Wrinkling behaviour of biaxial non-crimp fabrics during preforming

The necessary lightweighting of the transport sector to meet emission reduction targets can be helped through the expanded use of composites. However, for the high volume production of composites to be cost-effective, it is needed that they can be manufactured through automated liquid composite moulding (LCM). Furthermore, the defects that occur during the initial preforming stage of LCM, notably wrinkles, are a key obstacle preventing automation and adoption of LCM, because wrinkles significantly compromise the component performance, and because there is currently no reliable method for mitigating them. To pave the way towards wrinkling mitigation during preforming, this thesis aims to characterise the wrinkling behaviour of non-crimp fabrics (NCFs) as well as to investigate how the wrinkling severity is affected by the tool geometry. These aims are achieved through both experimental and numerical approaches. Firstly, experimental forming tests are conducted to characterise the mechanisms, severity and variability of wrinkling for a ±45° biaxial NCF during preforming, considering four contrasting benchmark geometries. Secondly, a large dataset of forming simulations for various tool geometries is generated and used to investigate the effect of geometry on wrinkling severity, and to develop a deep learning based surrogate model for rapidly predicting the fabric wrinkling over a given tool geometry. The results demonstrate that two macroscale wrinkling mechanisms exist for this NCF and that the most severe wrinkles occur consistently via lateral fabric compression during material draw-in rather than tow compression at shear-lockup. Furthermore, they show that the wrinkling variability is significant and is especially apparent for multi-layer forming. Additionally, the tool geometry is shown to have a substantial effect on wrinkling with more tapered geometries leading to less severe wrinkling. Lastly, the surrogate model is demonstrated to achieve similar predictions to the finite element simulations but at a much lower computational cost, thus enabling the optimisation of component geometry for minimal wrinkling.