Você está na página 1de 47

Lecture 01

Explicit Dynamics Basics

16.2 Release

Workbench LS-DYNA
(ACT Extension) Training
2016 ANSYS, Inc. November 25, 2016 0
Fields of application for explicit FE-programs
STATIC QUASI STATIC DYNAMIC

PUNCH

BLANK Blankholder

DIE

structural problems metal forming impact

SF=0 SF 0 S F = ma

IMPLICIT METHOD
EXPLICIT METHOD

2016 ANSYS, Inc. November 25, 2016 1


Why Use Explicit Dynamics?

Implicit and Explicit refer to two types of time integration


methods used to perform dynamic simulations
Explicit time integration is more accurate and efficient for
simulations involving
Shock wave propagation
Large deformations and strains
Non-linear material behavior
Complex contact
Fragmentation
Non-linear buckling

Typical applications
Drop tests
Impact and Penetration

2016 ANSYS, Inc. November 25, 2016 2


Why Use Explicit Dynamics?
Typical Values for Solid Impacts
VELOCITY LOW HIGH

Deformation Global Local

Response Time ms - s s - ms

Strain <10% >50%

Strain Rate < 10 s -1 > 10000 s -1

Pressure < Yield Stress 10-100 x Yield Stress

2016 ANSYS, Inc. November 25, 2016 3


Why Use Explicit Dynamics: Applications

Crashworthiness analysis
Full car crash
Car component analyses
Crash in ALL vehicle industries
Car
Truck
Bus
Train
Ship
Aircraft

2016 ANSYS, Inc. November 25, 2016 4


Why Use Explicit Dynamics: Applications

Manufacturing process simulation


Deep drawing
Hydro forming
Superplastic forming
Rolling
Extrusion
Stamping
Machining
Drilling

2016 ANSYS, Inc. November 25, 2016 5


Why Use Explicit Dynamics: Applications

Pipe whip (ANSYS News 3/93):


Impact of a pipe with a rotational velocity of 50 rad/sec
Extremely fast run time compared to implicit solution

2016 ANSYS, Inc. November 25, 2016 6


Why Use Explicit Dynamics: Applications

Stress wave propagation.


2500 solid elements.

2016 ANSYS, Inc. November 25, 2016 7


Why Use Explicit Dynamics: Applications

Bird-strike simulation
Simulate an impact from
a moving bird to the blades
of a rotator machinery

Bus roll-over simulation

2016 ANSYS, Inc. November 25, 2016 8


Theory

2016 ANSYS, Inc. November 25, 2016 9


1 DOF system equation of motion

k
p(t)
M
c equilibrium: fI fD fS p(t)
inertia force: f M
I
u(t) displacements
damping force: f Cu
D

inertia force fI elastic force: f K u


S

elastic force fS
external forces p(t)
M

damping force fD

equation of motion : M (t ) C u (t ) K u(t ) p(t )

equation of motion depends on time t time discretization necessary!


2 possibilities: implicit or explicit time integration

2016 ANSYS, Inc. 10-


November 25, 2016 10
Comparison explicit vs. implicit
implicit time integration e.g. Newmark-method

The equations of motion are evaluated at time tn+1 (i.e. at the end of the current time step)

Mn1 n1 Cn1 u n1 Kn1 un1 pn1

explicit time integration e.g. central difference scheme

The equations of motion are evaluated at time tn (i.e. at the begin of the current time step)

Mn n Cn u n Kn un pn

2016 ANSYS, Inc. November 25, 2016 11


Basic Formulation Implicit Dynamics

The basic equation of motion solved by an implicit transient dynamic


analysis is
mx cx kx F (t )
where m is the mass matrix,
c is the damping matrix,
k is the stiffness matrix, and
F(t) is the load vector.
At any given time, t, this equation can be thought of as a set of
"static" equilibrium equations that also take into account inertia
forces and damping forces.
The Newmark or HHT method is used to solve these equations at
discrete time points. The time increment between successive time
points is called the integration time step.

2016 ANSYS, Inc. November 25, 2016 12


Basic Formulation Implicit Dynamics

For linear problems:


Implicit time integration is unconditionally stable for certain integration
parameters.
The time step will vary only to satisfy accuracy requirements.

For nonlinear problems:


The solution is obtained using a series of linear approximations
(Newton-Raphson method), so each time step may have many
equilibrium iterations.
The solution requires inversion of the nonlinear dynamic equivalent
stiffness matrix.
Small, iterative time steps may be required to achieve convergence.
Convergence tools are provided, but convergence is not guaranteed for
highly nonlinear problems.

2016 ANSYS, Inc. November 25, 2016 13


Basic Formulation Explicit Dynamics

The basic equations solved by an Explicit Dynamic analysis


express the conservation of mass, momentum and energy in
Lagrange coordinates. These, together with a material model
and a set of initial and boundary conditions, define the
complete solution of the problem.
For Lagrange formulations, the mesh moves and distorts with
the material it models, so conservation of mass is automatically
satisfied. The density at any time can be determined from the
current volume of the zone and its initial mass:

0V0 m

V V

2016 ANSYS, Inc. November 25, 2016 14


Basic Formulation Explicit Dynamics

The partial differential equations which express the conservation of


momentum relate the acceleration to the stress tensor ij:
xx xy xz
x bx
x y z
yx yy yz
y b y
x y z
zy zz
z bz zx
x y z

Conservation of energy is expressed via:

e
1
xx yy yy zz zz 2 xy xy 2 yz yz 2 zx zx

xx

2016 ANSYS, Inc. November 25, 2016 15


Basic Formulation Explicit Dynamics

For each time step, these equations are solved explicitly for
each element in the model, based on input values at the end of
the previous time step
Only mass and momentum conservations are enforced.
However, in well posed explicit simulations, mass, momentum
and energy should all be conserved.

2016 ANSYS, Inc. November 25, 2016 16


Basic Formulation Explicit Dynamics

The Explicit Dynamics solver uses a central difference time


integration scheme. The semi-discrete equations of motion at
time n are:
Mx P F H
n n n n

where M is the diagonal mass matrix,


x are the components of nodal acceleration,
n

Pn are the external and body forces,


Fn is the stress divergence vector,
Hn is the hourglass resistance.
With the accelerations at time n - determined, the velocities
at time n + at direction i (i = 1,2,3) are found from
n 1 2 n 1 2
xi xi t n
n
xi

2016 ANSYS, Inc. November 25, 2016 17


Basic Formulation Explicit Dynamics

Finally the positions are updated to time n+1 by integrating the


velocities
n 1 n 1 2
x i xi t n1 2
n
xi
Advantages of using this method for time integration for
nonlinear problems are:
The equations become uncoupled and can be solved directly
(explicitly). There is no requirement for iteration during time
integration
No convergence checks are needed since the equations are uncoupled
No inversion of the stiffness matrix is required. All nonlinearities
(including contact) are included in the internal force vector

2016 ANSYS, Inc. November 25, 2016 18


Scheme of an explicit FE-program

loop over all time steps

2016 ANSYS, Inc. November 25, 2016 19


implicit vs. explicit time integration

Required hardware resources

Example: 90000 Shell-Elements = ca. 550000 Degrees of Freedom

Memory Requirement in MByte:

Implicit Front-Solver 48 MB Explicit 36 MB


Sparse 373 MB
PCG 414 MB

Disc Requirement in MByte:

Implicit Front-Solver 9260 MB Explicit 0 MB


Sparse 1930 MB
PCG 1060 MB

2016 ANSYS, Inc. November 25, 2016 20


Stability Time Step

To ensure stability and accuracy of the solution, the size of the time step
used in Explicit time integration is limited by the CFL (Courant-Friedrichs-
Levy[1]) condition.
This condition implies that the time step be limited such that a
disturbance (stress wave) cannot travel further than the smallest
characteristic element dimension in the mesh, in a single time step.
Thus the time step criteria for solution stability is
h
t f
c min
where t is the time increment, f is the stability time step factor, h is the
characteristic dimension of an element and c is the local material sound
speed in an element

2016 ANSYS, Inc. November 25, 2016 21


Stability Time Step

The element characteristic dimension, h, is calculated as follows


in LS-DYNA:

Hexahedral The volume of the element divided by the area of the


/Pentahedral largest side
Tetrahedral The minimum distance of any element node to its
opposing element face
Quad Shell The area of the element divided by the maximum edge or
diagonal length
Tri Shell The area of the element divided by the maximum edge
length
Beam The length of the element

[1] R. Courant, K. Friedrichs and H. Lewy, "On the partial difference equations of mathematical
physics", IBM Journal, March 1967, pp. 215-234

2016 ANSYS, Inc. November 25, 2016 22


Stability Time Step

The time steps used for explicit time integration will generally be much
smaller than those used for implicit time integration
e.g. for a mesh with a characteristic dimension of 1 mm and a material sound
speed of 5000 m/s. The resulting stability time step would be 0.18 -seconds. To
solve this simulation to a termination time of 0.1 seconds will require 555,556 time
steps
The minimum value of h/c for all elements
h
in a model is used to calculate the time step. t f
c min
This implies that the number of time steps
required to solve the simulation is dictated
by the smallest element in the model.
h
Take care when generating meshes for Explicit
Dynamics simulations to ensure that one or two
very small elements do not control the time step

2016 ANSYS, Inc. November 25, 2016 23


Stability Time Step and Mass Scaling

The maximum time step that can be used in explicit time integration is
inversely proportional to the sound speed of the material and therefore
directionally proportional to the square root of the mass of material in an
element
l n , min E
1 1 m t specified and c
t c (1 2 )
c Cii VC ii
(t specified ) 2 E
n
l n (1 2 )
2

where Cij is the material stiffness (i=1,2,3), is the material density, m is


the material mass and V is the element volume
Artificially increasing the mass of an element can increase the maximum
allowable stability time step, and reduce the number of time increments
required to complete a solution

2016 ANSYS, Inc. November 25, 2016 24


Stability Time Step and Mass Scaling

Mass scaling is applied only to those elements which have a


stability time step less than a specified value. If a model
contains relatively few small elements, this can be a useful
mechanism for reducing the number of time steps required to
complete an Explicit simulation
Mass scaling changes the inertial properties of the portions of
the mesh to which scaling is applied. Be careful to ensuring
that the model remains representative for the physical
problem being solved

2016 ANSYS, Inc. November 25, 2016 25


Mass Scaling

How to get a recommended mass scaling:


- Start the simulation for at least one time step, search for the 100 smallest timesteps
in file d3hsp:
100 smallest timesteps
----------------------
element timestep
shell 107136 0.57161E-03
shell 107679 0.57161E-03
shell 100631 0.60030E-03
shell 100659 0.60030E-03
...
shell 106479 0.73850E-03
shell 108365 0.74094E-03
shell 106739 0.74094E-03

- Difference between smallest and largest time step size in this list this is a criterion
for nonuniform mesh mesh may be improved (where are the smallest elements and
why?)
- Added mass must not be to large, otherwise a small element may get heavy nodes,
although the total mass increase is important. may yield to unrealistic stresses and
strains near small elements.

2016 ANSYS, Inc. November 25, 2016 26


Mass Scaling
Different possibilities to check the added mass:

a) total added mass over time for the whole model in ASCII file GLSTAT
(Added Mass)
b) Total percentage mass increase in mass for the whole model in ASCII file
GLSTAT (% Mass Increase) Attention: The existance of large rigid masses
may reduce the percentage mass increase significantly!
c) Added mass over time for each part in ASCII file MATSUM (Added Mass)
Attention: for Parts with many many elements this may not be very meaningfull.
d) Element based added mass as a Fringe plot inFcomp Misc - time step size
(only if *DATABASE_EXTENT_BINARY, STSSZ=3 is set).
e) Nodal added mass as Fringe plot inFcomp Misc - mass scaling (only if
*DATABASE_EXTENT_BINARY, MSSCL=1 is set)
f) Nodal percentage mass increase as Finge plot in Fcomp Misc - mass
scaling (only if *DATABASE_EXTENT_BINARY, MSSCL=2 is set)
recommended !!

2016 ANSYS, Inc. November 25, 2016 27


Elastic Waves

Different types of elastic waves can propagate in solids depending on


how the motion of points in the solid material is related to the direction
of propagation of the waves [Meyers].
The primary elastic wave is the longitudinal wave. Under uniaxial stress
conditions (i.e. an elastic wave travelling down a long slender rod), the
longitudinal wave speed is given by:
E
c0

For the three-dimensional case, additional components of stress lead to a


more general expression for the longitudinal elastic wave speed
K4 G
cP 3

2016 ANSYS, Inc. November 25, 2016 28


Elastic Waves

The secondary elastic wave is the distortional or shear wave and its speed
can be calculated as
G
cS

Other forms of elastic waves include surface (Rayleigh) waves, interfacial


waves and bending (or flexural) waves in bars/plates [Meyers]

Meyers M A, (1994) Dynamic behaviour of Materials, John Wiley & Sons, ISBN 0-
471-58262-X

2016 ANSYS, Inc. November 25, 2016 29


Plastic Waves

Plastic (inelastic) deformation takes place in a ductile metal when the


stress in the material exceeds the elastic limit. Under dynamic loading
conditions the resulting wave propagation can be decomposed into
elastic and plastic regions [Meyer].
Under uniaxial strain conditions, the elastic portion of the wave travels at
the primary longitudinal wave speed whilst the plastic wave front travels
at a local velocity
d
c plastic d

For an elastic perfectly plastic material, it can be shown [Zukas] that the
plastic wave travels at a slower velocity than the primary elastic wave, so
an elastic precursor of low amplitude often precedes the stronger plastic
wave
K
c plastic

2016 ANSYS, Inc. November 25, 2016 30
Comparison explicit vs. implicit
implicit time integration e.g. Newmark-method

The equations of motion are evaluated at time tn+1 (i.e. at the end of the current time step)
Characteristics : - equilibrium must be satisfied at time tn+1
- thus necessary to solve a large system of equations
- iteration within time step, convergence may be a problem
- few but large time steps
- time step size depending on frequencies of interest
- CPU time per time step depends on equation solver
- one step method, self starting

explicit time integration e.g. central difference scheme

The equations of motion are evaluated at time tn (i.e. at the begin of the current time step)
Characteristics: - equilibrium at time tn, non-equilibrium at time tn+1
- accelerations calculated to shift the system towards balance
- no large system of equations to solve
- usually no problems with convergence
- only conditional stable, time step must be small enough
time step size depends on highest natural frequency
- many but very small time steps
- two step method; not self starting

2016 ANSYS, Inc. November 25, 2016 31


Explicit versus Implicit
Factors Influencing Numerical Costs
Implicit Explicit

Model size (number of DOF) Model size (number of DOF)


Size respectively grade of nonlinearity Size of the critical time step
Number of time steps to simulate - characteristic element length
- sound of speed in the material
(Youngs moduli & density)
Size of the physical time window to be
simulated (termination time)

2016 ANSYS, Inc. November 25, 2016 32


Reduced Integration Formulation
A reduced integration element is an element which has a minimum
number of integration points. A reduced integration brick element has
one integration point at its centroid. A reduced integration shell has one
in-plane integration point, but still has multiple integration points through
the thickness of the shell.

Fully integrated elements are typical in implicit ANSYS. In ANSYS


WORKBENCH LS-DYNA, fully integrated brick elements have eight
integration points and fully integrated shells have four in-plane
integration points (with multiple points through the thickness).

Reduced integration saves CPU time by minimizing element processing.


Therefore, it is the default formulation most often used in ANSYS
WORKBENCH LS-DYNA.

2016 ANSYS, Inc. November 25, 2016 33


Reduced Integration Formulation

In addition to saving CPU time, single point integration elements are


also extremely robust in large deformation. ANSYS LS-DYNA elements
can undergo much greater deformations than standard ANSYS
implicit elements.

Two basic disadvantages of reduced integration elements are:


Deformations with zero energy modes are possible (Hourglassing).
The accuracy of stress results is directly related to the number of
the integration points.

2016 ANSYS, Inc. November 25, 2016 34


Hourglassing
Hourglassing is a zero-energy mode of deformation that oscillates at a
frequency much higher than the structures global response.
Hourglassing modes result in stable mathematical states that are not
physically possible. They typically have no stiffness and give a zigzag
deformation appearance to a mesh.
All under-integrated isoparametric elements (one Gauss point) have
hourglassing present. Single-point (reduced) integration elements
with linear displacement functions are prone to zero energy modes
(hourglassing).
Use default unless additional documentation is consulted. (e.g., see
Review of Solid Element Formulations Erhart.pdf)
If the overall hourglass energy is more than 10% of the internal
energy of a model, the results are suspect. Determining the level of
hourglass energy can be found from LS-DYNA ASCII output files
GLSTAT and MATSUM. In some cases, even an hourglass ratio of 5%
can be considered excessive.

2016 ANSYS, Inc. November 25, 2016 35


Hourglassing

Zero energy deformations for the one-point integrated solid element:

This mesh distortion produces no strain or volume change in the


mesh. Hourglass control brings additional stiffness or viscous
damping to minimize these non-physical, zero energy modes.

2016 ANSYS, Inc. November 25, 2016 36


Hourglassing
Minimizing hourglassing in ANSYS WORKBENCH LS-DYNA:
Avoid single point loads, which are known to excite hourglass
modes. Since one excited element transfers the mode to its
neighbors, point loads should not be applied. Try to apply loads
over several elements as pressures, if possible.
Refining the mesh often reduces hourglass energy, but a larger
model corresponds to increased solution time and larger results
files.
Use fully integrated elements, which do not experience
hourglassing modes. However, penalties in solution speed,
robustness, and even accuracy may result, depending on the
application. Alternatively, a few fully integrated seed
elements may be dispersed through the mesh to minimize
hourglassing.
beams are not affected by hourglassing. The higher order tet
element is not subject to hourglass modes, but it is not as
robust as the lower order tet .
2016 ANSYS, Inc. November 25, 2016 37
Hourglassing
Minimizing hourglassing in ANSYS WORKBENCH LS-DYNA
(continued)
Globally add elastic stiffness to reduce hourglass energy. This
can be done for the entire model by increasing the
hourglassing coefficient in Hourglass Control defined in
Analysis Settings:
Stiffness hourglass control is
recommended for problems
deforming with lower
velocities (e.g., metal forming
and crash).

Care should be used when increasing the hourglassing


coefficient. Values above 0.15 have been found to over-
stiffen the models response during large deformations and
cause instabilities.

2016 ANSYS, Inc. November 25, 2016 38


Hourglassing
Minimizing hourglassing in ANSYS WORKBENCH LS-DYNA (continued)
Locally reduce hourglassing in high risk areas of a model without
dramatically changing the models global stiffness. The added
Hourglass Control by Body is used to apply hourglass control only to a
specific material.

LS-DYNA locally applies


hourglass control on a Part
ID basis (not on a material
basis), so any Part with the
specified material will have
this hourglass control.
LS-DYNA ID 5 is often used
to reduce hourglassing.

2016 ANSYS, Inc. November 25, 2016 39


Control Hourglass Deformation
In order to avoid such hourglass instabilities, a set of corrective forces are
added to the solution
The corrective forces are called as Hourglass Damping
Always recommended for reduced-integrated solid/shell elements
To specify Hourglass locally or Globally
Recommend stiffness hourglass control,
LS-DYNA ID=4, with hourglass coefficient
QM = 0.03 for metal and plastic parts.
Recommend viscosity-based hourglass control
for foams and rubbers (LS-DYNA ID =2 or 3) or
hourglass formulation 6
In soft materials, stiffness-based hourglass
control causes overly stiff response even
with a reduced hourglass coefficient.

2016 ANSYS, Inc. November 25, 2016 40


Control Hourglass Deformation

Always check hourglass energy from Material output


(MATSUM) and Global data (GLSTAT)
The Hourglass Energy should be much less than the Internal Energy
If hourglass energy is very high, consider to
Refine the mesh in your model
Re-run the model in double precision

2016 ANSYS, Inc. November 25, 2016 41


Hourglass control
Recommendation for *HOURGLASS and. *CONTROL_HOURGLASS

for shell elements ihq=4 (stiffness form, default settings)


for solid elements (in general) ihq=5 (stiffness form, default settings)
for solid elements (foam, honeycomb) ihq=3 (viscous form, default settings)
for solid elements (elastic) ihq=6,qm=1.0 (stiffness form)
for solid elements (plastic) ihq=6,qm=0.01-0.001 (stiffness form)
for solid elements (rubber, viscoelast.) ihq=7,qm=1 (stiffness form)

Note: ihq=6,7 is a special solid element formulation according to Belytschko-Bindemann


Danger: Default ihq=1 it not orthogonal to rigid body rotation do not use !

2016 ANSYS, Inc. 42-


November 25, 2016 42
Hourglass Control for Solid
3 forms of viscous HG control for solids
Hourglass coefficient should be less than 0.15
Type 1: standard (cheapest)
Type 2: Flanagan-Belytschko (default)
Type 3: Flanagan-Belytschko with exact volume integration
(better for skewed elements)
2 forms of stiffness HG control for solids
Hourglass coefficient should not exceed 0.1; 0.03 is better
Type 4: Flanagan-Belytschko
Type 5: Flanagan-Belytschko with exact volume integration
Types 6, 7, 9, 10 (see next 2 slides)
Hourglass coefficient can range from 0.1 to 1.0

2016 ANSYS, Inc. November 25, 2016 43


Why use Explicit?

No convergence problems in highly nonlinear


problems
No equilibrium iteration needed
Material failure and erosion easy to model
High frequencies are naturally resolved because of
small time steps
Implicit-explicit switching capability for efficiency
Suited to a wide range of complex nonlinear problems

2016 ANSYS, Inc. November 25, 2016 44


Extend the Range of Structural Problems
Drop test simulations
(short time dynamic range, high frequencies)
Problems including complex contact situations
(large geometrical nonlinearities)
Problems including sophisticated material damage and failure
(large nonlinearities, element erosion)
Load limit analyses
(large deformations, large nonlinearities)
Manufacturing simulations
(large deformations, large nonlinearities)
High-speed Dynamic analyses
(failure, fragmentation, blast wave-structure interaction)

2016 ANSYS, Inc. November 25, 2016 45


Workshop 1 time step & Hourglass

Goal:
Understand effects of Time Step & Hourglass

Walkthrough
2016 ANSYS, Inc. November 25, 2016 46

Você também pode gostar