XII Nonlinear Analysis

This chapter includes material from the book “Practical Finite Element Analysis”. It also has been
reviewed and has additional material added by Matthias Goelke.

18.1 Introduction
In this chapter, we discuss the practical aspects of nonlinear static Finite Element Analysis.

But how do we know that our problem is nonlinear? The best way is to look at the load-displacement
respose of one or more characteristic load introduction points. As discussed in an earlier chapter on
Types of Analysis, when the structural response (deformation, stress and strain) is linearly proportional
to the magnitude of the load (force, pressure, moment, torque, temperature etc.), then the analysis
of such a structure is known as linear analysis. When the load to response relationship is not linearly
proportional, then the analysis falls under nonlinear analysis (see figure below). For example, when a
compact structure made of stiff metal is subjected to a load relatively lower in magnitude as compared
to the strength of the material, the deformation in the structure will be linearly proportional to the
load and the structure is known to have been subjected to linear static deformation. But most of
the time either the material behavior is not linear in the operating conditions or the geometry of the
structure itself keeps it from responding linearly. Due to the cost or weight advantage of nonmetals
(polymers, woods, composites etc.) over metals, nonmetals are replacing metals for a variety of
applications. These applications have nonlinear load to response characteristics, even under mild
loading conditions. Also the structures are optimized to make most of its strength, pushing the load
level so close to the strength of the material, that it starts behaving nonlinearly. In order to accurately
predict the strength of the structures in these circumstances, it is necessary to perform a nonlinear
analysis.

Linear Response:
P1/d1 = P2/d2
P2
Nonlinear Response:
P2 P1/d1 ≠ P2/d2

P1 P1
Load

Load

d1 d2 d1 d2

Comparison of Linear and Nonlinear Response

As discussed in the earlier chapter, the stiffness matrix relating to the load and response is assumed to
be constant for static analysis; however, all the real world structures behave nonlinearly. The stiffness
matrix consists of geometric parameters such as length, cross sectional area and moment of inertia,
etc. and material properties such as elastic modulus, rigidity modulus etc. The static analysis assumes
that these parameters do not change when the structure is loaded. On the other hand, nonlinear static
analysis takes into account the changes in these parameters as the load is applied to the structure.
These changes are accommodated in the analysis by rebuilding the stiffness matrix with respect to
the deformed structure (i.e. altered properties) after each incremental load application. Although,
the world is nonlinear, it should be mentioned that in many cases the assumption of linearity is valid
and in that way a linear analysis can be done instead. Also, from a computation point of view, it is a
much less expensive approach.

18.2 Comparison of Linear and Nonlinear FEA

Sr. No. Characteristic Feature Linear Problems Nonlinear Problems
1. Load - Displacement Relation Displacements vary linearly It is nonlinear. Thus stiffness
with applied loads. Thus varies as a function of load.
Stiffness is constant . Changes in Displacements can be very large
geometry due to displacement and hence an updated reference
are assumed to be small and state called as “Updated
hence are ignored. Original or Lagrangian” approach is used
undeformed state is always used sometimes .
as a reference state.
2. Stress-Strain Relation Linear up to the proportional It is non-linear function of
/ elastic limit . Properties such stress-strain and time. These are
as Young’s Modulus are easily difficult to obtain and requires
available. lot of additional experimental
material testing.
3. Scalability Applicable . That is if a 1 N force Not applicable.
causes x units of displacement
, then 10 N magnitude of force
will cause 10x displacement
4. Stress and Strain Measures Engineering or Nominal Stress Can be several such as :
–strain . 1. True strain / Natural Strain
(logarithmic) – True / Cauchy
Stress.
2. Green / Almansi strain and
Kirchoff – Piola Stress.
5. Superposition Applicable. That is a Not applicable .
combination of load cases is
possible.
6. Reversibility The behaviour of the structure The final state after removal of
is fully reversible upon the loads is different from the initial
removal of external loads. state. Due to this superposition
This also means that loading of load cases is not possible.
sequence is not important and Load history is very important.
the final state is unaffected by
the load history.
7. Solution scheme The load is applied in one step Load is split into small
with no iterations (Except for increments with iterations
solvers like ABAQUS ) performed to ensure that
equilibrium is satisfied at every
load increment.
8. Computational Time Small Large
9. User’s Interaction with the Least required Requires lot of monitoring
software as the software may fail to
converge sometimes.
18.3 Types of Nonlinearity
There are three basic source of nonlinearity; Geometric Nonlinearity, Material Nonlinearity, and
Contact Nonlinearity.

1) Geometric Nonlinearities

Geometric nonlinearity may be related to 1) Large strain 2) Large rotation, and 3) Large deformation

Geometric nonlinearity takes into account that the geometry cross section may change as a result of
large deformation (in linear static analysis the cross section is considered to be constant). Thus, the
equilibrium equations must be written with respect to the deformed structural geometry. Additionally,
applied loads may change their direction as they increase, as when pressure inflates a membrane (Ref.
Concepts and applications of finite element analysis; R.D. Cook et. al, p. 595).

Linear Elastic

Non linear Material (small displacements)

Large displacements and rotations (small strain; linear or nonlinear material)

Large displacements, rotations, and large strain (linear or nonlinear material)

Contact

Images after K.J. Bathe, Finite Elemente Methoden

Mathematically this means that we are considering the higher order derivatives in the displacement
-strain relationships as follows:

εx = ∂u/∂x + ½. [ (∂u/∂x)2 + (∂v/∂x)2 + (∂w/∂x)2]

εy = ∂v/∂y + ½. [ (∂u/∂y)2 + (∂v/∂y)2 + (∂w/∂y)2]
εz = ∂w/∂z + ½. [ (∂u/∂z)2 + (∂v/∂z)2 + (∂w/∂z)2]
εxy = ½. [ (∂u/∂y) + (∂v/∂x) + (∂u/∂x)(∂u/∂y) + (∂v/∂x) (∂v/∂y) + (∂w/∂x)(∂w/∂y)]
εyz = ½. [ (∂v/∂z) + (∂w/∂y) + (∂u/∂y)(∂u/∂z) + (∂v/∂y) (∂v/∂z) + (∂w/∂y)(∂w/∂z)]
εxz = ½. [ (∂w/∂z) + (∂w/∂x) + (∂u/∂z)(∂u/∂x) + (∂v/∂z) (∂v/∂x) + (∂w/∂x)(∂w/∂z)]

In Linear FEA, we consider only the first order terms. Another distinguishing feature of geometric
nonlinearity is that we form the equilibrium equations with respect to the deformed geometry, which
is not known in advance . The large displacement problem may be associated with a small strain or
large strain, and the displacement strain relationship may or may not be linear.

Example 1 : A large cantilever beam subjected to an end moment rolling up into a circle. This is a
very standard test case for verifying almost every nonlinear FEA Code.

Before deformation

After deformation

A Large cantilever performing a roll up maneuver.

Rotation at free end is 2/ radians or 360 degrees .

Example 2 : An adhesive bonded joint between two plates subjected to shear loading. At large
shear loads, the joint undergoes a very large rotation in the middle portion.

Adhesive bonded joint undergoing large rotation .

It can be observed that large rotation is a typical characteristic of geometric nonlinearity.

2) Material Nonlinearities
 Elastic –plastic

Stress
Elastic-Perfectly plastic

Strain

Nonlinear Hyperelastic
Elastic

Stress

Strain
20 %
Features of Material Nonlinearity

All engineering materials are inherently nonlinear as it is not feasible to characterize a nonlinear
material by a single constitutive law for the entire range of environmental conditions such as loading,
temperature and rate of deformation. We can idealize or simplify the material behavior to account
for only certain effects which are important for the analysis. The linear elastic (also called Hookean)
material assumption is the simplest of all. The material is nonlinear elastic if the deformation is
recoverable and plastic if it is irrecoverable. If the temperature effects on the material properties are
important, then the coupling between the mechanical and thermal behavior should be properly taken
into consideration through thermo-elasticity and thermo-plasticity. If the strain rate has significant
effects on the material, we have to consider the theories of visco-elasticity and visco-plasticity. An
example of material nonlinear behavior has been given in the above figures.

A Brief classification can be given as follows:

1. Nonlinear Elastic
2. Hyperelastic
3. Elastic-Perfectly Plastic
4. Elastic-Time independent plastic
5. Time dependent plastic ( Creep )
6. Strain rate dependent elasticity –plasticity
7. Temperature dependent elasticity and plasticity

If we observe the stress-strain curves in the above figures, then material nonlinearity can be classified
further as follows .
1. Linear Elastic-Perfectly Plastic
2. Linear Elastic- Plastic. The plastic part in the stress-strain curve is time independent and can be
analysed into two main types :
a. Elastic – Piecewise linear plastic. ( as shown in following figures)
b. Elastic – actual stress strain curve information.

σ
σ2
E3
Stress

Stress
E2
σ

E1
E1

Strain Strain
(a) Elastic Material (b) Elastoplastic Material
Stress Strain Curve for Material with Linear and Nonlinear Properties

3. Nonlinear elastic model characterizing materials with no fixed definition of yield point such as
say plastic but the strain still limiting well below say 20 % .

4. Hyperelastic materials such as rubber undergoing very large displacements. A typical application
is a gasket material .

3) Boundary Nonlinearities/Contact Nonlinearity

Boundary nonlinearity arises when boundary conditions in a FE model change during the course of
the analysis. The boundary conditions could be added to or removed from the model due to boundary
nonlinearity as the analysis progresses. This kind of nonlinearity typically involves contact sets in the
model which could get engaged or disengaged as a response to applied loads. The load transfer
mechanism via a contact pair is a complicated phenomenon. Researchers have developed several
theories to describe the load transfer using contact sets. A typical contact algorithm in FE codes works
as follows:

First, check for either open or closed contact surfaces. This is done by measuring the distances
between the nodes on the surfaces forming the contact pairs. If a contact is closed, then apply reaction
forces to the nodes in contact to keep them from penetrating through each other. If a contact is
open, then no load is transferred between those nodes. The reaction forces applied to the nodes are
calculated based on stiffness of the materials of each of the surfaces in the contact set. Since load
transfer through a contact set is not constant and it requires continuous check to determine the open
or closed status of the contact, this analysis can not be handled by linear static analysis codes. More
accurate results can be obtained by checking the open or closed status of the nodes in the contact set
more frequently.
18.4 Stress-Strain Measures for Nonlinear Analysis
This is a major distinguishing factor between linear static and nonlinear FEA. In linear analysis we
always use the engineering stress – strain definitions , whereas in nonlinear analysis typically the true
stress and train is used. An excellent summary about stress/strain relationships is also available at
http://dolbow.cee.duke.edu/TENSILE/tutorial/node3.html.

We will examine these various stress and strain definitions through a simple one-dimensional example.

l
l0 ∆l

A one dimensional rod undergoing axial deformation

1) Engineering strain and engineering stress.

Engineering strain is a small strain measure, which is computed using the original geometry. The
engineering strain measure is a linear measure since it depends on the original geometry, (i.e. length)
which is known beforehand. It is limited to small rotations of the material because a moderate rigid
body rotation will result in non-zero strains.
∆l
εx = —
l0

Engineering stress (σ), is the conjugate stress measure to engineering strain (ε). It uses the current
force F and the original area A0 in its computation.
F
σ= —
A0

2) Logarithmic strain and true stress.

Logarithmic strain / natural strain / true strain is a large strain measure, which is computed as
l
dl £ l ¥
εl = ∫ — = Ln ¤ —
l l0 l0 ¦
0

This measure is a nonlinear strain measure since it is a nonlinear function of the unknown final
length. It is also referred to as the log strain (or true strain) and represents an additive strain measure
as compared to linear strain. Let us consider a bar with say initial length of 1m undergoing
deformation in 3 steps as follows :

Step 1 : deformation from 1 m to 1.2 m

Step 2 : deformation from 1.2 to 1.5 m
Step 3 : deformation from 1.5 to 2 m

In the following table we compare the engineering strains against true strains and one can clearly
see that the additivity is retained in only true strain and hence this one should be used in
nonlinear analysis.
 Step Engineering Strain True Strain
1 0.2 Log (1.2/1) = 0.1823
2 0.3/1.2=0.25 Log (1.5/1.2) = 0.2231
3 0.5/1.5=0.33 Log (2/1.5) = 0.2876
Addition of strains in 0.78 0.6931
all the steps
Total : From 1 to 3 (2.0 – 1.0 )/1.0 = 1.0 Log(2/1) = 0.6931
Conclusion Engg. Strain is not additive and doesn’t True Strain is additive and thus
preserve the history of deformation preserves the history of deformation
correctly . correctly.

True stress (τ) is the conjugate 1-D stress measure to the log strain (εl ), which is computed by dividing
the force F by the current (or deformed) area A. This measure is also commonly referred to as the
Cauchy stress.
F
τ= —
A

3) Green-Lagrange strain and 2nd Piola-Kirchoff stress.

Green-Lagrange strain is another large strain measure, which is computed in 1-D as

1 £ l2 - l02 ¥
εG = — ————
2 ¤ l02 ¦

This measure is nonlinear because it depends on the square of the updated length l , which is an
unknown. A computational advantage of this strain measure, over the log or Hencky strain, is that it
automatically accommodates arbitrarily large rotations in large strain problems. The conjugate stress
measure for the Green-Lagrange strain , is the 2nd Piola-Kirchoff stress (S). It is computed in 1-D by
l0 F
s= — —
l A0

It should be noted that this stress has little physical meaning.

18.5 Solution Techniques for Nonlinear Analysis

This image illustrates the FE Model of a SUV tire and

consists of over 300,000 Hybrid Linear Brick/Prism
Elements with 1.2 million DOF. Tire simulation comprises
all kinds of non-lineralty: material, large deformation
and contact, which demands a well structured FE Model
with optimum number of elements. The contour plot is
of the Footprint Pressure distribution of the tire when
subjected to vehicle load is also shown.

Non Linear Analysis Application – Tire Analysis

( Image Source : Altair Calendar 2008,
Courtesy : Apollo Tyres Ltd.)

As we learned earlier, the stiffness matrix in a nonlinear static analysis needs to be updated as the
analysis progresses. How is this accomplished by an FEA code? This will be described next.

When any structure is subjected with an external load, internal loads are generated within the structure
to react the load. These internal loads are caused by stresses inside the structure. For the structure
to be in equilibrium, each node in the FE model representing the structure should be in equilibrium.
The condition of equilibrium is checked by determining the residual load, which is defined as the
difference between the external and internal load. For the structure to be in perfect equilibrium, the
residual load must be zero, which is a very unlikely situation in nonlinear analysis. The FEA codes
assume that the nodes are at equilibrium if the residual load is negligibly small. The FEA codes provide
a user interface to be able to select the acceptable residual load criteria. Let’s understand how FEA
codes iterate to achieve an equilibrium condition at a node.

Let’s consider point 1 in the load-displacement curve for a node in a FE model, which is in equilibrium;
meaning the difference between external load (P1) and internal load (I1)is zero. When the external
load is increased by small amount to P2, the incremental displacement using linear elastic theory is:

d2-d1 = k1 (P2-P1)

Where k1 is the stiffness based on the configuration at the point. The FEA code now calculates the
internal load at point 2, using the configuration properties at point 2 and compares the residual load
at point 2, which is given by:

R2 = P2-I2
 k1 k2
3’ k3
2

3’
2’
P

d
The Newton Raphson procedure

If R2 is more than the acceptable residual load, the FEA code locates a new point 3 using the linear
relationship:

d3-d2 = k2 (P2-P1)

Then a new stiffness matrix, k3, is formed based on the configuration at point 3 and the corresponding
internal load, I3, is calculated. The residual load at point 3 becomes:

R3 = P2-I3

If R3 meets the residual load criteria, the solution is said to have converged for that particular increment
and each attempt of calculating the residual load for an increment is called iteration. Each iteration
in a nonlinear static analysis process involves the same amount of calculation as in a static analysis.
Thus it becomes quite obvious that a nonlinear static analysis takes a lot more computational time
than a static analysis. Once the solution is converged for a load increment, the FEA code accepts it
as equilibrium and increases the load by further increments. This process is repeated until the entire
load is applied to the structure. The procedure given here is a stiffness update every step and is called
Newton Raphson method (see http://en.wikipedia.org/wiki/Newton%27s_method). Another
common option is to update the stiffness after a number of steps and this is called Modified Newton
Raphson method.

In the earlier sections we realized that the geometric nonlinearity can be accounted in the analysis
calculations by applying the load in several steps. This is done in such a way that the structure
geometry is updated to take the deformed shape (due to prior load) before applying the next step of
the load. In the cantilever beam example, the accuracy of the results can be improved by applying the
load in more than two steps. Similarly, material nonlinearity can be accounted for by representing the
stress-strain curves by many small straight segments. The more segments in the stress-strain curve,
the more accurate will be the results.
 Acceptable Result at
Reasonable Cost

Theoretical Result

Results
Computationally
Expensive

Model Refinement
Cost of Accuracy

If a load is applied in too many steps or the material stress-strain curves are represented by too
many segments, the computational cost will be high. As the number of load steps in a geometric
nonlinearity solution and the number of stress-strain segments in a material nonlinearity solution
tends to reach infinite, the FEA results will approach towards the theoretical results. After a certain
point, the significant improvement is not observed with a further increase in load steps or with added
data points in stress-strain curve. This threshold point can be determined by running several runs on
the same model with increasing the number of load steps or the number of stress-strain data points
and comparing the improvement in results.

FE analysts control the number of steps in a nonlinear static analysis by specifying maximum increment
(% of total load) in the nonlinear parameter card. The solver never exceeds the specified maximum
increment during the analysis. FE analyst are also required to specify the minimum step, which signals
termination of analysis if the step needs to be smaller than minimum increment for continuation of
the analysis.

18.6 Issues Related to the Convergence of Newton Raphson Method

A sudden change in stiffness or a sudden change in load can cause convergence difficulty. To settle
this issue, it is always better to break sudden changes down into a series of many small incremental
changes i.e. we must use ramped loading, small time steps. In the case of the Newton-Raphson
method, it will converge if the starting configuration is inside the radius of the convergence circle. If
Xstart is inside the radius of the convergence, the solution will converge; if it is outside, the solution
diverges as shown in the figures below.

Load Load
Diverging
Converged

F F

xstart x Displacement xstartx Displacement

Problems with convergence of Newton Raphson Method.

If your starting configuration is outside the radius of convergence, two techniques can help you obtain
a converged solution as :

Load Load

F F

u u
ustart ustart
(a) (b)
Remedies to obtain convergence

a. Apply the load incrementally to move the target closer to the start as shown in (a).

b. Use convergence-enhancement tools to enlarge the radius of convergence as shown in (b).

Experienced users typically combine both strategies to obtain convergence.

18.7 Essential Steps to Start with Nonlinear FEA

1. Learn first how the software works on a simple model before you use a nonlinear feature
which you haven’t used . Also guess how your structural component will behave, i.e. check
for available studies, reports and benchmarks .

2. Try to understand the software’s supporting documentation, its output and warnings.

3. Know what you are looking for. Prepare a list of questions you think your analysis should
be able to answer. Design the analysis, including the model, material model, and boundary
conditions, in order to answer the questions you have in mind.

4. Keep the final model as simple as possible. A linear analysis done first can provide a lot of
information such as where are the high stresses in the model, where the initial contact may
occur, and what level of load will introduce plasticity in the model. The results of the linear
analysis may even point out that there is no need for a nonlinear analysis. Examples of such a
situation include the yield limit not being reached, there is no contact, and the displacements
are small.

5. Verify and validate the results of the nonlinear FEA solution. Verification means that “the
model is computed correctly” from the numerical point of view. Wrong discretization with
respect to the mesh size and time stepping are common errors. Validation asks the questions
if “the correct model” is computed e.g. the geometry, material, boundary conditions,
interactions etc coincide with the one acting in reality.

6. Try to look into the assumptions made with respect to the structural component, its
geometry behavior with respect to large strain (On/Off ), look into different material models
if the earlier model is unable to give you a result you expect (sometimes software only
make some models compatible with commonly used elements and in this case you might