See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225474334

Artificial neural network and finite element modeling of nanoindentation tests

Article in Metallurgical and Materials Transactions A · July 2002

DOI: 10.1007/s11661-002-0027-3

CITATIONS READS
62 876

4 authors, including:

Ana Muliana Rami Haj-Ali

Texas A&M University Tel Aviv University
178 PUBLICATIONS 2,841 CITATIONS 159 PUBLICATIONS 3,847 CITATIONS

SEE PROFILE SEE PROFILE

Ashok Saxena
University of Arkansas
230 PUBLICATIONS 4,485 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Soft Materials, Soft Robotics, Eco Flex View project

Micromechanics View project

All content following this page was uploaded by Rami Haj-Ali on 21 May 2014.

The user has requested enhancement of the downloaded file.

Artificial Neural Network and Finite Element Modeling of
Nanoindentation Tests
ANASTASIA MULIANA, REJANAH STEWARD, RAMI M. HAJ-ALI, and ASHOK SAXENA

This study first presents two-dimensional (2-D) axisymmetric and three-dimensional (3-D) finite
element (FE) models of nanoindentation tests. Calculated load-displacement curves from the FE
models are compared with the load-displacement curves from nanoindentation measurements on
annealed copper. Numerical parametric studies are also performed to examine the effect of indenter
geometry and the material’s stress-strain behavior on the load-displacement response. The 2-D and
3-D FE load-displacement curves compare well with the measured results on annealed copper. The
second aspect of this study introduces a new modeling approach for indentation tests using artificial
neural networks (ANN). In this approach, ANN models are generated to approximate the FE load-
displacement curves for a wide range of material and geometric parameters. The ability of the ANN
models to predict the indentation response is examined against other FE results not used as part of
the training data. These models are shown to accurately predict the load-displacement behavior of a
nonlinear homogeneous material as well as one with a hard film, such as an oxide film, on a relatively
soft substrate. It is shown that the monotonic indentation load-displacement response during loading
contains ample information for the ANN model to extract material flow properties of the indented
material.

I. INTRODUCTION
THE strength, fatigue, and fracture properties of multi-
phase metallic alloys strongly depend on the in-situ elastic
and plastic flow properties of the individual constituents of
the microstructure. However, the in-situ mechanical proper-
ties of the individual constituents are difficult to determine.
The objective of this study is to extract stress-strain behavior
of a local region of a material using the load-displacement
behavior obtained during a nanoindentation test.
The general indentation problem has old roots in elasticity
and mechanics of solids. Prandtl[1] developed a slip-line field
solution for a flat punch on a semi-infinite medium. Hill et
al.[2] also considered a wedge indentation under plane strain
conditions. Ishlinky[3] obtained a solution for spherical
indentation. Shield[4] used an axisymmetric indentation
model by a flat cylindrical punch. Tabor[5] studied the plastic
deformed zone in elastic-plastic material as a result of spheri-
cal cavity expansion with an internal hydrostatic pressure,
and Johnson[6] extended this study to include the effect of
tip geometry. However, it is difficult to analytically describe
the complex stress-strain field due to inelastic response and
include detailed geometric parameters that may exist at dif-
ferent scales of the material.
Different numerical techniques have been developed and
can be used in indentation problems. Bhattacharya and Nix[7]
ANASTASIA MULIANA, Graduate Research Assistant, and RAMI M.
HAJ-ALI, Assistant Professor, are with the School of Civil and Environmen-
tal Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0355.
Contact e-mail: rami.haj-ali@ce.gatech.edu REJANAH STEWARD, for-
merly Graduate Research Assistant, School of Materials Science and Engi-
neering, Georgia Institute of Technology, is with the Department of Material
Science Engineering, University of Tennessee-Knoxville, TN 37996.
ASHOK SAXENA, Professor and Chair, is with the School of Materials
Science and Engineering, Georgia Institute of Technology.
Manuscript submitted July 26, 2001.
used the finite element (FE) method to solve for an indenta-
tion problem with an axisymmetric cone. The cone shape
is constrained to have the same volume as that of a pyramid-
shaped indenter for a given depth. Their FE hardness results
compared well with experimental data. They also presented
an elastoplastic analysis of axisymmetric conical indentation
and showed that the shape of the plastic zone strongly
depends on the indenter angle, Young’s modulus, and yield
stress.[8] The FE analyses seem to predict an upper bound
of the experimental results, while the spherical cavity model
seems to predict a lower bound behavior. Wang and Ban-
gret[9] used three-dimensional (3-D) FE models to investigate
the indentation response in a coated system. Sun et al.[10]
developed an axisymmetric FE model to simulate the inden-
tation of a hard coating on a softer substrate using a conical
indenter. The axisymmetric indenter is constrained to have
the same projected area of the Berkovich indenter, as a
function of the indentation depth. Lichinchi et al.,[11] Fivel
et al.,[12] and Knapp and Follstaedt[13] used 3-D and axisym-
metric FE models for different indentation problems in lay-
ered and homogeneous materials. A complex deformation
field is shown in their sixfold symmetric 3-D FE model for
verifying their axisymmetric model.[11]
The focus of the FE and analytical studies thus far has been
to predict the load-displacement curves given the plastic flow
properties. In this study, an artificial neural network (ANN)
approach has been used to solve the inverse problem of
extracting the plastic flow properties from measured load-
displacement behavior.
The ANN is a collection of computational cells (neurons)
that communicate through different connections with
adjusted weights. A forward-pass-type network consists of
a number of layers with several cells. Each cell is usually
connected with all cells in both the previous and the next
layers. The first and last layers include the input and output
cells. The ANN learning and training is achieved by
adjusting the connection strengths (weights) between the

METALLURGICAL AND MATERIALS TRANSACTIONS A VOLUME 33A, JULY 2002—1939

optimization variables are the weights. Knowledge represen-
tation can be achieved with different ANN models.[14,15] The
ANN is an effective tool when relationships and approxima-
tion of variables are needed from a given large volume of
data.[16,17,18] Complex nonlinear relationships can be identi-
fied using different ANN structures and training methods.
The domain of approximation is usually determined during
the training along with the connection weights. The ANNs
are robust, noise and fault-tolerant, and are capable of
generalizations.
The ANN has been applied in many engineering applica-
tions. The ANN has also been used to model nonlinear
constitutive behavior of different materials.[19–24] An indirect
method, called autoprogressive training, for training neural
network material models from structural tests, was recently
presented by Ghaboussi et al.[16] Structural test responses,
for example, load vs deflection data, are used to train the
neural network to learn the inelastic response and behavior
of the material. A fundamental premise of this method is
that structural tests usually generate a large number of spatial
patterns of stresses and strains that can be used for training.
Another complementary method to the autoprogressive train-
ing was proposed by Haj-Ali et al.[17,24] New micromechani-
cal material models are generated using ANN. The response
of the heterogeneous material, with and without induced
states of damage, is first generated using detailed FE micro-
mechanical models. The multiaxial effective material
response is captured by the trained ANN micromodels.
Currently, there are no well-developed experimental tech-
niques for measuring the mechanical properties, such as
modulus elasticity, hardness, strength, and toughness, of the
various microstructural constituents that are present in the
alloy at micron and submicron length scales. This article
presents two-dimensional (2-D) axisymmetric and 3-D non-
linear FE models for indentation. The FE results are com-
pared with experimental test performed on annealed copper
material. The results are in the form of force-displacement
profiles during the loading-unloading process. The first part
of this study describes nanoindentation tests performed on
annealed copper. The second part includes 2-D and 3-D FE
models with different parametric studies. The indentation
response is examined with different parametric studies,
including the effect of elastoplastic material properties, the
shape of the indenter, and a changing thickness of a hard
film on a soft substrate.
Finally, ANN models for indentation are proposed along
with different verification examples. In these examples, shal-
low indents are produced experimentally and simulated with
FE models. A major contention of the proposed ANN models
is that the data, from relatively shallow indentation tests
without unloading, contain sufficient information that allow
the trained ANN models to predict both elastic modulus and
plastic hardening of the material using load-displacement
indentation response.
II. EXPERIMENTAL RESULTS ON COPPER
In this investigation, the Nanoindenter-II manufactured
by MTS (Oak Ridge, TN), equipped with a Berkovich tip
is used to apply loads in the millinewton (mN) range to
produce indentations of the nanometer scale depth on a
copper sample and generate a load-displacement curve.
Information obtained is analyzed usually by applying the
2—VOLUME 33A, MONTH 2002
Fig. 1—An optical macrograph showing indentation matrix on copper. Note
the different size of the indents.
Oliver–Pharr method, which approximates hardness and
elastic modulus using empirical relationships. The unloading
data retrieved from the tests are fit to a power-law relation
between load and displacement.[25] The stiffness is deter-
mined from the differentiation of the power law at the maxi-
mum penetration depth. Using an empirically determined
indenter shape function, the hardness and the effective modu-
lus are determined. Unlike conventional microhardness
measurements, the advantage of the shape function is that
it does not require measurement of each indent.
Copper was selected as the model material to demonstrate
and compare the modeling approach because it is single
phase and isotropic. Specimen preparation consisted of
grinding and fine polishing to a 1-mm finish to minimize
surface artifacts. After careful specimen preparation, each
specimen was mounted into a specimen chamber and secured
for testing. Several approaches were taken to obtain data
from the copper sample. The first approach consisted of
creating 12 indents in a rectangular array at various depths
(Figure 1). The separation distance between each indent is
50 ␮m. In segment 1, the nanoindenter was instructed to
make two indents up to 1500-nm depth and immediately
unload. The various load-displacement curves are presented
along with the FE indentation results described in Section III.
III. FINITE ELEMENT MODELS AND RESULTS
The present study is based upon different FE modeling
strategies used for simulating the nanoindentation process
using 2-D axisymmetric and 3-D models. The simulations
are aimed to generate the force-displacement relationship
using a conical-shaped rigid indenter. The 2-D axisymmetric
and 3-D FE models are performed in ABAQUS* FE code.[26]
*ABAQUS is a registered trademark of Hibbitt, Karlsson, & Sorensen,
Inc., Pawtucket, RI.
The indentation process is simulated both during loading
and unloading. During the loading process, the indenter is
pushed into the specimen under load or displacement control,
and the changing depth is monitored. For a given displace-
ment, the corresponding load is calculated from the reaction
force of the indenter (displacement control). Incremental
METALLURGICAL AND MATERIALS TRANSACTIONS A
plasticity with isotropic hardening is used to model the mate-
rial behavior of annealed copper. The uniaxial stress-strain
curves are assumed to follow the Ramberg–Osgood stress-
strain relationship expressed as stress vs strain or strain vs
stress by the following equations:
␴0 1 n
⑀⫽ ⫹ n␴
E k
冦
E⑀, ⑀ Ɐ ⑀0
␴⫽ 1 [1]
k⑀p p冢 冣,
n ⑀ Ɑ ⑀0
⑀p ⫽ ⑀ ⫺ ⑀0
where E is Young’s modulus, n is hardening power, and (␧0,
␴0) is the yield point.
The properties used for annealed copper material are
Young’s modulus, E ⫽ 107.6 GPa; Poisson’s ratio, v ⫽
0.245; hardening power, n ⫽ 1.85; k ⫽ 0.32 GPa; and yield
stress, ␴0 ⫽ 34.5 MPa.
Contact with sliding is used to model the interaction
between the rigid indenter and the block surface. The rigid
indenter and the specimen surfaces are defined within the
FE procedure using a “master-slave” contact scheme.
A. Axisymmetric Models
The 2-D FE models consist of four-node axisymmetric
elements with reduced integration (CAX4R). A typical axi-
symmetric model and its geometry are depicted in Figure
2. The region of contact part is 1000 nm along the width
of the specimen surface (9000 nm). A fine mesh is used
under the contact area and near the tip of the indenter. The
mesh is progressively coarser further away from the tip. All
nodes at the bottom of the model are prevented from moving
in the x and y directions. Nodes along the axisymmetric axis
are constrained in the x direction.
A mesh convergence study was first performed in order
to determine an adequate number of nodes and elements.
This is done by running different models with increasing
number of elements until the difference between a two con-
secutive models was negligible. The 2-D mesh that is used
as a result of the convergence study includes a block with
815 nodes and 761 elements. The characteristic length for
the elements located directly under the indenter is on the
order of 1 to 10 nm.
A conical-shaped indenter is used in the 2-D models and
is defined as a rigid surface with a conical angle of 70.5
deg. The tip of the cone is trimmed by adding a sphere with
a radius R ⫽ 20 nm. The tip radius used in the 2-D simula-
tions may not exactly match the tip radius of the actual
Berkovich indenter. This can lead to initial mismatch
between the FE model and test results at indentation depths
close to the radius magnitude. Loading is achieved by mov-
ing the indenter in the negative y direction in order to pene-
trate the block. A second unloading step is performed, and
the indenter is released back to its initial position.
B. Three-Dimensional Models
The 3-D indentation model and geometry are depicted in
Figure 3. The symmetry of the model can be realized by
METALLURGICAL AND MATERIALS TRANSACTIONS A
examining the projection view of the Berkovich indenter,
as shown in Figure 3(c). The shaded area (1/6 symmetry)
can be used as a 3-D model with the proper symmetry
boundary conditions. The FE model includes 11,970 eight-
node brick elements with reduced integration (C3D8R). The
tip area of the FE model consists of 315-wedge elements
(C3D6) located at the center of the indented block. The
elements in the area under the contact surface have a charac-
teristic length of 10 nm. Both the Berkovich and the cone-
shaped indenters are used in the 3-D models. They are mod-
eled as rigid surfaces. The cone-shaped indenter has the
same dimensions as the indenter in the axisymmetric model.
Loading is applied by moving the indenter in the negative
z-axis direction in order to create the penetration in the
block. A second unloading process is applied by releasing
the displacement and moving the indenter back to the ini-
tial position.
C. Analysis of Experimental and FE Results
Figure 4 shows a comparison between the measured load-
displacement response during several nanoindentation tests
and the 2-D and 3-D FE predictions for annealed copper.
As mentioned before, the 3-D FE analysis uses the Berkovich
indenter, and the 2-D FE model uses an equivalent cone
indenter. The FE results, solid and dashed lines in Figure 4,
compare favorably with the repeated experimental data. The
FE-simulated load-displacement behavior includes an inden-
tation up to 250 nm, followed by the unloading step. It is
noted that the 2-D and 3-D curves begin to deviate from
each other and from the test results after an indentation depth
of 200 nm, and the 2-D FE response is consistently softer
than the 3-D model. This may be attributed to differences
in mesh refinement near the tip zone or other factors relating
to the difference between the indenters. Recall that the only
constraint that is satisfied between the two models is only
the common projected contact area. These differences are
explored in more depth in the following discussion.
In depth examination of the mesh deformation shows that
the number of elements in the 2-D model are sufficient to
capture a smooth deformed shape compared to the 3-D
model. The latter is stiffer due to smaller number and larger
aspect ratio of the elements used. Further, comparing the
effective Mises stress contours for the 2-D and 3-D models,
Figures 5 and 6, respectively, show comparable levels of
stress away from the tip region. The maximum stress magni-
tudes, as expected, are in the areas under the tip and have
values of 1.95 and 1.09 GPa for the 2-D and 3-D models,
repectively, at indentation level of 250 nm. The stresses
appear more spatially distributed in the depth direction than
the 2-D case. The situation is reversed when comparing the
contours for the effective plastic strain. The magnitude for
the plastic strain is larger for the 3-D case and is localized
in a small area under the Berkovich tip. Thus, sufficient
differences exist between the 3-D and 2-D simulations to
rationalize the small differences between the resultant load-
displacement behavior. Nevertheless, the 2-D simulations
seem to provide an accurate overall representation of the
measured trends. Thus, the parametric studies described in
Section III–have been performed using the 2-D model.
VOLUME 33A, MONTH 2002—3
Fig. 2—Geometry and two-dimensional (axisymmetric) FE model of the indented medium.
D. Parametric Studies
Three parametric studies are presented using the 2-D axi-
symmetric model. The first study examines the indentation
response as a function of different indenter angles (␣ ⫽ 30,
40, 50, 60, and 70.5 deg). The cone angle, ␣ ⫽ 70.5 deg,
is the equivalent angle that maintains the same projected
area of a cone equal to that of a Berkovich pyramid-shaped
indenter. The annealed copper material is modeled using
incremental plasticity with isotropic hardening material. The
Ramberg–Osgood constants and the elastic constants used
in the analysis are described in Section II. Four intermediate
loading levels (1200, 3600, 7200, and 12,000 nN), which
are fractions of the maximum applied load, are used to
illustrate the different responses when loads are decreased
to fractions of these intermediate values. The results are
summarized in Figure 7. The force-displacement response
is dependent on the indenter angle especially at large dis-
placements. Sharper indenter tips create more deformation,
and their indentation curves are relatively shallower for the
same applied load. At small displacements, the effect of the
indenter angle is not pronounced because the behavior is
more dependent on the spherical indenter tip geometry,
which is the same for all indenter angles.
The variation in load-displacement diagram with different
indenter angles shows that it is a variable that can be chosen
based on the material and microstructural characteristics.
However, it is then necessary to adjust the hardness and
stress-strain behavior interpreted from the measurements.
4—VOLUME 33A, MONTH 2002
For example, for hard materials, smaller indenter angles may
be more suitable to reach larger depths for a given maximum
force capacity.
The second parametric study involved changing the strain-
hardening exponent of the indented material. The power
hardening in the Ramberg–Osgood uniaxial stress-strain
relationship of Eq. [1] is used with different values (n ⫽
1.25, 1.5, 1.75, 2, 2.5, 5, and 10), while the indenter angle is
held constant at ␣ ⫽ 70.5 deg. Two predetermined unloading
levels are selected (1200 and 7200 nN). The effect of the
material hardening is pronounced even at very low levels
of deformation, as shown in Figure 8. Small changes in n
are significant enough to be resolved by the resultant load-
displacement diagrams. It is then clear that the nanoindenta-
tion technique can be used to determine the stress-strain
behavior of metallic materials. However, a methodology is
needed to solve the inverse problem of determining the
stress-strain behavior from measured load-displacement
characteristics. This will be addressed in greater detail in
Section IV on ANN.
The third parametric study includes models with different
oxide coating thicknesses. The goal of this investigation is
to simulate the effect of the often unknown oxidation layer
thickness that exists in the specimens used for nanoindenta-
tion testing. The coating material is assumed as the stiffer
material, with Young’s modulus being 20 times larger than
the annealed copper (E ⫽ 2151 GPa). The coating is assumed
as elastic and the substrate as elastoplastic materials. The
METALLURGICAL AND MATERIALS TRANSACTIONS A
Fig. 3—(a) through (c) Geometry and FE mesh of 3-D (1/6 symmetry) model with Berkovich indenter.
Fig. 4—Indentation load-displacement curves from 3-D FE with Berkovich
indenter, 2-D model with equivalent cone indenter, and experimental results.
conical indenter equivalent to the Berkovich is used in a
2-D axisymmetric model. The indentation results are shown
in Figure 9 for different coating thicknesses. As expected, a
consistent softening response exists with decreasing coating
thickness. With additional studies, such as the preceding, it
may be possible to correct the load-displacement diagrams
for the substrate if the oxide thickness is known. It also
shows that it is necessary to conduct the nanoindentation
tests as soon as possible to prevent the results from being
influenced by the oxide coatings, especially in certain materi-
als, such as copper.
METALLURGICAL AND MATERIALS TRANSACTIONS A
Fig. 5—Mises stress contours for 2-D (axisymmetric) FE model (indenta-
tion ⫽ 250 nm).
IV. ANN INDENTATION MODELS
The topology of the neural network used in this study has
two hidden layers. These are constrained to have the same
number of neurons (nodes). This number usually determines
the capacity of the neural network. The neural-network train-
ing algorithm includes different criteria that allow ANN size
growth by adding nodes in the hidden layers. Training is
VOLUME 33A, MONTH 2002—5

Fig. 6—Mises stress from FE model of 3-D (1/6 symmetry) medium with
Berkovich indenter (indentation ⫽ 250 nm).
Fig. 7—Load-displacement curves of different indenters with changing
cone-head angle.
Fig. 8—Load-displacement indentation curves with changing material
hardening exponent.
usually initiated with a small number of nodes. Additional
nodes are then added if the error level is not reduced within
a specified number of iterations. Random weights for an
ANN are initially assigned.
6—VOLUME 33A, MONTH 2002
Fig. 9—Effect of different coating thickness on the overall load-displace-
ment curves.
Fig. 10—ANN architecture and its hidden layers’ evolution during training
of indentation response of films, with different moduli and thicknesses, on
a copper substrate.
A program was written[17,24] for training the type of feed-
forward ANN. The classical back-propagation method is
used for determining the weight connections by minimiza-
tion of the total error. Nodes in the hidden layers are simulta-
neously added, and the new connection weights are achieved
by “freezing” the old weights for a number of iterations
before continued training. An intermediate training step is
added where training is limited only to the 20 pct of the
worst data cases. The equation for the error is defined as a
quadratic sum of the Euclidian norm of all error vectors.
This is expressed as
1
error ⫽
2 兺 㥋Tn ⫺ On㥋2
n⫽1
[2]
where T and O are the input and output vectors.
A class of simple feed-forward neural networks is used
in this study for the different indentation models. An example
is schematically depicted in Figure 10 for the ANN model
of indentation of a hard film on a soft substrate. It consists
of an input layer, two hidden layers with several nodes or
neurons, and an output layer. Three input variables (⌬, Ef /Ec ,
and tf) are used in this case, which represent the indentation
displacement, the film-substrate stiffness ratio, and the thick-
ness of the film. The output of the ANN is the indentation
force. Each neuron in a given layer has weighted connections
to every neuron in the next layer. The neural network is
METALLURGICAL AND MATERIALS TRANSACTIONS A

Fig. 11—ANN and FE indentation response as a function of the hardening
parameter n.
trained with a large set of input and output vector pairs that
are generated from the different FE models. The connection
weights are modified during the training until the ANN
yields a force response that satisfactorily agrees with the
correct FE simulation results. The term back propagation
refers to the training algorithm, where the observed differ-
ence in the force output, ANN vs FE, is used to modify the
connection weights.
The selection of an appropriate set of training cases and
procedure is very important. As in any approximation, there
is not a clear method to generate a comprehensive a priori
estimate of the required set of training cases before knowing
the outcome and convergence of the training process. Several
training strategies with different data sets have to be initially
applied. The training data must contain the knowledge that
describes the indentation behavior within a finite domain of
input and output. This is needed to allow the trained neural
network to generalize the indentation response that is implic-
itly included in the training data. The trained neural network
is then examined, and its response is compared, often visu-
ally, with the full range of trained and untrained (verifica-
tion) data.
The term “ANN prediction” is reserved for ANN response
for cases that were not used in the pretraining stages. This
is used in order to examine the ANN’s ability to associate
and generalize a true physical response that have not been
previously “seen.” A good prediction for these cases is the
ultimate verification test for the ANN models. These tests
have to be applied for (input and output) response within
the domain of training. It should be expected that ANN will
produce poor results for data that are outside the training
domain.
ANN Case Studies
In order to demonstrate the proposed ANN modeling of
indentation response, three different indentation problems
are defined, and ANN models are generated for each of
these cases. The FE models are used to generate indentation
data, both for training and verification purposes. The first
case is for indentation in a homogeneous medium that has
different hardening parameter. The Ramberg–Osgood uniax-
ial stress-strain relation is used in the form of
METALLURGICAL AND MATERIALS TRANSACTIONS A
Fig. 12—ANN and FE indentation response for different yield stress values
and a fixed hardening power.
n
␴ ⑀ ⑀0 ⫽ 0.00032
冢 冣 冢冣
␴0
⫽
⑀0 ␴0 ⫽ 0.0345 GPa
[3]
The first ANN model has an input vector with two compo-
nents, ⌬ and n, and the output is the indentation force. The
FE simulation is trained with monotonic response that is
generated using displacement control. Figure 11 describes
the FE and ANN indentation response for a wide range of
hardening parameters. The case with hardening parameter,
n ⫽ 1.8, is not used as part of the ANN training. Instead,
the ANN is trained with the other cases. The prediction of
ANN is compared with the FE results and is shown in thick
dashed and solid lines, respectively. The trained ANN has
clearly captured the indentation behavior for both data cases
(training and verification). It is interesting to note that con-
vergence difficulties are experienced for the elastic case
(n ⫽ 1), where the analysis is terminated due excessive
deformation at about an indentation level of 70 nm. However,
the ANN response beyond this level of indentation is clearly
reasonable, and the ANN was not degraded due to the lack
of data in this range. This can only add to the capability of
ANN in generalizing the data.
The second case study is similar to the first; however, the
yield stress is not specified and is part of the ANN input.
To this end, the ANN input vector has three variables: dis-
placement, hardening parameter, and yield stress (⌬, n, and
␴0). The output is the indentation measured force. Indenta-
tion simulations are performed using all combinations of
hardening and yield stress parameters:
n ⫽ 1.7, 1.8, 1.9, 2, 2.1, 2.2, (2.25), 2.3, and 2.4
␴0 ⫽ 0.03, 0.032, (0.0325), 0.033, 0.034, (0.0345), 0.036,
0.037, and 0.04 GPa [4]
E ⫽ 107.562 GPa
where the values in parenthesis are not used in the training
cases and are reserved for verification.
Figures 12 and 13 show the indentation load-displacement
response from the FE and ANN models after training. The
ANN training is successful and can be verified visually by
examining how well the dashed curves are close to the FE
solid lines. It is interesting to note that, despite the load-
deflection variation not being large for the response cases
in Figure 12, the trained ANN response is still very close
VOLUME 33A, MONTH 2002—7
Fig. 13—ANN and FE indentation response for different hardening powers
and a fixed yield stress.
Fig. 14—ANN prediction compared to FE indentation response for different
yield stress and hardening values (that were not used in the training).
to the FE. Figure 14 shows the ANN predicted response
with the FE analysis for cases that are not part of the training
process. It is clear that the ANN model is capable of pre-
dicting the indentation results for a wide range of material
yield and hardening values.
The ultimate verification would be to solve for the various
constants in Eq. [1], e.g., ␴0, k, and n, from the measured
load-displacement behavior during the loading portion of
the nanoindentation test and for E from the unloading portion
of the curve. To solve for these several constants and to
have confidence in the data, it may be beneficial to vary the
indenter geometry and depth systematically to insure the
integrity of the predicted constants. This will be pursued in
future studies to further develop this approach.
The third case study is concerned with constructing an
ANN model for load-displacement indentation response of
different, hard-elastic films on a soft copper substrate. The
hard films are defined with elastic modulus and thickness.
The ANN has an input vector with three components: dis-
placement of indenter, thickness of film, and film-substrate
elastic stiffness ratio (⌬, tf , and Ef /Es). The indentation force
is generated using different 2-D FE models for combinations
of film thicknesses and elastic stiffness ratios:
tf ⫽ 10, 15, 20, 25, (28), 30, 35, and 40 nm
8—VOLUME 33A, MONTH 2002
Fig. 15—FE indentation response as a function of coating thickness and
for fixed film-substrate Young’s modulus ratio.
Fig. 16—ANN and FE indentation response for different film-substrate
Young’s modulus ratio and fixed coating thickness.
Es ⫽ 107.562 GPa
␴0 ⫽ 0.0345 GPa
[5]
Ef
⫽ 1, (1.5), 2, (2.5), 3, 5, (6), 7, (8), and 10
Es
n ⫽ 2.1
where the values in parenthesis are not used in the training
cases and are reserved for verification.
Figure 15 shows the FE indentation response for cases
with Ef /Es ⫽ 3 and different film thicknesses. The analysis
is terminated at a relatively low level of indentation, ⌬ ⫽
100 nm, in order to examine the ability of a trained ANN
to capture and distinguish the different curves within a small
area of response. The case with tf ⫽ 10 nm did not converge
beyond ⌬ ⫽ 80 nm because of a tight, applied convergence
tolerance. No attempt was made to relax the convergence
criteria on residual forces and other contact algorithm param-
eters for two reasons. The first is that based on the last case,
this lack of convergence and partial result should not prevent
the ANN from generating a good model. The second reason
is to assure that all FE results are generated with the same
convergence criteria. Figure 16 shows that there is excellent
METALLURGICAL AND MATERIALS TRANSACTIONS A

Fig. 17—ANN prediction compared to FE indentation response for different
film-substrate Young’s modulus ratio and fixed coating thickness.
response of the trained ANN compared with the FE curves
for the training cases with tf ⫽ 40 nm and different stiffness
ratios, Ef /Es . The quality of the ANN for predicting the
indentation response of a coated soft substrate is shown in
Figure 17 for cases that are not part of the training process,
tf ⫽ 28 nm and Ef /Es ⫽ 1.5, 2.5, 6, and 8. The ANN models
for all cases except for Ef /Es ⫽ 8 show excellent predictive
capability. This can be explained by examining the different
stiffness ratios used in the training cases in Eq. [5]. Unlike
the stiffness ratios used for verification, values in parenthe-
sis, the last one is within a wider interval, between 7 and
10. This can be solved by adding more training cases; how-
ever, the overall performance of the ANN is clearly
acceptable.
V. SUMMARY AND CONCLUSIONS
Finite element modeling is an effective tool to simulate
the indentation response at the nanoscale level. This study
presents 2-D and 3-D numerical FE models. Numerical para-
metric studies are performed for other indentation problems
to examine the effect of material and geometric parameters
on the overall load-displacement response. The 2-D and 3-
D FE load-displacement curves compare well with available
test results on annealed copper. While the 2-D models are
accurate and computationally affordable, the use of 3-D
models is also needed in cases where multiphase materials
are investigated.
The ANN models are proposed in the second part of this
study to model the indentation response as a function of
different material and geometric parameters. The ability of
the ANN models to predict the indentation response is exam-
ined against other FE results that are not used as part of the
training data. The proposed ANN models show excellent
predictive capability for indentation of a nonlinear homoge-
neous material as well as for a hard film on a soft substrate.
This is demonstrated by generating ANN models with non-
linear material parameters in the form of Ramberg–Osgood
stress-strain relations as well as geometry of the indented
material system.
METALLURGICAL AND MATERIALS TRANSACTIONS A
View publication stats
Finally, most indentation tests on macro or micro/scales
are mainly concerned with unloading slope because it can
be related to analytical expressions for the elastic stiffness
and hardness of the material. However, a major motivation
of this study is to advocate the use of the loading portion
of the load-displacement response during an indentation test.
It is shown that this type of response does include masked
information on the indented material that can be uncovered
using the proposed ANN modeling approaches.
ACKNOWLEDGMENTS
We acknowledge the assistance of Ms. Laura Reister, Oak
Ridge National Laboratory, with nanoindentation testing.
This work was partially supported by the NSF, through the
Civil and Mechanical Systems (CMS) Division, and under
Grant No. 9876080.
REFERENCES
1. L. Prandtl: Goettinger Nachr, Math. Phys. Kl., 1920, pp. 74-85.
2. R. Hill, E.H. Lee, and S.J. Tupper: Proc. R. Soc. London, 1947, vol.
A188, pp. 273-90.
3. A. Ishlinsky: J. Appl. Math. Mech., 1944, vol. 8, p. 201.
4. R.T. Shield: Proc. R. Soc. London, 1955, vol. A233, pp. 267-87.
5. D. Tabor: Rev. Phys. Technol., 1970, vol. 1, pp. 145-79.
6. K.L. Johnson: J. Mech. Phys. Solids, 1970, vol. 18, pp. 115-26.
7. A.K. Bhattacharya and W.D. Nix: Int. J. Solids Struct., 1998, vol. 24
(12), pp. 1287-98.
8. A.K. Bhattacharya and W.D. Nix: Int. J. Solids Struct., 1991, vol. 27
(8), pp. 1047-58.
9. H.F. Wang and H. Bangert: Mater. Sci. Eng., 1993, vol. A16, p. 343.
10. Y. Sun, T. Bell, and S. Zheng: Thin Solid Films, 1995, vol. 258, pp.
198-204.
11. M. Lichinchi, C. Lenardi, J. Haupt, and R. Vitali: Thin Solid Films,
1998, vol. 333, pp. 278-86.
12. M.C. Fivel, C.F. Robertson, G.R. Canova, and L. Boulanger: Acta
Mater., 1998, vol. 46 (17), pp. 6783-94.
13. J.A. Knapp and D.M. Follstaedt: Surface Coating Technol., 1998, vols.
103–104, pp. 268-75.
14. J. Hertz, A. Krogh, and R.G. Palmer: Introduction to the Theory of
Neural Computations, Addison-Wesley, Redwood City, CA, 1991.
15. H. Ritter, T. Martinez, and K. Schulten: Neural Computation and Self-
Organizing Maps: An Introduction, Addison-Wesley, Redwood City,
CA, 1991.
16. J. Ghaboussi, D.A. Pecknold, M.-F. Zhang, and R.M. Haj-Ali: Int. J.
Num. Methods Eng., 1998, vol. 42, pp. 105-26.
17. R.M. Haj-Ali, D.A. Pecknold, and J. Ghaboussi: Modeling and Simula-
tion Based Engineering, Proc. Int. Conf. on Computational Engi-
neering Science, ICES98, Atlanta, GA, Oct., S.N. Atluri and P.E.
O’Donoghue, eds., Tech Science Press, pp. 551-57.
18. R.M. Haj-Ali, K.E. Kurtis, and A.R. Sthapit: ACI Mater. J., 2001,
vol. 98 (1), pp. 36-43.
19. J. Ghaboussi, J.H. Garrett, and X. Wu: Proc., NUMETA-90, Int. Conf.
on Numerical Methods in Engineering: Theory and Applications,
Swansea, England, pp. 701-17.
20. J. Ghaboussi, J.H. Garrett, and X. Wu: J. Eng. Mech. Div., ASCE,
1991, vol. 117 (1), pp. 132-53.
21. G. Ellis, C. Yao, R. Zhao, and D. Penumadu: J. Geotechn. Eng., ASCE,
1995, vol. 102, pp. 429-35.
22. R. Pidaparti and M. Palakal: AIAA J., 1993, vol. 31, pp. 1533-35.
23. J.-W. Oh, I.-W. Lee, I.-T. Kim, and G.-W. Lee: ACI Mater. J., 1999,
vol. 96 (1), pp. 61-67.
24. R.M. Haj-Ali, D.A. Pecknold, J. Ghaboussi, and G.Z. Voyiadjis: J.
Eng. Mech. Div., ASCE, 2001, vol. 127 (7), pp. 730-38.
25. W.C. Oliver and G.M. Pharr: J. Mater. Res., 1992, vol. 7, p. 1564.
26. Hibbitt, Karlsson, & Sorensen, Inc., ABAQUS, Version 5.8, User’s
Manual, Pawtucket, RI, 1998.
VOLUME 33A, MONTH 2002—9