Predicting the local solidification time using spherical neural networks

Abstract Castings are predestined for the application of structural optimization, but to date, integration process simulation into optimization is limited due high computational cost and therefore often neglected at beginning design process. This leads need surrogate models, which allow a fast simplified evaluation proposals during in order improve integration. article introduces novel approach that estimates solidification time randomly created geometries solely based on casting geometry. The uses ray-tracing methods calculate distance function along preset directions. estimated calculated using Spherical Convolutional Neural Network (CNN). training data obtained by several thousand simulations toolkit commercial software combined with further augmentation. model experimentally validated five different sand