See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/319699682

Calibration of Eurocode design models of thin-walled cylinder under bending

with full scale tests

Conference Paper  in  ce/papers · September 2017

DOI: 10.1002/cepa.429

CITATIONS READS

2 501

4 authors:

Dirk Jan Peters Adam J. Sadowski

Delft University of Technology Imperial College London
15 PUBLICATIONS   28 CITATIONS    72 PUBLICATIONS   608 CITATIONS   

SEE PROFILE SEE PROFILE

J. Michael Rotter Andreas Taras

The University of Edinburgh ETH Zurich
244 PUBLICATIONS   4,546 CITATIONS    67 PUBLICATIONS   369 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Granular Solids and Silos View project

Shell buckling design philosophy View project

All content following this page was uploaded by Adam J. Sadowski on 29 September 2017.

The user has requested enhancement of the downloaded file.

EUROSTEEL 2017, September 13–15, 2017, Copenhagen, Denmark

Calibration of Eurocode design models of thin-walled cylinder under

bending with full scale tests
Dirk Jan Peters a, Adam Sadowski b, Michael Rotter b, Andreas Taras c
a
Technical University Delft, [email protected]
b
Imperial College London
c
Bundeswehr University Munich

ABSTRACT
Design formulas for thin-walled cylinders in bending show remarkable differences in various codes,
including versions of the Eurocodes. Thin-walled, high grade steel cylinders are used in piles, e.g.
in combined walls for quay wall construction or in large mono-piles for mooring structures or
offshore foundations. These piles experience a dominant loading in bending, and may be designed
with Eurocode formulas for shell buckling of prismatic cylinders.
In recent years, a relatively large number of new results of full scale experiments has become
available. These results can be combined with test results from the literature and can be used for
calibration of the design model formulas for thin-walled tubes, in particular in the range that has a
high susceptibility to elastic or plastic local buckling. The diameter range of the tests is 400 to 900
mm. The d/t ratios range from 40 to 110 and the steel grades from S235 to S460.
New capacity-based formulas for Reference Resistance Design (RRD) of cylinders in bending have
been developed from extensive GMNIA calculations (geometrically and materially nonlinear
analyses with imperfections included) that included many variations of the relevant parameters and
buckling pre-conditions. A calibration against the tests forms a potentially valuable confirmation of
those design rules. The methodology of the calibration is based on statistical procedures that prove
sufficient reliability according to the Eurocode requirements.
Similar to other calibration exercises, it appears that the difference between the nominal and real
material strength is, amongst other aspects, an important input parameter for the calibration result.
The evaluation performed here shows that the published Eurocode EN 1993-1-6 2007 strength-
based formulas produce conservative results and that the new RRD amendment which produces
less-conservative design values can still be safe.
Keywords: Local buckling, Eurocode & Codification, Stability, Cold-formed, Experimental
verification, finite element method
1 INTRODUCTION
Tubular steel piles are often used in civil engineering structures like retaining walls or dolphin piles.
They are loaded with axial loads, radial loads and in bending. Bending is often the dominant load.
For reasons of design economy, their diameter-to-thickness ratio (d/t) is often high, which makes
them potentially susceptible to local buckling. The Eurocodes EN 1993-1-1 [1] and EN 1993-1-6
[2] provide design rules for tube bending. Like non-tubular sections, cross-sections are classified as
plastic, compact, semi-compact and slender, depending on the capacity and deformation capacity of
the sections. For tubes this classification is based on the slenderness ratio
d
t 2
with d is the external diameter in mm,
t is the wall thickness in mm,
ε is a factor depending on fy:  2  f ref f y ,
fy is the yield strength in MPa, fref = 235 MPa.
Currently the classification boundaries in the EN 1993-1-1 [1] are 70 ε2 < d/t < 90 ε2 for semi-
compact tubes, that can safely resist bending moments with first yield in the outer fibre. Beyond
© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
90 ε2 the tubes are susceptible to local buckling, which means that practical irregularities in
geometrical shape, weld geometry and connections of tube parts can induce instability below the
yield limit.
In the past, the EN 1993-1-1 [1] design rules for other instability cases, like column buckling and
local torsional buckling, have been investigated and calibrated with experiments. Slender tubes are
classed as cylindrical shells and so are treated as shell structures for which the Eurocode 1993-1-6
[2] is written. Due to the complex geometry of thin shell structures and the great difficulty in
undertaking experiments that relate to practical construction, much dependence on computer models
has become necessary. The calculation method GMNIA has become a common basis for advanced
designs. For cylindrical shell buckling experiments can be carried out but were not available in
sufficient quantities in the relevant range in 2007. A calibration against EN 1990 Annex D [3] was
not performed.
The EN 1993-1-1 and 1993-1-6 rules for tubular sections have usually been perceived as
conservative, and a number of investigations have been executed more recently, including
experimental verification.
In this paper the focus is on a comparison between predictions and experimental outcomes for tubes
of the sizes normally used for foundation piles (i.e. with diameters between 500 and 1200 mm).
2 METHODOLOGY
2.1 Literature review
Many authors have addressed the issue of reliable information on the classification of sections.
Sections that are susceptible to local buckling have a lower resistance, and show a less desirable
brittle type of failure [6]. Sections that can be loaded up to or slightly beyond first yield should be
treated as Class 3 sections.
The limit between Class 3 and Class 4 sections is treated differently in different codes. Rather
remarkable differences can be seen [7]. With the exception of two American codes, the boundary
between Class 3 and Class 4 in the various codes is between 90 2 and 130 2.
Designers should be aware that the information of the class boundary in different codes are not
necessarily comparable. Different codes use different partial safety factors, and this can cause one
code to appear to be less stringent on a class boundary, yet still produce conservative design
resistances. However, the statistical verification of these different codes cannot be checked.
2.2 Slenderness limit papers
Many authors have studied the available tests to define the slenderness limits according to the
accepted section classification. Elchalakani et al [6] investigated cold-formed circular hollow
sections and concluded that the upper limit of Class 3 sections should be at d/t = 133 2 (using fref =
235 MPa). Lanhui Guo et al [7] investigated very slender members with a low yield strength, and
concluded that this same Class 3 upper limit could be d/t = 150 2. They compared their results
with literature values and concluded that the Australian code [8], which has the class boundary at
130 2 is a good choice. None of these authors seem to have based their class boundary choice on a
statistical treatment, including a reliability assessment, of the strength. They each use the
traditionally accepted process of drawing a line as a close lower bound on the test results.
Gardner et al [9] collected a series of 153 tests from the literature and proposed a ‘mean curve’, as
an average of the cloud of test observations. They treated three sub-groups of hot-finished, cold-
formed tubes and fabricated tubes as a single group to obtain unified slenderness limits. They then
followed the statistical evaluation of EN 1990 [3] and found an offset factor M* of 1.24 for a
Eurocode ‘design curve’. They did the same to find an AISC/AS design curve. The resulting
Eurocode design curve crosses the first yield line at d/t 2 = 100 and the AISC/AS design curve at
135. This difference arises due to the different code verification process, and is not based on a lack
of data or other reasons.
Gardner et al [9] used as a ratio of mean to nominal yield strengths a value of 1.16 and coefficients
of variation of yield strength and geometrical properties 0.05 and 0.02 respectively. They did not
© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
differentiate between the sub-groups, although there appears to be a significant difference between
the fabricated and cold formed sections group, and the hot-formed sections group (Fig. 1).

1.1

1
M experiment / Mp

1.06
0.9

0.96 Fabricated, collected Gardner

0.8
Hot-finished, collected Gardner
0.98
0.7 Cold-formed, collected Gardner

0.6

0.5
50 70 90 110 130 150 170 190 210
Dd / t 2

Fig. 1. Data points used in Gardner et al. [9] with global average and averages of subgroups
2.3 Capacity based design rules
For slender tubes in Class 4, the most commonly used European capacity based design procedure is
given in EN 1993-1-6 [2] Annex D.1, written for cylindrical shells. As can be seen in Fig. 2 the
current design rules of EN 1993-1-1 [1] covering Classes 1 to 3, shows a large discontinuity at the
boundary between Classes 2 and 3 (at d/t2 = 70) and there is a smaller one at the boundary where
the two standards [1] and [2] meet at the boundary between Classes 3 and 4 (at d/t2 = 90). This
step is partially also caused by the difference between M0 and M1. The further drop at
approximately d/t2 = 160 is caused by passing the rather strange limit value of E/fy = 500 (EN
1993-1-6 clause D.1.2.1 (7)).
A new general design procedure called Reference Resistance Design (RRD) has been developed by
Rotter [10,11]. The specific parameters in RRD for tubes under uniform bending were developed
from very extensive GMNIA nonlinear finite element calculations, including all geometries, all
amplitudes of imperfection, all steel grades and all possible strain hardening moduli [12] and are
presented as a new Annex E.1 in EN 1993-1-6 [13]. The result is shown as ‘Annex E.1’ (2017) in
Fig. 2. This method has been accepted as an amendment and published, and is now applicable to
current designs. It is evident that this design method is much more continuous and also less
conservative. This paper aims at calibration of this Annex E.1 design rule, thus providing a simple,
safe and economic design rule for cases of pure bending of tubes.

0.8 EC3-1-1 (2005)

M/Mp

EC3-1-6 (2007)
Norsok N-004 (2004)
0.6 Annex E.1 (2017), with gM = 1.1
My / Mp, for fy / fref = 1.5

0.4
50 70 90 110 130 150 170 190 210 230 250 270
d/t 2

Fig. 2. Design rules, including partial factors M0 and M1 plotted in design chart

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
For reference purposes the design rule of Norsok N-004 [14] is also plotted in Fig. 2. This capacity
based design formula at first seems to cross the first yield limit (Class 3 to 4 boundary) at around
160  2, but [14] includes a varying value of M with slenderness starting at 1.15 and progressively
increasing to 1.3, which changes the intersection to below 90  2.
The three design rules given here are for pure bending, possibly combined with axial force, and are
as shown on in the graph not yet applicable to other load cases like internal pressure, loads exerted
by sheet piling between the primary tubular piles, energy absorption by large strains and special
cases of wall thickness transitions and local force introduction, all of which may be needed in the
piling standard [15].
2.4 Evaluation of experimental results
The method of calibration consists of a comparison of experimental results with predictions, known
as model results, using the following procedure. For every experiment result under bending
Mexperiment / Mp a model point Mmodel / Mp is calculated.
The experiment result is expressed as the measured maximum bending moment Mexperiment, just
before bucking (Fig. 4), normalised by the full plastic bending moment Mp, which is based on the
measured yield strength of the specimen.
The values of the measured yield strength are considered adequately represented by the Rp0,2 and
Rt0,5 values, of which Rt0,5 is favoured. The measured ductility values of the tubes of the Combitube
programme (see Section 3) lie between 1.5 and 4, with the majority between 1.5 and 3. The
measured axial strains at buckling lie between 0.0025 to 0.006. The steel strength properties are
measured as Rp0,2 and at Rt0,5. The parallel value of 0.2% proof stress corresponds to a total strain of
0.0019 + 0.002 = 0.0039 for fy = 400 and to 0.0029 + 0.002 = 0.0049 for fy = 600 MPa. As a result
the values of Rp0,2 and Rt0,5 are very similar. Both are good for ductility values between 1.5 and 2.5.
In these strain ranges, the effects of strain hardening do not need to be considered as affecting
global bending capacity. For plastic local bending of a tube wall, which occurs during local
buckling, strain hardening plays a significant role.
The full plastic moment can be calculated with fy based on the measured value of Rt0,5%.
fy
Mp  d 3   d  2t 3    d  t 2 tf y (1)
6  
with d and t as defined in Fig. 3.
The measured strain just before buckling buc may be expressed as curvature with
Cmeasured = 2buc / d, and normalized as C / Ce with
2 fy
Ce  (2)
Ed
Note that the full plastic moment can only be reached in Class 1 and 2 sections (plastic and
compact). Most piles are Class 3 or 4 (semi-compact or slender). A typical analytical bending
moment curvature relationship, including the Brazier ovalisation effect [16], for tubes with
d = 914 mm, t = 12/16/25 mm and fy = 600/500/400 MPa respectively are plotted in Fig. 4.
The variation in the experiment results is calculated because the known variation Vrt of basic
variables is needed as an input in a reliability assessment [3]. For this value a value of 0.08 was
chosen here, based on (0.0752 + 0.0242), using the known variations in the yield strength and wall
thickness respectively, which were measured in this study. Using the procedure of EN 1990,
Annex D [3], a partial factor M can be established. This is a “model factor”, representing the
inherent variability in the structure alone, rather than the variability of the material. When
comparing the experimental and model values, the measured yield strength in tests is used as input
for the model points. For these calibrations, the mean steel over-strength must be used (Section 5).

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
 1
1.2
1 0.8
t
0.8 0.6

M / Mp
M / Me
0.6
d 0.4
0.4

0.2
0.2
0 0
0 1 2 3 4 5 6 7
C / Ce

Fig. 3. Definition diagram for d and t Fig. 4. Theoretical moment curvature diagrams and typical
measured moment curvature diagrams, noting the point
Mexperiment / Mp and Cmeasured / Ce.

3 EXPERIMENTAL FAILURE DATA

In 2012, experiments were performed on empty tubes and fully sand-filled tubes loaded under pure
bending to investigate the influence of the sand-fill on the local buckling resistance [17]. The tests
were meant to test the behaviour and effect of the sand-fills for piles in combined walls. A series of
tests on empty piles was also performed to provide a reference. The experiments were verified using
FEM calculations to understand the failure mechanisms occurring in the tube and the sand. The
tests were commissioned by CUR, a Dutch joint research organization, prepared and evaluated by
RHDHV, sponsored by Port of Rotterdam and by Arcelor. They were carried out as notional four-
point bending tests on simple supports. The tubes were horizontal with vertical loading. The central
moment region was 12 m long with side moment gradient zones of 5.5 m. Each test specimen was
12 m long with end-plates and bolted connections to the side zones. The central moment region was
loaded using measured weights applied using flexible straps around each tube, deviating from the
ideal 4-point uniform bending condition with a slightly increased moment in the middle of the span.
All tubes were precisely surveyed and the geometrical deviations were measured and could be
classified according to the EN 1993-1-6 [2] Fabrication Tolerance Quality Classes (FQC) A, B or
C. The steel properties were found using tensile tests. All piles failed by local buckling.
The sand filled piles also failed in local buckling but at higher loads and higher deformation levels.
The buckled shape of each sand filled pile was typically an outward buckle. The empty tubes failed
with the classic diamond buckle shapes. All failures occurred in the middle zone.
In 2014 new experiments were carried out in the Port of Rotterdam on 914 mm diameter piles
installed in the harbour bed [18]. The tests prepared and evaluated by RHDHV and W+B, funded
by the Port of Rotterdam and were meant to test a proposed design method for energy absorbing
dolphin piles. After installation, the embedded part of each pile was mostly filled with soil. Eight
piles with lengths of 20 to 22 m were installed and loaded laterally up to the point of excessive
yielding or local buckling failure. Most of these piles failed by local buckling beneath the harbour
bed, but one pile buckled where the cross-section was empty, so is also used here (Table 1).
Between 2012 and 2014 a number of experiments were carried out as a part of the EU-funded
Combitube programme, performed at the University of Karlsruhe, Germany and Delft University,
the Netherlands. The test data and results are summarized below.
The Karlsruhe tests were as uniform bending with a small compressive force, the lever arm being
4.0 m and tested length 10.5 m. The Delft tests were four-point bending tests with a 8.1 m middle
zone and 3.6 m end zones. The full specimen was 15.3 m long between sling supports. Both the
load introduction points and the supports allowed ovalisation.

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
 Table 1. Summary of test results CUR project executed in Rotterdam in 2012
Mexperiment Test weld Tolerance
d t Grade d/t fy d/t fy/fref measured ecr C/Ce
/ Mp ID type Class
[mm] [mm] [MPa]  10-3
508 5 S235 101.6 305 131.9 0.96 2.50 1.72 A3 spiral B
508 5 S235 101.6 314 135.8 0.96 4.20 2.81 A4 spiral A
508 5 S235 101.6 307 132.8 1.00 3.51 2.40 A6 spiral A
508 7.1 X60 71.5 503 153.0 0.98 5.98 2.50 B2 spiral A
508 7.1 X60 71.5 503 153.0 0.99 6.61 2.76 B5 spiral A
508 7.1 X60 71.5 470 143.0 1.03 4.42 1.98 B6 spiral B
610 6.3 S235 96.8 317 130.6 0.86 4.55 3.01 C2 spiral B*
610 6.3 S235 96.8 329 135.6 0.82 4.68 2.99 C3 spiral B*
610 6.3 S235 96.8 312 128.5 0.95 5.78 3.89 C6 spiral A
711 6 X65 118.5 502 253.4 0.82 3.95 1.65 D1 long A
711 6 X65 118.5 502 253.4 0.70 3.82 1.59 D2 long A
711 6 X65 118.5 502 253.4 0.81 3.08 1.29 D3 long B
914 12 76.2 350 113.4 0.95 3.67 2.20 FD7 Spiral
* = oval shape Class B was tested in an upright 0-position and not in the flat ᴑ-position

Table 2. Summary of test results COMBITUBE project tests executed at Karlsruhe University [21]
d/t M/ measured Test Tolerance
d t Grade d/t fy C / Ce weld type
fy/fref Mp ebuc ID Class
[mm] [mm] [MPa]  10-3
820 11 X52 73.5 443 138.6 0.85 5.88 2.79 K1 spiral
spiral, coil weld in the
820 11.1 X52 72.9 398 123.5 0.85 4.09 2.16 K2
middle
820 9.1 X52 89.1 299 113.4 1.03 3.83 2.69 K3 spiral
820 7.8 X52 104.1 462 204.7 0.80 2.80 1.27 K4 spiral
spiral, coil weld in the
820 7.5 X52 108.3 418 192.6 0.78 1.25 0.63 K5 **
middle
spiral, coil weld in the
863 8.4 X52 101.7 415 179.6 0.80 2.85 1.44 K6
middle
spiral, girth weld in the
863 8.4 X52 101.7 444 192.1 0.79 2.61 1.24 K7 C
middle
863 8.6 X52 99.3 408 172.4 0.64 2.32 1.19 K8 spiral
** Significant initial dimple

Table 3. Summary of test results COMBITUBE project tests executed at Delft University [21]
d/t M/ measured Tolerance
d t Grade d/t fy C / Ce Test ID weld type
fy/fref Mp ebuc Class
[mm] [mm] [MPa]  10-3
1066 16.4 X70 65.1 534 148 0.92 5.67 2.23 T1 spiral A
1067 9 X60 118.3 407 205 0.74 2.79 1.44 T2 spiral A
1069 9 X60 118.7 392 198 0.74 2.05 1.10 T3 spiral B
1065 9.2 X60 116.2 441 218 0.83 3.14 1.49 T4 spiral A
1070 9 X60 118.4 399 201 0.82 2.86 1.51 T5 spiral A
1066 16.3 X70 65.3 529 147 0.87 4.88 1.94 T6 spiral B
1068 16.3 X70 65.4 614 171 0.73 4.60 1.57 T7 spiral *** A
1068 9.1 X60 117.4 452 226 0.75 2.95 1.37 T8 spiral A
1068 16.3 X70 65.4 614 171 0.81 6.12 2.09 T9 spiral A
1070 13.1 X52 81.6 331 115 0.9 3.40 2.16 T10 spiral B
1068 12.9 X52 82.8 352 124 0.83 4.36 2.60 T11 spiral A
1069 9.1 X60 117.1 448 223 0.75 2.59 1.22 T12 spiral + girth weld A
1070 9.2 X60 116.3 443 219 0.75 2.72 1.29 T13 spiral + girth weld B
1068 9.8 X60 108.8 505 234 0.78 2.74 1.14 T14 long A
1070 14.8 X70 72.3 536 165 0.87 6.19 2.42 T15 long A
*** Curvature measurement disturbed

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
 Table 4. Properties and results of full scale tests found in other literature
d t d/t  buc fy fu Mp Mexperiment / Mp Author Exp #
[mm] [mm]  10 -3
[MPa] [MPa] [kNm]
152.4 7.9 19.3 24.4 517 601 85.4 1.259 i 1
762 8.3 91.8 3.8 404 601 1905 1.004 ii C45P00
152.5 1.5 101.7 3.6 342 511 11.7 0.957 iii 25
152.5 1.5 101.7 4.29 342 511 11.7 0.966 26
122 1.5 81.3 14.8 365 515 7.95 0.975 25b
122 1.5 81.3 4.3 365 515 7.95 0.914 25c
609.6 6.35 96 2.27 357 505 825 0.868 iv 1
509 11.2 45.4 15 479 568 1330 1.133 v B1
514.7 17.5 29.3 11 459 533 1987 0.917 B2
507.9 19 26.7 17.4 474 527 2154 1.066 B3
525.6 23.6 22.3 23.6 450 534 2678 1.006 B4
273.1 5.9 46.1 14.3 381 161 1.047 vi 10a
273.1 8.9 30.7 42.5 334 208 1.150 10b
406.4 6.6 61.5 8.05 339 358 1.069 16a
508 6.5 78.4 3.71 377 616 0.948 20a
711 19 37.4 24.2 553 630 5033 1.179 vii A1
711 19 37.4 22.3 553 630 5033 1.072 viii A2
168.3 4.8 35.2 10 369 47.4 1.022 3
273.1 5.6 49.1 9.7 306 123 1.039 5
406.4 6.4 64 6.6 309 316 0.936 8
114.3 4 28.8 12 309 15.0 1.077 9
273.1 5.6 49.1 9.8 306 123 0.993 10
508 7.9 64.3 386 763 0.898 ix UGA508
508 7.9 64.3 386 763 1.008 UGR508
324 6.4 51 22 359 232 0.910 UGA324
272.2 14.86 18.3 9.05 290 286 1.065 x 1
273.9 7.8 35.1 9.33 304 168 0.962 4
273.9 5.61 48.8 10.9 404 479 163 0.932 7
273.1 3.53 77.4 8.87 288 313 73.9 0.877 13
272.9 2.46 110.8 4.46 310 55.8 0.864 17
168.3 3.2 53 7.9 442 550 38.6 1.037 xi 2
610 15.1 40.4 451 607 2411 1.256 xii LN2
610 15.1 40.4 557 772 2977 1.188 HN3
610 12.7 48 553 752 2506 1.126 HN6
610 12.7 48 557 637 2524 0.985 LN5
508 7.9 64.3 9.2 386 763 0.860 xiii UGA20W-2
324 6.4 51 6.8 358 231 0.973 UGA12W
610 7 87.3 3.66 440 1120 0.970 xiv B2
611 14.9 41 12.2 470 2489 0.973 B5
762 9.4 81.1 5.2 552 2939 0.907 xv 1
762 15.7 48.5 11.2 620 5422 0.959 4

i) Carr. MacRae. Bruton (2009) ix) Mohareb. Elwi. Kulak and Murray (1994)
ii) DelCol. Grondin. Cheng and Murray (1998) x) Sherman (1976)
iii) Douwen. van. Gresnigt. and Stark (1974) xi) Stephens. Olson and Rosenfeld (1991)
iv) Gresnigt (1977) xii) Suzuki. Endo. Yoshikawa and Toyoda (2001)
v) Gresnigt and van Foeken (1998) xiii) Yoosef-Ghoodsi. Kulak and Murray (1994)
vi) Jirsa. Lee. Wilhoit and Merwin (1972) xiv) Zimmerman. Stephens. DeGeer and Chen (1995)
vii) Kang. Yoo. Ahn. Cho and Yoon (2007) xv) Zimmerman. Timms. Xie and Asante (2004)
viii) Korol (1979) Cited from [20] and [22]

The measured ratios mean / nominal yield strength of the various subsets are on average 1.25 for
the Rotterdam tests, 1.15 for the Karlsruhe tests, and 1.08 for the Delft tests. The high value for the
Rotterdam tests is due to the presence of material grades S235, which is very uncommon for tubes.
Although the steel involved was graded as S235 it could have been accepted as grade S275. If this
S235 specimen is omitted the average for the Rotterdam tests is 1.15. A more detailed examination
shows that some of the tested materials in the Karlsruhe and Delft series were far below their grade.
In practice they would have been rejected for construction purposes. If all specimens with a ratio
© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
below 0.98 are omitted the average is found to be 1.145 with a standard deviation of 0.085 (CoV =
0.075) (Fig. 5). There is no correlation between high yield strength and low wall thickness (Fig. 5).
8 1.4
All tests average =
7

Yield strength: real / nominal

Karlsruhe, 1.145
6 Delft, stdev =
R'dam 0.085
5
4 1
3
2
1
0 0.6
0.8 0.9 1 1.1 1.2 1.3 1.4 0.9 1 1.1
yield strength: real / nominal Wall thickness: real / nominal

Fig. 5. Normal distribution of ratios real vs nominal yield strength in the experiments and plot of measure yield vs
measured wall thickness variation
4 FIELD DATA ON MATERIAL YIELD
To test whether the ratios of real to nominal yield strength for the test tubes are an adequate
representation of this ratio in practice a large number of material certificates were collected,
showing steel grades, measured yield, tensile strength etc.
A collection of 446 certificates was analysed of steel tube material used in real projects over the last
5 years. The steel had a range of origins, giving a good reflection of the current globalized market
for steel production and trade. The certificates cover European and API 5L grades. The total
population shows an average of ratio real to nominal yield strength of 1.12. The statistical
distribution of the bigger dataset is quite close to normal, although skewness to high values can be
seen (Fig. 6 p-p-plot). The ratio decreases slightly for higher grades, from 1.18 for S355 and X52 to
1.1 for X70. One of the largest subsets is for S460, whose average is 1.11, and which also gave
unacceptably low values for 3% of the cases. The average of the remainder is 1.12. If this same
group is considered as representative of a nominal grade of X70, the average ratio becomes only
1.038 with 10% dropping below unity. The average of the remaining 90% is 1.074 (Fig. 6).
This situation is not uncommon, and makes it less clear that assessments should systematically rely
on average over-strength values based on general steel production figures. The analysis results give
the impression that a 15 or 17% average over-strength on yield, which has been used for calibration
studies in the past, is too high. Although arguments could be found use an even smaller value, it is
proposed here to use an average over-strength of 12% in this calibration analysis, being a 1.12 ratio
of average measured to nominal yield strength.
70 2
P-P plot of all collected data
60 All collected
certificates lognormal
1
50
Normality

40 normal
average = 0
30 1.12,
stdev = 0.091
20
-1
10

0 -2
0.8 1 1.2 1.4 1.6 0.8 1 1.2 1.4 1.6
Yield strength: real / nominal Yield strength: real / nominal

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
45 60
40 All collected All collected
S460 50 S460
35 certificates certificates average =
30 40 changed for 1.058,
average = X70 stdev = 0.079
25 1.11,
stdev = 0.083 30
20
15 20
10
10
5
0 0
0.8 0.9 1 1.1 1.2 1.3 1.4 0.8 0.9 1 1.1 1.2 1.3
Yield strength: real / nominal Yield strength: real / nominal

Fig. 6. Normal distribution of ratios measured to nominal yield strength in 446 certificates from constructed projects

5 COMPARISON OF PREDICTIONS AND EXPERIMENTAL RESULTS

The initial geometrical deviations of most of the tube specimens were carefully measured. They
were assessed against the FQC definitions A, B and C of [2]. For some experiments no appropriate
measurements are known. For tests in the range d/t > 90  2, the relationship between the ratio
Mexp / Mp and d/t 2 is shown in Fig. 7, with all tests classified, where possible, according to their
FQC, together with a line showing the average of all results. The Karlsruhe K5 test was excluded
because this tube lay outside Class C and would be rejected in a construction project. The average
resistances appear to correspond to the tubes with Class B. Class A tests deviate not more than 1%
from the mean. The single Class C point is 5% below the mean. It is therefore proposed here to
treat the entire test data as a homogeneous group of FQC Class B. Differences in resistance for
different FQCs will still appear in the outcomes of the design rule, since these were based on FEM
calculations with thoroughly explored variations of initial geometrical deviations [12].
1.3
All experiments above 90
1.2
FQC class A
1.1 FQC class B
M experiment / Mp

FQC class C
1

0.9

0.8

0.7

0.6
70 100 130 160 190 220 250
D/t  2

Fig. 7. Fabrication Tolerance Quality Classes of test specimens with d/t > 90  2
6 CALIBRATION OF THE MODEL
All the available experimental results are plotted in Fig. 8. About the same average and scatter is
seen in each of the identified groups. The new tests (open markers) contribute significantly to a
validation at higher d/t. The sources of the many data points from [9] is unknown, but by inclusion
in Fig. 8 they confirm the trend. They are omitted from the current calibration analysis due to lack
of source data. The grey markers show the new experiments evaluated using measured yield
strengths from the literature (Table 4). The calibration analysis is thus based on 76 remaining tests.
First the mean of the data points is shown, conveniently described in this range by an exponential
function. This fitted curve is close to parallel to the ‘model curve’ derived from Annex E.1 [13].
© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
The procedure given in EN 1990 Annex D [3] entitled ‘Design assisted by testing’ is used here to
determine the safe separation of the model from the mean experimental outcome. This procedure
evaluates model predictions of a single outcome, here the resistance in bending, against
experimental results, using least-squares matching and accounting for known variations of input
parameters. The procedure can be used in two ways: to find a characteristic value and partial factor
or to find a design value directly. Here the procedure is applied both ways.
The procedure includes the effect of having too few results, thus leaving the real CoV uncertain, but
that is not relevant here. From the population of experiments the variation in the relevant input
parameters were found. This variation was found to vary with d/t 2, with a smaller CoV at higher
values of d/t 2. This corresponds to the fewer experiments in this range. Moreover the model
predicts that a 5% reduced wall thickness has a larger effect on the resistance at high d/t 2 than at
lower values. The test population was therefore treated as one group with a constant value of Vrt.
Using the procedure and adopting the value of Vrt = 0.08, the characteristic (5%) boundary is found
to be rk / rt = 0.87 and the boundary for the ultimate limit state at rd / rt = 0.723. These values
indicate that the partial factor M should be taken as 1.20.
If instead calibration is done to conform to the current Eurocode value of M1 = 1.1, the calibration
must focus on a correct positioning of the predictions (model) when including the factor M1 relative
to the design curve at the position rd / rt. To do this as accurately as possible, the non-symmetrical
distribution of the experiment results has been taken into account. Based on EN 1990 Annex D [3] a
point with Xd = d mx (1 + kd,n Vx) with kd,n = 3.04 must be found, corresponding to a probability of
1.2 × 10-3. This point is found at rd / re = 0.741 for the normal distribution and 0.772 for the log-
normal (Fig. 9). Using these values to ‘correct’ the required position of rd to become rd / rt = 0723
+0.772 – 0.741 = 0.754, the result can be expressed as rd / re = 0.754 / 1.018 = 0.74 (Fig. 9).
The position of the model results (using M1 = 1.1) relative to the experimental results is rd = 0.86 re
(Fig. 10). Using the above assessment of the effect of over-strength, there will be a shift of 12% in
measured yield relative to nominal yield, leading to 0.86/1.12 = 0.78. A correction that consists
simply of a uniform adjustment of the average is sufficient because the variation of the yield
strength has already been included in the calculation above. The Annex E.1 [13] prediction, based
on the GNMIA calculations [12], is then found to be above the required position by a factor of
0.78 / 0.74 = 1.05.
1.2

1.1 y = 1.218e-0.0021x
Delft
1
M experiment / M p

R'dam empty
0.9 Karlsruhe
R'dam dolphin empty
0.8 Collected Gardner, fabricated
Collected Gardner, cold-formed
0.7
Other literature
0.6 Annex E.1 (2017) FQC cl B, unfactored
neglected test Karlsruhe K8
Average of the experiments
0.5
50 70 90 110 130 150 170 190 210 230 250 270
d/t  2

Fig. 8. Experimental results compared with model

A simple application of this finding could be to amend the value of M, as applied to shell buckling,
to the value M =1.15, which is already closer than the 1.2 found in the procedure above.
Alternatively, since the procedures of [3] have so far not been used on shell buckling test results, it
may be found in due course that the value of M for shells should depend on both the geometry of
the shell and the stress condition involved, since the imperfection sensitivity and thus the scatter in
test results depend on the stress state [11,19]. There is no evidence that the previously chosen M1 =

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
1.1, found by calibration procedures applied to steel structural member tests, should also apply to
shell buckling conditions.

10000
re = 1.018 rt 14

Distribution of 76 experiment results

Normal
8000 12

0.868
rk Log-normal
re = M experiment

rd 10

0.860
6000
8

0.772
4000 6

0.741
4
2000
2

0
0 0 0.7 0.8 0.9 1 1.1 1.2 1.3
0 2000 4000 6000 8000 10000
Individual experiment result / average
rt = M model Annex E.1
Fig. 9. Statistical analysis results

1.2

1.1 y = 1.218e-0.0021x

1 Series1
M experiment / M p

Delft
0.9
R'dam empty

0.8 Karlsruhe
R'dam dolphin empty
0.7 Other literature
Annex E.1 (2017) FQC cl B, g_M = 1.15
0.6
Average of the experiments

0.5
50 70 90 110 130 150 170 190 210 230 250 270
d/t  2

Fig. 10. Experimental results used in the calibration procedure compared with Annex E.1 [13], including M = 1.15
7 SUMMARY AND ACKNOWLEDGMENT
The Eurocode EN 1993-1-6 [2] for shell buckling has been improved with the simplified and
elegant method of Reference Resistance Design [10,11], already implemented for cylinders under
uniform pure bending and spherical shells. The range of application of the method is very large
indeed, but even as applied to the special case of bending cylinders (Annex E.1) is still rather large:
from plastic, d/t = 50 2, to extremely thin shells (d/t = 2000). In recent years a large number of tests
have been executed on large steel tubes for piling and pipeline applications, which include spiral
welded and longitudinal welded tubes, including splices with girth welds and covering the range d/t
ratios range from 40 to 110 and the steel grades from S235 to S460.
These experiments have here been used to calibrate the predictions of Annex E.1 [13]. The
conclusions are that the trend of model follows the experimental results very well. The required
reliability of structural resistance according to EN 1990 [3], with a failure probability of 10-3, is
quite stringent. This calibration leads to a recommended model factor of 0.95 for the design of
fabricated welded tubular piles loaded in pure bending. The RRD method is a remarkable
improvement in terms of both simplification and economy of design. In the range of d/t = 902 to
2002, the calibrated method is approximately 20% less conservative than the 2007 published
© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)
EN 1993-1-6 [2]. The calibrated model line intersects the of My = 0.786 Mp line at d/t = 1152,
which confirms that the class boundary between Class 3 and Class 4 sections can indeed become
significantly higher than 902.
It is also noted that this conclusion is based on the assumption that the current value of M = 1.1,
drawn from tests on the buckling of structural members, should be reconsidered for shell buckling,
where imperfection sensitivity often leads to much wider scatter in test results. Given a
modification to M = 1.15, the existing Annex E.1 would meet the calibration requirement.
The Dutch contractors Dimco and VSF are acknowledged for sharing the procurement lists, steel
certificates and material test data sheet of their recent projects.

REFERENCES
[1] EN 1993-1-1. Eurocode 3: Design of steel structures. Part 1-1: General rules and rules for buildings. CEN. 2011
[2] EN 1993-1-6. Eurocode 3: Design of steel structures. Part 1-6: General - Strength and Stability of Shell
Structures. CEN. 2007.
[3] EN 1990 Annex D. Annex D of Eurocode 0: Basis of Design, CEN, Brussels. 2002.
[4] Simpson, B., Powrie, W. (2011). Embedded retaining walls, theory, practise and understanding, 15th
International Conference on soil mechanics and geotechnical engineering, Istanbul, 2001
[5] Gresnigt A.M. & Karamanos S.A. (2008). “Local buckling and deformation capacity of tubes in steel structures”,
Balkema, London.
[6] Elchalakani M., X.L. Zhao, R. Grzebieta (2004), Bending tests to determine slenderness limits for cold-formed
circular hollow sections, Journal of Constructional Steel Research (58), 1407-1430
[7] Lanhui Guo, Shijun Yang, Hui Jiao (2013). Behavior of thin-walled circular hollow section tubes subjected to
bending, Thin-walled structures (73), 281-289
[8] AS4100-1998 (1998) Steel Structures, Australian Standard, Standards Association of Australia, Sydney.
[9] Gardner, L., K.H. Law, G. Buchanan (2014), Unified slenderness limits for structural steel circular hollow
sections, Romanian Journal of Technical Sciences, Applied Mechanics, 59(1-2), 153-163
[10] Rotter, J.M. The new method of Reference Resistance Design for shell structures, Proc. SDSS 2016, Int. Colloq.
on Stability & Ductility of Steel Structs, Timisoara, Romania, pp 623-630, 2016.
[11] Rotter, J.M. Advances in understanding shell buckling phenomena and their characterisation for practical design,
in Recent Progress in Steel and Composite Structures, Eds M.A. Gizejowski, A. Kowlowski, J. Marcinowski and
J. Ziolko, CRC Press, Taylor and Francis, London, pp. 2-15, 2016.
[12] Sadowski, A.J. and Rotter, J.M. Extensive studies of capacity curves for tubes under global bending, Combitube
WP2 Report, Res. Rept RR12 02, Inst. Infrastruct & Environmt, Univ. Edinburgh, Edinburgh UK, 2012.
[13] AM-1-6-2013-13 Annex E.1 cylindrical shells under uniform bending, Amendment AM-1-6-2013-13 to EN
1993-1-6, Approved by CEN TC250 SC3, 2013.
[14] NORSOK N-004 Design of steel structures. Standards Norge, Lysaker, Norway. 2004.
[15] EN1993-5. Eurocode 3: Design of steel structures. Part 5: Piling. CEN. 2008
[16] Brazier, L.G. (1927), “On the flexure of thin cylindrical shells and other ‘‘thin sections’’. Proceedings of the
Royal Society. Series A 116. 104–114.
[17] Peters D.J., Broos E.J., Gresnigt A.M., Van Es S.H.J. “Local Buckling Resistance of Sand-filled Spirally Welded
Tubes”. The Twenty-fifth International Ocean and Polar Engineering Conference. ISOPE. 2015
[18] Peters D.J., missing reference to second Port of Rotterdam tests
[19] Rotter, J.M. and Schmidt, H. (eds) Stability of Steel Shells: European Design Recommendations: Fifth Edition
Revised Impression, Publication P125-2, European Convention for Constructional Steelwork, Brussels, 2013.
[20] Es S.H.J. van (2016), “Inelastic local buckling of tubes for combined walls and pipelines”. Delft University of
Technology.
[21] Gresnigt, A.M., Es, S.H.J. van, Bijlaard F.S.K. et al (2015), “Bending resistance of steel tubes in CombiWalls
COMBITUBE final report.
[22] Mohr W. (2003), “Strain-based design of pipelines”. Report Project No. 45892GTH to U.S. Department of
Interior. Minerals Management Service and U.S. Department of Transportation. Research and Special Programs.
EWI.

© Ernst & Sohn Verlag für Architektur und technische Wissenschaften GmbH & Co. KG, Berlin ∙ CE/papers (2017)

View publication stats