Machine Learning in Aerodynamics: Clustering and Autoencoders for CFD Applications

In the past decade, Machine Learning (ML) has emerged as a prominent topic in various scientific fields, including Fluid Mechanics. This dissertation delves into ML algorithms in Computational Fluid Dynamics, with a specific emphasis on unsupervised methods.In a first part, a well-established clustering algorithm is repurposed for a novel application: physical-based domain decomposition. This is crucial in various fields, ranging from the shape optimization to the decomposition of aerodynamic forces. The ML method demonstrated its ability in overcoming the limitations of existing deterministic algorithms employed to date.In a second part, Autoencoders (AE) take center stage for real-time predictions of flow fields around wing sections. The investigation on AEs begins with the assessment of their accuracy in predicting aerodynamic flow fields, particularly in strongly non-linear regimes. Subsequently, it also addresses some of the current challenges in ML techniques, encompassing the interpretability of the learning procedure, the volume of data required for algorithm training, and the Uncertainty Quantification of AE predictions. This part is pivotal for ensuring the safe deployment of ML technologies in real-world systems.The proposed applications, in this dissertation, serve as evidence of the significant impact that ML can wield on aerodynamic analyses in the near future.