Multi-fidelity probabilistic optimisation of composite structures

Abstract

In this thesis, novel multi-fidelity modelling-based probabilistic optimisation methods are presented to address the computational challenge of stochastic design philosophies applied to complex aircraft composite structures. Novel multi-fidelity formulations developed in this thesis, blending High-Fidelity Model (HFM) and Low-Fidelity Model (LFM), are shown to significantly improve computational efficiency by making use of machine learning techniques, such as Artificial Neural Networks (ANN) and Non-linear Auto-Regressive Gaussian Process (NARGP). To further improve the computational efficiency compared to the conventional probabilistic optimisation methods, a multi-level optimisation approach and a new sampling strategy to collect training data points are incorporated into the multi- fidelity formulations for the first time. In the developed optimisation methods, the HFM covers part of the design space whilst the LFM explores the whole design space to fill the lack of high-fidelity information. This improvement enables the multi-fidelity formulations to request a much smaller number of high-fidelity information causing considerable computational costs. Several engineering examples such as aircraft mono-stringer composite panels are used to demonstrate the accuracy and computational efficiency of the developed methods when used with different reliability and robustness analysis techniques, including Monte Carlo Simulation (MCS), the First-Order Reliability Method (FORM) and the Second-Order Reliability Method (SORM). The composite panels are subjected to mechanical and thermomechanical loads to show the broad range of potential applications. It is shown that the newly developed multi-fidelity probabilistic optimisation methods offer substantial computational time savings ranging from 50 % to 70 % and levels of error typically less than 1 % when compared with traditional probabilistic optimisation methods. Results demonstrate that the newly developed multi-fidelity probabilistic optimisation methods herein provide significant computational benefits and accurately predict the influence of uncertainties associated with design and manufacturing stages. As a result, the presented methods confidently carry out reliability-based and robust design optimisation of large-scale and complex aircraft composite structures.