3D Fluid Reverse Modeling via Neural Network Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for capturing and modeling three-dimensional fluid flow fields from real-world data are challenging due to the complexity of equipment required and the difficulty in accurately reconstructing internal flow fields, especially when dealing with complex scenes and uncalibrated fluid surface motions, which limits their applicability and efficiency.
Innovation Solution
A three-dimensional fluid reverse modeling method based on physical perception that combines deep learning with traditional physical simulation, using a two-step convolutional neural network to encode and decode spatiotemporal features from surface motion, estimate fluid parameters, and reconstruct flow fields, thereby simplifying the process and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex devices and techniques (synchronous cameras, staining solutions, laser equipment) are used to capture three-dimensional flow fields, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent replaces complex mechanical measurement devices (synchronous cameras, laser equipment, staining solutions) with a data-driven machine learning system that processes simple fluid surface motion images. The neural network model learns to infer three-dimensional flow field characteristics from two-dimensional surface observations, substituting physical measurement complexity with computational intelligence.
Solution Approach 2:
The patent creates a virtual copy of the complex measurement system through a trained neural network model. Instead of physically capturing flow fields with complex equipment, the system learns from training data to generate accurate flow field reconstructions from simple images, effectively copying the measurement capability in software rather than hardware.
2Ease of operation
If traditional graphics-based geometric reconstruction is used from fluid videos, then ease of operation is improved, but manufacturing precision (reconstruction accuracy) deteriorates
Solution Approach 1:
The patent replaces traditional graphics-based geometric reconstruction algorithms with a machine learning-based inference system. The neural network model, trained on paired data of surface images and corresponding flow fields, directly predicts flow field characteristics from surface motion, achieving both operational simplicity and high reconstruction accuracy.
3Adaptability or versatility
If manual parameter adjustment through trial-and-error is used for fluid re-simulation, then adaptability is improved, but productivity deteriorates
Solution Approach 1:
The patent replaces manual trial-and-error parameter adjustment with an automated machine learning system. The neural network model automatically determines optimal fluid parameters by learning from training data, enabling rapid adaptation to different fluid scenes without manual intervention while maintaining or improving re-simulation quality.
Solution Approach 2:
The system performs self-service by automatically determining fluid parameters through the trained model without requiring manual trial-and-error adjustment. The model independently learns and adapts to different fluid characteristics, freeing users from time-consuming manual parameter tuning while maintaining versatility.
4Measurement precision
If forward physical simulation with reverse parameter optimization is used for parameter determination, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent inverts the traditional approach by training the model forward during training (learning from ground truth data) and then using it for direct inference during application. This reverses the time-consuming iterative optimization process, achieving both high parameter accuracy and fast execution by pre-learning the mapping from images to parameters.
Solution Approach 2:
The patent performs preliminary action by pre-training the neural network model on large datasets of fluid simulations with known parameters. This upfront training phase captures the complex relationships between surface motion and flow field characteristics, so that during actual use, parameter determination occurs instantly without iterative optimization, significantly reducing time loss.
Data Source
AI summary
A three-dimensional fluid reverse modeling method based on physical perception. The method comprises: encoding a fluid surface height field sequence by a surface velocity field convolutional neural network to obtain a surface velocity field at a time t; inputting the surface velocity field into a pre-trained three-dimensional convolutional neural network to obtain a three-dimensional flow field, wherein the three-dimensional flow field includes a velocity field and a pressure field; inputting the surface velocity field into a pre-trained regression network to obtain fluid parameters; and inputting the three-dimensional flow field and the fluid parameters into a physics-based fluid simulator to obtain a time series of the three-dimensional flow field. The requirements for real fluid reproduction and physics-based fluid reediting are met.