Abstract:

Resin Transfer Moulding (RTM) is a cost-competitive manufacturing process for carbon fibre-reinforced polymer (CFRP) composites, used for structural applications that require lightweight design because of their enhanced properties. However, internal cavities and porosities appear at the injection stage due to poor fibre impregnation, leading to part rejection and unit cost increase. In order to make the manufacturing process insensitive to processing distortions, Digital Twins are employed towards Smart Manufacturing, being surrogate modelling one of the key enabling technologies, as it accelerates the RTM virtual simulation and reduces its computational costs.

In this work, a surrogate model based on a densely connected deep encoder-decoder is employed to overcome current surrogate models limitations and address a more realistic variety of cases encountered in the industry, such as different mould geometries (flat and three-dimensional), meshes (regular and irregular), boundary conditions (injection and venting port location) and permeability dissimilarities on the fibre preforms (multiple localized rectangular permeability dissimilarities and gaussian random permeability fields). Trained on a generated synthetic dataset, it predicts resin flow front and pressure scalar field for several instants of time during injection, given as input information a permeability map, a pressure build-up factor and simulation time, accelerating simulations on a 96% – 99%, from minutes to seconds.