Co-authors​ :

 Fernández JOAQUÍN , Baumela LUIS , Carlos GONZÁLEZ (SPAIN) 

The authors of this paper provide an innovative digital twin (DT) to analyse the resin transfer moulding (RTM) production process of structural composites. A polymer resin is injected into a closed mould in RTM to saturate a dry textile preform. RTM is one of the most used techniques for manufacturing high-performance structural composites. However, RTM is susceptible to production disturbances, including dry regions and porosity that lead to defects, much like other liquid moulding processes. It is crucial to monitor the injection of resin and the evolution of resin flow to minimise faults resulting from processing disruptions and ensure constant quality and process dependability. In this case, the DT is concentrated on identifying non-homogeneous resin flow caused by race-tracking channels that re-route resin flow to the mould's exit gates, resulting in dry regions and non-impregnated areas. Two surrogate models built on encoder/decoder deep learning architectures make up the DT core, offering the quick and precise response required for interrogation during manufacturing. With the help of a series of five pressure sensors positioned throughout the mould surface, the first surrogate serves as the disturbance detector, delivering an instantaneous depiction of the fabric permeability. The second provides real-time visualisation of two quantities of interest (QoI): the pressure field inside the mould and the flow progress. The two surrogates were trained using synthetic data produced by high-fidelity multi-physics simulations of the flow progress in a porous preform using Darcy's law and OpenFoam. With a consultation time of less than 50 ms, errors in the surrogates' pressure field predictions are less than 1%, allowing for encapsulation in the digital twin. The study outlines two potential applications for the digital twin: a posteriori mode after the process is finished and an on-the-fly mode during manufacturing. The DT performance was assessed by comparing the answer to a series of RTM experiments for various race-tracking circumstances.