Determination of needling process satisfying stiffness requirements of 3D needled composites

The traditional trial-and-error method is expensive and time-consuming to determine the needling process parameters satisfying the stiffness requirements of the 3D needled composites. A new design method for these unknown needling process parameters is proposed using a convolutional neural network (CNN) surrogate model, in which a series of stress distribution images of representative volume cell (RVC) for 3D needled composites under different loads are implemented. These stress distribution images reserve both material stiffness and needling process information. The CNN surrogate model can be efficiently established to obtain the relationship between the stress distribution images and the needling process parameters. The accuracy of the training samples and the test samples in the CNN surrogate model reaches 93.85% and 87.50% respectively. An artificial RVC can be directly constructed corresponding to the stiffness properties. The unknown needling process parameters can be determined using the CNN surrogate model and the stress distribution images of the artificial RVC. The result shows the stiffness properties of 3D needled composites with the needling process parameters obtained by the CNN surrogate model are in good agreement with the experimental results, where the maximum relative error is 7.66%. This method provides a new way to design the needling process parameters satisfying the specific stiffness requirements of the 3D needled composites.