Draping Tool Generation for Image-Based Surrogate Draping Models using Free Form Deformation and Topography Modification

In the context of composite manufacturing process development, digital twins are often used to
conduct parametric studies in order to determine the optimum process parameters without the need of
extensive trial-and-error experiments. Process simulation using finite element (FE) method has become
a well-established tool for performing such tasks. However, the major drawback of this method is the
high computational cost (especially in the context of preforming when draping dry materials), which
restricts calculations to a limited set of configurations that must be carefully selected by experts.
Therefore, modern approaches increasingly focus on surrogate modeling using artificial intelligence.
One possible surrogate model architecture is the image-based U-Net model that maps input images to
resulting output images. When trained with sufficient data, this kind of model is able to predict results
for different draping parameters quite accurately for draping tool geometries included in the training
data (see [1]). When it comes to unknown tool geometries (that were not within the training data),
however, predictions become unreliable. This paper describes the generation of a parametric tool that
covers a range of different tool geometries in order to expand the prediction accuracy for unseen
geometries. By training the surrogate model on draping tools with different similarities to a reference, it
can interpolate within this similarity range, enabling accurate predictions for unknown draping tools that
lie between known configurations. The parametric tooling uses a similarity measure to evaluate the
degree of similarity between two tools of different similarities to a certain reference tool. The goal is to
use these generated tools within a routine to automatically create training data and then train the
surrogate using the simulation results for the varying tools. The developed parametric draping tool uses
Free Form Deformation and the Downhill Simplex method to achieve shapes of different similarities.