The materials manufacturing industry has evolved rapidly with technological advances and digitization, bringing a new era in composite materials manufacturing. In recent years, resin transfer moulding (RTM) has gained much attention due to its ability to produce high-quality composite parts. However, RTM still faces some challenges that can affect the final product’s quality. This thesis explores data-driven algorithms to create a digital twin (DT) model to analyse the manufacturing process of structural composites by RTM. The data-driven models are designed to enhance the RTM process and predict the final product’s quality in real-time.

A dry textile preform is impregnated during RTM with a polymer resin injected in a closed mould. The DT looks out for inhomogeneous resin flow caused by zones of varying permeability in the fibres. Dry spots and a lack of impregnation result when resin flow is misdirected to the outlet gates of the mould, as can happen in areas of high permeability. The DT core contains two models based on encoder/decoder deep learning architectures, providing the fast/accurate response necessary for interrogation during manufacturing. The first model acts as the disturbance detector, providing the on-the-fly representation of the fabric permeability with only information gathered by five pressure sensors distributed over the mould surface. The second offers a real-time picture of a set of quantities of interests (QoI): the flow progress and the pressure field inside the mould. Both surrogates were trained with synthetic data generated by highfidelity multi-physics simulations of the flow progress in a porous preform by following Darcy’s law. Errors in the pressure field predictions of the surrogates are lower than 1% with consultation time <50 ms enabling encapsulation in the DT. Its performance was evaluated by comparing response against RTM experiments for different race-tracking scenarios.

The thesis findings will significantly contribute to the development of next-generation manufacturing technologies, demonstrating the effectiveness of DT and data-driven algorithms in improving the quality and efficiency of the RTM process. In particular, the study has contributed significantly to the field by designing a specific loss function for the problem and training models with synthetic data, which was validated using experimental data. Additionally, the study has considered the minimum number of synthetic data required to achieve reliable results and developed a methodology to capture data quadratically, considering the variation in velocity at the beginning and end of the process. Furthermore, the study has addressed the multitasking problem of predicting various quantities of interest, providing an efficient and effective solution to this problem.

The insights gained from this research will greatly benefit the manufacturing industry. They can enhance product quality and reduce manufacturing costs in other manufacturing processes. By leveraging the potential of digital twins and data-driven algorithms, manufacturers can optimize their production processes, minimize defects, reduce waste, and improve the overall quality of their products. Ultimately, this research will contribute to the advancement of the manufacturing industry, driving innovation and creating new opportunities for growth and development.