More than half a century after the first application of composite materials in aircraft, the accurate prediction of their failure remains a pressing and unresolved issue. Another important limitation continues to be the low transverse strength of unidirectional composite plies, which can lead to premature failure in common laminates such as the cross-ply. A tool that promises improvements on both of these fronts is computational modeling. With ever increasing computing power, today’s multi-scale-models are able to simulate the various failure modes across the relevant length scales, and allow a new level of optimization by accurately determining the contributing material parameters.

To fully exploit this capability, we propose a ‘smart’ framework, combining Computational Modeling, Design of Experiments, and Neural Networks. It is developed with the aim to create analytical surrogate models of complex material properties, in a fully automated way, and based on a minimum amount of computer simulations. While such a framework can be universally applied, we will showcase our results for one possible application: the prediction and optimization of the transverse strength of a unidirectional composite ply.

Generating Statistical Volume Elements of the material at microscale, we use a Computational Micromechanics model to compute the ply’s strength under transverse loading. By Design of Experiment principles, we then explore the interdependent influences of the main geometrical and material parameters. Finally, we obtain an analytical surrogate model of the transverse strength by employing a Neural Network.

Putting to use the developed ‘smart’ framework, the effects of the following parameters were studied: constituent strengths (matrix, fiber-matrix-interface), fiber volume fraction, shape of fiber cross-section (circular or non-circular), and fiber diameter. The results confirm and quantify the primary influence of the constituent strengths, followed by the secondary effect of the fiber volume fraction. Non-circular fiber cross-sections were found to increase transverse compressive strength, and mixing of different fiber diameters may increase slightly both tensile and compressive strength.

In addition to presenting the resulting surrogate models, challenges in the development and automation of the framework will be discussed, and the case will be made for wider application of such a framework.

A novel continuum damage mechanics model for 2D woven fabrics has been developed and implemented in a VUMAT subroutine for Abaqus/Explicit. The model takes into account the shear non-linearity, the toughening mechanisms associated to tensile failure, and the influence of shear stresses in the initiation and propagation of tensile damage.

The non-linear shear behavior is reproduced by means of a Ramberg-Osgood equation. Permanent deformation, and the degradation of the secant shear modulus associated to the accumulation of matrix damage have been coupled to the formulation of this unidimensional plasticity law. The physical reference required to define completely the shear constitutive response proposed can be obtained from the experimental tensile test of ±45 off-axis coupons. Failure initiation in tension is identified using Hashin's quadratic failure criterion, which accounts for the interaction of shear stresses in the promotion of tensile failure. A bilinear softening relation was selected to represent the combination of fibre breakage and toughening mechanisms characteristic of the intralaminar tensile fracture of these materials. Effort was placed on the development of a procedure for calibrating the softening relation associated to this failure mode. Specifically, this work addresses the applicability of linking the definition of a bilinear tensile damage law in homogenized continuum damage mechanics models for woven composites to the shape of a crack growth resistance curve measured with compact tension tests.

The constitutive model was validated including it in a set of finite element models of unnotched and open hole coupons with different multistacking sequences, under quasi-static tensile loading conditions, and comparing the results obtained by the simulation of the coupons against an experimental benchmark. A good correlation was achieved between the ultimate strength predicted with finite element analysis and the experimental data.