Reducción de complejidad computacional en simulaciones de dinámica de fluidos mediante aprendizaje profundo y representación simbólica

Guerrero Barberán, Jose Antonio

Reducción de complejidad computacional en simulaciones de dinámica de fluidos mediante aprendizaje profundo y representación simbólica

Abstract

This work is a comparative study that explores the application of Deep Learning and Symbolic Regression to reduce computational complexity in fluid dynamics simulations through an increase in resolution task in the spatial dimensions of the simulations. The work primarily focuses on addressing the challenges in computational fluid dynamics (CFD) by leveraging advanced machine learning techniques. It delves into the fundamentals of partial differential equations, particularly the Navier-Stokes equation, and discusses various computational solutions, including the use of deep neural networks, convolutional networks, and symbolic regression models. A dataset of CFD simulations that represent a turbulent flow from random initial conditions is computed using direct numerical methods. This work offers a comparative analysis studies different methodologies, models and architectures that can be used in the superresolution process, highlighting their effectiveness in producing highresolution simulations from lower-resolution data. Emphasis is placed on understanding the trade-offs between computational cost, accuracy, and the potential of AI-driven methods in CFD. This research also includes an analysis of the computational complexity involved and the predictive capabilities of the implemented models, contributing valuable insights into the integration of AI in fluid dynamics and a set of different architectures that can take advantage of the potential of machine learning.

Citation

Guerrero Barberán, J. (2024). Reducción de complejidad computacional en simulaciones de dinámica de fluidos mediante aprendizaje profundo y representación simbólica. Universitat Oberta de Catalunya (UOC). https://hdl.handle.net/10609/149570

This citation was generated automatically.

Full item page