Modern engineering design workflows are inherently iterative, with each iteration typically requiring new sets of computationally expensive evaluations while previously generated data is discarded. This thesis proposes a methodology for reusing data from previous design iterations within a multi-fidelity (MF) Gaussian Process (GP) surrogate modelling framework. The data from a previous design iteration, referred to as the source, is fused with a limited number current design evaluations, referred to as the target, using Multi-Fidelity Co-Kriging (MFCK). The source data serves as the low-fidelity input and the target data as the high-fidelity input within the MFCK. The squared Pearson correlation coefficient is adopted as an a priori metric for assessing LF source quality, while a post-hoc discrepancy share diagnostic is introduced to evaluate the contribution of the LF source from the trained model’s internal perspective.
The methodology is validated on two structural design problems of increasing complexity: a three-dimensional cantilever beam problem with a readily available analytical solution and a six-dimensional stiffened cylindrical shell problem, evaluated using finite element analysis. The findings suggest that the effectiveness of the approach depends primarily on the quality of the source data: fusion of well-correlated designs can substantially reduce the number of required target evaluations while maintaining comparable precision and improved robustness. Poorly correlated designs, on the other hand, can degrade the performance below the singlesource counterpart.
... Read more

Inverse design is a concept where one can design a structure via a property-first approach, where the properties of a system act as precursor information to guide the search for a viable design. This concept is classified as one of two: indirect or direct. While the former is no different than a forward-based approach using traditional optimization, a direct inverse design framework attempts to invert the process and directly map the desired properties to a suitable design candidate, which is unconventional. As such, the design process can be positively radicalized, with regards to the required computational resources when designing multiple structures.

In this work, the use of Bayesian machine learning, more specifically Bayesian optimization using Gaussian process regression, is employed to construct a direct inverse design framework using the design of mass-optimized fixed-wing aircraft ribs against buckling as the validation case study in 1, 6, and 10 dimensions. The results are compared to an indirect approach using the same algorithms and demonstrate that barring certain limitations, which are not inherent flaws of the direct approach, the framework designed not only demonstrates sound potential in inverting the forward map, but outperforms the indirect approach when designing multiple structures.

This research equally provides a stepping stone towards future research possibilities in the same field, all culminating in the improvement of the multi-disciplinary design process.
... Read more

In the renewed race for the Moon, there will be a high demand for transportation services to and from the lunar surface. The next generation of landers needs to have many capabilities, such as supporting longer exploration and global access, sustainability and versatility. ARCH-E, developed together with Airbus Defence and Space UK, fills this need as a reusable, crew/cargo-capable lander architecture. The lander's requirements and market are analysed, mission trajectory and landing are designed, followed by detailed subsystem design. The operations, risks, costs and sustainability considerations are assessed and highlighted. Finally, the design overview of a feasible lunar landing system is presented. Further design activities are recommended and a project timeline is laid out. ... Read more

This study investigates the development and application of meta-models for crashworthiness assessment of helicopter structures and components. It aims to address the challenges associated with scarcity of data from computationally expensive simulations and experimental drop-tests, and enable the use of surrogates in a crashworthiness optimization framework. Two predictive approaches utilizing Machine Learning techniques are compared to predict and assess the energy absorption of tubular metallic structures for different cross-section configurations. The first approach directly predicts energy absorption, while the second predicts load-displacement curves, from which energy absorption is derived. Results indicate that certain regressors, such as the Transform Target Regressor, the Decision Tree Regressor and the Poisson Regressor, consistently achieve high accuracy in predicting load-displacement curves and energy absorption across the evaluated tubular samples. A low-fidelity model able to provide less accurate but computationally inexpensive information is then introduced. The influence of low-fidelity data is investigated when it serves as additional input alongside high-fidelity data during the training phase of the surrogate model, through a comparative analysis. The research's findings suggest the efficiency of Machine Learning in representing structural behaviour under crushing conditions and highlight the potential for further enhancements through the integration of low-fidelity data, thereby holding promise for extending the methodology to more complex structures. ... Read more