Professional Documents
Culture Documents
Greg Black
Lafayette College
Easton, Pennsylvania
Advisor:
Dr. Jennifer Righman
University of Delaware
Newark, Delaware
Submitted to NSF-REU
11 August 2006
Abstract
The goal of this report is to evaluate a hydrocode, which is a type of computer
program, called AUTODYN for the use of modeling blast loads on bridge sections. Blast
modeling is necessary due to the threats posed by terrorist attack and current technology
options presented by AUTODYN which set it apart from other hydrocodes and other
available software. These include the benefits of it graphical interface, modeling options
and remapping capabilities. Meanwhile, its large demand on memory for complex
models creates issues in the modeling phase, before the models can actually be analyzed.
Yet if the user can get past the quirks of the program and work within the memory limits
ABSTRACT .................................................................................................................................................. 2
1 INTRODUCTION ..................................................................................................................................... 5
2 INTRODUCTION TO BLASTS .............................................................................................................. 5
2.1 EXPLOSIONS ......................................................................................................................................... 6
2.2 CONWEP............................................................................................................................................... 7
3 INTRODUCTION TO HYDROCODES ................................................................................................. 8
3.1 MODELING TECHNIQUES ...................................................................................................................... 9
3.1.1 Structured vs. Unstructured Solvers .......................................................................................... 11
3.1.2 Lagrange Solvers ....................................................................................................................... 13
3.1.3 Euler Solvers.............................................................................................................................. 14
3.1.4 Other Solvers ............................................................................................................................. 16
3.2 INTRODUCTION TO AUTODYN.......................................................................................................... 17
3.2.1 Material Models......................................................................................................................... 18
3.2.2 Parts........................................................................................................................................... 20
4 MODELING AND RESULTS................................................................................................................ 23
4.1 AUTODYN MODELS ......................................................................................................................... 23
4.2 RESULTS AND DISCUSSION ................................................................................................................. 27
4.2.1 Pressure in the Slab ................................................................................................................... 28
4.2.2 Deflection in the Slab................................................................................................................. 32
4.2.3 Effective Strain in the Slab......................................................................................................... 33
4.2.4 Pressure in the Air ..................................................................................................................... 34
4.2.5 Conclusions and Suggestions for Future Investigation.............................................................. 37
5 ACKNOWLEDGEMENTS .................................................................................................................... 38
6 REFERENCES ........................................................................................................................................ 38
APPENDIX – MODELING NOTES ........................................................................................................ 40
List of Figures
List of Tables
Table 4.1 – Models........................................................................................................................27
Table 4.2– Gauge #1 Initial Peak Pressure, Same Air Element Size............................................29
Table 4.3 – Gauge #2 Initial Peak Pressure, Same Air Element Size...........................................29
Table 4.4 – Gauge #1 Initial Peak Pressures, Same Slab Element Size.......................................31
Table 4.5 – Gauge #2 Initial Peak Pressures, Same Slab Element Size.......................................31
1 Introduction
The events of 9/11 continue to have a lasting effect on the US and the world.
Everyone on the planet has been affected in some way or another. The implications of
the vulnerability of the nation’s infrastructure to terrorist attack are a concern that should
be shared by all engineers. If bridges and other structures may be subjected to severe
loads from explosions or other sources, then it is the engineer’s responsibility to prepare
for them.
between structures and explosions. Yet methods for explosive testing are limited due to
cost and permissions for experimental results. Therefore, with modern advances in
This paper will evaluate a hydrocode program called AUTODYN for the use of
user interactions with the program as well as the accuracy of the several simulations run
in the program. It will also provide the reader with a brief overview of blasts or
explosions in order to provide some background on the subject as well as a basis for the
2 Introduction to Blasts
This section will discuss a few of the basic properties of explosions. Once these
ideas are understood, the interactions between explosions and structures can be more
easily discussed. It will also discuss ConWep, a blast calculation program distributed by
the United States government (Robert, 2007), which will be used to evaluate the
performance of AUTODYN.
2.1 Explosions
Figure 2.1 depicts a few of the basic characteristics of a simple explosion in air.
There is the charge (a), the pressure wave (p) and the standoff distance (r). The main
component that any explosion requires is some type of fuel or charge such as TNT.
When ignited, this charge rapidly releases energy in the forms such as heat, sound or
pressure waves (Robert, 2007). The pressure wave expands out from the charge. The
leading edge of this wave is sometimes called the “shock front” and will generally have
the highest pressure in the wave at any given point in time (Wilkinson et al., 2003). The
standoff distance is basically the distance from the center of the explosion to any object
or point of interest.
follow the same general pattern, so long as there isn’t any reflection from nearby objects.
This pattern, called an overpressure curve (Wilkinson et al., 2003) can be seen in Figure
2.2, below. The main components of the overpressure curve are the detonation (a),
arrival time (b), peak pressure (c), and time duration (c to e). The detonation can be
considered as time 0, while the arrival time is the time that it takes for the pressure wave
to reach the point of interest (Robert, 2007). Once the peak pressure is reached, it
immediately starts to decay and the time it takes the pressure to return to normal is called
the time duration (Wilkinson et al., 2003). As the material in the blast wave expands
outward it can leave a void, creating a region with pressure lower than normal
atmospheric pressure (Robert, 2007). The size, shape and material of the charge, as well
as the stand off distance will all determine the magnitude and shape of this curve. In
addition to the above factors, the blast wave and the pressure involved can reflect off of
surfaces in various directions, and cause further fluctuations in pressure at a single point.
1.0
c
0.9 SoD = 1000 mm
Quantity: 1 kg
0.7
0.6
0.5
0.4
0.3
0.2
d
0.1
a e
b
0
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Time, t, ms.
Figure 2.2 – Standard Pressure vs. Time Curve for an Explosion (Robert, 2007)
2.2 ConWep
States government. Users can input a charge size and standoff distance and receive
pressure for that point in relation to time as output. It also allows users to receive
pressure data after interaction with simple structures such as plates and shells. ConWep
is not guaranteed to give a 100% accurate result, but it has been compared to hand
A main disadvantage of ConWep comes from the fact that pressure curves can
only be obtained for one point at a time. Also ConWep has limited structural interaction
component.
3 Introduction to Hydrocodes
While the damage level produced by blasts is what makes them so critical for
examination, it is this same nature that makes experimental studies expensive and
difficult. In addition, the dynamic, time dependent nature of the loads produced by blasts
increases the complexity further, especially when compared to simple static loads. Large
scale tests can require millions of dollars in investments (Zukas, 2004). Therefore,
anyone doing experimental blast research needs an almost bottomless source of funding.
For the purposes of this project, the costs of destroying a bridge or even simple structure,
as well as having the permission to do so, make such testing out of reach. Therefore,
One such alternative has been made possible through advanced computer
programs called hydrocodes. What is a hydrocode and what is it used for? Zukas defines
a hydrocode as “a computer program for the study of very fast, very intense loading on
therefore ignoring material strength, which is the origin of the term hydrocode. This
method was used because the pressures generated by experiments often greatly surpassed
the strength of the materials (Zukas, 2004). Also, while many of the calculations
performed by hydrocodes could be done by hand or even with the use of a calculator, the
shear number of calculations involved in even simple problems makes the use of
Modern hydrocodes, including AUTODYN do a great deal more than model the
hydrodynamic behavior of materials, but the name has stuck. They can make use of a
variety of methods to model different material behaviors. In addition to their use in blast
modeling, hydrocodes have been used to evaluate structures for aircraft impacts, to
simulate vehicle crashes and even design sports equipment (Zukas, 2004).
The following sections will detail some of the methods hydrocodes use in
The systems used for modeling in hydrocodes are based on finite element and
finite difference techniques. How these techniques are implemented may vary, just as the
computer code used to write these programs will not be exactly the same. However,
there are many similarities due to their common basis. It is these common aspects which
will be discussed below. Usually the first step in modeling is to divide the problem up
into a finite system of nodes and elements as seen below in Figure 3.1. The configuration
of these systems, as well as how properties such as mass, energy and material strength are
dealt with is the main way of distinguishing between various methods. Lagrangian codes
and Eulerian codes are the two basic methods which are implemented in hydrocodes such
as AUTODYN.
(EOS) which are based in classical continuum dynamics. An EOS “relates the density (or
volume) and internal energy (or temperature) of the material with pressure”(Anderson,
1987). It does this by applying the principles of conservation of mass, momentum and
energy. For example, uniform gases would typically be modeled with an EOS based on
the Ideal Gas Law. Other relationships help to describe the nature of the material to be
modeled by relating the stress and strain to each other based on material properties.
These can incorporate strain rate, work hardening, thermal softening and other things
which can affect material properties and behavior. A typical order for calculations can be
Using these relationships, the modeling program will advance the calculations
forward a short period, called a timestep, and then perform this sequence of calculations
again. Since the timestep is an important variable, the program will often have a method
to calculate it on its own. This calculation incorporates the speed of sound in the material
(soundspeed), element size and some type of safety factor to prevent the timestep from
becoming to large (Zukas, 2004). Smaller safety factors result in smaller timesteps and
therefore more accurate the calculations. Smaller timesteps will require the performance
of more calculations to reach the same point in time. Therefore element size not only
As has already been mentioned, the setup of the grid of elements and nodes is
critical to the modeling process. Generally, smaller element sizes will allow for more
accurate calculations, while larger element sizes sacrifice accuracy for rapidity and
simplicity in the calculation process. Size is not the only factor to consider in mesh
generation though. This is where the differences in structured and unstructured solvers
become involved.
which form the elements and nodes are set up in a way to ease calculations with a regular
numbering scheme (Mathis et al., 2005). In addition, the elements formed will tend to be
quadrilaterals (at least in 2D). AUTODYN, for example, uses a coordinate system in
which each line is assigned a number in structured parts (Figure 3.3). In fact, any part
progressive calculations easier to perform, since related nodes and elements are next to
Unstructured meshes lack organization. The lines which form the elements and
nodes do not have a regular numbering scheme. The elements formed can have any
meshes seems to be their ability to accurately represent both the surface geometry and
macrostructure of a material (Mathis et al., 2005). However, since they lack the
organization of structured meshes, more time is spent determining which nodes and
elements are related. The only way to use an unstructured mesh in AUTODYN is to
or macrostructures are involved, while structured grids are preferred when the geometry
is simple.
As mentioned above, Lagrange solvers are one of the basic models used in
hydrocodes. One of the distinguishing features of a Lagrange code is that the grid it uses
is created so that cell boundaries occur at free surfaces and material boundaries. Another
is that during calculations, the mesh will distort to match the distortion of the material, as
seen in Figure 3.5 below. In a typical Lagrange mesh, coordinates, velocities, forces, and
masses are associated with the corner nodes, while stresses, strains, pressures, energies
and densities are centered within the cells (Birnbaum et al., 1999).
involved. Severe distortion of the mesh can result in inaccuracies, negative densities and
extremely small timesteps (Figure 3.6). In order to avoid this it is may be necessary to
eliminate the overly deformed cells by manually redrawing the mesh or “rezoning”
(Birnbaum et al., 1999). Therefore they are typically not used for models which involve
Figure 3.6 – Example of Normal Mesh and (a)-(d) Examples of Problematic Mesh Distortion
(AUTODYN, 2005b)
One of the advantages of a Lagrange mesh is that it moves with the material.
Lagrange solvers are often used in impact models (where two solid objects collide), as
Euler solvers, the other basic hydrocode model, differ from Lagrangian solvers in
a few basic ways. Instead of confining the grid to only the objects being modeled, Euler
solvers place a grid over the space in which the materials can move. As the calculation
progresses, the material of interest will move while the grid remains stationary (Figure
3.6). Individual nodes and cells are basically observing as the material being modeled
flows by. In a typical Euler model, the centers of the cells are used as interpolation
points for all variables, unlike Lagrange models as described above (Birnbaum et al.,
1999).
The main problems with Euler codes are with the amount of elements they
require, and their poor handling of geometry. Since you are not only modeling the object
of interest, but the space around that object, more elements and therefore more memory
and more time can be required than a standard Lagrange model. Also since the grid does
not distort with the object of interest, it becomes more difficult to track the various
components of a part, and therefore observe how a single piece behaves over time.
Therefore, Euler models are typically not used to model solid objects.
The advantage of Euler solvers is that they do not deform and therefore are not
subject to the limitations imposed by deformation in Lagrange solvers. They can also
allow the mixing of different materials inside the cells. Therefore the shape of material
surfaces is not completely limited by element size. They are used when a problem
involves high levels of deformation or fluid flow (i.e. gases and liquids), while Lagrange
solvers are normally used to model solids which don’t experience severe deformation.
Since the two basic solvers listed above each can have difficulty in modeling
These alternatives become particularly useful in modeling situations in which Euler and
Arbitrary Lagrange Euler (ALE) solvers are a combination of the two basic
solvers. They make use of an “automatic rezoning” technique in which the deformation
of the grid is limited (Birnbaum et al., 1999). The different parts of an ALE will act more
like a Lagrange or more like a Euler solver depending on the limitations the user puts in.
The main problems with ALE solvers are the amount of user input required and their
handling of contact surfaces. They perform well in modeling solids, fluids or gases so
Structural solvers (note this is a different term from the structured solvers
mentioned earlier), such as STAAD, which are made to handle beams, rods, and shells,
can also be used. These models are formulated to deal with specific geometries and
therefore handled the calculations more easily than a Lagrange or Euler solver could.
Shell solvers, for instance, are designed to handle thin structures. Therefore, it is
assumed to be in a biaxial state of stress, ignoring the component along its thickness, and
the timestep is only controlled by the length of the cells (Birnbaum et al., 1999). Since
the timestep just depends on the length, the program can run through the calculations
There are other gridless techniques which avoid the issues that occur with large
deformations. Smooth Particle Hydrodynamics (SPH) is one such technique which has
its basis in Lagrange solvers (meaning its nodes move as the part moves). The SPH
solver does not use cells or elements. Instead, as the name implies, SPH materials are
treated as if they are made up of a group of particles. This is similar to what a Lagrange
completely unstructured solver. An example can be seen below (Figure 3.7). The main
disadvantage of SPH solvers is that they are a relatively new method, and therefore less
Various programs have their own particular way of putting together and
implementing the techniques mentioned above, while retaining many of the basic
principles. The main benefit that can be attributed to AUTODYN (at least among
hydrocodes) is its graphical user interface (GUI) for creating the models, running the
simulations and observing the results. Other programs can model complex situations just
as well as, or even better than, AUTODYN. However, for users not accustomed to
inputting and receiving data via lines of code, AUTODYN becomes a useful tool. This
section will detail some of the specifics of using AUTODYN in general terms, without
The Materials button gives access to a range of options which basically allow the
user to specify which types of materials they wish to model and what type of situation is
expected. The user can input the information for a new material or make use of the
Here it is very important to stress that the user understand what type of behavior
they expect the material to encounter and therefore understand which particular material
model(s) may be appropriate. Real life data, including duplicable experiments, is critical
to establishing accurate models. In the Modify section, the user can specify the equation
of state for the material and specify the appropriate variables for that substance. This
allows the user to determine whether the material is a gas, metal, porous material,
explosive, polymer, etc and input the properties of that particular substance. A strength
subsection lets the user input properties of a structural nature such as tensile strength.
The Failure subsection deals with how that particular material will fail as well as
how AUTODYN will display that failure. Of particular importance in this case is how
the model deals with damage. Concrete for example has its own particular failure model
which can make use of a crack softening model in addition to other failure modes.
Damage in concrete is rated on a scale from 0 to 1 in relation to the strain at failure, with
0 being a fully intact cell and 1 being a fully failed cell (AUTODYN, 2005b). Users can
employ a visualization technique in Plots under Mater Status which will display which
The next subsection deals with erosion of the material. It allows the user to
determine whether or not an element will erode and at what point it will do so. When an
element reaches a specified strain limit, it is “eroded.” This means that the element is
“either discarded, or…transformed from a solid element to a free mass node disconnected
from the original mesh”(Birnbaum et al., 1999). Basically, the element breaks down and
either disappears or breaks off from the rest of the part and behaves like a particle (Figure
3.9). This option is noteworthy because it is one method the program uses to prevent the
errors that normally result when Lagrange parts suffer from large distortions.
Figure 3.9 – Example of Erosion, Note: Eroded nodes are in yellow (Birnbaum et al. 1999)
In most of these sections the user can define their own governing equations or
constants for the material. So it seems that there is a great deal that AUTODYN can do
3.2.2 Parts
The Part section of AUTODYN is actually where most of the “modeling” takes
place. A part is basically any distinct component of the system the user wishes to model.
Other sections will allow the user to define things such as initial conditions and
boundaries, but they won’t actually apply it to the model until they are assigned in the
parts section.
In the Part section, the user creates a part by either going through the process of
importing an object from another program, or by creating a whole new part. To create a
new part, first, the user specifies the type of solver they wish to use (i.e. Lagrange or
Euler). Next, the use creates the overall shape and mesh, which is the system of elements
and nodes which make up the part. The last thing the user does is to decide what type of
As already mentioned, AUTODYN does allow the user to import parts from other
programs such as LS-DYNA. The manuals that come with the program provide the user
with examples as to how this is done. However, this further complicates the process,
since the user would need to learn another program. Since one of the benefits of
AUTODYN is the comparative ease in which models are created, it seems unlikely that
an inexperienced user would prefer another program. Therefore, unless the user is more
experienced with the use of another program or wishes to make an unstructured part, as
different ways. For instance in order to create an explosion from a charge it is necessary
to use a Multimaterial Euler solver, which can be used with most materials instead of the
Ideal Gas Euler solver, which can only be used with gas materials. Another example
would be the case of boundary conditions. Euler solvers automatically treat their
wishes to allow material and other data to flow out of an Euler grid.
AUTODYN also gives the user the option of collecting data over time for certain
locations via gauges, while automatically doing so for parts or for materials as a whole.
Gauges can be placed anywhere on the model and can either be fixed or moving. A fixed
gauge will stay in one spot during the calculation no matter what the parts or materials
are doing. Meanwhile a moving gauge will stay fixed to the material or element it starts
at and record data as that material moves and deforms. Here it is important for the user to
know what type of data they are interested in since certain variables require the user to
AUTODYN also has the capability to “remap” the results of one model onto
another. This is beneficial since the user can take the results from a small part with a
relatively fine grid and basically load data such as pressures, and velocities into a larger
part with a coarser grid. Therefore, this larger model should then have more accurate
results than if it was run from the same starting point as the smaller model. Yet there are
limits to AUTODYN’s remapping capabilities. It only works for certain solvers, and
anything but Euler Ideal gas. How well remapping actually performs is evaluated later,
computer with 3.25 GB of RAM and a 3.60GHz Xeon CPU. Any attempt to make a part
with more than about 2 million elements either caused the program to shut down with a
models with parts approaching 2 million elements left the user with limited visualization
options. Basically, clicking rather simple options in the plots section such as contours,
grid, or nodes and elements, caused the program to either freeze with a “Memory
allocation error” or be so slow that the user must sit and wait before they can do anything.
Also, it should be noted that the slide and movie generation capabilities of
AUTODYN are fairly inconsistent from day to day. Often attempts to create an image or
animation from AUTODYN results simply produce fuzzy pictures, while other times
these built in functions will work fine. Usually, there is no obvious reason as to why
Run times for AUTODYN models also vary a great deal depending on the
complexity of the model. A simple wedge model could probably run for thousands of
AUTODYN does offer the option of using “parallel processing” which would
allow a network of computers to work on solving a single problem. This would allow
calculations to be performed more rapidly, and thus the running of the model would take
less time. However, it appears that the limiting factor is in the creation and visualization
of the model and the displaying of results, which can only be done on a single computer.
Therefore, it seems unlikely that “parallel processing” would be beneficial at this point in
time unless the user desires to run a particularly complicated model for an extended
period time.
This section will discuss the development of the AUTODYN models used in this
research. It will also cover the results of those models and their implications for
The standard AUTODYN model that was used is seen below in Figure 4.1 while
exact dimensions can be seen in Figure 4.2. This model was developed to be
thick slab that is fixed on both ends. The charge is based on a 100lb TNT charge as
representation of a vehicular bomb at a standoff distance of about 4 feet. The charge was
1.63g/cm3. Note that the center of the charge is located at the origin and is also the point
of detonation.
Concrete Slab
Air
TNT
Y
X
1700 1700
2440
1220 1220
2000
1220
300
4000
300
2000
Figure 4.2 – Standard Slab, Air and Charge Model (Dimensions in mm) (Robert, 2007)
In each instance the slab, air and charge were made out of square elements. The
air part that surrounds the slab was created to make sure that the explosion would be able
to take place, and interact with the entire slab. It was modeled as a Multimaterial Euler
part so that the explosion could be generated from a charge. The air is also at standard
atmospheric pressure, and about 15 degrees Celsius at the start of the calculation. The
boundary condition transmit was applied to all four sides of the air part to allow the flow
The only 2D slab model that had different dimensions for the air part consisted of
2mm elements in the slab and 2mm elements in the air. In that model the air of the lower
half is shortened to only 400mm below the origin and only 1600mm on each side. A 3D
model made of 20mm elements in the slab and in the air was also generated which was
shortened to only 1000mm below the origin. This was done to keep the model size under
2 million elements in each case. Models which had a remapped wedge simply did not
have the regular 300mm charge in the model. The remapped wedge itself is an axial
symmetric Multimaterial Euler part 1200mm long with the origin set as the detonation
point.
The concrete slab was modeled using the standard material properties of 35MPa
2.314g/cm3, and porous soundspeed of 2920m/s. It should also be noted that the concrete
is not reinforced. The slab was created as a Lagrange part since AUTODYN allows
Euler parts to interact with Lagrange parts. In addition, the slab had boundary conditions
were applied to the end nodes. These boundary conditions were constant, zero velocities
#3 are each 100mm from the edge while Gauge #2 is in the center of the slab. All three
gauges are centered vertically. These gauges were placed so that there would be no
interference from the boundary conditions and so they would remain within the slab in
Figure 4.3 – (a) Moving Gauges in Slab (b) Fixed Gauges in Air
Some models also contain fixed gauges located at 1000mm (Gauge #4), 1200mm
(Gauge #5) and 1220mm (Gauge #6) vertically above the detonation point (Figure 4.3
(b)). These gauges were placed when it became apparent that the planar models were
producing peak pressures higher than anticipated by either past ConWep or hand
calculations (see Results and Discussion section). A full list of the relevant models is
provided below (Table 4.1). For more details on model generation see the Appendix.
Table 4.1 – Models
This section will discuss the results of the models run in AUTODYN and their
will also discuss some results of the attempts to use a different failure model (crack
softening). Finally it will also discuss the outcomes of the remapping attempts.
Initial models kept the air elements a constant size (5mm), while varying the size
of the slab elements. Figures 4.4 and 4.5 are graphs which display the pressure results for
Gauge #1 and Gauge #2 of these initial tests while Tables 4.2 and 4.3 compare the initial
peak pressures. Gauge #3 showed similar results to Gauge #1. Note that mmel stands for
millimeter elements.
These results seem to indicate that the calculations diverge as slab element size
increases. However, they do not do so to a significant degree until 20mm elements are
used. A cursory examination of the graphs also seems to indicate that larger slab
elements will smooth over the sharper peaks obtained from the finer meshes.
60000
5mmel air +
50000 2mmel slab
5mmel air +
5mmel slab
40000
Pressure (kPa)
5mmel air +
10mmel slab
30000
5mmel air +
20mmel slab
20000
5mmel air +
40 mmel slab
10000
0
0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80
Time (ms)
Figure 4.4 – Pressure vs. Time for Gauge #1, Same Air Element Size
250000
5mmel air +
2mmel slab
200000
5mmel air +
5mmel slab
Pressure (kPa)
150000
5mmel air +
10mmel slab
100000
5mmel air +
50000 20mmel slab
5mmel air +
0
0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60
40mmel slab
Time (ms)
Figure 4.5 – Pressure vs. Time for Gauge #2, Same Air Element Size
Table 4.2 – Gauge #1 Initial Peak Pressure, Same Air Element Size
Peak Pressure at Percent Difference (%) from
Model
about .43ms (kPa) 5mmel air + 2mmel slab
5mmel air + 2mmel slab 24297 N/A
5mmel air + 5mmel slab 23733 2.32
5mmel air + 10mmel
22984 5.40
slab
5mmel air + 20mmel
21927 9.76
slab
5mmel + 40mmel slab 20361 16.20
Table 4.3 – Gauge #2 Initial Peak Pressure, Same Air Element Size
Peak Pressure at Percent Difference (%) from
Model
about .3ms (kPa) 5mmel air + 2mmel slab
5mmel air + 2mmel slab 141527 N/A
5mmel air + 5mmel slab 144418 2.04
5mmel air + 10mmel
144754 2.28
slab
5mmel air + 20mmel
146853 3.76
slab
5mmel air + 40mmel
132268 6.54
slab
Three models used varied the air element sizes while keeping the size of the slab
elements constant (2mm). Figures 4.6 and 4.7 are graphs which display the pressure
results for Gauge #1 and Gauge #2 of these models while Tables 4.4 and 4.5 compare the
initial peak pressures. Again, Gauge #3 shows similar results to Gauge #1.
This data appears to show that as the element size of the air changes, the
calculations become very inconsistent. The peaks of the two models with smaller
elements are different magnitudes while the peak for the 10mm element air model
60000
50000
2mmel air +
40000 2mmel slab
Pressure (kPa))
5mmel air +
2mmel slab
30000
10mmel air +
2mmel slab
20000
10000
0
0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80
Time (ms)
Figure 4.6 – Pressure vs. Time for Gauge #1, Same Slab Element Size
250000
200000
2mmel air +
Pressure (kPa)
2mmel slab
150000
5mmel air +
2mmel slab
100000
10mmel air +
2mmel slab
50000
0
0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60
Time (ms)
Figure 4.7 – Pressure vs. Time for Gauge #2, Same Slab Element Size
Table 4.4 – Gauge #1 Initial Peak Pressures, Same Slab Element Size
Peak Pressure Percent Difference (%) from 2mmel air
Model
(kPa) 2mmel slab
2mmel air + 2mmel slab 20153 N/A
5mmel air + 2mmel slab 23733 17.76
10mmel air + 2mmel
24300 20.58
slab
Table 4.5 – Gauge #2 Initial Peak Pressures, Same Slab Element Size
Peak Pressure
Model Percent Difference (%) from 2mmel air
(kPa)
2mmel air + 2mmel slab 217839 N/A
5mmel air + 2mmel slab 141527 35.03
10mmel air + 2mmel
220626 1.28
slab
Comparing the Figures 4.6 and 4.7 to Figures 4.4 and 4.5 suggests that the size of
the air elements effects the calculations much more dramatically than the size of the slab
elements. This is unfortunate, since as the largest part the air already requires more
elements. If the air elements need to be fairly small in order to obtain accuracy, then this
will limit the size of the model that AUTODYN can make due to the memory issues
presented earlier. Therefore, it is critical to examine other variables in order to see what
level of effect this inconsistency in pressure may have on the ultimate behavior of the
slab. If this obstacle can be avoided, then it will make creating larger, more complicated
models easier.
The graph of the results of the deflection analysis can be seen in Figure 4.8. The
graph contains data from models with various element sizes in both the air and the slab.
Visual analysis of this graph seems to agree with results of the pressure analysis. While
the deflections for the models with the same air appear to be almost uniform, the
25
10mmel air +
2mmel slab
20
5mmel air +
40mmel slab
5mmel air +
15 20mmel slab
5mmel air +
10mmel slab
10
5mmel air +
5mmel slab
5mmel air +
5
2mmel slab
2mmel air +
2mmel slab
0
0 0.1 0.2 0.3 0.4 0.5 0.6
Time (ms)
Figure 4.8 – Deflection Comparison for the Back Center of the Slab
4.2.3 Effective Strain in the Slab
The results of the analysis of this third variable (effective strain) agree with those
presented by the analysis of pressures and deflections. Figure 4.9 and 4.10 are graphs
which display the pressure results for Gauge #2. The results for Gauge #1 and Gauge #3
agree with the results for Gauge #2. They all show that the size of the air elements
affects the AUTODYN models to a much greater degree than the size of slab elements.
0.1
0.09
5mmel air +
0.08 40mmel slab
0.07
5mmel air +
20mmel slab
0.06
5mmel air +
Effective Strain
10mmel slab
0.05
5mmel air +
0.04 5mmel slab
0.02
0.01
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Tim e (m s)
Figure 4.9 – Effective Strain vs. Time for Gauge #2, Same Air Element Size
0.1
0.09
0.06
2mmel air +
Effective Strain
2mmel slab
0.05
0.04
10mmel air +
0.03
2mmel slab
0.02
0.01
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Tim e (m s)
Figure 4.10 – Effective Strain vs. Time for Gauge #2, Same Slab Element Size
Further investigations into the effects of air on the modeling used the results of
the fixed gauges. The main focus of this is on the results produced by gauges located
1000mm from the charge center. The results of these studies can be seen below in Figure
4.11.
60000
40000
10mmel Air No Slab 2D
Pressure (kPa)
Planar Spherical
ConWep
20000
10000
0
0.19 0.21 0.23 0.25 0.27 0.29 0.31 0.33 0.35
Time (ms)
Figure 4.11 – Initial Pressure Results for 1000mm from Charge Center
The results of these tests were not promising at first but it may be better to explain
how and why the tests were performed, chronologically. First, the results of the “10mmel
air and 2mmel slab” model were compared to those from ConWep, the established blast
program mentioned earlier. Although the arrival times are fairly close, the peak pressure
from the AUTODYN model is 5 times as large as the ConWep value. The presence of
the slab could have caused the higher pressure if it built up on the surface, yet the results
of the 10mmel air model without the slab match the one with the slab almost perfectly.
The shape of the charge may have been another factor since ConWep normally deals with
spherical blasts. However, the results of a model with a charge with a circular cross
section show that, do not match either the square charge results or the ConWep results.
Figure 4.11 also shows the results of the simple 1D 10mmel wedge, which
actually matches the ConWep peak pressure nicely. This suggests that perhaps it is
simply the planar models in AUTODYN which have trouble matching the expected
results. Therefore, it seems important to study the effects of remapping a wedge onto a
planar model.
Figure 4.12 shows the results of a remap of a 1200mm long 1mmel wedge onto a
20mmel air and 20mmel slab model. Figure 4.13 is a comparison of the results of the
model with the remapped wedge with the ConWep and 10mmel wedge results at a
distance of 1000mm from the charge center. The wedge is remapped onto a model with
larger air elements than all of the other 2D planar models and yet the results of this model
match much better (~6.5% difference from ConWep) to expected results than those
models. This suggests that remapping is the may the best way to obtain accurate,
Figure 4.12 – Remap of Wedge onto 20mmel Air and 20mmel Slab Model
20000
18000
16000
10mmel 2000mm
14000
Wedge
Pressure (kPa)
12000
10000
ConWep
8000
6000
20mmel air + 20mmel
4000 slab Wedge Remap
2000
0
0.19 0.21 0.23 0.25 0.27 0.29 0.31 0.33 0.35
Time (ms)
Figure 4.13 – Comparison of 10mmel Wedge, ConWep and Remapping Results at 1000mm from
Center of Charge
First and foremost it appears that a fine mesh for the air is required to achieve
accurate results. However, due to the limits in model size imposed by AUTODYN, a fine
mesh will restrict models to small sizes. In addition it appears that may produce
inaccurate results when modeling a charge with a planar cross section. Remapping,
Future work with the intention of developing bridge models should look more
other materials, and parts such as the girders themselves. The options for failure models
for concrete and whatever other parts are brought in need to be further examined for
appropriateness and accuracy. Finally, real experimental data should be obtained to more
In addition to the advisor for this project Dr. Jennifer Righman, the author would
also like to thank Renee Robert and Evan Brodsky for working with him on this project,
as well as Dr. Jack Gillespie and Dr. Bazle Gama, of the University of Delaware Center
6 References
Anderson, Charles E. Jr. (1987). “An Overview of the Theory of Hydrocodes.” Int. J.
AUTODYN (2005a). “SPH User Manual & Tutorial: Revision 4.3.” Century Dynamics
Birnbaum, Naury K., Francis, Nigel J., & Gerber, Bence I. (1999). “Coupled Techniques
Hayhurst, Colin J. Clegg, Richard A., Livingstone, Iain H. & Francis, Nigel J. (1996).
were being created and run. There are notes for most models and recordings of errors,