Engineering Design Optimisation Using Computational Fluid Dynamics and Human-AI Collaboration

Abstract

The primary contributions of this thesis are a comparison of a ducted winglet to two other geometries and a user study which investigated the relationship between engineers and an optimisation algorithm. With an ever increasing emphasis on sustainable and renewable energy production, new technology can increase the efficiency of existing infrastructure such as wind turbines. Aerodynamic devices known as winglets can be fitted to the end of wind turbine blades to increase efficiency and reduce negative downstream effects. A patented winglet introduces a duct that channels freestream air from the bottom of the winglet to the top. The effectiveness of this design is simulated using computational fluid dynamics and compared to two alternative designs across a range of angles of attack. The patented winglet had lower induced drag across a range of angles of attack. Following the comparison, optimisation methods including evolutionary algorithms are employed to further increase the efficiency of the winglet. These algorithms can be used when engineers would otherwise rely on intuition, preconceptions, and theory to try and find optimal designs. The manner in which engineers utilise and engage with an evolutionary algorithm known as MAP-Elites is evaluated through a user study. 12 participants were given 20 minutes each to design a car that could travel over an inclined course with obstacles as far as possible. Their behaviour was compared to survey answers they provided before and after designing cars. Participants who engaged with the MAP-Elites algorithm outperformed baseline designs created by the computer and participants who did not engage with the algorithm. Participants were more likely to use the MAP-Elites algorithm even if they were unaware they were using it or if they had stated that they did not trust optimisation algorithms. Shortening the optimisation cycle is explored during the optimisation of the ducted winglet. The effects of increasing uncertainty in the results are explored through two studies on the robustness of evolutionary algorithms and a study on 2-dimensional aerofoils. The studies show that the optimisation cycle can be reduced to a certain extent while maintaining the original ranking of the designs.