Friday 14 March 2025
The quest for efficient and accurate simulations of complex fluid flows has been a long-standing challenge in science and engineering. From understanding ocean currents to designing more aerodynamic aircraft, grasping the intricacies of fluid dynamics is crucial for tackling many real-world problems. However, traditional methods often rely on computationally expensive and time-consuming simulations, making it difficult to explore large design spaces or predict flow behavior under various conditions.
A recent paper proposes a novel approach that combines machine learning with physics-informed neural networks (PINNs) to tackle this challenge. The authors developed an algorithm that uses elliptic input features (EIFs) to train a ML model, which can then be used to predict the flow field around complex geometries with high accuracy and speed.
The EIFs are derived from the governing equations of fluid dynamics, such as Navier-Stokes, and encode the underlying physics of the flow. By using these features as inputs to the ML model, the authors were able to train a neural network that can accurately predict the flow behavior around a Joukowski airfoil, a complex geometry with a curved upper surface and a sharp trailing edge.
The key innovation lies in the use of PINNs, which are designed to satisfy the physical laws governing the system being modeled. In this case, the ML model is trained to minimize the difference between the predicted flow field and the actual solution of the Navier-Stokes equations. This ensures that the predictions made by the model are not only accurate but also physically meaningful.
The authors demonstrate the effectiveness of their approach by simulating the flow around a Joukowski airfoil at various angles of attack and Mach numbers. They show that their ML model can accurately predict the flow behavior, including the formation of vortices and separation bubbles, with high precision.
Moreover, the authors highlight the potential benefits of this approach for real-world applications, such as reducing the computational cost of CFD simulations by using PINNs to initialize the solver or accelerating the convergence of iterative solvers. They also suggest that their method can be extended to other complex systems, such as turbulence modeling and multiphase flows.
The development of this novel approach has significant implications for a wide range of fields, from aerospace engineering to environmental science. By combining the strengths of machine learning and physics-informed neural networks, researchers can now tackle complex fluid flow problems with unprecedented accuracy and speed, opening up new possibilities for innovation and discovery.
Cite this article: “Accurate and Efficient Simulations of Complex Fluid Flows using Machine Learning and Physics-Informed Neural Networks”, The Science Archive, 2025.
Fluid Dynamics, Machine Learning, Physics-Informed Neural Networks, Pinns, Elliptic Input Features, Eifs, Navier-Stokes Equations, Flow Field Prediction, Aerospace Engineering, Environmental Science
