Virtual testing and optimisation of aerospace composites using representative coupon data

Abstract

Composite materials, particularly carbon-fibre reinforced plastics, have been increasingly used in the aerospace industry due to their outstanding specific strength and stiffness properties. To minimise costs, material waste and improve production rates, the design, certi cation and quality control of large composite structures relies on extensive testing of small-scale coupons, which may be unrepresentative of the nal large-scale component. Unrepresentative coupon data can result in over-conservative, or at worst, unsafe designs of the final component, which are commonly avoided using restrictive industrial guidelines that limit the full potential of composite materials. The objective of this thesis is therefore twofold: to investigate and develop experimental and numerical methodologies for the generation of more representative coupon data; and to use these methodologies for the exploration and optimisation of composite laminate designs, including unconventional solutions not bounded by conservative guidelines.

An in-depth investigation of a resin edge treatment, used for suppression of edge effects in short beam coupons, is conducted using experiments, computed tomography and bespoke finite element models. The treatment can be easily applied to coupon tests, producing more representative strength predictions of large-scale components, where edge effects are either non-existent or less significant. A numerical version of the treatment is proposed for the generation of more representative coupon data using finite element models.

In order to aid the generation of reliable and representative coupon data, a general mesoscale finite element modelling framework is proposed for virtual testing of composite laminates. With this framework, very good agreement is found between simulations and experiments, with a maximum relative difference of 5% in predicted and experimental strengths for a variety of laminates. Very good qualitative agreement is also found between experimental and simulated failure morphologies.

The proposed finite element modelling framework is used along with a comprehensive testing campaign to investigate the case of open-hole tension coupon tests, considering a variety of unconventional laminate designs, under both on-axis and misaligned loading, and with suppressed and unsuppressed edge effects (i.e with or without edge treatment). It is found that unconventional laminates with significant ply blocking can outperform conventional designs due to beneficial accumulation of sub-critical interlaminar and intralaminar matrix damage, but are also more prone to edge-driven failure, particularly under misaligned loading. In such cases, coupon-level predictions with unsuppressed edge effects are found to largely misrepresent large-scale components. Conversely, by suppressing edge effects, strength predictions increase in up to 80%, bringing into question the validity of standard coupon-level testing procedures for some applications.

Lastly, a novel data-driven Bayesian optimisation framework is proposed for the design of stronger stacking sequences, whilst enforcing other design requirements, typical in the aerospace industry, such as speci ed in-plane sti ness, laminate balance and symmetry. The framework is the first to incorporate high-fidelity finite element modelling of composite laminates within a data-driven optimisation methodology. The optimisation framework is applied to the stacking sequence selection of open-hole tension coupons, and benchmarked using three cases studies of increasing complexity, including both conventional and unconventional laminates, on-axis and misaligned loading, and suppressed and unsuppressed edge effects. It is shown that the optimisation framework can find relatively high-performance laminates, including unconventional stacking sequences, from a small number of iterations and in instances where most global optimisation methods are intractable, due to the large number of finite element model evaluations required. As such, the proposed optimisation framework can be used as a pragmatic and computationally efficient early-design tool.

Date of Award	17 Nov 2021
Original language	English
Awarding Institution	University of Bath
Supervisor	Andrew Rhead (Supervisor) & Richard Butler (Supervisor)