You are on page 1of 118

Chapter 12 - Nonlinear Analysis

CHAPTER 12
Important issues regarding nonlinear analysis of structures are described.Three types
of nonlinearities are introduced with an emphasis on geometrical and material
nonlinearities. Nonlinear formulations for one dimensional bars and beams are
described, as well as generalization to multi-dimensional problems. The most recent
solution techniques are presented. Applications are placed on beams, frames, plates
and shells.

Nonlinear analysis

page
12.1 Introduction.............................................................................................................................................................. 12.2
12.1.1 General........................................................................................................................................................ 12.2
12.1.2 A nonlinear geometrical problem............................................................................................................... 12.5
12.1.3 Nonlinear material behaviour..................................................................................................................... 12.14
12.2 Stiffness relationship for beam with axial force................................................................................................... 12.17
12.2.1 General........................................................................................................................................................ 12.17
12.2.2 Comparison of alternative stiffness matrices for lateral deformations of a bem with axial force.............. 12.18
12.3 Formulations for nonlinear geometrical behaviour of bars and beams with axial and lateral deformation 12.20
12.3.1 General........................................................................................................................................................ 12.20
12.3.2 Methods with updated coordinates............................................................................................................. 12.23
12.3.3 Total Lagrangian formulation for a beam with axial and lateral deformation............................................ 12.28
12.3.4 Generalization............................................................................................................................................. 12.35
12.4 Nonlinear material behaviour................................................................................................................................ 12.36
12.4.1 One dimensional case................................................................................................................................. 12.36
12.4.2 Generalization............................................................................................................................................. 12.42
12.4.3 Cyclic plasticity, shakedown and ratchetting............................................................................................... 12.43
12.5 Solution techniques.................................................................................................................................................. 12.45
12.5.1 General........................................................................................................................................................ 12.45
12.5.2 Load increnmental methods........................................................................................................................ 12.48
12.5.3 Iterative methods........................................................................................................................................ 12.57
12.5.4 Combined methods..................................................................................................................................... 12.56
12.5.5 Advanced solution procedures................................................................................................................... 12.58
12.5.6 Direct integration methods......................................................................................................................... 12.63
12.6 Applications............................................................................................................................................................. 12.67
12.6.1 General........................................................................................................................................................ 12.67
12.6.2 Beams and frames....................................................................................................................................... 12.67
12.6.3 Plane stress, plates and shells..................................................................................................................... 12.74
12.7 Analysis of accidental load effects...................................................................................................................... 12.84
12.6.1 General........................................................................................................................................................ 12.84
12.6.2 Fires and explosions................................................................................................................................... 12.86
12.6.3 Ship impacts................................................................................................................................................ 12.88
Appendix A Solution of the differential equation of a beam with axial load........................................................... 12.94
Appendix B General formulation for geometrically nonlinear behaviour.............................................................. 12.99
Appendix C Plasticity theory....................................................................................................................................... 12.106
Reffrences ................................................................................................................................................................... 12.116

12.1
Chapter 12 - Nonlinear Analysis

12 Nonlinear Analysis

12.1 Introduction

12.1.1 General

Linear versus nonlinear analysis

Structural analysis – including the finite element method – is based on the following
principles:

¾ Equilibrium(expressed by stresses)
¾ Kinematic compatibility (expressed by strains)
¾ Stress-strain relationship

So far, the analysis has been based on the assumptions that

¾ Displacements are small


¾ The material is linear and elastic

When the displacements are small, the equilibrium equations can be established with
reference to the initial configuration. Moreover, this implies that the strains are linear
functions of displacement gradients (derivatives).

The linear elastic stress-strain relationship corresponds to Hooke’s law.

The relationship between load and displacement for structures with nonlinear
behaviour may be as shown in Fig. 12.1.

When the ultimate strength of structures that buckle and collapse is to be calculated,
the assumptions about small displacements and linear material need to be modified. If
the change of geometry is accounted for, when establishing the equilibrium equations
and calculating the strains from displacements, a geometrical nonlinear behaviour is
accounted for. Various examples are given in Fig. 12.1. In section 12.1.2, a
quantitative example is completely washed out.

Analogously, material nonlinear behaviour is associated with nonlinear stress-strain


relationship. An example is given in Section 12.1.3.

Finally, nonlinearity may be associated with the boundary condition, i.e. when a large
displacement leads to contact. Boundary non-linearity occurs in most contact
problems, in which two surfaces come into or out of contact. The displacements and
stresses of the contacting bodies are usually not linearly dependent on the applied
loads. This type of non-linearity may occur even if the material behavior is assumed
linear and the displacement are infinitesimal, due to the fact that that the size of the
contact area is usually not linearly dependent on the applied loads, i.e. doubling the
applied loads does not necessarily produce double the displacement. If the effect of
friction is included in the analysis, then slick-slip behaviour may occur in the contact
area which adds a further non-linear complexity that is normally dependent on the
loading history.
12.2
Chapter 12 - Nonlinear Analysis

Fig. 12.1c shows a typical contact problem of a cylindrical roller on a flat plane.
Initially the contact is at a single point, and then spreads as the load is increased. The
increase in the contact area and the change in the contact pressure are not linearly
proportional to the applied load. Another example is shown in Fig. 12.1c where the tip
of the cantilever comes into contact with a rigid surface.

a) Response of a thin plate/shell (e.g. due to water pressure or explosion pressure)

b) Response of a column subjected to lateral and axial compressive load

c) Representation of contact
Figure 12.1 Typical nonlinear geometrically behaviour.
12.3
Chapter 12 - Nonlinear Analysis

Table 12.1 Comparison of linear and non-linear analysis [Becker, 2001]

12.4
Chapter 12 - Nonlinear Analysis

Reasons for nonlinear stress analysis

There are several areas where nonlinear stress analysis may be necessary (Moan et al.,
2002):

¾ Direct use in design for ultimate and accidental collapse limit states.
Modern structural design codes refer to truly ultimate failure modes and
not only first yield and analogous modes.
¾ Use in the assessment of existing structures whose integrity may be in
doubt due to (a) visible damage (crack, etc.) concern over corrosion or
general ageing. The above will largely relate to the ultimate limit state
because, in many cases, the serviceability limit state will already have been
exceeded and yet key question still remain such as: Is the structure safe?
Should it be repaired and if so, how will any proposed strengthening work?
Can it be kept in service for a little time longer?
¾ Use to help to establish the causes of a structural failure.
¾ Use in code development and research: (a) to help to establish simple
‘code-based’ methods of analysis and design, (b) to help understand basic
structural behaviour and (c) to test the validity of proposed ‘material
models’.

With the new generation of inexpensive yet powerful computers, solution cost is no
longer the major obstacle it has been. However, the complexity of nonlinear stress
analysis still remains to provide the ‘expert’ as well as the unwary novice with many
headaches.

Nonlinear analyses are applied in all the ways mentioned above. However, a
significant increase in the use of nonlinear stress analyses in the assessment of existing
structures is envisaged and eventually in the direct design of more routine
structures.This will occur as hardware becomes cheaper and faster and software
becomes more robust and user-friendly.

It will simply become easier for an engineer to apply direct analysis rather than code-
based charts. However, problems will arise because the latter often include ‘fiddle
factors’ relating to experience, uncertainty, etc. The advent of more computer-based
analysis procedures will lead to the need for a ‘surrounding’, probably computer-
based, ‘code’ to incorporate the ‘partial factors’ including those factors (often now
hidden) relating to the degree of uncertainty of the analysis. The analysis would have
to be directly embedded in a statistical reliability framework.

12.1.2 A nonlinear geometrical problem

Geometrical nonlinearity may be illustrated by the bar system shown in Fig. 12.2a
(Bergan and Syvertsen, 1978). See also Crisfield (1991).

12.5
Chapter 12 - Nonlinear Analysis

a) Geometry

b) Deformation and equilibrium for small displacements ( r)

c) Deformation and equilibrium for large displacements ( r)

Figure 12.2 Two-bar systems

Linear model

When r is small compared to h, the axial strain in the bar is

r sin α o r
ε = = sin α o cos α o
A / cos α o A
(ε is positive when the bar shortens)

and the axial force becomes:

EA
S = EAε = sin α o cos α o ⋅ r
A

S is positive in compression. Equilibrium as referred to the initial geometry, gives

2 EA
R = 2 S sin α o = sin 2 α o cos α o ⋅ r (12.1)
A
or
R = Kr
where

2 EA 2
K= sin α o cos α o
A

The stiffness K is constant, implying a linear relationship between force and


displacement.
12.6
Chapter 12 - Nonlinear Analysis

If the angle αo is small (αo << 1)

sinαo ∼ αo, cosαo ∼ 1


i.e.
2 EAα o2
R= r (12.2)
A

Nonlinear model (large deformations)

The true axial shortening for finite (not small) value of r is

A A
Δ= −
cos α o cos α
and the strain is (positive in compression)

Δ cos α o
ε= =1 −
A / cos α o cos α

Equilibrium for the deformed truss is:

⎛ cos α o ⎞
R = 2 S sin α = 2 EAε sin α = 2 EA sin α ⎜1 − ⎟ (12.3)
⎝ cos α ⎠

By introducing
h−r A A
sin α = , cos α = , cos α o =
A 2 + (h − r ) 2 A 2 + (h − r ) 2 A2 + h2

Eq. (12.3) may be written as

2 EA ⎛ h ⎞ ⎛⎜ A A ⎞
⎟r
R= ⎜ − 1⎟ − (12.4)
A ⎝ r ⎠ ⎜ A 2 + (h − r ) 2 A + h2
2 ⎟
⎝ ⎠
or
R = K (r ) ⋅ r

The stiffness now depends upon the displacement r and the force-displacement
relationship (12.4) is nonlinear.

For small angles α and αo:


h−r
sin α ≈ α ≅ tgα =
A
2
1 1⎛h−r⎞
cos α ≈ 1 − α 2 ≈ 1 − ⎜ ⎟
2 2⎝ A ⎠
2
1 1⎛h⎞
cos α 0 ≈1 − α 02 ≈ 1 − ⎜ ⎟
2 2⎝A⎠
12.7
Chapter 12 - Nonlinear Analysis

Introducing these expressions into Eq. (12.4) yields

⎛ h r ⎞⎛ h 1 r ⎞
⎜ − ⎟⎜ − ⎟
2 EA ⎝ A A ⎠ ⎝ A 2 A ⎠ 2 EA ⎛ h r ⎞ ⎛ h 1 r ⎞
R= 2
r≈ ⎜ − ⎟⎜ − ⎟r
A 1⎛h⎞ A ⎝ A A ⎠⎝ A 2 A ⎠
1− ⎜ ⎟
2⎝A⎠

2
⎛h⎞
when assuming that ⎜ ⎟ << 1.
⎝A⎠

h
By introducing ≈ α o , the following equation results:
A

2 EA 2 ⎛ r ⎞ ⎛ r ⎞
R= α o ⎜1 − ⎟ ⎜1 − ⎟ r (12.5)
A ⎝ h ⎠ ⎝ 2h ⎠

or
R = K (r )r (12.6)
where
2 EA 2 ⎛ r ⎞ ⎛ r ⎞
K (r ) = α o ⎜1 − ⎟ ⎜1 − ⎟
A ⎝ h ⎠ ⎝ 2h ⎠
(12.7)
2 EA 2 EA 2 ⎛ r ⎞r
= αo + α o ⎜ − 3⎟ = K o + K g
A A ⎝h ⎠h

The first term of Eq. (12.7) is the linear stiffness term; see Eq. (12.2), while the second
term is a correction due to nonlinear geometrical effects. The stiffness relationship is
a third degree polynomial, plotted in Fig. 12.3a.

Eq. (12.7) does not always give unique solutions for a given load. Fig. 12.3b shows
that three equilibrium points (A, B, C) may correspond to a given load level.
This also means that it would not be possible to follow the load-displacement curve by
increasing the load. Actually when the point D is reached, the solution will jump to E,
which is a stable equilibrium condition. This phenomenon is called snap-through.

a) Load-deflection b) Possible equilibrium conditions

Figure 12.3 Load-deflection characteristics of two-bar system.


12.8
Chapter 12 - Nonlinear Analysis

Eq. (12.5) expresses a formulation of equilibrium between external loads (Rext = R)


and internal loads Rint = K(r) r. The stiffness K(r) is denoted secant stiffness.

The Eq. (12.6) can be solved, i.e. r can be determined for a given R, analytically.
However, in general this is not possible and iterative methods need to be used. Then it
is convenient to express the equilibrium condition, Eq. (12.5) on a differential form:

d
dR = ( K (r )r )dr = K I dr (12.8)
dr
where
d
K I (r ) = ( K (r )r )
dr
is denoted the tangent stiffness or incremental stiffness. The formulation (12.8) allows
a solution by initial value problem or incremental methods. Such a method may be
combined with or replaced by iterative methods. This topic is discussed in Section
12.5.

The incremental stiffness KI( r) for the problem defined by Eqs (12.6 – 12.7) is

2 EA d ⎛⎛ r ⎞⎛ r ⎞ ⎞
K I (r ) = α o ⋅ ⎜ ⎜1 − ⎟ ⎜1 − ⎟ r ⎟
A dr ⎝ ⎝ h ⎠ ⎝ 2h ⎠ ⎠
2 EA 2 ⎛ r 3⎛ r ⎞ ⎞
2

= α o ⎜1 − 3 + ⎜ ⎟ ⎟ (12.9)
A ⎜ h 2 ⎝ h ⎠ ⎟⎠

2 EA 2 6 EA 2 r ⎛ r ⎞
= αo + α o ⎜ − 1⎟ = K o + K G
A A h ⎝ 2h ⎠

Here, Ko is the linear (initial) stiffness, while KG represents the change in incremental
stiffness due to change of geometry by deformation. It is sometimes called geometric
stiffness.

Stiffness concepts

The different stiffness concepts (Ko, KG, KI, K) are shown in Fig. 12.4.

Ko- linear (initial) stiffness


K - secant stiffness
KI - incremental (tangent) stiffness
KG-geometrical stiffness (negative in
the fig.)

a) At a stable point b) At an unstable point

Figure 12.4 Stiffness definitions.


12.9
Chapter 12 - Nonlinear Analysis

As illustrated in Fig. 12.4b the incremental stiffness in an unstable point is zero, or

dR = K I dr = ( K o + K G )dr = 0

This condition is analogous to the stability criterion in case of linearised buckling,


which for a beam is:

( K − λ P Kσ ) r = 0

as addresses in Chapter 2.9.4.

This means that buckling occurs when the “load factor” λP is large enough that λPKG
equals the elastic stiffness, K.

Further comments about the two-bar problem

Assume that the bars initially have an axial force, S0 (positive in compression). When
an external load, R is applied, the total axial force in each bar is denoted by S. The
axial force imposed due to the load, R is then S – S0, and the following equation
applies

S − S0 = EAε = Aσ (12.10)

And the following equation replaces Eq. (12.1)

2 EA 2
R = 2 S sin α = sin α 0 ⋅ cos α 0 ⋅ r + 2 S0 sin α (12.11)
A

where the first term is as derived before. In particular it is noted that

h−r
sin α ≈ α ≈ and cos α ≈ cos α 0 ≈ 1.0
A

K(r) in Eq. (12.7) becomes

K (r ) = K 0 + K g + K s (12.12)

Where the additional term is

h−r r
K s = 2S0 = 2S0α 0 (1 − ) (10.12a)
A h

KI in eq. (12.9) becomes

KI = K0 + KG + Kσ (12.13)

where

12.10
Chapter 12 - Nonlinear Analysis

2S0
Kσ = − (12.13a)
A

These terms constitute


¾ the linear stiffness, K0
¾ the initial displacement(geometric) stiffness, KG
¾ the initial stress stiffness, Kσ

Ko is familiar from small displacement structural analysis. KG reflects the effect of


changing deplacement on the stiffness and Kσ is the effect on the stiffness of the initial
member forces. Initial means existing prior to further displacement.

K0 and Kσ are essential to any linear stability analysis wherein the bucling load is
found which gives an internal force (e.g. N) such that the stiffness K0 + Kσ is singular.
Kσ is also called the geometric stiffness because it represents the change in the forces
maintaining the structure (e.g. the bar) in equilibrium, which results from a rotation,
with the internal forces (e.g. S0) remaining constant.

It is interesting to consider the special care that h = 0, with a pretension force S0 = − S0


(Where S0 is positive in tension). Noting that α 0 = h / A the governing equation,
Eq. (12.5), with account of the geometric stiffness, Eq. (12.13 a) becomes

⎡ EA r 2 ⎤ 2 S0
R=⎢ ( ) r + r
⎣ A A ⎥⎦ A

which describes the force-displacement relationship for a pretensined cable structure.


It is observed that the incremental stiffness has two terms – one initial deplacement
(KG) and one initial stress term, namely:

3EA S
KG = ( 3
r ), Kσ = 2 0
A A

R(N)

800
R, r S0 (Tension) Formatted: Font: 10 pt

S0 = 104
600

Increasing linearity
a = 2l
400 a = 5000 mm
S0 = 102
E = 200×106 N/mm 2
200
A = 0.25 mm 2
h = 0 mm
0
10 20 30 40 50 r,mm S0 = 102 or 104 N
Fig.12.5 Load-displacement relationship for a pretension cable with lateral load.

12.11
Chapter 12 - Nonlinear Analysis

dR
It is also interesting to establish KI directly by K I = for the total problem. Based
dr
on the equilibrium equation

h−r
R = 2S
A
and,

⎛h 1 r⎞
S − S0 = EAε ≅ EA ⎜ − ⎟r
⎝A 2 A⎠
dS
the drivative becomes :
dr

dS ⎡h r ⎤
= EA ⎢ − ⎥
dr ⎣A A⎦

Then, the tangent stiffness is

dR d ⎡ h − r ⎤
KI = = 2S
dr dr ⎢⎣ A ⎥⎦
dS h − r d h−r
=2 + 2S ( )
dr A dr A

2 EA ⎛ h r ⎞⎛ h − r ⎞ 1
= ⎜ − ⎟⎜ ⎟ − 2 S0
A ⎝ A A ⎠⎝ A ⎠ A

2 EA ⎛ h ⎞ 2 EA ⎡ hr ⎛ r ⎞ ⎤
2 2
S0
= ⎜ ⎟ + ⎢− 2 + ⎜ ⎟ ⎥ − 2
A ⎝A⎠ A ⎢⎣ A ⎝ A ⎠ ⎥⎦ A

= K 0 + K G + Kσ (12.14)

2EA
K0 = (12.15a)
A

2 EA ⎛ h ⎞
2
⎡⎛ r ⎞ 2 r⎤
KG = ⎜ ⎟ ⎢⎜ ⎟ − 2 ⎥ (12.15b)
A ⎝A⎠ ⎢⎣⎝ h ⎠ h ⎥⎦

S0
Kσ = −2 (12.15c)
A

Another problem is concerned with a cable with a pretension force N0 and its own
weight and an initial geometry defined by w0.

12.12
Chapter 12 - Nonlinear Analysis

u
N
N0

N0
w0 (Initial geometry)
u
Fig. 12.6. Load-displacement relationship for a catenary cable.

Comments on strain expressions

In the present derivations the engineering strain, ε defined by ε = ( A − A 0 ) / A 0 has been


used. As discussed later other strain definition may be applied – much as the Green’s
strain, ε G discussed subsequently. However, for the bar or cable discussed above, it
follows by definition that

1 A 2 − A 20 1 A − A 0 A + A 0 1 1
εG = 2
= = ε (ε + 1 + 1) = ε + ε 2 ≈ ε
2 A0 2 A0 A0 2 2

for structural materials where typically ε ≤ 0.003. Also, with small strains the length
of the stress is the same, i.e.

A A
=
cos α 0 cos α

Moreover, the cross-sector area remains the same.

Also the engineering stress σ can be assumed to the same as the continued mechanics
stress. Cauchy stress and 2nd Piola- Kirchoff stresses in local coordinates .

Generalization

The stiffness expressions (12.16, 12.18) can be generalized to systems with many
degrees of freedom

K (r )r=R (12.16a)
implying

K I (r)dr = ( K o + K G (r) ) dr = dR (12.16b)

where R and r are load and displacement vectors, respectively. Also, the various
stiffness concepts indicated in Fig. 12.4, can also be generalized.

12.13
Chapter 12 - Nonlinear Analysis

Eq. (12.16a) expresses equilibrium between external loads, R and internal (reaction)
forces, Kr. The formulation (12.16a) with secant stiffness, however, is not very
practical. The differential formulation (12.16b) may be written on a finite incremental
form

K I Δr = ΔR (12.17a)
implying

Δr = K −1 ΔR (12.17b)

ΔR and Δr are corresponding increments in load and displacement, respectively. With


a given condition (r, R), KI can be calculated and the displacement increment Δr due
to a load increment, ΔR can be calculated by Eq. (12.17b).

12.1.3 Nonlinear material behaviour

Sofar a linear (elastic) relationship between stress and strain has been assumed.
Material tests of metals show that the linearity does not apply when the stress exceeds
a level, σP proportionality limit. Above this level a nonlinear elasto-plastic condition
prevails. See the two typical relationships in Fig. 12.7.

In the example with the mild steel the stress reaches a plateau, yield stress level, σY
until the stress again increases. This phenomenon is denoted hardening.

a) Mild steel b) High strength steel, aluminium

Figure 12.7 Stress-strain curves for metals.

Unloading from a stress condition above σP takes place along a straight line parallel
with the initial linear stress-strain relationship, as shown by the dashed line in Fig.
12.7. When the stress is zero, a residual plastic strain, εP remains (see Fig. 12.7a).

The nominal or engineering strain is defined as the ratio of the change of length over a
given gange length to the original length, l0 as follows

l − l0 l
εe = = −1 ( 12.18)
l0 l0

12.14
Chapter 12 - Nonlinear Analysis

Comment on nominal and true stress and strain

In uniaxial tests used to obtain the stress-strain curve for a given material, the test-
piece cross-sectional area and length change with the onset of plasticity. Assuming
that the load remains constant on the test-piece, the stress will continue to rise, not fall,
as the neck develops before final fracture. Therefore it becomes necessary to take
account of the ‘current’ cross-sectional area and length, rather than refer to the original
(nominal) dimensions of the test-piece. In order to obtain a better understanding of
plastic flow behavior, ‘true’ stresses ans strains are defined.

To take account of the change in the dimensions of the test-piece, the true stress, σtrue
is defined as the force over the current (instantaneous) area, A, as follows:

F
σ true = (12.19)
A

The true strain (also called natural or logarithmic strain), εtrue, is defined as follows:

l dl
ε true = ∫ (12.20)
l0 l

Where l is the current (instantaneous) length. Using the incompressibility (constant


volume) assumption in the plastic flow of metals, i.e. Al = A0l0 , the following
relationship between the nominal and current dimensions is obtained:

A0l0 A
A= = 0 (12.21)
l 1+ ε0

Substituting this value of A in equation in equation (12.19), the true stress can be
expressed as follows:

F (1 + ε 0 )
σ true = = σ (1 + ε 0 ) (12.22)
A0

Integrating the expression for the true strain in equation in equation (12.20) gives:

⎛l ⎞
ε true = ln ⎜ ⎟ = ln(1 + ε 0 ) (12.23)
⎝ l0 ⎠

which is the total (logarithmic) strain between the original and current limits of length.

Therefore, the true plastic strain component can be obtained by subtracting the elastic
strain from the total strain, as follows:

σ0
ε plastic = ε true − ε elastic = ln(1 + ε 0 ) −
(1 + ε 0 ) (12.24)
E
If the unaxial stress-strain curve in Fig. 12.7, which is derived from the load-extension
readings from a uniaxial tension test, is re-plotted with true-stress vs. true strain, as
12.15
Chapter 12 - Nonlinear Analysis

shown in Fig. 12.8, it can be seen that the curve continues rising beyond the point
where necking appears. The curve clearly indicates a ‘strain hardening’ effect, i.e. the
material becomes harder as the strain is increased. It shouldbe noted that the true
stress-strain curve is strictly only valid up to the onset of necking, since the formation
of neck gives rise to a complex state of stress which is no longer uniaxial, i.e. the
stress is not simply the force divided by the cross-sectional area.

Fig. 12.8. Nominal and true stress and strain curves.

The elastic strain is expressed by Hooke’s law: ε e = σ / E

Figure 12.9 Definitions of material properties.

To deal with nonlinear material properties additional terminology needs to be


introduced.

The stress in point A, may be expressed as:

σ = ES ε (12.25)

where ES is the secant modulus, which depends upon the stress (strain) level.

When loading is introduced at A, the change of stress, Δσ can be obtained from

Δσ = ET Δε (12.26)

12.16
Chapter 12 - Nonlinear Analysis

where ET is the tangent modulus.

By unloading from A Hookes law applies:

Δσ = E Δε (unloading) (12.27)

Effect of nonlinear material behaviour on structural behavior

The effect of elasto-plastic behaviour may be illustrated by considering buckling of a


simply supported column with length, A . The critical buckling stress is then (according
to F. Engesser, 1889) obtained by replacing E with the tangent modulus, ET in the
expression for the Euler buckling stress:

π 2 ET
σ Cr = ,λ =A / I / A (12.28)
λ2

ET depends upon the stress σ = σCr (see Fig. 12.7). Eq. (12.28), therefore, could be
solved by iteration by first assuming σCr, calculating ET corresponding to σ = σCr,
calculating a new σCr by Eq. (12.28); use new σCr to calculate ET etc. This example is
to illustrate that nonlinear material behaviour (plasticity) affects the behaviour. A
more refined approach is needed to calculate the ultimate strength accurately.
Moreover, the effect of initial imperfections and residual stresses need to be accounted
for.

Other non-linear material problems

Material non-linearities are classified into three categories:

¾ Time-independent behaviour such as the elastic-plastic behavior of metals in


which the structure is loaded past the yield point as briefly described above.
¾ Time-dependent behaviour such as creep of metals at high temperatures in
which the effect of variation of stress/strain with time is of interest and a power
law stress-strain relationship is often used.
¾ Viscoelastic/viscoplastic behaviour in which both the effects of plasticity and
creep are exhibited. Here the stress is dependent on the strain rate and the
material behaviour can be represented by a combination of a spring and a
dashpot.

12.2 Stiffness relationship for beam with axial force

12.2.1 General

In Section 12.1.2 the stiffness relationship for a two-bar problem was established.

This section deals with element stiffness relationships (S = kv) which account for the
effect of change of geometry. When such element relationships have been established,
the relationship for an entire structural system can be established by the direct stiffness
method
12.17
Chapter 12 - Nonlinear Analysis

S i = k i v i ⇒ R = Kr
Only elasto-plastic behaviour will be pursued herein.

In Section 2.9.3 the stiffness matrices for trusses and beams with axial forces have
been derived based on assumed polynominal shape functions and the principle of
vertical work. The result was written as

S = (k 0 + k σ ) v (12.29)

(when axial forces in tension are defined as positive)

where
k 0 – stiffness matrix corresponding to small displacements.
k σ - initial stress (geometric) stiffness matrix.

For beams with an axial force it is actually possible to exactly establish the stiffness
relationship. This approach is based on the solution of the govering differential
equation, as illusrated in Appendix A.

It is interesting to compare the approxiomate stiffness matrix obtained in Section 2.9.3


with the exact one obtained in Appendix A.

12.2.2. Comparison of alternative stiffness matrices for lateral deformations of a


beam with axial force

In Section 2.9.4 the stiffness relationship involving lateral deformation and rotations
for a beam with axial force. The exact stiffness relationship is derived based on the
differential equation in Appendix A.

It is observed that the exact stiffness matrix, k is the same as for a beam without axial
force, but with each stiffness term multiplied by a φi-function, which depends upon
axial force. The analytical expressions for these φi-functions are given by Eq. (A.16),
and they are plotted as a function of

P PA 2 ⎛ A P π ⎞
ρ= = 2 ⎜β = = P ⎟⎟ (12.30)
PE π EI ⎜ 2 EI 2
⎝ ⎠

in Fig. 12.10 . Note that PE (Euler load) does not depend on the boundary condition of
the beam.

12.18
Chapter 12 - Nonlinear Analysis

Figure 12.10 Stability (Livesley) functions.

It is observed that all φi-functions are equal to 1.0 for β = ρ = 0, i.e. a case with zero
axial force.

For compressive axial force (positive ρ), all φi’ s (and the stiffness), except φ4,
decrease with increasing axial force. For tensile axial force, all φi’ s, except φ4,
increase with the magnitude of P.

The exact stiffness matrix thus established based on linear elstic material behaviour is
applied in computer programs like USFOS.

The stiffness matrix obtained in Section 2.9.4 is approximate because the interpolation
functions do not satisfy the differential equation. However, accurate results may be
obtained by using several finite elements.

Now, typical terms in the stiffness matrix k’NL in Eq. (A.19) may be compared to the
corresponding term of k’L = k’ – k’σ .

For instance,

12 EI 12 EI 12 EI 12 P
k ' NL (1.1) = φ 5 ≈ 3 [1 − ρ ]= 3 − 2 (12.31a)
A 3
A A π A

(by visual consideration of the diagram in Fig. 12.10. A formal linearization may be
made by considering analytical expressions for the φi-functions.)

12 EI 6P
k 'L (1.1) = 3
− (12.31b)
A 5A

which shows that a linearization of k’NL(1.1) with respect to ρ yields a result very close
to k L' (1.1) . The same applies to the other terms.

12.19
Chapter 12 - Nonlinear Analysis

12.3 Formulations for non-linear geometrical behaviour of bars and beams

12.3.1. General

Numerical solutions of problems involving geometric non-linearity (GNL) usually


attempt to replace the continuous non-linear displacement by a series of linearised
increments of displacements. The theories used in the analysis of GNL problems using
the FE method are not familiar to most engineers and, unfortunelately, suffer from a
multitude og jargon terms – usually terms named after the scientists and
mathematicians who first invented them, e.g. Lagrange, Green, Cauchy, Almansi,
Piola-Kirchoff, etc.

In this section some of features of GNL problems are discussed in order to provide a
brief introduction to this type of non-linearity. Further information on all aspects of
GNL analysis using FE can be found in the textbooks by Crisfield [1991, 1997], and
Zienkiewicz and Taylor [1991].

Classifications of GNL problems

Most engineers associate GNL problems with ‘large’ displacements effects such as
buckling. However, GNL problems can also involve small displacements (in fact,
GNL theory often predicts less displacement than the corresponding linear load-
displacement theory). In order to classify GNL problems, it is best to focus on whether
strain is assumed small or large, as this has a direct influence on which non-linear
definitions of stresses and strains are used.

(a) Small strain GNL problems

These problems are associated with small or large rotations. Problems in which small
rotations occur include shallow struts, shells and arches deflected by a transverse load,
clamped circular plates under transverse point loads and shallow spherical caps.
Examples of large rotation problems include a fishing rod bent under the weight of a
heavy fish, buckling of an imperfect Euler strut, and a deep arch.

(b) Large strain GNL problems

These are the most complex of all GNL problems and are usually associated with
metal forming and manufacturing processes, such as deep drawing of drink cans,
forging, extrusion and rolling. With large strains, it is also important to model material
non-linearity such as plasticity. An exception is rubber which can undergo very large
strains, of the order of unity, but remains elastic. This type of behaviour is called
‘hyperelastic’ or ‘non-linear elastic’ behaviour. The constitutive equations for rubber
can be derived from the expressions for the potential energy density.

Definitions of stresses and strains in GNL problems

The conventional definition of ‘engineering strain’ may not be adequate when dealing
with GNL problems because it measures the change in length over the original
(undeformed) length. A more suitable definition is one which takes into account the
new length, such as the so-called ‘logarithmic strain’. Two strain definitions which
12.20
Chapter 12 - Nonlinear Analysis

have been widely used in GNL problems are called the ‘Green strain’ and ‘Almansi
strain’. These strains are based on the ‘square’ of the length, and have been shown to
be very effective in dealing with a wide range of GNL problems, including large strain
problems

As in the strain measures, the ‘engineering’ or ‘nominal’ stress, defined as the force
divided by the original undeformed area, may be inappropriate for use in GNL
problems in which the cross-sectional area may exhibit large changes. Instead, as in
material non-linearity, a ‘truestress’ (also called ‘Cauchy stress’) can be defined as the
force divided by the ‘current’ cross-section area, rather than the original area.

Another feature of GNL problems is that the relationships between stresses and strains
have to be carefully defined. In conventional elasticity equations, stresses are linked to
strains through the constitutive law, i.e. Hooke’s law. In GNL problems, in addition to
the constitutive equations, stresses are usually associated with the corresponding
strains using the ‘virtual work theorem’ or ‘total potential energy theorem’. This is
important in large strain problems, where such stresses are called ‘work-conjugates’ to
the corresponding strains.

Conservative and non-conservative (follower) loads

A conservative load is that which always applies in a fixed direction regardless of the
deformation of the body. A typical example is a gravitational load, which aways
applies vertically. A non-conservative (follower) load is one which changes its
direction during the deformation, i.e. it follows the deformation of the body, e.g. an
internal pressure in a vessel changes its position and direction as the vessel deforms, in
order to remain perpendicular to the surface. Figure 12.11 shows a schematic
representation of conservative and non-conservative loads.

Fig. 12.11. Conservative and non-conservative loads.

12.21
Chapter 12 - Nonlinear Analysis

Formulations

Formulation of geometrical non-linear problems requires choice of reference systems


for describing the structures geometry and deformations.

The most common modes of describing the deformations of solids and fluids are the
Eulerian and the Lagrangian approaches:
¾ Eulerian description of motion is also denoted spatial description, because it
refers to what happens at a certain place in space. In the description the current
coordinates x1, x2, x3 and time t are the independent variables. This description
is especially suited for hydrodynamics.

¾ Lagrangian description of motion refers to what happens at a material particle.


Hence, this description is also denoted material description. Independent
variables are the initial coordinates x1, x2, x3 and time t.

The Lagrangian description of motion is commonly preferred in structural mechanics


since the initial configuration is usually known. This formulation will be adopted in
the present work.

Adopting a Lagrangian description, the geometry at a certain load condition is to refer


it to
¾ the initial geometry/fixed global coordinate system
¾ assume that the loading takes place in a stepwise manner and refer each
element to a local coordinate system that follow the structure during
deformation. This is called a corotational system . The coordinates are
updated.

The purpose is to show how incremental relations like

ΔR = K I Δr

can be established for various types of structures. However, since the assemble of the
global stiffness is straightforwardwhen the element relationship S = kv is known, the
focus will be on the element relationship.

It is interesting to note that the total Lagrange formulation for the two-bar problem
corresponds to Eq. 12.13.

I = K 0 + K G + Kσ
K TL (12.31c)

The updated Lagrange formulation corresponds to

I = K 0 + Kσ
K UL (12.31d)

where S (positive in tension) is the actual S at any load step.

A simplified version of the latter reference system is described in Section 12.3.2. The
Section 12.3.3 the use of the first type of reference system will be demonstrated.

12.22
Chapter 12 - Nonlinear Analysis

12.3.2 Method with updated coordinates

General

In this procedure the element stiffness relationships are first determined in a local
system and are then transformed into a fixed, global coordinate system before the
global stiffness relation is assembled. The location of the local coordinate systems
would have to be updated when deformations cause the geometry to change. In this
method the non-linear geometrical effects are accounted for by continuously changing
the transformation matrices.

It is assumed here that each element behave in a linear manner when it is referred to a
corotational coordinates. This implies that small deformations on the local level are
assumed.

Element relationships

Fig. 12.12 shows an arbitrary plane element in its initial and deformed condition. The
element is connected to global nodes a and b and a local axis defined by a straight line
through these two points. The nodal forces (S) (in the local system) are defined in the
conventional manner. The deformations of the element are completely described by
the axial elongation, u and rotation ωa and ωb, which are referred to the chord ab (Fig.
12.12b). By means of these three deformation parameters the force vector S can be
uniquely obtained from

⎡ EA ⎤
⎢ 0 0 −
A ⎥
⎢ ⎥
⎡ S1 ⎤ ⎢− 6 EI −
6 EI
0 ⎥
⎢S ⎥ ⎢ A 2 A2 ⎥
⎢ 2 ⎥ ⎢ 4 EI 2 EI
⎢S ⎥ ⎢ 0 ⎥⎥ ⎡ω a ⎤
⎢ω ⎥ = k~~
S=⎢ 3⎥ = ⎢ A A v (12.32)
EA ⎥ ⎢ b⎥
⎢S 4 ⎥ ⎢ 0 0 ⎥ ⎢⎣ u ⎥⎦
⎢S 5 ⎥ ⎢ A ⎥
⎢ ⎥ ⎢ 6 EI 6 EI
⎢⎣ S 6 ⎥⎦ ⎢ 2 0 ⎥
A A2 ⎥
⎢ 2 EI 4 EI ⎥
⎢ 0 ⎥
⎣ A A ⎦

a) Nodal forces b) Deformations

Figure 12.12 Forces and deformations in a local coordinate system.

12.23
Chapter 12 - Nonlinear Analysis

The vector ~v describes the deformations relative to the local coordinate system xy. In
addition the element will be subject to rigid body motions (translation and rotation)
that correspond to the location of the local xy-system relative to the global x y -system,
and are defined by the global coordinates of nodes a and b. Rigid body motions do not
imply internal forces in the element. Hence, the relationship (12.32) does not depend
on rigid body motions and is valid for any position of the local system. The reason
~
why k has dimensions 6 by 3 is that the three components of the rigid body motion is
removed from Eq. (12.32).

The assumptions of small displacements implies that u, ωa and ωb are small.

The next step is to express the global nodal forces, S when the global displacement v
are known. S and v are defined in Fig. 12.13a and b.

a) Nodal forces b) Displacements

Figure 12.13 Global nodal forces and displacements. All angles in for figure are
positive.

The relationship between the nodal forces defined in the global and local coordinate
system is given by

S = TT S (12.33)

where

⎡ cosθ sin θ 0 ⎤
⎢− sin θ cos θ 0 0 ⎥
⎢ ⎥
⎢ 0 0 1 ⎥
T=⎢ ⎥
⎢ cos θ sin θ 0⎥
⎢ 0 − sin θ cos θ 0⎥
⎢ ⎥
⎢⎣ 0 0 1⎥⎦

and θ is the angle between the x - and x-axes (Fig. 12.12). By combining Eqs (12.32,
12.33) gives
~
S = T T S = T T k~
v =k * ~
v (12.34)

12.24
Chapter 12 - Nonlinear Analysis

~
k* = T T k can be expressed explicitly by carrying out the matrix multiplication by
hand.

If the local deformations ~v are expressed by the global displacements v Eq. (12.34)
can be transformed into a relationship between S and v . This can be achieved as
follows.

The coordinates of nodal points a and b in deformed condition, are:

xa = x a 0 + v1
ya = y a0 + v2
(12.35)
xb = xb 0 + v 4
yb = y b 0 + v5

where x a 0 etc. are the coordinates of the initial condition. The chord length ab in
deformed condition is obtained as

A= ( x b − x a )2 + ( y b − y a )2 (12.36)

Relative to the initial condition the rotation of the chord axis ab (Fig. 10.15) is

ψ =θ −θ 0
⎛ yb0 − y a 0 ⎞
θ 0 = arcsin ⎜⎜ ⎟⎟ (12.37)
⎝ A0 ⎠
⎛ yb − y a ⎞
θ = arcsin ⎜ ⎟
⎝ A ⎠

where A 0 is the element length in the initial condition. To determined θ0 and θ in a


unique manner, the sign of cosθ0 and cosθ needs to be checked.

The local displacements are then obtained (from Fig. 12.12b, 12.13b) as:

⎡ω a ⎤ ⎡v3 + ψ ⎤
v = ⎢⎢ω b ⎥⎥ = ⎢⎢v6 + ψ ⎥⎥
~ (12.38)
⎢⎣ u ⎥⎦ ⎢⎣ A − A 0 ⎥⎦

Eqs (12.34, 12.38) make it possible to find the element forces S for any known
displacements, v . This relation for an individual element can then be used to express
the relationship for the structural systems in the following manner.

Structural system relationships

The relationship between the displacement vector for element No. i and the global
displacement vector r is expressed by the kinematic relation:

12.25
Chapter 12 - Nonlinear Analysis

vi = air (12.39)

Suppose that the structure is subjected to given external forces, corresponding to a


vector R. The principle of virtual displacements yields

∑ (v )
~ i T
Si =~
rTR
i

where the (~) indicates virtual displacements. Since virtual displacements are
compatible
~i
v = ai ~
r

which yields

~
rT ∑ (a )
i
i T
Si =~
rTR

Since ~
r can be arbitrary, the equilibrium condition becomes

∑ (a )
i
i T
Si =R (12.40)

The left-hand side of Eq. (12.40) is a function of the displacement vector, through Eqs
(12.34, 12.38, 12.39). Eq. (12.40) is a non-linear equilibrium equation. The left-hand
side expresses the sum of internal forces from each element in the structure.

To find the displacements r that fulfill Eq. (12.40) requires iteration. If r’ is a


displacement vector that does not fulfill Eq. (12.40), the corresponding element forces
S i do not satisfy this equation. Hence, this displacement condition implies a set of
unbalanced forces or residual forces, given by:

R r = ∑ ai ( ) T
S ′i − R (12.41)

Rr represents the additional forces required to fulfill the global equilibrium in this case
(with r’).

R + R r = ∑ (a i ) T S ′ i (12.42)
i

As indicated in Section 12.1.2 and discussed more in detail later, an incremental


stiffness relation Eq. (2.103), will be applied. In the following it will be shown how
this relationship can be established.

In a local coordinate system the incremental stiffness relation for an element can be
obtained as shown in Section 2.9.4 and may be written as: (when a tensile axial force
is considered positive)
dS = ( k 0 + k σ ) dv = k I dv (12.43)
12.26
Chapter 12 - Nonlinear Analysis

where S and v is defined as in Fig. 12.14 and the S vector correspondingly.

Figure 12.14 Local degrees of freedom for an element.

k0 and kσ are the elastic and geometrical stiffness matrices, respectively as defined by
Eqs (2.110) and (2.111). However, now the axial force is considered positive if in
tension (this explains the +sign before kσ in Eq, (12.43)). This force can be calculated
from the axial strain by:
u
P = EA (12.44)
A0

By transformation to a global system

d S = T T dS = T T k I dv = T T k I dv = k I dv (12.45)

where
k I = TT k I T

S and v are defined in Fig. 12.12 and T in Eq. (12.33). Kinematic relationship and
static equilibrium for a system of elements can be expressed as:

dv i = a i dr, dR = ∑ a i d S i ( )
i

The incremental system stiffness then becomes

dR = ∑ a i ( )T
dS i = ∑ a i ( ) T
k iI a i dr
i i

or
dR = K I dr (12.46)
where
K I = ∑ ai ( )T
k iI a i
i

Eqs (12.40, 12.46) form the basis for obtaining solutions of the geometrically non-
linear problem, by the methods described later in Section 12.4.

The method described in this section is relatively simple in use since most of the
matrices are well known from linear analysis.

12.27
Chapter 12 - Nonlinear Analysis

Two practical issues should be noted when implementing this method in a computer
program. The stiffness matrices k0 and kσ could be calculated based on initial element
length A 0 , rather than the updated length, A , without loss of accuracy. In this way the
element stiffness matrices can be calculated once and stored and retrieved when
needed. However, the axial force, P in kσ need to be updated. The second issue is that
the coordinate updating, Eq. (12.35), element length, Eq. (12.36) and rotations, Eq.
(12.36) need to be done accurately to avoid that round off errors reduce/destroy the
accuracy and cause convergence problems in the iterative solution.

This formulation presented in this section is often called “Updated Langrange


formulation”. However, this is not quite true, since corotated coordinate system in
general will be a curvilinear one at not a cartesian one as in this formulation. The
formulation is, therefore, an approximately updated Lagrangian formulation.
Alternatively it may be called a method based on corotated coordinates or corotating
coordinates.

12.3.3 Total Lagrangian formulation for a beam with axial and lateral
deformations
In the previous section a local coordinate system was assumed to follow the rigid body
displacements as the structure deforms. The deformations are referred to these
updated coordinate systems. Contrary to this, the total Lagrange formulation is based
on a fixed coordinate system. When calculating displacements and strains, the
expressions used in previous sections for the strain then need to be modified.

In this section a beam undergoing axial and lateral deformations will be used to
illustrate the total Lagrangian formulation. Classical linear beam theory will serve as
basis for this. This is because shear deformation contribute little to the behaviour of
beams that have large deformations.

Fig. 12.15 shows a beam element undergoing large displacements. The displacements
of a plane beam are uniquely described by the displacement u and w of the neutral
axis. For a case with small displacements the strain at a location (x, z) is

ε x ( x, z ) = u, x − z ⋅ w, xx ( x) (12.47)

If the displacements and rotations are large (but the strains are small), the strain due to
displacement in the x-direction remains unchanged. However, the large deflection,
w(x) causes an additional axial strain. This effect can be estimated as follows:

a) Element with large displacement b) Geometric locations for a differential


element of length dx

Figure 12.15 Kinematics for beam with large displacements.

12.28
Chapter 12 - Nonlinear Analysis

Let a small lateral displacement w = w(x) take place. Thus each differential length dx
is changed to a new length, ds, where ds > dx because the distance between supports is
not allowed to change. From Fig. 12.14b.

⎛ w, 2 ⎞
(
ds = 1 + w, 2x )
1/ 2
dx ≈ ⎜⎜1 + x ⎟⎟ dx
2 ⎠
(12.48)

where the latter approximation comes from the first two terms of the binomial
expansion. The approximation is valid is w, 2x << 1 , which restricts this development to
small rotations. Axial membrane strain, εm in the bar due the large deflection, w is
therefore

ds − dx w, 2x
εm = ≈ (12.49)
dx 2

In the linear theory of elasticity we ignore terms of order w, 2x . But here we seek the
consequences of retaining the more important of the higher-order terms that linear
theory neglects. We are taking a physical approach to formulating these terms. They
may also be obtained by a systematic procedure of linearization, as will be touched
upon in Appendix C.

This means that a beam with axial displacement of the neutral axis u = u(x) and lateral
displacement w = w(x) has a strain of

1
ε x = u, x + w, 2x − z ⋅ w, xx (12.50)
2

It is noted that if the axial strains are not small, an additional terms needs to be
included in Eq. (12.50). The strain then becomes

1 1
E xx = u, x − z ⋅ w, xx + u, 2x + w, 2x (12.51)
2 2

This is the so-called Green’s strains as presented in Appendix B. For metal structures,
however, the additional term 12 u , 2x is negligible.

Another issue is the effect of an initial lateral deflection, w (x) of the beam. A beam
with initial lateral deflection may also be considered a shallow arch. In this case the
length of an arch with a projected length along the x-axis equal to dx would be
( )
ds = 1 + 12 w , 2x dx , according to Eq. (12.40). When an additional lateral deformation,
w(x) occurs, the resulting arch length becomes

[
ds* = 1 + (w , x + w, x ) ]
2 1/ 2 ⎡ 1 2⎤
dx ≈ ⎢1 + (w , x + w, x ) ⎥ dx
⎣ 2 ⎦
(12.52)

and the membrane strain due to lateral direction becomes

12.29
Chapter 12 - Nonlinear Analysis

ds * − ds 1
εm = = w , x w, x + w, 2x (12.53)
ds 2

For metal beams with initial lateral deflection, w (x) the Green strain becomes

1
E xx = u , x − z ⋅ w, xx + w , x w, x + w, 2x (12.54)
2

Alternatively Exx may be written as

E xx = u, x − zw, xx + (w, x + w, x ) 2 − w, 2x

Stresses

Stresses may be defined with reference to the deformed structure or its initial
configuration. True stresses referred to the deformed configuration are denoted σij
(Eulerian or Cauchy stresses). Stresses referred to the initial configuration are denoted
Sij (2nd Piola-Kirchhoff stresses). The latter stress is consistent with the Green’s strain
that also refers to the initial configuration. If the Green’s strain is applied the stresses
should therefore be Piola-Kirchoff stresses.

For the one-dimensional case the Piola-Kirchoff stress is related to the Eulerian stress
(defined in the deformed configuration) as:

∂X ⎛ ∂u ⎞
S xx = σ xx = ⎜1 − ⎟ σ xx
∂x ⎝ ∂x ⎠

where X and x are coordinates of the initial and deformed configuration, respectively.

In metal structures the strains are, but the rotation may be large and nonlinear
geometric effects. In such cases Sxx and σxx are for all practical purposes, considered
to be equal.

Virtual work (Principle of virtual displacements)

Assume that a beam structure in an equilibrium position is given a virtual axial and
lateral displacement, δu and δw, which are kinematically consistent with the boundary
conditions. By neglecting volume forces the equation of virtual work may according
to Appendix B.2 the equation be expressed a

∫S
V
xx δ E xx = ∫ qu δu ( x)dx + ∫ q w δw( x ) dx
A A
(12.55)

It is noted that Exx and Sxx are expressed by the displacements u and w according to
Eq. (12.54) and Hooke’s law for an elastic beam.

By applying the approach in Appendix B the incremental form of the virtual work can
be written as

12.30
Chapter 12 - Nonlinear Analysis

∫ ΔS
V
xx δ E xx dV + ∫ S xx Δ δ E xx dV + ∫ Δ S xx Δ δE xx dV
V V

⎡ ⎤ (12.56)
= ∫ Δ q u δu dx + ∫ Δ q w δw dx − ⎢ ∫ S xx δ E xx dV − ∫ q u δu dx − ∫ q w δw dx ⎥
⎢⎣V ⎥⎦

If the configuration Cn is in equilibrium the parenthesis on the right hand side of Eq.
(12.56) will vanish according to Eq. (12.55). However, due to approximations in the
solution procedure equilibrium will not be generally satisfied. Hence, the terms in the
parenthesis on the right hand side will serve as equilibrium correction terms.

Finite element model

The incremental stiffness relation for an element can then be established based on Eq.
(12.56) and choice of interpolation functions for the displacements u =[u, w]T. This
implies choice of nodes and degrees of freedom in each node. One option is shown in
Fig. 12.16, based on:

- quadratic polynomial for u, with the parameter vu = (u1, u2, u3)T


- cubic polynomial for w, with the parameters vw = (w1, θ1, w2, θ2)T

The initial lateral deflection, w may be described by the same type of polynomial as
w.

In matrix notation the displacement may be written as

u ( x) = N u v u
w( x) = B w v w (12.57a-c)
w ( x) = N w v w

Figure 12.16 Stiffener elements.

If the matrix notation of Eqs (12.57a-c) are introduced, Exx in Eq. (12.54) may be
written as

E xx = N u , x v u − z ⋅ N w , xx v w + v Tw N Tw , x N w , x v w
1 T T (12.58a)
+ vw Nw,x Nw,x vw
2
or
12.31
Chapter 12 - Nonlinear Analysis

1 T T
E xx = Bv + v Tw N Tw , x N w , x v w + vw Nw ,x Nw,x vw (12.58b)
2
where

B = [N u , x , N w , xx ]
⎡v ⎤ (12.58c)
v=⎢ u⎥
⎣v w ⎦

The first variation in strain, δExx due to a variation in displacement δv is

1
δ E xx = B δ v + v Tw N Tw' x N w , x δ v w + v Tw N Tw , x N w , x δ v w (12.59)
2

(Note: δ E xx = E xx (v + δ v ) − E xx (v ) , which is most easily performed on the scalar


expression (12.59) first. The high order term (δw)2 is neglected.)

From one configuration to another, which is close to the first one, the change in strain
variation is

Δδ E xx = Δv Tw N Tw , x N w , x δ v w (12.60)

The stress increment between the two configurations becomes

ΔS xx = DΔE xx (12.61)

where D is the incremental material stiffness coefficient, elastic or elasto-plastic. See


Section 12.4.

By recognizing the integrand of the first integrand of Eq. (12.56) is a scalar, it may be
written as:

∫ ΔS
V
xx δ E xx dV = ∫ δ E xxT D ΔE xx dV
V
(12.62)

When introducing the strain expression (12.58 b-c) into Eq. (12.62), the expression is
simplified if the following notation is introduced:

v wt = v w + v w (12.63)

Then, by using Eq. (12.62)

12.32
Chapter 12 - Nonlinear Analysis

∫ ΔS xx δ E xx dV = δ v T ∫ (B T D B dV )Δv
V V

+δ v ∫ (B )
D v Twt N Tw , x N w , x dV Δv w
T T

(
+ δ v Tw ∫ N Tw , x N w , x v wt D B dV Δv ) (12.64)
V

(
+ δ v Tw ∫ N Tw , x N w , x v wt D v Twt N Tw , x N w , x dV Δv w)
= δ v T k 00 Δv + δ v T k 0 L Δv w
+ δ v Tw k T0 L Δ v + δ v w k LL Δv w
By introducing

⎡v u ⎤
v w = [0 I ] ⎢ ⎥ = Hv (12.65)
⎣v w ⎦

Eq. (12.64) may be written as

∫ ΔS xx δ E xx dV = δ v T (k 00 + k 0 L H + H T k T0 L + H T k LL H )Δv
V (12.66)
= δ v T k 1Δ v

The second term on the left hand side of Eq. (12.56) becomes

∫ Δδ E xx (
S xx dV = δ v Tw ∫ N Tw , x N w , x S xx dV Δv w )
V V

= δ v Tw k Lσ Δv w = δ v T H T k Lσ HΔv (12.67)
= δ v k 2 Δv
T

The third term on the left-hand side of Eq. (12.56) is neglected as a higher order term.

The load term of Eq. (12.56) may be formulated in matrix notation in a similar manner
as for linear analysis. The resulting virtual work during an increment may be written
as

T
∫ Δq u δ u dx + ∫ Δq w δ w dx = δ v ΔS (12.68)

When neglecting the third term on the left hand side and the parenthesis on the right
hand side, Eq. (12.56) hence becomes

δ v T [(k 1 + k 2 ) Δv − ΔS ]= 0 (12.69)

Before completing the derivation the nodal set of parameters, v may be rearranged
from

v = [u1 , u 2 , u 3 , w1 ,θ 1 , w2 ,θ 2 ]
T
(12.70a)

12.33
Chapter 12 - Nonlinear Analysis

to
v ' = [(u , w,θ )1 ; (u , w,θ )2 ; u 3 ]
T
(12.70b)

The connection between v and v’ is given by the matrix G:

v = Gv ' (12.71)

Such a rearrangement involves a matrix with elements which are 0 or 1.

Introducing

k1' = G T k 1G; k '2 = G T k 2 G


k 'I = k '1 + k '2 (12.72)
ΔS = G ΔS T

Eq. (12.69) turns into

δ v 'T ( k 'I Δ v '− ΔS ') = 0 (12.73)

Since the virtual displacement δv’ is arbitrary, the principle of virtual work gives

k 'I Δv ' − ΔS ' = 0 (12.74)

This is the stiffness relation on incremental form for the beam element. The (9x9)
symmetric matrix is k 'I is called incremental element stiffness matrix. k '1 is denoted
initial displacement matrix (which includes the linear small displacement stiffness
matrix) or large displacement matrix and the other contribution k ' 2 is known as initial
stress matrix or geometric matrix.

In a similar manner the incremental stiffness relation for a beam element based on the
updated Lagrangian formulation can be obtained. In this case the resulting
incremental element stiffness matrix may be written as:

k ' = k '0 + k 'σ (12.75)

where the stiffness matrix k '0 is that due to small deflection theory, namely the first
term on the left hand side in Eq. (12.75), and k σ' is the (linearized) geometric stiffness
matrix, resulting from second integral on the left hand side.

In linear analysis of plane structures the behaviour in plane (axial) behaviour and out
of plane or lateral (beam bending) is uncoupled. In geometrically nonlinear behaviour,
axial and lateral behaviour is coupled. This can be seen for a Total Lagrangian
Formulation from the fact that mean strain depends upon u,x as well as 0.5 w, 2x .

12.34
Chapter 12 - Nonlinear Analysis

Comments

Various finite element models may be established based on the choice of interpolation
polynomials for u and w. Since the axial strain would vary significantly over the
element undergoing large deflections, the displacement u should be approximated by a
quadratic or higher degree polynomial, while w would be interpolated by a cubic
polynomial. Moreover, this choice of displacements reduces the effect of
“selfstraining”, a phenomenon resembling “shear looking” for membrane elements.

When the interpolation polynomials are given, the stiffness matrix is calculated by
means of the expressions (12.73, 12.72, 12.67, 12.64-12.66).

Internal degrees of freedom, e.g. associated with one of the d.o.f. associated with the
displacement u are normally eliminated by static condensation, see Chapter 9.

Finally, the global incremental stiffness relation is obtained by the direct stiffness
method.

12.3.4 Generalization

In the previous section nonlinear geometrical formulations for one dimensional


bars/beams are treated.

The one dimensional formulation presented above can be generalized to two


dimensional problems, i.e. thin plates subjected to in-plane and lateral loads. This
generalization involves

¾ linearized problem in terms of a stiffness and geometric stiffness matrix that


can be used for buckling analysis of plates subjected to in-plane forces per unit
length: Nx = h⋅σx, Ny = h⋅σy and Nxy = h⋅τxy, where h is the plate thickness.
¾ Updated or Total Lagrangian Formulation for geometrically nonlinear
problems.

Linear buckling analysis

Analogous with the expression (2.106) for a beam the virtual work done by in-plane
forces Nx. Ny and Nxy acting on a plate is

Wext = ∫ (N x w x x y , y w, y + N xy (w, x w, y + w, y w, x )) dA
~, w, + N w
~ ~ ~

(12.76)
[
~, w
=∫ w ~, ⎡ N x
] N xy ⎤ ⎡ w, x ⎤
y ⎢
N y ⎥⎦ ⎢⎣ w, y ⎥⎦
x dA
⎣ N xy

Tensile forces are considered to be positive.

By introducing the interpolation polynomial w = Nv

12.35
Chapter 12 - Nonlinear Analysis

⎡ w, x ⎤ ⎡ N, x ⎤
⎢ w, ⎥ = ⎢ N, ⎥ v = Bσ v (12.77)
⎣ y⎦ ⎣ y⎦

the external work done is written as

Wext = v T ⎡⎣ ∫ BσT G Bσ dA⎤⎦ v (12.78)

where

⎡ Nx N xy ⎤
G=⎢ (12.78a)
⎣ N xy N y ⎥⎦

By equating internal and external work it is found that the element stiffness matrix

k = k '0 + k σ' (12.79)

where

k0’ is the stiffness matrix for amall displacement theory as given in Chapter 7.
k σ' = ∫ BσT G Bσ dA (12.79 a-b)

Geometrical nonlinear effects can be formulated for two-dimensional structures, such


as plates and shells according to the same principles as for bars and beams.

Reference is made to textbooks such as Crisfield (1991), Hughes (2000), NAFEMS


(1992), Taylor and Zienkiewicz (2000).

12.4 Nonlinear material behaviour

12.4.1 One dimensional case

In Section 12.1.3 typical one-dimensional stress-strain relationships for metals were


shown, and it was pointed out that a nonlinear relationship occurs when the stress
exceeds the proportionality limit.

A material is called nonlinear if stresses σ and strains ε are related by a strain-


dependent matrix rather than a matrix of constants. Thus the computational difficulty
is that equilibrium equations must be written using material properties that depend on
strains, but strains are not known in advance. Plastic flow is often a cause of material
nonlinearity. The present section deals with formulation of elastic-plastic problems by
considering the one dimensional case.

Assume that yielding has already occurred; then a strain increment dε (between points
A and B) takes place (Fig. 12.17a). This strain increment can be regarded as

12.36
Chapter 12 - Nonlinear Analysis

composed of an elastic contribution dεe and a plastic contribution dεp, so that dε = dεe
+ dεp. The corresponding stress increment dσ can be written in various ways

( )
dσ = Edε e = E dε − dε p ; dσ = Et dε ; and dσ = H ' dε p (12.80)

where H’ is called the plastic tangent modulus as given by ∂σ / ∂ε p . (Note that σ =


H(εp ).) Substitution of the first and third of Eqs. (12.80) into the second yields

1 ⎛ E ⎞
H '= or Et = E ⎜1 − ⎟ (12.81)
1
Et − 1
E ⎝ E + H'⎠

where Et is the tangent modulus. When written in this form, the expression for Et is
similar to a more general expression used for multiaxial states of stress. If E is finite
and Et = 0, then H’ = 0, and the material is called “elastic-perfectly plastic”.

a) Stress-strain plot in uniaxial stress, b) Kinematic and isotropic


idealized as two straight lines, where hardening rules.
σy is the stress at first onset of yielding.

Figure 12.17 characteristic features of one-dimensional stress-strain relationships.

A summary of elastic-plastic action in uniaxial stress is as follows:

1) The yield criterion states that yielding begins when |σ| reaches σY, where in
practice σY is usually taken as the tensile yield strength. Subsequent plastic
derformation may alter the stress needed to produce renewed or continued yielding;
this stress exceeds the initial yield strength σY if Et > 0.

2) A hardening rule, which describes how the yield criterion is changes by the history
of plastic flow. For example, imagine that the material first has been loaded to point B
and then unloading occurs from point B to point C in Fig. 12.17a. With reloading
from point C, response will be elastic until σ > σB, when renewed yielding occurs.
Assume then that unloading occurs from point B and progresses into a reversed
loading as shown in Fig. 12.17b. If the yielding is assumed to occur at |σ| =σB the
hardening is said to be isotropic. However, for common metals, such a rule is in
conflict with the observed behaviour that yielding reappears at a stress of approximate
magnitude σB - 2σY when loading is reversed. Accordingly, a better match to
observed behaviour is provided by the “kinematic hardening” rule, which (for uniaxial
stress) says that a total elastic range of 2σY is preserved.
12.37
Chapter 12 - Nonlinear Analysis

3) A flow rule can be written in multidimensional problems. It leads to a relation


between stress increments dσ and strain increments dε. In uniaxial stress this relation
is simply dσ = Et dε, which describes the increment of stress produced by an
increment of strain. Note, however, that if the material has yet to yield or is
unloading, then dσ = E dε (e.g., in Fig. 12.17a, complete unloading from point B leads
to point C and a permanent strain εp).

The discussion in the foregoing paragraph does not require that post-elastic response
be idealized as a straight line. In other words, Et, need not be constant.

Bar structures with vriable cross-section area

As a simple application of one-dimensional plasticity, imagine that a tapered bar is to


be loaded by an axial force P (Fig. 12.18). Material properties are those depicted in
Fig. 12.17. The bar is modeled by two-d.o.f. bar elements, each of constant cross
section. For elastic conditions, the element stiffness matrix is given by Eq.(2.48),
where E = dσ/dε when |σ| < σY. Upon yielding, the stress-strain relation becomes Et =
dσ/dε. Accordingly, letting Eep represent the “elastic-plastic” stiffness, the element
tangent-striffness matrix is expressed by:

AE ep ⎡ 1 −1⎤
kT = ⎢ −1 1 ⎥ (12.82)
L ⎣ ⎦

where Eep = E if the yield criterion is not exceeded or if unloading is taking place, and
Eep = Et if plastic flow is involved.

In numerical solutions, material may take the transition from elastic to plastic within
an iterative cycle of the solution process. For example, imagine that dε spans εD to εA
in Fig. 12.17a. The problem of “rounding the corner” can be addressed by combining
E and Et according to the fraction m of the total step dε that is elastic. Thus let

εY − ε D
E ep = mE + (1 − m) E t where m = (12.83)
εA −εD

a) Geometry and model b) Load-displacement incrementation

Figure 12.18 Tapered bar subjected to axial loading

12.38
Chapter 12 - Nonlinear Analysis

Alternatively, by substituting stresses for strains and using the fictitious stress σ*
=EεA, we can write m in terms of stresses, as m = (σY - σD)/(σ* - σD). Refinements of
this scheme are possible.

As another simple elasto-plastic problem consider the single degree of freedom


elastoplastic problem shown in Figure 12.19 . Each bar is constructed of an identical
material which has bilinear elastic perfectly plastic behavior. Fig. 12.17a with Et = 0.

Each bar is of cross-sectional area A and elastic modulus E. The yield stress values for
the material in the bars are σ Y(1) , σ Y(2) and σ Y(3) where σ Y(1) < σ Y(2) < σ Y(3) . When the force
f is applied all behaviour is initially elastic. As the load is increased the stress in
element 1, reaches the yield and can carry no further increase of stress. Hence, the
internal resisting force emanating from element 1 from this stage onward is

p (1) = σ Y(1) A (12.84)

and element 1 loses its stiffness for an increasing load.

As the load is increased the additional load must be balanced by additional internal
resisting forces in elements 2 and 3. Eventually, the stress in element 2 reaches the
yield stress and this element also yields, loses its stiffness and can carry no further
increase of load. Thus, all extra load must be taken by element 3.

Finally, element 3 yields and loses its stiffness and the structure can take no further
load. The load at which this element yields is

f ult = A(σ Y(1) + σ Y(2) + σ Y(3) ) (12.85)

Thus, the source of nonlinearity in the elastoplastic problem occurs in the evaluation
of the stress in the elements and hence in the internal resisting force. The stress is
effectively a nonlinear function of the displacements.

The load-displacement curve for this problem is shown in Figure 12.19 and consists of
a piecewise linear curve with the end of each segment signaling the onset of yield of a
new element until failure of the whole structure.

The slope of the load-displacement curve at any stage og the analysis is called the
‘tangential stiffness’ KT. When elastic

KT = 3EA / A (12.86)

As the first bar yields,

KT = 2 EA / A (12.87)

When the second bar yields


KT = EA / A (12.88)
12.39
Chapter 12 - Nonlinear Analysis

and finally, after the last bar yields

KT = 0

a) Three-bar b) Equilibrium c)Load-displacement curve

Fig. 12.19 Three-bar elastoplastic problem

Solution procedures for elasto-plastic problems will be discussed in Section 12.5.

Beam structures

The deformation of a beam subjected to axial force and bending is described by


assuming that plane sections remain plane. This implies that the strain is given by Eq.
(12.54). The corresponding stresses need to be calculated in each layer (co-ordinate z
referred to the centroid (neutral axis)) at a longitudinal location, x, by using the
incremental expression: dσ = E t dε . In Section 12.4.2 a more accurate expression for
beams with thin-walled cross-sections will be given for the incremental stress-strain
relationship.

Even fo Elasto-plastic material behaviour plane section remain plane. Hence the strain
generally can be written as

ε = εm − z ⋅κ (12.89)

where εm and κ is the membrane strain and curvature, respectively. In Section 12.4.2
εm and κ have been described by their interpolation functions in terms of the co-
ordinate x. The volume integrals that express the stiffness matrices, Eqs. (12.64,
12.67), can be carried out as follows
~ ~ ~
∫ ε Dεdv = ∫ ∫ [ε m − z ⋅ κ ) D(ε m − zκ ) dA]dx (12.90)
v xA

For an elastic beam with D=E and z is defined with reference to the centroid the
integral (12.90) may be written as

~ ~ 2~
∫ ε DεdV = ∫ EAε m ε m dx + ∫ ( ∫ Ez κ κdA) dx
v x x A
(12.91 )
= ∫ EAε~m εdx + ∫ EIκ~κdx
x x

12.40
Chapter 12 - Nonlinear Analysis

This is because for instance ∫ zε~mκdA = 0 , when z is referred to the centroid.


A
For a beam with Elasto-plastic material behaviour the bending stress depends upon the
strain e.g. as indicated in Fig. 12.20. Eq. (12.90) then needs to be integrated
numerically and would include coupling terms of for instance the form:
~
∫ zD ( z )ε mκdV
v

σ
N.A. ∫ σ dA = 0 Formatted: Font: 10 pt
Formatted: English (U.S.)

ε
H
∫ σ ⋅ zdA = M Formatted: English (U.S.)
(no axial force) Formatted: English (U.S.)
elastic elasto-plastic
Formatted: Font: 10 pt
a) stress-strain curve b) stress distribution in beam section for Formatted: English (U.S.)
various level of bending moment (Note: the
Formatted: English (U.S.)
bending strain is ε b = − z ⋅ H , i.e. varying
Formatted: English (U.S.)
linearly over the cross-section) Formatted: English (U.S.)

Fig. 12.20 Stress distribution in a beam cross-section as a function of bending moment


(no axial force)

It should be noted that the integral (12.90) also would include coupling terms of z is
not defined with refrence to the centroid.

The integration over the cross-section must be carried out with check of the stress
level to decide whether:

D = E (elastic)
D = Et (elasto-plastic), Eqs. (12.80, 12.81)

and needs to be carried by numerical integration by either standard Gaussian or by


Lobatto integration (Appendix, Crisfield (1991)).

In the formulation described above the stress and strain is described over the height of
the beam- as a formulartion of z. Moreover, the relevant material properties in
different sections along each beam element need pursued. However, if the behaviour
of beams and frames under lateral (and axial) loading is observed, it is often seen that
failure occurs by the formation of plastic hinges. Simulated by this observation more
efficient methods can be developed, based on the assumption that plastic hinges
develop at predefined locations. Moreover, the elasto-plastic behaviour in a hing is
handled by using axial force, N and bending moment, M instead of pursuing the
stresses (and strains) in each fiber level of the corss-section.

Future comments on the use of the plastic hinges and generalized forces (M,N) are
provided in Section 12.6.2.

12.4.2 Generalization

12.41
Chapter 12 - Nonlinear Analysis

In the same manner as for one-dimensional case elasto-plastic behaviour of metals in


multidimensional stress state is characterized by

¾ An initial yield condition, i.e. the state of stress for which plastic deformation
first occurs.
¾ A hardening rule which describes the modification of the yield condition due
to strain hardening during plastic flow.
¾ A flow rule which allows the determination of plastic strain increments at each
point in the load history.

It is assumed that the material is isotropic, which implied that the stiffness properties
are independent of orientation at a point.

A review of elasto-plastic theory for multi-dimensional stress states based on isotropic


hardening is given in Appendix C. It is shown that the relationship between stress and
strain increments may be written as follows

dσ ij = Dijkl
ep
dε kl (12.92a)

where

ep
Dijkl = E ijkl − β sij s kl (12.92b)

2νG
Eijkl = 2Gε ije + δ ij ε kke (12.92c)
1 − 2ν

9G 2
β= (12.92d)
σ ( H '+3G )
2

σ = 3
2 sij sij = 3
2 σ ij σ ij − 12 (σ kk ) 2 (12.92e)

where the Einstein’s summation convention is applied i.e.


n n n n n
ep
Dijkl dε kl = ∑∑ Dijkl
ep
dε kl ; sij sij = ∑∑ sij2 ; σ kk = ∑σ kk ; σ m = 13 σ kk . (12.92f)
k =1 l =1 i =1 j =1 k =1

sij denotes the deviatoric, or reduced stress tensor

sij = σ ij − 13 δ ij σ kk (12.92g)

where δij is equal to 1 if i=j and equal to 0 if i±j.

It is seen that Eq. (12.92a ) may be simplified to the following form if one dimensional
plane stress condition is considered with dεxx ≠ 0 and dεyy = dεzz = 0 and ν= 0.3

12.42
Chapter 12 - Nonlinear Analysis

⎛ E(1 − ν) 9G 2 ⎞
dσ xx = ⎜ − ⎟ dε xx
⎝ (1 + ν)(1 − 2ν) (H '+ 3G) ⎠
(12.93)
⎛ 1.33E 2 ⎞
≈ ⎜1.33 − ⎟ dε xx
⎝ H '+ 1.16E ⎠

It is seen that Eq. (12.93) resembles the second of the expressions (12.81).

In dealing with thin-walled metal structures, a plane stress condition is more relevant.
The incremental stress-strain relationships for one- and two-dimensional conditions
can be obtained from Eq. (12.92a-d). The resulting expressions are shown in
Appendix C.

12.4.3 Cyclic Plasticity, Shakedown and Ratchetting

When some materials are subjected to uniaxial tension beyond yield, then unloaded
and reloaded in compression, it is found that the yield stress in compression is less
than the equivalent value in tension. This effect is called the Bauschinger effect and
occurs because of the permanent strains and residual stresses remaining after the first
yield point is reached. These residual stresses add to the reversed stresses in
compression loading thus lowering the second yield point. Figure 12.21(a) shows a
typical ‘hysteresis loop’ which forms when reversed loading is applied to a metal,
where the Bauschinger effect is exhibited by the fact that the yield stress, σ Y ( c ) in
compression is less than σ Y ( t ) .

In isotropic hardening it is assumed that during cyclic loading in which the load
changes from tensile to compressive, the yield point and the effects of work hardening
are the same in tension and compression (i.e. no Bauschinger effect), whereas in
kinematic hardening the yield point in compression is lower.

Referring to Figure 12.21(a), the first yield occurs at point A and then the material
hardens up to point B. Upon unloading from point B, the material follows a straight
line from B to C and the second yield occurs at point C. With continued load cycles
between fixed limits, a stable hysteresis loop may be observed.

Cyclic hardening or softening can also be observed in some metals. Under fully
reversed constant amplitude stress-controlled experiment, it is observed that the strain
amplitude either decreases or increases with each cycle. Similarly, under a strain-
controlled loading, the stress amplitude either decreases or increases with each cycle.
It is often convenient to represent this type of cyclic hardening or softening by a
‘backbone’ or ‘cyclic’ stress-strain curve, which is drawn through the tips of the stable
hysteresis loops, as shown in Figure 12.21(b).

The accumulation of plastic strains in cyclic loading is particularly important in ‘low


cycle fatigue’ problems, i.e. when the number of load cycles is usually less than
10,000 cycles. The accumulation of strains, which eventually leads to failure, is
usually caused by a mechanical or a thermal cyclic load, or a combination of both.

12.43
Chapter 12 - Nonlinear Analysis

σy(t)

σy(c)

(a) Hysteresis loop

(c) Cyclic stress-strain curve

(d) Elastic Shakedown

Figure 12.21: Cyclic and reversed loading

Three important phenomena can be observed in cyclic loading when the amplitude is
kept constant:

(i) Elastic Shakedown


This occurs when the plastic strain in the cycle is relatively small, i.e. the total
strain is less than twice the yield strain (the strain when the stress reaches the
yield stress). Referring to Figure 12.21(c), yield first occurs at point A and
strain hardening causes the stress to rise to point B. When unloading occurs,
the behaviour is linear elastic from B to C, with no further yielding.
Therefore, in subsequent cycles, provided the applied load does not go below
point C, the material behaves elastically, i.e. it moves up and down the line
BC with no further development of plastic strains. Thus the material has
‘shaken down’ to a stabilizing condition, i.e. the structure is assumed to have
‘settled down’ to an elastic state.

12.44
Chapter 12 - Nonlinear Analysis

(ii) Ratchetting
Depending on the load level, in some loading situations where a constant
amplitude of stress is imposed on the material a stable hysteresis loop may
not be reached. Instead, plastic strains keep on accumulating incrementally
with each cycle, leading to eventual failure. This mechanism is called
‘ratchetting’, also known as ‘cyclic creep’, and can occur due to a cyclic
thermal loading under a constant mechanical load. This occurs due to a cyclic
thermal loading under a constant mechanical load. This phenomenon is often
observed in materials which exhibit a difference in the yield stress between
load cycles of tension and compression, such as cast iron and most
composites.

(iii) Alternating plasticity


This phenomenon occurs in some cyclic load situation, where the behaviour
can settle down to a state where the plastic strains in ech cycle are equal and
opposite, and there is a progressive increase in total strain. It should be noted
that in real-life applications, alternating plasticity is of practical concern since
a limited amount of incremental material damage occurs in each cycle of
reversed plasticity. Such damage can be correlated to the equivalent plastic
strain.

12.5 Solution techniques

12.5.1 General

Characteristic features of static non-linear response

Two different types of structural non-linearities have primarily been described in this
chapter, namely geometrical and material non-linearities. A third non-linearity is
associated with boundary conditions, e.g. so-called contact problems. Characteristic
features of various types of nonlinear response are illustrated in Fig. 12.22.

The response diagrams illustrate three “monotonic” types of response: linear,


hardening, and softening. Symbols F and L identify failure and limit points,
respectively.

A response such as in (a) is characteristic of glass and certain high strength composite
materials.

A response such as in (b) is typical of cable, netted and pneumatic (inflatable)


structures, which may be collectively called tensile structures. The stiffening effect
comes from geometry “adaptation” to the applied loads. Some flat-plate assemblies
also display this behaviour initially.

A response such as in Fig. 12.22c is more common for structural materials than the
previous two. An almost linear regime is followed by a softening regime that may
occur slowly or suddenly. Alternative “softening flavours” are given in d – g)
12.45
Chapter 12 - Nonlinear Analysis

The diagrams, d – g illustrate a “combination of basic flavours” that can complicate


the response as well as the task of the analyst. Here B and T denote bifurcation and
turning points, respectively.

The snap-through response (d) combines softening with hardening following the
second limit point.

The response branch between the two limit points has a negative stiffness and is
therefore unstable. (If the structures is subject to a prescribed constant load, the
structure “takes off” dynamically when the first limit point is reached). A response of
this type is typical of slightly curved structures such as shallow arches or shells.

The snap-back response (e) is an exaggerated snap-through, in which the response


curve “turns back” in itself with the consequent appearance of turning points. The
equilibrium between the two turning points may be stable and consequently physically
realisable. This type of response is exhibited by trussed-dome, folded and thin-shell
structures in which “moving arch” effects occur following the first limit point; for
example cylindrical shells with free edges and supported by end diaphragms.

In all previous diagrams the response was a unique curve. The presence of bifurcation
(popularly known as “buckling” by structural engineers) points as in (f) and (g)
introduces more features. At such points more than one response path is possible. The
structure takes the path that is dynamically preferred (in the sense of having a lower
energy) over the others. Bifurcation points may occur in any sufficiently thin structure
that experiences compressive stresses.

Bifurcation, limit and turning points may occur in many combinations as illustrated in
(g). One striking example of such a complicated response is provided by thin
cylindrical shells under axial compression.

12.46
Chapter 12 - Nonlinear Analysis

(a) (b) (c) Formatted: English (U.S.)


Formatted: English (U.S.)
Example problem Formatted: English (U.S.)
N
N Formatted: English (U.S.)
N
ks

(d) (e) (f) (g) Formatted: English (U.S.)


Formatted: English (U.S.)
Figure 12.22 Characteristic features of nonlinear response: Formatted: English (U.S.)
Formatted: English (U.S.)
Basic response patterns: (a) Linear until brittle failure,
(b) Stiffening or hardening,
(c) Softening.

More complex response patterns:(d) snap-through,


(e) snap-back,
(f)&(g) bifurcation combined with limit points and snap-back

While in linear analysis the solution always is unique, this may no longer be the case
in non-linear problems. Thus the solution achieved may not necessarily be the
solution sought. For instance, for a given load R in the case shown in Fig. 12.22d)
there may be three different displacement states.

12.47
Chapter 12 - Nonlinear Analysis

Non-linear equations

The resultant of internal forces can be expressed as

i
R int = ∑ (a i ) T S
i

and the total equilibrium can be expressed as

R int = R

Hence, the equations that need to be solved are formulated in terms of a total and an
incremental equation of equilibrium

∑ (a )
i
i T
Si = R (12.94a)

K I (r ) dr = dR (12.94b)

For a given external load, R the displacement vector r is sought.

Various techniques for solving these non-linear problems exist. Herein three types of
methods will be briefly described, namely:

¾ incremental or stepwise procedures


¾ iterative procedures
¾ combined methods

The basic principles of these methods will first explained in Sections 12.5.2 – 12.5.4
with reference to problems where the R-r relationship is monotonously increasing with
r. In Section 12.5.5 a more general approach to deal with particular problems
associated with limit points, will be described.

Physical insight into the nature of the structural problem is essential in addition to
pure mathematical knowledge to obtain a successful method of solution. Often a
combination of different methods is used to achieve optimal efficiency/accuracy.

12.5.2 Load incremental methods

Incremental methods provide a solution of the non-linear problem by a stepwise


application of the external loading. For each step the displacement increment, Δr is
determined by Eq. (12.94b). The total displacement is obtained by adding
displacement increments. The incremental stiffness matrix, KI is calculated based on
the known displacement and stress condition before a new load increment is applied,
and is kept constant during the increment.

This method is also called Euler-Cauchy method. For load increment No. (m+1) is
may be expressed as
12.48
Chapter 12 - Nonlinear Analysis

ΔR m +1 = R m +1 − R m
Δrm +1 = K I (rm )−1 ΔR m +1 (12.95)
rm +1 = rm + Δrm +1

with the initial condition r0 = 0.

In this way the load may be incremented up to the desired level.

The method is illustrated for a single degree of freedom in Fig. 12.23.

Figure 12.23 Euler-Cauchy incrementing.

It is noted that the method does not include fulfillment of the total equilibrium
equation, Eq. (12.94a). For this reason total equilibrium will not be fulfilled. This is
illustrated by the deviation between approximate and true K(r)r = R in Fig. 12.23.

The accuracy may be increased by reducing the load increment. Also, the load
increment should be adjusted according to the degree of non-linearity. Computer
programs commonly include procedures for automatical choice of load increment.

Example 1 Incremental solution of two-bar problem in Section 12.1.2

In Section 12.1.2 the exact solution to a two-bar problem (for small h/A ratio) was
presented. Also, the incremental stiffness, KI was given by explicit formula Eq.
(12.9). In practice, analytical solutions can rarely be provided. Rather, numerical
solutions need to be used.

12.49
Chapter 12 - Nonlinear Analysis

The problem in Section 12.1.2 may be solved by an ”updated Lagrange”, corotational


formulation based on the element for the bar in Section 2.9.4, i.e. with a stiffness
relationship given by Eq. (2.103).

Eq. (2.103) is given for the element in Fig. 2.16 and needs to be transformed to a
single d.o.f., r2 refer to the global degrees of freedom, shown in Fig. 2.16. By
establishing the stiffness matrix k and introducing the boundary condition (zero
displacement) in the support nodes, a two d.o.f. system (r1 and r2) results. Due to
symmetry only r2 is needed to describe the problem in Fig. 12.2. The resulting
stiffness matrix for the symmetric system in Fig. 2.16 is

EA 2 P 2
R 2 = 2( s − c ) r2
A/c A/c

where P is the axial (compression) force in the member and


c = cos θ
s = sin θ

and θ is the angle between the element axis and the horizontal axis.

The trigonometric functions may be approximated by

θ3 θ5
sin θ = θ − − −
6 120
θ2 θ4
cos θ = 1 − + −
2 24
θ3 2θ5
tg θ = θ + + +
3 15

If it is assumed that θ2 is negligible relative to 1.0, the following approximations


should be retained when these functions are used in products and sums

sin θ = θ, cos θ = 1, tg θ = θ

However, when differences between trigonometric functions are calculated, two terms
in the expansions need to be retained. In the following, this will only apply to the
expression for cos θ.

For the system in Fig. 12.2 the angle θ = α is small and the incremental stiffness
relationship may be written as:

EA 2 P
ΔR = 2( α − ) Δr
A A

Global equilibrium is described by

R = 2 Ps = 2 Pα

12.50
Chapter 12 - Nonlinear Analysis

A solution can be obtained by the following incremental approach:

After computational cycle (j-1) the following information is available


j −1
Displacement: r j −1 = ∑ Δri

h − r j −1
Geometry: α j −1 ≅ α0
h
j −1
Axial force: Pj −1 = ∑ ΔRi /( 2 sin α j −1 )
i =1

Next step is calculated as follows:


EA 2 Pj−1
Displacement increment ΔR j = 2[ α j−1 − ]Δrj = K I Δrj
A A
Update total displacement rj = rj-1 + Δr j

h − rj
Updated geometry αj = α0
h
R j−1 + ΔR j
Updated axial force Pj =
2α j

The incrementation should be initiated by the geometry corresponding to α 0 and the


axial P0 = 0.
The load may be incremented by conveniently expressed as a fraction Δλ j of the
ultimate capacity of the present system, i.e.

3
ΔR j = EAα 03 Δλj
8

The load incremental factor Δλ is chosen as a number between 0 and 1.


The results are shown in Fig. 12.24. Ex.1 based on E= 210 Gpa, A = 0.0001 m2 and
α0 = 0.02. The load incremental factor Δλ = 0.5 and 0.1.

12.51
Chapter 12 - Nonlinear Analysis

Fig. 12.24 Load-displacement relations obtained by load incrementation.

Incrementation with equilibrium corrections

A simple improvement of the Euler-Cauchy method can be achieved by an equilibrium


correction. Consider the condition after the m’th step – the total load is Rm and
calculated displacement is rm. However, Eq. (12.94a) is not fulfilled. The unbalanced
or residual forces are then given by

R r = ∑ (a i ) T S i (rm ) − R m = R int (rm ) − R m (12.96)


i

The unbalanced forces may be accounted for by adding them to the next load
increment, when rm+1 is to be calculated. This means that the external loads are
reduced so that global equilibrium is restored. This principle of equilibrium correction
is illustrated in Fig. 12.25 for a single d.o.f. system.
Formally the method may be expressed as follows

ΔR m+1 = R m+1 - R m
R eq = R m - R int (rm )
Δrm+1 = K I (rm )-1 ΔR m+1 - K I (rm )-1 (R int (rm ) - R m ) = K I (rm )-1 ⎡⎣ΔR m+1 + R eq ⎤⎦
rm+1 = rm + Δrm+1
(12.97)

Figure12.25 Euler-Cauchy procedure with equilibrium correction.

The additional effort required in this modified Euler-Cauchy method consists in


calculating the internal force vector Rint(rm).

Euler-Cauchy’s method is based on a single point, rm by calculation of KI, and is


denoted one-step method. In principle, multi-step methods are envisaged, but in
practice it would not be optimal to go beyond a two-step method.

Example 1- continued
12.52
Chapter 12 - Nonlinear Analysis

Equilibrium corrections may be used in the Example 1 in the following manner:

The simple load incrementation procedure used in Example 1 may be improved by


equilibrium corrections.

The unbalanced force, Rr for the two-bar problem may be calculated according to Eq.
(12.96). The total load after increment j is then estimated by:

j
R j = ∑ ΔR j
i =1

Rint may be calculated as follows

Rint (r j ) = 2 Pj sin α j ≈ 2 Pj α j

when the displacement rj and axial force, Pj in each bar is calculated as follows:

j
rj = ∑ Δri ≅ A(α 0 − α j )
i

ΔA EA
Pj = EA ⋅ = (A / cos α 0 − A / cos α j )
A / cos α j A / cos α j
EA 2 EA rj rj
≈ (α 0 − α 2j ) = (2α 0 − ( ) 2 )
2 2 A A

Hence
rj rj rj
R int (rj ) = EA(2α 0 − ( ) 2 )(α 0 − )
A A A

Δri and ΔRi are incremental

ΔR i
rj = ∑ Δri = ∑
K I (ri −1 )

Since K I (ri −1 ) varies there is no simple explicit relationship between rj and R j = ∑ ΔRi

The unbalanced force after increment j is

⎛ rj r j ⎞⎛ rj ⎞
Rr =EA⎜⎜ 2α 0 − ( ) 2 ⎟⎟⎜⎜ α 0 − ⎟⎟ − R j
⎝ A A ⎠⎝ A⎠
where

j j
r j = ∑ Δri , R j = ∑ ΔRi
i =1 i =1

12.53
Chapter 12 - Nonlinear Analysis

The geometry and axial force, and hence, the incremental stiffness to be used to
calculate the next displacement increment, is the same as described in Example 1 in
Section 12.5.2.

Example 2 Incremental solution of elasto-plastic bar problem (after Cook et al., 1988)

The purpose is to calculate the load-displacement relationship for the bar in Fig.
12.18a, by using an incremental approach with small but not infinitesimal strains, so
that dε becomes Δε. A numerical representation of the stress-strain relation must be
stored, so that σ, E and Et can be obtained for any ε. The algorithm outlined below
requires that we also store, and update after each computational cycle, the nodal
displacements, r element strains ε, and element stresses σ. With two-d.o.f. bar
elements, σ and ε are constant over each element length L.

1. For the first computational cycle (j = 1), assume Eep = E for all elements. Apply
the first load increment, ΔR 1 .

2. Using the current strains, determine the current Eep in each element. Use Eq.
(12.82) to obtain kI for each element i.

Obtain the tangent stiffness matrix, K I ( j −1) = ∑ (a i ) T k it ( j −1) . Solve


K I ( j −1) Δr j = ΔR j for Δr j . From Δ rj obtain current strain increments Δε for each i
j

element (i).

3. Optional. If any elements make the elastic-to-plastic transition, use Eq. (12.83) to
revise Eep for each such element, and go back to step 2. Without changing the
applied load ΔR j repeat steps 2 and 3 until convergence, which may be defined as
Δε being less than a prescribed fraction of the accumulated total ε in every
element. These operations represent secant-stiffness iterations within one of the
load steps of the tangent-stiffness procedure.

4. Update: r j = r j −1 + Δr j
and for each element, ε ij = ε ij −1 + Δε ij
and σ ij = σ ij + Δσ ij where Δσ j = ( E ep ) j Δε j .

For the first cycle (j=1), initial values (subscript j – 1) of displacement, strain, and
stress are typically all zero if one starts from the unloaded configuration, but are
nonzero if one starts from a state in which plastic action impends.

5. Apply the next load increment and return to step 2.


6. Stop when ∑ ΔR j reaches the total applied load.

Three cycles of the foregoing algorithm are depicted in Fig. 12.18b. Each cycle
produces a line segment whose slope corresponds to the current stiffness. Drift from
the exact path can be reduced by using smaller load increments, by exercising step 3
previously discussed, and by using “corrective loads,” which are discussed below.
12.54
Chapter 12 - Nonlinear Analysis

Step 3 can be avoided by using load increments ΔR j that bring a single element to the
verge of yielding as each load increment is added. This is easily accomplished by
scaling the incremental tangent-stiffness solutions.

12.5.3 Iterative methods

The most frequently used iterative method for solving non-linear structural problems
is the Newton-Raphson method.

The Newton-Raphson algorithm to solve x for the problem: f(x) = 0 is

f (x n )
x n +1 = x n −
f '(x n )

where f '(x n ) is the derivative of f(x) with respect to x, at x = xn.

f ( x n ) / tgθ = f ( x n ) / f ' ( x n )

f(x)
f(xn)
θ
xn
f ( xn ) f ( xn )
=
tgθ f '( xn )
Fig. 12.26 Newton-Raphson algorithm

This approach can be generalized to solve Eq. (12.92a):

In this case KI(r) represents the generalisation of the ∂f / ∂x in Newton’s method for a
single d.o.f. See also discussion in Section 12.1.2.

Eq. (12.92a) is solved by the iteration formula

rn +1 − rn = Δrn +1 = K I−1 (rn ) (R − R int ) (12.98)


or
rn +1 = rn − K I−1 (rn ) (R int − R)

The basic principle for this iteration is illustrated in Fig. 12.27 for a single d.o.f.
system.

This method requires that KI is established and that Δrn+1 is solved from
12.55
Chapter 12 - Nonlinear Analysis

R − R int = K I(n) Δrn +1 (12.98a)

in each iterative step. This is time-consuming. By updating KI less frequently


reduced efforts are needed. Since this approach implies only a limited loss of rate of
convergence, such modified Newton-Raphson iteration is beneficial.

Fig. 12.28 illustrates two alternative for modified Newton-Raphson methods, one with
no updating of KI and one method where KI is updated after the first iteration.

Fig. 12.27 Newton-Raphson iteration.

a) No updating of KI b) KI updated after 1. iteration

Figure 12.28 Modified Newton-Raphson methods for single d.o.f.

The iteration is stopped when the accuracy is acceptable. The convergence criterion
may be based on the change of displacement from one iteration to the next. The
convergence criterion may be expressed by

|| rn +1 − rn || < ε (12.99)

where || ⋅ || is a vectornorm and ε is a small, positive number, say, of a magnitude, of


the order 10-2–10-4. The vectornorm is a measure of the size of the vector. There are
different vectornorms that may be applied. One alternative is the modified Euclidean
norm defined by:

12.56
Chapter 12 - Nonlinear Analysis

∑ (r )
1 N
2
|| r ||= k / rref (12.100 )
N k =1

where N is the number of components in the vector r and rref is a reference size, e.g.
max(ri )
.
N

12.5.4 Combined methods

Incremental and iterative methods are often combined.

The external load is applied in increments and in each increment equilibrium is


achieved by iteration. Fig. 12.29 illustrated a combination of Euler-Cauchy
incrementation and a modified Newton-Raphson iteration.

Figure 12.29 Combined incremental and iterative solution procedures

The procedure is carried out by applying loading according to Eq. (12.95) followed by
iteration at each load level by using Eq. 12.98). Commonly a modified Newton-
Raphson method is used keeping the gradient KI constant during several iteration
cycles. Iteration is stopped according to the criterion ( 12.99).

As long as the load curve are monotonically increasing with displacement, the
methods described are efficient. However, if there is an extremal point in the load-
displacement curves (as e.g. in Fig. 12.3) special procedures need to be adopted to
achieve acceptable results.

12.57
Chapter 12 - Nonlinear Analysis

12.5.5 Advanced solution procedures

General

The solution procedure described so far are a combination of incremental load coupled
with full or modified Newton-Raphson iterations. Because the plastic flow rules are
incremental in nature elasto-plastic problems should strictly be solved using small
incremental steps. For, no matter how accurately flow rules and keeping on the yield
surface may be satisfied within an increment, the solution is only in equilibrium at the
end of each increment after equilibrium iterations. However, often acceptable
solutions can be obtained with large steps.

Although incremental-iterative techniques provide the basis for most nonlinear finite
element computer programs, additional sophistications are required to produce
effective, robust solution algorithms. An extensive of more refined methods are
discussed e.g. in Chapter 9 of Crisfield (1991). In this section a brief review of such
methods is given.

For instance, severe difficulties are encountered when load incrementation methods
are used to pass a limit point, L (Fig. 12.22c),i.e. when the target stiffness becomes
zero. Using incrementation in displacement instead of load can solve this problem.
This approach will be effective for problems characterised by Fig. 12.22(c-d) for
which the load is uniquely determined by the displacement. However, displacement
incrementation will fail at turning points (T) (“snap-back point”) e.g. in Fig. 12.22e).

In the present section emphasis will be placed on arc-length techniques for solving
these problems. Prior to their introduction, analysts either used artificial springs,
switched from load to displacement control or abandoning equilibrium iteration in the
close vicinity of the limit point. In relation to structural analysis, the arc-length
method was originally introduced by Riks [1972] and Wempner [1971] with later
modifications being made by a number of authors.

Before describing such methods, one may ask why we need to pursue the response
beyond a limit point (L) in Fig. 12.22c). After all, the limit point represents the
ultimate strength. There are several reasons:

i) In many cases it may be important to know not just the collapse load, but
whether or not this collapse is of a “ductile” or “brittle”nature.

ii) The structure with the characteristic displayed in Fig. 12.22 may represent a
component in structure. The ultimate behaviour of a redundant structure
consisting of such components, would depend upon the post-ultimate beyond
limit point, L) behaviour of the component.

Method

As a starting point the global equilibrium equation is written as:


12.58
Chapter 12 - Nonlinear Analysis

g (r, λ ) = R int (r ) − λR ref = 0 (12.101 )

where Rref is a fixed external load vector and the scalar λ is a load level parameter.
Equation (12.101) defines a state of “proportional loading” in which the loading
pattern is kept fixed. Non-proportional loading will be briefly mentioned later in this
section.

The essence of the arc-length method is that the solution is viewed as the discovery of
a single equilibrium path in a space defined by the nodal variables, r and the loading
parameter, λ. Development of the solution requires a combined
¾ incremental (also called predictor)
¾ iterative (also called corrector)
approach.

Many of the materials (and possibly loadings) of interest will have path-dependent
response. For these reasons, it is essential to limit the increment size. The increment
size is limited by moving a given distance along the tangent line to the current solution
point and then searching for equilibrium in the plane that passes through the point thus
obtained and that is orthogonal to the same tangent line (Fig. 12.30c).

In Fig. 12.30c the arc-length control strategies in the solution of nonlinear equations
are illustrated and compared with load and displacement control. For instance if load
incrementation is applied, the iterations are carried out to correct the displacements.
When the arc-length method is applied the itereations are carried out with respect to
both the load and displavements.

Fig.12.30 Geometric representation of different control strategies of non-linear


solution methods for single d.o.f. An increment is made along a tangential path, SP.
Correction to reach g = Rint - λRref = 0 is obtained by iteration controlled by the
(hyper)plane c = 0
a) load control, b) state control, c) arc-length control

The arc length is formulated as an additional variable involving both the load and
displacement. The increment in the load-displacement space can be described by a
displacement vector Δr and a load increment parameter Δλ , such that ΔR = Δλ R ref .
This formulation results in an additional equation to be solved. The advantage of the
12.59
Chapter 12 - Nonlinear Analysis

extra equation is that the solution matrix never becomes ‘singular’ even at the limit
points. Therefore, the solution matrix is re-assembled with N+1 variables, where N is
the total number of the variables (degrees of freedom) of the system. However, the
disadvantage is that, in some FE formulations, the solution matrix becomes
unsymmetric, which may incur an increase in computing time and/or computer
storage, particularly for very large problems. First the increment (predictor) from the
“First point” is made along the tangent. Then, this solution is corrected iteratively to
reach the “Second point” and so on.

Several methods exist to obtain the arc length, for example by making the iteration
path follow a plane perpendicular to the tangent of the load-displacement curve, as
shown in Figure 12.31. Alternatively, instead of a normal plane, more sophisticated
paths such as spherical or cylindrical planes can be followed, and the solution matrix
can be manipulated to become symmetric (see, for example, Crisfield [1991]).

Load factor λ
iteration

increment

Figure 12.31: Schematic representation of the arc-length technique.

A geometrical interpretation of the incremental iterative approaches by Riks-Wempner


and Ramm is sketched in Fig. 12.32. While in Ramm’s method the iterative corrector
is orthogonal to the current tangential plane during the iteration, it is orthogonal to the
incremental vector ( Δr0 , Δλ 0 ) in the Riks-Wempner methods.

a) Riks-Wempner’s method b) Ramm’s method

Fig. 12.32 Arc-length control methods (Crisfield, 1991)


12.60
Chapter 12 - Nonlinear Analysis

An alternative iterative method is so-called orthogonal trajectory iterations (Fried,


1984). The first step in this method can be illustrated by reference to Fig. 12.30. The
first iteration is then assumed to be orthogonal to the vector S’P’ instead of SP.

The resulting iterative solution will appear as shown in Fig. 12.33.

Haugen (1994) found that this method was more efficient than the normal plane
iterations.

Figure 12.33 Arc-length method with orthogonal trajectory iterations.

Automatic incrementation

To achieve computational efficiency the load increment Δλ should be chosen


depending upon the degree of nonlinearity of the problem. Methods have been
established based on the curvature of the nonlinear path (den Heijer and Rheinboldt,
1981) or the so-called current stiffness parameter (Bergan et al, 1978):

Δr1T ΔR 1 Δλ i2
S pi = (12.102)
Δλ12 ΔriT ΔR i

S ip refers to increment No.i.


The initial value of S ip ( S 1p ) is 1.0. For stiffening system it will increase. For softening
system it will decrease. If S ip changes sign the sign of the increment should be
changed.

Numerical experiments show that nearly the same number of iterations were requested
to restore equilibrium when the increments were chosen according to the approach of
Bergan et al.( 1978).

12.61
Chapter 12 - Nonlinear Analysis

Ramm (1981) proposed another approach for estimating the necessary increment Δλ
(load incrementation) or Δλ (for arc-length method). The new arc-length, A n is
obtained by
1/ 2
⎛I ⎞
ΔA n = ΔA 0 ⎜⎜ d ⎟⎟ (12.103)
⎝ I0 ⎠
where ΔA 0 is the “old” arc-length, and Id and I0 are the desired number of iterations
(given as input) and the number of iterations when the old arc-length was used. This
approach requires a suitable estimate of the initial arc-length.

An alternative tactic is to apply load incrementation for early increments and switch to
arc-length control once a limit point is approached.

The current stiffness parameter can be used to decide the switch from load
incrementation (or displacement control) to the arc-length method. An alternative
indicator of when the limit point is approached is the check of negative values on the
diagonal of the incremental stiffness matrix, i.e. negative pivot elements in the
solution algorithm.

In particular the current stiffness parameter may be used to control the solution
strategy at limit points or bifurcation points. Alternative changes may be made when
the current stiffness is below a limit value, namely
- the sign of the incrementation is changed
- iteration may be suppressed and a simple incrementation may be used. Iterations
are then resumed when S ip increases beyond a specific limit (see Fig. 12.34).

Fig. 12.34 Possible choice of solution algorithm for a problem with limit point

Non-proportional loading

The solution procedures in this chapter have been based on the equilibrium
relationship of (12.101) which implies a single loading (or displacing) vector, Rref, is
proportionally scaled via λ. For many practical structural problems, this loading
12.62
Chapter 12 - Nonlinear Analysis

regime is too restrictive. For example, we often wish to apply the “dead load” or
“self-weight” and then monotonically increase the environmental load. Even more
general load conditions may be required. Fortunately, many such loading regimes can
be applied by means of a series of loading sequences involving two loading vectors,
one that will be scaled (the previous Rref) and one that will be fixed ( ( R ref ) . The
external loading can then be represented by

R = R ref + λR ref (12.104)

so that the out-of-balance force vector becomes

g = R int − R ref − λR ref (12.105)

12.5.6 Direct integration methods

General

Up to now the methods for directly solving the statistic nonlinear equation have been
based on incrementation of loads or displacements. Possibly combined with iterative
methods. These are often considered standard methods for solving nonlinear problems
(e.g. in ABAQUS).

An alternative approach is to use so-called finite difference methods for direct


integration of the dynamic equation of motion :

Mr (t) + Cr (t) + Kr (t) = R (t) (12.106)

to solve the static problem : Kr = R . Nonlinear structural effects make K a function


of r, K (r ) .This means that the loading R is increased (artificially) or as a function of
time. The loading time needs to be sufficiently long so that the inertia and demping
forces do not have an effect on the behaviour on the static problem that is to be solved.

A finite difference approximation is used when the time derivatives of (12.106)


(
r and r ) are replaced by differences of displacement (r) at various instants of time.
The direct integration methods are alternatives to modal methods, and they can be
used to successfully treat both geometric and material non-linearities.

The finite difference methods are called explicit if the displacements at the new time
step, t + Δt , can be obtained by the displacements, velocities and accelerations of
previous time steps.

r (t + Δt ) = f {r (t ), r (t ),  r (t − Δt ),...}
r (t ), r (t − Δt ), r (t − Δt ),  (12.107)

or

ri +1 = f {ri , ri ,  ri −1 ,...}


ri , ri −1 , ri −1 , 

12.63
Chapter 12 - Nonlinear Analysis

This is as opposed to the implicit finite difference formulations where the


displacements at the new time step, t + Δt , are expressed by the velocities and
accelerations at the new time step, in addition to the historical information at previous
time steps.

ri +1 = f {ri +1 ,  ri ,...}
ri +1 , ri , ri ,  (12.108)

Many of the implicit methods are unconditionally stable and the restrictions on the
time step size are only due to requirements of accuracy. Explicit methods, on the other
hand, are only stable for very short time steps.

Central difference method

To illustrate this approach, one of the explicit solution methods, the central difference
method is described in the following. The central difference method is based on the
assumption that the displacements at the new time step, t + Δt , and the previous time
step, t − Δt ,can be found by Taylor series expansion.

Δt 2 Δt 3
ri +1 = r0 (t ) + Δt ri + ri +
 ri + ...
 (with r0 (t ) = ri ) (12.109)
2 6

Δt 2 Δt 3
ri −1 = ri − Δt ri + ri −
 ri + ...
 (12.110)
2 6

The terms with time steps to the power of three and higher are neglected. Subtracting
Eq. (12.110) for Eq. (12.109) yields :

ri +1 − ri −1 = 2Δt ri (12.111)

Adding Eq. (12.109-110) yields :

ri +1 + ri −1 = 2r + Δt 2ri (12.112)

Rearranging Eq. (12.111-112), the velocities and accelerations at the current time step
can be expressed as:

1
ri = {ri +1 − ri −1} (12.113)
2Δt

1
ri =
 {ri +1 − 2ri (t ) + ri −1} (12.114)
Δt 2

Finally inserting Eqs. (12.113-114) into the dynamic equation of motion Eq. (12.106)
gives:

12.64
Chapter 12 - Nonlinear Analysis

⎧ 1 1 ⎫ 1 1
⎨ 2 M+ C ⎬ ri +1 = R i (t ) − K ri (t ) + 2 M {2ri − ri −1} + C ri −1
⎩ Δt 2 Δt ⎭ Δt 2 Δt
(12.115)

If the mass matrix, M, and the damping matrix, C, are diagonal, the equations will be
uncoupled, and the displacements at the next time step, t + Δt , can be optained
without solving simultaneous equations.

The characteristic features of Eq. (12.115) are best illustrated by an example. Let us
consider a system with three global directions of freedom. The mass matrix, M and
damping matrix Care assumed to be diagonal.

Eq. (12.115) may then be written as:

⎧ ⎡ M 11 0 0 ⎤ ⎡C11 0 0 ⎤ ⎫ ⎡ r1(i +1) ⎤ ⎡ R1(i ) ⎤ ⎡ K11 K12 K13 ⎤ ⎡ r1( i ) ⎤


⎪ 1 ⎢ 0 M ⎥ 1 ⎢ ⎪⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎨ 2 ⎢ 22 0 ⎥+ ⎢ 0 C22 0 ⎥⎥ ⎬ ⎢ r2(i +1) ⎥ = ⎢ R2( i ) ⎥ − ⎢⎢ K 21 K 22 K 32 ⎥⎥ ⎢ r2( i ) ⎥
⎪ Δt ⎢⎣ 0 0 M 33 ⎥⎦
2Δt
⎢⎣ 0 0 C33 ⎥⎦ ⎪⎭ ⎢⎣ r3( i +1) ⎥⎦ ⎢⎣ R3(i ) ⎥⎦ ⎢⎣ K 31 K 32 K 33 ⎥⎦ ⎢⎣ r3(i ) ⎥⎦

⎡ M 11 0 0 ⎤ ⎧ ⎡ r1( i ) ⎤ ⎡ r1(i −1) ⎤ ⎫ ⎡C11 0 0 ⎤ ⎡ r1(i −1) ⎤
1 ⎢ ⎥ ⎪ ⎢ ⎥ ⎢ ⎥⎪ 1 ⎢ ⎢ ⎥
+ 2 ⎢ 0 M 22 0 ⎥ ⎨ 2 ⎢ r2( i ) ⎥ − ⎢ r2( i −1) ⎥ ⎬ + ⎢ 0 C22 0 ⎥⎥ ⎢ r2( i −1) ⎥
Δt 2Δt
⎢⎣ 0 0 M 33 ⎥⎦ ⎪⎩ ⎢⎣ r3(i ) ⎥⎦ ⎢⎣ r3(i −1) ⎥⎦ ⎪⎭ ⎢⎣ 0 0 C33 ⎥⎦ ⎢⎣ r3(i −1) ⎥⎦
(12.116)

The first equation in Eq. (12.121) is explicitty written as :

⎧ 1 1 ⎫
⎨ 2 M 11 + C11 ⎬ r1( i +1) = R1 (t ) − K11r1( i ) − K12 r2( i ) − K13 r3( i )
⎩ Δ t 2 Δ t ⎭ (12.117)
1 1
+ 2 M 11 {2r1(i ) − r1(i −1) } + C11r1(i −1)
Δt 2 Δt

This shows that ri ( i +1) can be directly, explicity determined by the response at time t.
There is no coupling between displacements, rj (i +1) at the time t + Δt .

Because the expressions for the displacements are explicitly given, there is no need to
invert the tangent stiffness matrix at every time step. The explicit method also has the
advantage of drastically reducing the need for computer memory capacity. The
stiffness forces, or internal force vector, can be found by summation of element
contributions. The global stiffness vector, K, need not to be stored in the computers
core memory.

As already mentioned, Eq. (12.115) is conditionally stable and requires that

2
Δt < (12.118)
ωmax

12.65
Chapter 12 - Nonlinear Analysis

where ωmax is the highest natural frequency of

det(K − ω 2 M ) = 0 (12.119)

The maximum frequency of Eq. (12.119) is bounded by the maximum frequency of


the constituent unassembled and unsupported elements. When finding the maximum
natural frequency of an element, one will see that the time step, Δt , must be short
enough that information does not propagate across more than one element per time
step. The maximum allowable time step will therefore be limited by a characteristic
length, λe , of the element and the acoustic wave speed, c.

λe
Δt < (12.120)
c

Higher order elements yield higher maximum frequencies and should be avoided when
doing explicit integration. Many alternative methods exist. See. e.g.

Solution of static problems

The explicitmethod is very well suited to treat dynamic problems. As indicated above
the method can also be used to solve static problems.

It is obvious that the periode of the loading, or the amount of time for the loading to
reach it’s maximum value, must be much larger than the largest eigenperiod to avoid
dynamic effects as determined from the lowest eigenfrequency found for Eq. (12.119).
The response of the structure is also dependent on the magnitude of the loading, not
only on the period, and this complicates the picture. In addition, failures due to
collapse or cracking of parts of the structure will cause vibrations. These events will
not be captured by a traditional static analysis.

All effects taken into account; if the time of the loading to reach its maximum level is
conservatively chosen to be 30 times the longest eigenperiod of the system, the
dynamic effects have shown to be negligible.

Another problem with explicit analyses is that post-collapse behaviour cannot be


traced if the loading is given as applied forces. In many cases this can be avoided by
switching to displacement control. If displacement control is not possible or desirable,
implicit solution procedures using arc length solution methods can be used.

An advantage with the explicit solution procedure is that it is very easy to use. The
user of an explicit finite element program is left with the difficulty of applying loads
sufficiently slowly to avoid dynamic phenomena and sufficiently fast to avoid too
large computional efforts times.

In static analyses, and even in dynamic analyses, the computational time van be
considerably reduced by changing mass densities in elements. The time step will be
governed by the smallest element in the model. Artificially increasing the mass of
small elements will reduce the acoustic wave speed and hence allow longer time steps.
Similarly very large elements can be given mass reduction and hence be less affected
by inertia forces. Systematic increase and reduction of element masses can be
12.66
Chapter 12 - Nonlinear Analysis

performed to improve computational efficiency, but the details in these methods will
not be elaborated on.

12.6 Applications

12.6.1 General

In practical structural analysis it is good practice that a nonlinear analysis is preceeded


by a linear analysis, that provides a basis for planning the nonlinear analysis.

The linear analysis may be e.g. a stress analysis or buckling analysis.

The stress analysis provides information about which nonlinearities which may be
important, i.e. where stresses exceed yield limit and implies that elasto-plastic
behaviour should be considered.

In case of compressive loading, buckling may be an important issue. A linear


buckling analysis based on elastic and geometric stiffness matrices, then provides
buckling loads and buckling modes. Knowledge about the elastic buckling load
(“Euler load”) can, together with code formulations for ultimate strength yield
information about the ultimate capacity to be expected. The buckling modes are
crucial for the introduction of initial imperfections for the nonlinear analysis. Usually,
a linear combination of linear buckling modes are used to describe the imperfection
pattern, and the maximum value is scaled to some given imperfection tolerance level.

12.6.2 Beams and frames

General computer programs

General purpose finite element programs for nonlinear analysis, such as ABAQUS,
LSDYNA etc. contain several options for beam models.

Special purpose programs

The program system USFOS was initially developed at NTNU/SINTEF to deal with
nonlinear analysis of frames and trussworks. Later it has been extended also to cover
plates and shells. The basic for USFOS is summarized in this section. More details
may e.g. be found in Skallerud and Amdahl (2002).

USFOS may be used to assess the ultimate global capacity of space frame structures
and to document the residual strength of such structures. The formulation of beam
models allows the use of very coarse finite element modelling of the structure, but still
obtain results good accuracy. Local flexibility of tubular joint is included through a
simplified, but very efficient formulation. The formulation gives very good results
compared with shell analysis of the joint, but requires no special modelling of the joint

12.67
Chapter 12 - Nonlinear Analysis

geometry. The flexibility characteristics are calculated form the information already
present in the design model of the structure.

The program is based on an updated Lagrange (incremental-iterative) procedure. It


uses a (nonlinear) Green strain formulation (Section 12.3.3) . Thus, the USFOS beam
element is valid for large displacements, but restricted to moderate strains. The
influence of axial force on the bending stiffness of the element is introduced by the
nonlinear terms in the Green strain formulation. The tangent and secant stiffness
matrices are then obtain by introducing interpolation functions for the element
displacements. The shape functions in USFOS are taken as the exact solution of the
4th order differential equation for a beam column, i.e. the stability or Livesly functions
(Section 12.2 and Appendix A). With these shape functions all integration in the
element stiffness expressions is carried out analytically, giving closed form solutions
for the nonlinear elastic stiffness matrix.

In addition to accounting for the coupling between axial displacements and lateral
rotations on the element level, large displacement effects are also taken into account
by updating the local reference co-ordinate system at each step.

Material nonlinearities are modelled by plastic hinges at element ends or at element


midspan. The control of element plastification is a control of stress resultants (forces
and moments) against the total plastic capacity of the cross section, instead of local
stresses. Possible unloading of plastic hinges into the elastic range is checked at each
load step. Elasto-plastic element properties are established based on the flow theory.
The plasticity formulation is based on interaction formulae for plastic capacity of
stress resultants.

Fig. 12.35 USFOS formulation: One finite element per structural element.

12.68
Chapter 12 - Nonlinear Analysis

Hinges may occur at beam-ends and at midspan. In the latter case the original element
is subdivided into two sub elements. The extra nodal point is introduced automatically
and eliminated by static condensation before adding into the global stiffness matrix.

The difference between yield hinge formulations have been investigated (Hellan et al.
(1994 -1995), and show significant differences. Elastic-perfectly plastic models over-
predict the column buckling capacity of ideally straight tubulars. For yield-hinge
models incorporating first fiber yield, gradual plastification and strain hardening, the
buckling capacity depends on the plasticity parameters given to model the transition
from elastic to plastic behaviour. In general, these formulations may be slightly
conservative for stocky columns, and slightly un-conservative for slender columns.
For both these formulations, geometrical imperfections and residual stresses for a
member can be accounted for by introducing equivalent initial imperfections in the
element formulation to ensure exact fit to any given column curve, e.g. Eurocode 3,
which is based on extensive test results. However, particular assessment of the
imperfection mode for systems analysis is required. By comparing various
imperfection patterns, it was found that imperfection in the direction of the global load
yields the lowest capacity. This is discussed in more detail in Hellan et al. (1994).

Joints can in principle be modeled by shell elements. Systematic FE studies of linearly


elastic joints have resulted in parametric formula for the joint flexibility (Ultiguide,
1999).

Under extreme loads, the nonlinear deformations of the joint and failure characteristics
can influence the disposition of forces and the overall structural response. Failure of
tubular joints generally involves some combination of the following local and global
modes :

¾ Local plastic deformation (yield) of the chord around the brace intersection
¾ Cracking in the chord at the weld toe (and propagation to severance)
¾ Local buckling in compression areas of the chord
¾ Ovalisation of the chord cross-section
¾ Beam shear failure across a gap K joint chord
¾ Beam bending of the chord especially for T/Y action

The specific response depends upon the type of joint (T/Y, X, K; simple, stiffened,
grouted; etc), the loading (Axial – tension/compression; bending – in-plane
bending/out-of-plane bending, etc) and the joint geometry parameters ( β , γ , etc ).

Fig.12.36b illustrates typical load-deformation responses for axially loaded joints as


seen in isolated component tests performed with idealized boundary conditions.

12.69
Chapter 12 - Nonlinear Analysis

a) member behaviour b) Joint behaviour c) Joint model

Figure 12.36: Typical behaviour of members and joints in jacket platforms

Several methods are in use to model joint behaviour in non-linear analyses as


summarized in Ultiguide (1999). One method is to model the joint behaviour by a non-
linear spring between the brace end and the center of the chord. Spring properties can
then determined from test data, FE analyses or parametric formulae. Alternatively, the
springs can be replaced by a beam-column element with the ultimate capacities
determined from the capacity equations for the joint (Fig. 12.36c) given in appropriate
codes. The joint behaviour may be modeled by a plastic potential, with interaction
between the axial force, in-plane bending and out-of-plane bending. It is, however, an
open question whether the interaction in the post-ultimate range is adequately
described in this way. Formulations have also been published that accounts for brace-
to-brace interaction by adding ‘beam’ elements between the brace ends.

It is important that the nonlinear analysis method can predict failure in accordance
with recognized failure criteria. Hence, besides modeling of stiffness and capacity,
yield characteristics, post ultimate behaviour and ductility limits (if applicable) should
be represented, including local failure modes such as local denting, local buckling,
joint overload, joint fracture, etc. Local buckling may be accounted for by various
approaches. One alternative is to modify the stress-strain relationship by a drop in the
stress when the critical strain εcrit, is reached (Fig. 12.37a). Another alternative is to
simulate local buckling by using a model for a dented tube (Taby and Moan, 1985).
This is done by assuming an initial dent and accounting for the growth of the buckle as
the member deforms (Fig. 12.37b). The same model is applied in the analysis of
structures with dents imposed by e.g. accidental loads. General shell FEM
formulations may be able to capture local buckling through the shell modeling, but
initial geometric imperfections are generally needed to initiate the local buckle
correctly. Obviously, a global analysis based on a model which accommodates local
buckling modes by shell elements will be very time consuming.

a) Equivalent stress-strain b) Dented member model

Figures 12.37: Models for damaged members


12.70
Chapter 12 - Nonlinear Analysis

An automatic load algorithm is implemented. At each step the occurrence of new


plastic hinges is examined. The step length is scaled to make the stress resultants
satisfy the yield criterion exactly.

Equilibrium iterations may be performed including correction to the yield surface due
to drift-off of stress resultants during finite increments.

Possible unloading of plastic hinges into the elastic range is checked during the load
incrimination step. If unloading occurs, the element stiffness is modified and the
system stiffness recalculated before the step is completed.

For a detailed description of theory and implementation, reference is made to the


theory manual of the program (Søreide et al., 1994). Moan et al (1985) compared
results obtained by the USFOS formulations with those obtained by “conventional” a
FE formulation (Engseth, 1986).

Ultimate limit state checks of structures have traditionally been carried out by linear
global analyses to calculate member forces; and only include inelastic and second
order effects in the determination of the ultimate capacity of the individual
components. The strength of materials formulations used in codes for component
limit states have been extensively validated against test results. This methodology
hence, focuses on the first failure of a structural component and not the overall
collapse of the structure, which obviously is of main concern in view of the failure
consequences. Tools for nonlinear analysis that includes second order geometrical and
plasticity effects, provides a means to account for possible re-distribution of the forces
and subsequent component failures until system collapse, and, hence, a more realistic
estimate of the ultimate capacity.

Initially, such methods were developed for calculating the residual strength of systems
with damage (progressive limit state checks of NPD). More recently, such methods
are also applied for re-assessment of ageing structures to determine the consequences
of fatigue induced fracture of members in connection with inspection planning.

Examples are presented in dr.ing.thesis by Ø. Hellan (1995) and Hellan et al (1994).

Fixed platform analyses are carried out by modeling the pile-soil behaviour by
equivalent linear or nonlinear concentrated springs or, distributed springs along the
piles, or continuum (finite element) model that represent stiffness and foundation,
capacity, appropriately using the material properties in the different soil layers. Most
of the existing pile-soil models are empirical or semi-empirical based on the
acquisition of data from a limited number of large diameter pile tests from early
1970’s until now. Although these empirical models have provided practical tools for
the designer of the offshore pile-soil foundation, they are associated with some
uncertainties, see e.g. Horsnell and Toolan (1996) and Nadim and Dahlberg (1996).

Soils exhibit nonlinear behaviour, even at low load levels, which needs to be
accounted for. The response calculations are carried out as integrated pile-soil-
structure analyses. A typical finite element model of the platform in Fig. 12.38a is
shown in Fig. 12.38c.
12.71
Chapter 12 - Nonlinear Analysis

(261)

(363)

(463)
(455)
(456)

a) b)

c) d)

Figure 12.38: Global ultimate behaviour of a North Sea jacket

The effects of global seabed scour and local scour in granular soils, and the partial loss
of soil-pile contact in cohesive soils should be accounted for. When modeling the
individual piles in a pile group, nonlinear soil P-y and T-z curves have to be adjusted
to account for pile group effects.

Cyclic loads cause deterioration of the lateral bearing capacity as indicated in Fig.
12.38. The soil capacity and the nonlinear P-δ characteristics given in most codes
represent the fully degraded properties of the soil, based on the cyclic (hysteretic)
behaviour. The capacity under monotonic, static loading can be significantly higher, as
shown in Fig. 12.38. High loading rates (compared to laboratory test loading) can also
contribute to increase the capacity up to 40% for lateral loading and 50% for axial
loading for wave loading. It is noted that this strength increase only applies to the
dynamic (variable) part of the soil reaction.

Analysis strategy and solution procedure

To facilitate the interpretation and verification of ultimate strength predictions a


systematic approach to the non-linear analysis is necessary. For each scenario a
12.72
Chapter 12 - Nonlinear Analysis

sequence of analyses of increasing complexity and increasing realism should be


performed, beginning with the structure alone; then introducing, for example, soil-
structure interaction and non-linear joint characteristics in turn, as appropriate. Also, a
static analysis is done before a dynamic one, and a monotonous analysis before a
cyclic one. At each stage the results should be examined and the influence of the
modeling changes examined.

Various options for solution of the non-linear problem are used. For static problems an
incremental iterative, proportional loading algorithm is applied. Load increment size is
adjusted when plastic hinges are formed, predefined maximum displacement
increment is exceeded and when load limit points are passed. In the latter case arc
length iterations is used as one option. Even though automatic solution controls are
generally available, manual intervention may be required. Sudden changes in
stiffness, rapid unloading and alternative equilibrium conditions in the non-linear
region may cause numerical instability and present challenging problems for all
analysis software; small step sizes may be required to coax the solution.

Apparent solution difficulties should be investigated in terms of the non-linear events


being predicted. Failure of an analysis does not necessarily mean that the ultimate
strength of the system has been reached. Solution should be continued until a clearly
defined peak or sustained limit load has been reached and the post-ultimate response
characteristics determined.

Case studies

As demonstrated e.g. by Moan, Azadi and Hellan (1997) the choice of pile-soil model
can affect the load distribution in the structures and, hence, the failure mode and
corresponding ultimate strength. The most important issue is, of course, that a pure
linear pile-soil model would not represent a possible soil failure and, hence
overestimate the system strength if the pile-soil is the critical part of the system. For
the jacket in Fig. 12.37a with plugged piles the pile-foundation is not critical. Yet the
difference in jacket failure mode when using a linear instead of a nonlinear model,
results in an ultimate load which is about 15% smaller for the former case.

Determination of the global ultimate capacity by monotonically increasing wave


loading (pushover analysis) has become a well-established approach.

As mentioned above, systems analysis is particularly of interest to demonstrate


robustness in connection with structures that are damaged due to accidental loads of
fatigue fracture. It is found that the nonlinear approach especially yields more realistic
predictions of the ultimate strength of a damaged system than a conventional approach
based on a linear global analysis and an ultimate strength assessment of the structural
component.

To illustrate the effect of damage the jacket in Fig. 12.38 is considered. The ultimate
capacity is normalized with respect to F100. For this case a linear spring model is
applied for the pile-soil. Damage in terms of removal of individual braces is
considered as indicated in Fig. 12.38b. Table 12.2 shows that failure of the braces 261
and 463 for broad side loading does not reduce the ultimate strength, and, most
importantly, for all cases with a single brace failure the reserve capacity is at least

12.73
Chapter 12 - Nonlinear Analysis

0.7 x 2.73 = 2.08 times the 100 year characteristic load, while the normal total safety
factor for design checks of components of offshore structures is about 1.5.

Table 12.2: Residual strength of damaged North Sea jacket. Linear pile-soil
behaviour. Wave height incremenation
Loading and damage condition

Broad side loading End-on loading


Brace 261 263 Brace Brace 463 Brace 455 Brace 456

Ultimate strength
Fult / FH100 2.73 2.73 2.73 2.89 2.89
Residual strength
Fult (d) / Fult 1.0 0.76 1.0 0.91 0.85
Note: Fult – ultimate strength, Fult (d) – ultimate strength of damaged platform

Conclusions

For a realistic evaluation of the structural safety of offshore structures with complex
three-dimensional and stochastic sea loads and complex structural geometries, the
global ultimate strength needs to be calculated, yet, considering relevant local failure
modes of members and joints. The platforms have to be analyzed in intact and damage
condition. The efficiency of the calculation methods is then detrimental, especially in a
design context, when repeated analyses are required. Members are modeled as beam
elements and joints by spring elements. It is shown how local failure modes, especially
local buckling of members and joints, are accommodated by a phenomenological
approach.

The sensitivity of the global ultimate strength to damage of individual members is


examined.

12.6.3 Plane stress and bending of plates and shells

Formulations

The one-dimensional formulations in terms of strain, Exx, stress, Sxx or their


incremental form, need to be generalized for multidimensional problems. The basis
for this generalization is given in appendices. To a large extent this generalization
corresponds to the generalization of the linear theory from one – to e.g. two-
dimensional problems. For instance, the expressions for strains and stresses when
geometrical nonlinearities are considered, can be established by introducing
“correction terms” on the linear strain expression in the same way as for Exx as
discussed in Section 12.3.3.

However, the generalization of elasto-plastic theory from one-to e.g. two-dimensional


problems include the following issues:
¾ yield criterion (von Mises)
¾ hardening rule
¾ flow rule

12.74
Chapter 12 - Nonlinear Analysis

The von Mises yield criterion is well known. The generalization of isotropic hardening
rule is relatively straight-forward (see Appendix C), other hardening rules are more
complex and are not described in the Appendix C.

The generalization of the flow rule is the most important issue in the generalization.
The main starting point is Eq. (C.15) which follows from Drucker’s postulate. Eqs.
(C.19) shows that the plastic strain increment is proportional to the equivalent stress
increment, dσ and the deviatoric stress, sij. Eq. (C.30) expresses the stress increment
as a function of total strain increment. It is this relationship that is used in multi-
dimensional elasto-plastic finite element formulations.

Elasto-plastic analysis of stresses and strains in a corner of a large frame composed


of box members (Moan and Nordsve, 1979)

The purpose of this example is to study the behaviour of a corner in a large frame (Fig.
12.38a). The analysis has to relevance to serviceability and ultimate limit state
dimensioning criteria.

The structure analyzed is a plane idealization of the structure shown in Fig. 12.39c.
Both an unstiffed configuration (No. I) and a stiffened one (No. II) are considered. In
configuration No. II only the effective part of traversal panels and flanges of the actual
panel are included, see Figs. 12.39c-d. Buckling stiffeners are smeared and contribute
to the effective plate thickness. The plate material is assumed to be ideally elasto-
plastic with yield stress σ Y = 360N/mm 2 . The loads are introduced according to
simple beam theory on the boundary AA, see Fig. 12.39d. The level of the load is
expressed by the nominal stress level according to beam theory in section BB. The
external moment and shear forces at AA are such that the maximum equivalent stress
in the outer fiber in BB is η ⋅ σ Y . The average shear stress in BB is 0.18 η ⋅ σ Y and the
average normal stress in CC 0.58 η ⋅ σ Y . η is a usage factor. Two load conditions are
considered, namely η = 0.63 (Condition A) and η = 0.84 (Condition B). These usage
factors correspond to typical values allowed for the operational and extreme load
condition for marine steel structures.

12.75
Chapter 12 - Nonlinear Analysis

a) Gravity platform b) Deck structure and columns

c)Corner of large frame d) Plane stress symmetric model.


Configuration I: unstiffed
Configuration II: stiffened

e) Mesh of constant strain elements and (possible) bar elements, and uniaxial model of
material.

Fig. 12.39. Idealization of a corner in a large frame composed of box members.

12.76
Chapter 12 - Nonlinear Analysis

The finite element solution is obtained with an idealization of the panel based on 288
constant strain plane stress elements. Out-of-plane displacements are not considered.
Stiffeners are represented by bar elements, totaling 77. The resulting mesh is shown in
Fig. 12.39e. Because the nonlinear behaviour is very local in the present case, it is
computationally advantageous to use the so-called substructure approach. The two
substructures used, denoted by No.1 and 2, are shown in Fig. 12.39e.

A selection of the results obtained in this study is displayed in Fig. 12.40 and Table
12.3.

Fig. 12.40 shows how the yield zone develops in the two configurations. It is observed
that the initial extent of the yield zone for Configuration I can be estimated fairly well
by using the yield condition in conjunction with stresses calculated according to the
elasticity theory. However, the gradual local softening of the structure due to
plastification influences the further development of the yield zone.

Fig. 12.40. Yield zones in substructure No.2, see Fig. 12.39e.

In a plastic zone, strain is obviously a better measure of strength than stress. Table
12.3 displays the maximum strains found. For both configurations considered the
maximum strain is far below fracture strain, which may be of the order of 10ε Y . This
is because the plastic flow is contained in a fairly large elastic medium. The location
HS1 and HS2 are the most highly strained zones in Configuration I and II,
respectively. In particular the shear straining at HS2 in Configuration II should be
noted.

The residual equivalent stress when unloading from maximum loading is displayed in
Table 12.3.

12.77
Chapter 12 - Nonlinear Analysis

Table 12.3: Characteristic stress and strain responses in corner of a large frame.

Elasto-plastic analysis
Elastic analysis referring to max. loading (η = 0.84 )
Elastic stress Max. equivalent Max. residual equivalent stress
Configuration concentration plastic strain: after unloading :
factor: ε max
p
σ max

σ max / σ ref ε max


p
/ εY σ max

/σY
I 1.91 1.6 0.62
II 1.78 2.1 0.52

σ max maximum pointwise equivalent stress


σ ref = ησ Y maximum equivalent stress in Section BB, see Fig. 12.39d.
σ ∗
max maximum pointwise equivalent residual stress.
σY yield stress
εY strain in uniaxial specimen at the yield point.
ε p
max maximum pointwise accumulated equivalent plastic strain.

The above case study may serve as a basis for a discussion of the adequacy of linear
elastic stress analysis and possible other, improved methods of analysis to represent
actual limit states for a steel structure.

The traditional way of assessing the strength of a structure is by ensuring that the
stresses obtained in a linear elastic analysis are less than the yield stress. Commonly
the analysis is performed by means of a beam or a frame model, which represents the
gross stress variation. If the stress analysis is based on a finite element method, the
very local stresses can be calculated. Therefore, due to the high elastic stress
concentration present in most steel structures, the stresses obtained in a linear elastic
analysis may exceed the yield level even or operational loads. In the above case study
the maximum equivalent stress for Configuration I and II at Load Level A ( η = 0.63 )
is, respectively, 1.12σ Y and 1.20σ Y . This fact does not necessary imply an inadequate
design. This is because ductile collapse strength depends on the average stress over a
region and not the point wise maximum value – given that sufficient ductility allows a
redistribution of stresses to comply with the stress distribution1 at ultimate collapse.
However, fracture strength under single overload or repeated large loads (low cycle
fatigue) depends on the local response. Even in that case the stresses obtained in a
linear analysis may not be particularly relevant as a measure of strength. A better
measure of strength is the maximum plastic strain, as calculated in a elasto-plastic
analysis.

Nor does a linear elastic stress analysis reflect the fact that the local yielding caused
during overloads influences the subsequent growth of a crack under relatively small
amplitude (high cycle fatigue conditions). The possible local yielding may blunt the

1
A ductile collapse of structures with in-plane compressive or shear stresses are associated with loss of
stiffness due to out-of-plane displacements, and an accurate representation may require incorporation of
plasticity as well as large deflection effects.
12.78
Chapter 12 - Nonlinear Analysis

crack tips and/or change the residual stress pattern. Regarding the possible change of a
residual stress field, it should be noted that current (high cycle) fatigue design rules
have calibrated the strength data according to the assumption that there will be a
tensile yield zone adjacent to the weld, see e.g. Gurney and Maddox (1972).

Large deflection, elasto-plastic behaviour of plate-strip subjected to lateral loading.

The purpose of the present example is to investigate the structural behaviour of a plate
subjected to local lateral load such as a wheel load on a helideck, or on the deck of a
ro-ro carrier etc. The actual design criterion may be stated in terms of an ultimate or
serviceability limit state condition.

A typical stiffened panel is shown in Fig. 12.41. If an ultimate strength criterion is


used for the stiffeners and girders the corresponding strength can be relatively
accurately estimated by a consideration of plastic-hinge theory. If the design criterion
is formulated in terms of a serviceability requirement allowing no yielding; the
structural analysis may be properly accomplished according to linear elastic beam
theory. Therefore, the attention here is focused on the load-carrying behaviour of the
plating. A plate-strip of unit width with material and geometric properties as shown in
Fig. 12.42 is considered. The material is assumed to be mild steel with no strain
hardening. The geometries correspond to slenderness ratios, b / t of 40 and 60.

Fig. 12.41 Stiffened panel with categorization of regions according to structural


behaviour.

The plate-strip is considered to be simply supported on the stiffeners. Two


configurations of horizontal restraint are analyzed, namely complete constraint against
movement and an elastic restraint corresponding to the membrane flexibility of the
adjacent spans, see Fig. 12.42.

The lateral load is uniform with intensity q over the central 200 mm of the loaded
span. The extreme load intensity q = 2.0 N / mm 2 may refer to an accidental, crash
landing of a helicopter of such a magnitude that the wheel legs collapse.
q = 1.2 N / mm may be a more normal load during the landing of a helicopter.

The plate-strips are modeled by beam elements of equal length, b/2, with due
consideration of the symmetry of the problem. The thickness of the plate is divided by
the layers.

12.79
Chapter 12 - Nonlinear Analysis

Fig.12.42 Geometrical and material data for plate-strip.

Typical load-deflection curves obtained by a finite element large deflection, elasto-


plastic analysis are displayed in Fig. 12.43. They are compared with solutions obtained
according to small deflection theory combined with linear elastic and rigid plastic
representation of the material behaviour. Due to the development of membrane action
in a real plate-strip undergoing large deflections, a prediction by a small displacement
theory will underestimate the stiffness and strength significantly. According to this
theory the first yielding occurs at q = 0.35 N / mm 2 , and plastic collapse at
q = 0.53 N / mm 2 . The more precise structural representation predicts initial yielding
to take place at about q = 0.54 N / mm 2 (for b / t = 40; σ Y = 320 N / mm 2 ) and no
danger of collapse even at q = 2.0 N / mm 2 .

Fig. 12.43 Load-deflection curves for a plate-strip of thickness


t = 10 mm; b = 400 mm, and yield stress σ Y = 320 N / mm 2 .

12.80
Chapter 12 - Nonlinear Analysis

Table 12.4 summarizes characteristic response values of various plate-strip


configurations as obtained by a large deflection, elasto-plastic analysis. It is noted that
ultimate capacity is not explicitly given in Table 12.4, because ductile collapse
mechanisms are not developed for any configuration at a load intensity less than
2.0 N / mm 2 . However, for the case b / t = 60 and σ Y = 240 N / mm 2 yielding over the
complete cross-section is developed at q = 1.8 N / mm 2 . By sufficient additional
deflection a stable equilibrium can be maintained even at q = 2.0 N / mm 2 . The
maximum strain is reported because it may be a measure for ultimate capacity as
defined by fracture. (Another point here is that for the two cases exhibiting the
maximum strains strain hardening will occur and limit both maximum deflections and
strains.) Maximum total and permanent deflection are given because they are relevant
as an SLS criterion. It is observed that a load implying a plastic collapse mechanism
according to small deflections theory barely seems to lead to yielding and permanent
deflections when the true finite deflections are taken into account. It is particularly
noted that the membrane boundary condition is an important parameter as regards total
deflection. However, the permanent deflection appears to be less dependent upon the
stiffness of the membrane spring used in this case study (see Fig. 12.42).

Table 12.4 Responses of plate-strips of different configurations subjected to a lateral


load, see Fig. 12.42.
Case Response quantities
Membrane Maximum Permanent
central Maximum
Geometric boundary central
Material Load deflection, deflection strain,
parameter conditions; property σ ε max
q wmax wp
b/t spring Y

stiffness, k: ε max / ε Y
wmax / t wp / t
40 ∞ 240 1.2 0.95 0.64 5.7
40 ∞ 240 2.0 1.36 1.14 15.6
40 ∞ 320 1.2 0.81 0.35 3.1
40 ∞ 320 2.0 1.14 0.76 5.5
40 E/40 320 1.2 1.10 0.41 3.8
40 E/40 320 2.0 1.49 0.79 6.1
60 ∞ 240 1.2 1.48 0.90 6.6
60 ∞ 240 2.0 2.20 -1) 17.5
1)
The numerical solution procedure is not able to follow the unloading path for this
case

It is generally expected that the plate-strip analysis yields conservative results. The
inherent conservativism should be explored by a proper two-dimensional analysis
before it is applied in design.

The above study has the following design implications. If sufficient membrane
restraints can be mobilized by the surrounding structure, the ultimate ductile strength
of a laterally loaded plate can be very high. Therefore, the total or permanent
deflections may be governing in the dimensioning. A reasonable serviceability
requirement on maximum allowable permanent deflection could be of the same order
12.81
Chapter 12 - Nonlinear Analysis

of magnitude as the deflections caused during fabrication processes. If the structural


component can be subjected to in-plane compressive loads after the lateral load, the
possible reduction strength due to the permanent deflections must be observed.

Stiffened plates

The stiffened panel is a fundamental structural component in ship hulls. Due to the
simplicity in fabrication and excellent strength to weight ratio, it is also widely used in
civil engineering, bridge, aerospace, offshore and other engineering fields design of
ship structures. Fig. 12.44 shows a typical stiffened plate structure.

Figure 12.44 A stiffened panel

The main purpose of the plates is to transfer the hydrostatic loads (the difference
between external and internal pressure) to the stiffeners, which again, through beam
action, transfer the loads to the transverse girders. These are parts of the transverse
frames of the hull girder. From the vertical girders the loads are transferred to the
heavy longitudinal girders or as membrane stresses in the side of the hull.

The stiffened panel in ships is generally subjected to combined in-plane and lateral
pressure loads. In plane loads include biaxial compression/tension, biaxial in-plane
bending and edge shear, which are mainly induced by overall hull girder bending and /
or torsion of the vessel. Lateral pressure loads are due to water pressure and/or cargo.
These loads components are not always applied simultaneously, but more than one
normally exist and interact .

Theoretically, the primary modes of overall failure for a stiffened panel can be
categorized into the following five groups, namely:
¾ Excessive plasticity of the plate after overall (grillage) buckling occurs in the
elastic regime

12.82
Chapter 12 - Nonlinear Analysis

¾ Yielding along plate-stiffener intersection after local buckling or collapse of


plating between stiffeners occurs
¾ Column or beam-column type collapse of the plate-stiffener combination as
representative of the stiffened panel after local buckling of collapse of plating
between stiffeners occurs
¾ Local buckling of stiffener web after local buckling or collapse of plating
between stiffeners occur
¾ Tripping of stiffener after local buckling or collapse of plating between
stiffeners occur

For a stiffened panel, the plate flange of a panel stiffener is usually not fully effective
because plate buckling results in a non-uniform stress distribution (see Fig. 12.45).

The rationale of effective width of buckled plating is such that the stress distribution of
the stiffener and its associated effective plating can be regarded as uniform and equal
to the peak stress σmax (Fig. 12.45). It would be incorrect to use the panel overall
average stress, which is given as σmean.

The effects of plate buckling are usually taken into account in design codes by means
of the effective width approach. As illustrated in Fig. 12.45 the regions close to the
stiffener are considered to be effective in carrying load while the regions remote from
the stiffeners are considered to be fully ineffective in resisting compression. The
ultimate load is generally considered to be reached when the maximum stress reaches
the yield stress. Many different effective width equations have been proposed.
Faulkner simple and useful expression is:
be 2 1
= − 2
b β β

, where β is the plate slenderness parameter and is defined as:

b σY
β=
tp E

Real stress distribution


σmax

Effective
stress distribution
b
be
σmean
b

Fig. 12.44 Stress distribution in a plate in post-buckling regime

12.83
Chapter 12 - Nonlinear Analysis

12.7 Analysis of accidental load effects

12.7.1 General

As mentioned above, the ALS check is a survival check of the structural system which
is damaged due to accidental actions or abnormal strength. Accidental actions are
caused by human errors or technical faults, and include fires and explosions, ship
impacts, dropped objects, unintended distribution of variable deck loads and ballast,
change of intended pressure difference. Most notable in this connection is of course
accidental loads such as ship impacts, fires and explosions which should not occur, but
do so because of operational errors and omissions.

The accidental actions and abnormal conditions of structural strength are supposed to
be determined by risk analysis, see e.g.Vinnem (1999), by accounting for relevant
factors that affect the accidental loads. In particular, risk reduction can be achieved by
reducing the probability of initiating event; leakage and ignition (that can cause fire or
explosion), ship impact, etc. or by reducing the consequences of hazards. Passive or
active measures can be used to control the magnitude of the accidental event and,
thereby, its consequences. For instance, the fire action is limited by sprinkler/inert gas
system or by fire walls. Fenders can be used to reduce the damage due to collisions.

ALS checks apply to all relevant failure modes as indicated in Table 12.5. Account of
accidental loads in conjunction with the design of the structure, equipment as well as
safety systems is a crucial safety measure, to prevent accidents to escalate. Typical
situations where direct design may affect the layout and scantlings are indicted in
Table 12.6.

Table 12.5 Safety criteria


Limit states Description Remarks
Ultimate (ULS) - Overall “rigid body” stability Different for bottom –
- Ultimate strength of structure, supported, buoyant, ---.
mooring or possible foundation Component design check
Fatigue (FLS) - Failure of joint Component design check
depending on residual system
strength after fatigue failure
Accidental - Ultimate capacity1) of damaged System design check
collapse (ALS) structure (due to fabrication
defects or accidental loads)
or operational error
1)
Capacity to resist “rigid body” instability or total structural failure

12.84
Chapter 12 - Nonlinear Analysis

Table 12.6 Examples on accidental actions for relevant failure modes of platforms

Structural concept Failure mode Relevant accidental action


Fixed platform Structural failure All
Structural failure All
Floating Instability • Collision, dropped object, unintended
platform pressure …, unintended ballast that
initiate flooding
Mooring system •Collision on platform
Strength • (Abnormal strength)
Tension leg Structural failure All
platform Mooring - slack • Accidental actions that initiate flooding
system - strength • Collision on platform
• Dropped object on tether
• (Abnormal strength)

Table 12.7 Design implications of accidental loads


Direct design of
- structure (to avoid progressive structural failure or flooding)
- equipment & protective barrier (to avoid damage and escalation of
accident)

Load Structure Equipment Passive protection system


Fire Columns /deck Exposed equip. Fire barriers
(if not protected) (if not protected)
Explosion Topside Exposed equip. Blast / Fire barriers
(if not protected) (if not protected)
Ship impact Waterline structure Possibly exposed Possible fender systems
(…. subdivision) risers, …
(if not protected) (if not protected)
Dropped object Deck Equipment on deck, Impact protection
Buoyancy elements risers and subsea
(if not protected)

Different subsystems, like:


¾ loads-carrying structure & mooring system
¾ process equipment
¾ evacuation and escape system
are designed according to criteria given for the particular subsystems. For instance,
safety criteria for structural design are given in terms of ULS, FLS and ALS with
specific target levels, and, hence, implies a certain residual risk level.
ALS is carried out by checking the system strength after e.g. the effect of accidental
actions with annual exceedance probability of 10-4 (as determined by risk analysis).
A complete identification of a “Design Accidental Event” should also include an
estimate of the probability of occurrence. For each physical phenomenon (fire,
explosions, collisions, ..) there is normally a continuous spectrum of accidental events.
12.85
Chapter 12 - Nonlinear Analysis

A finite number of events has to be selected by judgement. These events represent


different action intensities at different probabilities. The characteristic accidental load
on different components of a given installation, could be determined as follows:
¾ establish exceedance diagram for the load on each component
¾ allocate a certain portion of the reference exceedance probability (10-4) to each
component
¾ determine the characteristic load for each component from the relevant load
exceedance diagram and reference probability.

Iin view of the uncertainties associated with the probabilistic analysis, more pragmatic
approach would normally suffice. Yet, significant analysis efforts are involved in
identifying the relevant design scenarios for the different types of accidental loads.

For each design accident scenario the damage imposed on the offshore installation
needs to be estimated, followed by an analysis of the residual ultimate strength of the
damaged structure in order to demonstrate survival of the installation. To estimate
damage , i.e. permanent deformation, rupture etc of parts of the structure, nonlinear
material and geometrical structural behaviour need to be accounted for. While in
general nonlinear finite element methods need to be applied, simplified methods, e.g.
based on plastic mechanisms, are developed and calibrated using more refined
methods, to limit the computational effort required.

In the following sections the estimation of damage due to accidental loads will be
exemplified.

12.7.2 Fires and explosions

The dominant fire and explosion events are associated with hydrocarbon leakage from
flanges, valves, equipment seals, nozzles etc. Commonly the effect of 40 - 60
scenarios need to be analyzed. This means that location and magnitude e.g. of relevant
hydrocarbon leaks, likelihood of ignition, as well as combustion and temperature
development(in a fire) and pressure-time development (for an explosion) needs to be
estimated, followed by a structural assessment of the potential damage.

The fire thermal flux may be calculated on the basis of the type of hydrocarbons,
release rate, combustion, time and location of ignition, ventilation and structural
geometry, using simplified conservative semi-empirical formulae or
analytical/numerical models of the combustion process. The heat flux may be
determined by empirical, phenomenological or numerical method
(SCI,1993;BEFETS,1998). Typical thermal loading in hydrocarbon fire scenarios may
be 200- 300 kW/m2 for a 15 min – 2 hours period. The structural effect is primarily
due to the reduced strength with increasing temperature.

In case of explosion scenarios the analysis of leaks is followed by a gas dispersion and
possible formation of gas clouds, ignition, combustion and development of
overpressure. Tools such as FLACS, PROEXP, or AutoReGas are available for this
effort. Typical overpressures for topsides of North Sea platforms is 0.2-0.6 barg, with
a duration of 0.1-0.5 s.

The damage due to explosion should be determined with due account of the nonlinear
and dynamic character of the action effects. Simple, conservative single degree of
12.86
Chapter 12 - Nonlinear Analysis

freedom models may be applied. In particular cases where simplified methods have
not been calibrated, nonlinear time domain analyses based on numerical methods like
the finite element method should be applied. A recent overview of such methods may
be found in Czujko (2001). For instance the behaviour of the topside structure of the
6-legged North Sea jacket shown in Fig. 12.46 under blast loading, has been studied.
Fig. 12.47 shows the failure mode of the stiffened lower deck. The analysis was
carried out by LS-DYNA using strain-based rupture criteria (Czujko, 2001). As
indicated in this figure the final failure is a rupture.

Fire and explosion events that result from the same scenario of released combustibles
and ignition should be assumed to occur at the same time, i.e. to be fully dependent.
The fire and blast analyses should be performed by taking into account the effects of
one on the other. The damage done to the fire protection by an explosion preceding the
fire, should be considered.

Fig. 12.46 Layout of topside of 6-legged North Sea jacket

Fig. 12.47 Failure mode of lower deck in topside structure in Fig. 12.46.

12.87
Chapter 12 - Nonlinear Analysis

12.7.3 Ship impacts

Ship impacts on fixed platforms could cuase reduction of structural strength and
possible progressive structural failure. For buoyant structures the impact damage can
lead to flooding and, hence, loss of buoyancy. The measure of damage in this
connection is the maximum indentation implying loss of watertightness. However, in
case of large damage, reduction of structural strength is also of concern for floating
structures.

Ship collision loads are characterised by a kinetic energy, described by the mass of the
ship, including hydrodynamic added mass and the speed of the ship at the instant of
impact. If the collision is non-central, i.e. the contact force does not go through the
centre of gravity of the platform (installation) and the ship, a part of the kinetic energy
may remain as kinetic energy after the impact. The remainder of the kinetic energy has
to be dissipated as strain energy in the installation and, possibly, in the vessel.
Generally this involves large plastic strains and significant structural damage to either
the installation or the ship or both.

The most probable impact locations and impact geometry should be established based
on the dimensions and geometry of the offshore structure and vessel, and should
account for tidal changes, operational sea-state and motions of the vessel and structure
which has free modes of behaviour. Impact scenarios should be established
representing bow, stern and side impacts on the structure as appropriate

The collision problem comprises both internal mechanics related large, inelastic
deformations at the point of contact as well as global hull bending of struck vessel and
interaction with the surrounding fluid (added mass, viscous forces etc.). A fully
integrated analysis is fairly demanding. It is, therefore, often found convenient to split
the problem into two uncoupled analyses, namely, the external collision mechanics
dealing with global inertia forces and hydrodynamic effects, and internal mechanics
dealing with the energy dissipation and distribution of damage in the two structures.
Only the latter issue is pursued herein. This involves estimating how the energy is
shared among the installation and the ship. The structural response of the ramming
ship and installation can formally be represented as load-deformation relation ships as
illustrated in Fig. 12.48. The strain energy dissipated by the ship and installation
equals the total area under the load-deformation curves. The total energy dissipation
may be expressed by:

w s, max w i, max
E s = E s,s + E s,i = ∫ R s dw s + ∫ R i dw i
0 0

where Ri and Rs are the resistance of installation and ship, respectively; and dwi and
dws are the deformation of installation and ship. As the load level is not known a
priori an incremental procedure is generally needed. It is customary to establish the
load-deformation relationships for the ship and the installation independently of each
other assuming the other object infinitely rigid. Fig. 12.48b shows approximate
resistance-indentation for ships. This approach may imply severe limitations, because
both structures will inevitably dissipate some energy regardless of their relative
strength. Care should therefore be exercised that the load-deformation curves
calculated are representative for the true, interactive nature of the contact between the
two structures.
12.88
Chapter 12 - Nonlinear Analysis

50
Broad side
Rs Ri D = 10 m D
40 = 1.5 m

Impact force (MN)


30
D
Stern end
Es,s Es,i 20 Stern corner D = 10 m
= 1.5 m
10 D
dws Ship FPSO dwi Bow
0
0 1 2 3 4
Indentation (m)
a) Impact force-intendation b) Impact force- intendation for
supply vessel impact on rigid
cylinrical column
Fig. 12.48 Energy absorption based on force-intendation relationship

Based on the relative energy absorption capabilities of the installation and ship, the
design of the installation different design principles may be distinguished; namely:
strength design; ductility design , or shared-energy design. As indicated in Fig.12.49
the distribution depends upon the relative strength of the two structures. Strength
design implies that the installation is strong enough to resist the collision force with
minor deformation, so that the ship is forced to deform and dissipate the major part of
the energy. Ductility design implies that the installation undergoes large, plastic
deformations and dissipates the major part of the collision energy. Shared energy
design implies that both the installation and ship contribute significantly to the energy
dissipation.

Ductile Shared-energy Strength


design design design
E n ergy d issipation

ship

installation

Relative strength - installation/ship

Fig 12.49 Ship impact design principle based on reløative energy sharing between ship
and installation

From the calculation point of view strength design or ductility design is favourable. In
this case the response of the «soft» structure can be calculated on the basis of simple
considerations of the geometry of the «rigid» structure. For instance Fig. 12.48b can
be used when the ship is soft while the platform is rigid to carry out a strength design
of the platform in the case of supply vessel impact. In shared energy design both the
magnitude and distribution of the collision force depends upon the deformation of both
structures. The analysis has to be carried out incrementally on the basis of the current
deformation field, contact area and force distribution over the contact area. It is the
current weaker structure that is forced to deform most, whereas the damage of the
12.89
Chapter 12 - Nonlinear Analysis

other may remain virtually unchanged during an incremental step. The relative
strength of the two structures may vary both over the contact area as well as over time.

Recent advances in computers and algorithms have made nonlinear finite element
analysis (NLFEM) a viable tool for assessing collisions. There are generally two
methodologies available: implicit analysis and explicit analysis. Implicit
methodologies require solution of equation systems. This places demands on the
equation solver and the computer capacity especially in terms of memory resources.
Explicit systems do not require equation solving. Equilibrium is solved at element
level. However, to maintain stability, very small time steps are needed. . Explicit
methodologies based computer codes include ABAQUS/Explicit, DYTRAN, LS-
DYNA, PAM-CRASH and RADIOSS, and implicit methodologies based codes
include ABAQUS/Standard, ANSYS, MARC and NASTRAN.

The internal accident mechanics involve yielding, crushing, tearing or fracture. Any
non-linear FEM mesh for simulating the internal accident mechanics needs to be fine
enough so as to capture such highly non-linear characteristics. It is found that a
particularly fine mesh is required in order to obtain accurate results for components
deforming by axial crushing. Higher order elements generally provides better accuracy
and allow a less finer mesh, but they require more computational effort. The
importance of mesh fineness or element types has been studied by many investigators
(e.g. Amdal & Kavlie 1992, ). It is observed that a very large number of elements is
required in order to obtain accurate results for components deformed by axial crushing
forces. Accounting for realistic size and boundary conditions of FE models is also
crucial. Analytical formulae derived for evaluating structural damage characteristics
(e.g., failure patterns) may be used to determine relevant mesh size. For instance, it is
recommended that more than eight (rectangular plate-shell type) finite elements are
necessary to capture the structural crushing pattern within a half length of one
structural fold, see Figure 12.50. Available analytical formulations for predicting the
length of structural fold are suggested e.g. by Amdahl & Kavlie (1992). It is cautioned,
however, that these formulae were derived for different crushing patterns to different
structural geometries.

H
H
H

Figure 12.50: A thin-walled structure crushed under predominantly axial compressive


loads and cut at its mid-section (Paik & Thayamballi 2002)

An improved modeling technique for fracture behavior may be to introduce a double


set of nodes such that the elements are allowed to separate once the critical stress is
attained (Amdahl & Stornes 2001).

12.90
Chapter 12 - Nonlinear Analysis

The critical strain for fracture depends heavily on the stress-strain measure as well as
the mesh size. Various options exist for the stress-strain relationship. Most often
engineering stress-strain relationship or true stress-strain relationship is used. The true
stress-strain relationship can model the physical process more accurately than the
engineering stress-strain relationship, but it is more complex as the change of the
element volume needs to be involved. Experience obtained thus far indicates that the
difference between these two approaches can be neglected up to ultimate stress as long
as the load–displacement relation and the associated energy dissipation are concerned.
The dependency of fracture strain on element size has been studied e.g. by Simonsen
& Lauridsen (2000). By comparing tensile tests with numerical simulations they
concluded that a fairly small element mesh is required to capture the features of the
tensile test. An empirical relationship between fracture strain and element size has
been derived by Lehmann et al (2001), based on thickness measurements of structural
elements from actual collisions.

A major challenge in NLFEM analysis is prediction of ductile crack initiation and


propagation. This problem is not yet solved. Crack initiation and propagation should
be based on fracture mechanics analysis, using the J-integral or Crack Tip Opening
Displacement method rather than simple strain considerations. A difficulty in this
connection is that the strain depends upon the element mesh. The simplest approach
to the problem is to remove elements once the critical strain is attained. This is fairly
easily done in an explicit code because there is no need to assemble and invert the
effective system stiffness matrix. However, deleting elements disregards the fact the
large stresses can be maintained parallel to the cracks. An improved modelling is to
introduce a double set of nodes such that the elements are allowed to separate once the
critical stress is attained. A drawback with a double set of nodes is that the potential
location of cracks needs to be defined prior to analysis.

Paik et al.(2003) compared FE analyses with test results obtained by The Association
of Structural Improvement of Shipbuilding Industry of Japan (ASIS 1993). One of the
collision test models is a double side structure model made of mild steel. It was
impacted from outer side shell by a 82.32 kN weight (striking bow) freely fallen from
a height of 4.8m. The weight struck the double hull model at a speed of 9.7 m/s.

In FE modelling, the element size should be fine enough so that deformation patterns
be properly captured in the analysis. It is desirable that the shape of the element is
rectangle and the aspect ratio of an element is near 1.0. While the deformation patterns
of steel plates under axial compression at the ultimate limit state have a sinusoidal
shape, ship collisions and grounding cause more complex deformation patterns
involving folding and tearing as well as localized yielding. To investigate the effect of
mesh fineness, two meshes are considered: one is a coarse mesh usually applied for an
ultimate strength analysis and the other is a fine mesh more suitable for a collision
analysis.

For an ultimate strength analysis, five elements between a stiffener spacing may be
enough to capture the collapse pattern of the plating between stiffeners, i.e., with one
element size of 80mm for the collision test model considered. However, for a collision
analysis involving structural folding, a finer mesh is required. As previously noted in
Section 3.5.2, at least eight elements are needed within a half length of one structural
fold, see Figure 11. In the present benchmark study, the fold length was estimated by

12.91
Chapter 12 - Nonlinear Analysis

H = 0.983b 2 / 3 t 1 / 3

where b = width of plating between stringers, t = plate thickness, H = a half fold


length. With b=2000mm and t=7mm for the test model, a half length of one fold is
H = 298.5 mm. Therefore, it is recommended that one element size must be smaller
than 298.5/8=37.3 mm so that at least 11 elements are necessary between a stiffener
spacing (i.e., 400mm). In the present benchmark study, 13 elements were used for
modelling of plating between stiffeners, which corresponds to the element size of
400/13=30.77mm.

Fracture strain as obtained by the tensile coupon test was used for both types of
element sizes which are relatively large. The strain rate sensitivity effect on the
material yield stress of mild steel was accounted for using the Cowper-Symonds
formula (with C = 40.4 and q = 5 ) , while the effect of strain rate on fracture strain
was not considered (since LS-DYNA does not deal with it).

Two types of material models are considered: model I for accounting for the strain-
hardening effect but neglecting the neck effect and model II accounting for both
strain-hardening and necking effects.

Figure 12.51 shows the two types of finite element modelling for the collision test
model, namely a coarse mesh for the ultimate strength analysis and a fine mesh for the
collision analysis. Figure 12.52 shows the deformation patterns obtained from the two
types of finite element sizes.

Figure 12.51: Two types of finite element models with meshes: coarse mesh for the
ultimate strength analysis (upper) and fine mesh for the collision analysis (lower)

12.92
Chapter 12 - Nonlinear Analysis

(a) Coarse mesh, penetration of 500mm (b) Fine mesh, penetration of 500mm

(c) Test, penetration of 800 mm (ASIS 1993)

Figure 52: Deformation patterns obtained by the mesh type with material model II (a) coarse
for the ultimate strength analysis, (b) fine for the collision analysis and (c) by the test

Figure 12.53 compares the force-penetration curves (left) and the absorbed energy-
penetration curves (right). It is evident that the fine mesh for the collision analysis
captures the folding mechanism more properly than the coarse mesh for the ultimate
strength analysis. In this example, the difference between material models I and II is
negligible because rupture associated with necking is not a dominant failure mode.
1.6 ① Experiment
4 ① Experiment ②③ ② Coarse mesh with Material modelⅠ
② Coarse mesh with Material modelⅠ ③ Coarse mesh with Material model Ⅱ
③ Coarse mesh with Material model Ⅱ ④ ④ Fine mesh with Material modelⅠ
④ Fine mesh with Material modelⅠ ⑤ Fine mesh with Material model Ⅱ
⑤ Fine mesh with Material model Ⅱ 1.2 ③
3 ②
Energy (MJ)

① ⑤
Force (MN)

0.8 ⑤
2 ④

1 0.4

0 0
0 200 400 600 0 200 400 600
Penetration (mm) Penetration (mm)

Figure 1253: Collision force versus penetration (left) and absorbed energy versus
penetration (right)

12.93
Chapter 12 - Nonlinear Analysis

Appendix A

Solution of the differential equation of a beam with a axial load

A.1. Differential equation

The differential equation for a beam with axial force is (Eq. 2.105)

∂2 ⎛ ∂2w ⎞ ∂2w
⎜⎜ EI 2 ⎟⎟ + P 2 = q ( x) (A.1)
∂x 2 ⎝ ∂x ⎠ ∂x

For constant EI, this equation may be written as

∂4w 2 ∂ w
2
q( x) 2 P
+ k = ;k = (A.2)
∂x 4
∂x 2
EI EI

The general solution of Eq. (A.2) is composed of a homogenous and particular


solution. The homogenous solution is obtained by solving with respect to w(x) for
q(x) ≡ 0.

x x
wh ( x) = C1 sin 2 β + C 2 cos 2 β + C 3 x + C 4 for compressive P (A.3)
A A

wh ( x) = C1 sinh (kx ) + C 2 cosh (kx ) + C 3 x + C 4 for tensile P (A.4)

where
A P
β= (A.5)
2 EI

and Ci are constants to be determined by boundary conditions. The particular solution


satisfies Eq. (A.1) and would, hence, depend upon q(x).

For example, with

q( x) = α 1 + α 2 x

the particular solution becomes

1 ⎛1 1 3⎞
w p ( x) = ⎜ α 1 x + α 2 x ⎟ + Ax + B
2
(A.6)
P⎝2 6 ⎠

The constants A and B can be chosen arbitrarily, but will be influenced by the choice
of C3 and C4 in the homogenous solution, since the boundary conditions should be
satisfied by the total solution:

w( x) = wh ( x) + w p ( x) (A.7)

12.94
Chapter 12 - Nonlinear Analysis

Sofar, the exact solution for a problem with distributed load, q(x) has been established.

Now the solution for a beam with end moments and shear forces will be established,
see Fig. A.1. This case will serve as basis for obtaining the stiffness relationship for a
beam with axial force

Figure A.1 Beam with axial force, with no lateral load.

The solution for this problem is obtained in another manner than by solving Eq. (A.2)
directly.

The moment at location C may be expressed by moment equilibrium with respect to C


of the left hand part of the beam:

M ( x) = − M A + P ( w − w A ) + Qx (A.8)

The moment-curvature relationship is given by Eq. (2.77) as

∂2w
M = − EI (A.9)
∂x 2

By combining Eqs (A.4, A.5)

∂2w
− EI = − M A + P ( w − w A ) + Qx (A.10)
∂x 2

or
∂2w P 1
+ w= ( Pw A + M A − Qx) (A.11)
∂x 2
EI EI

Eq. (A.11) is a 2nd order differential equation as opposed to the 4th order one in Eq.
(A.2). The homogeneous solution of Eq. (A.7) for a positive P is:

P
wh = C1 sin kx + C 2 cos kx ; k 2 = (A.12)
EI

The solution for a negative P is similar, with hyperbolic functions instead of


trigonometric ones.

The total solution (for P>0) may be written as:

12.95
Chapter 12 - Nonlinear Analysis

1
w( x) = C1 sin kx + C 2 cos kx + ( Pw A + M A − Qx) (A.13)
P

A.2 Stiffness matrix with stability function

The exact solution for a beam with an axial force (and no lateral load) is

⎛ x⎞ ⎛ x⎞ 1
w( x) = C1 sin ⎜ 2 β ⎟ + C 2 cos⎜ 2 β ⎟ + ( Pw A + M A − Qx)
⎝ A⎠ ⎝ A⎠ P
(A.14)
2β ⎛ x⎞ 2β ⎛ x⎞ Q
w , ( x) = C1 cos⎜ 2 β ⎟ − C 2 sin ⎜ 2 β ⎟ +
A ⎝ A⎠ A ⎝ A⎠ P

A P
where β =
2 EI

This solution applies for an arbitrary beam as shown in Fig. A.2, with transverse
displacement and rotations of the ends (wA, θA) and (wB, θB), end moments MA, MB
while the shear force is constant and equal to Q (due to no lateral load, q(x)).

Based on the solution (A.14), kinematic boundary conditions and equilibrium


equations, a relationship between forces MA, MB and Q and displacements wA, θA, wB
and θB can be established as shown in as follows.

Figure A.2: Beam with axial force.

The boundary conditions are

MA
w(0) = C 2 + w A + = wA
P
2β Q
w '(0) = C1 − = −θA
A P
MA Q
w(A) = C1 sin 2β + C 2 cos 2β + w A + − A = wB
P P
2β 2β Q
w '(A) = C1 cos 2β − C 2 sin 2β − = −θB
A A P

The first two equations result in:


12.96
Chapter 12 - Nonlinear Analysis

⎛Q ⎞ A M
C1 = ⎜ − θ A ⎟ ; C2 = − A
⎝P ⎠ 2β P

The last two equations then become:


MA QA ⎛ s ⎞ s
(1 − c) − ⎜⎜1 − ⎟⎟ + w A − w B − θ A A =0
P P ⎝ 2β ⎠ 2β
MA Q
2 βs − (1 − c) − θ A c + θ A = 0
PA P
where
s = sin 2 β , c = cos 2 β

Solving these equations with respect to MA and Q yields:

PA 1 ⎡ 2β ⎤
MA = (1 − c) ( wB − w A ) + ( s − 2 β c)θ A + (2 β − s )θ B ⎥ (A.15a)
4 β 1 − c − β s ⎢⎣ A ⎦

P ⎡ 2β s ⎤
Q= (w B − w A ) + (θA + θB )(1 − c) ⎥ (A.15b)
2(1 − c − βs) ⎢⎣ A ⎦

Moment equilibrium requires

M A + M B − P(w B − w A ) − QA = 0

which yields

M B = − M A + P(w B − w A ) + QA
PA 1 ⎡ 2β ⎤ (A.15c)
= (1 − c)(w A − w B ) + (2β − s)θA + (s − 2β c)θB ⎥
4β 1 − c − βs ⎢⎣ A ⎦

The expressions (A.15a-c) may be simplified by using the so-called φi-functions,


defined by:

A P
φ1 = β cot gβ ; β=
2 EI
1 β2
φ2 =
3 1 − φ1
1 3
φ3 = φ1 + φ2 (A.16)
4 4
1 3
φ4 = − φ1 + φ2
2 2
φ5 = φ1φ2

By using these functions Eq. (A.15a-c) is transformed into:


12.97
Chapter 12 - Nonlinear Analysis

6EI 4EI 2EI


MA = φ2 (w B − w A ) + φ3 θ A + φ 4 θB
A2 A A
6EI 4EI 4EI
M B = 2 φ2 (w B − w A ) + φ 4 θA + φ3θB (A.17)
A A A
12EI 6EI 6EI
Q = 3 φ5 (w B − w A ) + 2 φ2 θA − 2 φ2 θB
A A A

By means of Eq. (A.17) the stiffness matrix for an element with 4 d.o.f. can be
established.

The relations (A.17) may then be used to establish the stiffness relationship S = kv,
with Si and vi defined as shown in Fig. A.3b, by observing that:

S 2 = −Q, S 3 = M A , S 5 = Q, S 6 = M B
(A.18)
v 2 = w A , v3 = θ A , v 5 = wB , v 6 = θ B

By combining these relations with Eq. (A.18) the resulting equations are written on
matrix form as:

⎡S 2 ⎤ ⎡ 6φ 5 − 3Aφ 2 − 6φ 5 − 3Aφ 2 ⎤ ⎡v 2 ⎤
⎢S ⎥ ⎢− 3Aφ 2A φ 3
2
3Aφ 2 A 2φ 4 ⎥⎥ ⎢v ⎥
⎢ 3 ⎥ = 2 EI ⎢ 2 ⎢ 3⎥ (A.19)
⎢S 5 ⎥ A 3 ⎢ − 6φ 5 3Aφ 2 6φ 5 3Aφ 2 ⎥ ⎢ v5 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣S 6 ⎦ ⎣− 3Aφ 2 A 2φ 4 3Aφ 2 2A 2φ 3 ⎦ ⎣v 6 ⎦

or

S , = k ,NL v ,

a) 6 d.o.f. model b) 4 d.o.f. model

Figure A.3 Beam element with axial force.

12.98
Chapter 12 - Nonlinear Analysis

Appendix B

General formulations for geometrically nonlinear behaviour

B.1 Continuum mechanics

General

The purpose of the present appendix is to briefly define some fundamental concepts of
the mechanics of continua, restricted to three-dimensional Euclidean spaces.

Coordinates are means of identifying or labeling material particles. The simultaneous


position of the set of particles comprising the body is called the configuration of the
body. The motion of the body is a continuous sequence of configurations in space and
time.

Figure B.1 Deformation of a body.

Two configurations of a body in motion are shown in Fig. B.1. The initial
configuration is denoted C0. P0 indicates the initial location of a material particle. The
current configuration is denoted C, and P marks the present location of the material
particle.

The original configuration is described by the coordinates xi. To describe the motion
og the body two choices of coordinates are preferred:

¾ the same rectangular Cartesian coordinates are used to describe both the
original reference and deformed configurations (e.g. xi in Fig. B.1).
or
¾ the frame of reference is distorted in such a way that the coordinates of Xi of a
particle have the same numerical values xi as the reference configurations.
Such coordinates that follow deformations become survilinear (called
convective coordinates).

For the present purpose the first choice is most appropriate and is hence, used.

If the reference configuration within this choice is the actual initial configuration at t0
= 0, and the independent variables are the coordinates xi and time t, then the
description of motion is called material description or Lagrangian description. The
alternative is spatial description whose independent variables are the present position
xi occupied by the particle at time t and the present time t. The spatial description,
12.99
Chapter 12 - Nonlinear Analysis

also called the Eulerian description focuses attention on fixed points of space instead
of on a given points in the body. The description is most used in fluid mechanics. We
shall limit ourselves here to material or Lagrange description.

However, it is necessary to relate the coordinates xi of the initial configuration with the
coordinates Xi of the displaced body. The relation between the two coordinates is
given by the displacement as illustrated in Fig. B.2. In component form the relation is:

X i = xi + u i (B.1)

where ui = ui(xi, t)

Figure B.2 Definition of displacement.

Strains

Alternative expressions for strain and stress appropriate for large deformation
problems are envisaged (e.g. Crisfield, 1991). They include the Almansi and Green
strains. The Green (-Lagranage) strain is commonly applied. It is defined by the
equations

1 ⎛ ∂u ∂u j ∂u k ∂u k ⎞
Eij = ⎜ i + + ⎟
2 ⎜⎝ ∂x j ∂xi ∂xi ∂x j ⎟
⎠ (B.2)
= (u i ' j + u j 'i + u k 'i u k ' j )
1
2
3
when adopting the Einstein summation convention that u k u k = ∑ u k2 .
i =1

The Almansi strain is defined with reference to the deformed geometry.

By defining ui by u1 = u, u2 = v and u3 = w, Eqs (B.2) may be written fully as:

12.100
Chapter 12 - Nonlinear Analysis

E xx = u , x +
1 2
2
( )
u , x + v, 2x + w, 2x

1 2
(
E yy = v, y + u , y + v, 2y + w, 2y
2
)
1
(
E zz = w, z + u , 2z + v, 2z + w, 2z
2
) (B.3)

E xy = u , y + v, x + (u , x u , y + v, x + v, y + w, x w, y )
E yz = v, z + w, y + (u , y u , z + v, y + v, z + w, y w, z )
E zx = w, x + u , z + (u , z u , x + v, z + v, x + w, z w, x )

The initial terms in Eqs (B.2) are the customary engineering definitions of normal and
shear strain (εx = u,x, etc.). The added terms, in parentheses, become significant if
displacement gradients are not small. Green-Lagrange strains are zero for a rigid-body
rotation of any magnitude. In Eqs (B.2), all displacement derivatives are computed in
the original coordinate system, regardless of how large a rigid-body rotation may be
supposed on the deformations. This is the “total Lagrangian” approach, in which all
displacements are measured in a reference frame that is stationary rather than attached
to the deforming structure. The stationary coordinates may also be called “material
coordinates” and may be denoted in some papers by uppercase labels X, Y and Z (or
xi).

Green normal strains correspond to defining the strain of a line segment by the
equation

1 ⎡⎛ ds * ⎞ ⎤
2

E = ⎢⎜ ⎟ − 1⎥ (B.4)
2 ⎢⎣⎝ ds ⎠ ⎥⎦

where ds and ds* are respectively the initial and final lengths of the line segment. If
ds ≈ ds*, Eq. (B.3) reduces to the usual small-strain approximation, E = E (ds*-ds)/ds.

Stress

Consider the body in deformed state. In Fig. B.3 Δf is the force acting on the surface
element Δa and n is the unit normal vector at Δa with outwards direction.

Figure B.3 Surface force on deformed body.

12.101
Chapter 12 - Nonlinear Analysis

Assume that the ration Δf/Δa tends to a definite limit at Δa tends to zero. The force t
per unit area acting on the surface in the limit condition is denoted traction

Δf df
t = lim = (B.5)
Δa →0 Δa da

The equilibrium equation at the surface of the body gives the definition of the Eulerian
(Cauchy) stress tensor, σij

σ ij ni = t j (B.6)

In Eq. (B.6) tj and ni are, respectively, components of t and n. σij refers to the current
configuration which is a natural physical concept. However, stresses must be related
to strains. In the Lagrangian description of motion the strains Eij are referred to the
original position of the particles. Therefore, it is convenient to define stresses to the
initial area. The stress tensor should be energy conjugate with the strains Eij. Since Eij
is symmetric, the stress tensor should also be symmetric. These stresses, Sij are
denoted Piola-Kirchhoff stresses, and can be expressed by the Cauchy stress, σlk as
follows

∂x i ∂x j
S ji = σAk / det | ∂x i / ∂X j | (B.7)
∂Xk ∂XA

where det | ∂xi/∂Xj | is the determinant of the deformation matrix {∂xi/∂Xj}. For
normal strains this determinant is equal to unity.

Equilibrium equations

It can be shown (e.g. Crisfield, 1991) that equilibrium equations for a body with
volume forces can be formulated with reference to the initial geometry as

∂ ⎡ ⎛ ∂u i ⎞⎤
⎢ S jk ⎜⎜ δ ik + ⎟⎟⎥ + ρ 0 F0i = 0 (B.8)
∂x j ⎣ ⎝ ∂x k ⎠⎦

Virtual work

Eq. (B.8) expresses the equilibrium of a body in the direction i. The body is acted
upon by surface tractions and body forces. The surface of the body under
consideration may be thought of as consisting of two areas. On area S1 the surface
tractions are specified, while the displacements are specified on the area S2. By giving
the body from its equilibrium position a virtual displacement δui, which is
kinematically consistent with the boundary conditions, the following equation may be
derived:

12.102
Chapter 12 - Nonlinear Analysis

1 ⎛ ⎛ ∂u i ⎞ ⎛ ∂u ⎞ ⎛ ∂u ∂u k ⎞ ⎞
⎜δ ⎜ ⎟ +δ ⎜ j ⎟⎟ + δ ⎜ k ⎟ ⎟ dV
2 V∫ ⎜⎝ ⎜⎝ ∂x j
S ⎜ ∂x
ij ⎟ ⎜ ∂x ∂x ⎟ ⎟
⎠ ⎝ i ⎠ ⎝ i j ⎠⎠ (B.9)
− ∫ Ti δ u i dS − ∫ ρ Fi δ u i dV = 0
S1 V

The index zero has been omitted here as everything is referred to the original
configuration. Ti denotes the prescribed surface tractions. By introducing the Green’s
strain tensor of Eq. (B.2) in Eq. (B.9) the virtual work equation may be written as

V
∫S ij δ Eij dV − ∫ Ti δ u i dS − ∫ ρ Fi δ u i dV = 0
S1 V
(B.10)

B.2 Principle of virtual displacement on incremental form

Total Lagrange formulation

In this section an incremental form of the virtual work equation will be shown.

Fig. B.4 shows the spatial coordinate system xi and the body in three different
configurations. C0 is the initial configuration. Cn is some deformed configuration and
Cn+1 is a configuration close to Cn.

Figure B.4 Initial and deformed configuration of a body.

The virtual work principle (Eq. (B.10)) in configuration Cn+1 reads

∫S δ Eij* dV = ∫ Ti * δ u i* dA + ∫ Fi * δ u i* dV
A*
ij (B.11)
v S V

In Eq. (B.11) superscript * denotes quantities in configuration Cn+1. The superindex A


on the stress S ijA denotes stress due to external loads. The relations between these
quantities in configurations Cn+1 and Cn are
S ijA* = S ijA + ΔS ijA
E ij* = Eij + ΔEij
u k*,i = u k ,i + Δu k ,i (B.12a-e)
Ti * = Ti + ΔTi
Fi * = Fi + ΔFi
12.103
Chapter 12 - Nonlinear Analysis

The Δ denotes increment when moving from Cn to Cn+1. The virtual displacment fields
for configurations Cn and Cn+1 are assumed to be identical so that δ u i* = δ u i . The
virtual work principle for configuration Cn is expressed as in Eq. (B.10). Introducing
Eqs (B.12) into Eq. (B.11) yields an incremental form of the virtual work principle as

∫ ΔS δ Eij dV + ∫ S ijA δ ΔEij dV + ∫ ΔS ijA δ ΔEij dV


A
ij
V V V
(B.13)
⎛ ⎞
= ∫ ΔTi δ u i dS + ∫ ΔFi δ u i dV − ⎜⎜ ∫ S ijA δ Eij dV − ∫ Ti δ u i dS − ∫ Fi δ u i dV ⎟

S V ⎝V S V ⎠

If the body in configuration Cn is in equilibrium, the parenthesis on the right hand side
of Eq. (B.13) will vanish according to Eq. (B.10). However, due to approximations in
the solution procedure, configuration Cn will generally not be in equilibrium before
onset of further loading. The terms in the parenthesis will then act as an equilibrium
correcting term. The variation of Green’s strain terror, Eq. (B.4), is

δ Eij = 12 (δ u i , j + δ u j ,i + u k ,i δ u k , j + δ u k ,i u k , j ) (B.14)

(Note that δ Eij = Eij (u i + δ u i ) − Eij (u i ) .)

The increment of Green’s strain tensor when moving from configuration Cn to Cn+1 is

ΔEij = 1
2 ( Δu i,j + Δu j,i + Δuk ,i uk ,j + uk ,i Δuk ,j + Δuk ,i Δu k ,j ) (B.15)

The variation of the increment of Green’s strain tensor reads

δ ΔEij = 12 (δ u k , j Δu k ,i + δ u k ,i Δu k , j ) (B.16)

The constitutive relation is assumed to have the following form

ΔS ijA = C ijkA ΔE kA (B.17)

where C ijkA is the material tensor. The material modeling will be discussed in Section
10.4.

Eq. (B.17) may also be written as a vector-matrix relation if the components


ΔS Aij ( ΔEkl ) are collected in vectors, and Cijkl in a matrix.

Updated Lagrangian formulation

In the updated Lagrangian description the displacement increments are referred to the
preceding configuration Cn. The large displacements and rotations are taken into
account by updating the geometry. Hence, the nonlinear terms may be neglected in
Eqs (B.14 – B.15). They are, however, the only terms contributing to the variation of
the strain increment, see Eq. (B.17).
12.104
Chapter 12 - Nonlinear Analysis

The discussion has revealed that the equations resulting from the virtual work
principle on incremental form as expressed in Eq. (B.13) are nonlinear in Δui.
Linearization is obtained by neglecting second order terms. This is done by omitting
the third integral on the left hand side of Eq. (B.13) as well as by using the linearized
strains

δ Eij = 12 (δ u i , j + δ u j ,i )
(B.18a-b)
ΔEij = 12 (Δ u i , j + Δu j ,i )

Finally, the linearized form of the virtual work principle in incremental form can be
written as

∫C
V
ijkA ΔEkA δEijdV + ∫ S δΔE dV
V
ij ij

⎛ ⎞
= ∫ ΔT δu dS
S
i i + ∫ ΔF δu dV
V
i i − ⎜ ∫ S ijδEijdV − ∫ Ti δ u i dS −
⎝V S
∫ F δu dV ⎟⎠
V
i i

(B.19)
where Sij are the total stresses within the structure.

12.105
Chapter 12 - Nonlinear Analysis

Appendix C Plasticity theory

In the same manner as for the one dimensional case, elasto-plastic behaviour of metals
in a multiaxial stress state can be characterized by

¾ An initial yield condition, i.e. the state of stress for which plastic deformation
first occurs.
¾ A hardening rule which describes the modification of the yield condition due
to strain hardening during plastic flow.
¾ A flow rule which allows the determination of plastic strain increments at each
point in the load history.

It is assumed that the material is isotropic, which implies that the stiffness properties
are independent of orientation at a point.

In this section a brief review of the plasticity theory will be given. More details may
be found in textbooks like Crisfield(1991), Chapter 6.

C.1 The von Mises initial yield condition

The yield condition of a material defines the limit of purely elastic behaviour under
any combination of stresses. Many initial yield conditions have been proposed.
Experiments indicate that the von Mises yield condition best represents material
behaviour of most metals. Another advantage of the von Mises criterion is its simple
continuous function of stress components which makes it especially attractive to
numerical analysis.

The mathematical expression for the von Mises initial yield surface reads

f = σ −σY = 0 (C.1)

where f is the loading function and σY is the initial uniaxial yield stress of the material.
The equivalent stress σ is given by

σ = 3
2 sij sij = 3
2 σ ijσ ij − 12 (σ kk ) 2 (C.2)

where the Einstein’s summation convention is applied i.e.


n n n
sij sij = ∑∑ sij2 ; σ
i =1 j =1
kk = ∑σ
k =1
kk ; σ m = 13 σ kk .

sij denotes the deviatoric, or reduced stress tensor

sij = σ ij − 13 δ ij σ kk (C.3)

where
⎧1 for i = j
δ ij = ⎨
⎩0 for i ≠ j

12.106
Chapter 12 - Nonlinear Analysis

C.2 Hardening rule

By using the von Mises yield criterion, the yield criterion in connection with
hardening may be written as:

σ Y = H (ε p
) (C.4)

where the equivalent plastic strain is

ε p

ε = ∫ dε
p p
(C.5)
0

with the equivalent plastic strain increment

dε p = (2 / 3)dε p
ij dε p
ij (C.6)

(again using Einstein’s summation convention).

It is also convenient to introduce the plastic work, Wp:

ε ijp ε ijp

W = ∫ σ ij dε = ∫ σ dε
p p
ij
p
(C.7)
0 0

The hardening may be described by different models

¾ isotropic
¾ kinematic or generally anisotropic manner as illustrated for a one-dimensional
and two-dimensional condition in Fig. C.1.

The choice between models especially matters if the load condition/stress is reversed
or cyclic.

a) One-dimensional b) Two-dimensional

Figure C.1 Hardening models


12.107
Chapter 12 - Nonlinear Analysis

Experiments with metals show that a phenomenon denoted Bauschinger effect occurs.
This means that the material yields a lower stress level when the loading is reversed
than during the initial loading. As shown e.g. for the one-dimensional case in Fig.
C.1a, this feature is not captured by the isotropic hardening.

In the present section only isotropic hardening is considered.

The yield condition is written as

f = σ − H (ε p ) = 0 (C.8)

where σ is given by Eq. (C.2).

C.3 The flow theory of plasticity

The relation ship between stress and (plastic) strain may be obtained mainly by the
two major different plasticity theories, namely the deformation theory and the flow
theory. Experiments show that the flow theory is the better one when treating
problems with general loading paths, e.g. reversed and cyclic loading. Therefore, the
flow theory of plasticity is described herein.

The flow theory of plasticity yields an incremental relationship between Cauchy


stresses σij and true strains εij. For small strains the same constitutive law can be used
between Piola-Kirchhoff stresses and Green’s strains.

It is a fundamental assumption that the total strains εij may be decomposed into elastic
components ε ije and plastic components ε ijp

ε ij = ε ije + ε ijp (C.9)

The elastic components of strain are related to the stresses by Hooke’s law.

1+ ν ν
εije = Cijkl σ kl = 1
9K δijσ kk + 2G
1
sij = σij − δijσ kk (C.10a)
E E

or inversely

3νK
σij = E ijkl εekl = 2Gεije + δijε ekk (C.10b)
1+ ν

where the compression modulus K, modulus of elasticity, E and shear modulus, G are
such that:

3(1 − 2ν ) K = 2(1 + ν )G = E (C.10c)

The plastic mode is assumed to be incompressible, implying

ε iip = 0
12.108
Chapter 12 - Nonlinear Analysis

As was expected in Eq. (C.1) for initial yielding it is assumed that in general there
exists a loading function f. Setting f equal to zero defines a yield surface that bounds
the elastic range. The mathematical expression for the yield surface may be written

f (σ ij , ε ijp , κ ) = 0 (C.11)

for all σij and ε ijp components.

Eq. (C.11) defines the initial as well as subsequent yield surfaces depending upon the
values of the variables. κ is a hardening parameter. Different values of f define
different stress states

f < 0 ; elastic state of stress


f = 0 ; plastic state of stress (C.12a-c)
f > 0 inadmissible

Taking the differential of f from a plastic state gives

∂f ∂f ∂f
df = dσ ij + p dε ijp + dκ (C.13)
∂σ ij ∂ε ij ∂κ

again by invoking Einstein’s summation convention.

Then, three different loading conditions can be defined:

∂f
dσ ij < 0 ; f = 0 during unloading
∂σ ij
∂f
dσ ij = 0 ; f = 0 during naytral loading (C.14a-c)
∂σ ij
∂f
dσ ij > 0 ; f = 0 during loading
∂σ ij

The final assumption is that the so-called Drucker’s postulate is valid. This implies
that the yield surface is convex. Moreover, it means that the plastic strain increment
vector is directed along the outward normal to the yield surface. The normality
implication may be expressed as

∂f
dε ijp = dλ (C.15)
∂σ ij

where dλ is a nonnegative constant. Eq. (C.15) indicates that the loading function may
be taken as a plastic potential.

According to Eq. (C.14b), the function value of f remains unchanged, like zero, from
one plastic state to another.
12.109
Chapter 12 - Nonlinear Analysis

Since,

κ = κ (ε ijp ) (C.16)

the function f is dependent upon two sets of variables, the stresses and plastic strains

(
f = f σ ij , ε ijp ) (C.17)

Hence, Eq. (C.13) may be written as

∂f ∂f
df = dσ ij + p dε ijp = 0 (C.18)
∂σ ij ∂ε ij

By substituting Eq. (C.18) into Eq. (C.15) yields

∂f
dσ ij
∂σ ij
dλ = − (C.19)
∂f ∂f
∂ε mn ∂σ mn
p

From Eqs (C.2, C.8)

∂f ∂σ 3 sij
= = (C.20)
∂σ ij ∂σ ij 2 σ

and

∂f ∂ ∂
= ( )
σ − H (ε p ) = − p H (ε p ) ( )
∂ε ij
p
∂ε ij
p
∂ε ij
(C.21)
∂H ∂ε p ∂W p 1
=− p = − H ' σ ij
∂ε ∂W ∂ε ij
p p
σ

where the plastic work, Wp (Eq. (C.7)) has been utilized.

By noting

∂f ∂σ
=
∂σ ij ∂σ ij

and

σijsij = σijσij − 13 σijδijσ m = σijσij − 13 σ2m = 23 σ 2

12.110
Chapter 12 - Nonlinear Analysis

as well as using Eqs (C.20 – C.21) to rewrite the demoninator of Eq. (C.19), dλ may
be written as:

∂σ
dσ ij
∂σ ij dσ
dλ = − = (C.22)
⎛ σ ⎞ ⎛ 3 sij ⎞ H'
⎜⎜ − H ' ij ⎟⎟ ⎜⎜ ⎟⎟
⎝ σ ⎠⎝ 2 σ ⎠

Now, the derivative of the hardening parameter H with respect to plastic strain, ε p
can be obtained from uniaxial test. This is obtained from Eq. (10.94) by replacing σ
and εp with the equivalent quantities σ and ε p for multidimensional cases.

By combining Eqs (C.15, C.20, C.22) the following expressions for plastic strain
increments result:

3 sij
dεijp = dσ (C.23)
2 H 'σ

These expressions are called the Brandtl-Reuss equations for an isotropically


hardening material, obeying the von Mises yield criterion.

It is seen from Eq. (C.23) that the relative increments of yielding is determined by the
relative magnitude of total stresses. The stress increments only relate to the scalar
dσ . It may also be convenient to express the plastic strain increments by the stress
increments.

The differential dσ in Eq. (C.25) is then expressed as follows:

∂σ 3
dσ = dσ kl = s kl dσ kl (C.24)
∂σ kl 2σ

Then by inserting in Eq. (C.23)

dεijp = α ⋅ sijs kl dσ kl (C.25)

where

9
α=
4σ H ' (σ )
2

Besides the relation between plastic strain and (equivalent) stress increment it is also
of interest in finite element formulations to have the relationship between plastic strain
increments and the total strain increments, i.e. including elastic increments.

By reformulating Eq. (C.19) and introducing Hooke’s law:

12.111
Chapter 12 - Nonlinear Analysis

⎛ ∂f ∂f ⎞ ∂f
dλ ⎜ p ⎟ = − dσij
⎝ ∂ε mn ∂σ mn ⎠ ∂σij
∂f
= − (E ijkl dεekl ) (C.26)
∂σij
∂f ⎛ ∂f ⎞
= − E ijkl ⎜ dε kl − dλ ⎟
∂σij ⎝ ∂σ kl ⎠

Introducing the following relations:

Eijkl s kl = 2Gsij
(C.27)
Eijkl = E klij

and applying Eqs (C.20, C.21) gives

3Gs kl dε kl
dλ = (C.28)
σ (H '+3G )

By Eqs (C.15, C.20) the plastic strain increment now becomes

9Gsij s kl
dε ijp = dε kl (C.29)
2σ 2
(H '+3G )
This is the relationship between plastic and total strain.

Finally, the incremental stress-strain relationship than gets the following form:

dσ ij = E ijkl dε kle ( Hooke' s law)


(
= Eijkl dε kl − dε klp ) (C.30)
= Dijkl
ep
dε kl

where

9G 2 sij s kl
ep
Dijkl = Eijkl − (C.31)
σ 2 (H '+3G )

It is noted that Dijkl is symmetric

Dijkl = Dklij

because both terms in the expression ( C.31) are symmetric.

12.112
Chapter 12 - Nonlinear Analysis

C.4 Incremental stress-strain relationships for oriented bodies made of thin


plates

Beam

The following assumption are made:

Strains:

γ xy = dγ xy = 0 ; γ yz = dγ yz = 0
(C.32)
γ zx = dγ zx = 0

Stresses:

σ xy = dσ xy = 0 ; σ yz = dσ yz = 0
(C.33)
σ yy = dσ yy = 0 ; σ zz = dσ zz = 0

By taking into consideration these assumptions, the following stress-strain relation can
be found

⎛ dσ xx ⎞ ⎛ D xxxx D xxyy D xxzz ⎞ ⎛ dε xx ⎞


⎜ ⎟ ⎜ ⎟⎜ ⎟
⎜ dσ yy ⎟ = ⎜ D yyxx D yyyy D yyzz ⎟ ⎜ dε yy ⎟ (C.34)
⎜ dσ ⎟ ⎜ D D zzzz ⎟⎠ ⎜⎝ dε zz ⎟⎠
⎝ xx ⎠ ⎝ zzxx D zzyy

or

dσ ij = Dijkl
ep
dε kl (C.35)

or in matrix notation

dσ = Ddε (C.36)

Eq. (C.33) and Eq. (C.34) gives, when symmetry of stiffness matrix is taken into
consideration

dσ xx = (D xxxx −
D xxyy
(D D zzzz − D zzxx D yyzz )
D yyyy D zzzz − D yyzz
2 yyxx


D xxzz
(Dzzxx D yyyy − D yyxx D yyzz ) ) dε xx (C.36a)
D yyyy D zzzz − D yyzz
2

The deviatoric stress components expressed with total stresses become


s xx = 23 σ xx ; s yy = s zz = − 13 σ xx (C.37)

12.113
Chapter 12 - Nonlinear Analysis

Eq- (C.31) and Eq. (C.37) now give the following expressions for the stiffness
components of Eq. (C.34)

9G 2s 2xx E(1 − ν) 4G 2 σ2xx


D xxxx = E xxxx − = −
σ 2 (H '+ 3G) (1 + ν)(1 − 2ν ) σ 2 (H '+ 3G)
Eν 2G 2 σ 2xx
D xxyy = D yyxx = − 2
(1 + ν)(1 − 2ν) σ (H '+ 3G)
D xxzz = D zzxx = D xxyy (C.38)
E(1 − ν) Gσ 2 2
D yyyy = D zzzz = − xx
(1 + ν)(1 − 2ν) σ 2 (H '+ 3G)
Eν G 2 σ 2xx
D yyzz = D zzyy = − 2
(1 + ν)(1 − 2ν) σ (H '+ 3G)

By combination of Eq. (C.36) and Eq. (C.38) the stress-strain relation can be reduced
to an expression of the form

dσ xx = Ddε xx (C.39)

Plate

As is generally done for thin plate theory, the following case will be considered.

Strains:

γ zx = dγ zx = 0 ; γ yz = dγ yz = 0 (C.40)

Stresses:

σ zx = dσ zx = 0 ; σ yz = dσ yz = 0
(C.41)
σ zz = dσ zz = 0

The stress-strain relation can then be reduced to a 4x4 expression

⎛ dσ xx ⎞ ⎛ D xxxx D xxyy D xxxy D xxzz ⎞ ⎛ dε xx ⎞


⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜ dσ yy ⎟ ⎜ D yyxx D yyyy D yyxy D yyzz ⎟ ⎜ dε yy ⎟
⎜D (C.42)
⎜ dσ ⎟ D xyyy D xyxy D xyzz ⎟ ⎜ dε ⎟
⎜ xy ⎟ ⎜ xyxx ⎟ ⎜ xy ⎟
⎜ dσ ⎟ ⎜D D zzzz ⎟⎠ ⎜ dε ⎟
⎝ zz ⎠ ⎝ zzxx D zzyy D zzxy ⎝ zz ⎠

or
dσij = Dijkl dε kl + Dijzz dε zz (C.43)
and

dσ zz = D zzkl dε kl + D zzzz dε zz (C.44)

This equation system may be contracted to the following expression


12.114
Chapter 12 - Nonlinear Analysis

⎛ Dijzz D zzkl ⎞
dσ ij = ⎜⎜ Dijkl − ⎟⎟ dε kl (C.45)
⎝ D zzzz ⎠

It is seen that the system (10.111) is symmetric.

The deviatoric stress components for this two-dimensional case becomes

s xx = 13 ( 2σ xx − σ yy ) ; s yy = 13 ( 2σ yy − σ xx )
(C.46)
s xy = σ xy ; s zz = − 13 ( σ xx + σ yy ) =− σ m

In Eq. (10.108) the stiffness matrix elements are

9G 2s 2xx E(1 − ν) 9G 2s 2xx


D xxxx = E xxxx − = −
σ 2 (H '+ 3G) (1 + ν)(1 − 2ν ) σ 2 (H '+ 3G)
Eν 9G 2s xx s yy
D xxyy = D yyxx = − 2
(1 + ν)(1 − 2ν) σ (H '+ 3G)
Eν 9G 2s xx σ m
D xxzz = D zzxx = + 2
(1 + ν)(1 − 2ν) σ (H '+ 3G)
9G 2s xx σ xy
D xxxy = D xyxx = −
σ 2 (H '+ 3G)
E(1 − ν) 9G 2s 2yy
D yyyy = −
(1 + ν)(1 − 2ν) σ 2 (H '+ 3G)
Eν 9G 2s yy σ m
D yyzz = D zzyy = + 2
(1 + ν)(1 − 2ν) σ (H '+ 3G)
9G 2s yy σ xy
D yyxy = D xyyy = −
σ 2 (H '+ 3G)
9G 2 σ m σ xy
D zzxy = D xyzz =
σ 2 (H '+ 3G)
9G 2 σ2xy (C.47)
D xyxy = G −
σ 2 (H '+ 3G)
E(1 − ν) 9G 2 σ 2m
D zzzz = − 2
(1 + ν)(1 − 2ν ) σ (H '+ 3G)

For the equivalent stress σ the following expression comes out of Eq. (C.2)

σ = σ xx2 + σ yy2 − σ xxσ yy + 3σ xy2 (C.48)

and the equivalent plastic strain increment from Eq. ( C.6 ) becomes
dε p
= 4
3
(dε p2
xx
2
) 2
+ dε yyp + dε xxp dε yyp + 13 dγ xyp (C.49)

12.115
Chapter 12 - Nonlinear Analysis

References
Amdahl, J. and Kavlie, D. (1992). “Experimental and numerical simulation of double
hull stranding”, DNV-MIT Workshop on Mechanics of Ship Collision and Grounding,
Høvik, September.

Amdahl, J. and Stornes, A. (2001). “Energy dissipation in aluminium high-speed


vessels during collision and grounding”, Proceedings of ICCGS’01, Copenhagen, 203-
219.

ASIS (1993). The Conference on “Prediction Methodology of Tanker Structural


Failure & Consequential Oil Spill”, Association of Structural Improvement of
Shipbuilding Industry of Japan, III.39-III.47.

Bell, K. (1994). “Matrix Statics”, Tapir, Trondheim (in Norwegian).

Becker, A.A. (2001). “Understanding Non-linear Finite ElementAnalysis through


Illustrative Benchmarks”, NAFEMS, Glasgow.

BEFETS: “Blast and Fire Engineering for Topside Systems”, Phase 2, SCI Publication
No. 253, Ascot, UK, 1998.
Bergan, P.G. and Syvertsen, T.G. (1978). ”Buckling of Columns and Frames”, Tapir,
Trondheim (in Norwegian).

Bergan, P.G., Horrigmoe, G., Kråkeland, B. & Søreide, T.H., (1978), “Solution
techniques for non-linear finite element problems”, Int. J. Num. Meth. Engngn., 12,
1677-1696.

Cook, Robert D., Malkus, David S. and Plesha, Michael E. (1989) “Concepts and
Applications of Finite Element Analysis” John Wiley & Sons, ISBN 0-471-84788-7.

Crisfield, M.A. (1991). “Non-linear Finite Element Analysis of Solids and


Structures”, Vol. 1, J. Wiley & Sons, Chichester.

Cullen, The Hon. Lord, 1990. ‘The Public Inquiry into the Piper Alpha Disaster’,
London: HMSO.

Czujko, Jerzy (Ed.) (2001). “Design of Offshore Facilities to Resist Gas Explosion
Hazard.” Engineering handbook. First edition. CorrOcean ASA, ISBN 82-996080.

Den Heijer, C. & Rheinboldt, W.C. (1981), “On step-length algorithms for a class of
continuation methods”, SIAM J. Num. Analysis, 18, 925-948.

Emami Azadi, M., Moan, T. and Hellan, Ø., (1997). “Nonlinear Dynamic vs. static
Analysis of Jacket Systems for Ultimate Limit State Check”, proc. Int. Conf. on
Advanced in Marine Structures, DERA, Dunfirmline.

Engseth, A.G. (1985). “Finite element collapse analysis of tubular steel offshore
structures”, Div. of Marine Structures, Report UR-85-46, Norwegian Institute of
Technology, Trondheim, Norway.

12.116
Chapter 12 - Nonlinear Analysis

ISO 13819, (1994), ‘Petroleum and Natural Gas Industries – Offshore Structures –
Part 1: General Requirements’, (1994), ‘Part 2: Fixed Steel Structures’, (2001), Draft,.
Int. Standardization Organization, London.

Fried, I. (1984) ”Orthogonal trajectory accession on the nonlinear equilibrium curve”


J. Comp. Methods Appl. Mech. Engng. Vol.47, pp 283-297.

Haugen,B. (1994) ” Buckling and stability problems for thin shell structures using
high performance finite elements” Ph.D. thesis, Univ. of Colorado, Boulder.

Hellan, Ø. (1995) “Nonlinear pushover and cyclic analyses in ultimate state design and
reassessment of tubular steel Offshore Structures”, Dr.ing.thesis, Report MTA 1995:
108, Department of Marine Structures, Norwegian Institute of Technology, Norway.

Hellan, Ø., Moan, T. and Drange, S.O. (1994) “Use of Nonlinear Pushover Analysis in
Ultimate Limit State Design and Integrity Assessment of Jacket Structures” Proc.
Behaviour of Offshore Structures, BOSS, Boston, Massachusetts, vol. 3, pp. 323-345.

Horsnell, M.R. and Toolan, F.E., (1996), ‘Risk of Foundation Failure of Offshore
Jacket Piles’, Paper No. 28th Offshore Technology Conf.., Houston, 381-392.

Moan,T and Nordsve, N.T. (1979) “Numerical Prediction of Ultimate Behaviour of


Marine steel Structures”, Int. Symp. on Marine Technology, NTH, Trondheim.

Moan, T. and Amdahl, J. (2001). "Risk Analysis of FPSOs, with Emphasis on


Collision Risk", Report RD 2001-12, American Bureau of Shipping, Houston
(restricted).

Moan, T., Amdahl, J. Engseth, A.G., and Granli, T. (1985), , “Collapse behaviour of
trusswork steel platforms, “ Proc. Int. Conference on Behaviour of Offshore
Structures, BOSS, Delft, Holland.

Moan, T., Amdahl, Hellan, Ø. (2002), “Numerical Analysis for Ultimate and
Accidental Limit State Design and Requlification of Offshore Platforms“ invited
papers, Fifth World Congress on Computational Mechanics, Viena, Austria, ISBN 3-
9501554-0-6, http://www.wccm.tuwien.ac.at

NAFEMS (1992), “Introduction to Nonlinear Finite Element Analysis”, E.Hinton (ed.)


NAFEMS, Glasgow.

Nadim, F. and Dahlberg, R. (1996). ‘Numerical Modelling of Cyclic Pile Capacity in


Clay’, Proc. OTC Conf., 1, pp. 347-356.

Lehmann, E., Egge, E.D., Scharrer, M. and Zhang, L. (2001). ”Calculation of collisions
with the aid of linear FE models”, Proceedings of PRADS’01, Shanghai, 1293-1300.

Paik, J.K. and Thayamballi, A.K. (2002). “Ultimate limit state design of steel-plated
structures”, John Wiley & Sons, Chichester, U.K.

Paik, J.K. et al. (2003). “Report of ISSC Committee V.3, Collision and Grounding”,
Proc. 15th ISSC, SanDiego, Published by Elsvier, Amesterdam.
12.117
Chapter 12 - Nonlinear Analysis

Ramm, E. (1982) “The Riks/Wempner approach – an extension of the displacement


control method in non-linear analysis”, Non-linear Computational Mechanics, ed. E.
Hinton et al., Pineridge Swansea, pp. 63-86.

Ramm, E., (1981)“Strategies for tracing the non-linear response near limit-points,
Non-linear Finite Element Analysis in structural mechanics”, ed. W. Wunderlich,
Springer-Verlag, Berlin, pp 63-89.

Riks, E. (1972) “The application of Newton’s method to the problem of elastic


stability”, J. Appl. Mech. 39, 1060-6.

SCI(1993) “Interim Guideance Notes for the Design and Protection of Topside
Structures against Explosion and Fire”, Stell Construction Institute, Document SCI-P-
112/503, London.

Skallerud, B. and Amdahl, J. (2002) “Nonlinear analysis of offshore structures”,


Baldock: Research Studies Press, 2002

Simonsen, B.C. and Lauridsen, L.P. (2000). “Energy absorption and ductile fracture in
metal sheets under lateral indentation by a sphere”, International Journal of Impact
Engineering, 24, 1017-1039.

Søreide, T.H., Amdahl, J. Granli, T. and Astrup, O.C. (1986), “Collapse analysis of
framed offshore structures”, Paper OTC 5302 Offshore Technology Conference,
OTC, Houston, Texas.

Søreide, T.H., Amdahl, J., Eberg, E., Holmås, T. and Hellan, Ø. (1994), “USFOS – A
computer program for progressive collapse analyses of steel offshore structures;
Theory manual”, report STF71 F88038, 6th revision, SINTEF Structures and Concrete,
Trondheim, Norway.

Taby, J. and Moan, T. (1985) “Collapse and Residual Strength of Damaged Tubular
Members”, Proc. 4th In. Conf. on the Behaviour of Offshore Structures, BOSS Conf.,
Elsevier, Amsterdam.

Taylor, R. and Zienkiewicz, O.C. (2000), “The finite element method”,Vol. 1,2,3, 5th
Edition

Ultiguide (1999), “Best Practice Guidelines for Use of Nonlinear Analysis Methods in
Documentation of Ultimate Limit States for Jacket Type Offshore Structures", DNV-
SINTEF-BOMEL, Oslo - Trondheim - London.
Vinnem, J.E. (1999) “Offshore Risk Assessment”, Kluwer Academic Publishers,
Doordrecht.
WOAD: “Worldwide Offshore Accident Databank”, Det Norske Veritas, Oslo, 1996.
Wempner, G.A. (1971), “Discrete approximations related to nonlinear theories of
solids”, Int. J. Solids & Structs., 7, 1581-1599.

Wolf, J.P. (1994). ‘Foundation Vibration Analysis using Simple Physical Models’,
PTR Prentice Hall, Englewood Cliffs.
12.118

You might also like