Journal of Constructional Steel Research 70 (2012) 137–152

Contents lists available at SciVerse ScienceDirect

Journal of Constructional Steel Research

Development of component stiffness equations for bolted connections to

RHS columns
A.Y. Park, Y.C. Wang ⁎
School of Mechanical, Aerospace and Civil Engineering, University of Manchester, UK

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

Article history: This paper presents the derivation and validation of analytical equations for prediction of initial stiffness in
Received 27 October 2010 bolted endplate connections to RHS columns based on the component method of joint characterisation.
Accepted 2 August 2011 Analytical derivation of the equations is presented along with explanation of assumptions and simplifications
Available online 28 September 2011
used to reduce complexity. A comparative study with finite element simulation results shows that the
equations for the RHS in transverse tension component can be used for all realistic joint geometries with an
Keywords:
Initial stiffness
error of less than 10%. Validation against experimental results from France et al. (1999) show that when used
Component based method in conjunction with existing equations given in Eurocode 3 Part 1.8 for other joint components such as
Joint endplate and bolts, predictions of joint initial stiffness had an average 9.2% error and a maximum 21.4% error
Hollow section which is acceptable for practical use. This is a significant improvement over existing equations where range of
Analytical method validity is severely limited and there is large error in predictions.
© 2011 Elsevier Ltd. All rights reserved.

1. Introduction nections to enable using the Component Method approach. However,

The advantages of using inherent stiffness and bending moment
resistance of joints in steel framed structures to resist sway (semi-
rigid design) or to reduce beam size (partial strength design) are well
rehearsed. As a result, a large number of research studies have been
conducted over many years to develop methods to quantify joint
behaviour. Among a variety of methods that have been considered,
including experiments, finite element simulation, curve fitting, the
component based method has emerged as being most favoured
because this method combines the flexibility of being able to deal with
any change in joint detail yet sufficiently simple for implementation in
practical design.
In the component based method, a joint is represented by a small
number of components, each representing one part of the joint due to
one single action. The action of each component is characterised by a
force–displacement relationship. The force–displacement relationships
of all components are then assembled, based on satisfying the
equilibrium conditions of the joint, to give the desired joint character-
istic. For steel frame design under serviceability limit state of deflection
control or ultimate limit state of sway stability, the joint rotational initial
stiffness is of primary importance. To obtain this joint quantity, the
stiffness of the joint components should be quantified.
For joints between I-section beams and H-section columns, the
Eurocode EN 1993-1-8 [3] gives equations for characterisation of
force–displacement relationships of all components of bolted con-
⁎ Corresponding author.
E-mail address: yong.wang@manchester.ac.uk (Y.C. Wang).
despite advantageous structural properties of using tubular columns,
EN 1993-1-8 does not cover joints to tubular columns.
Developing a method to calculate the rotational stiffness of steel
beam to Rectangular Hollow Sections (RHS) is the objective of this
paper. The flat surface of RHS allows conventional endplate
connections to be used in conjunction with blind bolting systems
such as Flowdrill for easy construction [1]. This type of joint can
develop significant initial stiffnesses close to those of welded joints
without the associated cost of weld detailing. This type of joint will be
investigated in this paper.
Jaspart et al. [8] appear to be the only one to have conducted
research to develop an analytical method to quantify the initial
stiffness of steel beam to RHS column joint using bolts. However, their
method has a number of shortcomings, including limited range of
applicability, large errors, and complex equations. Further comments
will be made later in this paper on the error and applicability of the
method of Jaspart et al. As an illustration of the limited range of
applicability of this method,
Fig. 1 shows the limitations imposed by the method where joints
with more than two bolt rows fall outside of the acceptable geometric
range. According to these limitations, 19 out of 20 tests of France et al.
[5–7] will be outside the ranges of this method. Clearly these
limitations are too restricting for this method to be useful as a
potential design method.
The principle objective of the new research conducted in the
paper is to develop a new method of calculating the joint stiffness
that is much less restricting than the currently available method
from Jaspart et al. [8], can achieve much better accuracy, and is much
simpler to use.

0143-974X/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2011.08.004
138 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

where c = height of joint tension zone, Lstiff= width of RHS

Fig. 1. Example of limitations imposed by Jaspart et al. method (2004) against France et al. testing (1999) for RHS face in transverse tension stiffness component.

2. Derivation of formulae where

2.1. Notation mπb

am =
at Bolt hole separation 2a
a Equivalent yield width
b Height of RHS column For an infinitely long plate, b → ∞, and therefore cosh[am] → ∞ and
c Height of joint tension zone tanh[am] → 1. Eq. (1) reduces to:
dRHS Depth of RHS section
D Flexural rigidity of a plate 2



a P ∞ Sin mπc Sin mπx
E Young's modulus ðw1 Þy = 0 = ∑ a a
ð2Þ
2D π 3 m = 1 m3
kr Rotational stiffness of RHS sidewalls
Mkr Moment at sides due to stiffness of sidewalls, kr
P Bolt load Consider the two bolts on the same row as the joint component,
S Ratio of sum to infinity to first term the deflection at the centre of the plate, x = a / 2, should be used. This
tw Thickness of RHS web (connecting face) is the main difference between the new derivations and that of Jaspart
wRHS Width of RHS section et al. [8]. By treating each bolt row as one joint component, this allows
z Lever arm any number of bolt rows to be used in the joint. In contrast, in Jaspart
et al., it was assumed that all bolt rows in tension were considered
together as one combined joint component by means of the combined
2.2. Unfilled column height of the tension zone. The deflection at the centre of this tension
zone was then calculated. Not only was this representation of joint
Although the method of Jaspart et al. [8] has a number of component inflexible (e.g. not realistically able to deal with joints that
shortcomings as explained in the introduction section, it provides a have more than two bolt rows), it also made the calculation method
good starting point for deriving the equations for the stiffness of the RHS much more complicated and more difficult to obtain simplified
face in transverse tension due to bolt loads. In this study, as shown in Fig. analytical solutions.
2, the complex 3D column section is reduced to a 2D plate for the
connecting face, supported by rotational springs of magnitude kr
representing the contribution of stiffness by the RHS sidewalls.
As shown in Fig. 3, deflection, w, is obtained by considering the P P P P
deflection w1 of concentrated loads acting on a plate with simply
supported edges and then adding a negative deflection w2 due to a
kr kr
restraining bending moment at the sides imposed by the sidewall at at
stiffness.
The analytical solution for the deflection of a rectangular plate
dRHS a a
with simply supported edges subject to a single concentrated
transverse load is derived by Timoshenko [9] as the following:


mπx ! wRHS
a2 P ∞ Sin mπc
a Sin a am
ðw1 Þy = 0 = ∑ Tanh½ am − ð1Þ
2D π3 m = 1 m3 Cosh2 ½am Fig. 2. Simplification of 3D model to 2D plate model.
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152 139

Mkr Mkr
c1 =
(a-at)/2
b/2 b/2 b/2
c2 =
(a+at)/2
O

w1
X
+ O

w2
X
= O

w1+w2
X

b/2 at b/2 b/2

a a
Y
a Y Y

Fig. 3. Geometry of deflection components w1 and w2.

Assuming initial deflections occur in the elastic range, we can use the sum to infinity can be replaced by a factor, S, multiplied by the first
superposition to get the combined deflection due to two bolts (see Fig. term with a negligible loss in accuracy to give:
3), one at c1 = (a − at) / 2 and the other c2 = (a + at) / 2 with the load P
shared evenly between both bolts. To achieve this, we can modify Eq.   ha π i
6a2 P 1−v2 S Cos t
(2) to the following: 2a
ðw1 Þx = a = 2;y = 0 = ð8Þ

c mπ
c mπ
 E π3 tw
3
2
a P ∞ Sin 22a + Sin 22a Sin mπc
ðw1 Þy = 0 = ∑ a
ð3Þ
2D π3 m = 1 2 m3 To consider the w2 component, the deflection due to a bending
moment equivalent to the side wall rotational stiffness per unit
Substituting x = a / 2, c1 = (a − at) / 2, and c2 = (a + at) / 2 into Eq. length, kr, we take the solution in the following form:
(3), we can simplify the deflection due to both bolts at the centre of
the plate as: ∞ hmπyi
h i h i w2 = ∑ X2m Cos ð9Þ
ða−atÞmπ ða + atÞmπ
mπc m=1 a
a2 P ∞ Sin 2a + Sin 2a Sin
ðw1 Þx = a = 2;y = 0 = 3
∑ a
2Dπ m = 1 2 m3
where
As
mπx
mπx mπx −mπx mπx −mπx
X2m = Am e a
+ Bm e a + Cm e a + Dm e a ð10Þ
h i h i a a
Sin ða−a2at Þmπ + Sin ða + at Þmπ
2a
ha mπi hmπi
≡ Cos t Sin ;
2 2a 2 Weynand et al. [10] give the following solution for the coefficients.

      

mπ2
at mπ Am = aAwim emπ kr mπ 2emπ 2D + akr mπ + a −1 + e2mπ kr
a2 P ∞ Sin Cos
ðw1 Þx = a = 2;y = 0 = ∑ 2 2a
ð4Þ       2
2Dπ3 m = 1 m3
+ 2D 1 + e2mπ mπ Cos½mπ = 4D2 −1 + e2mπ m2 π2
 
As even terms of m give sin[mπ / 2] = 0, we can consider odd values − 4aDkr mπ 1−e4mπ + 4e2mπ mπ
only and reduce the above to   
+ a2 k2r 1 + e4mπ −2e2mπ 1 + 2m2 π2

mπ
a2 P ∞ Cos at2a
ðw1 Þx = a = 2;y = 0 = ∑ ð5Þ
2Dπ3 m = 1;3;5; :: m3      
Bm = − aAwim kr 2D  −1 + e2mπ mπ + akr 1 + e2mπ ð−1 + 2mπÞ
where      
+ emπ 2D −1 + e2mπ mπ + akr −1 + e2mπ −2mπ Cos½mπ 
  2  
E tw 3  4D2 −1 + e2mπ m2 π 2 −4aDkr mπ 1−e4mπ + 4e2mπ mπ
D=   ð6Þ
12 1−v2   
+ a2 k2r 1 + e4mπ −2e2mπ 1 + 2m2 π2
Inserting Eq. (6) into Eq. (5), we get
    
mπ mπ 2mπ
  Cm = − aAwim e kr mπ 2e ð2D + akr Þmπ + a −1 + e kr

at mπ
6a2 P 1−v2 ∞ Cos      2
ðw1 Þx = a = 2;y = 0 = ∑ 2a
ð7Þ + 2D 1 + e2mπ mπ cos½mπ = 4D2 −1 + e2mπ m2 π2
E π3 tw 3 m = 1;3;5 m3
 
4mπ 2mπ
−4aDkr mπ 1−e + 4e mπ
As the sum does not converge, the contribution of each term has   
+ a2 k2r 1 + e4mπ −2e2mπ 1 + 2m2 π2
been investigated in the following section and it can be shown that
140 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

    
Dm = − aAwim emπ kr emπ 2D −1 + e2mπ mπ from Eq. (15), we get
    
+ akr −1 + e2mπ −2mπ + 2D −1 + e2mπ mπ 
   tw3 11:5
kRHS;concrete filled = a π ð16Þ
+ akr 1 + e2mπ ð−1 + 2mπÞ Cos½mπ a2 Cos½ 2a
t
 2:024 S−1
  2  
 4D2 −1 + e2mπ m2 π2 −4aDkr mπ 1−e4mπ + 4e2mπ mπ
   2.4. H-section column minor axis
+ a2 k2r 1 + e4mπ −2e2mπ 1 + 2m2 π2
Due to the similar nature of H-section webs in transverse tension,
the equation for RHS face in transverse tension is also valid for bolted
⌊ ⌋
2 cmπ
a PSin
Awim = a joints to H-section column minor axes by replacing rotational stiffness
2D m3 π3 contribution of the RHS sidewalls kr, with rotational stiffness
contribution of the H-section flanges with the following
where

4 E I 1:5wRHS + dRHS 2EI
kr = kr = ð17Þ
dRHS 2:0wRHS + dRHS D

Derivation of kr is given in Appendix A.

The same equations will be used. However, since one bolt row is a Derivation is given in Appendix A.
joint component, x = a / 2 and y = 0. This allows considerable
simplifications to be made to Eq. (9) to give the following solution: 2.5. Determining correction factor S for convergence of summation

ðw2 Þx = a = 2;y = 0 To quantify the ratio of sum to infinity and the first term of the sum
 2 mπ   h i
18a3 kr P −1 + v2 ∞
e 2 −1 + emπ ð−1 + Cos½mπÞ Sin ða−aat Þmπ (S in Eqs. (14) and (15), for calculating deflection w1), a range of
= ∑m = 1 2    
Eπ2 t3w m ð1+emπ Þ2 Emπtw3−6akr −1+e2mπ +2emπ mπ −1+ v2 bolthole separation to clear column face width ratios, at/a, were
ð11Þ plotted. As displayed in Fig. 4, there is a relationship for the ratio that
can be represented by a simple quadratic equation with the
It can be established that there is a negligible contribution from terms coefficients optimised for usage within the practical joint geometry
m larger than 1. Therefore by taking the first term only, Eq. (11) can be range of 0.2 b at/a b 0.8. This equation is:
simplified to:
  h i S = 0:143ðat =aÞ−0:306ðat =aÞ + 1:076 ð18Þ
3 π = 2 π 2 2
36a e −1 + e kr P −1 + v Sin ða−at
a
Þπ

ðw2 Þx = a = 2;y = 0 = −     
Eπ 2 t3w ð1 + eπ Þ2 Eπt3w −6akr −1 + e2π + 2eπ π −1 + v2

 2 As shown in Fig. 5, for the practical range 0.2 b at/a b 0.8, the error is

π
a3 kr P −1 + v2 Cos a2a
t less than 0.07% by replacing the summation of terms to infinity by a
ðw2 Þx = a = 2;y = 0 =−    ð12Þ single term. This leads to a reduction of overall complexity and thus
E tw3 4:7E tw3 −10:5akr −1 + v2
makes it less susceptible to calculation error.

Combining Eq. (8) and (12) to get the final deflection, w:

2.6. Comparison of full and simplified equations
 ha π i  
π
2 2 3 2 2
6a P 1−v S Cos t a kr P −1 + v Cos a2a
t

ðw2 Þx = a = 2;y = 0 = 2a −    A comparison of the full and simplified equations derived in the
E π3 tw 3 E tw3 4:7E tw3 −10:5a kr −1+ v2 previous section is made to determine the acceptability of the
ð13Þ simplifications made by checking the error in predictions. For the
comparison, the following equations for the deflection are used
To simplify further, we can take v = 0.3 to get: where full refers to the analytically derived equation without
ha π i approximations and simplified refers to the full equations which
5:46a2 P S Cos t
π has been simplified by means of approximations and assumptions for
2a − a3 kr P Cos a2a
t

ðw2 Þx = a = 2;y = 0 =   common usage (i.e. v = 0.3).

E π3 tw 3 E tw 11:5a kr + 5:7E tw3
3

ha π i !
2
a P Cos t 3
ðw2 Þx = a = 2;y = 0 = 2a − 2:024a kr S−a kr E S tw ð14Þ
E tw 3 11:5a kr + 5:7E tw3

Taking the reciprocal of this value and dividing by E/P to get the
stiffness in the component method format,
!
3 3
tw 11:5a kr + 5:7E tw
kRHS;unfilled = ha π i ð15Þ
a2 Cos t 2:024 a kr S−a kr + E S tw3
2a

2.3. Concrete-filled column

For the concrete-filled column case, we can assume that the

concrete infill prevents rotation at the sidewalls. Therefore, kr = ∞, Fig. 4. Ratio of sum of odd terms to infinity over the first term of the sum.
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152 141

M M
C C

d d

r
r
A B A B
r = 2tw
Fig. 5. Error between full sum and factored first term.
w/2 w/2
Simplified cross section Actual cross section
Full equation: Fig. 7. Simplified and actual RHS cross sections.


  atmπ
6a2 P 1−v2 Cos
∞ 2a
∑m = 1;3;5;::
E π 2 tw3 m3
 2 mπ   h i
18a3 krP −1 + v2 e 2 −1 + emπ ð−1 + Cos½mπ Þ Sin ða−ata Þmπ
addition to this, the relationship in errors of kr calculation on the
∞
− 2 3
∑m = 1  mπ 2
   overall stiffness calculation is investigated to determine if the errors
Eπ tw m ð1+ e Þ Emπtw −6akr −1 +e
2 3 2mπ
+ 2e mπ −1 + v2
mπ

from usage of the simplified model for derivation of kr is significant

(Fig. 7).
Simplified equation:
The difference in cross section geometry is shown in Fig. 7. For the

 actual cross section geometry, the root radius of the RHS corners was
2 atπ !
a P Cos assumed to be equal to twice the RHS thickness.
2a 2:024a kr S−a kr + E S tw3
This validation work is divided into three components. Firstly, the
E tw3 11:5a kr + 5:7E tw3
analytical equations derived by Jaspart et al. are validated against FEM
results for the simplified section to determine if FEM provides an
Fig. 6 shows the errors that arise from usage of the simplified accurate calculation of the RHS sidewall stiffness. Secondly, the same
equation in comparison to the full equation for various values of at/a. FEM techniques are used for the actual cross section model and the
It can be seen that there is up to approximately 0.6% error for very errors in using the analytical equation based on the simplified model
small and large values of at/a, however for the practical range of joint are determined. Finally, the total error on the prediction of the bolt
geometries mentioned in the following section, 0.2 ≤ at/a ≤ 0.8, the row deflection caused by an error in the prediction of kr is quantified
simplified equations gives the solution to within 0.4% error of the full for the realistic minimum and maximum limits of the geometric ratio
analytical equations. In general, it can be accepted that the error from at/a, 0.2 and 0.8.
simplifications and approximations is insignificant. The analytical equation was validated against FEM results obtained
from a 2D beam analysis using Abaqus 6.7-1 [4]. Geometry and
2.7. Validating sidewall stiffness simplification against ABAQUS boundary conditions are exactly as shown in Fig. 7. A reference bending
modelling moment of 100 Nm was applied for calculation of the stiffness. The

As presented in Appendix A, the derivation of the sidewall spring

stiffness coefficient, kr, is conducted using a simplified 2D model
which neglects the curvature of the RHS corners. It is necessary to
determine the error in calculation of the kr coefficient caused by using
this simplified model compared to the actual section with curved
corners. This is especially true for thick columns which have a larger
root radius and therefore deviates more from the simplified model. In

Fig. 6. Error between full and simplified equations for various at a values. Fig. 8. Geometry of t-stub test T1 [2].
142 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

Table 1
Errors from usage of simplified cross section for kr derivation.

Geometry RHS width variation RHS depth variation RHS thickness variation

wRHS (mm) 200 300 400 200 200 200 200 200 200 200 200 200
dRHS (mm) 200 200 200 200 300 400 200 200 200 200 200 200
tw (mm) 10 10 10 10 10 10 4 5 8 10 12.5 16
E (1 × 108 N/m2) 210 210 210 210 210 210 210 210 210 210 210 210
I (1 × 10−8 m4) 8.33 8.33 8.33 8.33 8.33 8.33 0.533 1.04 4.27 8.33 16.4 34.1
Root radius (mm) 20 20 20 20 20 20 8 10 16 20 25 32
a (mm) 160 260 360 160 260 360 184 180 168 160 150 136
a/tw 16 26 36 16 16 16 46 36 21 16 12 8.5

Analytical Eqn.
Stiffness (kNm/rad) 291.7 284.4 280.0 291.7 200.0 153.1 18.7 36.5 149.3 291.7 569.7 1194.7

FEM simplified
Stiffness (kNm/rad) 290.8 283.6 279.3 290.8 199.7 153.0 18.7 36.4 149.1 290.8 567.0 1185.5
kr Error 0.3% 0.3% 0.2% 0.3% 0.1% 0.1% 0.0% 0.1% 0.2% 0.3% 0.5% 0.8%

FEM curved
Stiffness (kNm/rad) 289.1 279.3 284.1 289.1 203.6 157.1 19.4 37.9 151.3 289.1 556.0 1136.1
kr error 0.9% 1.8% − 1.4% 0.9% − 1.8% − 2.5% − 4.0% − 3.8% − 1.3% 0.9% 2.5% 5.2%
Total error
– for at/a = 0.2 0.02% 0.11% − 0.15% 0.02% − 0.08% − 0.18% − 0.14% − 0.12% − 0.04% 0.02% 0.06% 0.10%
– for at/a = 0.8 0.02% 0.10% − 0.13% 0.02% − 0.07% − 0.16% − 0.12% − 0.11% − 0.04% 0.02% 0.05% 0.09%

Positive error = Stiffness overestimation; Negative error = Stiffness underestimation.

investigation was separated into three sections: variation in RHS width,
variation in RHS depth, and variation in RHS thickness.
The results from this investigation are presented in Table 1. Firstly,
it can be seen there is a negligible error in the prediction of kr between
the analytical equation and the simplified FEM models. This shows
that the FEM techniques are suitable for modelling the sidewall
stiffness for the actual cross section. When comparing the analytical
equation against the FEM model with curved edges however, there is
a larger error for prediction of the kr coefficient. The largest errors
were found for the lower bound and upper bound variations in the
RHS thickness which had errors of −4.0% (4 mm thickness) and 5.2%
(16 mm thickness) giving a range of error of approximately ±5%. This
coincides with the upper and lower bound values of the a/tw ratio of
46 and 8.5 although errors of up to −2.5% are also found in the
variation of RHS depth showing that other factors also affect the
calculation of kr.
We can put this into perspective by determining the effect that
errors in kr has on bolt row deflection calculation. The effect of kr error
on bolt row deflection error was investigated for the two realistic
minimum and maximum cases of at/a = 0.2 and 0.8 to give a
generalised understanding of the error across the realistic range of
geometric parameters. It can be seen that in all cases the error in kr
calculation does not have a significant effect on the bolt row deflection
calculation. For the RHS sections investigated, the maximum error
was 0.16% with the majority of sections having less than 0.1% error. It
can be concluded that the simplified cross section used for derivation
of the kr is acceptable for all practical section sizes.
2.7. Conclusion
In this section, derivation of the RHS face in transverse tension and
H-section web in transverse tension is presented. Although partly
based on work in Jaspart et al. [8] and Weynand et al. [10], the
approach used in this study of calculating stiffness on a per bolt row
basis overcomes the shortcomings of equations presented in Jaspart et
al. [8] which calculates stiffness as a group of bolt rows while severely
limiting the range of applicability to joints with two bolt rows
Various simplifications are applied to the derived equations and
then checked for consistency against the results of the full equations.
It was found that there were errors of less than 0.4% in the realistic
range of joint geometries while greatly improving the simplicity of the
equations and thus the likelihood of calculation errors during the
design process. Checks were also conducted on the geometrical
simplifications used by Weynand et al. [10] on the derivation of the
sidewall stiffness coefficient. It was found that errors of up to
approximately 5% exist by using the simplified equations for sidewall
stiffness, however this level of error has a negligible effect on the
overall error of the RHS stiffness component.
3. Parametric study of component stiffness equations using FEM
3.1. Introduction
The equations derived for the initial deflection of bolt rows in
tension are investigated across a variety of joint geometries to
determine the variation in their validity and their range of
applicability. From this it is possible to determine restrictions on
their applicability to extreme cases, namely for small and large values
of the ratios at/a and tw/b for which there may be noticeable
differences between actual and predicted behaviour. This is especially
true for thick sections with large values of tw/b for which actual
behaviour may deviate significantly from equations based on thin
plate theory.
3.2. Validation of ABAQUS simulation
To demonstrate that the FEM simulations conducted in this
section are accurate, this section compares FEM results with test
results for a t-stub assembly in tension by Bursi and Jaspart [2] using
the general finite element analysis software ABAQUS 6.7-1 [4]. While
the t-stub and hollow section column is not a direct comparison,
there is similarity in geometrical features and the deformed shape
that makes it a suitable comparison. The non preloaded T-stub
connexion test T1 as shown in Fig. 8 was used as a reference for this
purpose.
Due to the validation of the initial stiffness equations being
of interest in this section, the validation of the modelling focused on
the accuracy of the load–displacement behaviour in the elastic range.
For this reason, various simplifications were made to the model
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152 143

A comparison of the load–displacement curves can be seen in Fig.

10 where it is clear that there is an exact replication of displacement
behaviour replicated with the FEM for the elastic range where there is
linear load–displacement relationship. While there is some discrep-
ancy at higher loads, this is both expected and insignificant as
convergence in the early deflection range is of interest and plastic
behaviour was not accounted for in the material properties. The
modelling techniques used in this section will form a basis for the
parametric study in the following section of this paper.

3.3. Validation of RHS face in transverse tension component against

ABAQUS simulation

To investigate the validity of this component, the general finite

element software ABAQUS was used to simulate an unfilled RHS
column with a single bolt row in tension. Validation of the simulations
was assessed by comparing results against a t-stub test conducted by
Bursi and Jaspart [2] which is presented in Section 4.2. Fig. 11 shows
the FEM model for the joint component under consideration.
To simplify the model, the bolt loads are applied to reference nodes
that are coupled to the bolthole surface to simulate flowdrill bolt
behaviour. A column length of 1 m is used with ends fully fixed and
this is reduced to a quarter model with appropriate symmetry
boundary conditions to reduce model complexity. The model is
discretised using the 8-node cubic brick element with reduced
integration as described in the previous section. The relative
deflection of the column face at the centre is obtained by taking the
deflection of the column face at the centre and subtracting the
deflection at the sidewalls.
Fig. 12 gives a plot of errors in prediction of deflection for varying
Fig. 9. One-eighth model and mesh of t-stub test T1.
values of at/a across different column sections and bolthole sizes. In
general, the analytical equations provide predictions of initial
deflections across different column configurations with generally
less than 10% error in the practical range of at/a. The ratio at/a is
including using only elastic material properties and very small
physically limited to a minimum of approximately 0.1–0.15 and a
loading.
maximum of 0.8–0.85 (depending on column thickness). For the two
The main parameters that needed validation were the type of
cases involving the column with a = 0.136; tw = 0.016 and thus a
mesh element, mesh discretisation, and use of symmetry to reduce
relatively small a/tw value of 8.5, there is greater error in deflection
the model complexity. The 8-node linear brick with reduced
predictions at higher values of at/a. The largest error is approximately
integration was chosen for all components of the model while the
15% at at/a = 0.8 for a/tw = 8.5. It is also clear that there is a slight
mesh was discretised with 4 mesh layers across the edge of the t-stub
deviation in predictions when different bolt hole sizes are used for a
thickness for sections in bending. The bolt and bolthole perimeters
given column but the difference is insignificant.
were discretised using 32 elements along their edges. The actual mesh
From this parametric study it is possible to conclude that the
used can be seen in Fig. 9. The point load, F, was applied to a reference
analytical equations provide a good prediction of initial deflections
node which was then coupled to the web section. To reduce model
throughout different column configurations with generally less than
complexity the t-stub was subdivided across 3 planes and applied
10% error in the range of at/a which is physically possible and for the
the relevant symmetry boundary conditions to give a one-eighth
range of a/tw values for standard RHS column sections which are
model.
commonly available.
Table 2 gives a summary and comparison of geometrical limits
against those given by Jaspart et al. [8]. The last restriction of Jaspart
c
et al., 0.05 b b 0.20, is particularly limiting as shown in Fig. 1 as it
Lstiff
realistically limits the range of validity to joints with only two bolt
rows. Because of treating each individual bolt row as one joint
component, the new derivations in this paper have eliminated this
limitation.

3.4. Validation of H-section web in transverse tension component

against ABAQUS simulation

To investigate the validity of this component, as in the last

section, the general finite element software ABAQUS was used to
simulate a H-section column web with a single bolt row in tension as
found in minor axis joints to H-sections. Validation of the simulations
Fig. 10. Load–displacement comparison of t-stub test T1 and FEM. was assessed by comparing results against a t-stub test conducted by
144 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

Coupling to inner bolt

hole surface

Reference point

Fig. 11. FEM model for RHS bolt row in transverse tension component (quarter model).

Fig. 12. Error in predictions of deflection at various levels of at/a against FEM results for RHS face in transverse tension.

Bursi and Jaspart [2] which is presented in Section 4.2. Fig. 13 shows taking the deflection of the web face at the centre and subtracting the
the FEM model for the joint component under consideration. deflection at the web edges.
The same simplifications and approach to developing the FEM Fig. 14 gives a plot of errors in prediction of deflection for varying
were used as in the previous section for the RHS column validation. values of at/a across different column sections and bolthole sizes. In
The relative deflection of the column face at the centre is obtained by general, the analytical equations provide predictions of initial
deflections across different column configurations with generally
less than 25% error in the practical range of at/a. The ratio at/a is
physically limited to a minimum of approximately 0.1–0.15 and a
maximum of 0.8–0.85 (depending on flange thickness and section
Table 2
depth).
Range of validity for proposed equations.
From this parametric study it is possible to conclude that the
Geometric ratio Proposed equations Equivalent in Jaspart analytical equations provide a prediction of initial deflections
(2004)
throughout a variety of column configurations with generally less
Equivalentyieldwidth 12 b a/tw b 27.7* (b10% error) Lstiff than 25% error in the range of at/a which is physically possible.
10 b b 50
Thickness 8.5 b a/tw b 27.7* (b 15% error) tc
*Upper bound may be
increased with further study
4. Validation of stiffness equations against France testing (1999)
but little practical usage
Boltholespacing 0.2 b at/a b 0.8 (b 10% error)
0.08 b
b
b 0.75
4.1. Calculation of joint stiffness
Equivalentyieldwidth 0.1 b at/a b 0.8 (b 15% error) Lstiff
Tensionzoneheight N/A c
0.05 b b 0.20 France et al. [5–7] carried out some tests on bolted endplate
Equivalentyieldwidth Lstiff
connections to empty and concrete filled rectangular hollow sections
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152
Fig. 13. FEM model for H-section web in transverse tension component (quarter
model).
using Flowdrill bolts. In this section, these tests will be used to assess
accuracy of the new equations for calculating the component stiffness
for the tube under transverse tension.
One of the assumptions made is that the RHS in transverse
compression component is assumed to be near infinite for initial
stiffness and is therefore not included as an effective component in
the approach proposed in this research. This is believed to apply for all
practical joints in which a high level of stiffness is desired, as the
width of the compression zone will typically be equal to or greater
than the deformable width of the RHS yield area as shown in Fig. 15.
This means that the majority of the loading will be transferred directly
to the RHS sidewalls which have near infinite stiffness. This effect is
even stronger when the endplate thickness is greater than the RHS
thickness which will usually be the case.
Another assumption made in the derivation of the RHS stiffness
equations is that deflections are taken at the centre of the RHS section
as opposed to the centre of the bolt holes. It can also be argued that the
endplate will have a constricting effect on this RHS face and thus
Fig. 14. Error in predictions of deflection at various levels of at/a against FEM results for H-section web in transverse tension for use with minor-axis joints.
145
Compression zone width
RHS yield width
Fig. 15. Example of joint with compression zone wider than RHS yield area width.
deflections will be limited to that of the bolt holes. For this reason, a
comparison of results using both approaches is given in Section 4.3.
For the unfilled columns, all bolt rows except for the row closest to the
flange on the compression side of the beam are assumed to be in tension
as in Eurocode 3 Part 1.8 [3]. For the concrete-filled column, all bolt rows
above the beam flange in compression are considered to be in tension for
the calculation of initial stiffness. This is due to the concrete-filling
preventing deflections in the compression zone and thus lowering the
centre of rotation to that of the beam flange in compression.
Fig. 16 shows how the joints tested by France et al. [5–7] is
represented by the component based method if using the new
derivations of the paper. Each individual bolt row is treated as a tension
component. As has been explained, this approach is different to that
of Jaspart et al. [8] who treated the entire bolt group as one combined
column tension zone. Because of this change, all tests of France et al.
[5–7] can be assessed using the new derivations as opposed to the
Jaspart et al. equations which practically limits the applicability to joints
with two bolt rows (i.e. one bolt row in tension) of which there is only
one in the France tests. The RHS column in compression component is
considered insignificant for initial stiffness calculation and is therefore
not included in the approach proposed in this study.
Thus, incorporating equations already existing in EC 1993–1.8 for
joint components related to the endplate and bolts, the effective
components of bolted endplate joints to RHS columns for joint initial
stiffness calculation are given in Table 3.
4.2. Results
Table 4 gives a summary of the results obtained from application
of the above equations to all 20 tests by France et al. [5–7] excluding
those with partial endplates, of which there are 14 joints with flush
146 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

Fig. 16. Differences in approach for component based method of bolted endplate joints to RHS columns.

endplates and 6 joints with extended endplates. Of these, 14 joints carried out as if the Jaspart et al. equations could be applied to all 20
are to unfilled columns and 6 joints are to concrete-filled columns. France tests. These calculation results are shown in Table 3 and they
Joint initial stiffness is obtained from the secant stiffness of the origin gave an average of approximately 40% and maximum 80% error in
and first data point of the moment–rotation curves, as shown in Fig. prediction.
17.
As shown in Table 4, the prediction of joint initial stiffness when 4.3. Alternative joint stiffness formulations
applying the proposed component method equations gives an average
error of 8.9% and a maximum of 18.8%. This level of error is similar This section investigates the applicability of the assumptions given
in magnitude to that found in the validation and parametric in Section 4.1 regarding the formulation of the joint stiffness
study of the RHS in transverse tension component conducted with equations. This includes determining whether the location of
FEM. This validation study suggests that the equations derived for the deflection for derivation of the RHS face in transverse tension stiffness
RHS in transverse tension component for determining initial joint component should be taken at the centre of the RHS face or at the bolt
stiffness and the selection of effective components is suitable for holes and whether usage of the RHS in transverse compression
application to both flush/extended endplates to unfilled/concrete- component is necessary due to the fact that the compression zone is
filled columns covering a range of geometries. For comparison, generally equal to or larger than the RHS yield width in practical joints
although the Jaspart et al. equations are outside the range of validity and therefore is likely to have very high stiffness.
for the c/Lstiff ratio for most of the France tests, calculations were The following approaches were considered for the joint stiffness
formulations:
Table 3
Summary of effective components for calculating initial stiffness. (a) Taking deflections at centre of RHS face, infinite compression
zone stiffness
1) Bolts in tension (EN 1993-1-8)
(b) Taking deflections at centre of RHS face, with compression zone
k2, indv = 1.6As/Lb stiffness*
2) Endplate in bending (EN 1993-1-8) (c) Taking deflections at bolt centres, infinite compression zone
stiffness
k2, indv = 0.9lefftp3/m3
(d) Taking deflections at bolt centres, with compression zone
3) RHS in transverse tension*
stiffness*


3
f 1 tw
kRHS;indv =
atπ *Infinite compression zone stiffness is used for joints with
a2 S cos
2a
concrete-filled RHS
where
! The following are the equations that are derived for these
11:5akr + 5:7Etw3 11:5 approaches:
f1 = 3
for unfilled or for concrete-filled columns
2:024akrS−akr þ EStw 2:025S−1
RHS in transverse tension, taking deflections at bolt holes.
S = 0.143(at/a)2 − 0.306(at/a) + 1.076

This is obtained by using x = (a − at) / 2 in Eqs. (3) and (11) of
4EI 1:5wRHS + dRHS 2EI
kr = for RHS; for H-section minor axis Section 2.2.
dRHS 2:0wRHS + dRHS B
0

1
ða−at Þmπ ða−at Þmπ
Equivalent component stiffness from individual bolt row stiffness (EN 1993-1-8) Cos Cos
B6a2 Pð1v2 Þ ∞
4a 2a C
ðwÞx = ða−at Þ = 2;y = 0 =B
@ E π 3 tw3 ∑ C
A
keq = ∑ kindvhr/zeq m¼1;3;5;:: m3

where   at π at π ha π i
ða−at Þπ
 
∑r keff;r hr
2 18a2 e 2a −at 1 + eπ −1 + e a + að−1 + eπ Þð1 + e a Þ krPð−1 + v2 Þ2 Cos t
2a
zeq = −    
∑r keff;r hr Eπ2 tw3 ð1 + eπ Þ2 Eπtw3 −6akr −1 + e2π + 2eπ π −1 + v2

Joint initial stiffness (EN 1993-1-8)

Ez2 RHS in transverse compression, taking deflections at centre of RHS face.
Sj;ini =
∑keq
This is derived by taking deflections of a uniform distributed load
Where lever arm, z, is as defined in Fig. 6.15 of EN 1993-1.8.
with width and height equal to the compression zone dimensions
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152 147

Table 4
Comparison of initial stiffness predictions and values from France et al. (5–7) testing.

Endplate type With concrete Test result Jaspart et al. Error (%) Using new equations Error (%)
infill (kNm/rad) equations (kNm/rad)

Test 2 Flush – 12,500 7247 42.1% 14,358 − 14.9%

Test 4 Flush – 5200 3342 35.8% 4942 5.0%
Test 5 Flush – 3700 2646 28.6% 4179 − 12.9%
Test 6 Flush – 6200 4883 21.4% 5793 6.6%
Test 7 Flush – 2120 2388 − 12.5% 2173 − 2.5%
Test 8 Flush – 23,000 6222 73.1% 26,227 − 14.0%
Test 10 Flush – 1960 1685 14.2% 1648 15.9%
Test 18 Flush – 5350 3538 34.0% 5417 − 1.3%
Test 25 Flush – 17,500 7911 55.1% 16,796 4.0%
Test 26 Flush – 7700a 6119 21.0% 8038 − 4.4%
Test 19 Extended – 31,000b 16,061 32.2% 25,793 16.8%
Test 20 Extended – 60,000 20,827 65.6% 53,873 10.2%
Test 21 Extended – 125,000 25,922 79.5% 146,478 − 17.2%
Test 23 Extended – 65,000 20,825 68.0% 55,392 14.8%
Test 14 Flush Yes 25,000 26,019 − 4.1%
Test 15 Flush Yes 9700 10,141 − 4.5%
Test 16 Flush Yes 4500c 4697 − 4.4%
Test 17 Flush Yes 3700 4492 − 21.4%
Test 22 Extended Yes 100,000 104,185 − 4.2%
Test 24 Extended Yes 100,000 104,119 − 4.1%
Average 41.7% Average 9.2%
Maximum 79.5% Maximum 21.4%

Notes
Positive error for underestimation of stiffness, negative error overestimation of stiffness.
Average error is calculated as the mean value of absolute errors.
a
Initial portion of curve lacks smoothness. 1st data point gives near infinite stiffness, 2nd–4th data points give 27,000 which is much larger than Test 25 which is identical except
for a thicker column section. 4th–5th data points give 7700 so use this value.
b
First 3 data points give near infinite stiffness so use 4th–5th data points.
c
Adjusted from 11,000 kNm/rad. Should be much less than Test 15 stiffness so first data point ignored and stiffness calculated from subsequent data points to get 4500 kNm/rad.

with simply supported edges as in Fig. 18. The contribution of the Note: v = height of compression zone, v = shear modulus
sidewall stiffness was then applied as in the derivation in Appendix A For an infinitely long column (i.e. b = ∞) and taking x = a / 2, y = 0,
for the unfilled column. The width of the simply supported plate is q = P / (u × v) we get
equal to the RHS yield width. Due to the inherent difficulty in deriving
4 2

 mπv!
analytical equations for cases where the compression zone width is 48a Pð1−v Þ ∞ Sin mπu
ðw1 Þx = a=2;y = 0 = ∑ 2a 4a

larger than the RHS yield width, compression zone width was taken as Eπ5 tw3 u v m = 1;3;5;:: m3
equal to the RHS yield width for such cases.
The analytical solution for the deflection of a rectangular plate Adding the contribution of the sidewall stiffness, w2 from Eq. (12)
with simply supported edges subject to a uniform distributed load is of Section 2.2,
derived by Timoshenko [9] as the following:
 mπv!
48a4 Pð1−v2 Þ ∞ Sin mπu
hmπui hmπxi
 ðw1 Þx = a=2;y = 0 = ∑ 2a 4a
w1 =
4qa4 ∞
∑
ð
dπ 5 m = 1;3;5;::
ð−1Þðm−1Þ = 2
m5
Sin
2a
Sin
a ð 1−
Cosh mπy
a
Cosh½am ðCosh½am−2ym Eπ 5 tw3 uv m = 1;3;5;:: m3

t π
hmπyi a3 kr Pð−1 + v2 Þ2 Cos a2a
+ ym Sinh½am−2ym + am
Sinh½2ym
2 Cosh½am Þ +
cosh½am−2ym mπy
2 Cosh½am a
Sinh
a ÞÞ −  
Etw3 4:7 E tw3 −10:5akr −1 + v2


where
RHS in transverse compression, taking deflections at bolt holes
mπb mπv This is derived in a similar manner to above taking the deflection of
am = ; ym =
2a 4a a rectangular plate with simply supported edges subject to a uniform
distributed load at x = (a − at) / 2 and then adding the contribution of
the sidewall stiffness at x = (a − at) / 2 in Eq. (11) of Section 2.2.

 !
4
48a Pð1−v Þ ∞
2 ðm−1Þ = 2
ð−1Þ hmπui mπða−at Þ mπv
ðwÞx = a=2;y = 0 = ∑ Sin Sin
Eπ 5 tw3 uv m = 1;3;5;:: m5 2a 2a 4a

  ha π i
ða−at Þπ   at π at π
18a2 e 2a −at 1 + eπ −1+ e a + að−1 + eπ Þð1 + e a Þ krPð−1 + v2 Þ2 Cos t
2a
−    
Eπ2 tw3 ð1 + eπ Þ2 Eπtw3 −6akr −1 + e2π + 2eπ π −1 + v2

These approaches are used to calculate joint stiffness in the joints

Fig. 17. Example calculation of initial stiffness from joint moment–rotation curve.
tested by France et al. [5–7] to determine the validity in the
assumptions used and results are presented in Table 5.
There is a clear distinction between the accuracy of approaches
used. Approach (a) that takes deflections at the centre of the RHS face
and an infinite compression zone stiffness, is the most accurate with
148 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

Mkr Mkr

b/2 b b/2

v
O

w1
X
+ O

w2
X
= O

w1+w2
X

b/2 u b/2 b/2 u

a a
Y
a Y Y

Fig. 18. Simplified 2D plate model for compression zone deflection derivation.

both lowest average and maximum errors. Also, there is no trend of
stiffness overestimation or underestimation. Approach (c) which
takes deflections at the bolt holes show significant stiffness over-
estimation in all cases while Approaches (b) and (d) which
incorporate the compression zone stiffness have stiffness underesti-
mation in all cases. For this reason, it is believed that Approach (a)
which is the basis for the assumptions given in the previous section, is
appropriate.
5. Conclusion
This paper presents a new derivation of equations for the
stiffness of RHS bolt rows in transverse tension. In this method,
each bolt row on the RHS face is treated as a joint component and
its stiffness under transverse tension is derived. For calculating the
joint rotational stiffness, this paper has demonstrated that the RHS
tension zone deflection should be calculated at the centre between
the bolts and the compression zone should be assumed to possess
infinite stiffness. Combining this with existing equations for bolts in
tension and endplate in bending given in EN 1993-1-2 [3], it is
possible to predict the initial stiffness of bolted endplate connec-
tions to RHS column sections. Improvements over the existing
equations of Jaspart et al. [8] include widely increased range of
validity, accuracy of predictions and simplicity in implementation.
Validation against experimental testing by France et al. [5–7] shows
that the proposed approach gives predictions with an average of
9.2% and a maximum 21.4% error. This is considered acceptable for
practical usage including serviceability limit state design of semi-
continuous frames, because the frame sway deflection will not be
particularly sensitive to the above level of change in joint rotational
stiffness. The new method for calculating the joint rotational
stiffness is summarised in Table 3.

Table 5
Comparison of initial stiffness predictions using different approaches.

With concrete France test result (a) Error (b) Error (c) Error (d) Error
infill (kNm/rad)

Test 2 – 12,500 − 14.9% 71.8% − 82.3% 56.1%

Test 4 – 5200 5.0% 69.8% − 51.6% 52.7%
Test 5 – 3700 − 12.9% 58.8% − 82.9% 44.4%
Test 6 – 6200 6.6% 73.4% − 43.5% 48.2%
Test 7 – 2120 − 2.5% 68.7% − 70.1% 52.4%
Test 8 – 23,000 − 14.0% 65.6% − 78.7% 38.9%
Test 10 – 1960 15.9% 60.8% − 36.0% 38.1%
Test 18 – 5350 − 1.3% 69.8% − 71.4% 52.0%
Test 25 – 17,500 4.0% 79.6% − 84.5% 52.9%
Test 26 – 7700a − 4.4% 76.3% − 98.0% 50.1%
Test 19 – 31,000b 16.8% 86.0% − 57.7% 70.2%
Test 20 – 60,000 10.2% 85.9% − 72.4% 66.9%
Test 21 – 125,000 − 17.2% 83.5% − 122.4% 52.0%
Test 23 – 65,000 14.8% 86.6% − 63.6% 68.6%
Test 14 Yes 25,000 − 4.1% − 4.1% − 4.6% − 4.6%
Test 15 Yes 9700 − 4.5% − 4.5% − 5.1% − 5.1%
Test 16 Yes 4500c − 4.4% − 4.4% − 4.5% − 4.5%
Test 17 Yes 3700 − 21.4% − 21.4% − 22.1% − 22.1%
Test 22 Yes 100,000 − 4.2% − 4.2% − 18.0% − 18.0%
Test 24 Yes 100,000 − 4.1% − 4.1% − 17.9% − 17.9%
Average 9.2% 54.0% 54.4% 40.8%
Maximum 21.4% 86.6% 122.4% 70.2%

Notes
Positive error for underestimation of stiffness, negative error overestimation of stiffness.
Average error is calculated as the mean value of absolute errors.
a
Initial portion of curve lacks smoothness. 1st data point gives near infinite stiffness, 2nd–4th data points give 27,000 which is much larger than Test 25 which is identical except
for a thicker column section. 4th–5th data points give 7700 so use this value.
b
First 3 data points give near infinite stiffness so use 4th–5th data points.
c
Adjusted from 11,000 kNm/rad. Should be much less than Test 15 stiffness so first data point ignored and stiffness calculated from subsequent data points to get 4500 kNm/rad.
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152 149

Appendix A. Formation of stiffness matrix equations for derivation of kr

A1. Derivation of kr for rotational stiffness contribution of RHS sidewalls

This subsection gives the derivation of the kr coefficient for rotational stiffness contribution of RHS sidewalls that is used in the stiffness
characterisation of bolted endplate joints to RHS sections. The kr coefficient replaces the complex 3D geometry of the RHS face in tension and
replaces it with a 2D model with rotational springs at edges to replace the rotational stiffness contribution of sidewalls as shown in Fig. 2. This is
the approach adopted in Jaspart et al. [8] and yields the same equation; however the original work does not give the derivation of this equation so
it is included for reference.
For the derivation of this coefficient, the simplified model as shown in Fig. 19 was used. Symmetry was taken advantage of to simplify the
model to a 2 beam model. Rotational stiffness, kr, is calculated as the ratio of the applied moment, M, to the rotation, θ, both at Node 3 and
therefore derivation requires determining the relationship between these two values.
The following stiffness matrix was formed for this model:
Beam 1
L1 = w / 2
2 3
EA EA
6 L 0 0 − 0 0
7
6 1 L1 7
6 7
6 12EI 6EI 12EI 6EI 7
6 0 0 − 3 7
6 L31 L21 L1 L21 7
2 3 6 72 3
P1X 6 7 x1
6 6EI 4EI 6EI 2EI 7
6 P1Y 7 6 0 0 − 2 76 y1 7
6 7 6 L1 7
6 7 6 L21 L1 L1 76 θ1 7
6
M1
7 =6 76 7
6 7 6 EA 76 7
4
P2X
5 6 EA 76 x2 7
P2Y 6− 0 0 0 0 74 y2 5
6 L1 L1 7
M2 6 7 θ2
6 12EI 6EI 12EI −6EI 7
6 7
6 0 − − 0 7
6 L31 L21 L31 L21 7
6 7
6 7
4 6EI 2EI 6EI 4EI 5
0 0 − 2
L21 L1 L1 L1

Beam 2
L2 = d
2 3
12EA 6EA 12EI 6EI
6 L3 0 − − 3 0 − 2 7
6 2 L22 L2 L2 7
6 7
6 EA EA 7
6 0 0 0 − 0 7
6 L2 L2 7
2 3 6 72 3
P2X 6 7 x2
6 6EI 4EI 6EI 2EI 7
6 P2Y 7 6 − 0 0 76 y2 7
6 7 6 L2 7
6 7 6 L22 L2 L22 76 θ2 7
6
M2
7 =6 76 7
6 7 6 76 7
4
P3X
5 6 12EI 6EI 12EI 6EI 76 x3 7
P3Y 6− 3 0 0 74 y3 5
6 L2 L22 L32 L22 7
M3 6 7 θ3
6 7
6 EA EA 7
6 0 − 0 0 0 7
6 L2 L2 7
6 7
6 7
4 6EI 2EI 6EI 4EI 5
− 2 0 0
L2 L2 L22 L2

P P P P
kr kr
at at
dRHS a a

wRHS
Fig. 19. Usage of kr to quantify rotational stiffness contribution of RHS sidewalls.
150 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152

Model
L1 = w / 2; L2 = d

2 3
EA EA
6 L 0 0 − 0 0 0 0 7 0
6 1 L1 7
6 7
6 12EI 6EI 12EI 6EI 7
6 0 0 − 0 0 0 7
6 7
6 L31 L21 L31 L22 7
6 7
6 6EI 4EI 6EI 2EI 7
6 0 − 0 7
6 0 0 0 7
2 3 6 L21 L1 L21 L2 72 3
P1X 6 7 x1
6 7
6 P1Y 7 6 EA EA 12EI −6EI 12EI 6EI 76 y 7
6 7 6− 0 0 + 3 0 − 0 − 76 1 7
6 M1 7 6 L1 L22 76 7
6 7 6 L 1 L2 L22 L32 76 θ1 7
6 P2X 7 6 76 x2 7
6 7 6 12EI 6EI EA 12EI −6EI EA 76 7
6 P2Y 7 = 6 0 − 3 − 2 0 + 3 0 − 0 7 6 7
6 7 6 L1 L1 L2 L1 L1 L2 76 y2 7
6 M2 7 6 76 θ1 7
6 7 6 76 7
6P 7 6 6EI 2EI 6EI 6EI 4EI 4EI 6EI 2EI 7 6 7
6 3X 7 6 0 − 2 − + 3 0 76 x3 7
4P 5 6 L21 L1 L21 L2 2 L2 74 y3 5
7
3Y 6 L2 L1 L2
M3 6 7 θ
6 12EI 6EI 12EI 6EI 7 1
6 0 − 3 7
6 0 0 0 0 7
6 L2 L1 L23
L2 7
2
6 7
6 7
6 EA EI 7
6 0 0 0 0 − 0 0 0 7
6 L2 L2 7
6 7
6 4EI 7
4 6EI 2EI 6EI 5
0 0 0 − 2 0 0
L2 L2 L22 L2

The following boundary conditions and loads are used to solve for M3 and θ3:
x1 = 0
θ1 = 0
x2 = 0
y2 = 0
x3 = 0
M2 = 0
P1Y = 0

Solving for θ2,

M2 = P1Y

2
2EIw ðð2θ2 + θ3 Þ + 4dðθ2 w + 3y1 ÞÞ 24EIðθ2 w + 4y1 Þ
=
d w2 w3

θ2 w
θ2 = −
d + 2w

Putting this value into M3,


θ3 w
2EI − + 2θ3
2EIðθ2 + 2θ3 Þ d + 2w 2EIθ3 ð2d + 3wÞ
M3 = = =
d d dðd + 2wÞ

Dividing M3 by θ3 to get kr:

M3 2EIð2d þ 3wÞ
=
θ3 dðd + 2wÞ

4EIðd + 1:5wÞ
∴kr =
dðd + 2wÞ

A2. Derivation of kr for rotational stiffness contribution of H-section flanges

This section gives the derivation of the kr coefficient for rotational stiffness contribution of the H-section flange for usage in the
stiffness characterisation of bolted endplate joints to H-sections columns (Minor axis). The kr coefficient replaces the complex 3D geometry of the
 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152 151

B
P P P P
kr kr
at
at
a a

D
Fig. 20. Usage of kr to quantify rotational stiffness contribution of H-section flange.

H-section web in tension and replaces it with a 2D model with rotational springs at edges to replace the rotational stiffness contribution of flanges
as shown in Fig. 20.
The simplified model as shown in Fig. 21 is used to determine spring stiffness of the H-section flange. Symmetry was taken advantage of to
simplify the model to a 2 beam model. Rotational stiffness, kr, is calculated as the ratio of the applied moment, M, to the rotation, θ, both at Node 2
and therefore derivation requires determining the relationship between these two values.
As the rotation of a cantilever with an end moment of M is given by

ML
θ=
EI

Solving for θ2,

MBeam1 D M D
θ2 = θ2;Beam1 + θ2;Beam2 = + Beam2
2EI 2EI

As M2 = MBeam1 + MBeam2,

M2 D
θ2 =
2EI

As kr = M2 / θ2:

2EI
∴kr =
D

M
Node
3

Beam
2
d

Beam
1
Node Node
1 2
w/2
Fig. 21. Simplified model for calculation of RHS sidewall rotational stiffness.
152 A.Y. Park, Y.C. Wang / Journal of Constructional Steel Research 70 (2012) 137–152
References
[1] British Steel Tubes Pipes. SHS jointing — flowdrill & Hollo-bolt. British Steel Tubes
and Pipes; 1997.
[2] Bursi OS, Jaspart JP. Benchmarks for finite element modelling of bolted steel
connections. J Constr Steel Res 1997;43:17–42.
[3] CEN. Eurocode 3 — design of steel structures part 1–8: design of joints. British
Standards Institute; 2005.
[4] Dassault Systèmes (2007). ABAQUS User's Manual Version 6.7. Dassault Systèmes.
[5] France JE, Davison JB, Kirby PA. Moment-capacity and rotational stiffness of
endplate connections to concrete-filled tubular columns with flowdrilled
connectors. J Constr Steel Res 1999;50:35–48.
[6] France JE, Davison JB, Kirby PA. Strength and rotational response of moment
connections to tubular columns using flowdrill connectors. J Constr Steel Res
1999;50:1–14.
[7] France JE, Davison JB, Kirby PA. Strength and rotational stiffness of simple
connections to tubular columns using flowdrill connectors. J Constr Steel Res
1999;50:15–34.
[8] Jaspart J, Pietrapertosa C, Weynand K, Klinkhammer R. Development of a full
consistent design approach for bolted and welded joints in building frames and
trusses between steel members made of hollow and/or open sections —
application of the component method — CIDECT Report 5BP-4/04. Universitè de
Liegé; 2004.
[9] Timoshenko. Theory of plates and shells. 1st Edition. New York and London:
McGraw-Hill Book Company, Inc.; 1940. p. 158.
[10] Weynand K, Jaspart J, Ly L. Application of the component method to joints
between hollow and open sections — CIDECT Final Report 5BM. Universitè de
Liegé; 2003.