You are on page 1of 20

Non-linear transient dynamics analysis

DYNA_NON_LINE

Code_Aster, Salome-Meca course material


GNU FDL licence (http://www.gnu.org/copyleft/fdl.html)

Introduction to non-linear transient computing


for structural dynamics
Different kind of nonlinearities
Constitutive laws, large displacements, contact…

Spatial description
Direct (physical DoF) or modal projection

Nonlinear direct dynamics in Code_Aster

Syntax of the DYNA_NON_LINE operator

Differences between STAT_NON_LINE and DYNA_NON_LINE

Damping representation

Some advices for a proper use of DYNA_NON_LINE

Numerical applications
2 - Code_Aster and Salome-Meca course material GNU FDL Licence
Different sort of non-linearities (1)

Contact-friction
Single DoF oscillator with perfect plasticity constitutive relation

k M Fext
x  Fext  sign( x) * min( k . x , Fs )
M. 
F < Fs
Shock oscillator


M .x  Fext  k .x  kc x  gap Fext

Large transformations: pendulum


 3 5
I . 
   M . g.sin    M . g.(   ...)
6 5!

3 - Code_Aster and Salome-Meca course material GNU FDL Licence

Different sort of non-linearities (2)

Nonlinear constitutive relations


f
Plasticity (steel) d  d e  d p d p  d 


Viscoplasticity (steel)
Norton F
Hoff - Rabotnov - Lemaître  p   or  p  G  , Vi , T 


Damage (concrete)

E  E1  D with D
  f F , D,V 
Rabotnov - Kachanov ~
Chaboche ext i

4 - Code_Aster and Salome-Meca course material GNU FDL Licence


Different sort of non-linearities (3)
Nonlinear behavior for Civil Engineering dynamics computations
Global law for reinforced concrete shell elements: GLRC (R7.01.32)
Parameters identification: DEFI_GLRC (U4.42.06)

F
 p   or  p  G  , Vi , T 


Compatible with excentered reinforcements finite-elements

Beams and columns: PMF (multifiber beam elements) with suitable constitutive relations
(MAZARS, VMIS_CINE_GC…)

5 - Code_Aster and Salome-Meca course material GNU FDL Licence

Different sort of non-linearities (4)

Large transformations
Large displacements: Green-Lagrange strain tensor (when  > few %)

1  u u j uk uk 
ij   i   
2  x j xi xi x j 
.

Several formulations (unlike small perturbations)


Lagrangian, based on the Green-Lagrange strain tensor ans the second Piola-Kirchhoff stress tensor
Eulerian, based on the Almansi (strain) and the Cauchy (stress) tensors
Updated lagrangian formulation (for fast transient dynamics): simple but can be inaccurate (curvature effects in
shells / membrane effects)

6 - Code_Aster and Salome-Meca course material GNU FDL Licence


From linear to non-linear analysis

Governing FE equations, at each time step:


M.x  C.x  K T .x  F ext

Tangent stiffness operator KT (constitutive relations): nonlinear

Cut-off frequency can be difficult to define


Not only dependent of sollicitations and linear eigenmodes

Eigenfrequency are variables


Nonlinear modal analysis

7 - Code_Aster and Salome-Meca course material GNU FDL Licence

Numerical methods for non-linear analysis

Direct transient response with DYNA_NON_LINE


Localized non-linearities: shocks, friction
Limited size of the discretized system (< 500,000 DoF)
Nonlinearities as internal forces:
Non-linear constitutive relations: plasticity, damage…
full implicit representation
Geometric nonlinearities: large displacements
Implicit time integrators: Newmark family, HHT, Krenk, -method
Explicit time integration schemes: central difference, Tchamwa-Wielgosz

Modal transient response with DYNA_VIBRA


Only localized nonlinearities (quasilinear system)
Low frequency responses
Nonlinearities in right hand terms of
Modal reduction for fast computation
equations: explicit representation
Explicit time schemes: Euler, De Vogelære, adaptive
Implicit time schemes: Newmark

8 - Code_Aster and Salome-Meca course material GNU FDL Licence


Numerical methods for nonlinear analysis (1)

Modal decomposition (RITZ)


Moderate and localized nonlinearities (most of the structure remains linear)
Low frequency phenomenon

Direct transient method (test-case sdld31a)

Implicit time schemes: often used


Global and strong nonlinearities
Medium to large size (available parallelized solvers) and medium frequency phenomenon

Explicit (fast transient dynamics): specific computations with Code_Aster


Global and strong nonlinearities, except some contact algorithm
Medium sized problems (suboptimal code optimization for explicit) and high frequency phenomenon (wave
propagations)
When implicit solving does not converge

9 - Code_Aster and Salome-Meca course material GNU FDL Licence

Numerical methods for nonlinear analysis (2)

b=(g+1/2)2/4
NEWMARK time integration family b

p  2 1  
 U n 1  p U n  bt 2 U
 ,  U n  U n  t U n  t   b  U n ,
 n 1
2 
Unconditional
 avec  stability
 U
 p U   gt U  , p
 U n  U n  t 1  g U n .
n 1 n n 1   
Unstable
zone
Conditional
Average acc.
stability
Linear acc.
Fox Goodwin
Central diff.
0 g=1/2 g
Linear equilibrium equation
Displacement Speed Acceleration
X  U n 1 
XU n 1

XU n 1

 M  b t K 2
 M  bt 2 K  A  M  bt 2 K ,
A  , A  , 

 bt 2 gt
  B  F ext
 n 1  K U n .
p

 1  1
B  F n 1  b t 2 M U n .
ext p p

B  F n 1 K U n  gt M U n .
ext p

 
10 - Code_Aster and Salome-Meca course material GNU FDL Licence
Numerical methods (3)
Implicit direct transient dynamics
HHT scheme (Hilber-Hughes-Taylor 1977)
Numerical dissipation in high frequency domain, second order accuracy
Based on a Newmark scheme: modified average acceleration (first order)

b
1   2 1
g   0    0,3
4 2
Modification of equilibrium equation: “average” between t n and tn+1 :
  1   F  F  1   F  F
MU
int int ext ext
n 1 n 1 n n 1 n

Linear equilibrium solving:


AU n 1  B with :
 M  bt 2 K 1
 Newmark A  B  F n 1 
ext p
M U n,
bt 2 bt 2

 bt 2 1   K
B  1   F n 1 
 HHT M 1
A M U n   F n  KU n .
ext p ext

 bt 2 bt 2

Krenk scheme: similar to -method (order 1)


Parameter k ~ 2 . Q (no dissipation for k = 1, increasing with k)

11 - Code_Aster and Salome-Meca course material GNU FDL Licence

Numerical methods (4)


Implicit direct transient dynamics
Numerical damping due to time integration scheme (test-case
sdld31a)

12 - Code_Aster and Salome-Meca course material GNU FDL Licence


Numerical methods (5)
Implicit direct transient dynamics
Solving with nested loops algorithm
Time loop (like linear case)
NEWTION iterations (like STAT_NON_LINE operator)

Equilibrium verified at each time step

Explicit scheme: the NEWTON loop disappears

Constitutive relation solving (local loops: at each Gauss point)

Itérations à matrice constante Itérations à matrice tangente

K (Xit+dt ) +  K(Xit+dt ). X = R
Xi+l t+dt = Xit+dt + X

Repeat until convergence Non convergence des itérations à Convergence des itérations à

K (Xit+dt). X = R - K (Xit+dt)


matrice constante matrice tangente

Xi+lt+dt = Xit+dt + X

13 - Code_Aster and Salome-Meca course material GNU FDL Licence

Numerical methods (6)


Explicit direct transient dynamics
Time integration schemes and solving method
Central difference (from Newmark family with b= 0 and g = 1/2):
no dissipation
Conditional stability: critical time step (CFL condition)

tcrit = 2 / wmax with wmax : higher eigen pulsation of the discretized system

Other interpretation: tcrit ~ l min EF / c with:

l min FE : smallest caracteristic length

c : wave celerity (traction : c2 = E / r)

Tchamwa-Wielgosz: HF dissipation (like HHT)


Parameter: f = 1.05 (default value)
= 1 : no damping (but not equivalent to central difference)
Acceleration is the primal unknown for equilibrium resolution
Solving operator = mass matrix (lumped for numerical efficiency)
14 - Code_Aster and Salome-Meca course material GNU FDL Licence
DYNA_NON_LINE operator syntax
Like STAT_NON_LINE, DYNA_NON_LINE arguments are non-assembled matrices and vectors,
unlike linear dynamics operators (DYNA_VIBRA)

Before DYNA_NON_LINE

Boundary conditions definition

Constitutive relation: INCREMENT (COMP_INCR)

Strain tensor: PETIT / PETIT_REAC / GROT_GDEP / SIMO_MIEHE /

GDEF_HYPO_ELAS / GREEN_REAC / GDEF_LOG

Initial conditions

Options for the Newton algorithm

Convergence criterion

Solver choice (direct or iterative / sequential or parallel)

Time integration scheme

(Eigenvalues calculation on tangent updated matrices)

Options for results storage

After DYNA_NON_LINE
15 - Code_Aster and Salome-Meca course material GNU FDL Licence

Damping representation (1)

Rayleigh damping (for each material)

C = .K + b.M (well suited for linear modal dynamics)


Defined with DEFI_MATERIAU and AMOR_ALPHA =  , AMOR_BETA = b
Relation with modal damping coefficient: 2 x (w) =  / w + b . W
Parameters can be fitted with 2 methods
1. Mean value xbetween w1 and w1
2. Enforcing xvalue at w1 and w2

Choice for the stiffness matrix K:


AMOR_RAYL_RIGI = 'TANGENTE' (default) or 'ELASTIQUE'
ELASTIQUE: for elastic matrix: keeps constant Rayleigh damping (best choice for GLRC)
TANGENTE: for tangent matrix: Rayleigh damping decreases when nonlinearities appear, due to constitutive
relation
16 - Code_Aster and Salome-Meca course material GNU FDL Licence
Damping representation (2)

Modal damping: N mod


C  a  K  K 
i 1
i i i
T
with: ai = 2 xi / (ki/wi)

• Syntax
AMOR_MODAL(
MODE_MECA = mode, (eigenmodes)
AMOR_REDUIT = l_amor, [l_R] (xi list)
REAC_VITE = ‘OUI’, [DEFAUT] (update at each Newton’s iteration, or not)
)

Remarks
• Modal analysis on the linear system needed before NL calculation

• Modal damping terms are explicited in equilibrium equations: time step


may have to be reduced in order to insure stability, even with an
implicit time-schemes like Newmark

17 - Code_Aster and Salome-Meca course material GNU FDL Licence

Damping representation (3)

Localized dampers: dashpots


Affected on discrete elements (POI1 or SEG2): in AFFE_MODELE
MODELISATION = ‘DIS_T’ / ‘DIS_TR’

Damping values with AFFE_CARA_ELEM, keyword: DISCRET


CARA = ‘A_T_D_N‘ / ‘A_TR_D_N’ / ‘A_T_D_L’…

Remark
Those dashpots are taken into account in DNL only if AMOR_ALPHA is defined
(even if its value is 0), except in NEW11 release (11.2.7 version or above)

Absorbing boundaries (half-space media)


Defined on some boundaries, using AFFE_MODELE
MODELISATION = '3D_ABSO'

18 - Code_Aster and Salome-Meca course material GNU FDL Licence


Damping representation (4)
Numerical damping due to time integration scheme

Newmark: modified average acceleration (HHT and


MODI_EQUI='NON')
g12/4;d1/2 
Parameter:  (ALPHA = -0.3 as default value)
 = 0: average acceleration (NEWMARK): undamped and 2nd order accuracy
Damping increases when  decreases and only first order accuracy

Full HHT (HHT with MODI_EQUI='OUI')


Same parameter:  (ALPHA = -0.3 as default value)
 = 0: average acceleration (NEWMARK): undamped and second order
Damping increases when  decreases and remains second order

19 - Code_Aster and Salome-Meca course material GNU FDL Licence

Damping representation (5)


Numerical damping due to time integration scheme (cont.)

Q-method or Krenk (THETA_METHODE or KRENK)


Parameters: k ~ 2 . Q (undamped if k = 1, damping increasing with k)
Well suited for nonregular problems (first order accuracy)

Tchamwa-Wielgosz (explicit):
Parameter: f (PHI = 1.05 as default): no dissipation if PHI = 1

20 - Code_Aster and Salome-Meca course material GNU FDL Licence


Damping representation (6)
Numerical damping due to time integration scheme (cont.)

Damping representation for a 1 DoF linear system


Idea: HF damping and no damping in LF range
Complementary to structural damping (Rayleigh…)

21 - Code_Aster and Salome-Meca course material GNU FDL Licence

Some advice for a proper use of


DYNA_NON_LINE
Read Reference documentations (at least R5.05.05)

Read documentation U2.06.13

If loadings are time dependent (for instance: accelerograms in


seismic analysis)
Sufficiently regular functions:
Sampling is correct (small time steps)
C2, or at least C1

Avoid some quasi-static artifices like excessively stiff materials


(often used as simplified representation of reinforcements)

22 - Code_Aster and Salome-Meca course material GNU FDL Licence


Some advice for a proper use of
DYNA_NON_LINE (cont.)

Initial conditions
Non-regular loadings: artificial oscillations
Computing from an initial static pre-stressed state (effect of gravity)
First step: quasi-static calculation with STAT_NON_LINE

Time step value: very strong physical meaning


Criterion based on desired cut-off frequency
Criterion based on prescribed conditions
Criterion based on wave propagation: Courant condition
Stability condition for explicit time schemes

23 - Code_Aster and Salome-Meca course material GNU FDL Licence

Some advice for a proper use of


DYNA_NON_LINE (cont.)

Damping/dissipation: in this order

Material (behavior) dissipation


Dissipation in links, joints and assemblies (friction)
Modal damping (values from experiments)
Rayleigh damping (often fitted on modal damping values)
HF damping from the time scheme (if needed)

24 - Code_Aster and Salome-Meca course material GNU FDL Licence


Some advice for a proper use of
DYNA_NON_LINE (cont.)
Time integration scheme choice (test-case sdld31a)

Implicit: often the best choice in Code_Aster


Low to medium frequency problems
Tight respect of equilibrium (can leads to convergence problems)
Almost all the quasistatic methods are available (except continuation…)
Different time schemes: Newmark family, Full HHT, Krenk, -method…

Explicit: fast dynamics (wave propagations): CPU time consuming


High frequency problems
Stability conditions (named Courant or CFL): T ~ l min EF / c
No convergence problem (if CFL is insured) but the numerical solution accuracy has to be checked

HF dissipation: HHT or Tchamwa-Wielgosz

25 - Code_Aster and Salome-Meca course material GNU FDL Licence

Some advice for a proper use of


DYNA_NON_LINE (cont.)

Specificities of explicit computations

Stability condition (CFL): T ~ l min EF / c

Solving using acceleration (unlike displacement used with implicit schemes)


Mass matrix inversion: lumping is recommended
Displacement boundary conditions (Dirichlet) are expressed in acceleration terms

Rayleigh damping
Only proportional to the mass matrix (stiffness proportional terms tends to lower the CFL condition value)

Computational efficiency is low with Code_Aster (compared to specialized codes like LS-
DYNA or EUROPLEXUS), except with modal reduction

No “exact” contact algorithm: only penalization method

26 - Code_Aster and Salome-Meca course material GNU FDL Licence


Some advice for a proper use of
DYNA_NON_LINE (cont.)
Chaining operators
STAT_NON_LINE  DYNA_NON_LINE: OK
DYNA_NON_LINE implicit  explicit: OK
DYNA_NON_LINE explicit  implicit: MACRO_BASCULE_SCHEMA (cf. sdnv100j) because
a specific balancing algorithm has to be used, in order to avoid artificial numerical
oscillations

Post-processing and results storage


Syntax: like STAT_NON_LINE (with the addition of velocity and acceleration fields)
Feature: large number of time steps: filtering of storage
Archiving: storage of full fields only for some time steps
Explicit time schemes: archiving time step / computation time step = 10 to 100
Implicit time scheme: archiving time step / computation time step = 1 to 10
Observation: storage of evolutions on some nodes of the model (at each time step)

27 - Code_Aster and Salome-Meca course material GNU FDL Licence

Some advice for a proper use of


DYNA_NON_LINE (cont.)

Optimizations for large problems


Parallel computing could be used
Scalability (speedup) is usually less efficient than in quasistatic problems because of the
very large number of time steps
Mumps solver is more robust (direct solver)
Petsc solver (iterative solver) offers a better speedup but it can fail
Speedups are correct with 4 to 8 CPU
Archiving of result (with OBSERVATION)
Splitting long transient simulations with POURSUITE
Storing base in /scratch
In some cases, maxbase value has to be increased
Memory allocation should be large in order to avoid out-of-core behavior (writing memory data
on filesystem)
Using .mess informations (after each operator):
Statistiques mémoire(Mo): 15521./8920./1603./248. (VmPeak/VmSize/Optimum/Minimum)
Documentation U1.03.03 for details about memory management

28 - Code_Aster and Salome-Meca course material GNU FDL Licence


Straight transient solution: Friction problem

Friction block with a spring (test-case SDND100)

Ft=10, Fn=1, U0=8.5E-4, Xloc=Z

Fn  M g  kn gap T  K X T   FN


MX
k
m U0
Ft   Fn

29 - Code_Aster and Salome-Meca course material GNU FDL Licence

Different NL transient studies (1)

Piping
Plasticity (on beam or pipe FE)

Raft uplift
Shocks with friction
Initial state computed with STAT_NON_LINE

Cables pinching
Shocks and large displacements
Initial state computed with STAT_NON_LINE

30 - Code_Aster and Salome-Meca course material GNU FDL Licence


Different NL transient studies (2)

Concrete buildings or dams

Damage in concrete (ENDO_ISOT_BETON) and plasticity in steel

Global laws like GLRC (shell elements)

Soil-structure interaction
Deconvolution
Raft uplift

Fluid-structure interaction
Added masses
Sloshing effects

Nonlinear soil behavior

31 - Code_Aster and Salome-Meca course material GNU FDL Licence

Different NL transient studies (3)

Large steel tanks: buckling with FSI using DYNA_NON_LINE


Steel plasticity + orthotropic carbon fiber reinforcements
Acoustic fluid (compressible and inviscid) with free surface
Large displacements and structural instability: buckling
Rnom.=5,7 m Conical top

Ring 2

1,985 m

2,005 m 8

7 Ring 1
2,005 m

2,005 m 6
16 m Max. water
5 level
2m
=15,7 m
10,12 m 2m 4 Buckling mode
2m 3
(push-over method)
2m
2

1
Bolts for
fastening
32 - Code_Aster and Salome-Meca course material GNU FDL Licence
Different NL transient studies (4)
Large steel tanks: buckling with FSI using DYNA_NON_LINE
FSI coupled model (u,p,f) formulation (Ohayon)
Pressure values on deformed shape (amplified) of fluid domain
T = 0,1 s T=1s T=3s

33 - Code_Aster and Salome-Meca course material GNU FDL Licence

Different NL transient studies (5)

Circular tunnel excavation


Soil constitutive relation: Drucker-Prager
Convergence is very difficult with an implicit method (STAT_NON_LINE or
DYNA_NON_LINE)
 Explicit pseudo-dynamic method (increased mass terms)
Loading: deconfinement
2D mesh: from 1,500 to 60,000 nodes

57 m

R = 3m 57 m
Keystone Y

Right foot
34 - Code_Aster and Salome-Meca course material X GNU FDL Licence
Different NL transient studies (6)

Circular tunnel excavation


Some results
Quasi-static method: convergence until 94-96 % of deconfinement
Explicit method: 99 % deconfinement reached

9,0E+05

M1 : HHT
8,0E+05
M4 : HHT
7,0E+05 M1 : STAT
temps CPU consommé (s)

M4 : STAT
6,0E+05

M4 : DIFF_CENT en poursuite de STAT


5,0E+05
M4 : DIFF_CENT en poursuite de HHT

4,0E+05
M2c : DIFF_CENT en poursuite de STAT
M2c : STAT : M2c
3,0E+05

2,0E+05

1,0E+05

Shear bands 0,0E+00


0,6 0,65 0,7 0,75 0,8 0,85 0,9 0,95 1
taux de déconfinement

CPU time comparison


35 - Code_Aster and Salome-Meca course material GNU FDL Licence

Different NL transient studies (8)


Structure in large rotations
Pivot link in the up and left corner
Subject to gravity
No damping
2 schemes are compared
Central difference (explicit)
Average acc. (implicit)

36 - Code_Aster and Salome-Meca course material GNU FDL Licence


Evolutions in Code_Aster
Computations control: energetic balance analysis

Damping enhancements
Rayleigh (other stiffness matrix choice: secant…)
Dissipation control during transient computation…

Time step adaptation


Event driven
CFL updating during computation

Distributed computations (at solver level with MUMPS, PETSc,


FETI, or with transient subdomain methods)

37 - Code_Aster and Salome-Meca course material GNU FDL Licence

For more information


http://www.code-aster.org

Other Code_Aster training


Nonlinear methods (focused on constitutive relations, XFEM, contact)
Dynamics

Documentations: R5.05.05, U4.53.01, U2.06.13, U2.06.10,


R7.01.32 (GLRC)

Test-cases (names beginning with sdn of fdn…)

Book
Non-Linear Finite Element Analysis - A Short course taught by T. J. R. HUGHES and T.
BELYTSCHKO - Zace Services Ltd - ICE Division

38 - Code_Aster and Salome-Meca course material GNU FDL Licence


End of presentation

Is something missing or unclear in this document?


Or feeling happy to have read such a clear tutorial?

Please, we welcome any feedbacks about Code_Aster training materials.


Do not hesitate to share with us your comments on the Code_Aster forum
dedicated thread.

39 - Code_Aster and Salome-Meca course material GNU FDL Licence

You might also like