Proceedings of OMAE2006

25th International Conference on Offshore Mechanics and Arcticof

Proceedings Engineering
OMAE2006
June 4-9, 2006, Hamburg, Germany
25th International Conference on Offshore Mechanics and Arctic Engineering
June 4-9, 2006, Hamburg, Germany

OMAE2006-92591
OMAE2006-92591

DESIGN OF PRESSURE HULLS USING NONLINEAR FINITE ELEMENT ANALYSIS

John R. MacKay
Defence Research and Development Canada – Atlantic
Delft University of Technology

Malcolm J. Smith
Defence Research and Development Canada – Atlantic

Neil G. Pegg
Defence Research and Development Canada – Atlantic

ABSTRACT the safety factors to account for non-addressed design issues

Through the use of nonlinear finite element analysis
(NLFEA), submarine pressure hull designs could potentially
be based on calculated limit states that include the full geo-
metric complexity of the structure, and real-world effects such
as build imperfections. In addition, NLFEA could provide a
rational means of assessing the effects of in-service damage on
structural performance. Analysis of pressure hulls using 3D
NLFEA is not currently supported in design codes, primarily
because the uncertainty regarding the accuracy of the method
has not been quantified. Defence Research and Development
Canada (DRDC), the R&D branch of the Canadian Navy, is
undertaking work to develop a partial safety factor for 3D
NLFEA of pressure hulls, by comparison of numerically cal-
culated collapse pressures to experimental results. Data from
experiments previously conducted at various institutions will
be augmented by a pressure hull testing program currently
being undertaken by a joint project of DRDC and the Ministry
of Defence of the Netherlands. The development of NLFEA
modeling and analysis guidelines, as well as a revised de-
sign/analysis procedure, will be discussed, as well as a history
of the DRDC submarine structures research program.
INTRODUCTION
The design of externally loaded pressure vessels, such as naval
submarines, commercial submersibles, and autonomous un-
derwater vehicles, has traditionally been carried out using a
combination of analytical methods, which include design for-
mulae, empirical curves, and numerical methods. Most design
codes [1, 2] use a conventional working stress approach, un-
dertaking an analysis for each mode of failure. A single safety
factor, or several partial safety factors (PSF), accounting for
uncertainties in the loading, fabrication, and strength of the
pressure hull, is applied to the characteristic strength in order
to determine the allowable working pressure. The analytical
methods rely on many simplifying assumptions and require
such as interactive collapse, local deformations, variable mate-
rial properties and geometry, and the effect of penetrations and
secondary structure on collapse.
Many shortfalls of traditional methods of design can be over-
come using numerical methods, specifically 3D nonlinear fi-
nite element analysis (NLFEA). 3D NLFEA is a powerful
tool, which allows more realistic modeling of geometry, mate-
rial, and structural behaviour. This same flexibility is also a
deficiency of FEA. Analysts have a range of choice, not only
regarding the appropriate level of geometric detail and mate-
rial models, but also with respect to element type, boundary
conditions, meshing parameters, and method of solution. Rea-
sonable variation in the available choices causes scatter in
results.
There is also inherent uncertainty in the finite element method,
as it is based on theory that involves various simplifications
and assumptions. In addition, the quality of results obtained
using finite element calculations is directly related to the ele-
ment mesh used, as well as the choice of solution parameters
such as the load step size in nonlinear calculations. A further
uncertainty with NLFEA arises in the definition of failure, or
collapse in the case of pressure vessels under external loading,
which is not necessarily a straightforward determination.
These uncertainties can be accounted for in a PSF for the
NLFEA of pressure vessels.
The use of nonlinear finite element analysis will not eliminate
scatter in the prediction of pressure hull collapse pressures.
NLFEA actually introduces an extra uncertainty, the human
factor, to the design process. The real advantage of NLFEA is
that it would allow designs to be based on the limit states of a
geometrically accurate representation of a structure, as op-
posed to the current practice of applying first yield criteria to
an idealized structural representation.

1 © 2006 Defence Research and Development Canada

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 The incorporation of 3D NLFEA into the pressure vessel de-
sign process can only be accomplished by comparing NLFEA
results with experimental measurements. This evaluation
should cover, not only a range of failure modes, but also an
array of FE modeling parameters that are considered accept-
able and appropriate. This paper outlines a proposal to study
the accuracy of the finite element method in predicting the
ultimate strength of pressure hulls, as well as providing a
framework to include NLFEA in the design process through
the use of revised safety factors and analysis procedures.
However, before this is discussed, an overview of pressure
hull collapse behaviour and the methods currently used to pre-
dict it will be presented, as well as a summary of recent work
at DRDC on the assessment of pressure hull compartment
damage.
TRADITIONAL PRESSURE HULL DESIGN METHODS
The primary load-bearing structure of a submarine is the pres-
sure hull, which is designed to withstand the compressive
forces associated with hydrostatic pressure. The most efficient
geometries for resisting these compressive forces are circular
cross-sections, and thus, pressure hulls are typically composed
of a combination of ring-stiffened cylinders and cones, with
spherical or torispherical domes at either end. Figure 1 shows
a typical pressure hull structure.
Ring-Stiffened Ring-Stiffened Cylinder
Cone
Internal Ring-Stiffeners
Load Bearing
Bulkhead Torospherical
Endcap
FIGURE 1: TYPICAL PRESSURE HULL STRUCTURE
Traditional design methods are based on the assumption that a
compartment, which may consist of cylindrical, conical and
torispherical sections, can be approximated as a uniform ring-
stiffened cylinder terminated by rigid bulkheads. The ideal-
ized cylinder is based on the most pessimistic dimensions,
resulting in an inherent conservatism associated with the tradi-
tional design methods.
The compressive nature of the design loads means that failure
due to instability (i.e. buckling) is probable. In general, the
external pressure causing elastic buckling or elasto-plastic
collapse governs the ultimate strength of a ring-stiffened cyl-
inder [3].
When considering elastic buckling (i.e. ignoring the effects of
plasticity and nonlinear geometry) of ring-stiffened cylinders,
failure is categorized as either interframe or overall, depending
on the number of circumferential and longitudinal lobes (Fig-
ure 2). Overall buckling is characterized by a global failure of
the combined shell and ring-stiffeners and typically occurs
with 2 to 4 circumferential lobes. Interframe buckling occurs
at the shell between ring-stiffeners and is usually associated
with a relatively large number of circumferential lobes (5 to
20).
Real pressure hulls are designed to make use of the full elastic
range of the material, with collapse occurring elasto-
plastically rather than elastically. The failure point is defined
2 © 2006 Defence Research and Development Canada
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
as the highest point of the load-deflection curve, but in elasto-
plastic collapse this point is not sharply delineated as it is in
elastic buckling. Elastic and elasto-plastic collapse modes
tend to be similar for overall modes, but can differ greatly for
interframe modes (Figure 2). This terminology can nonethe-
less be applied to elastic and elasto-plastic modes alike. For
typical pressure hull structures, the value of external pressure
to cause elasto-plastic collapse is significantly less than the
elastic buckling pressure.
Elastic overall collapse (n=2) Elastic interframe collapse (n=6)
Elasto-plastic interframe collapse
FIGURE 2: COLLAPSE MODES OF RING-STIFFENED
CYLINDERS
Design for Interframe Collapse
Interframe collapse pressure is typically determined using a
curve based on empirical data, normalized with respect to ana-
lytically calculated yield pressures. Test data has been col-
lected [4] for a range of ring-stiffened cylinders, all with out-
of-circularities of less than 0.5% of the radius, as is typically
required for design. The collapse pressure is calculated using
a curve that is fitted to either the mean [1] or lower bound [2]
of the data. The collapse pressure is plotted against the mini-
mum elastic interframe buckling pressure, as predicted by von
Mises. Both collapse and buckling pressures are normalized
by dividing by the pressure at which the mean circumferential
stress of the plating equals yield.
Use of the empirical curve inherently takes account of elasto-
plastic collapse and the build imperfections associated with
real cylinders, such as out-of-circularity (OOC) and residual
stresses.
Design for Overall Collapse
Overall collapse pressure is typically determined using an
elasto-plastic method, which includes the effects of material
plasticity and residual stresses due to cold bending of the
frames. Cold bending residual stresses can cause a significant
reduction in collapse pressure, particularly if frame yielding
precipitates failure.
In overall elasto-plastic collapse, a buckling instability devel-
ops following initial yielding of either a frame or the plating in
way of a frame. Once the initial yielding has occurred, large
deflections of the frame become possible and total collapse
follows soon after.
The procedure for determining the overall elasto-plastic col-
lapse pressure was developed by Kendrick [5]. Kendrick’s
analysis begins with a non-linear differential equation for a
single ring frame and one bay of attached plating, in which a
harmonic OOC shape and residual stresses due to cold bend-
 ing of the frame and plating are included. The equation can be
solved for a given external pressure by finite differences. The
pressure is increased until a solution can no longer be found;
the last converged solution provides the collapse pressure of
the ring frame and its plating. For a complete description of
the elasto-plastic method for determining overall collapse see
Ref. [1, 6].
Design Safety Factors
The collapse pressures discussed in the previous sections are
related to the allowable working pressure of a pressure hull
through safety factors that were developed through a combina-
tion of experiments and past experience with pressure hull
design. Some design codes use a single safety factor to ac-
count for all uncertainties, while other codes use a PSF ap-
proach. Typical PSFs account for:
1. Departures from specified design parameters and discrepan-
cies between the as-built structure and the model.
2. Uncertainties in the strength of the model, including ap-
proximations in the analysis methods and subjective uncertain-
ties from correlating experimental data with analysis results.
This PSF is unique to the type of failure being considered.
3. Uncertainties in the loads, including accidental overloading
and other discrepancies in the load estimation.
For the pressure hull structure, collapse pressures are typically
determined for each of four failure modes: interframe collapse,
overall collapse, dome-end collapse, and frame tripping. The
allowable working pressure is determined by dividing the cal-
culated collapse pressures by the appropriate factor(s), and
minimizing with respect to the various failure modes.
STRUCTURAL MODELING AND STRENGTH
ASSESSMENT OF PRESSURE HULLS AT DRDC
In recent years, DRDC has been investigating the effect of
different kinds of damage on the structural capability of pres-
sure hull compartments. Two particular problems have been
the subjects of extensive study: (1) the effect of pressure hull
indentation damage and (2) the effect of pressure hull external
corrosion. The latter problem has led to the initiation of an
experimental program to determine the effects of corrosion
damage on pressure hull strength, which will be discussed in a
later section.
DRDC has jointly funded, with the UK Ministry of Defence,
the development of SubSAS (Submarine Structural Analysis
Suite) [7], a specialized software tool for both design-level
and finite element analysis of pressure hull structures. This
tool has already proven to be effective in reducing the time
required for pressure hull collapse analysis, and has been used
extensively in the pressure hull damage assessments described
below.
Indentation Damage Assessment
This analysis focused on predicting the effect of an inward
indentation of the shell plating on the structural capability of a
pressure hull. Three-dimensional FE models of a damaged
ring-stiffened cylinder were developed using shell elements.
3 © 2006 Defence Research and Development Canada
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
The indentation was incorporated into the models by deform-
ing the mesh to conform to the damage. The indentation was
modeled between ring-stiffeners and had a maximum depth
less than the shell thickness. The ends were constrained to
represent the rigidity of the end structures. OOC was intro-
duced at the design level, ±0.5 percent of radius, with various
shapes.
Figure 3 shows pressure-deflection plots for damaged and
undamaged models calculated with the ANSYS nonlinear
FEA program. Analyses were conducted using standard load
control, as well as the arc length nonlinear scheme, with both
methods yielding similar results. The indicated deflection is
the maximum radial deflection in the model, which for the
damaged model occurs at the centre of the dent. Both the
damaged and undamaged pressure hulls failed in an interframe
collapse mode. At a given pressure, deflections are signifi-
cantly increased by the existence of the dent. Note that these
deflections are in addition to the pre-existing permanent de-
formation in the damaged plating.
∆Pc
∆Py
Pressure
Undamaged
Dented
Max radial deflection (mm)
FIGURE 3: EFFECT OF DENT ON COLLAPSE PRESSURE
AND YIELD PRESSURE
The pressure-deflection plots also show that the change in the
yield pressure (∆Py) is much larger than the change in the col-
lapse pressure (∆Pc). The yield pressure results were con-
firmed with linear stress predictions. Thus, the primary con-
cern would not be that the damage could precipitate a prema-
ture collapse of the pressure hull, but rather that localized
yielding would cause the dent to grow in size, and that under
repeated loadings this growth would not reach a stable limit.
Corrosion Damage Assessment
This analysis was conducted to determine the effects of re-
duced shell thickness due to corrosion on the strength of a
pressure hull compartment. In our assessments corrosion was
modeled as reduced thickness of hull plating. Two different
types of FE models were used: 1) a 3D shell model of the
pressure hull compartment, and 2) a 3D solid model of a sin-
gle representative ring-stiffener and associated bay of shell
plating.
Compartment Model – The pressure hull compartment
model was comprised of outer shell, ring stiffeners, and a
torispherical dome. Constraints were used in place of a water-
tight bulkhead. Various models were constructed, using dif-
ferent types of elements (4-node, 8-nodes shells), material
 properties (stress-relieved, cold bent), and shapes and ampli- interframe collapse can occur. And since the shell plating also
tudes of OOC. serves as an outer flange for the ring stiffeners, it also lowers
the bending stiffness of the rings, thus promoting overall col-
Corrosion damage was applied as a single rectangular area of
lapse of the compartment.
reduced thickness. Three cases of shell reduction were stud-
ied: 1) variable shell thickness in the corroded region (average Both analytical and FEA results indicated that interframe col-
thickness reduction of 4.1%), 2) a uniform thickness reduction lapse pressures were more greatly affected than overall col-
of 8.7%, and 3) a uniform thickness reduction of 14%. lapse pressures. However, collapse pressures for both modes
were reduced. Reductions in collapse pressure due to corro-
The assessments were performed in the following stages:
sion ranged between 2% and 19%, depending on the combina-
1. Traditional analytical design methods were used to calculate tion of shell thickness, OOC, and material model.
the effect of uniform shell reductions.
Nonlinear analysis of models using 8-node shell elements re-
2. A baseline for the FE calculations was established using sulted in collapse pressures that were consistently lower (be-
models of the undamaged compartment. tween 0.4% and 1.4%) than their 4-node counterparts with
twice the element density. Models with corrosion showed
3. The strength of the corroded compartment was evaluated greater discrepancy between companion 8-node and 4-node
using a series of FE models with the three different distribu-
models than those without corrosion, however this difference
tions of the hull thinning described above. was never greater than 1.4% for any combination of material
A wide variety of OOC shapes were applied to the models to and OOC.
ensure that all possible collapse modes could be predicted.
Ring Model – A single ring-stiffener and bay of plating was
This study was concerned with effect of corrosion on overall
created using solid elements in order to study the effect of
collapse pressure, and thus OOC was applied individually for
small, isolated corrosion scars. One-quarter of the ring cir-
modes 2 through 6, as well as a combination of these modes
cumference was modeled with symmetry boundary conditions,
with random phase angles. Measured values of OOC were
effectively representing an infinitely long ring-stiffened cylin-
also used.
der with a repeating corrosion pattern. Single corrosion scars
Two material models were used: 1) an idealized stress- of various sizes and locations were applied to the models
relieved condition with perfectly elastic – perfectly plastic (Figure 5) and collapse pressures were compared to the non-
stress-strain relationship, and 2) a multi-linear material model corroded model. OOC was applied in two forms in order to
that included the strength-reducing tendency of residual encourage either interframe or overall collapse.
stresses due to cold bending.
More than one hundred different FE models were analyzed,
each representing a unique combination of thickness, OOC
shape, and material behaviour. Collapse pressure were deter-
mined taking into account both geometric and material nonlin-
earities. The arc length solution method in ANSYS was used
to calculate the collapse pressures of the various models. This
technique allowed the model to be loaded up to and beyond
the point of collapse, allowing the failure mode to be identi-
fied more easily. Both overall and interframe collapse modes
were predicted as shown in Figure 4.

FIGURE 5: RING MODEL WITH CORROSION SCAR (HIGH-

Interframe Collapse Overall Collapse
FIGURE 4: EXAMPLE COLLAPSE MODES FOR A PRES-
SURE HULL COMPARTMENT
It was found that corrosion has two separate effects on the
collapse behaviour of a compartment. It weakens the shell
plating between frames, thus lowering the pressure at which
LIGHTED IN INSET)
Corrosion scars were idealized by subtracting a volume of
solid from the shell plating, such that a circle was inscribed on
the curved surface of the shell and the scar depth varied curvi-
linearly up to a maximum value at the centre. The maximum
depth of the scar was held constant for all analyses, at 13% of
the shell thickness. Two corrosion parameters were varied for

4 © 2006 Defence Research and Development Canada

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 these analyses; namely, the diameter and location of the scar. Experimental Program at DRDC
Scars were located either at mid bay or directly over the stiff-
Previous Testing Program – DRDC and the Netherlands
ener.
Ministry of Defence have collaborated on an experimental
Nonlinear FE analyses were carried in the following manner: study of the effects of OOC and secondary structure on the
collapse behaviour of pressure hulls [8]. This study, con-
1. A baseline was established by determining the collapse
ducted in the early 1990s, was also used to validate the accu-
pressure of the ring with no corrosion.
racy of NLFEA in predicting pressure hull collapse. The pro-
2. The worst-case scenario of a uniform shell reduction of ject consisted of the fabrication, testing, and analysis of 5
13% over the entire ring was analyzed. aluminium cylinders designed to fail in overall collapse. The
specimens were constructed of machined aluminium tubing
3. The collapse pressure and pressure to cause first yield of the
and had external bar stiffeners (Figure 6).
material was determined for the various isolated scars.
40
As with the compartment models, the collapse pressures for
the ring models were determined using the arc length nonlin-
ear solution procedure in ANSYS. Both geometric and mate- 110 5.5
rial nonlinearities were considered, and the collapse pressure
8
was taken as the peak of the load-displacement plot.
2.7
The corrosion scars, which varied in diameter from approxi-
mately 6.5 to 79% of the stiffener spacing, were found to have
a minor effect on the collapse pressure. The greatest reduc-
tions occurred for interframe failures with scars at mid-bay. 520

This resulted in a locally reduced bending stiffness of the shell (All Dimensions in Millimeters)

between stiffeners, leading to reductions in collapse pressure

of at most 1% for the largest scar. Scars located over the stiff-
ener resulted in a maximum reduction to collapse pressure of
0.6%.
Models designed to fail in an overall mode were only slightly
affected by even the largest corrosion scars. Scars at both
mid-bay and the stiffener were found to locally reduce the
overall bending stiffness of the ring, resulting in a maximum
reduction to collapse pressure of 0.2%. For all cases, the re-
duction to collapse pressure did not approach that caused by a FIGURE 6: PLAIN BAR-STIFFENED CYLINDER
uniform thinning of the shell, which was 23% and 7.5% for
interframe and overall failures, respectively. Three ‘plain’ models (Figure 6), and two models with simu-
lated decks or tank tops (Figure 7), were tested. OOC in an
Corrosion scars were found to significantly decrease the value overall mode-3 shape was mechanically applied to all of the
of pressure causing first yield of the material. This was par- specimens, excluding one of the cylinders with a simulated
ticularly severe for the case of interframe failures with the scar deck. The cylinder specimens were tested to failure under
at mid-bay, which showed an average reduction of 22% to the hydrostatic pressure at the high-pressure testing facility at
yield pressures. Yielding occurred at the centre of the scar, DRDC.
and the average reduction to yield pressure was greater than
that for the uniformly thinned model (20%). It was found that
collapse of the corroded models did not immediately follow
first yielding of the shell, but that yielding occurred well be-
fore collapse, at 70% of the ultimate strength of the structure.
The effect was lesser for the case of corrosion over the stiffen-
ers with a 6% average reduction in yield pressures.
As with collapse pressure, the reduction in yield pressure was
found to be less acute in the overall failure models, showing
average reductions of 7.5% and 0% for scars at mid-bay and at
the stiffener, respectively.
The conclusions of this study were that small, isolated corro-
sion scars result in a slight reduction in collapse pressure;
however, pressures causing material yield may occur well in
advance of collapse. As with the case of local indentation
damage, the possibility exists that a pressure vessel could sus-
tain permanent local deformation while operating below the FIGURE 7: BAR-STIFFENED CYLINDER SPECIMEN WITH
collapse load, which could grow with repeated loadings. SIMULATED DECK

5 © 2006 Defence Research and Development Canada

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 The results of these experiments showed that the specimens
failed in the predicted overall collapse shape (Figure 8). It
was found that OOC has a significant effect on strength, as
shown by a 15% decrease in the collapse pressure of the
model with tank top and OOC, when compared to a similar
model with no OOC. However, the tank top itself did not
have a significant effect on the overall collapse pressure.
FIGURE 8: PLAIN BAR-STIFFENED CYLINDER AFTER
TESTING
Experimental results and the most recent NLFEA calculations
are summarized in Table 1. The original nonlinear FE calcula-
tions [8] showed that the method consistently overestimated
collapse pressure by less than 7%. However, these analyses
were carried out on models that did not include the effects of
residual stresses.
TABLE 1: EXPERIMENTAL AND NLFEA RESULTS (REF.
[9]) FOR BAR-STIFFENED CYLINDERS
Recent calculations [9] show that the process of mechanically
applying OOC to the cylinders results in residual stresses,
which, when simulated using NLFEA, were found to increase
the collapse pressure by almost 10%, compared to models
with the same level of OOC but no residual stresses. When
this was combined with an improved mesh and material
model, NLFEA was found to overestimate the collapse pres-
sure by less than 10%.
Current Testing Program – DRDC and the Netherlands
Ministry of Defence have recently begun a second collabora-
tive experimental research program to investigate the effects
of damage due to corrosion on the structural integrity of pres-
sure hulls, with the goal of developing submarine structural
safety and maintenance guidelines.
This project involves the fabrication, destructive testing, and
numerical analysis of up to 30 small-scale aluminium ring-
stiffened cylinders. It will examine the effects of material loss
due to corrosion on the collapse pressure of pressure hulls.
The effect of geometric imperfections and residual stresses
will also be studied. These types of ‘damage’ will be studied
in a parametric fashion to determine their effects in isolation
and when combined. The test models will comprise various
levels and patterns of simulated corrosion scarring, types of
collapse (interframe and overall) and pressure hull configura-
tions (internally and externally stiffened). The results of these
tests will help assess the accuracy of the damage modeling
method previously described.
The first phase of the testing program, which has been com-
pleted, examined cylinders stiffened with external T-bars.
These cylinders are of similar dimensions as the previous bar-
stiffened specimens, and can be lumped into two groups: short
cylinders, which were designed to failure locally (interframe
collapse), and long cylinders, which were designed to fail
globally (overall collapse). Three short cylinders, two having
simulated corrosion on the shell (Figure 9), and two long cyl-
inders, one with ‘dog-bone’ corrosion on the central flanges
(Figure 10), were manufactured for the first phase of testing.

Specimen Collapse Pressure (MPa) Error (%)

Experiment NLFEA
1
Plain Cylinders
CYL1 7.14 7.43 4.1
CYL2 6.23 6.80 9.1
CYL3 6.49 6.78 4.5
2
Cylinders with Tank Top
DECK1 7.88 8.57 8.8
DECK2 6.68 6.81 1.9
Notes on nonlinear finite element calculations:
1. MARC load-control nonlinear analysis; 4-node shell elements; tying equations
to account for effects of double area at base of web; pressure loading at hull,
except for width of web, and at ends of web; includes the effect of residual
stresses due to process of inducing OOC (3 circumferential lobes); multi-linear
material model based on measured stress-strain curve (isotropic hardening);
result is last converged solution (0.01 MPa load step)
2. Same as Note 1, except no residual stresses were included as specimens
were heat treated after welding of deck; Specimen DECK1 did not have me- FIGURE 9: SHORT T-STIFFENED CYLINDER WI TH SHELL
chanically induced OOC
CORROSION (INSET)

6 © 2006 Defence Research and Development Canada

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 Nonlinear FE calculations [9] predicted the collapse pressures
of these specimens within ±6%. The results of these analyses,
as well as the experimental collapse pressures, can be found in
Table 2.

TABLE 2: EXPERIMENTAL AND NLFEA RESULTS (REF.

[9]) FOR T-STIFFENED CYLINDERS

Specimen Collapse Pressure (MPa) Error (%)

Experiment NLFEA
1
Undamaged Cylinders
SHORT2 7.87 7.40 -5.9
LONG1 9.05 9.00 -0.5
2
Cylinders with Corrosion
SHORT3 6.77 6.60 -2.6
SHORT4 6.94 6.60 -4.9
LONG2 8.59 8.90 3.6
Notes on nonlinear finite element calculations:
1. MARC load-control nonlinear analysis; 4-node shell elements; no correction for
double area at web base and web-flange intersection; pressure applied at the
hull; no residual stresses as specimens were not deformed after fabrication;
bilinear (perfectly elastic-perfectly plastic) material model with isotropic harden-
ing; value of yield used was 250 MPa versus measured value of approximately
300 MPa; result is last converged solution (0.1 MPa load step); measured values
of OOC were implemented in the mesh
2. Same as Note 1, except no OOC imperfection was used for analysis; approxi-
FIGURE 10: LONG T-STIFFENED CYLINDER WI TH ‘DOG- mate simulation of corrosion using existing mesh
BONE’ FLANGE CORROSION (INSET)
It is important to note that these are preliminary analyses, and
Corrosion damage of the short cylinders was in the form of a the FE modelling was not as rigorous as that for the results
square patch (34x34mm) of shell in the central bay with a re- shown in Table 1. For instance, at the time of calculation,
duced thickness of 25%. This type of damage resulted in an measured imperfections were only available for undamaged
average reduction in collapse pressure of 13% from the un- specimens, and measurements of the material properties were
damaged state. Corrosion damage to the central flanges of not available at all. Tensile test data have since shown the
the long cylinder resulted in a reduction in collapse pressure of yield stress of the aluminium to be significantly greater than
5% from the undamaged state. the nominal value (300 MPa vs. 250 MPa), which partially
accounts for the under-prediction of collapse pressure in most
Both visual inspection after testing, and examination of the cases. See the notes below Table 2 for further analysis de-
strain gauge data, confirmed that the cylinders collapsed in the tails.
predicted modes (i.e. interframe or overall). However, obser-
vation of the tested cylinders has revealed that collapse oc- NLFEA IN PRESSURE HULL DESIGN AND
curred catastrophically for all specimens (see Figure 11), such ANALYSIS
that the exact nature of collapse (e.g. number of circumferen-
tial lobes) could not be determined with any degree of cer- The use of NLFEA has not been implemented in pressure ves-
tainty. As a result, a study to improve the testing apparatus sel design codes due to the absence of correlation between
and procedure will be conducted before the next phase of test- NLFEA and experimental results. The goal of this project is
ing. to address this lack of data by providing NLFEA results for
comparison with experimental values. Standards will be pro-
posed for the modeling and analysis of pressure hulls using
NLFEA, which, in combination with a PSF derived from the
experimental-NLFEA correlation, will allow the incorporation
of nonlinear finite element calculations into the design proc-
ess. The various aspects of this research project are described
in the following sections.

Experimental Database
The first step to establishing a NLFEA PSF is to assemble a
database of experiments conducted on the collapse of ring-
FIGURE 11: SHORT T-STIFFENED CYLINDER WITH COR- stiffened cylinders under hydrostatic pressure. Kendrick [4]
ROSION AFTER TESTING has previously constructed a database of this sort for strictly

7 © 2006 Defence Research and Development Canada

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 interframe failures, which has been the basis for the interframe more complex methods, such as relating Pc,FE and Pc,Exp using
collapse prediction in most external pressure vessel design a combination of geometric factors or analytically calculated
codes [1, 2]. Additionally, Ellinas and Supple [10] have col- values.
lated experimental results for a variety of failure modes with
The development of the PSF can be demonstrated using the
the purpose of comparison with predictions from various de-
limited data available from the previous and current Canada-
sign codes. Both of these collections were compiled over two
Netherlands research programs described above. This data
decades ago, and thus, additional research has been conducted
does not represent the full capabilities of the finite element
since [11, 12]. An effort will be made to obtain the previously
method for predicting collapse pressures, as some of the
collected experimental information, which must be combined
analyses were undertaken with incorrect or incomplete physi-
with an up-to-date search for collapse studies conducted in the
cal data. However, the data does represent the scatter that is
last twenty years. The test results from the Canada-
intrinsically part of finite element modeling due to the various
Netherlands experimental programs described in this paper
choices available regarding modeling and solution methods, as
will also be included in the database.
well as the variable amount of information available to the
analyst.
NLFEA Modeling Guidelines
Figure 12 plots the finite element collapse pressures of the 10
Modeling guidelines must be established in order for the PSF
aluminium cylinders against the corresponding experimental
determined in this study to be useful. Applying a PSF to fu-
results (see Table 1 and Table 2). An alternative method of
ture finite element results can only occur if the modeling is
representing the same data is shown in Figure 13, which plots
performed in a manner consistent with that adhered to in the
the FE collapse pressure, normalized with respect to the ex-
development of the factor.
perimental value, against the experimental results. Both of
The finite element analyst must make various choices when these figures show that the FE calculated collapse pressures
creating and solving finite element models of pressure hulls. were within ±10% of the experimental values, and exhibit
These include assumptions made when creating the geometric good scatter about the line of zero error (diagonal line in Fig-
and/or finite element model, as well as choice of material ure 12 and the horizontal line, Pc,FE/Pc,Exp = 1, in Figure 13).
models and method of solution. The effects of these choices
on the outcome of the analysis must be studied in order that
modeling guidelines be established. The various techniques of
finite element modeling will be studied and compared with
experimental results to establish a code of best practice.

Determination of Partial Safety Factor

There are several sources of scatter in the results of 3D
NLFEA that must be accounted for by a PSF. The first is the
departure of the finite element method from the analytical
solution. There is also the variability in the departure of the
FE model from the true geometry, material behaviour, and
boundary conditions of the actual specimen being analyzed.
In other words, it is impossible to create an exact FE represen-
tation of the real model, regardless of mesh size or level of
detail. Scatter in FE predictions is partially the result of vary-
ing levels of detail and precision in the measurement of the
physical model. Another source of scatter is the aforemen-
tioned human factor. The analyst has a wide range of accept-
able and reasonable choices when creating and solving the FE
model. These choices can be limited by the use of modeling
guidelines, which will reduce but not eliminate the associated FIGURE 12: EXPERIMENTAL AND NLFEA CALCULATIONS
scatter in results. OF COLLAPSE PRESSURE
Once a database of suitable experimental results has been
compiled and FE modeling guidelines have been established, A margin of error of ±10% in the prediction of collapse pres-
the task of constructing and analyzing the FE models will be- sures by NLFEA is equivalent to a PSF for the method equal
gin. The results of the nonlinear finite element analyses will to 1.1. Of course, additional PSFs accounting for discrepan-
be used to determine a single PSF, applicable to both inter- cies in the model from the as-built structure and over-loading
frame and overall failure, to account for the uncertainties in would also have to be applied in order to determine the allow-
the modeling and analysis methods. Determination of the PSF able working pressure.
may be as simple as taking the upper bound of a plot of The development of the safety factor in the preceding para-
Pc,FE/Pc,Exp vs. Pc,Exp, where Pc,FE is the collapse pressure pre- graphs has been presented as an example of the procedure, and
dicted using NLFEA and Pc,Exp is the experimentally deter- does not imply that the best results that can be achieved by
mined collapse pressure. Calculation of the PSF may involve NLFEA are within ±10% of actual collapse pressure. As men-

8 © 2006 Defence Research and Development Canada

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 tioned previously, a study will be conducted to determine best
practice and modeling guidelines, which may serve to reduce
this level of error. Additionally, a more advanced method of
interpreting the NLFEA data (e.g. normalizing data with re-
spect to analytically calculated values) may further reduce the
margin of error. The demonstrated approach doesn't take into
account experimental error, an additional uncertainty, which
would also have to be included in the determination of the
PSF.
FIGURE 13: EXPERIMENTAL AND NORMALIZED NLFEA
CALCULATIONS OF COLLAPSE PRESSURE
Revised Design Procedure
The primary goal of this research is to provide a pathway that
would allow the collapse pressure calculated using 3D
NLFEA to replace the analytically calculated values for over-
all and interframe collapse in the design of pressure hulls.
This can be achieved by the determination of a PSF account-
ing for uncertainties in nonlinear finite element modeling and
analysis methods, specifically for the prediction of the col-
lapse of ring-stiffened cylinders. The PSF will be established
by modeling with as much realism as possible. The use of the
measured shape and material properties, when available, will
result in a safety factor that closely reflects the capability of
NLFEA to accurately predict collapse pressures. However, in
the design process, the engineer does not have the benefit of
measured geometry and material behaviour; thus, the design
procedure must be revised to prescribe important considera-
tions, such as OOC, when using NLFEA.
SUMMARY
DRDC sees a requirement for a greater understanding of the
nonlinear finite element method in predicting pressure hull
collapse. Ultimately, DRDC has the goal of including this
method in the design procedure by the use of an appropriate
partial safety factor. Establishing such a safety factor requires
a multi-year research effort, as evidenced by the scope of work
outlined in this paper.
REFERENCES
[1] Defence Procurement Agency, 2001, “SSP 74 Design of
Submarine Structures”, Sea Systems Publication No. 74, De-
9 © 2006 Defence Research and Development Canada
Downloaded From: http://proceedings.asmedigitalcollection.asme.org/ on 01/30/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
fence Procurement Agency, Sea Technology Group, United
Kingdom.
[2] British Standards Institution, 1980, “BS 5500 British Stan-
dard Specification for Unfired Fusion Welded Pressure Ves-
sels, Issue 5”, British Standards Institution, United Kingdom.
[3] Kendrick, S., 1985, “Ring-Stiffened Cylinders Under Ex-
ternal Pressure”, in R. Narayanan, (Ed.), Shell Structures, Sta-
bility and Strength, Elsevier Publishers, New York, pp. 57-95.
[4] Kendrick, S., 1970, “Externally Pressurized Vessels”, in
S.S. Gill, (Ed.), The Stress Analysis of Pressure Vessels and
Pressure Vessel Components, Pergamon Press, Toronto, pp.
405-511.
[5] Kendrick, S., 1979, “The Influence of Shape Imperfections
and Residual Stress on the Collapse of Stiffened Cylinders”,
Proceedings of the Conference on Significance of Deviations
from Design Shape, Mech. Eng. Publ. Ltd. for Inst. of Mech.
Eng., London, pp. 25-35.
[6] Smith, M.J. and MacKay, J.R., 2005, “Overall Elasto-
Plastic Collapse of Ring Stiffened Cylinders with Corrosion
Damage”, International Journal of Maritime Engineering, 147
Part A1, pp. 53-61.
[7] MacKay, J.R., 2004, “Submarine Structural Analysis
Suite: SubSAS (Version 2.3.1) – Tutorial Manual”, DRDC
Atlantic TM 2004-228, Defence R&D Canada – Atlantic,
Dartmouth, Nova Scotia.
[8] Bosman, T.N., Pegg, N.G., and Keuning, P.J., 1993, “Ex-
perimental and Numerical Determination of the Nonlinear
Overall Collapse of Imperfect Pressure Hull Compartments”,
Proc. Warship ’93, International Symposium on Naval Sub-
marines 4, RINA.
[9] The Netherlands Organisation for Applied Scientific Re-
search TNO, 2005, unpublished results.
[10] Ellinas, C.P. and Supple, W.J., 1984, “Buckling Design
of Ring-Stiffened Cylinders”, Journal of Waterway, Port,
Coastal and Ocean Engineering, 110 (4), pp. 413-431.
[11] Seleim, S.S. and Roorda, J., 1985, “Buckling Behaviour
of Ring Stiffened Cylinders: Experimental Study”, SM Paper
No. 195, Solid Mechanics Division, University of Waterloo,
Waterloo, Ontario.
[12] Ross, C.T.F., Haynes, P., Seers, A., and Johns, T., 1995,
“Inelastic Buckling of Ring-Stiffened Circular Cylinders un-
der Uniform External Pressure”, American Society of Me-
chanical Engineers, Petroleum Division (Publication) PD, 70,
pp. 207-215.