Advanced Finite Element Analysis in Structural Design

D Iosif
GHD Pty Ltd, Perth, Australia

ABSTRACT: This paper discusses the use of numerical methods (Finite Element Modelling ) in the civil and
industrial design. The applications detailed herewith employ nonlinear analysis capabilities, structural
dynamics and use various material models. Three representative studies are presented to illustrate the
versatility of Finite Element Analysis (FEA) in solving problems associated with faults in structural design,
design optimization and Standards’/Code compliance.

1 INTRODUCTION
1.3 Tank Vibration
1.1 Large Pipe Bends for Water Industry
High vibration levels were observed and recorded on
GHD has undertaken a number of Finite Element a spent surge liquor tank typical to Alumina
Analyses [7] on a range of large pipe segmented Industry. Such vibrations are known to lead to
bends on behalf of a Water Authority Client. The catastrophic fatigue failures and therefore must be
aim of this study was to develop an alternative contained.
design methodology in the belief that current GHD designed structural modifications to the
standards such as AS 4041 was producing tank based on the analysis of the natural frequencies
uneconomical designs. of the tank and its internals, stress levels, mode
The results of the study indicate that localised shapes and participation factors.
peak stresses approximately twice as much as the
normal pipe hoop tension develop on the inside of
the mitre cut joint (intrados). The major finding of 2 TECHNICAL APPROACH AND FE
this study was that the intrados stress was MODELLING
independent of the segment length, which is in
2.1 Large Pipe Bends for Water Industry
disagreement with current Standards in Australia
(AS 4041) and the United States (ANSI/AWWA A literature review of relevant design models either
C208). An alternative design methodology was currently or previously relevant to segmented pipe
proposed such that the peak stress on the inside of bends was undertaken as a preliminary phase. The
the mitre cut joint be limited to yield based on the key design codes and their approach to the design of
Von Mises Criterion for combined stress. mitre bends is essentially given by:
• AWWA C208 1996 [2] and 2001 [3]. These
standards recommend dimensions only and do not
1.2 Dubai Drydocks address working pressures or admissible stress.
GHD was contracted to conduct FE analyses to Reference is made to AWWA M11 [4] for dealing
investigate and assess the structural integrity of with these parameters.
Dubai Drydocks Caissons and provide a more • An Official Note to C208 [1] was given which
economical optimized re-design of the caissons’ recommends the following relationship be used for
walls. Complex analyses of Ultimate Loadcase R/D <2.5 (see Figure 1 for geometry)
Combinations concluded that a number of redundant PD D θ A
walls can be removed without affecting the t= ( tan + )
structural integrity of the Caissons and substantially ASa 3 2 2
reducing Client’s repairs and maintenance costs. where:
 design methods. Figure 2a shows the Working
Pressure vs. Segment Length A for a bend of OD
P = pressure 1400 and t=11 mm. This figure compares the above
standards with the FEA analysis and the current
A = segment length as in Figure 1 Water Authority standard rating.
D = diameter

Sa = admissible stress (0.5 x yield stress of material)

θ = segment angle as shown in Figure 1

Figure 1. FE model and terminology

• For R/D = 2.5 and above, the bend can be treated
as a straight pipe with admissible stress given by
maximum admissible hoop stress. In terms of
maximum admissible hoop stress, AWWA M11 [4]
sets this limit to 0.5 x yield stress based on steady
state pressure while ANSI/ASME B31.3 and AS
4041 sets this limit to a maximum of 0.67 x yield
stress based on full working pressure.
• According to AWWA M11 [4] a safety factor of
at least two to ultimate tensile stress is
recommended for maximum working pressure. In
other words a bend designed for Working Pressure
“WP” should not reach UTS at pressure “2 x WP”.
AS 4041 - Appendix I nominates a safety factor on
UTS of 2.35.
• AS4041-1998 Pressure Piping Clause 3.15.4.3
recommends multiple mitre bends angle of cut to be
less than 22.5° and allowable working pressures to Figures 2a, 2b, 2c: Standards’ recommendations
be the lesser calculated from two equations.
Equation (1) is independent of the segment length The curves represented in Figure 2a, 2b and 2c show
and is the most limiting and generally to less than quite different allowable maximum working
half of normal hoop stress. Equation (2) is based on pressures depending on which method of analysis is
the same parameter as C208 (1) above and is used.
dependent on segment length A. In Figure 2a the plots show that for AS 4041, the
equation 1 criteria in all cases set the maximum
In the light of the abovementioned Standards, admissible pressure. This equation does not take
calculations have been performed to plot various into account the dimension A in the segmented bend.
The two upper curves are based on AWWA C208 - the bond between the rebar and the concrete, the
M11 and AS 4041 equation 3.15.4.3(2). These relative size of the concrete aggregate compared to
relationships include in their formulation: segment the rebar diameter, and the mesh. A reasonable
length A and the dependency on bend radius R ie starting point for relatively heavily reinforced
(R=f(A)). concrete modeled with a fairly detailed mesh is to
assume that the strain softening after failure reduces
Three distinct diameter bends were analysed for the stress linearly to zero at a total strain of about
various segment length “A” including: 10 times the strain at failure. The strain at failure in
• FEA model 1: OD 1400, t=11mm, A=400 mm standard concretes is typically 10-4, which suggests
• FEA model 2: OD 1400, t=11mm, A=1050 mm that tension stiffening that reduces the stress to zero
• FEA model 3: OD 1400, t=11mm, A=725 mm at a total strain of about 10–3 is reasonable.
• FEA model 4: OD 1000, t=8mm, A=300 mm
• FEA model 5: OD 1000, t=8mm, A=600 mm
• FEA model 6: OD 1000, t=8mm, A=900 mm
• FEA model 7: OD 500, t=5mm, A=200 mm
• FEA model 8: OD 500, t=5mm, A=400 mm The choice of tension stiffening parameters is
• FEA model 9: OD 500, t=8mm, A=600 mm important in ABAQUS/Standard since,
These nine models were analysed for a wide range generally, more tension stiffening makes it
of internal working pressures (WPs). Since some easier to obtain numerical solutions. Too little
WPs produced stress above the yield point of the tension stiffening will cause the local cracking
material, the non-linear module of AbaqusTM was failure in the concrete to introduce temporarily
used. unstable behavior in the overall response of
The following material properties were assigned
the model. Few practical designs exhibit such
to the models:
Young’s modulus 210,000 MPa behavior, so that the presence of this type of
Poisson’s ratio 0.29 response in the analysis model usually
For OD 1000 and OD 500 bends: indicates that the tension stiffening is
Yield stress =300 MPa with strain hardening to UTS unreasonably low.
410 MPa
For OD 1400:
Yield stress 250 MPa with strain hardening to UTS
410 MPa
The models consist of shell elements with 6 degrees Figure 3: Tension stiffening model
of freedom and five integration (Gauss) points
across thickness. The FE models were built with
pipe legs of approximately 10 times the diameter of As the concrete cracks, its shear stiffness is
the bend and were constrained as shown in Figure 1. diminished. This effect is defined with the SHEAR
RETENTION option by specifying the reduction in
2.2 Dubai Drydocks the shear modulus as a function of the opening strain
across the crack.
Effects associated with the rebar/concrete interface,
such as bond slip and dowel action, are modeled A number of two caisson types are presented in this
approximately by introducing some “tension paper as seen in figures 6 and 7. The complexity of
stiffening” into the concrete modeling to simulate the rebar reinforcement is illustrated in Figure 5.
load transfer across cracks through the rebar. This is to our knowledge, the first full 3D Finite
Element analysis attempted in the world for the
The postfailure behavior for direct straining across modeling of large dry-dock Caissons which takes
cracks is modeled with the TENSION STIFFENING into consideration all aspects of steel-concrete
option, which allows the user to define the strain- interaction under combined loading.
softening behavior for cracked concrete. This is
achieved by specifying a postfailure stress-strain The primary loads were as follows:
relation or by applying a fracture energy cracking 1. Earth Pressure
criterion.
Postfailure stress-strain relation 2. Internal Water Pressure
Specification of strain softening behavior in 3. External Water Force
reinforced concrete generally means specifying the
postfailure stress as a function of strain across the a. One Side (i.e. dock empty)
crack. The tension stiffening effect depends on such b. Both sides
factors as the density of reinforcement, the quality of
4. Deck Surcharge
5. Hauling-in (one side)
a. Position 1 (one end of caisson) –
Force perpendicular to hauling in
beam toward edge
b. Position 2 (Central) – Force
perpendicular to hauling in beam
toward edge
c. Position 3 (opposite end of caisson to
Position 1) – Force perpendicular to
hauling in beam toward edge
d. Central – Force along hauling in
beam, up page direction
6. Crane load
Figure 4: Solid model
7. Ship load
8. 150 T Bollard load
a. 150 T force applied perpendicular to
cope beam towards edge
b. 150 T force applied along cope beam,
down page direction
9. 250 T Bollard
a. Each bollard, 250 T force applied
perpendicular to hauling in beam,
away from edge
b. Each bollard, 250 T force applied at
45 degrees to hauling in beam, away
from edge, up page direction.
c. Each bollard, 83 T force applied
Figure 5: Detail showing reinforcement (rebars) in
perpendicular the hauling in beam,
walls and bollard
towards edge
d. Each bollard, 83 T force applied
along hauling in beam, up page
direction
10. Capstan
a. 25 T applied perpendicular to hauling
in beam, towards edge
b. 25 T applied perpendicular to hauling
in beam, away from edge
c. 25 T applied along hauling in beam
up page direction.
11.Fender Load
Figure 6: Finite Element model of Caissons 1-1 (1). • The vibration experienced by the tank shell
Rebars were modeled as embedded elements. below 8 Hz suggests the existence of
“forced vibration”.

• When the level of the liquid inside the tank is

at 14 m, the first natural frequency of the
tank shell in the absence of attached baffles
is just above 8 Hz.

• Baffles’ normal modes start from as low as 2

Hz. The baffles can therefore be excited at
f >2 Hz and reach resonance if forcing
frequency is the same with any of the baffles’
natural frequency. As baffles’ natural
frequencies between 2 and 8 Hz are
numerous and closely spaced, with large
participating effective mass, the resonance of
the baffles becomes possible.

Figure 7: Finite Element model of Caisson 1-1(4). We can therefore assume that the forced vibration
Interaction rebar-concrete modeled. that excites tank’s shell below 7-8 Hz can occur:
a) Internally: via the baffles which start to vibrate
under a forcing harmonic load F=F0sin(ωt)
produced by the liquid intake-discharge, sloshing
2.3 Tank Vibration or air pressure pulses
b) Externally: via the piping system attached to the
This project aimed at developing structural external shell of the tank vibrating at forcing
modifications to the tank structure (Figure 8) in frequencies.
order to contain tank’s strong shell vibrations below c) as a combination of both
8 Hz.
The immediate aim was to shift baffles’ natural
frequencies to higher values as currently they are
free to vibrate at frequencies as low as 2 Hz with
modes very closely spaced in frequency domain,
large modal participation factors and effective mass.

According to the Client’s vibration measurements of

the tank shell, large frequency peaks were recorded
at frequencies below 7 Hz.

3 RESULTS
3.1 Large Pipe Bends for Water Industry
The results of the nine FEA models are summarised
as follows:
• The peak stress occurs over a narrow bond of
the inside joint of the mitre cut, termed the
intrados. The peak stress is approximately
double the normal hoop stress.
• The development of stress at the intrados is
Figure 8: Finite Element model of the tank and its independent of the segment length A, for A
internals (deflected shape) greater than 0.25OD. Therefore an increase
in segment length would not result in any
At this point it is worth mentioning the following: economies. As the stress at the intrados
indicates localised outward deformation
under pipe pressure it is not surprising that
 extending the segment length has no effect at Figure 9: Von Mises stress OD100 A300mm
the intrados.
• Stress at intrados was found to be dependent
of A only for values of A<0.25OD
• The FEA results differ from the
ANSI/AWWA C208 Technical Note and
AS4041 – 3.15.4.3 Equation 2 where
allowable stresses are dependent on the
Segment Length A as well as diameter,
thickness, mitre cut angle and working
pressure.
• The existing requirements in WA on
segmented bends is slightly under designed
for OD1400 (figure 2a) but conservative for
OD1000 and OD500 (figures 2b and 2c) if
the Von Mises Criterion is adopted.

Figures 9-11 present the results of the FEA of a

1000OD bend, A300mm (6a) and A900mm (6b).
The Von Mises stress of 300 MPa is reached the
same WP = 2.0 MPa regardless the dimension of
segment “A”. This shows that the bends in this class
are capable of sustaining a higher internal pressure
than the one current in WA where the standard
design specifies WP = 1.8 MPa (figure 2b) and
substantially higher (32%-38% higher) than the WP
of recommended by AS 4041 and AWWA M11. The Figure 10: Von Mises stress OD1000 A900mm
old practice based on the Standards was to increase
segment “A” in order to allow an increase in internal
pressure. This procedure has been now abandoned in
the light of recent results. Figure 6c shows the
complex deformation pattern in the bend under
internal pressure magnifies 500 times.

Figure 11: Deformation pattern (x500 times)

3.2 Dubai Drydocks

Some of the displaced patterns for the two caissons
are presented in Figures 12 and 13.
Figure 12 shows the displaced pattern under three
primary loads namely Earth Pressure, Internal Water
pressure and deck surcharge while Figure 13 shows
the deformation pattern of Caisson 1.1-(4) under end of the removal step. The forces are ramped
Ultimate Load Combinations down gradually to ensure that element removal has a
LC_A= 1.4 x (L1 + L2 + L3a) + 1.6 x (L6 + L9a + smooth effect on the model. No further element
L10b) + 2.0 x L11 calculations are performed for elements being
where loads L# are the primary loads defined in removed, starting from the beginning of the step in
Section 2.2 which they are removed.
The simulation of the Ultimate State loadcases
showed that the removal of certain walls may affect
the structural integrity of caissons while the removal
of others is perfectly safe. The walls were
“remodeled” in such a manner to allow compliance
with Standards.

Figure 12: Caisson 1-1(1). Displaced pattern under

Loads 1+2+4

Figure 14: LC_A. Maximum principal stress

3.3 Tank Vibration

3.3.1 As-Is Tank

Tank analysis filled to 14 metres and all internals

present.
The Von Mises stress levels obtained as a result of
hydrostatic pressure and self weight are below
admissible stress levels. Maximum stress in the tank
shell is around 80 MPa; higher values are predicted
Figure 13: Caisson 1-1(4). Displaced pattern under around the bottom nozzles.
Ultimate Loadcase Combination The dynamic analysis of the tank predicts natural
frequencies which are very closely spaced. At low
The maximum tensile stress in concrete for LC_A frequencies between 2 and 8 Hz, the baffles tend to
is localized over a small area as shown in Figure 14. vibrate excessively due to a high number of closely
The concrete relies on reinforcement in the areas spaced normal modes. Tank’s shell also vibrates in
where the tensile stress is larger than 3MPa. Based the areas adjacent to baffles. The baffles vibrate
on maintenance and operational consideration, either independently (<5 Hz, Figure 15) or
certain walls in the caissons were to be removed and simultaneously all three or in pairs of two (>5-7 Hz).
the existing walls remodeled. In the analysis, just At frequencies above 8 Hz tank’s shell vibrate all
prior to the removal step, the FEA code stores the around its circumference which suggests that the
forces/fluxes that the region to be removed is vibration in the shell is not controlled only by the
exerting on the remaining part of the model at the baffles.
nodes on the boundary between them. These forces These results agree very well with the results
are ramped down to zero during the removal step; described derived in previous work when the baffles
therefore, the effect of the removed region on the were removed to independently derive the natural
rest of the model is completely absent only at the frequencies of the tank’s shell.
Tank analysis filled to 5 metres and all internals Liquid level 14 m
present. The advantage of the proposed design is that the
As in the previous case it is predicted that at low tank and its baffles do not have any natural
frequencies (below 5 Hz), the baffles have the frequency under 5.5 Hz.
largest vibration amplitudes. Tank’s shell vibrates
around and in line with the baffles. The baffles’ first natural frequency is at 5.5 Hz,
At frequencies above 5 Hz (see Figues 16) the followed by 6.5 and 6.8 Hz. All these 3 modes are
shell vibrates quite independently (of the baffles) localised in nature (occur at the baffles edge close to
around its circumference as baffles continue to the tank wall and between two consecutive existing
display large mode shapes. stiffeners) and unless excitation force (of a nature
unknown to us) is extremely strong, resonance of the
baffles and subsequent vibration of the shell will be
Tank analysis filled to 14 metres and no baffles.
non-existent.
The objective of this analysis was to derive the
natural frequencies of the shell in the absence of the From the measurements taken by Client, forcing
baffles as in previous two analyses it was difficult to frequencies above 7 Hz induce relatively low levels
ascertain how the tank’s shell behaves without being of vibration (relatively low peaks) so that stiffening
excited by the baffles. the baffles using connecting plates was found to be a
The first natural frequency of the tank is at 7.8 Hz feasible solution.
and occurs in the roof only. This correlates
accurately with the natural frequency of the roof However, the baffles and adjacent tank shell have
measured by the Client. There are no natural two normal modes at 7.2 and 9 Hz which can be
frequencies of tank shell below 7.8 Hz. excited. If these two are not excited, the next
Tank’s shell natural frequencies occur above 8 dangerous frequency is at 9.6 Hz, usually too high to
Hz and are not as closely spaced as in previous two produce considerable vibration in the tank shell.
cases. Indeed, subsequent measurements taken over a
Modal participation factors are high which shows period of one year did nor record any vibrations on
that tank’s shell can be excited between 8-10 Hz the stiffened tank.
independent of the baffles’ vibration. This is a very
important result which suggests that both the baffles The absence of tank’s shell circular rings (stiffeners)
and tank shell should be stiffened. is reflected in some loss of structural stiffness. If
only the baffles are stiffened, the baffles and tank
3.3.2 Stiffened Tank shell can vibrate strongly starting from f = 6 Hz
whereas in the presence of the circular rings the
A new stiffening system was proposed and shell can vibrate more predominantly after 9 Hz if
subsequently implemented in order to reduce excited by forcing frequency.
baffles’ vibration and shown in Figure 17a,b,c.

• tank shell stiffeners: The shell of the tank Liquid level 7 m

was stiffened at elevations 9000mm,
13000mm, 14500mm, 16250mm and 17750 For a 7 metres liquid level the baffles vibrate and
mm. The stiffening system comprised of can easily be excited at values as low as 6 Hz
100x100x6 L profiles radially welded to the despite stiffening the shell and baffles.
tank shell through the use of 60 plates The tank shell vibrates freely with the baffles if
(250x150x10). In doing this the radial space forcing frequency is present and resonance of the
between two consecutive plates is 583 mm tank shell at around 7 Hz can easily be reached.
which is less that the minimum recorded From the structural dynamic point of view it is
width of a vibration amplitude. not recommended to run the tank with a liquid level
well below 14 m. In practice, the level of liquid is
• baffles stiffeners – through the use of a not allowed below 14 m level with running pump.
rectangular plates 550x1250x12 welded on
the passive side of the baffles, spaced at 1100
mm from one another mm connecting the
baffles with the tank shell
Figure 15: Localised natural frequency of baffles

Figure 16: Natural frequency of the tank shell at 5

Hz when tank is emptied to 5 m liquid level.

Figure 17 (a,b,c): Structural modifications to the

baffles and tank’s shell.

3.4 References
1. American Water Works Association, Official Note
Addendum to ANSI/AWWA C208 – 83 Standard for
Dimensions for Fabricated Steel Water Pipe Fittings
2. ANSI/AWWA C208-96 AWWA Standard for
Dimensions for Fabricated Steel Water Pipe Fittings
3. ANSI/AWWA C208-01 AWWA Standard for
Dimensions for Fabricated Steel Water Pipe Fittings
4. AWWA M11 Manual of Water Supply Practices.
Steel Pipe – A Guide for Design and Installation
5. AS 4041. Pressure Piping ASME B31.3 – 1996
Edition. Process Piping
6. API Recommended Practice 1111 – Limit State
Design
7. AbaqusTM version 6.4.1: ABAQUS Inc.

3.5 Conclusions
Finite Element Modeling constitutes a valuable tool
in the design and analysis of various structures to
Standard requirements. The usefulness of FEA
extends to exploring new design alternatives and
evaluating multiple design approaches that entail
complex loading, nonlinear materials and complex
geometries. The FEA programs can concentrate on
that handful of crucial variables adjusting them up to
find an optimum solution.

The future
The next step beyond “design optimization” is
“behavioral modelling”. As the software explores
multiple combinations it organises the results in a
tree-like structure of cause-and-effect relationship.
Major variables are represented as “big branches”
while sub-branches show the options available using
various parameters. Such design approach can be
databased and overall design can be revised almost
instantly.