Fibre Distribution Characterization and Properties Prediction of Discontinuous-Fibre Reinforced Thermoplastic Composites Based on a Data-Driven Approach

Mechanical properties of discontinuous-fibre reinforced polymer matrix composites (DFR PMCs) are highly dependent on their fibre distribution. The fibre distribution inside the DFR PMCs is usually characterized and analyzed separately by different features such as fibre orientations and fibre lengths. There is no established method to characterize fibre packing states within the DFR PMCs, and the links between different fibre distribution features are rarely addressed. In this study, the concept of fibre cell is proposed to provide an alternative to characterize the fibre distribution within DFR PMCs. The concept involves dividing a fibre domain into smaller cells based on the location of each fibre and characterizing the fibre distribution within the fibre domain based on the fibre cell properties. Compared with traditional characterization methods, the proposed method is able to include fibre packing information. Moreover, all fibre distribution features can be characterized in a joint manner instead of separately by different fibre distribution features. The proposed method is also merged into a machine learning (ML) framework to deduce the feature-property correlation via an artificial neural network. The ML framework is used as a surrogate model for fast homogenized elastic modulus predictions.