CN-122020882-A - PINN-based gear point cloud statics stress simulation method and system

CN122020882ACN 122020882 ACN122020882 ACN 122020882ACN-122020882-A

Abstract

The invention discloses a PINN-based gear point cloud statics stress simulation method and system, and relates to the technical field of solid mechanics simulation analysis. The method comprises the steps of obtaining and importing a target gear geometric model, carrying out sparse grid rapid finite element analysis on the target gear geometric model, deriving global grid and boundary information, establishing a point cloud model based on the global grid and the boundary information, determining boundary conditions, establishing PINN a physical information neural network based on the point cloud model and the boundary conditions, constructing a loss function, carrying out multiple stage network training to obtain a PINN prediction result, and outputting the prediction result as a simulation result. The method can directly utilize the point cloud data, realize the accurate prediction of the physical field through a staged training strategy, and remarkably improve the convenience, accuracy and engineering practical value of complex structure analysis.

Inventors

DING HAOYUAN
SHI LUBING
LIU XIAOKUN
XUE JIAQI
LIN RUIYI
Peng teng

Assignees

郑机所(郑州)传动科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260109

Claims (10)

1. A PINN-based gear point cloud statics stress simulation method is characterized by comprising the following steps: acquiring and importing a target gear geometric model, carrying out sparse grid rapid finite element analysis on the target gear geometric model, and deriving global grid and boundary information; Establishing a point cloud model based on the global grid and the boundary information, and determining boundary conditions; and constructing PINN a physical information neural network based on the point cloud model and the boundary condition, constructing a loss function, performing multiple stage network training to obtain a PINN prediction result, and outputting the prediction result as a simulation result.
2. The PINN-based gear point cloud hydrostatic force simulation method according to claim 1, wherein the boundary conditions comprise grid information, displacement and stress information of boundaries.
3. The method of claim 2, further comprising applying corresponding displacement constraints on the grid of the fixed surface and applying corresponding stress conditions on the grid of the stressed surface.
4. The method for simulating gear point cloud hydrostatic forces based on PINN of claim 3, further comprising computing a control equation residual, a displacement boundary residual, and a force boundary residual in a universe.
5. The PINN-based gear point cloud hydrostatic force simulation method as claimed in claim 4, wherein the loss function is formed by weighting three parts of a control equation residual error, a displacement boundary residual error and a force boundary residual error.
6. The PINN-based gear point cloud hydrostatic force simulation method of claim 1, wherein the method for acquiring and importing the target gear geometric model comprises the following steps: And obtaining a gear geometric model file provided by a user or completing parameterized three-dimensional geometric modeling of the gear transmission system by utilizing computer aided design software to obtain a target gear geometric model.
7. The PINN-based gear point cloud statics force simulation method according to claim 1, further comprising the following steps: and modifying the initial boundary conditions and the model parameters, performing network training again, and outputting simulation results.
8. The gear point cloud statics atress simulation system based on PINN is characterized by comprising a target importing and analyzing module, a point and boundary determining module and a neural network analyzing module, wherein: The target importing and analyzing module is used for acquiring and importing a target gear geometric model, carrying out sparse grid rapid finite element analysis on the target gear geometric model, and deriving global grid and boundary information; the point and boundary determining module is used for establishing a point cloud model based on the global grid and the boundary information and determining boundary conditions; the neural network analysis module is used for constructing PINN physical information neural networks based on the point cloud model and boundary conditions, constructing a loss function and carrying out multiple stage network training to obtain a PINN prediction result; And the verification output module is used for judging whether the prediction result reaches the preset precision, and if so, outputting the prediction result as a simulation result.
9. An electronic device, comprising: A memory for storing one or more programs; A processor; The method of any of claims 1-7 is implemented when the one or more programs are executed by the processor.
10. A computer readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, implements the method according to any of claims 1-7.

Description

PINN-based gear point cloud statics stress simulation method and system Technical Field The invention relates to the technical field of solid mechanics simulation analysis, in particular to a PINN-based gear point cloud statics stress simulation method and system. Background The Physical Information Neural Network (PINN) is used as a front edge method for combining scientific calculation and deep learning, and the core idea is to embed a partial differential equation for controlling a physical process into a training target of the neural network so as to guide a data-driven learning process by utilizing a physical rule. The effectiveness of the method is significant for realizing high-precision and high-efficiency simulation analysis of complex engineering structures. Currently, standard PINN methods based on mainstream frameworks (e.g., deepXDE) have formed a relatively fixed flow paradigm. To further improve performance, the research community has developed various strategies for improvement, such as introducing adaptive weight balancing mechanisms to coordinate magnitude differences of different loss terms, using conservative PINN (cpiinn) to handle multi-scale or intermittent problems through domain decomposition, and developing Variance PINN (VPINN) based on weighted residue method to improve numerical stability. However, the existing PINN technology exposes significant limitations in solid-state mechanical analysis for real engineering scenarios. Current applications are mostly limited to the problem of line elasticity of two-dimensional or simple three-dimensional regular geometries. When an actual structure containing complex geometric features is processed, the prior art faces serious challenges in the aspects of complex curved surface configuration point sampling, boundary condition application and the like, so that the solving precision is obviously reduced, the prior art seriously depends on accurate mathematical definition of the boundary conditions, and all boundaries of the whole calculation domain, such as a displacement constraint surface and an external force acting surface, are required to be given by clear mathematical expressions. For the practical engineering problems of complex boundary shape or various load working conditions, the applicability of the boundary condition is greatly limited by the strong dependence of the boundary condition, and the practical situation of unknown boundary cannot be effectively processed. Disclosure of Invention In order to overcome the problems or at least partially solve the problems, the invention provides a PINN-based gear point cloud statics stress simulation method and a PINN-based gear point cloud statics stress simulation system, which can directly utilize point cloud data from CAD/CAE software without explicit programming definition of complex geometric shapes and boundary conditions thereof, realize accurate prediction of a physical field through a staged training strategy, and remarkably improve convenience, accuracy and engineering practical value of complex structure analysis. In order to solve the technical problems, the invention adopts the following technical scheme: In a first aspect, the invention provides a PINN-based gear point cloud statics stress simulation method, which comprises the following steps: acquiring and importing a target gear geometric model, carrying out sparse grid rapid finite element analysis on the target gear geometric model, and deriving global grid and boundary information; Establishing a point cloud model based on the global grid and the boundary information, and determining boundary conditions; and constructing PINN a physical information neural network based on the point cloud model and the boundary condition, constructing a loss function, performing multiple stage network training to obtain a PINN prediction result, and outputting the prediction result as a simulation result. According to the method, the point cloud data of boundary perception is introduced, and the standard geometric model is directly utilized, so that dependence on the analysis function to describe complex boundaries is eliminated, and the preprocessing threshold and the application cost are reduced. More importantly, the invention mixes the loss function and the staged training strategy, and the physical constraint is deeply embedded into the learning process, so that the solving result is ensured to strictly meet the physical law/physical formula and boundary condition, and the physical rationality and numerical stability of the knowledge are obviously improved. Finally, the method realizes high-efficiency and high-reliability simulation of complex structures such as a gear system and the like while maintaining high precision, and provides a reliable physical engine for digital twinning and intelligent design. Based on the first aspect, further, the boundary condition includes grid information, displacement, stress information of the bo