A comparison between robust design and digital twin approaches for Non-Crimp fabric (NCF) forming

There is growing interest in adopting digital twin systems within the field of composites manufacturing. However, given the current limitations in measuring variability and accurately simulating complex defects, it remains questionable as to whether the high costs of building a digital twin are justified. In this paper, a case study is conducted on simulation-driven optimisation of the forming of non-crimp fabric (NCF). A robust design strategy (a one-time optimisation that is robust to variabilities of the material and process) is compared with a digital twin approach (active control is conducted based on real-time optimisation, accounting for in-situ measurements of variabilities). An optimisation method based on a Gaussian process (GP) surrogate model, active learning, dimension reduction and gradient boosting is developed. This method enables the optimisation of complex forming processes with a very small dataset, built from large simulation models. Both strategies significantly reduce the wrinkling level and improve process robustness.