CN-122021394-A - Fluid simulation acceleration system based on AI model

CN122021394ACN 122021394 ACN122021394 ACN 122021394ACN-122021394-A

Abstract

The invention discloses a fluid simulation acceleration system and method based on an AI model, and relates to the field of fluid simulation technology and artificial intelligence. The system utilizes the AI model to predict and replace a high-complexity calculation process in fluid simulation by constructing a fluid characteristic extraction module, an AI prediction model training module, a simulation data adaptation module and an acceleration simulation execution module, and solves the technical problems of low calculation efficiency, long time consumption, high hardware resource consumption and the like caused by the traditional fluid simulation dependent numerical value solution. According to the invention, the internal law of fluid motion is learned through the AI model, so that the quick prediction of a simulation result is realized, the fluid simulation efficiency is improved by 50% -90% on the premise that the simulation precision meets the engineering requirement, and the simulation method is suitable for various scenes requiring fluid simulation such as aerospace, automobile engineering, hydraulic engineering and the like, and has a wide application prospect.

Inventors

WANG YIDING
LI YUANYI
YUAN DEKUI
SUN JIAN

Assignees

天津大学

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (8)

1. An AI model-based fluid simulation acceleration system, comprising: The fluid characteristic extraction module is used for acquiring initial parameter data of fluid simulation, preprocessing the initial parameter data, and extracting key characteristic parameters of fluid movement, wherein the key characteristic parameters comprise fluid density, viscosity, flow velocity distribution, boundary condition parameters and geometric structure parameters; The AI prediction model training module is used for constructing an AI prediction model based on historical fluid simulation data, wherein the historical fluid simulation data comprises historical initial parameters, corresponding numerical solution simulation results and key characteristic parameters; the simulation data adaptation module is used for receiving initial input parameters of the fluid to be simulated, calling the fluid characteristic extraction module to extract key characteristics of the fluid to be simulated, and converting the key characteristics into an input format which can be identified by the AI prediction model; The accelerated simulation execution module is used for inputting the adapted input data into a trained AI prediction model, outputting a simulation result predicted value of the fluid to be simulated through the AI prediction model, directly outputting the predicted value as a final simulation result if the error of the predicted value meets a preset threshold value, taking the predicted value as an initial iteration value of a numerical solution if the error does not meet the preset threshold value, starting a simplified numerical calculation process, and outputting a corrected simulation result.
2. The AI model-based fluid simulation acceleration system of claim 1, wherein the preprocessing of the fluid feature extraction module comprises data cleaning to remove outliers and missing values, data normalization to unify parameter magnitudes, and data dimension reduction to preserve core features, the data dimension reduction employing Principal Component Analysis (PCA) or Linear Discriminant Analysis (LDA) algorithms.
3. The fluid simulation acceleration system based on an AI model of claim 1, wherein the AI prediction model adopts a hybrid deep learning architecture and comprises a feature encoding layer, a time sequence prediction layer and a result decoding layer, wherein the feature encoding layer adopts a Convolutional Neural Network (CNN) to extract spatial features, the time sequence prediction layer adopts a long and short term memory network (LSTM) or a gate control circulation unit (GRU) to capture time sequence correlation of fluid movement, and the result decoding layer adopts a fully connected network to output a simulation result predicted value.
4. The AI model-based fluid simulation acceleration system of claim 1, further comprising a model optimization module that optimizes the AI prediction model through adaptive learning rate adjustment, regularization, and an ensemble learning strategy that employs weighted fusion of prediction results for a plurality of sub-models, the weights being dynamically adjusted based on validation set accuracy for the sub-models.
5. The fluid simulation acceleration method based on the AI model is characterized by comprising the following steps: The method comprises the following steps of S1, data acquisition and preprocessing, namely collecting fluid simulation historical data in different scenes, wherein the fluid simulation historical data comprises initial parameters, boundary conditions, geometric parameters and corresponding numerical solution simulation results; s2, constructing and training an AI prediction model, namely constructing a mixed deep learning model integrating space feature extraction and time sequence prediction based on the physical characteristics of fluid movement, training the model by utilizing a training data set, monitoring the model precision by verifying the data set, and adjusting model parameters by adopting a self-adaptive optimization algorithm until the model precision meets the preset requirement; S3, adapting data to be simulated, namely acquiring target object parameters to be subjected to fluid simulation, including geometric structure parameters, fluid physical parameters and boundary condition parameters, extracting key features of the target object through a feature extraction algorithm, and converting the key features into data matched with an input format of an AI prediction model; S4, AI accelerates simulation execution, namely inputting the adapted target object data into a trained AI prediction model to obtain a simulation result predicted value, calculating an error of the predicted value and a numerical solution reference value, outputting the predicted value as a final simulation result if the error is less than or equal to a preset threshold value, taking the predicted value as an initial iteration value of the numerical solution if the error is greater than the preset threshold value, executing a simplified numerical calculation flow (reducing iteration times or simplifying calculation grids), obtaining a corrected simulation result and outputting the corrected simulation result; And S5, model iterative updating, namely collecting new simulation data (comprising an AI prediction result and an actual verification result), carrying out incremental training on the AI prediction model regularly, optimizing model parameters, and improving the adaptation capability of the model to complex scenes.
6. The AI model-based fluid simulation acceleration method of claim 5, wherein the adaptive optimization algorithm in step S2 adopts Adam algorithm or RMSProp algorithm, and the evaluation indexes of the model accuracy include Mean Absolute Error (MAE), root Mean Square Error (RMSE) and determination coefficient (R 2 ), and the preset requirement is R 2 is greater than or equal to 0.92 and RMSE is less than or equal to 5%.
7. The AI-model-based fluid simulation acceleration method of claim 5, wherein the simplified numerical computation process in step S4 includes optimizing computation meshing based on AI prediction results, performing coarse meshing computation on the fluid motion plateau region, and preserving fine meshing computation on the high-gradient region, while taking AI prediction results as initial values of a numerical solution, and reducing the number of times required for iterative convergence.
8. The AI-model-based fluid simulation acceleration method of claim 5, wherein the incremental training in step S5 is performed in an online learning mode, and only the newly added data is updated with local parameters, so that full-scale data retraining is avoided, and the time cost of model updating is reduced.

Description

Fluid simulation acceleration system based on AI model Technical Field The invention relates to the crossing field of fluid simulation technology and artificial intelligence, in particular to a fluid simulation acceleration system and method based on an AI model. Background The fluid simulation technology is widely applied to various industrial fields such as aerospace, automobile design, hydraulic engineering, energy power and the like, and is characterized in that a Navier-Stokes equation is solved through a numerical solution (finite volume method, finite element method and the like), and the motion state of fluid and the interaction with a solid boundary are simulated. However, the traditional fluid simulation technology has obvious technical bottlenecks that on one hand, in order to ensure the simulation precision, a calculation domain needs to be finely grid-divided, so that the calculation amount grows exponentially, the simulation process often needs to take hours or even days, the product research and development period is seriously influenced, on the other hand, the fluid movement (such as turbulence and multiphase flow) in a complex scene has strong nonlinearity and time sequence correlation, a numerical solution method needs a large amount of iterative calculation to converge, extremely high requirements are put forward on hardware calculation resources (CPU and GPU), and the simulation cost is increased. In the prior art, in order to improve the fluid simulation efficiency, modes of simplifying a physical model, reducing grid precision or optimizing a numerical algorithm and the like are generally adopted, but the methods usually have the cost of sacrificing the simulation precision, and are difficult to meet the requirements of high-precision engineering simulation. In recent years, the rapid development of artificial intelligence technology provides a new thought for solving the contradiction, and the internal law of fluid motion is learned through an AI model, so that the rapid prediction of simulation results is realized, and the simulation efficiency is expected to be greatly improved on the premise of ensuring the precision. However, the existing fluid simulation method based on AI still has the following problems that firstly, the adaptation of an AI model to the physical characteristics of the fluid is insufficient, the prediction precision is difficult to meet engineering requirements, secondly, the model generalization capability is poor and is only suitable for fluid simulation of specific scenes, thirdly, a complete AI prediction-numerical correction closed loop is not formed, and the simulation requirement under complex working conditions is difficult to cope with. Therefore, a fluid simulation acceleration system and method based on an AI model with high precision, efficiency and generalization capability are needed. Disclosure of Invention The invention aims to overcome the defects of the prior art, provides a fluid simulation acceleration system and a fluid simulation acceleration method based on an AI model, and solves the technical problems of long time consumption and high resource consumption of the traditional fluid simulation by constructing an AI prediction model fused with physical characteristics and combining a hybrid simulation mode of 'AI fast prediction and numerical value accurate correction'. The technical scheme adopted by the invention is that the fluid simulation acceleration system based on the AI model comprises: The fluid characteristic extraction module is used for acquiring initial parameter data of fluid simulation, preprocessing the initial parameter data, and extracting key characteristic parameters of fluid movement, wherein the key characteristic parameters comprise fluid density, viscosity, flow velocity distribution, boundary condition parameters and geometric structure parameters; The AI prediction model training module is used for constructing an AI prediction model based on historical fluid simulation data, wherein the historical fluid simulation data comprises historical initial parameters, corresponding numerical solution simulation results and key characteristic parameters; the simulation data adaptation module is used for receiving initial input parameters of the fluid to be simulated, calling the fluid characteristic extraction module to extract key characteristics of the fluid to be simulated, and converting the key characteristics into an input format which can be identified by the AI prediction model; The accelerated simulation execution module is used for inputting the adapted input data into a trained AI prediction model, outputting a simulation result predicted value of the fluid to be simulated through the AI prediction model, directly outputting the predicted value as a final simulation result if the error of the predicted value meets a preset threshold value, taking the predicted value as an initial iteration value of a numerical so