Missouri University of Science and Technology

Scholars' Mine

International Specialty Conference on Cold- (2012) - 21st International Specialty Conference

Formed Steel Structures on Cold-Formed Steel Structures

Aug 24th, 12:00 AM - Aug 25th, 12:00 AM

Design Capacity Determination Assisted by Testing Based on

LRFD Method
Y. Q. Li

L. P. Wang

Z. Y. Shen

Follow this and additional works at: https://scholarsmine.mst.edu/isccss

Part of the Structural Engineering Commons

Recommended Citation
Li, Y. Q.; Wang, L. P.; and Shen, Z. Y., "Design Capacity Determination Assisted by Testing Based on LRFD
Method" (2012). International Specialty Conference on Cold-Formed Steel Structures. 2.
https://scholarsmine.mst.edu/isccss/21iccfss/21iccfss-session12/2

This Article - Conference proceedings is brought to you for free and open access by Scholars' Mine. It has been
accepted for inclusion in International Specialty Conference on Cold-Formed Steel Structures by an authorized
administrator of Scholars' Mine. This work is protected by U. S. Copyright Law. Unauthorized use including
reproduction for redistribution requires the permission of the copyright holder. For more information, please
contact [email protected].
 Twenty-First International Specialty Conference on Cold-Formed Steel Structures
St. Louis, Missouri, USA, October 24 & 25, 2012

Design Capacity Determination Assisted by Testing Based on

LRFD Method

Y.Q. Li1, L.P. Wang2 and Z.Y. Shen1

Abstract

In some special circumstances of structural design, adequate calculation models

are not available in current specifications or they cannot be used directly for new
materials or structural configurations. In these cases, tests of the prototype units
may be accepted as an alternative to calculation. If this alternative procedure is
adopted, corresponding requirements and evaluation of test results should apply.
In this paper, a complete procedure of determining the design capacity of the test
specimens based on statistical analysis and Load and resistance factor design
(LRFD) method was presented: characteristic value of experimental resistance
was obtained from the minimum value of test results, and the design value of
experimental resistance was determined with proper resistance partial coefficient
to achieve the level of reliability required for the relevant design situation.
Finally, the preconditions and applications of this method were discussed.
Keywords: testing; characteristic value; design value; probability; reliability
analysis; LRFD

1
Professor, Department of Building Engineering & State Key Laboratory for Disaster
Reduction in Civil Engineering, Tongji University, Shanghai, China
2
Graduate Student, Department of Building Engineering, Tongji University, Shanghai,
China

841
 842

Introduction

In some circumstances of structural design, for example, in the promotion of

structural innovation and the application of new material and new structural
configurations, there are no existing methods for reference. An accepted way is
to test on the structures or members to determine the design capacity, such as the
design value of resistance. Relevant provisions can be found in structural
standards of several countries and the calibration procedures are well established.
In the European code
BS EN 1990:2002, the conception of ‘design assisted by testing’ was elaborately
presented, in which the plan and implementation of test as well as evaluation of
test results are given. In Australian /New Zealand Standard for Cold-formed
Steel Structures AS/NZS 4600:2005, testing for assessment or verification is
introduced and the method for determining the design capacity of specific
products and assemblies is given. In the American Iron and Steel Institute
Specification for the design of cold-formed steel structural members: 2001,
structural tests are divided into three categories among which the first class is
called ‘tests for determining structural performance’.
Currently, the methodologies of design assisted by testing in the above standards
are already used in practice. It seems there are some differences in application
among different standards. In this paper, firstly, the theory and methodologies in
EN 1990 and AS/NZS 4600 were briefly reviewed, and then the authors were
trying to find an alternative solution to the problem. In the proposed method,
statistical method was adopted to determine the characteristic value of
experimental resistance from the minimum value of test results, further the
design value of experimental resistance was determined with proper resistance
partial coefficient to achieve the level of reliability required for the relevant
design situation. Finally, the preconditions and applications of the method were
discussed.
Review of the methodology about design assisted by testing

BS EN 1990:2002
Assessment of the design capacity via the characteristic value
In BS EN 1990:2002, there are two ways provided to obtain the design value,
one is via the characteristic value:
X k (n ) d
X d  d  m 1  knVX  (1)
m m X
 843

where, d is the design value of the conversion factor; X k (n ) is the characteristic

value including statistical uncertainty for a sample of size n with any conversion
factor excluded;  m is partial factor for material property; mX is the mean of n
sample results; VX is the coefficient of variation of X. kn is given in Table 1.
Table 1 Values of kn for the 5% characteristic value (Normal distribution)
n 1 2 3 4 5 6 8 10 20 30 

VX known 2.31 2.01 1.89 1.83 1.80 1.77 1.74 1.72 1.68 1.67 1.64

VX -- -- 3.37 2.63 2.33 2.18 2.00 1.92 1.76 1.73 1.64

unknown

In Eq. (1) the coefficient of variation, VX is usually unknown in advance and it

is estimated from the sample, or is known from prior knowledge. Student’s T
distribution is implemented to accounting for the limited number of samples.
Direct assessment of the design value
Directly assessment of the design value from experimental results is defined as
X d = hd mX {1 - kd ,nVX } (2)

where d covers all uncertainties not covered by tests. k d ,n is given in Table 2.

Table 2 Values of k d ,n for the ultimate load state design value (Normal
distribution)
n 1 2 3 4 5 6 8 10 20 30 
VX known 4.36 3.77 3.56 3.44 3.37 3.33 3.27 3.23 3.16 3.13 3.04

VX
-- -- -- 11.40 7.85 6.36 5.07 4.51 3.64 3.44 3.04
unknown

The value of kd ,n in Table 2 gives the design value a probability of observing a

lower value of about 0.1% according to intended reliability index.
It is found that the method presented in the EN code is very sensitive to the
observed standard deviation  R which is not known in most cases. A suggested
way to achieve this is by choosing a proper distribution for the standard
deviation, while this procedure is relatively complex in some circumstances.
AS/NZS 4600:2005
In AS/NZS 4600 (2005), the method for determining the design capacity of test
 844

specimens is based on testing of a group of repeated specimens. The design

capacity (Rd) is determined by dividing the minimum value of experimental
resistance Rmin by a modification factor kt considering the variability of
structural units. The equation is
R 
Rd   min  (3)
 kt 

It is emphasized that this method does not apply to the testing of structural
models nor to the establishment of general design criteria. Modification factor is
given in Table 3.
Table 3 Factors ( kt ) to allow for variability of structural units in AS/NZS 4600
Coefficient of variation of structural characteristics (Vsc)
No. of units to
be tested (n) 5% 10% 15% 20% 25% 30%

1 1.20 1.46 1.79 2.21 2.75 3.45

2 1.17 1.38 1.64 1.96 2.36 2.86

3 1.15 1.33 1.56 1.83 2.16 2.56

4 1.15 1.30 1.50 1.74 2.03 2.37

5 1.13 1.28 1.46 1.67 1.93 2.23

10 1.10 1.21 1.34 1.49 1.66 1.85

100 1.00 1.00 1.00 1.00 1.00 1.00

In Table 3, n is the sample size; the coefficient of variation of structural

characteristics can be calculated via the formula Vsc  k 2f  km2 , where km is the
coefficient of variation of material property; kf is the coefficient of variation of
geometric fabrication.
In AS/NZS 4600, it seems that the modification factor kt covers all kinds of
relevant uncertainties, such as the sample size of test, material property, and
model uncertainty which may be vague and lead to unsafe or uneconomic results
in some cases.
A new method for assessment of design capacity

Based on the idea of method for determining the design capacity involved in EN
code and AS/NZS code, this paper attempted an alternative way to this problem
to simplify the calibration procedure. Also the new method was elaborately
 845

developed to make it closer connected with the general principle of limit state
design based on LRFD.
Determination of characteristic value of experimental resistance
Experimental results of the structural property like resistance of individual
specimens take on discreteness of different level (Jane E 1996) due to variation
of the structural characteristics including the effect of variation of material
strength, geometrical fabrication, external actions and construction quality, etc.
(Shao-Fan Chen 2005).
Generally, structural resistance is a function of several relevant variables.
Normal distribution is often used for the distribution of resistance; however, this
assumption is regarded as a relatively conservative one. (ISO 2394:1998) In this
paper, lognormal distribution has been adopted for structural resistance, which is
regarded to be more suitable (Ji-Hua Li 1990).
In this procedure, two main assumptions were made: (1) experimental results of
individual specimen Xi (i=1 …n)are independent from each other; (2) the
experimental resistance of test specimens follows lognormal distribution
expressed as ln X ~ N   ,  2  , where   E[ln( X )] ,  2  D[ln( X )] is the mean value
and variance of the logarithm of experimental resistance, respectively. The
probability density function of lognormal distribution is:
1  (ln x   )2 
f  x；，   exp   ，x  0 (4)
x 2  2 2 

If the distribution mean value and variance of experimental resistance is

expressed using the mean value  R and variance of resistance  R 2 respectively,
the relation of  R ,  R 2 and  ,  2 is:
  R2
 = ln（ ）， 2 = ln（1+（ ）） (5)
R 2 R2
1+（ ）
R2

Given the probability distribution of experimental resistance, the characteristic

value with a prescribed survival probability RK can be determined
mathematically. In normal conditions, the characteristic value is assumed to be
the 0.05fractile of a Normal distribution. Thus RK with a survival probability of
95% can be expressed with equation: P  ln x   -1.645   5% Namely,
P  x  e -1.645   5% , hence,

RK  e -1.645 (6)

Here a key issue arises: for a specified group of tests, the sample mean value
 846

mR and the sample standard deviation sR should be adopted in Eq. (4) and Eq.
(5), however usually we don’t know their values before the tests are conducted.
In this paper, the distribution mean value  R and standard deviation  R of
experimental resistance are taken in place of mR and sR . The difference was
accounted for by determining the characteristic value from the minimum value
of the tested specimens instead of the average value. The more unpredictable
minimum value is assumed to well represent the deviation in real testing as well
as the influence of statistical uncertainty due to a limited sample size on the
distribution model of experimental results.
In the process of determining characteristic value from the minimum value，an
equation was assumed:
RKt  Rmin kt (7)

where Rmin is the minimum value of experimental resistance, kt is the

modification factor, RKt is characteristic value of the experimental resistance
determined from Rmin .
In order to make RKt has at least the same survival probability with RK, it should
satisfy the relation
P  RKt  RK   95% (8)

Case 1: one specimen tested

In the circumstance that only one specimen is tested, the minimum value is just
the experimental value. Substitute Eq. (6) and Eq. (7) into Eq. (8), the following
relation is obtained
P  x  kt e -1.645   95% (9)

where x is variable representing the experimental resistance.

Mathematically, 0.95 fractile of the probability distribution of the experimental
resistance follows equation P  ln x   +1.645   95% , we can get
 R2
3.29 ln（1+（ ）
R2 3.29 ln（1+（Vsc 2）
kt  e 3.76
e e (10)

Vsc  (11)


where Vsc is the coefficient of variation of structural resistance, according to

the assumption above in this method, Vsc can be determined from prior
 847

information Vsc  k 2f  km2 . Where km is the coefficient of variation of material

property; k f is the coefficient of variation of geometric fabrication. So the
values of kt can be obtained and they are given in Table 4.
Case 2: more than two test specimens tested
For two test specimens and above, experimental resistance X i (i=1 …n) of
individual specimens are sorted according to their value. It is known
mathematically that the minimum experimental value symbolized by X (1) is a
random variable named as Minimal Order Statistics with the probability density
function expressed as (Cheng-Yi Pan 1993):
n n1  (ln x   ) 2 
f *1  x   1  F  x   exp    (12)
x 2  2 2 

where n is the number of specimens. F(x) is the lognormal distribution function,

which is:
1  (ln t   ) 2 
F  x  
x
exp   dt (13)
0
t 2  2 2 

Substitute Eq. (6) and Eq. (7) into Eq. (8), we’ll have
P  x1  kt e -1.645   95% (14)

where x1 is variable representing the minimum value of experimental

resistance, with probability density function expressed in Eq. (12).
Integration was performed to Eq. (14), and then using Eq. (12), we can get:
(1.645 s ) ln（1+（Vsc 2）
kt  e (15)

The indication of symbols is same to that described in case1. So the values of kt

can be obtained and they are given in Table 4.
With the value of kt , characteristic value of resistance can be determined from
the minimum value via the equation RKt  Rmin kt . It is shown in Table 4 that when
the number of specimens is small or variation of structural resistance is big, the
value of kt is larger. Hence, influence of limited number of specimens is
rationally accounted for in kt .
 848

Table 4 Value of kf considering the variation of structural resistance

Coefficient of variation of resistance ( Vsc )

No. of units to
be tested (n) 5% 10% 15% 20% 25% 30%

1 1.21 1.46 1.75 2.11 2.52 3.02

2 1.15 1.33 1.52 1.75 2.00 2.29

3 1.13 1.27 1.42 1.60 1.79 2.01

4 1.11 1.23 1.37 1.51 1.67 1.85

5 1.10 1.21 1.33 1.45 1.59 1.74

10 1.07 1.14 1.22 1.30 1.39 1.48

100 1.00 1.01 1.01 1.01 1.01 1.02

Evaluation of design capacity based on LRFD

According to the principle of load and resistance factor design, the characteristic
value RKt should be divided by the appropriate factors to obtain the design
value Rd .According to Chinese design codes (GB50068-2001), it is required that
the reliability index of any kind of members designed by current code should be
no less than the target reliability index  presented in Table 5. The specific
procedure of solving of  R can be referred on (Xin-Pei Zhang 2001).
Table 5 Reliability index (  ) of structural members based on Ultimate Limit
State Design
Safety grades
Facture type

Ductile 3.7 3.2 2.7

Brittle 4.2 3.7 3.2

In this paper, the following two steps are taken to obtain a reasonable and
feasible resistance partial coefficient.
Step 1: Perform a tentative analysis on the resistance partial coefficient using the
least square method in which the error between the characteristic value of
resistance RK determined with the practical formula presented in current codes
and that determined by direct probability reliability method RKt is the minimum.
In this article, two types of combination cases were considered according to
“Load code for design of building structures GB50009-2001” as did in (Yuan-Qi
 849

Li 2007).
The practical design formula for the load combination of dead load, live load
and wind load is presented as
 G SG  （ Q SQ   W SW ） RK  R
K K K
(16)

where  G is the partial coefficient of dead load; SG is the characteristic value K

of effect of dead load;  is the combination coefficient;  Q is the partial

coefficient of live load; SQ is the characteristic value of effect of live load;
K

 W is the partial coefficient of wind load; SW is the characteristic value of

K

effect of wind load; RK is the characteristic value of resistance;  R is the

resistance partial coefficient. Assuming  j  ( S L  SW ) SG (j=1, 2,…, n) and K K K

 k  SW S L (k=1, 2,…, m). Therefore, given the values of  j ,  k , the

K K

corresponding characteristic resistance designed with the practical formula

specified in current codes is
RK   R ( G SGK  （ Q SQK   W SWK）
) (17)

 R is unknown which will be solved by the least square method.

The practical design formula for the combination of dead load, live load is
presented as
 G SG   Q SQ  RK  R
K K
(18)

Assuming l  S L SG (l=1, 2,…, m), given the values of l , the corresponding

K K

characteristic resistance designed with the practical formula specified in current

codes is
RK   R ( G SGK   Q SQK ) (19)

On the other hand, characteristic value of resistance R tK , determined by direct

probability reliability method can be conservatively taken as R K , which can be
obtained by substituting Eq. (5) into Eq. (6)
R
ln( )1.645 ln(1VR2 )
1VR2
R tK  e (20)

where uR is the mean value of resistance, which can be obtained with a certain
target reliability index using the JC method (Second-order Moment Method)
referred on (Zai-Min Mou 1991) or a more practical method without iteration
process (Guo-Fan Zhao 1984). VR is the variation coefficient of resistance and
in this article.
 850

VR  Vsc  k 2f  km2 (21)

Based on the characteristic value of resistance RK expressed by Eq. (17) or Eq.

(19), and RKt calculated
m n
by Eq.n (20) the least square method was adopted by
assuming H   ( RKjk
t
 RKjk ) 2   ( RKlt  RKl ) 2 , further
k 1 j 1 l 1
m n n
H   ( RKjk
t
  Rt X jk ) 2   ( RKlt   Rt X l ) 2 (22)
k 1 j 1 l 1

In Eq. (22),  Rt is resistance partial coefficient for assessing design capacity

from RKt ; X jk   G SG   ( Q SQ   W SW ) ; X l   G SG   Q SQ .Let H /  Rt  0 ,  Rt can
K K K K K

be determined by:
m n n

 R t
Kjk X jk   RKlt X l
 Rt  k 1 j 1
m n n
l
，m  5, n  4 (23)
 X
k 1 j 1
2
jk   X l2
l

Referring to (Yuan-Qi Li 2007), it is assumed that  j  0.5,1, 2,3 ,  k  0.5,1, 2,3, 4

and l  0.5,1, 2,3 . The values of load partial coefficient and combination
coefficient are taken as  G  1.2 ,  Q  1.4 ,  W  1.4 ,   0.9 . Given the above
relevant parameters,  Rt was calculated by Eq. (23) as shown in Table 6.
Step 2: Based on the tentatively determined value of resistance partial coefficient
in the step1, in order to make it convenient in design practice, a linear
relationship was established between the resistance partial coefficient and the
expected reliability index. On the other hand, to meet the requirement that the
reliability index of any kind of members designed by current code should be no
less than the target reliability index (Jin-Long Chen 2005), the actual value of
 Rt was taken larger than the calculated value with the corresponding reliability
index. In view of these considerations,  Rt was conservatively recommended as
1.00 when   2.7 and 1.15 when   3.7 . For reliability index within the range of
2.7 to 3.7, a linear interpolation was adopted expressed as
 Rt (  )  1.0  0.15(   2.7) and the results were displayed in Table 6.

Table 6 Resistance partial coefficient under different target reliability index

Resistance partial   3.7   3.5   3.3   3.2   3.0   2.7
coefficient
Calculated  Rt 1.0689 1.0386 1.0097 0.9958 0.9689 0.9310

Recommended 1.15 1.12 1.09 1.075 1.045 1.0

With the recommended value of  Rt , the reliability indexes of structural

members in (Yuan-Qi Li, et, al. 2007) under different combinations of external
 851

loads can be obtained, as shown in Table 7, in which it was verified that the
recommended values of  Rt well satisfied the reliability requirement.
When the resistance partial coefficient  Rt is determined, design value of
resistance can be estimated by
Rd =RKt /  Rt (24)
where R t
K  Rmin kt .
Requirement and application

The method for determining the design capacity assisted by testing discussed in
this paper is based on testing of the prototype structures or members. There are
several requirements when it is used: For the test specimens, it is required that
the samples are representative that can truly reflect the geometry, detailing and
loading characteristics, etc. For the execution of testing, it is required that the
testing is repeatable and indispensable, in addition, the test conditions should be
as real as possible to the actual situation, such as avoiding improper loading
which will induce extra constraints to the deformation of the structure. For the
test results, there is also a limitation value of the variation. Referring to the
European specification for cold-formed members and sheeting (EN
1993-1-3:2006), no less than three repeated test specimens are required for
determining the design capacity. If the scatter of test value between individual
specimen and average value is within 10%, the design capacity can be directly
determined in accordance with the methods presented in this paper, otherwise a
group test of at least six specimens should be conducted to revise the results.
Design by testing is a method for establishing design values of resistance
properties for structural, elements and materials. The method in this paper is
based on a statistical evaluation of the test results which is consistent with the
concept of probabilistic design and LRFD. The scope of application covers the
cases that cannot be treated by information given in Codes of practice such as
lacking of theoretical models or data. Also it can be applied when existing
design formulae seem to be conservative or unsafe and derivation of new design
formulae is needed.
Summary

In this article, a method for assessing the design capacity assisted by testing was
theoretically derived, in which an important parameter “modification factor” is
obtained to determine the characteristic value of resistance from the minimum
value of testing results, and then based on LRFD method，proper values of
 852

resistance partial coefficient was recommended to determine the design capacity

via characteristic value. Finally, the preconditions and applications of this
method are discussed. So this paper provides reference for both structural design
and standard development.
The proposed method for evaluation of design capacity assisted by testing in this
paper is dedicated to provide an explicit procedure while alleviate the
complexity in current provisions. The method is consistent with the general
principle of LRFD method which appears be an effective way in solving some
practical design problems. On the other hand, there are still some issues need
further research. One is that in the above discussion, a uniform coefficient of
variation of resistance is used for different types of materials and components;
so that the resistance partial coefficient is very general which needs further
investigation before it can yield more satisfactory results.
Acknowledgements

The authors are grateful to the financial support by the research fund of State
Key Laboratory of Disaster Reduction in Civil Engineering
(No.SLDRCE09-B-01), the National Natural Science Foundation of China (No.
51078288), and Kwang-Hua Fund of College of Civil Engineering, Tongji
University.
References

AISI. (2001). “Specification for the design of cold-formed steel structural

members.” AISI/COS/NASPEC2001, Washington DC.
AS/NZS. (2005). “Cold formed steel structures.” AS/NZS 4600:2005,
Australian/New Zealand Standard.
BS. (2002). “Eurocode-Basis of structural design.” BS EN 1990:2002, British
Standard.
Cheng-Yi Pan, Ying-Hui He. (1993). “Principle and method of mathematical
statistics.” Tongji University Press, Shanghai.
EN. (2006). “Part 1-3: General rules —Supplementary rules for cold-formed
members and sheeting.” EN 1993-1-3:2006(E).
GB. (2001). “Unified standard of reliability design for building structures.”
GB50068-2001, Beijing.
Guo-Fan Zhao. (1984). “A practical method for structural reliability analysis.”
Journal of Building Structures, (3), 1-10.
ISO 2394. (1998). “General principles on reliability for structures”. ISO
2394-1998, International standard.
Jane E. Lundberg, Theodore V. Galambos. ( 1996). “Load and resistance factor
design of composite columns.” Structural Safety, 18(2-3), 169-177.
 853

Ji-Hua Li, Zhong-Min Lin, Kun-Zhen Ma, Zhuo-Min Shao, Ji-Fa Chen, De-Xin
Hu. (1990). “Probabilistic limit state design of building structures.” China
Architecture & Building Press, Beijing.
Jin-Long Chen, Si-Zuo Chen, Ying Wang. (2005). “Study of resistance partial
parameters of all kinds of components of steel structures.” Optimization of
Capital Construction, 26(6), 115-115.
Shao-Fan Chen (2005). “Principles of Steel Structural Design (Third Edition).”
China Science Press, Beijing.
Xin-Pei Zhang.(2001). “Reliability analysis and design of building structures.”
China Science Press, Beijing.
Yuan-Qi Li, Zu-Yan Shen, Lei Wang, Yan-Min Wang, Hong-Wei Xu. (2007).
“Analysis and design reliability of axially compressed members with
high-strength cold-formed thin-walled steel.” Thin-Walled Structures, 45(4),
473-492.
Zai-Min Mou, Xue-Ting Chen. (1991). “Probability limit state design method
for cold-formed thin-wall steel structures.” Journal of Building Structures, 12(3),
14-23.
Appendix. - Notation

kf : Coefficient of variation of geometric fabrication

km : Coefficient of variation of material property
kt : Modification factor
mR : Sample mean value of experimental resistance
n Number of specimens
Rd : Design value of the resistance
Rdc : Design value of resistance determined from the formulas provided by codes
RK : Characteristic value of the experimental resistance with a prescribed
survival probability
RKt : Characteristic value of the experimental resistance determined from Rmin
t
Rmax1 : Maximum value of experimental resistance
t
Rmax i
: The ith largest value of experimental resistance
Rmin : Minimum value of experimental resistance
Rt : Experimental resistance value
sR : Sample standard deviation of experimental resistance
SG : Characteristic value of effect of dead load
K

SQK : Characteristic value of effect of live load

 854

SWK : Characteristic value of effect of wind load

Vsc VR : The coefficient of variation of resistance
x1 : Variable representing the minimum value of experimental resistance
X (1) : Minimal Order Statistics of the experimental resistance

Xi : Experimental resistance for specimen i

 : Mean value of the logarithm of experimental resistance
 2 : Variance of the logarithm of experimental resistance
R : The mean value of resistance
 R 2 : The variance of resistance
 : Reliability index
 : Combination coefficient of different action
: Cumulative distribution function of the standardized Normal distribution
 G : Partial coefficient of dead load
 Q : Partial coefficient of live load
 R : Resistance partial coefficient
 Rt : Resistance partial coefficient determined from RKt
 W : Partial coefficient of wind load