AI-ordlista
Den kompletta ordlistan över AI
Navier-Stokes Equations
System of partial differential equations describing the motion of Newtonian fluids, combining conservation of mass and momentum, fundamental for flow modeling.
Physics-Informed Neural Networks (PINNs)
Neural network architecture integrating physical laws as constraints in the loss function, enabling the solution of partial differential equations without supervised training data.
AI-Assisted CFD
Hybrid approach combining traditional computational fluid dynamics with machine learning techniques to accelerate simulations and improve prediction accuracy.
AI-Based Turbulence Modeling
Use of neural networks to model turbulent subgrid scales, replacing traditional RANS models with data-driven approaches.
Model Reduction by Autoencoders
Dimensionality reduction technique using autoencoders to identify principal modes of fluid flows, enabling real-time simulations while preserving physics.
Neural Lattice Boltzmann Methods
Simulation approach where the collision and streaming operators of the Lattice Boltzmann method are replaced by neural networks to improve computational efficiency.
GANs for Flow Generation
Generative adversarial networks trained to produce realistic fluid flow fields, accelerating the generation of initial and boundary conditions for CFD simulations.
Hybrid AI-Physics Solvers
Algorithms combining traditional numerical solvers with deep learning models to correct errors and accelerate convergence of fluid simulations.
Transfer Learning in Fluid Dynamics
Adaptation of AI models pre-trained on certain flow regimes to predict behaviors under different conditions, significantly reducing computational costs.
U-Net for Flow Segmentation
U-shaped convolutional neural network architecture specifically optimized to identify and segment coherent structures in velocity and pressure fields of turbulent flows.
AI-Driven Topological Fluid Optimization
Application of reinforcement learning to optimize geometric shapes in fluid systems, minimizing pressure losses or maximizing mixing performance.
Neural Subgrid-Scale Models
Neural networks replacing traditional LES turbulence models to capture the effects of unresolved small scales directly in large-scale simulations.
Hydrodynamic Instability Prediction
Use of temporal models such as LSTM to predict the onset and evolution of instabilities in flows, enabling preventive active control.
Deep Collocation in Fluid Mechanics
Training method where collocation points are intelligently distributed in the physical domain to ensure accurate representation of the Navier-Stokes equations.
Variational Autoencoders in CFD
Generative models learning the latent distribution of flow states to enable interpolation, extrapolation, and sampling of new coherent fluid fields.
Mesh-Free Neural Solvers
Deep learning-based solvers eliminating the need for traditional meshing, using radial basis functions or other meshless approaches to solve flows.
Graph Neural Networks for Fluid Networks
Application of graph neural networks to model flows in complex channel or pipe networks where the system topology is naturally represented by a graph.
Active Flow Control with Reinforcement Learning
Use of reinforcement learning to develop optimal real-time control strategies, reducing drag or eliminating boundary layer separations.