THE LOSS OF THE SLEIPNER CONDEEP PLATFORM

I. Holand

SINTEF Structures and Concrete, Trondheim, Norway

Abstract. The Sieipner A Condeep Platform sprang a leak and sank under a ballasting test
operation outside Stavanger in August 1991. Model studies in a water basin with leakages in
different cells showed that leakage into a specific cell (cell 03. Fig I) is capable of explaining
the sinking scenario. When this result was compared with strength calculations. major cracking
in the walls of cell 03 was found to be the probable cause of the accident.

The general approach to ensuring structural safety is to base the design on criteria specified in an
accepted set of rules and standards. where unavoidable uncertainties are taken care of by safety
coefficients. and to avoid errors by quality assurance and control. Design criteria had been
specified according to regulations issued by the Norwegian Petroleum Directorate [8) and in a
Norwegian Standard [7). (To some extent previous editions were used.) 'TIle design criteria were
not found to be the reason for the accident.

A check of the global response analysis revealed serious inaccuracies in the interpretation of
results from the finite element analyses. giving a shear force in a critical section in cell walls that
was less than 60 % of the correct value. In addition. the reinforcement provided in specific cell
joints was inadequate. 'TIle reasons for tJle accident are therefore very likely to have been found.
Full-scale tests of the cell wall joint were undenaken. and the tests confirmed the strengths found
by calculations after the accident.

'TIle investigation demonstrates that current procedures for calculating stress resultants at
boundaries are not robust when irregular elements are used. Ways of improving postprocessing
routines for analysis and design of concrete structures are discussed.

1 The Sleipner A platform. Loss and investigations

The gravity base structure of the Sieipner A platform is a concrete structure placed at a depth of
82 m in the Non]l Sea for production of hydrocarbons. 'TIle platform (Fig I) is of the Condeep
type and has a base structure consisting of 24 cells. with a total base area of 16 000 m2• Four cells
were elongated to shafts supporting the platform deck. 'TIle triangular spaces between the main
cells are denoted tri-cells. 'TIle total concrete volume is 75 000 m 3• 'TIle first Condeep platform
(Beryl A) was placed in the Nortll Sea in 1975. and Sleipner A was number 12 in the Condeep
series.

Sleipner A sprang a leak and sank under a controlled ballasting operation in Gandsfjorden outside
Stavanger. Norway on 23 August 1991. 'TIle accident occurred when the platfoml had been
lowered to a depth of about 99 m. The personnel on board heard a bang and observed that water
25

GM.A. Kusten and MAN. Hendriks (edr.), DIANA Computational Mechanics '94, 25-36.
@ 1994 Kluwer Academic Publishers.
26

99.0 m
::us

32.0 m

O.Om

Detail A
Tri-cell

Fig 1 Sieipner A Gravity Base Structure. Water levels at failure

 27

was flowing into tIle D3 shaft. TIle water level was then about 32 m in D3. resulting in a water
head on the tri-cell walls below this level of 67 m. The top of the D3 shaft disappeared 17-18.5
minutes after the first strong bang was heard. On board the platform were 14 men, who all were
rescued by boats close by. Nobody was hurt.

TIle contractor (Norwegian Contractors) set up an Investigation Committee on the same day [4].
At the same time the operator (Statoil) established its own Investigation Group. SINTEF was
contracted by Statoil to support its investigation group by carrying out technical studies of the
reasons for tlle loss of the Sieipner A platform. Some of the conclusions of this study may be
found in [2] and [9]. TIlere are no divergences in the technical findings in the studies by the two
investigation bodies.

The platform was completely destroyed when it hit the bottom. A new platform was built in 1991-
92. for the most part using identical geometry but with an increased reinforcement in critical parts.
I! is now in production.

2 Sinking scenario and leakage estimation

Numerical simulations of the events from the point at which the first rupture occurred until the
tirst shaft top (D3) disappeared have been performed. Several parameters have been varied, and
the results from tlle simulations have been compared with observations. The comparisons have
shown that only a leak into the D3 shaft gives a simulated sinking that conforms well with the
observations.

For the top of shaft D3 (0 disappear in tlle observed time of about 18 lnin, the most probable leak
rate at the start of sinking is from 7.8 to 8.1 m3/s. As the difference in water level falls the leak
rate is reduced to the range 7.2 to 7.7 m3/s as the top of shaft D3 vanishes.

TIle possibility that the leakage path was through a fractured tri-cell into the D3 shaft was paid
particular attention. TIle inflow is limited by a concrete slab at the top of the tri-ceIl. through
which passes a pipe which limits the flow through one tri-cell to a rate somewhat lower than the
inflow deduced from the sinking process. Accordingly. two tri-cells have been assumed to be
cracked. with one half of the inflow through each.

A model of the D3 shaft and two tri-cells was built to model length scale I:60 and tested by
SINTEF Norwegian Hydrotechnical Laboratory. Cracks were placed in the partition walls between
the Iri-ceIl and the D3 shaft. near the corners of the tri-cells. All tests produced an observed
sinking time of 18 minutes. During tlus time a rise in the water level in D3 from the initial level
of 32 m was observed. and the corresponding leakage into 03 was calculated.

TIle model study contirmed that one crack. a few centimetres wide and several metres long. in
two of tile tri-cells adjacent to 03. crossing the original water surface in drill-shaft D3. can
explain all observations made during sinking.

The period from when overtopping of the shaft D3 began until the structure hit the sea floor
lasted nearly two minutes. For most of tlus time tlle structure was not entirely submerged. but was
being weighted down by the inflow. Once it disappeared below the surface, the structure
accelerated and reached the bottom in some 45 to 50 seconds. When the platform lut the sea floor.
at a depth of 180 m. the calculated velocity was 4.4 m/s. A report from a search of the site of the
loss shows that the shafts have broken off and lhe raft has imploded completely. Most probably
28

one or more concrete cells imploded just prior to the actual impact. about 20 m above the bOllom.
A video of the remains of the structure shows extensive crushing and spaIling. The event was
registered at seismological stations in Norway. and the computed magnitude of the impact
corresponds roughly to 3 on the Richter scale.

3 Finite element analyses

The platform design was analyzed using the program system MSC/NASTRAN Version 65 C and
the basic 8-noded tlnite solid element (CHEXA-8).

Fig 2 shows in the lower right corner the element mesh that was used. as generated by the
program. As Fig 2 shows. irregular elements were used in the tri-cell waIl. This is not in
accordance with normal practice, which is to modify the generated elements manually to avoid
irregular elements in highly stressed regions. TIle main tigure shows a planar tinite element model
that was used for a local analysis of the tri-cell in the investigation. The axial force N was given
values taken from the global analysis.

'L 1
2 x 1095 mm

z x

I~
3x964mm
•

,/'
N

Fig 2 Generated element mesh for tri-cell. Planar element model

 a, =2.19MN Pressure Pressure
q = 1 MPa q = 1 MPa
O2 = 2.059 MN
M" = 1.13MNm

a,x I Q2x

qI2.2.12MNm
8

,
,
\
, \,
I \ : B\;'
2059 I __ ~ __' ~A
O.60~ 0.605
I - 0.59 0.680

Parabola through A, B and C glvas: NASTRAN:

1.13 + 0.68 = 1.81 MNm
181 I 2.12 • 0.85

a,x=1.28MN a,X I a, = 0.59

Parabola through A, Band C givas M = 1.13 MNm
al= 1.28MN al,a 2 = 0.62

Fig 3 Computed shear force [MN/m] Fig 4 Computed moment [MNmlm]

~
30

Figs 3 and 4 show the results obtained for shear forces and moments using the NASTRAN
program for an analysis of the simplified model. Values in the midpoints were used and
extrapolated to the boundary.

Fig 3 shows considerable errors for shear forces in the mid-points of the elements. TIle parabolic
extrapolatjon to the boundary enlarges the error, with the result that the boundary shear force is
about 60 per cent of the beam theory solution. Fig 4 shows the corresponding results for the
moment. In tlus case, the clamping moment at tlle boundary depends on the representation of
stiffness. Hence, the sum of calculated moments at the boundary and at mid-span has been
compared with the sum p12/8 according to equilibrium, and the sum is found to be 15 % too
small.

The deviatjons from equilibrium are caused by irregular elements and by the extrapolation
scheme. A regUlar mesh will give the beam theory solution.

4 Strength of tri-cell walls and corners

The strength of the tri-cell walls and the corner areas where the tri-cell walls meet were analyzed
by fonnulas in standards, strut-and-tie models and fillite element analyses.

Full-scale models of a critical portion of tlle corner were also tested. A tri-cell and a specimen
with loading arrangement are shown in Fig 5. Water pressure in cracks is an important component
of tlle loading and was simulated by a modification of tlle ratio between tlle loadings from jacks
Hand V2 (see Fig 5) adjusted to a gradual increase of span from 4.4 to 5.8 m during the tests.
The lengths of 4.4 and 5.8 m correspond to the distances between tlle two inner and outer corners
of tlle tri-ceJl walls, respectively (see Fig 5). TIle crack propagation appears from the photos in
Fig 7.

TIle height of the specimens was chosen to be 680 mm, equal to about four times the distance
between the reinforcement units in the original design. TIle external dimensions shown in Fig 5
were the same for all specimens.

TIle test programme included ten specimens (Fig 6). Seven specimens (Y 1-Y6 and Y8, Fig 6a)
were reinforced as built, one (Y7, Fig 6a) with elongated T-headed bars in tlle haunch, two (Y9
and Y 10, Fig 6b) with comprehensive modifications to the reinforcement, according to a new
design of the structure. TIle experimental programme was worked out in cooperation between
SINTEF Structures and Concrete, Norwegian Contractors and Statoil. SINTEF was responsible
for the loading arrangement and the testing procedure.

Tests Willl such heavy specimens are a challenge witll respect to tlle loading equipment. Jack
capacities up to 1500 and 500 tons in the verucal and horizontal direction, respectively, were
needed to simulate realistic loadings of the specimens. TIle only place in Scandinavia Willl such
an actuator for verucal loading is Statens Provllingsanstalt in Bods, Sweden, and the tests were
performed there.
 31

'
VL
800
..,.l:./

.,."
..,.. .'ll
--specimen

Detail A (fig 1)

Fig 5 Test specimen and loading arrangement [3]

Specimens Specimen
Yl - Y8 Y9 u12c1c170
u25c1c170

u25c1c170

a b

Fig 6 Reinforcement in Y2. Y7 and Y9 (defonned bars K500TS with yield strength of
550 MPa) [3]
32

Y2

Y7

Y9

Fig 7 Cracking of three specimens denoted Y2. Y7 and Y9, after failure. Numbers on
specimens times ten correspond to water level at initiation of cracks [3]
 33

Two different situations are of interest:

Failure situation; to compare observed water head of 99 - 32 = 67 m at failure with

simulated water head causing rupture in test specimens

Design situation; submerged platform during mating of deck, water head 72.5 m, mainly
to verify redesign

A main problem in the "as built" design was that T-headed bars (see Fig 6a) used as transverse
reinforcement in the haunch were too short.

For YI-Y6 and Y8, which were designed to be the same as the original structure, and which are
referred to as the 'as built' specimens, the test variables were: the axial force level in the tri-cell
walls, the combination of moment and shear action in the joint, and the concrete strength. In Y7,
the T-headed bars were elongated 10 the compressive reinforcement at each side.

Specimens Y9 and Y 10 were related to the new design of the joint. The main difference compared
to the other specimens was the amount of shear reinforcement in the joint region.

Only one diagonal crack was observed at each side of the joint for the 'as built' specimens. These
cracks always passed just behind the anchorage plates of the T-headed bars. The initiation of the
cracks was very sudden, without any warning. The cracks immediately propagated the whole
distance between the upper loading point and the outer corner. The diagonal cracking levels
shown for Y7 correspond to initiation of cracks outside the longer T-headed bars.

TIle load level at initiation of diagonal cracks was higher for Y9 and Y 10 than for the "as built"
specimens, due to the improved detailing of the haunch reinforcement. The increased
reinforcement also resulted in more than one diagonal crack. TIle residual strength after cracking
was also considerably larger for Y9 and Y 10 than for the "as built" specimens. The capacities of
Y9 and Y 10 were satisfactory with respect to the design assumption.

TIle crack propagation and tlnal failure mode observed in the tests were also compared with
results of a non-linear FEM analysis. TIle analyses performed with stirrups show that the stresses
in the stirrups are low and show a nearly linear increase up to a level where the stresses rise to
yield stress within a few load steps with, in most cases, subsequent termination of the program.
TItis kind of behaviour is typical for structural parts without sufficient shear reinforcement.

Representative failure loads expressed in tenns of water heads are indicated in Fig 7 as:

failure load as built; 70 m as compared with Sleipner A failure at 67 m

failure load with elongated T-headed bar; 125 m and with reinforcement as redesigned; 155
m as compared with design load 72.5 m

TIle tests contlrmed that failure of one or more tri-cell joints was the most probable reason for
the accident, and that careful reinforcement design is crucial tor the strength of the joint.
34

5 Alternative method for calculating stress resultants

The errors made in the analysis of shear forces in the Sleipner A tri-cell demonstrated that current
methods may be unsatisfactory when irregular elements are used. It is imperative to avoid errors
of tllis kind. An obvious remedy wllich is frequently used is to avoid irregular elements in critical
sections. TIle utilization of irregular elements in tile critical section in the Sleipner case was not
in accordance with established practice.

Strength checking of reinforced concrete is based upon senli-empirical design formulae for
sections through structural elements. taking into account cracking and other non-linear effects.
mainly based on reduced-scale experiments for beams and columns. TIlese formulae are specified
in codes and standards. but have also been supplemented by special rules in the post-processors
to make them applicable for shell sections with three in-plane stress resultants. three bending
moments and two out-of-plane shear forces [I]. TIlUS. tile action effects are also needed in the
form of stress resultants to check that computed action effects do not exceed the strength at' the
reinforced concrete section.

However. the analyses do usually not provide stress resultants directly. In current usage, solid
elements are preferred for modelling of marine concrete structures. not because detailed linear
stresses are needed. but because the modelling of a complex geometry and the application of
water pressure on external surfaces is facilitated using solid elements. TIle procedure of computing
section forces referred to points on a shell nliddle surface then will be as follows:

TIle nodal displacements tram the global analysis for tlle load combination in question are
used to compute strains in each element in accordance with the interpolation functions
Stresses are computed from the constitutive law
Stresses in Gauss points are integrated over the thickness to stress resultants
Where design values at the element edge are needed. stress resultants at Gauss points are
extrapolated to the edge

TIle errors made in the Sleipner tri-cell demonstrate that even though trus approach may be
convenient. it is not robust enough to provide reliable results when used uncritically.

On the other hand. stress resultants wllich undoubtedly are in equilibrium with external loads are
available as element nodal forces. Section stress resultants based on nodal forces are thus very
robust in the sense that they always satisfy equilibrium. TIle arguments are here referred to a shell
surface. but are even more valid in irregular regions. as in ring beams at the top and bottom of
cell walls.

The idea of utilizing the nodal forces for dimensioning purposes was discussed in the infancy of
finite element methods. but has since remained in the background. Researchers at the Department
of Structural Engineering at the Norwegian Institute of Technology. however, partly because of
tlle experience from Sieipner. did start looking more closely into practical applications of such
principles [5.6J. The approach is described in some detail in [6). which is presented at the present
conference.

TIle uncertainty in distributed stress resultants based on nodal forces relates to the correctness of
nodal forces. and to the local distribution of nodal forces into distributed stress resultants related
to a shell surface. Of these two sources of uncertainties. the correctness of the nodal forces is
considered to be of greater interest. Errors in nodal forces are mainly due to two sources:
 35

inadequate representation of linear stiffness by the finite element model. and deviations in the
constitutive laws caused by the nonlinear behaviour of reinforced concrete.

When distributed stress resultants are based on nodal forces. it is obvious that the only concern
of the linear analysis is that the model should provide a correct representation of stiffness. All
comparisons of strains and stresses which have worried us for many decades are irrelevant.

TIle dimensioning process violates the global analyses. which are exclusively based on a linear
theory of elasticity. A linear analysis is used because of the complexity of a nonlinear analysis.
and even more because of the need to use the principle of superposition to find the most
unfavourable combination of load cases for the section in question.

In a subsequent step critical portions may be studied by local non-linear analyses. Since non-linear
analyses are also iterated to equilibrium. the approach also provides a robust representation of
distributed stress resultants in a non-linear analysis.

Thus. if stress resultants are based on nodal forces. the results of the analyses by the skew
elements shown in Figs 3 and 4 will. in spite of the inaccuracies in strains and stresses. give the
linear beam distribution for shear force (Fig 3) and the correct sum of moments q12/8 (Fig 4).
36

References

Grosch. H.. Brekke. D.-E.. Aldstedt. E.: Current Practice for Design of Offshore Concrete
Structures. Proceedings of tlle Third (1993) International Offshore and Polar Engineering
Conference. Singapore. 6-11 June 1993.

2 Gudmestad. O.T.. Holand, I.. Jersin. E.: Handbok i prosjektering av betongkonstruksjoner

til havs (Manual for Design of Offshore Concrete Structures). Statoil 1993 (in Norwegian)

3 Jakobsen. B, Gausel. E. Stemland, H. Tomaszewicz, A: Large Scale Tests on Cell Wall

Joints of a Concrete Gravity Base Structure. Third International Symposium on Utilization
of High Strength Concrete. Lillehammer 1993.

4 Jakobsen, B: The Loss of the Sleipner A Platform. Proceedings of the Second (1992)
International Offshore and Polar Engineering Conference. San Francisco. USA. June 1992.

5 Kvamsdal. T, Mathisen. K M. Okstad. K M. Aas. H B: Techniques for Reliable Calculation

of Sectional Forces. Department of Structural Engineering. University of Trondheim. The
Norwegian Institute of Technology. Report no. R-19-93. Trondheim 1993.

6 Kvamsdal. T. Matllisen. K M: Reliable Recovery of Stress Resultants. Proceedings of First

International Diana Conference on Computational Mechanics. Delft. The Netl1erlands 1994.

7 Norges standardiseringsforbund: NS 3473E Concrete structures. Design rules. 4tl1 edition.

Nov 1992.

8 Norwegian Petroleum Directorate: Regulations concerning loadbearing structures in tl1e

petroleum activities. 1992.

9 Rettedal. W.. Gudmestad. O.T.. Aarum. T.: Development of Concrete Platforms after
Sleipner A Sinking. OMAE 93. Glasgow June 1993.