Operated by Los Alamos National Security, LLC for NNSA MultiMat 2011 - 1 A cell-centered Lagrangian hydrodynamics method for multi-dimensional unstructured grids in curvilinear coordinates with solid constitutive models D.E. Burton, T.C. Carney, N.R. Morgan, S.R. Runnels, S.K. Sambasivan*, M.J. Shashkov X-Computational Physics Division * T Division Los Alamos National Laboratory MultiMat 2011 International Conference on Numerical Methods for Multi-Material Fluid Flows Arcachon, France September 5-9, 2011 Acknowledgements: U.S. DOE LANL LDRD Program A. Barlow, B. Despres, M. Kenamond, P.H. Maire, P. Roe LA-UR-11-04995 August 27, 2011
33
Embed
Operated by Los Alamos National Security, LLC for NNSA MultiMat 2011 - 1 A cell-centered Lagrangian hydrodynamics method for multi-dimensional unstructured.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 1
A cell-centered Lagrangian hydrodynamics method for multi-dimensional unstructured grids in
curvilinear coordinates with solid constitutive models
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 8
because
Does the discrete operator for the curl of the momentum equation vanish? Yes, because of a corner stress tensor!
∇z ⋅σ p →1Vz
N j ⋅σ pj
j
z
∑
∇p× ∇z ⋅σ( ) →1
Vp
−Ni( )
i
p
∑ × ∇z i( )⋅σ⎡
⎣⎤⎦
= 1Vp
1V
z c( )
−Ni( )× N j ⋅σ p
j
j
z i( )
∑⎛
⎝
⎜⎜
⎞
⎠
⎟⎟
⎡
⎣
⎢⎢
⎤
⎦
⎥⎥
i
p
∑
= 1Vp
1V
z c( )
−Nc( )× Nc ⋅σ p
c( )⎡
⎣⎤⎦
c
p
∑
=0
Nc ×Nc =0
We need to show that the difference equations satisfy
The second-order operators are evaluated on a staggered grid
ρ&u=∇gσ∇× ρ&u( ) =∇×∇gσ =0
∇p×∇z ⋅σ p
∇z ⋅σ p
Nc
σ p
c
In other words, the internal contributions to the curl integral (dotted) vanish, so that there are no internal sources of circulation
Nj
Ni
The key is that both integrals must see the same stress tensor in the corner
This would not be true for surface stresses
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 9
The same methodology yields similar results for other relationships for both the cell and the nodal control volumes
∇z × ∇puo( ) =0
∇z × ∇p ⋅σo( ) =0
∇z ⋅∇p ×uo( ) =0
∇z ⋅∇p ×σo( ) =0
∇p× ∇zup( ) =0
∇p× ∇z ⋅σ p( ) =0
∇p ⋅∇z×up( ) =0
∇p ⋅∇z×σ p( ) =0
∇p e ∇z
∇z
u
pσ p
uoσ o
∇p
∇z e ∇p
Cell
Nodal CV
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 10
To incorporate the Second Law into the discretization, we must first decompose the energy equation
The Second Law
In a closed system ,
the kinetic energy must dissipate into the internal, suggesting
It is sufficient (but not necessary) that
which is the “entropy condition”
Dissipation models similar to
are invoked to satisfy the entropy condition
Alternative variable (not a linearization)σ p
i = σ z +δ pzi σ
u pi = uz +δ pz
i u
d i = ni ⋅δpzi σ( )⋅δpz
i u≥0
&kz =1
Mz
Ni ⋅σ pi
i
z
∑⎡
⎣⎢
⎤
⎦⎥⋅uz
=&uz ⋅uz
&wz =1
Mz
σ z : Ni upi
i
z
∑⎡
⎣⎢
⎤
⎦⎥
=σ z : &γ
&dz =1
Mz
Ni ⋅δpzi σ ⋅δpz
i u( )i
z
∑
“Work”
Internalenergy
Totalenergy
Kineticenergy
Momentum
equation ni ⋅δpz
i σ : μδpzi u
( &τ =0)
&dz ≥0
&ez =&τ z−&kz
&wz + &d
⎧⎨⎪
⎩⎪
“Dissipation”
Mz&τ z = Ni ⋅σ p
i ⋅upi
( )i
z
∑
= Ni ⋅σ pi
i
z
∑⎡
⎣⎢
⎤
⎦⎥⋅uz +σ z : Ni up
i
i
z
∑⎡
⎣⎢
⎤
⎦⎥+ Ni ⋅δpz
i σ ⋅δpzi u( )
i
z
∑⎡
⎣⎢
⎤
⎦⎥
=Mz&kz + &wz + &dz
⎡⎣ ⎤⎦
Strain equation
uz
δpzi u = u p
i − uz
Ni
u pi
i
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 11
Organization of the presentation
Mimetic approach• Why a mimetic approach• Corner vs. surface fluxes• Conservation & ancillary equations• Curl & divergence expressions• Entropy & the energy equation
Nodal solvers & the entropy condition• Conventional approach• A new tensor approach• Spurious vorticity
Mimetic approach for axisymmetric (rz) geometry• The notion of a centroidal control volume• Axisymmetric equations
Concluding remarks
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 12
The entropy condition
and the momentum conservation law
are solved on a nodal
control volume to yield surface stress and velocity
Algorithmic roadmap: Part 2, the “Riemann” solution–We will show results from two entropy relations
uo
uz
∇zu
Linear construction from cell center to cell surface
uo
u
p
Riemann-like solution at the node
&vz
u
p
Integration of fluxes
Cell CV Cell CV
Nodal CV =Dissipation
region Nodal CV
d i = ni ⋅δ iσ( )⋅δ iu≥0
N i ⋅σ p
i
i
p
∑ =0
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 13
Substitute the dissipation expression
into the flux conservation law
and solve the matrix equation for velocity
then go back to solve for force
We first tried a conventional dissipation expression* that is mathematically sufficient to satisfy the entropy condition
0= Nifpi
i
p
∑
=up ⋅ Niμ i nini( )
i
p
∑ + Ni ni ⋅σ oi −μ iiuo
i( )
i
p
∑
= A[ ] up⎡⎣ ⎤⎦− B[ ]
u p⎡⎣ ⎤⎦= A[ ]−1 B[ ]
f
pi =ni ⋅σ o
c +μ ii upi −uo
i( ) ⋅ nini
σ o uo
σ p
u
p
F
p1
F
p1
It is not necessary to explicitly evaluate the corner stress tensor since only the forces are actually used
We could solve for the stress tensor but it would be non-symmetric (4 unknowns) since there are now 4 equations
To satisfy rotational equilibrium, the stress tensor must be symmetric. Could this be related to observed chevron modes?
δf =μ δu⋅n( ) n
fp1 =n1 ⋅σ p
c
fp2 =n2 ⋅σ p
c
σ pc =
σ xx σ xy
σ yx σ yy
⎡
⎣
⎢⎢
⎤
⎦
⎥⎥
* Maire 2007Carre et al 2009
Operated by Los Alamos National Security, LLC for NNSA
The Noh* rz problem on a polar grid – relative to SGH, CCH shows reduced wall heating & reduced Gibbs oscillations
9x100L20sexperimental
9x100L20s acoustic
Substantial wall heating
Gibbs phenomena
Significantly reduced wall heating
Density should reach 64
Den
sity
Distance
SGH tensor viscosity
SGH standard
CCH
Dissipation pushes the shock
ahead
SGH tensor
viscosity
SGHstandard
CCH
We CCH compare with SGH using:• Standard settings – as normally used• Nonstandard options – tensor viscosity
* Noh 1987** Lipnikov
Campbell & Shashkov 2001
Operated by Los Alamos National Security, LLC for NNSA
Sedov* rz is a sensitive test of energy conservation and symmetry –CCH result is clearly superior to SGH
Colors correspond to pressure
CCH 1D sphericalSGH does not
preserve symmetry
Cavity volumes suggest dissipation in tensor viscosity
L20s rz quadratic
SGH tensor viscosity
SGH standard
CCH
Den
sity
Distance
Very noisy
SGH tensor
viscosity
SGHstandard
CCH
* Sedov 1959
Note smooth mesh
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 16
N12.0 L20e.06.808
The dissipation condition worked well for many problems - until we tried the Taylor Anvil* and Howell** problems (with strength)
ν = 0.35, and yield stress σY= 400 MPa. The material is assumed to harden linearly with a plastic modulus of 100 MPa. The calculations are carried out up to a time of 80μs (at which point nearly all the initial kinetic energy has been dissipated as plastic work).
Standard mesh size ∆x = 0.1296mm and ∆y = 0.064mm which results in 200 points along the axial direction and 100 points along the radial direction.
δf =μ δu⋅n( ) n
Coarse 25x50 mesh
Chevron instability arises here with fine
meshing
* G.I. Taylor 1948** Howell & Ball 2002
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 17
The conventional dissipation condition produces a force
that is in the direction .
We believe the system of equations should be closed with a physical model instead of simply a sufficient condition
δf =n⋅δσ =μn n⋅δu( )
n
L =aδt
δL=δuδt
a
The stress jump at the discontinuity should be proportional to the strain rate
This can be expressed in an impedance form
For a planar shock & in the principal frame of the strain rate tensor with a basis vector and signal velocity then
so the stress jump reduces to
Note that this is independent of the grid!
a
δσ =μ ⋅δw
a
n
Note that in the dissipation expression is a component of the strain rate tensor not a vector
A corresponding vector exists
and it is only when the force is calculated that the grid becomes involved
This force is in the direction not
δu=δua
δu
δf =n⋅δσ→ μδu n⋅a
δu n
δu
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 18
The stress field is discontinuous, so we must explicitly enforce conservation of momentum
Substitute the dissipation expression
into the momentum conservation law
and solve for velocity directly
then go back to solve for the forces
The tensor dissipation condition does not require a matrix inversion
0= Fpi
i
p
∑
=up Niμci ni ⋅ac
i
i
p
∑ + Ni
i
p
∑ ni ⋅σ oi −μc
iuoi ni ⋅ac
i⎡⎣
⎤⎦
F
pi =Ni ⋅σ p
i =Ni ⋅σ oi +μc
i up−uoi
( ) Ni ⋅aci
u p =− Ni ni ⋅σ o
i −μciuo
i ni ⋅aci⎡
⎣⎤⎦
i
p
∑
Niμci ni ⋅ac
i
i
p
∑
σ o uo
σ p
c
u
p
The 2 forces are constructed from a symmetric viscous stress tensor
so that rotational equilibrium is satisfied
δf =μδu n⋅a
F
p1
F
p2
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 19
N14.0 L20e.0@ t=150
With the tensor dissipation, the Taylor anvil* ran without chevrons -We also ran a coarsely zoned Howell** problem without chevrons
N12L20e.1 sym graducrash@ t=30
Coarse zoning with conventional dissipation
crashes due to chevron cells
t = 30
With tensor dissipation runs to completion
t = 150
δf =μ δu⋅n( ) n
0 μs
150 μs
Elastic-plastic shell coasts inward until it stops
• 4 cm cylindrical• 3 cm spherical
Initial velocity field is divergence-free
Howell – 2Dxyusual setup
* G.I. Taylor 1948** Howell & Ball 2002
δf =μδu n⋅a
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 20
N14.2L20e.0 symQ2=0
N14.2L20e.1 symQ2=1
Preliminary testing suggests tensor dissipation does not seem to exhibit hourglass or chevron modes
Noh xybox grid
Noh xypolar grid
Sedov xybox grid
Saltzman xyt=0.75
t=0.80
t=0.85
t=0.90
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 21
Algorithmic roadmap: Part 1, Reconstruction – We looked at several schemes – this is based on Maire’s work
uo
uz
∇zu
Linear reconstruction from cell center to cell surface
uo
u
p
Riemann-like solution at the node
&vz
u
p
Integration of fluxes
Cell CV Cell CV
Nodal CV
Nodal CV =Dissipation
region
* Maire 2007
The finite volume integrals are conservative, but do not provide a distribution within the cell
Conserved quantities can be redistributed linearly through the centroid, without altering the total
The gradient is used to extrapolate from the cell center to the surface
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 22
N14.0 nc ncL20e.0 full gradu
Note: chevron at piston face fixed
With SGH & CCH, large scale spurious vorticity seems to be an issue - If we know the answer before hand, we can fix it. Should we?
We can obtain excellent answers in SGH & CCH if we know there should be no vorticity
The numerical scheme can damp vorticity in various ways:
• corner pressures1
• tensor viscosity2
• curl-Q3 • Dukowicz & Meltz4 • etc
Here, the limited velocity gradient used in the second order extrapolation is simply symmetrized
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 23
Unfortunately, practical problems do have vorticity – a stiffened algorithm can produce a bad answer!
THE ULTIMATE GOAL is a single method that produces satisfactory answers to all problems with the same “knob” settings, or better yet, no knobs
Basic methodFoot @ 6.66
DampedFoot @ 5.65
N14.0 L20e.15.649 not acceptable
Taylor anvil @ 80Coarsely zoned 25x50
Contours are effective stress
SaltzmanTaylor anvil
Both appear to have legitimate bending modes
Why & how should we inhibit one and not the
other?
Because the Saltzman result was produced by a planar shock and should be irrotational!
The numerical algorithm does not explicitly guarantee that planar shocks are irrotational,
so we need a vorticity limiter to enforce the condition
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 24
A dynamic vorticity limiter performs nearly as well in both cases as a knobbed switch that is based upon prior knowledge
N12.0 L20e.0
N14.2 L20e.3.1 N114.2 L20e.1sym
N14.2 L20e.3.16.663
N14.2 L20e.1 sym5.681
N14.2 L20e.0 full6.843
6.843 6.663 5.681
ψ =1 ψ =0
Basic method
Damped Dynamic
limiter
ψ =dynamic
Since shocks tend to be 2-3 cells wide, the
dynamic limiter is determined by the
minimum of the shock limiter
including adjacent cells
t=0.75
t=0.90
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 25
Organization of the presentation
Mimetic approach• Why a mimetic approach• Corner vs. surface fluxes• Conservation & ancillary equations• Curl & divergence expressions• Entropy & the energy equation
Nodal solvers & the entropy condition• Conventional approach• A new tensor approach• Spurious vorticity
Mimetic approach for axisymmetric (rz) geometry• The notion of a centroidal control volume• Axisymmetric equations
Concluding remarks
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 26
It is widely believed the mimetic equations in rz are conservative, but do not preserve 1D symmetry*
Conversely, it is also widely believed that the area weighted equations are flawed because they preserve symmetry at the cost of rigorous conservation
By considering a second order extension of the mimetic notion,
we show that the axisymmetric (rz) equations:• are mimetic• are conservative• do preserve symmetry• are appropriate for an infinitesimal region
about the centroid• are the canonical area weighted ones
That is, the canonical area weighted momentum & strain equations ARE the correct ones • Symmetry preservation:
on an equiangular polar grid, spherical loading should produce spherical results
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 27
Second order requires a shift in our thinking from “cell averages” to solutions at an integration point, the centroid
In first order schemes, the solution consists of cell averages of conserved quantities:
volume/strainmomentumtotal energy
and derived quantities:kineticinternal energystress
In second order, conserved quantities are redistributed linearly through the centroid without altering the totals
However, the kinetic & internal energies become nonlinear functions and cannot be cell averages
So the solution consists of values at the centroid, NOT cell averages
vz,γz
uz
τ z
kz=
uz2
2ez =τ z−kz
σ z =σ z γz,ez( )
Can there be a fundamental principal we are missing?
Yes, a “centroidal” control volume
uz
u
kkz
uz
u
kkz
First orderEnergy is flat within the cell
Second orderWe only know the kinetic & internal energy here
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 28
In planar geometry, construct a geometrically similar subcell centered at the centroid and scaled by a factor α
u p
%u p
N %N
uz
Geometrically, we have
We linearly interpolate coordinates & all fluxes between the surface and the centroid by the same factor, e.g.,
Flux is conserved by detailed balance between the “doughnut” and the “hole” for all values of α
In particular, the resulting equations for quantities at the centroid are identical to those for cell averages, e.g.,
%ρz = ρ z
%Vz = α 3Vz
%M z = α 3M z
%Ni = α 2Ni
%u p
i =uz +α upi −uz( )
%&vz=1%Mz
%Ni ⋅%upi
i
z
∑ →1
α3Mz
α3Ni ⋅upi
i
z
∑=&vz
The point:
The extended mimetic finite volume equations at the centroid are the same as those for the cell as a whole (at least in xy geometry)
This is important for understanding rz geometry
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 29ref: TN38z temp, TN 38(19)
The usual mimetic methodology is conservative in rz, but does not preserve symmetry
Consider a section of a thin “orange slice”
The conservation relations must now include contributions from the lateral surfaces
so that the mimetic evolution equations reduce to
N+ +N−=− Nc
c
z
∑ =−Azr
N+ ⋅uz+ +N−⋅uz
−=0
N+uz+ +N−uz
−=Az −ruz +ϕϕ uz ⋅r( )⎡⎣ ⎤⎦
N+ ⋅σ z+ +N−⋅σ z
−=−σ zϕϕAzr
N+ ⋅σ z+ ⋅uz
+( ) +N−⋅σ z
−⋅uz−
( ) =0
M z &vz = Ni ⋅upi
i
z
∑
Mz&γz = Niupi
i
z
∑ +Az −ruzc +ϕϕ uz ⋅r( )⎡⎣ ⎤⎦
Mz&uz = Ni ⋅σ pi
i
z
∑ −Azσ zϕϕ r
Mz&τ z = Ni ⋅σ pc ⋅up
c( )
i
z
∑
The usual mimetic momentum & strain equations do NOT preserve symmetry
but the volume & total energy do
Δϕ → 1
Vz = Rz Az
M z = constant
ρ z ≡M z
Vz
A
z
RΔϕ
x
y
r
N
N−
N+
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 30
However, we can write the mimetic rz equations for the centroid subcell - They do preserve symmetry and are conservative
The volume & total energy are the same as those for the entire cell and also preserve symmetry
M z %&vz = Nc ⋅upc
c
z
∑
Mz%&τ z = Nc ⋅σ pc ⋅up
c( )
c
z
∑
Then the subcell momentum & strain equations are conservative by detailed balance for all α, including the limit
After taking the limit α 0 and substituting for mass, the mimetic method yields symmetry preserving, area weighted expressions
ρzAz %&uz = Ai ⋅σ pi
i
z
∑ +1
Rz
r ⋅σ z −σ zϕϕ r( )
ρ zAz %&γ z = Aiu pi
i
z
∑ +1
Rz
ϕϕ uz ⋅ r( )
The point: The canonical area weighted equations* are in fact
• mimetic• conservative• preserve symmetry• appropriate for an infinitesimal
region about the centroid
M z %&uz = Ni ⋅σ pi
i
z
∑ −Azσ zϕϕ r
=Rz A i ⋅σ pi
i
z
∑ +Az r ⋅σ z −σ zϕϕ r( ) +O α( )
Mz%&γz = Niupi
i
z
∑ +Az −ruz +ϕϕ uz ⋅r( )⎡⎣ ⎤⎦
=Rz A iupi
i
z
∑ +Azϕϕ uz ⋅r( ) +O α( )
%Ni ⋅%ψpi
i
z
∑ → α 3 Rz A i ⋅ψpi
i
z
∑ +Azr ⋅ψz +O α( )⎧⎨⎩
⎫⎬⎭
One can show
* Wilkins, 1964
See Burton 1994 for a proof of symmetry preservation
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 31
Organization of the presentation
Mimetic approach• Why a mimetic approach• Corner vs. surface fluxes• Conservation & ancillary equations• Curl & divergence expressions• Entropy & the energy equation
Nodal solvers & the entropy condition• Conventional approach• A new tensor approach• Spurious vorticity
Mimetic approach for axisymmetric (rz) geometry• The notion of a centroidal control volume• Axisymmetric equations
Concluding remarks
Operated by Los Alamos National Security, LLC for NNSAMultiMat 2011 - 32
Summary: Because we are in relatively unexplored territory, we used a
mimetic approach to guide the formulation of the difference scheme
We reached many of the same conclusions as previous investigators (Despres, Maire, …), but differ on others.
All conservation and ancillary relations are satisfied• The GCL is satisfied by corner, not surface, fluxes• Curl-divergence relations are satisfied• Both the mechanical & viscous stress tensors are
symmetric & preserve rotational equilibrium• The area weighted equations in rz are conservative and
preserve symmetry
The conventional dissipation expression gave rise to chevron modes
• A new tensor dissipation model leads to a simpler nodal solution and does not appear to introduce chevron or hourglass modes
• A physics-based vorticity limiter seems promising for addressing large scale spurious vorticity
Historically, the weakest link in CCH has been the nodal motion
• The community seems to be converging on a solution
The mimetic approach provided firm guidelines in formulating CCH for:
• Material strength• Unstructured polytopal grids• Multi-dimensional formulation with