An Approach to the Generalized Displacement Control Method

C. E. Magna
Department of Civil and Environmental Engineering, University of California, Davis

ABSTRACT

The generalized displacement control method is one of the strongest numerical procedures
for nonlinear finite element analysis. For this method, the solution procedure considers all degrees
of freedoms in the analysis and includes the selection of the load incremental parameter. In the
earlier works on this topic, two parameters to detect the change in direction of loading: the
generalized stiffness parameter and the incremental load parameter.

The motivation of this paper is to present an iterative method widely used in problems of
nonlinear finite element at structural analysis level.

INTRODUCTION

Under abnormal or ultimate loading conditions, linear structural analysis equations are
incapable of reflecting the real behavior of structures. In such cases, structures behave nonlinearly,
according to changes of their stiffness. Due to this problem, an iterative incremental numerical
technique must be used to map out the load-displacement behavior of the structural system (model).
Generally, these solution methods can be classified roughly as pure (linear) incremental, direct
iteration, and iterative incremental methods.

A widely used scheme to solve nonlinear problems is the incremental analysis. The
incremental analysis considered three configurations feature called C , C , and C [1, 2]. The initial
configuration C is known as the initial condition of the problem, as an example this configuration
maybe a structure not subject to any external load. The C configuration is known from solving a
non-linear model, where the solution of this model is one that brings the required equilibrium.
Finally, the configuration C is the current configuration, which is not known. The incremental
analysis is based on the fact that the incremental step from C to C is assumed small, so a number
of simplifications can be applied to reduce the complexity of a structural model to a simpler
problem.

Known the mathematical model that governs the nonlinear problem, the need to find methods
of solution which show what is happening in the incremental step from C to C . The purpose of
this term paper is summarized an efficient solution method to resolving problems of models that
have some kind of nonlinearities.

SOURCES OF NONLINEARITY

The behavior of structural systems that consider non-linearity (at geometry or material level)
is possible to identify a series of common phenomena, which reflect the state in which the model is
for that configuration.

Figure 1: General characteristics of a nonlinear problem.

As shown in Figure 1, in an equilibrium path can be areas of hardening (OA, DE) and
softening (AD). We also can find snap-back points, which represent the states from which the
system tends to dissipate energy. Precisely these points in the nonlinear behavior of a model are
those that generate numerical problems in the various solution methods. Thus, a good solution
method must be able to manage these characteristic points to follow the non-linear equilibrium path
given.

In addition to these points, Figure 1 also shows areas of stability, where the system takes
energy, and areas of instability, where the system supplies energy which it is incapable to receive in
its current configuration. Then, a solution method must also be able to detect these states in the
structure, so the applied load can increase or decrease according to the state. Other interesting
behaviors appreciated in Figure 1 are hardening and softening of the system. Note that these
phenomena may be caused by purely geometric nonlinearities, without considering the non-linearity
of the materials.

INCREMENTAL METHOD OF SOLUTION

Incremental solution methods are based on the fact that the incremental steps are "small".
Due to this assumption, we can approximate a nonlinear function within that increment by a linear
function, simple to solve. So if we want to know a given configuration, we can reach it by the
succession of several incremental steps, limiting the non-linear problem by a sequence of linear sub
models [2]. Using the convention that the right superscript denotes the incremental step in which the
analysis is, we present the non-linear problem of a model based on the equilibrium formulation:

P F  U (1)

Where P are the external forces applied on the model and F the resisting forces of the
model, which depend on the displacements U . Using Taylor expansion of first order around the
initial configuration, in the step i 1, the equilibrium problem in its linear form can be written as:

∂F (2)
P F U U U
∂U

Now, if we consider that in step i 1 the equilibrium is satisfied, we can replace the resisting
forces F for the external force P . Then, the expression changes to:

K · ∆U ∆P (3)

Where the increments of load and displacement are defined as:

∆U U U   (4)
∆P P P (5)

K is the stiffness associated to the model for the last configuration known in equilibrium.
Finally, if we want to know the displacement and applied loads on the model for configuration C ,
we simply add up the increments:

(6)
U ∆U

(7)
P ∆P  
 Therefore, it can be appreciate the fact that pure incremental methods solve the problem of
non-linearity whereas the model is linear in sections or increments. The advantage of these methods
is their simplicity; however the linear approximation assumed leads to errors within each increment,
which are added throughout the history of configurations. This error results in an imbalance
between the applied forces and the resisting forces of the model. Additionally, these errors are never
estimated during the process, but simply neglected. This lack of pure incremental method is what
gives rise to the incremental-iterative methods.

ITERATIVE INCREMENTAL METHODS

Incremental iterative methods are those that perform iterations within a single incremental
step to ensure its convergence, which is expressed through the equilibrium of external forces with
the resisting forces. The equilibrium is achieved by correcting the error (non- equilibrium of the
forces) obtained by using linear estimators. Typically, an incremental iterative method consists on
the three stages listed below:

• Prediction: From the incremental equation described in eq. (3), for an increment of the
applied external force on the model, the displacement increments are predicted.
• Correction: Knowing the displacements it is possible to know the resisting forces. Then,
by assembling these forces on each increment, it is possible to know the resisting forces
present in the model.
• Equilibrium: At this stage we review the equilibrium between the applied forces and the
resisting forces of the model. The error obtained by setting the equilibrium is known as
the imbalance of forces. Using a convergence criterion is determined whether the
imbalance is small enough to advance another incremental step or if it is large enough to
make another iteration.
When any iterative incremental method is developed, we face two typical questions: (a) How
is determined the external load increment for each step? (b) Based on what we carry out the
iterations within an increment? The first question refers to the fact that an efficient algorithm is
capable to adapt to the condition of the given model, varying the load increment according to the
degree of nonlinearity of the model. The second question focuses on determining the convergence
characteristics that the iterative method should have and the way that the method manages the
points detailed Figure 1. These are two very different questions, but both find their answer on the
load parameter definition, which will be explained later.
 The following is the general formulation of an iterative incremental method. First, it is
important to explain the notation used in this formulation. A superscript on the right side of a
quantity will denote the incremental step to which we refer. A subscript on the right side of a
quantity will indicate the iteration to which we refer, within an incremental step.

For example the term:

will represent the external force applied to the model that exists in the incremental step i,
during iteration j. Using the notation described above, we present the linear approximation for the
incremental step i in the iteration j.

K · ∆U P F (8)

Where the initial conditions for the first iteration will be those given by the last iteration l of
the previous incremental step, such that:

K K

F F

U U

The external force applied on the model can be written as:

P P ∆P

Or expressed in terms of the load parameter, λ :

P P λ ·P (9)

This load parameter means which is the incremental load factor applied in iteration j of the
incremental step i, and P is the load reference vector. From expression (8), we solve ∆U and the
total displacement for that iteration is:

U U ∆U (10)

On the other hand, defining the imbalance of forces at iteration j 1 as:

R P F

The expression (8) can be rewritten as:

 K · ∆U λ ·P R (11)

And the above expression can be separated into two expressions:

K · ∆U P (12)

K · ∆U R (13)

Thus, solving the previous system, the displacement increments for iteration j will be:

∆U λ · ∆U ∆U (14)

Then, what characterizes the different iterative incremental methods is how is defined the
load parameter λ . It is now important to define a convergence criterion. This may be taken as the
norm of the vector of unbalance, which must be less than or equal to a certain tolerance.
T
R ·R tol.

Special remark: The Newton-Raphson Method.

One of the most powerful procedures for the solution of nonlinear problems is the Newton-
Raphson method, which is a load control method. Generally, the majority of the iterative algorithms
adapted to solve nonlinear equations may be considered as a variation of this technique.

The first iteration in this method is identical to the linear incremental method except at the
end of the first iteration, member forces are calculated and transformed into the global coordinates.
Then, unbalanced load vector between the applied external load and the internal nodal forces are
determined. Subsequent iterations are employed until a predefined convergence criterion is
satisfied. Generally, the Newton-Raphson technique uses two kinds of convergence criteria: force
based and displacement based [3]. A simple example can be seen in Figure 2.

Figure 2: Newton-Raphson solution scheme [5]

 Even though this method is one of the most used, it fails to surpass limit points (Figure1),
because the limit points have a tangent parallel or nearly parallel to the x-axis, then the method does
not converge [4, 5].

GENERALIZED METHOD OF DISPLACEMENT CONTROL

This method was proposed by Yang and Shieh in 1990, and it depicts a good solution for
nonlinear problems because it presents a great stability in the vicinity of the critical points of a
given model (snap-back points and boundary points). Additionally it presents the adaptability of the
load parameter, which reflects variations in the stiffness, and the ability to determine the direction
of load automatically.

The load parameter for the generalized method of displacement control is defined by the
concept of the generalized stiffness parameter (GSP). This is defined as:
T
∆U ∆U (15)
GSP T
∆U ∆U

The GSP parameter represents the stiffness of the structure. GSP starts with unity and
becomes smaller when the structure softens, and becomes greater when the structure hardens. It
turns out to be negative only for the incremental steps immediately after the limit points, while
remaining always positive for the other incremental steps.

Then, for the first iteration (j 1) within an incremental step load parameter is defined as:

.
λ  λ · |GSP| (16)

In the above expression, the final sign will depend on the sign of GSP and the sign of initial
load parameter of the previous incremental step. For the remaining iterations (j 2), the load
parameter is defined as:
T
∆U ∆U (17)
λ T
∆U ∆U
ALGORITHM FOR THE GENERALIZED METHOD OF DISPLACEMENT CONTROL

In this section the implementation of this method is present. First, the cumulative load
parameter, Λ , is defined as:

Λ Λ λ (18)

The algorithm follows as:

1. Select a reference load vector and a value for the first load parameter . Initialize the
displacement, applied forces and cumulative load parameter as value zero, that is to say:
0, 0 ,Λ 0.
2. Fort the first iteration (j 1) at any incremental step i:
• Compute the stiffness matrix of the model .
• Solve the equilibrium equation (12) and determine ∆ . For 1, take 1.
For 2, determine according to the expression (15).
• For 2, calculate with the equation (16). The sign of will be equal to the sign
of due to the sign of .
• Compute the displacement increments ∆ according to the equation (14). Notice
that ∆ 0 , because for the first iteration we have zero unbalance.
3. For the following iterations (j 2):
• Compute the stiffness matrix K .
• Solve the incremental displacement components from the expressions (12) and (13).
• Determine the load parameter λ from the expression (17).

• Calculate the displacement increments ∆U according to the expression (14).

4. Determine the applied external loads P according (9) and the total displacements U with
equation (10).
5. From U , calculate the resisting force of the model F .

6. Calculate the imbalance R as: R P F.

T
7. Check the convergence criterion adopted, for example: R ·R tol.
If it is true, move to a new incremental step. If it is not satisfied, make a new iteration.

Examples of the utilization of this method can be found in the literature of non-linear finite
element. To examine some of these examples, please see references [1] and [2].
CONCLUDING REMARKS

In a non linear finite element problem, special attention should be paid to the predictor stage
dealing with the solution of the structural displacements from the structural equilibrium equations.

The accuracy of an iterative incremental method is determined primarily by the equations for
recovering the element forces used in the corrector stage. To have a good representation of the
problem, the solution scheme should have the following characteristics: to be numerically stable in
the limit points; to have the ability of self-adjusting when the load increments are determined, this
will show the variation in structural stiffness; and to have an automatic reversing in the direction of
loading when a limit point is passed. The general displacement control method satisfies these three
requirements through the GSP parameter, which works as an indicator for guiding the iterations.

As it was mentioned above, the generalized method of displacement control has following
advantages, especially with respect to the Newton-Raphson method: (a) the numerical stability is
assured at the limit points, in these points the load increment parameter λ is not constant in the
iterative process; (b) the variation of the stiffness is taken into account in determining the load
increments via introduction of the GSP parameter; (c) if a change of sign in GSP is detected, the
direction of loading is automatically inverted.

REFERENCES

1. Yang, Y. B. , Leu, L. J. and Yang, Judy P.(2007), Key Considerations in Tracing the
Postbuckling Response of Structures with Multi Winding Loops, Mechanics of Advanced
Materials and Structures, 14: 3, 175 – 189.
2. Yang Y. B., Kuo S. R. (1994), Theory and Analysis of Nonlinear Framed Structures,
Prentice Hall, Singapore.
3. Morteza A.M., Sonmez M. (2008), Solution Techniques for Nonlinear Equilibrium
Equations, ASCE Structures Congress, Vancouver, Canada.
4. Wikipedia: http://en.wikipedia.org/wiki/Newton%27s_method
5. Plus magazine: http://plus.maths.org/issue9/puzzle/solution.html