Solving large optimisation problems for improved robustness in dry fabrics forming applications

Abstract

In this thesis, a series of digital methods aiming to enhance the robustness of
composite manufacturing processes are developed. Specifically, the dry Non-Crimp
Fabric (NCF) forming process (which is a crucial stage in Liquid Composite Moulding
(LCM)) serves as a case study to demonstrate the application of these methods. This
work addresses two major challenges: the high costs associated with process
optimisation to eliminate wrinkles, and the impact of material variability on
manufacturing robustness.
To overcome these challenges, a model-based process optimisation framework was
developed, combining a Finite Element (FE) simulator with a Gaussian Process (GP)
emulator. This strategy uses a minimum amount of computationally expensive FE
simulations to train a GP surrogate, which is then utilised to efficiently identify optimal
process parameters. Dimension reduction and active learning techniques were adopted
to maximise the data efficiency. The impact of material variability and its propagation
were experimentally investigated by using 3D Digital Image Correlation (DIC) system.
Finally, the proposed optimisation framework was utilised to enhance the robustness of
an NCF forming process. Two strategies were compared: a robust design approach,
involving one-time optimisation resilient to variabilities, and a digital twin approach,
featuring real-time optimisation based on in-situ measurements of material and process
conditions. The results demonstrated that both strategies significantly reduced wrinkling
levels and improved process robustness, validating the effectiveness of the proposed
methods.

Date of Award	4 Feb 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Jonathan P Belnoue (Supervisor), Stephen R Hallett (Supervisor), Adam J Thompson (Supervisor) & Tim Dodwell (Supervisor)