Zimmerling, ClemensClemensZimmerlingDörr, DominikDominikDörrHenning, FrankFrankHenningKärger, LuiseLuiseKärger2022-03-052022-03-052019https://publica.fraunhofer.de/handle/publica/25854010.1016/j.compositesa.2019.05.0272-s2.0-85066461437Manufacturing continuous fibre reinforced components often involves a forming process of textiles. Process simulations using Finite Element (FE) techniques allow for an accurate virtual formability assessment, but are typically time-consuming, especially for iterative design optimisations. To provide remedy, this work proposes machine-learning (ML) techniques as easy-to-evaluate approximations of FE-forming results. While previous studies focus on adjusting process parameters to achieve manufacturability, this work investigates local geometry variations. Initially, an ML-model is trained on FE-based forming examples in order to relate geometric features to forming results. During component formability assessment, an image-based recognition approach identifies manufacturing-critical regions. Then, the ML-model estimates forming results for each region individually. The validity of local formability assessment for a minimum mutual distance is based on Saint-Venant's Principle and is supported by FE-based verification. The overall approach is validated on a complex shaped box-geometry. Moreover, time-efficient exploration of local design alternatives to improve manufacturability is demonstrated.enfabrics / textilescomputational modellingstatistical properties / methodsformingmachine learning660620journal article