Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Composite Structures
journal homepage: www.elsevier.com/locate/compstruct
a r t i c l e i n f o a b s t r a c t
Available online 29 March 2008 This paper considers the capability of finite element (FE) modelling to accurately predict fastener
pull-through failure of composite laminates. Such failures are dominated by inter-ply delamination
Keywords: and through-thickness shear failure of the laminate and the common modelling approach is to use com-
Fastened joints putationally expensive, detailed three-dimensional models that include delamination for every ply inter-
Cohesive fracture face, fastener contact and prestress. This paper considers a simplified FE modelling strategy achieved
Composite failure through judicious use of symmetry boundary conditions, hybrid shell/solid modelling and reduced num-
bers of interfaces for delamination. The LS-DYNA FE software was used for this study using the available
composite material and cohesive failure models. The conclusion drawn from this work is that the use of
simplified FE models does have merit in modelling fastener pull-through provided the material is quasi-
isotropic and the boundary conditions are uniform around a circular perimeter. Additional work is how-
ever required to determine suitable cohesive properties and progressive shear failure parameters.
Ó 2008 Published by Elsevier Ltd.
Fig. 3. Failure specimen viewed from above and below after failure, from Gunnion
[2].
Table 1
Cohesive fracture implemented in LS-DYNA 971 [6]
Fracture model LS-DYNA definition Stacked shell with offset Stacked solid Zero thickness capability
*
1 CONTACT_AUTOMATIC_ONE_WAY_SURFACE_TO_SURFACE Type 6 No Yes Yes
*
2 CONTACT_AUTOMATIC_SURFACE_TO_SURFACE Type 8 Yes N/A N/A
*
3 MAT_184 Yes Yes Yes
*
4 MAT_185 Yes Yes Yes
*
5 MAT_186 Yes Yes Yes
294 D.J. Elder et al. / Composite Structures 86 (2008) 291–298
Table 2
LS-DYNA 3-D composite material models with failure
maximum stress failure criterion, allowing a progressive damage which allowed the number of solid elements through the thickness
with shear softening to occur in the vicinity of the fastener head. to be reduced from 12 to four.
A deficiency of this material is that it does not comply with the The resulting model for this study was a 5° 3-D axisymmetric
plastic flow law. However, as only a limited area of the model in hybrid solid/shell model with three cohesive fracture planes [10]
the vicinity of the fastener head was equipped with this material, as shown in Fig. 7. The adopted materials included a triangular
the error introduced was negligible. cohesive fracture model (*MAT_186) and a honeycomb material
implemented with user defined orthotropy and a maximum stress
3.3. Element formulations failure criterion (*MAT_26). This model was approximated from the
actual test configuration as per Fig. 6. The bolt preload was initi-
The appropriate element choice is essential for efficient and ated by nodal temperatures applied to a temperature dependent
accurate explicit FE simulations. The adoption of the honeycomb orthotropic material such that the shank of the fastener contracted
material resulted in two possible eight node, constant strain ele- in the Z-direction only inducing the preload into the fastener [11].
ment formulations being considered: the under integrated (UI) sin- A sensitivity analyses of the model proved that at the point of
gle gauss point element; and the selectively reduced (S/R) delamination, the non-linear membrane stress governed and so
integrated element. Although both elements are constant strain the adoption of a constant S/R element was considered suitable
formulations and do not accommodate accurate bending strains in terms of accuracy and computational efficiency. The pull-
in single through-thickness element configurations, the S/R ele- through load was applied to the fastener centreline as a constant
ments performed well in representing bending stiffness with this velocity (with initial ramp-up), with constraints as indicated in
configuration. Fig. 7 applied. To allow accelerated loading of the structure to fur-
ther shorten the analysis time, various linear damping elements
4. Simplified fastener pull-through model development were applied to minimise unwanted vibration. The final model
was analysed successfully with a run time of 20 min on a single
4.1. Model description CPU computer.
The aim of the modelling strategy was to produce a model con- 4.2. Error assessment
taining a realistic prediction of fastener pull-through failure with
an analysis time under 30 min on a single CPU computer. This Pivotal in producing a workable model with a minimal compu-
would allow the fast turnaround of analyses and the future use tation time was the use of symmetry. Gaining model size reduction
of the optimisation tool LS-OPT for system identification of the by the use of geometric symmetry generally produces errors for
cohesive fracture and material parameters. Although LS-DYNA is composite materials owing to the through-thickness non-homoge-
a general finite element code with both implicit and explicit capa- neity of composite lay-ups producing non-zero rotational deforma-
bilities, it is strongly orientated toward the simulation of large ex- tions at the lines of geometric symmetry [12]. The error estimation
plicit 3-D crash scenarios, occupant safety and metal forming. This associated with the adopted model configuration includes mate-
generally results in advancements in LS-DYNA being first imple- rial, boundary fixity and the use of a dynamic explicit computa-
mented for 3-D explicit analysis, which is the case for the cohesive tional process to model a quasi-static loading regime. The error
fracture models implemented in the current version of LS_DYNA. magnitudes were determined by comparing the results of two dif-
However, a full 3-D explicit analysis was considered infeasible ferent implicit analyses that considered geometric nonlinearity
due to the extremely long computation times required. It was esti- only. A full 3-D model was developed where each ply was mod-
mated that a full 3-D analysis of fastener pull-through would re- elled with eight node continuum shell elements. This element for-
quire run times in excess of 24 h on a single CPU computer. mulation allowed high aspect ratios to be used and accordingly
This necessitated the consideration of various means to simplify reduced the model size considerably when compared to the tradi-
the model to reduce computation time. Firstly, the approximate tional solid formulation that is restricted to aspect ratios of approx-
axisymmetry of the test specimen was utilised which allowed only imately 10. The model assumed that the fastener provided full
a 5° section to be modelled with axisymmetric boundary condi- rotational restraint to the area under the top and bottom fastener
tions. To avoid computational problems toward the centre of the flanges as per Fig. 8. This model allowed the magnitude of errors to
fastener, a 15° slice of the fastener shank was modelled. Secondly, be determined for stiffness, stress and boundary conditions. In
as material through-thickness shear failure and inter-ply delami- addition, a 2-D axisymmetric model with contact and fastener pre-
nation is dominant in the region around the fastener, only the cen- stress was considered as per Fig. 9 and allowed the sensitivities of
tral area of the laminate and fastener was modelled using solid bolt pretension to be explored.
elements, while the outer region was modelled using more compu- When considering unified perimeter boundary conditions, all
tationally efficient shell elements. Lastly, the number of interfaces models produced peak stiffness within 15%, while bending and
at which delamination was allowed was reduced to only three membrane stresses showed similar variations. The largest and
D.J. Elder et al. / Composite Structures 86 (2008) 291–298 295
Fig. 6. (a) Plan view of actual test configuration and (b) adopted 5° 3-D axisymmetric model.
perhaps most critical stress variation was associated with the shear combination of these attributes resulted in insignificant errors
stress with approximately 30% variation. Of particular interest was associated with dynamic considerations.
the error in material definition associated with the application of The dominant uncertainty in the experiment was associated
geometric symmetry. When comparing the true orthogonal lami- with the perimeter boundary conditions. It was intended in the
nate definition with the radial (axisymmetric) definition used in experiment design that the perimeter boundaries were to be firmly
the simplified model as shown in Fig. 10, the difference between clamped such that vertical and rotational constraint was estab-
the two was 5% for membrane and bending and 10% for shear. lished. This was achieved by an annulus shim plate and a screw
These variations between models tended to decrease as the non- clamping collar (refer Figs. 6 and 7). However, the conclusion
linear membrane load path became more dominant with increas- drawn from all analytical methods, including classical and numer-
ing deformation. ical, was that the perimeter boundary was closer to simply sup-
The error associated with the accelerated load was minimised ported than clamped. The effect of the extremes of boundary
using a ramped velocity loading at the fastener centreline with conditions from simply supported to fully restrained is shown in
an associated variable acceleration applied to the composite sec- Fig. 11. To normalize the model boundary conditions to the exper-
tion and damping elements to reduce spurious noise. The effective imental results, a partial restraint system was added to the simply
296 D.J. Elder et al. / Composite Structures 86 (2008) 291–298
Fig. 8. Implicit 3-D LS-DYNA model with continuum shell elements used for error determination.
Fig. 9. Implicit 2-D axisymmetric LS-DYNA model with fastener prestress included.
Fig. 10. Material direction: (a) as defined in the axisymmetric model compared with (b) the actual definition.
supported model, increasing its initial stiffness. This was intended to a model with no delamination (ND) as per Fig. 12. The compar-
to allow point A in Fig. 11 to be relocated to point A0 such that the ison between these two extremes indicated that at low deforma-
experimental and analysis results had common stiffness at the tions, the tangent stiffness of the WD model was approximately
elastic limit. 60% that of the ND model, which was consistent with 2-D axisym-
metric implicit solutions as per Fig. 13. In this area of deflection,
4.3. Effect of delaminations the structure is governed by bending stiffness. As the deflection in-
creases the membrane stiffness becomes the predominate compo-
To explore the sensitivity of the model stiffness to the cohesive nent and at the failure limit of 3.9 mm the tangent stiffness of both
shear parameter, a model with delamination (WD) was compared models are approximately equal. The ND model is approximately
D.J. Elder et al. / Composite Structures 86 (2008) 291–298 297
Fig. 13. Force versus stiffness for the explicit and implicit ND and WD models and
Fig. 11. Force versus deformation behaviour from experiment and analysis showing experiment.
the effect of boundary conditions.
20% stiffer and has a 30% greater stress when compared to the WD
model. From the comparison of the WD and ND models it can be
clearly seen that for the experimental curve to evolute from point
A0 to point B0 , material damage must be occurring in addition to the
delamination.
The deflected shape of the ND and WD models shows a signifi-
cant difference, with the WD model having a longer cord length be-
tween points 1 and 2 with greater curvature at point 1 as shown in
Fig. 14.