FEA Model Creation in LS-DYNA

The model of the coupon was drawn and meshed using Altair HyperMesh. The mesh size was 0.3 mm, or 200 elements along the 60 mm length of the samples, and 50 elements along the 15 mm width of the samples. This was set up using shell elements as they are compatible with all the material models and representative of the slender beam shape of the sample.

As the solver used in this simulation was LS-DYNA, the rollers of the three-point bend fixture were also modelled and meshed in HyperMesh. Figure 4-11 shows the initial model set up. The roller faces in contact with the sample were modelled.

Figure 4-11: HyperMesh model of the three-point bend test

To model a hybrid material in a quasi-static bending situation, several models, theories and assumptions are possible. These all have inherent advantages and disadvantages. Three different models were investigated in this specific context. The first model included the steel and the composite in the same part. The second model assumed the steel and the composite in separate parts. The final model assumed all individual layers, be it steel or composite, to be in separate parts. The model with the highest correlation of results with the experimental work was taken forward to the later stages of modelling. In all models, constraints, initial state and required outputs were defined; as well as the material characteristics.

Bottom rollers

Top roller

Assumption 1: Steel and Composite in the same part

Figure 4-12: LS-DYNA model of the three-point bend test, with a single layer of shell elements

The overruling assumption for this model is that the adhesive bond behaviour between the composite to the steel is perfect, i.e. displays no failure in the elastic region and therefore does not require separate modelling. The steel and the composite are modelled together in an overall PART_COMPOSITE, as in Figure 4-12. As explained in Chapter 3, the material cards used are MAT_02 and MAT_58 respectively. This material model is also used in the modelling of the benchmark materials, pure steel or composite alike. An underlying restrictive feature of PART_COMPOSITE is the assumption that the interface between its layers to be perfect, implying a perfect interface between the steel and the composite material. It does not recognise incompatibilities between material interfaces, and therefore won’t show true delamination failure modes.

Based on the vector directionality discussed previously, it is possible to define all six hybrid experimental cases separately and produce results. The layers are defined in the correct order and grown so that each sample can be represented.

Results for this type of model were within a 10% correlation to the experimental results, showing a high level of confidence in the model. Figure 4-13 shows the simulated and experimental load-extension graphs of the 0.8 mm hybrid samples. These are seen to present a high correlation, and provide confidence in the model

difference in the [45,-45]s hybrid samples is likely due to the slippage discussed

earlier.

Figure 4-13: Representation of correlation between the experimental (solid lines) and simulated (dotted line) for the hybrid 0.8 mm DP600 GFRP samples The two other assumptions were tested, however it was shown that Assumption 1 was the scenario that presented the highest level of correlation with the experimental model and the assumption taken forward through the simulations in LS-DYNA. The results are presented and discussed in detail in Chapter 4, Results. Once this assumption was selected it was then further verified through a second round of testing, where the results were predicted using the model and then checked experimentally. The correlation, as represented in Figure 4-13, was still found to be within 10 %, and the assumption carried forward to further simulations. The correlation is based on the values of stiffness gradient k calculated from the curves shown previously. The further two simulation set-ups tested are briefly presented and discussed as follows.

Assumption 2: Steel and Composite in separate parts

Figure 4-14: LS-DYNA model with two layers of shell elements: Steel and composite model

Two layers of shell elements were defined, one was given the properties of the steel and one was treated as a PART_COMPOSITE with four layers of composite. A contact was created between the two layers to simulate the presence of the adhesive and avoid layer penetrations. This contact was either simulated as a contact card or using a layer of solid elements with basic adhesive properties.

In both cases, the model results did not correlate closely enough (over 15 %) with the experimental results, the stiffness of the overall structure was overestimated, and computational time increased.

Assumption 3: Five different shell layers

Figure 4-15: Layers modelled individually

Five layers of shell elements were defined, one was given the property of the steel and four were given individual composite properties. The fibre orientation was

each fibre direction. Contacts were created between the layers simulating the presence of the adhesive and the nature of the matrix bond. This model had a high level of complication added to it, and inherent assumptions with every layer. This was also the most computationally expensive model. The results were the least convincing, the error was over 20 %. Additionally, this model did not profit from using MAT_58 and PART_COMPOSITE, as it did not use the in-built laminate function.