National Aeronautics and Space Jet Propulsion Laboratory

Administration California Institute of Technology

Validating Finite Element Models of

Assembled Shell Structures

Claus Hoff
Jet Propulsion Laboratory
California Institute of Technology

Presented at the WCCM 2006

Century City, CA
Thursday, July 20, 2006

1
National Aeronautics and Space
Outline Jet Propulsion Laboratory
Administration California Institute of Technology

• Problems with membrane rotations in assembled shell models

• Penalty stiffness for membrane rotations

• Physical stiffness for membrane rotations using shell elements

with 6 dof per node

• Connections avoiding rotations

• Conclusions

2
 Problems with Membrane Rotations in Shells
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

Ž Mindlin shells have no stiffness in the

membrane rotation θz normal to the
shell plane θz
4 3
Ž The rank deficiency can be z uy

automatically removed, for example y

using AUTOSPC in MSC.Nastran

1
Ž The membrane rotation may be loaded x ux
either by accident or intentionally, for 2

example, through stiffeners, spot

welds, rigid elements, concentrated
masses, etc.
Ž False load transfers or spurious modes
may occur
3
National Aeronautics and Space
Typical Shell Model in Aerospace
Jet Propulsion Laboratory
Administration California Institute of Technology

Terrestrial Planet Finder

Chronograph

• Elements
QUAD4, TRIA3,
few HEXAs and PENTAs

• Connectors
CELAS, CBUSH,
RBE2,
RBE3

4
National Aeronautics and Space
Typical Shell Model in Automotive Jet Propulsion Laboratory
Administration California Institute of Technology
BIW with acoustic cavity (courtesy of GM) 1.5 Mill. dof

312,142 GRID

260,348 CQUAD4

17,332 CTRIA3

10,812 CHEXA8

664 CPENTA6

1,880 CELAS2

397 CELAS1

390 CBEAM

14,336 RBE2

17 RBE3

2,310 MPC

5
 Penalty Stiffness for Membrane Rotations
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

A penalty term is added to the energy functional

∫ (Θ z − Ω z ) t dA
−6 1 2
Π p = 10 K6ROT G
2
A
with

1 ⎛ ∂u y ∂u x ⎞ θz
Ω z = ⎜⎜ − ⎟⎟ 4 3
2 ⎝ ∂x ∂y ⎠ z uy
y

1
K6ROT is a user parameter in MSC.Nastran, x ux
2
default value is K6ROT= 100.

6
 Good Value of the Penalty Stiffness
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology
Fine Meshes
1.0100 64x64
1.0000
A 32x32
0.9900
0.9800 16x16
SPC = 0.9700
123456
0.5 0.9600
0.9500
8x8
0.9400
0.9300
4x4
0.9200
0.5 0.9100

Normalized Tip Displacement

T= 1.e-4 0. 10. 100. 1000. 1e +4 1e +5 1e +6

Y
K6ROT
X 1.0 Coarse Meshes
1.0000
4x4
0.9000
0.8000
2x2
0.7000
0.6000
0.5000
1
0.4000
0.3000
0.2000
0.1000
0. 10. 100. 1000. 1.e +4 1.e +5 1.e +6
K6ROT 7
 Effect of Penalty Stiffness - Flagpole
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

Flagpole Example
K6ROT No. of First Comment
Zero Nonzero
Modes Mode

0. 7 2.4671e+6 spurious mode,

bad first

100. 6 1.2745e+6 no spurious,

good first

• Intentional loading of the drilling degree-of-freedom in shells may be

discredited as bad modeling practice.

8
 Effect of Penalty Stiffness – Ring with Stiffener
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

• Unintentional loading of the drilling degree-of-freedom occurs in some cases,

for example, in stiffened shells modeled with beams and offsets.

Cylindrical Ring with Stiffener Normal Modes

Offset
K6ROT Disturbance Comments

1 rigid body mode

Z R=10 246 No spurious modes
2
ε ≤1.8E−4
AUTOSPC catches
246 0.
Θ 2
singularity
246 n 2nd mode 9.667
246
246

246
0.
ε ≥1.9E−4 1 rigid body mode
5 spurious modes
7th mode 9.734

1 rigid body mode

Ring Modeled with CQUAD4 100. ε ≥1.9E−4 No spurious modes
Stiffener modeled with CBARS and offset 2nd mode 9.655

No I2 specified on PBAR
Offset direction and normal direction differ slightly

9
 Shells with 6 Degrees of Freedom per Node
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

The membrane displacements and the membrane rotations are coupled

with quadratic shape functions, for example, the displacement at the midpoint
of edge 1-2 is

um =
1
(u1 + u2 ) + 1 l1 (Θ z 2 − Θ z1 ) n1
2 8
θz
4 3
z uy
The following term is added to the energy functional y

Π Θ = G ∫ (Θ z − Ω z ) t dA
1 2
1
2 x ux
A 2

1 ⎛ ∂u y ∂u x ⎞
Ω z = ⎜⎜ − ⎟⎟
2 ⎝ ∂x ∂y ⎠
10
 Comparison of QUAD4 and QUADR
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

• QUADR is superior to QUAD4 for in plane shear and in plane bending

Straight Beam Shear load at tip

Tip deflection (normalized with theoretical value)

Rect
Version Rect Trap Skew
QUAD4 0.9929 0.0515 0.6323
QUADR 0.9926 0.9613 0.9491

Trap

Skew

11
 Comparison of QUAD4 and QUADR (cont.)
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

Normal Modes of a Straight Beam

Rect

Trap

Skew

In Plane Bending Frequency Element Rect Trap Skew

QUAD4 9.39 15.89 9.41

QUADR 9.39 9.42 9.39

12
National Aeronautics and Space
Modeling of Connectors Jet Propulsion Laboratory
Administration California Institute of Technology

Bolts or spot welds for nearly congruent meshes

2 x ( 3 x 9) Quad4s t = 1.0 mm E = 2.06 e+5 N/ mm2

Example
2 x 8 Spot Welds D = 2.0 mm ν = 0.3
Perturbation Δ y = 0.2 mm ρ = 0.785 e - 8 kg/mm2
demonstrates how
global results change
wave perturbation Δ y shift perturbation Δ y
when shell rotations
are coupled to
connectors

n
30 mm

n
y

x
t = 1.0 mm n
z

90 mm

Figure 6. Two Plates Connected with 16 Spot Welds

13
National Aeronautics and Space
Modeling of Connectors (cont.) Jet Propulsion Laboratory
Administration California Institute of Technology
n

• Comparing two modeling GA upper shell mid surface

options
n
– RBAR with no additional
constraints GB
lower shell mid surface

– Connector with constraints

• Membrane rotations of the
shell are not coupled to
the drilling rotations of the Error in first 6 Eigenvalues
bar
2
0
CWELD, shifted grids
-2 (not visible)
RBAR, shifted grids
-4
Error [%]

– The results from the perfectly CWELD, missaligned grids

aligned grids are the baseline -6 (not visible)
-8 RBAR, missaligned grids

– RBAR modeling causes errors -10

up to 12% in the first -12
eigenfrequencies -14
1 2 3 4 5 6
Eigenfrequency No.

Figure 7. Errors in the first 6 Eigenfrequencies

14
 Connections Avoiding Rotations
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

Interpolation constraints for translations z

⎧u ⎫ ⎧u ⎫
⎪ ⎪ ⎪ ⎪
⎨ v ⎬ = ∑ N I (ξ A , η A ) ⋅ ⎨ v ⎬ GA4 y
⎪w⎪ ⎪w⎪ GA3
⎩ ⎭A ⎩ ⎭I
GA
x
GA1 GA2
Kirchhoff constraints for rotations

∂w
θ xA = = ∑ N I , y ⋅ wI
∂y
Beam
∂w connected
θ yA = − = − ∑ N I , x ⋅ wI
∂x to shell

1 ⎛ ∂v ∂u ⎞ 1
θ zA = ⎜ − ⎟ = (∑ N I , x ⋅ v I − ∑ N I , y u I )
2 ⎝ ∂x ∂y ⎠ 2
15
 Connections Avoiding Rotations (cont.)
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

GBH4 GBi
GBH3 ¾ Spot weld is modeled with a
GB Shell B
GBH1
HEXA element
GBH2
¾ The HEXA is connected to
the shell elements through
GAH4
interpolation of translational
GAH1 GAH3
GAi dof only
GA

GAH2
Shell A
GAi GAi

GAH4 GAH3 GAH4 GAH3

GA GA

GAH1 GAH2 GAH1 GAH2

Coarse Mesh Fine Mesh

16
 Modeling Connectors (Results)
National Aeronautics and Space Jet Propulsion Laboratory
Administration California Institute of Technology

Body in White Model

• 3,712 CWELD Elements
• 3,562 connecting two parts
• 150 connecting three parts

Eigenfrequency Analysis:
Test Result
Old Spot Weld Modeling VW
New CWELD Element

Mode I Mode II Mode III

17
National Aeronautics and Space
Conclusions Jet Propulsion Laboratory
Administration California Institute of Technology

• Shell elements with membrane rotations (6 dof per node) are a

good choice in general
– Increase accuracy
– Provide realistic stiffness in membrane rotations

• For connectors (bolts, rivets, and spot welds), global results are
more accurate when coupling to shell rotations is avoided
– Generate constraints which involve only translational dof in shells

18