Predicting the fiber orientation of injection molded components and the geometry influence with neural networks

Abstract

The injection molding simulation of short fiber reinforced plastics (SFRP) is time consuming. However, until now it is necessary for predicting the local fiber orientation, to optimize the molding process and to predict the mechanical behavior of the material. This research presents the capabilities of artificial neural networks (NN) in predicting fiber orientation tensor (FOT) during injection molding processes, with a focus on enhancing computational efficiency compared to traditional simulation methods. Three NN architectures are compared based on simulated injection molded plates, with the goal of predicting the effect of the plate geometry on the local fiber orientation. Results indicate that NN outperform the baseline assumption of aligned fibers and demonstrate significant potential for accurate FOT prediction. The computational efficiency of NN, especially during the prediction phase, showcases a reduction in processing time by a factor of 104 compared to traditional simulation methods. This research lays a foundation for further exploration into the feasibility of NN in partly replacing time-consuming simulations for practical applications in injection molding processes.