FEA Modeling of Geometric Nonlinearities in Cantilever Beams

Jordan Radas
Kantaphat Sirison
Wendy Zhao

Abstract

The design and optimization of a geometrically nonlinear transversely loaded cantilever beam
are explored using ANSYS to better understand the theory behind large displacements. Finite
element formulation of the beam will highlight the development of the strain dependent
geometric matrix through derivations from beam kinematics. Both iterative methods and ANSYS
features for solving non-linear problems are discussed.

To investigate the effect of geometric non-linearity, an AL-2024 cantilever beam with a length of
96 inches, width of 19.625 inches and height of 1.625 inches is transversely loaded on the tip
with a force of 2500 lbs. For all mesh sizes, the tip displacement of ANSYSs non-linear solver
yields less displacement than that of the linear solver. This discrepancy resulted from the extra
geometric stiffness taken into account in the non-linear assumption. Lastly, small holes are
introduced within the beam to minimize materials usage. The beam is subject to a yield
constraint of 50 ksi. Through using Goal Driven Optimization method and non-linear
assumption, the beams volume was reduced from 3061.5 in3 to 2719.5 in3.

The analysis performed serves as a proof of concept that real-world situations involving large
structural displacements, such as diving boards and airplane wings, benefit from nonlinear
modeling considerations despite larger computing time.

Beam Kinematics

For three-dimensional solids, a one dimensional beam model simplification is used to describe
the behavior of individual cross sections in terms of strains and deformations evaluated on the
longitudal axis. This large deformation beam theory assumes that plane cross-sections remain
plane, cross-sections remain constant, and perfectly linear elastic material. Hence, the only
source of nonlinearity will be due to geometry

With small strain and large deformation, the total lagrangian finite element formulation will be
used. The beam element consists of 2 nodes. The 3 degrees of freedom at each node are
ux (x),uy (x) and (x) .

The motion of the particle, and consequentially the element, needs to be determined before
formulating
the strain displacement matrix, which is then used to obtain the element stiffness

matrix. First, the displacement field, which relates the position of the particle after deformation (
x,y ) to its position in the initial configuration (X,Y), is obtained for the beam element. This is
shown in Figure 1 in Appendix A. Then, the deformation gradient matrix of the beam is
calculated by differentiating the displacement field with respect to the X and Y referenced
coordinate/vector. Finally, the Green LaGrange strain tensor, which gives the square of the local

 change in distance, is calculated by matrix multiplication with the deformation gradient. These
steps are shown explicitly for the beam element in Appendix A Figure 2. The Green LaGrange
strain tensor is different from the Cauchy strain tensor because it does not assume infinitesimally
small strain. Hence, the formulation of the derivatives of the displacement field is different.

The green LaGrange strain is then characterized into the axial, shear and rotational components
as follows. This will be the basis of the strain matrix
e (1 u'x )cos u'y sin 1

1(1 u'x )sin u'y cos



'

The [B] matrix, which relates nodal displacement to the strain, can be calculated by
differentiating , and with respect to ux1, uy1 2. Before doing so, heavy trigonometry is
used to convert ux, ux,
and in terms of the nodal x,y and displacement. The result is:

cos a (1N1u'ax )sin

sin cos u'ysin N 2 a
cos

B sin b (1Nu'
cos sin
1b )cos u'cos N 2b
sin
Where
x y

0 0 1 0 0 1

Discretization of the linearization of the Weak Form

The principle of virtual work, using the Total Lagrangian formulation, is shown below. The right
side represents the internal work, described by the product of the Green Lagrange strain and PK2
stress, is balance by the external work from transverse loading
(EA GA EI )dx u q y
V0

Discretizing the weak from above using the strain displacement interpolation, which is the [B]
matrix above, yields the equation Ku = f. Where

k B 0
T
DBdx kgeometr ic kmater ial kgeometr ic
V

The material stiffness matrix takes the form below. The interpolation highlighted in yellow
relates shear and rotational strain to nodal displacement, and is exactly the same form as derived
in class. The axial strain interpolation also takes the exact same form as the one derived for an
axially loaded bar.
The need for the geometric stiffness matrix arises from the change in element stiffness due to
large nodal displacements. The geometric stiffness matrix is composed of two parts, the left from
axial deformation and the right from shear.

In contrast, for linear beam formulation, it is assumed that the axial and shear strain is small,
allowing the geometric stiffness to equal zero.

Newton Raphson Method

The presence of the geometric matrix means the total (tangential) stiffness matrix is dependent
on displacement. It is therefore impossible to analytically solve for the exact nodal displacement.
An iterative method as shown below is needed where u is the displacement increment, R is
the external loading and F is the nodal forces corresponding to the internal stress,

The final solution is in equilibrium, meaning the external loading and nodal forces are within
some predetermined tolerance. In cases where the tangent stiffness matrix in the ith iteration is
close to zero, the solution may take a long time to converge.

Substeps in ANSYS refer to the application of load in increments to obtain a solution within each
load step, which are adjusted to account for abrupt changes in loading on a structure. In each
substep, equilibrium iterations performed until convergence. There is an opportunity cost of
accuracy versus time when considering the ideal number of substeps. To help, ANSYS has an
automatic time stepping feature that chooses the size and number of substeps to optimize, as well
as a bisections method that activates to restart solution from last converged step if a solution does
not converge within a substep. Further analysis of various ANSYS feature available to assist in
aiding convergence can be found in Appendix B.
The restoring load F corresponds to the element internal load of the current stress state, and the
stresses of the element depend in turn on the large strains of the geometric nonlinearities. By
using the right polar decomposition theorem, the restoring load is found to be the product of the
rotation and shape change matrices. The logarithmic strain of an element can then be found by
taking the natural log of the shape change matrix. Using spectral decomposition on the shape
change matrix, the logarithmic strain can then be found to be the sum of the natural log of the
eigenvalue, or principal stretches, multiplied by the eigenvectors and the transpose of the
eigenvectors, or principal directions

ln[U ]
3
ln i ei ei T
i 1

ANSYS evaluates the restoring load by using an incremental approximation of the strain. This
increment is found by the equation
n R1/ 2 T ~n R1/ 2 ,
where R1 / 2 is the rotation matrix from the right polar decomposition of the deformation gradient
evaluated at the midpoint, and the midpoint geometry is the average of the current and previous
increments,
X 1 / 2 1 X n X n 1 .
2

The strain increment, ~n , is found at the midpoint as well

~ B u
n 1/ 2 n
Finally, the strain increment is added to the strain of the previous increment
n n n1.

Design Challenge: Olympic Diving Board

To investigate and verify the effects of the stiffness matrix component added in nonlinear finite
element analysis, a simple problem involving a cantilever beam under transverse loading
conditions resulting in large deflection is considered in Figure 1.

Figure 1: Problem Specification

Solid Beam Analysis

The beam is fixed at the left end and loaded with a -2500lb force in the z-direction to induce a
large deflection comparable to that generated by a divers impact, while the dimensions are
constrained to 8 x 19 5/8 x 1 5/8 to simulate a half-length Olympic diving board. A study
conducted by Burkhardt, Barton, and Garhammer of California State University monitored impact
and propulsion forces associated with plyometric exercises such as vertical jumps resulting in
forces almost six times the individuals body weight. The loading condition, being that of a 400lb
individual, presents an over-estimation of the force exerted by an Olympic diver in practice.
According to a major supplier, the diving boards are extruded from an aircraft grade aluminum
alloy treated to a tensile yield strength of 50ksi, justifying the use of Al 2024-T6 material
properties. A base case is established by analyzing the solid beam geometry using ANSYS to
compare linear and nonlinear results, where nonlinear solution settings are adjusted to account for
large displacement and the Newton-Raphson method provides a converged solution. Results of the
simulation are reported in Table 1 below, where the deflection decreases from the linear to
nonlinear solution as expected. High factors of safety indicate that the beam geometry may be
changed in order to reduce the amount of material used in production to reduce costs, while still
meeting yield criterion under the given loading condition.

Table 1: Solid Beam Results Comparison

Method Deflection [in.] Max Tensile Stress FoS Equivalent Stress FoS
Linear 8.0566in 2.3051 2.3144
Nonliear 8.0282in 2.3151 2.3195

Beam Optimization

Using the Goal Driven Optimization feature within ANSYS, the cross-section of the beam in the
xz-plane is altered in order to achieve a reduction in volume by adding a series of fifteen holes
spaced six inches apart as seen in the geometry of Figure 2.

Figure 2: XZ-Plane Geometry

The diameter of each odd-numbered hole is used as a design variable in the optimization process
with an allowed range of 1 - 1.5 inches, while the even numbered holes remain fixed at 1 inch, and
the beam is subjected to identical boundary and loading conditions as in the solid beam case. Given
the design parameters, ANSYS samples points within the user specified design space, minimizes
the sample space via statistical methods, and creates an approximate finite element analysis
solution for each point. After setting goals to minimize volume and meet factors of safety in max
tensile and equivalent stress of 1.25, the optimization presents three possible candidates that may
be verified by substitution into the modeler. The geometry of the candidate chosen is listed below
in Table 2, while Table 3 displays the linear and nonlinear solution results in which the same trend
for reduced deflection in the nonlinear case is present. The cost analysis in Table 4 demonstrates
that in this proof of concept example, the simple addition of holes into the geometry of the beam
creates an 11% saving on both material consumption and cost. For a review of the mesh
convergence and SHELL 181 element technology used in both analyses, refer to the Appendix.

Table 2: Optimized Geometry

Hole # D1 D3 D5 D7 D11 D13 D15 D17
Diameter [in.] 1.0593 1.2151 1.1939 1.3763 1.4521 1.4052 1.0653 1.4812

Table 3: Optimized Beam Results Comparison

Method Deflection [in.] Max Tensile Stress FoS Equivalent Stress FoS
Linear 8.7993in 1.2754 1.2555
Nonliear 8.7576in 1.2756 1.2757

Table 4: Cost Analysis

Parameter Solid Beam Optimized Beam Reduction
Volume [in.3] 3061.5 2719.5 342
Cost [USD] $2915.91 $2590.18 $325.73

Recommendations and Conclusions

Due to the uncertainty in the number of iterations required to reach a converged solution and the
large matrices involved with the calculation of geometric nonlinearities, these computations need
to be completed on a computer. As such, understanding the theory behind the process ANSYS
uses is important to the effective use of the software for the modeling of large deformations.

For future study, the optimization could look into larger changes to the structure of the diving
board. In addition, the small difference in the numerical results between the nonlinear and linear
are a result of the realistic loading conditions resulting in smaller displacement than the large
deformations that would make the error of linear approximations non-negligible. A smaller
cross-section, such as that of a fishing pole would provide a better case study of such necessity of
nonlinear geometry. Large deformations of cantilever structures are matters of safety for many
real-life applications, and the computation time necessary to account for the design and
optimization of a product.
References

Gavin, Henri P. "Geometric Stiffness Effects in 2D and 3D Frames." N.p., 2012. Web. 20 Apr.
2014.

Gavin, Henri P. "Structural Element Stiffness and Mass Matrices." N.p., 2014. Web. 20 Apr.
2014.

"Nonlinear Finite Element Methods (ASEN 6107) Course Material." Nonlinear Finite Element
Methods (ASEN 6107) Course Material. N.p., n.d. Web. 20 Apr. 2014.

Wriggers, P. Nonlinear Finite Element Methods. Berlin: Springer, 2008. Print.

Bathe, Klaus Jurgen. "Nonlinear Analysis." MIT OpenCourseWare. N.p., n.d. Web. 20 Apr.
2014.

Nelson, Thomas, and Erke Wang. "Reliable FE-Modeling with ANSYS." N.p., n.d. Web.
<http://ansys.tv/staticassets/ANSYS/staticassets/resourcelibrary/confpaper/2004-Int-ANSYS-
Conf-24.PDF>.

Kohnke, Peter, PhD. ANSYS Mechanical APDL Theory Reference. Canonsburg, PA 15317:
ANSYS, Inc, Nov. 2013. PDF.

Madenci, Erdogan, and Ibrahim Guven. The Finite Element Method and Applications in
Engineering Using ANSYS. New York: Springer, 2006. Print.

Springboards And More Duraflex 16-foot Modified Maxiflex Model "B" Diving Board. N.p., n.d.
Web. 30 Apr. 2014. http://www.springboardsandmore.com/proddetail.asp?prod=66-231-
330

E. Burkhardt, B. Barton, and J. Garhammer, Maximal Impact and Propulsion Forces During
Jumping and Explosive Lifting Exercises, Journal of Applied Sport Science Research, Volume
4 Number 3, 1990. http://www.csulb.edu/~atlastwl/Impact_Forces-Jumps&Lifts.pdf.
"ASM Material Data Sheet: Al 2024 - T6." ASM Aerospace Specification Metals Inc. N.p., n.d.
Web. 01 May 2014.
http://asm.matweb.com/search/SpecificMaterial.asp?bassnum=MA2024T6

"Aluminum Bare Plate 2024." Order Aluminum 2024 Plate in Small Quantities at
OnlineMetals.com. N.p., n.d. Web. 01 May 2014.
http://www.onlinemetals.com/merchant.cfm?pid=6983&step=4&showunits=inches&id=9
15&top_cat=0
Appendix A

Figure1

Figure 2

Figure 3
Manual Mesh Convergence
Mesh Size [in.] Solid Beam Deflection [in.] Optimized Beam Deflection [in.]
.025 8.6921
.05 8.6314
.075 8.5762
.1 8.5321
.125 8.1403 8.5026
.25 8.0255 8.2517
.5 8.0566 8.0334
1 7.8053
2 7.8074

SHELL 181 Element Technology

Appendix B Ansys Features

Ansys utilizes a modified version of this Newton Raphson method to iterate nonlinear problems.
Instead of updating the stiffness matrix every iteration, updates would only happen during the
first or second iteration of each substep, as seen in the comparison below.

Incremental Newton-Raphson Initial-Stiffness Newton-Raphson

Even though the modified Newton-Raphson method takes a larger amount of time to converge,
the modified method requires a smaller number of matrix inversions and reformulations. The
convergence characteristics are somewhat different because there are fewer elements that form
an approximate tangent matrix.

In addition to iterating until the residual forces and moments are below the selected criterion, the
user can also choose to instead iterate until the magnitude of the incremental displacements
and/or rotations between iterations is smaller than a chosen limit. While taking this option can
provide faster convergence in some cases, it does not ensure that the structure is in equilibrium
and the user needs to remember to check that there are no errors, such as below, from using just
displacement controlled iteration.
Another Ansys feature is the predictor, which uses previous history to improve the guess for the
next solution by picking the current DOF solution to start each time step for both the tangent
matrix and the restoring load.

{un,0 } = {un1} + {un}

Where {un} = displacement increment accumulated over the previous time step
n = current time step
t
= t n < 5
n1

When encountering convergence difficulties, Ansys uses adaptive descent to change from the
full tangent matrix to a "stiffer" one, before changing back as the solution converges. This can
only be used with the full Newton-Raphson method, and further increases the rate of
convergence by forming a secant matrix if > 0 .

[KTi ] = [KS ] + (1 )[KT ]

Where [KS] = secant (or most stable) matrix

[KT] = tangent matrix
= descent parameter
A further option to improve on the Newton-Raphson method is the line search option, which
scales the solution by the line search parameter, s, also a scalar value.

{ui+1} = {ui } + s{ui }

Where s = line search parameter, 0.05 < s < 1.0

This parameter is chosen automatically by minimizing the energy of the system. The scaled
solution is then uses as the starting solution for the next iteration.

gs = {ui }T({F(i-1)}-{R(s{ui })})

Where gs = gradient of the potential energy with respect to s

Appendix C Senior Design Questionnaire
1. What are the function(s) of your design?
The function of the design is to support the impact force of a divers vertical jump on
structural deflection while the process serves as a demonstration of nonlinear finite
element analysis.

2. What constraints related to the main function(s) must your design satisfy?
The design of the diving board must support flexibility and strength in order to displace
under the divers weight, propel them into the air, and meet failure criterion with
appropriate factors of safety. A comparison of the linear and nonlinear solution should
exhibit a decrease in deflection in going from linear to nonlinear as a result of the added
stiffness matrix component.

3. What are the performance objectives of your design? What must be optimized?
The goal of the design is to alter and optimize the geometry of the beam such that a
reduction in volume and material cost is achieved. A series of fifteen holes are implemented
into the cross-section and the diameter of each odd numbered hole is used as a design
variable in optimization.

4. What alternative design concepts were considered?

The alternative design concepts include all other possible geometries that satisfy the hard
constraints on factors of safety in stress such that the beam does not enter yielding. Other
design concepts that could have been considered are cuts of different geometry made in the
cross-section (width and height). However, we found holes to be structurally more
interesting

5. What analyses were used to select among these alternative design concepts?
Given the design variables and constraints, ANSYS Goal Driven Optimization is used to
select the best candidates.

6. Which concepts or skills learned in your coursework were applied to the design? Please
provide a list with each entry providing the department and number of the course, plus a brief
description of the particular concept or skill used.
MAE 3250 ANSYS Workbench
MAE 4701

7. What format did your design take? For example, is it a complete set of CAD drawings, a
working prototype, a full finished product, a system configuration, a process, or something
else?
The design format is an ANSYS simulation

8. Briefly evaluate your design, relative to its function(s), constraints, and objectives.
The design succeeds in meeting yielding criterion and achieves an 11% reduction in cost,
however, we believe it may be altered to produce greater savings and a more explicit
representation of the benefits in using nonlinear processes for situations involving large
displacement.
9. Describe each students role in the design project and in writing this report, if there are
multiple authors.
Jordan: ANSYS
Craig: Theory
Wendy (not for senior-design): Theory