; rei = the experimental resistance observed from the i-th test result; and rti =
the theoretical resistance calculated from gR(xi), where xi is a set of mean-measured input
parameters used in the i-th test. The intrinsic bias exists because the resistance function was not
calibrated based on an extensive database. By multiplying this bias correction term with the
resistance function, we can predict the unbiased resistance of gR(x) as follows:
r  bg R (x) (8)
where r = a predicted resistance for a given set of x after bias correction and  is the error of the
unbiased resistance function. The error term for each test result, i , is estimated as follows:
rei
i  (9)
brti
where rei = the experimental resistance observed from the i-th test result and rti = the theoretical
resistance obtained from gR(xi) using the mean measured input parameters. This error term
represents the modelling uncertainty of the unbiased prediction model from all possible sources for
prediction errors such as oversimplification in the design equations and the neglect of the size
effect.

The variance of r in Eq. 8 is obtained from two sources: 1) the variance of the error term (δ) in Eq.
8, and 2) the variance of the resistance function gR(x) based on the uncertainties in the input
parameters x. Assuming that r in Eq. 8 follows a lognormal distribution, the coefficient of variation
of r in Eq. 8 is estimated as follows:

Vr  V 
2
 Vrt 2  (10)

where Vδ = the coefficient of variation of the error term (δ); and Vrt = the coefficient of variation of
gR(x) due to the uncertainties in the input parameters x. Vδ can be statistically estimated from δi
(i=1,…,N) in Eq. 9, and Vrt can be estimated for each test result using a Monte Carlo simulation or
the first-order approximation of moments [20]. To calculate Vrt, we assume that the mean values of
the input parameters x are taken as the mean measured values in each test, and the coefficient of
variation of the input parameters is obtained from Table 1. Note that the coefficient of variation
(c.o.v.) of f’c is not directly provided by AS 3600/ NZS 3104 [22],[23], but the values of fcm
according to those of f’c are provided. Therefore, the c.o.v. of f’c was inferred from the definition
that f’c is the lower 5% fractile of fcm. The value for fcm was chosen as the average of the fcm values
for the entire test dataset.

Table 1. Coefficient of Variation (c.o.v.) of Basic Variables

Variables c.o.v. References
Yield strength of steel (fy) 0.07 [21]
Compressive strength of concrete (f’c) 0.10 [22],[23]
All linear dimensions (do, b, and L) 0.01 [3],[24]
Tube thickness (t) 0.10 [24]

Because we assume that r in Eq. 8 follows a lognormal distribution, the standard deviation of lnr
(σlnr) is calculated as follows:

ln r  ln 1  Vr 2  (11)
172 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

This standard deviation is used to calculate the target design value of the resistance (rd) for a target
reliability index β as follows:
rd  bg R (x) exp  k ln r  0.5ln2 r  (12)

where

k
k Vd 
2
  Vrt 2 
(13)
Vr 2
and kd = the fractile factor corresponding to β at the 75% confidence level, determined for a number
of test data N from a non-central t-distribution. Note that the target reliability index β considers
only the resistance, and ignores the effect of loads. In this case, AS 5104: 2005/ISO 2394:1998
[26],[27] recommends that β be empirically estimated as R × βt, where βt = the target reliability
considering both resistance and load effects and R = the First Order Reliability Method (FORM)
sensitivity factor when resistance is taken as 0.8. Therefore, the design value for resistance
corresponds to the product R = 0.8 × 3.8 = 3.04 (equivalent to the probability of the actual
resistance falling below the design resistance of 1 in 845 = 0.0012).

4.2 Determination of Capacity Factors

This section provides the calculation steps of the statistical method proposed by Johnson and
Huang [16] to estimate the multiple capacity factors for steel (ϕ) and concrete (ϕc). We assume that
these capacity factors are defined by the product of a material independent model factor (ϕm), and
material dependent factors ϕms for steel and ϕmc for concrete. In other words, ϕ = ϕm × ϕms for steel,
and ϕc = ϕm × ϕmc for concrete.

First, we calculate the material dependent factors (ϕms and ϕmc) by using the ratio between the
characteristic strength and the design strength of each material. For example, the characteristic
strength of structural steel is defined as follows:


f yk  f y exp 1.64 ln f y  0.5 ln2 f y  (14)

where

 ln f  ln 1  V f2
y
 y
 (15)

and V f y = 0.07 from Table 1. The design strength of steel for the target reliability index β is defined
as follows:


f yd  f y exp    ln f y  0.5  ln2 f y  (16)

Then, the material dependent factor for steel, ϕms, is obtained from the ratio of Eqs. 15 and 16 as
follows:
f yd 1
ms   (17)
f yk exp    1.64   ln f y 

The calculations in Eqs. 14-17 are repeated for concrete to calculate the material dependent factor
ϕmc, by replacing V f y with V fc = 0.015, as suggested in Table 1.
 Design Strength of Concrete-Filled Steel Columns 173

Next, we calculate the material independent model factor (ϕm) for the i-th test result as follows:
rdi  g R (i , xki , xri ) (18)

where rdi = the target design value obtained in Eq. 12 for each test result; ϴi = the capacity
reduction factors ϕ and ϕc for the i-th test; x ki = the characteristic strength of the materials (fy and
fc) for the i-th test; and x ri = the input parameters, except x ki , for the i-th test. In this equation,
for the i-th test, the only unknown term is the material independent factor ϕmi inside ϴi. This term
can easily be obtained using any numerical solver for non-linear programming. In this study, the
Active-Set Optimisation algorithm [29] is used. We repeat this calculation for the entire test dataset
and find the material dependent factor, ϕm by averaging ϕmi as follows:
1 N
m  mi
N i1
(19)

Finally, the capacity factors are calculated as follows:

  m  ms (for steel) and c  m  mc (for concrete) (20)

5. INVERSE ANALYSIS METHOD FOR THE RELIABILITY ESTIMATION

OF EXISTING DESIGN EQUATIONS

In this section, an approach that is the inverse of the Johnson and Huang’s method [16] for
estimating the reliability of existing design codes for given capacity factors is proposed. It can be
used when the capacity factors are known but the corresponding reliability index is unknown. This
inverse approach is useful in quantifying the conservatism of the design equations with given
capacity factors in terms of reliability.

This approach is based on inverting Johnson and Huang’s method, which was introduced in the
previous section. Let g R (, x k , x r ) be a resistance function, where xk = the characteristic
strength of the materials; xr = the input parameters, except xk ; and ϴ = the capacity factors which
are given. To calculate the reliability index of an equation with known capacity factors, the
following equation can be derived by combining Eqs. 12 and 18 for each test datum:
bgR (xi )exp  k ln r  0.5 ln2 r   gR (xki , xri , ) (21)

where all the terms on the left-hand side ( b , g R ( x i ) , k, and  ln r ) can be calculated using Eqs.
7-13, except the target reliability index β in Eq. 13, which is also included in kd in Eq. 13. This is
the only unknown term in this equation, and it is calculated as βi for each test result. The right-hand
side of Eq. 21 is the design resistance, which is calculated from the design equation with the given
capacity factors. Eq. 21 can be solved numerically for the only unknown term βi. Again, the
Active-Set Optimisation algorithm [29] is used as a numerical solver for this nonlinear equation.
This solving for the term βi in Eq. 21 is repeated for the complete test results, and the reliability of a
design equation with given capacity factors is calculated as the average of the βi values as follows:
1 N
  i
N i1
(22)

where N = the number of test results.

174 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

6. RESULTS AND DISCUSSIONS

6.1 Capacity Factor Calibration using Johnson and Huang’s Method

In this section, we first use Johnson and Huang’s method summarised in Section 4 to calibrate the
capacity reduction factors (ϕ and ϕc) in Eqs. 1 and 2. The preliminary work of this analysis can be
found in [25], in which Johnson and Huang’s method was used to calibrate the factors for CFST
stub columns according to the Eurocode format. For this calibration, we need to set the target
reliability index β required in Eqs. 13 and 17. The target reliability index is related to the expected
social and economic consequences of a design failure. According to AS 5104: 2005 [26] /ISO
2394:1998 [27], the suggested reliability index for ultimate limit-state design is t = 3.8, which
corresponds to a case in which the consequences of failure are great (the highest level) and the
relative costs of safety measures are moderate (Table 2). These are equivalent to ‘Consequence
Classes’ in EN 1990, and ‘Importance Levels’ in AS/NZS 1170 [28]. As mentioned earlier, the
procedure in Section 4 only considers resistance, ignoring the load effect, and the target reliability
index β is calculated as 0.8 βt = 3.04 by using a dominating resistance parameter, αR = 0.8, as
recommended in AS 5104: 2005 [26] /ISO 2394:1998 [27].

Based on this target reliability index, the calculations in Eqs. 7-20 are conducted, and the capacity
reduction factors for steel and concrete for rectangular and circular columns are calculated. The
results are reported in Table 3, along with the results of the intermediate steps.

Table 3. Results Summary for Target Reliability Index β = 3.04

Vr
Section types Member types b (mean) ϕ ϕc
Eq. (7) Eq. (10)
Stub columns 1.06 0.14 0.82 0.77
Rectangular
Long columns 1.10 0.16 0.82 0.76
Stub columns 1.03 0.15 0.78 0.73
Circular
Long columns 1.11 0.16 0.83 0.77

In this table, the constant bias term b refers to the conservatism embedded in a design equation
calculated using Eq. 7, and the average of the coefficient of variation of the resistance function Vr
(mean) refers to the accuracy of a given design equation, considering the modelling and parameter
uncertainties calculated using Eq. 10. Because b is defined in terms of the ratio between the
experimental resistance and the theoretical resistance, if the b of a design equation is close to 1,
the equation has almost no conservatism. In other words, the greater the capacity factor values are,
the more embedded conservatism there is. A large Vr (mean) value indicates large uncertainties in a
design equation. To achieve the same target reliability with a large value of Vr (mean), the capacity
factors should be decreased to achieve more conservatism.

From the results reported in the table, long rectangular columns require slightly smaller capacity
factors than long circular columns because they have a slightly smaller b value with the same Vr
(mean) values. Similarly, circular stub columns require the smallest capacity factors because they
have the lowest b value. The design equations for circular stub columns have almost no
conservatism in itself as its b value is 1.03, which is close to 1.

Based on the results of this table, the overall calibrated capacity reduction factors show values for
steel and concrete that are closer together than the values suggested in AS 5100, i.e., ϕ = 0.9, and ϕc
= 0.6. The capacity reduction factor of steel shows a reduction, but that of concrete shows increases
 Design Strength of Concrete-Filled Steel Columns 175

for both rectangular and circular members. This difference can be explained by 1) the fact that the
results in this paper were obtained based on an extensive up-to-date database with a large number
of test results; and 2) the fact that the capacity reduction factors in AS 5100 were originally
suggested for the separate use of steel and concrete, and they do not fully reflect the simultaneous
behaviours of concrete and steel, such as the confinement effect of steel tubes. In Figure 2, the
calibrated results of the capacity reduction factors in Eq. 20 are plotted for a range of target
reliability values between 2.5 and 4.2 to demonstrate the sensitivity of the results. These plots show
that the capacity reduction factors vary linearly according to the target reliability.

Based on these results, we cannot conclude that the current AS 5100 code should be immediately
updated because we do not know how much conservatism is embedded in the capacity factors
provided in AS 5100. However, the interaction between concrete and steel should be carefully
considered when the capacity reduction factors are updated because the values of the capacity
reduction factors are closer to one another than those in AS 5100 [3] for the overall range of
reliability indices.
1 1
Steel () Steel ()
0.9 Concrete (c) 0.9 Concrete (c)

0.8 0.8
Capacity factors

Capacity factors

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4
2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(a) Rectangular stub columns (b) Circular stub columns

1 1
Steel () Steel ()
0.9 Concrete (c) 0.9 Concrete (c)

0.8 0.8
Capacity factors

Capacity factors

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4
2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(c) Long rectangular columns (d) Long circular columns

Figure 2. Capacity Factor Versus Reliability Index for Different Member Types

6.2 Inverse Analysis Results using the Proposed Inverse Analysis Method
176 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

To compare the effects of the capacity factor values calculated using Johnson and Huang’s method
[16] and those provided in AS 5100, regarding the design equations’ reliability, the inverse analysis
method proposed in Section 5 is utilised. For these two cases, inverse analyses have been
performed, and the reliability for each section/member types has been calculated. The results are
listed in Table 4. In this table, we can first confirm that, for all cases, the capacity factors obtained
using Johnson and Huang’s method provide reliability indices similar to the target reliability index,
β = 3.04, as expected, while the capacity factors given by AS 5100 provide much higher reliability
indices that are close to 3.40. Here, the slight fluctuations in the values around 3.04 for the capacity
factors calculated from Johnson and Huang’s method are caused by the error produced by the
numerical solvers and the averaging procedures in Eqs. 19 and 22, where the scattering of the
calculated values is neglected. These results are also illustrated in Figure 3 as a bar graph. In this
graph, the dotted lines show the reliability index β = 3.04. Johnson and Huang’s method’s capacity
factors provide reliability around this dotted line while the capacity factors provided in AS 5100
provide greater reliability around a value of 3.4, indicating greater conservatism. In addition,
different section/member types of CFST columns provide different reliability indices based on type
because the design equations themselves have different embedded conservatisms and accuracies for
different column types.

The conservatism of the capacity factors provided by AS 5100 can also be clearly confirmed by the
following analyses. Here, we calculate the optimal capacity factors for the target reliability index, β
= 3.04, but we fix one of the capacity factors as the value provided in AS 5100 and calculate the
other factor. For example, we fix the capacity factor for steel as 0.9 and calculate the remaining
capacity factor for concrete, or we fix the capacity factor for concrete as 0.6 and calculate the
remaining capacity factor for steel. From this analysis, we can confirm that the design equations are
conservative if the calculated capacity factor has greater values than those in AS 5100 for the target
reliability index, β = 3.04, and vice versa. For these analyses, all the calculations in Eqs. 7-13 are
used in a same manner, but the remaining calculation steps in Eqs. 14-20 are modified so as to be
simpler because we only estimate a single capacity factor. The calculation steps and assumptions
used in Eqs. 14-17 are not needed in this analysis, because they are only needed for multiple
capacity factor calculation, and in Eq. 18, the unknown term ϕmi inside ϴi is replaced by ϕi for steel
or ϕci for concrete for the i-th test. Again, to solve for ϕi or ϕci, any numerical solver designed for
non-linear programming can be used, and in this study, the Active-Set Optimisation algorithm [29]
is used.

Table 4. Calculated Reliability Indices from the Inverse Analysis

Section types Member types Capacity factors Reliability index (β)
Johnson and Huang, ϕ = 0.82, ϕc = 0.77 3.06
Stub columns
AS 5100, ϕ = 0.9, ϕc = 0.6 3.43
Rectangular
Johnson and Huang, ϕ = 0.82, ϕc = 0.76 3.02
Long columns
AS 5100, ϕ = 0.9, ϕc = 0.6 3.44
Johnson and Huang, ϕ = 0.78, ϕc = 0.73 2.97
Stub columns
AS 5100, ϕ = 0.9, ϕc = 0.6 3.49
Circular
Johnson and Huang, ϕ = 0.83, ϕc = 0.77 3.00
Long columns
AS 5100, ϕ = 0.9, ϕc = 0.6 3.36

The analysis results are plotted in Figures 4 and 5, where one capacity factor is calculated for a
 Design Strength of Concrete-Filled Steel Columns 177

range of target reliability values between 2.5 and 4.2 when the other factor is fixed to the value
provided in AS 5100. Figure 4 shows the results when the capacity factor for steel is fixed at 0.9
and the factor for concrete is calculated. Figure 5 shows the results when the capacity factor for
concrete is fixed at 0.6 and the factor for steel is calculated. For all results in Figure 4, the
calculated capacity factor for concrete is always greater than 0.6 for the target reliability index, β =
3.04. This means that the capacity factors given in AS 5100 provide higher reliability than the
target reliability, implying more conservatism. This observation is also supported by Figure 5, and
the calculated capacity factor for steel is always greater than that provided in AS 5100 for the target
reliability index, β = 3.04. These results are consistent with the inverse reliability analysis results in
Table 4 and Figure 3.

Table 5. Conservatism and Uncertainties after applying Capacity Factors

Vr
(mean) Vr
Section types
Member
Capacity factors b Eq. (10) (mean)
types Eq. (7) (before applying Eq. (10)
capacity factors)
Stub Johnson and Huang, ϕ = 0.82, ϕc = 0.77 1.33 0.14
0.14
columns AS 5100, ϕ = 0.9, ϕc = 0.6 1.40 0.16
Rectangular
Long Johnson and Huang, ϕ = 0.82, ϕc = 0.76 1.43 0.16
0.16
columns AS 5100, ϕ = 0.9, ϕc = 0.6 1.54 0.18

Stub Johnson and Huang, ϕ = 0.78, ϕc = 0.73 1.37 0.15

0.15
columns AS 5100, ϕ = 0.9, ϕc = 0.6 1.50 0.15
Circular
Long Johnson and Huang, ϕ = 0.83, ϕc = 0.77 1.46 0.18
0.16
columns AS 5100, ϕ = 0.9, ϕc = 0.6 1.54 0.18

4.2
Johnson and Huang
4
AS5100
3.8
()
Reliability index 

3.6

3.4
 = 3.04
3.2

2.8

2.6

Rectangular stub Rectangular long Circular stub Circular long

Section/member types

Figure 3. Comparison of Reliability Indices for Different Capacity Factors

178 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

To see the conservatism of the capacity factors for both steel and concrete in AS 5100, similar
analyses have been performed by fixing the capacity factor for steel at 0.93, which is a slightly
increased value from that in AS 5100. This value was chosen to confirm if the capacity factor for
concrete is still greater than the value provided in AS 5100 when using this slightly increased value
in the capacity factor for steel. The analysis results are plotted in Figure 6, and again, for all
section/member types, the calculated capacity factor for concrete for the target reliability index, β =
3.04 is still greater than 0.6, which is the value provided in AS 5100. These results imply that the
capacity factors provided in AS 5100 are conservative compared to the target reliability suggesting
both capacity factors can be slightly increased to create an optimal design.

1 1

0.9 0.9

0.8 0.8
Capacity factors

Capacity factors
0.7 0.7

0.6 0.6

0.5 Steel () 0.5 Steel ()

Concrete (c) Concrete (c)
0.4 0.4
2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(a) Rectangular stub columns (b) Circular stub columns

1 1

0.9 0.9

0.8 0.8
Capacity factors

Capacity factors

0.7 0.7

0.6 0.6

0.5 Steel () 0.5 Steel ()

Concrete (c) Concrete (c)
0.4 0.4
2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(c) Long rectangular columns (d) Long circular columns

Figure 4. Capacity Factor Versus Reliability Index when

the Capacity Factor for Steel is Fixed at 0.9.
 Design Strength of Concrete-Filled Steel Columns 179

1.5 1.5
Steel () Steel ()
Concrete (c) Concrete (c)
Capacity factors

Capacity factors
1 1

0.5 0.5

2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(a) Rectangular stub columns (b) Circular stub columns

1.5 1.5
Steel () Steel ()
Concrete (c) Concrete (c)
Capacity factors

Capacity factors

1 1

0.5 0.5

2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(c) Long rectangular columns (d) Long circular columns

Figure 5. Capacity Factor Versus Reliability Index when
the Capacity Factor for Concrete is Fixed at 0.6

6.3 The Effect of Multiple Unbalanced Capacity Factors

Johnson and Huang’s method [16] and the inverse analyses in the previous sections assumed that
the modelling and parameter uncertainties were same before and after the application of the
capacity factors. However, they did not consider any additional uncertainties created after applying
the capacity factors to the theoretical equations. In the original form of a theoretical equation, no
capacity factors are considered, or it is assumed that all the capacity factors are equal to unity, and
all the modelling and parameter uncertainties are estimated based on this form. However, if we
apply different capacity factor values to different materials in a composite member, this will create
an imbalance between the predicted strengths of these materials. This imbalance will create
additional uncertainties because it will change the design equations’ accuracies and propagate
parameter uncertainties in a different manner. Therefore, in this study, we repeat the inverse
analysis carried out in the previous section by recalculating the parameter and modelling
uncertainties after applying the capacity factors. Because the application of imbalanced capacity
factors for steel and concrete strengths will change the accuracy of the design equations, the
uncertainties need to be updated accordingly.
180 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

1 1

0.9 0.9

0.8 0.8
Capacity factors

Capacity factors
0.7 0.7

0.6 0.6

0.5 Steel () 0.5 Steel ()

Concrete (c) Concrete (c)
0.4 0.4
2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(a) Rectangular stub columns (b) Circular stub columns

1 1

0.9 0.9

0.8 0.8
Capacity factors

Capacity factors

0.7 0.7

0.6 0.6

0.5 Steel () 0.5 Steel ()

Concrete (c) Concrete (c)
0.4 0.4
2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2
Reliability index () Reliability index ()

(c) Long rectangular columns (d) Long circular columns

Figure 6. Capacity Factor Versus Reliability Index when

the Capacity Factor for Steel is Fixed to be 0.93

To investigate the effect of these additional uncertainties created via the application of different
capacity factors for different materials, the inverse analysis proposed in Section 5 has been repeated,
but the resistance function gR(x) in Eqs. 7-10 have been replaced by g R (x, ) , where x = the
mean-measured input parameter values and ϴ = the applied capacity factors. Due to this change, the
values of b in Eq. 7 and Vr in Eq. 10 are updated as shown in Table 5. In this table, the constant
bias term b refers to the updated conservatism of a design equation, and Vr (mean) refers to the
updated accuracy of a design equation. For all section/member types, the b values are greater
than those in the previous inverse analysis because the design equations now have more
conservatism after applying capacity factors. The Vr (mean) values are increased or remain the
same after applying the capacity factors, because the accuracy of the imbalanced equations has
been changed due to the application of these factors.

After applying the capacity factors provided by AS 5100, the reliabilities for all section/member
types have been reduced to values close to the target reliability index, β = 3.04, except for the
equations for circular stub columns (Table 6). This exception occurred because the Vr (mean) value
for the equations for circular stub columns did not increase after applying the capacity factors
provided by AS 5100, while the Vr (mean) values for the other cases were increased. The increase
 Design Strength of Concrete-Filled Steel Columns 181

in the Vr (mean) values indicates an increase of modelling error or a decrease of accuracy, and this
reduces the reliability of equations. The application of different constant values to the steel and
concrete strength predictions in an unbalanced manner decreases the accuracy of the equations,
except for those related to circular stub columns. It is also noted that the application of the capacity
factors obtained from the Johnson and Huang’s method decreases the accuracy of the equations for
long circular columns and these values cannot achieve the target reliability index, β = 3.04, after
considering the increased uncertainties caused by the application of the capacity factors obtained
from Johnson and Huang’s method. The reliability indices after applying the capacity factors are
also represented as a bar graph in Figure 7.
4.2
Johnson and Huang
4
AS5100
3.8
()
Reliability index 

3.6

3.4

 = 3.04
3.2

2.8

2.6

Rectangular stub Rectangular long Circular stub Circular long

Section/member types

Figure 7. Comparison of Reliability Indices for Different Capacity Factors

after applying Capacity Factors

It is observed that the capacity factors for composite columns in AS 5100 yield almost the same
target reliability suggested in AS 5104: 2005 [26] /ISO 2394:1998 [27] after considering the
additional uncertainties created via the application of the capacity factors. These results show that
the current capacity factor values in AS 5100 for CFST columns need to be maintained. However,
this is due to the significant reduction in the accuracy of the design equations that reduce the large
conservatism embedded in these factors, which concludes that these factors are not optimally
calibrated for cost-safety balanced design. In addition, the results in this paper and the parameter
distributions shown in Figure 1 demonstrate that the existing design equations can be safely
extended to cover a wider scope than is currently considered in either EC 4 or AISC, as follows:

 Rectangular hollow sections

o 200 MPa < fym < 460 MPa
o 20 MPa < fcm < 100 MPa
o Depth h to width B ratio of the composite cross section 1.0 < h/D < 2.0
250
o Max h/t = 125
fy
 Circular hollow sections
o 180 MPa < fym < 680 MPa
o 20 MPa < fcm < 100 MPa
 250 
o Max D/t = 225  
 fy 
182 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

7. CONCLUSIONS

The aim of this paper was to attempt to improve the practical applications of CFST members by
re-calibrating the capacity reduction factors in AS 5100 [3] and investigate the effect of these
factors and the factors given by AS 5100. The factor calibration method developed by Johnson and
Huang [16] was used to re-calibrate the capacity factors for CFST columns in AS 5100, based on
the extensive database developed by Tao et al. [7], which contains 1,583 test results for CFST
columns. In addition, an inverse analysis procedure based on Johnson and Huang’s method was
proposed in order to estimate the reliability of design equations with known capacity factors. Using
these methods, first, the capacity factors for CFST columns in AS 5100 were calibrated, and the
results showed that the values of the capacity reduction factors are closer to each other than those in
AS 5100. Second, the proposed inverse analysis procedure was used to estimate the reliability of
the equations in AS 5100, and the results showed that the current capacity factors given in AS 5100
provide more conservatism, represented by a higher reliability than the target reliability suggested
in AS 5104: 2005 [26] /ISO 2394:1998 [27]. Lastly, the inverse analysis was repeated considering
the additional uncertainties created from the application of the capacity factors. It was observed that
the application of different constant values to the steel and concrete strength predictions in an
unbalanced manner decreases the accuracy of the equations in most cases, and the capacity factors
for composite columns in AS 5100 give almost the same target reliability as that suggested in AS
5104: 2005 [26] /ISO 2394:1998 [27]. After taking all the results into consideration, the current
capacity factor values in AS 5100 for CFST columns are adequate with regards to safety and can be
maintained to meet the target reliability suggested in the current design codes. However, it should
also be noted that this is due to the significant reduction in the accuracy of the design equations that
reduce the large conservatism embedded in these factors, which leads to the conclusion that they
are not optimal for cost-safety balanced design. For future research, further development of the
capacity factor calibration method would be advantageous in order to incorporate the uncertainties
caused by the imbalance between the predicted strengths of these materials as observed from the
analysis of the proposed inverse method for estimating the reliability of capacity prediction models.
It should also be noted that although these studies have been carried out for the purposes of the
Australian Standards, they have important ramifications for international codes of practice
regarding structural composite members such as with EC 4 [4], AISC [5], and the code of practice
in Hong Kong [6].

ACKNOWLEDGMENTS

This work is supported by the Australian Research Council (ARC) under its Discovery project
(Project No: DP120101944). The authors wish to thank Dr Douglas Goode, the University of
Manchester, UK, and Professor Roberto T Leon, Virginia Tech, USA for providing the initial
column databases.

REFERENCES

[1] Shams, M. and Saadeghvaziri, M.A., “State of the Art of Concrete-filled Steel Tubular
Columns”, ACI Structural Journal, 1997, Vol. 94, No.5, pp. 558-571.
[2] Chitawadagi, M.V., Narasimhan, M.C. and Kulkarni, S., “Axial Capacity of Rectangular
Concrete-filled Steel Tube Columns-DOE Approach”, Construction and Building Materials,
2010, Vol. 24, No. 4, pp. 585-595.
[3] Standards Australia International Ltd., “AS 5100.6:2004 Bridge Design, Part 6: Steel and
Composite Construction”, New South Wales, Australia, 2004.
 Design Strength of Concrete-Filled Steel Columns 183

[4] British Standards Institution, “BS EN 1994-1-1:2005 Eurocode 4: Design of Composite

Steel and Concrete Structures”, Part 1-1. London, UK, 2005.
[5] American Institute of Steel Construction, “Specification for Structural Steel Buildings.
ANSI/AISC 360-10”, Chicago, IL, USA, 2010.
[6] Buildings Department of Hong Kong, “Code of Practice for Structural Use of Steel”,
Buildings Department, the HKSAR Government. Hong Kong, 2005.
[7] Tao, Z., Uy, B., Han, L.H. and He, S.H., “Design of Concrete-filled Steel Tubular Members
According to the Australian Standard AS 5100 Model and Calibration”, Australian Journal
of Structural Engineering, 2008, Vol. 8, No. 3, pp. 197-214.
[8] Goode, C.D., “A Review and Analysis of over One Thousand Tests on Concrete Filled Steel
Tube Columns”, Proceedings of 8th International Conference on Steel-Concrete Composite
and Hybrid Structures, Harbin, China, 2006, August 12-15.
[9] Wu, F.Y., “Compressive Behaviour of Recycled Concrete-filled Steel Tubes”, Masters
Thesis, College of Civil Engineering, Fuzhou University, China, 2006. (in Chinese).
[10] Gupta, P.K., Sarda, S.M. and Kumar, M.S., “Experimental and Computational Study of
Concrete Filled Steel Tubular Columns under Axial Loads”, Journal of Constructional Steel
Research, 2007, Vol. 63, No. 2, pp. 182-193.
[11] Lue, D.M., Liu, J.L. and Yen, T., “Experimental Study on Rectangular CFT Columns with
High-Strength Concrete”, Journal of Constructional Steel Research, 2007, Vol. 63, No. 1,
pp. 37-44.
[12] Tao, Z., Han, L.H. and Wang, D.Y., “Experimental Behaviour of Concrete-filled Stiffened
Thin-walled Steel Tubular Columns”, Thin-Walled Structures, 2007, Vol. 45, No. 5, pp.
517-527.
[13] Yu, Z.W., Ding, F.X. and Cai, C.S., “Experimental Behavior of Circular Concrete-filled
Steel Tube Stub Columns”, Journal of Constructional Steel Research, 2007, Vol. 63, No. 2,
pp. 165-174.
[14] Yang, Y.F. and Han, L.H., “Concrete Filled Steel Tube (CFST) Columns Subjected to
Concentrically Partial Compression”, Thin-Walled Structures, 2012, Vol. 50, No. 1, pp.
147-156.
[15] Chang, X., Fu, L., Zhao, H.B. and Zhang, Y.B., “Behaviors of Axially Loaded Circular
Concrete-filled Steel Tube (CFT) Stub Columns with Notch in Steel Tubes”, Thin-Walled
Structures, 2013, Vol. 73, pp. 273-280.
[16] Johnson, R.P. and Huang D., “Statistical Calibration of Safety Factors for Encased
Composite Columns”, Composite Construction in Steel and Concrete III, ASCE, New York,
1997, pp. 380-391.
[17] Yu, Q., Tao, Z. and Wu, Y.X., “Experimental Behaviour of High Performance
Concrete-filled Steel Tubular Columns”, Thin-Walled Structures, 2008, Vol. 46, No. 4, pp.
362-370.
[18] Chen, Z., Zhu, J. and Wu, P., “High Strength Concrete and its Application”, Beiing, China,
Tsinghua University Press, 1996. (in Chinese).
[19] Gulvanesian, H. and Holicky, M., “Annex C – Calibration Proceedure”, Leondardo DaVinci
Pilot Project CZ/02/B/F/PP-134007, Handbook 2-Reliability Backgrounds, 2005.
[20] Ang, A.H.S. and Tang, W.H., “Probability Concepts in Engineering: Emphasis on
Applications to Civil and Environmental Engineering”, New York, John Wiley & Sons,
2007.
[21] Joint Committee on Structural Safety, JCSS Probabilistic Model Code,
http://www.jcss.byg.dtu.dk/, 2001.
[22] Standards Australia International Ltd., “AS 3600:2009 Concrete Structures”, New South
Wales, Australia, 2009.
[23] Standards New Zealand, “NZS 3104: 2003 Specification for Concrete Production”,
Wellington, New Zealand, 2003.
184 W.-H. Kang, B. Uy, Z. Tao and S. Hicks

[24] Standards Australia/Standards New Zealand, “AS/NZS 1163:2009 Cold-Formed Structural

Steel Hollow Sections”, Sydney, Australia, 2009.
[25] Kang, W.H., Uy, B., Tao, Z. and Hicks, S., “Statistical Safety Factor Calibration of Short
Concrete-filled Steel Tubular Columns”, Proceedings of the 22nd Australasian Conference
on the Mechanics of Structures and Materials, ACMSM 22, Sydney, Australia, 2012,
December 11-14.
[26] Standards Australia International Ltd., “AS 5104: 2005 General Principles on Reliability for
Structures”, New South Wales, Australia, 2005.
[27] International Organization for Standardization, “ISO 2394:1998 General Principals on
Reliability for Structures”, Geneva, 1998.
[28] Standards Australia International Ltd., “AS/NZS 1170.0: 2002 Structural Design Actions –
General Principles”, New South Wales, Australia, 2002.
[29] Nocedal, J and Wright, S., “Numerical Optimization. 2nd ed.”, Berlin, New York:
Springer-Verlag, 2006, ISBN 978-0-387-30303-1.