Abstract:Despite the plethora of deep learning frameworks for flow field prediction, most of them deal with flow fields on regular domains, and although the best ones can cope with irregular domains, they mostly rely on graph networks, so that real industrial applications remain currently elusive. We present a deep learning framework for 3D flow field prediction applied to blades of aircraft engine turbines and compressors. Crucially, we view any 3D field as a function from coordinates that is modeled by a neural network we call the backbone-net. It inherits the property of coordinate-based MLPs, namely the discretization-agnostic representation of flow fields in domains of arbitrary topology at infinite resolution. First, we demonstrate the performance of the backbone-net solo in regressing 3D steady simulations of single blade rows in various flow regimes: it can accurately render important flow characteristics such as boundary layers, wakes and shock waves. Second, we introduce a hyper-net that maps the surface mesh of a blade to the parameters of the backbone-net. By doing so, the flow solution can be directly predicted from the blade geometry, irrespective of its parameterization. Together, backbone-net and hyper-net form a highly-accurate memory-efficient data-driven proxy to CFD solvers with good generalization on unseen geometries.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **How to use the deep - learning framework to accurately predict the 3D flow field on complex geometric shapes (such as aircraft engine turbines and compressor blades), especially to achieve efficient and high - resolution prediction in irregular domains**. Specifically, although traditional computational fluid dynamics (CFD) solvers are accurate, they are very slow, especially when the fluid solver needs to be frequently called during the design optimization process. In addition, most of the existing deep - learning models can only handle the flow field prediction on regular grids and lack support for complex geometric shapes and irregular grids in industrial applications. To solve these problems, the author proposes a deep - learning framework based on Implicit Neural Representation (INR). The main features of this framework include: 1. **Backbone - Net**: - Regard the 3D flow field as a function from coordinates to flow - field features and model it through a neural network (called Backbone - Net). - This network inherits the characteristics of the coordinate - based multi - layer perceptron (MLP) and can represent the flow field with infinite resolution in domains of any topological structure. - By regressing the steady - state simulations of a single blade row under different flow conditions, it is verified that the Backbone - Net can accurately capture important flow - field features such as boundary layers, wakes, and shock waves. 2. **Hyper - Net**: - Introduce a Hyper - Net, which maps the surface grid of the blade to the parameters of the Backbone - Net. - In this way, the flow - field solution can be directly predicted from the blade geometry without considering its parameterization method. - The Backbone - Net and Hyper - Net together form an efficient and accurate data - driven surrogate model to replace the CFD solver and have good generalization ability on unseen geometries. ### Main contributions 1. **Implicit neural representation**: Propose an implicit neural representation method for CFD flow - field prediction, making the flow - field representation both compact and continuous. 2. **No additional pre - processing required**: This method does not need to interpolate the CFD flow field onto a Cartesian grid as in traditional methods, thus avoiding time - consuming and approximate pre - processing steps. 3. **Efficient prediction of complex flow fields**: Through the carefully designed architecture, it is shown that a small - scale coordinate MLP can faithfully predict the complete 3D flow field of turbomachinery configurations under various flow conditions, from subsonic to supersonic. 4. **Direct geometry - to - flow - field mapping**: Establish a direct mapping from blade geometry to aerodynamic flow fields, and the hyper - network can adjust the weights of the Backbone - Net according to any relevant information. 5. **Good generalization performance**: Experiments prove that this model can generalize well from a small number of samples. Through these innovations, the author aims to provide an efficient, accurate, and easy - to - deploy deep - learning framework to replace traditional CFD solvers, especially in complex industrial application scenarios.

Implicit Neural Representation For Accurate CFD Flow Field Prediction

Fast simulation of airfoil flow field via deep neural network

A Point-Cloud Deep Learning Framework for Prediction of Fluid Flow Fields on Irregular Geometries

Neural fields for rapid aircraft aerodynamics simulations

Predicting Pressure Fields from Incomplete Velocity Fields Based on Deep Convolutional Neural Network

Aero-Nef: Neural Fields for Rapid Aircraft Aerodynamics Simulations

Fast Transonic Flow Prediction Enables Efficient Aerodynamic Design

Fast Prediction of Flow Field Around Airfoils Based on Deep Convolutional Neural Network

FlowDNN: a Physics-Informed Deep Neural Network for Fast and Accurate Flow Prediction

Fast sparse flow field prediction around airfoils via multi-head perceptron based deep learning architecture

Investigation of physics-informed deep learning for the prediction of parametric, three-dimensional flow based on boundary data

The prediction of external flow field and hydrodynamic force with limited data using deep neural network

Unsteady Flow-Field Forecasting Leveraging a Hybrid Deep-Learning Architecture

Integration of Deep Learning and Computational Fluid Dynamics for Rapid Aerodynamic Force Prediction of Compressor Blades

Fast Flow Field Estimation for Various Applications with A Universally Applicable Machine Learning Concept

Predicting unsteady incompressible fluid dynamics with finite volume informed neural network

Deep Learning Methods for Reynolds-Averaged Navier-Stokes Simulations of Airfoil Flows

A fast three-dimensional flow field prediction around bluff bodies using deep learning

Deep learning-based predictive modelling of transonic flow over an aerofoil

A Novel Hybrid Deep Learning Method for Predicting the Flow Fields of Biomimetic Flapping Wings