Virtual rapid prototyping of materials with deep learning: spatiotemporal stress fields prediction in ceramics employing convolutional neural networks and transfer learning

Abstract

ABSTRACT Additive manufacturing offers a solution for producing advanced ceramics with complex geometries by enabling precise control over geometry, microstructure, and composition. By leveraging deep learning, rapid prototyping and evaluation of printed ceramic parts become feasible. This study employs convolutional neural networks and transfer learning to predict spatiotemporal fields in ceramics, using synthetic datasets generated from X-ray computed tomography (micro-CT) and finite element analysis. The novel approach integrates spatiotemporal factors into deep learning models, enhancing the prediction of stress and damage evolution, and ultimately providing deeper insights into material behaviour and performance. Additionally, we introduce a target loss training strategy, which focuses on achieving a specific accuracy level, thus reducing training time while maintaining high precision. The proposed deep learning framework achieves accurate predictions of spatiotemporal stress fields, and captures fracture initiation and propagation behaviours. This method facilitates rapid prototyping and advances material design and evaluation processes while significantly reducing experimental efforts.