Multiscale FE Modelling and ANN to Predict Mechanical Properties of Non-Crimp Fabric Composites With Manufacturing Induced Defects

Abstract

Fiber-reinforced plastic (FRP) composites comprising unidirectional non-crimp fabrics (UD-NCFs) have recently gained considerable traction for fabrication of liquid composite molded parts. However, the variations in the microstructure of UD-NCF composites poses challenges in predicting and optimizing their properties. Additionally, their mechanical properties are influenced by process-induced defects, including variations in the relative position and shape of the tows, in-plane tow misalignment, out-of-plane tow crimping, and non-uniform fiber volume fraction. Thus, there is a need to develop robust tools to predict the mechanical properties of NCF composites that capture the critical manufacturing-induced defects. In this study, a multiscale finite element (FE) modelling approach is proposed to predict the in-tow (micro) and lamina-level (meso) effective properties of UD-NCF composites in order to support the use of artificial neural networks (ANNs) for the same purpose. First, a microscopic analysis was conducted to capture the material micro and mesoscale structure and process-induced defects. Second, microscale and mesoscale FE models were constructed based on the microscopic analysis, where process-induced defects were included within the associated representative volume elements (RVEs). The developed multiscale modeling approach was intended to generate an adequate amount of reliable training and testing data for ANN models. Finally, micro and mesoscale ANN models were developed, trained, and tested to predict the relations between inputs (i.e., constituent properties and material structure parameters) and outputs (i.e., in-tow and lamina-level effective properties). The trained ANN models reduced the calculation time from hours in multiscale FE modeling to seconds without sacrificing accuracy.