Gear point cloud statics stress simulation method and system based on PINN

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 following steps: acquiring and importing a target gear geometric model, performing sparse mesh fast finite element analysis on the target gear geometric model, and exporting global mesh and boundary information; establishing a point cloud model based on the global grid and the boundary information, and determining a boundary condition; and constructing a PINN physical information neural network based on the point cloud model and the boundary condition, constructing a loss function to carry out multiple times of staged network training so as to obtain a prediction result of the PINN, and outputting the prediction result as a simulation result. According to the method, the point cloud data can be directly utilized, accurate prediction of the physical field is realized through a staged training strategy, and the convenience, accuracy and engineering practical value of complex structure analysis are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of solid mechanics simulation and analysis technology, and more specifically, to a method and system for simulating the static forces of gear point clouds based on PINN. Background Technology

[0002] Physical Information Neural Networks (PINNs), as a cutting-edge method integrating scientific computing and deep learning, are based on the core idea of ​​embedding partial differential equations controlling physical processes into the training objectives of neural networks, thereby using physical laws to guide a data-driven learning process. The effectiveness of this method is of great significance for achieving high-precision and high-efficiency simulation analysis of complex engineering structures.

[0003] Currently, standard PINN methods based on mainstream frameworks (such as DeepXDE) have formed a relatively fixed process paradigm. To further improve performance, the research community has developed a variety of improvement strategies, such as introducing an adaptive weight balancing mechanism to coordinate the magnitude differences of different loss terms, using conservative PINN (cPINN) to handle multi-scale or discontinuous problems through domain decomposition, and developing variational PINN (VPINN) based on the weighted residual method to improve numerical stability.

[0004] However, existing PINN technology exhibits significant limitations in solid mechanics analysis for real-world engineering scenarios. Current applications are mostly limited to linear elastic problems involving two-dimensional or simple three-dimensional regular geometries. When dealing with real-world structures containing complex geometries, existing technologies face severe challenges in sampling configuration points on complex surfaces and applying boundary conditions, leading to a significant decrease in solution accuracy. Furthermore, existing technologies heavily rely on precise mathematical definitions of boundary conditions; all boundaries of the entire computational domain, such as displacement constraint surfaces and external force application surfaces, must be given with clear mathematical expressions. For real-world engineering problems with complex boundary shapes or diverse load conditions, this strong dependence on boundary conditions greatly limits its applicability and makes it unable to effectively handle situations where the boundaries are unknown. Summary of the Invention

[0005] To overcome or at least partially solve the above problems, this invention provides a PINN-based method and system for static force simulation of gear point clouds. It eliminates the need for explicit programming to define complex geometric shapes and their boundary conditions, and can directly utilize point cloud data from CAD / CAE software. Through a phased training strategy, it achieves accurate prediction of the physical field, significantly improving the convenience, accuracy, and engineering practical value of analyzing complex structures.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a method for simulating the static forces of gear point clouds based on PINN, comprising the following steps: Obtain and import the target gear geometry model, perform sparse mesh fast finite element analysis on the target gear geometry model, and export the global mesh and boundary information; A point cloud model is established based on the global grid and boundary information, and the boundary conditions are determined. A PINN physical information neural network is constructed based on a point cloud model and boundary conditions. A loss function is constructed to perform multiple staged network trainings to obtain the prediction results of PINN, and these prediction results are output as simulation results.

[0007] This method introduces boundary-aware point cloud data and directly utilizes standard geometric models, eliminating the reliance on analytical functions to describe complex boundaries and reducing preprocessing barriers and application costs. More importantly, this invention combines loss functions with a phased training strategy, deeply embedding physical constraints into the learning process. This ensures that the solution strictly satisfies physical laws / formulas and boundary conditions, significantly improving the physical rationality and numerical stability of the solution. Ultimately, this method achieves efficient and highly reliable simulation of complex structures such as gear systems while maintaining high accuracy, providing a reliable "physical engine" for digital twins and intelligent design.

[0008] Based on the first aspect, the aforementioned boundary conditions further include various information such as boundary mesh information, displacement, and stress.

[0009] Based on the first aspect, it further includes: applying corresponding displacement constraints on the mesh of the fixed surface; and applying corresponding stress conditions on the mesh of the stressed surface.

[0010] Based on the first aspect, it further includes: calculating the residuals of the governing equations, displacement boundary residuals, and force boundary residuals over the entire domain.

[0011] Based on the first aspect, further, the above loss function is composed of three weighted parts: the residual of the governing equation, the residual of the displacement boundary, and the residual of the force boundary.

[0012] Based on the first aspect, the method for obtaining and importing the target gear geometry model described above further includes the following steps: Obtain the gear geometric model file provided by the user or use computer-aided design software to complete the parametric three-dimensional geometric modeling of the gear transmission system to obtain the target gear geometric model.

[0013] Based on the first aspect, the PINN-based gear point cloud static force simulation method further includes the following steps: Modify the initial boundary conditions and model parameters, train the network again, and output the simulation results.

[0014] Secondly, this invention provides a gear point cloud static force simulation system based on PINN, including a target import and analysis module, a point and boundary determination module, and a neural network analysis module, wherein: The target import and analysis module is used to acquire and import the target gear geometric model, perform sparse mesh fast finite element analysis on the target gear geometric model, and export the global mesh and boundary information. The point and boundary determination module is used to build a point cloud model based on the global grid and boundary information, and to determine the boundary conditions; The neural network analysis module is used to construct a PINN physical information neural network based on a point cloud model and boundary conditions, and to construct a loss function to perform multiple staged network trainings to obtain the prediction results of PINN, and output the prediction results as simulation results.

[0015] This system, through the coordinated efforts of multiple modules including target import and analysis, point and boundary determination, and neural network analysis, directly utilizes standard geometric models by incorporating boundary-aware point cloud data. This eliminates the reliance on analytical functions to describe complex boundaries, reducing preprocessing barriers and application costs. More importantly, this invention employs a hybrid loss function and a phased training strategy, deeply embedding physical constraints into the learning process. This ensures that the solution strictly satisfies physical laws / formulas and boundary conditions, significantly improving the physical rationality and numerical stability of the solution. Ultimately, this invention achieves efficient and highly reliable simulation of complex structures such as gear systems while maintaining high accuracy, providing a reliable "physical engine" for digital twins and intelligent design.

[0016] Thirdly, this application provides an electronic device including a memory for storing one or more programs; a processor; and, when the one or more programs are executed by the processor, implementing the method as described in any of the first aspects above.

[0017] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the first aspects above.

[0018] This invention has at least the following advantages or beneficial effects: 1. This invention introduces boundary-aware point cloud data and directly utilizes industry-standard geometric models from CAD / CAE, thus eliminating the reliance on analytical functions to describe complex boundaries and reducing the preprocessing threshold and application cost. 2. This invention employs a hybrid loss function and a phased training strategy to deeply embed physical constraints into the learning process, ensuring that the solution strictly satisfies physical laws / formulas and boundary conditions, and significantly improving the physical rationality and numerical stability of the solution. 3. While maintaining high precision, this invention achieves efficient and highly reliable simulation of complex structures such as gear systems, providing a reliable "physical engine" for digital twins and intelligent design. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a gear point cloud static force simulation method based on PINN, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the complete process of a gear point cloud static force simulation method based on PINN according to an embodiment of the present invention; Figure 3 A schematic diagram illustrating the training principle of PINN; Figure 4 This is a schematic diagram of the finite element method and existing PINN results; Figure 5 This is a schematic diagram of the gear point cloud and its boundary information in an embodiment of the present invention; Figure 6 This is a schematic diagram comparing the training results of PINN in this embodiment of the invention with the actual results output by existing finite element software; Figure 7 This is a schematic diagram of a gear point cloud static force simulation system based on PINN, according to an embodiment of the present invention. Figure 8 This is a structural block diagram of an electronic device provided in an embodiment of the present invention.

[0021] Explanation of reference numerals in the attached diagram: 100, Target import and analysis module; 200, Point and boundary determination module; 300, Neural network analysis module; 101, Memory; 102, Processor; 103, Communication interface. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0025] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0026] In the description of the embodiments of the present invention, "multiple" means at least two. Example

[0027] like Figures 1-2 As shown, in a first aspect, embodiments of the present invention provide a method for simulating the static forces of gear point clouds based on PINN, comprising the following steps: S1. Obtain and import the target gear geometric model, perform sparse mesh fast finite element analysis on the target gear geometric model, and export the global mesh and boundary information; Furthermore, this includes: obtaining a gear geometry model file provided by the user or using computer-aided design software to complete a parametric three-dimensional geometric model of the gear transmission system to obtain the target gear geometry model.

[0028] In some embodiments of the present invention, the geometric and physical parameters of the object of analysis are explicitly defined. Taking an isotropic linear elastic cuboid as an example, its geometric dimensions are set as length × width × height = 100mm × 100mm × 100mm. The material properties are set with reference to Q235 steel: elastic modulus E = 210 GPa, Poisson's ratio v = 0.3. The boundary conditions are set as follows: the bottom surface is fixed (i.e., the displacement of all nodes on the bottom surface is 0), the upper surface is subjected to a uniformly distributed vertical pressure load P = 100 MPa, and the front and rear sides are free surfaces. The governing equation of this problem is the linear elastic static equilibrium equation: ∇·σ + f = 0, and its constitutive relation obeys the generalized Hooke's law.

[0029] This invention supports two methods for inputting geometric data. The first is directly reading existing geometric model files provided by the user that conform to standard formats (such as STP / XT). The second is completing parametric 3D geometric modeling of the gear transmission system in computer-aided design (CAD) software (such as SolidWorks). The modeling object can cover a single rack, a single gear, or a gear pair. To obtain benchmark data for training the physical information neural network, the aforementioned CAD model file needs to be imported into finite element analysis software such as ANSYS Workbench for finite element analysis. To improve the efficiency of subsequent data generation, this method does not perform a full-size simulation of the entire gearbox, but only extracts the key meshing gear pairs and their adjacent supporting structures as analysis objects. A coarser mesh size is intentionally set on this sub-model to significantly shorten the computation time of a single finite element solution, thereby supporting rapid parametric analysis. The core purpose of this approach is to balance data generation efficiency with the accuracy of the physical field trend, rather than pursuing high precision in a single simulation. Simultaneously, key boundary conditions are set according to actual working conditions (such as applying fixed constraints to the inner wall of the gear and applying stress to the stressed tooth surface of the gear). After the finite element method (FEM) solution is completed, the structured information necessary for constructing the physical information neural network dataset is extracted. The exported data package should include: the three-dimensional coordinates of all nodes in the finite element model, representing the discrete geometry of the computational domain; the coordinates and displacement values ​​(usually 0) of all nodes with fixed constraints; and the coordinates and corresponding values ​​of all nodes with applied external loads. Taking a gear as an example, its exported point cloud structure (left) and corresponding boundary information (right) are shown below. Figure 5 As shown. The structured dataset prepared in this part provides learning samples with both engineering background and physical realism for the subsequent training of the physical information neural network.

[0030] S2. Establish a point cloud model based on the global grid and boundary information, and determine the boundary conditions; Furthermore, the aforementioned boundary conditions include various information such as boundary mesh information, displacement, and stress.

[0031] Furthermore, it also includes: applying corresponding displacement constraints on the mesh of the fixed surface; and applying corresponding stress conditions on the mesh of the stressed surface.

[0032] Furthermore, it also includes: calculating the residuals of the governing equations, displacement boundary residuals, and force boundary residuals across the entire domain.

[0033] S3. Construct a PINN physical information neural network based on the point cloud model and boundary conditions, and construct a loss function to perform multiple staged network trainings to obtain the prediction results of PINN, and output the prediction results as simulation results; the above loss function is composed of three weighted parts: control equation residuals, displacement boundary residuals and force boundary residuals.

[0034] In some embodiments of this invention, a fully connected neural network is constructed, with spatial coordinates (x, y, z) as input and the displacement field (u, v, w) to be solved as output. The network structure employs 5 hidden layers, each with 100 neurons, and the tanh function is selected as the activation function to accommodate higher-order derivative calculations. The loss function L is a weighted sum of the residuals of the governing equations, displacement boundary residuals, and force boundary residuals. The optimization process uses the Adam optimizer, with an initial learning rate set to 1e-3, and a total of 10,000 iterations.

[0035] Furthermore, it also includes: modifying the initial boundary conditions and model parameters, retraining the network, and outputting simulation results.

[0036] In some embodiments of the present invention, taking a cube as an example, the first training may involve fixing the bottom surface of the cube and subjecting it to pressure on the top surface; the second training may change the pressure on the top surface to tension, perform rapid finite element analysis, obtain a new set of data, and input it into PINN for training. The purpose of this retraining is to allow PINN to learn more force conditions, gain a more thorough understanding of the physical formulas, and output correct results under various force conditions.

[0037] In some embodiments of the present invention, after training is completed, the displacement and stress fields predicted by the network are queried at regular grid points within the computational domain. For example... Figure 6 As shown, the prediction results of PINN (right figure) are compared with the numerical solutions obtained from commercial finite element software (left figure). The results show that the displacement field prediction results are highly consistent with the finite element solutions; the displacement field deviates slightly in the concentrated region, but the overall distribution trend is correct. The displacement field comparison is as follows: Figure 4 As shown in the figure. This verifies the effectiveness and accuracy of the proposed method for solving linear elastic statics problems, laying the foundation for subsequent complex mechanical analyses.

[0038] This invention constructs a framework that deeply embeds physical laws into the neural network training process, and achieves efficient and high-precision parameterized prediction through a phased training strategy. This framework transforms the governing equations of linear elasticity into the network's training objectives using multinomial functions, ensuring that the network output is strictly constrained by physical laws.

[0039] To ensure the model outputs physically accurate displacement and stress fields under limited boundary conditions, solid mechanics formulas are incorporated as training loss terms to guide the training process. The embedding of physical information is achieved through four custom functions, which together form the basis of the training loss function: First, geometric equation embedding: This function automatically differentiates the gradient of the displacement field output by the neural network with respect to the input coordinates, obtaining the strain tensor. This process links the geometric information of the point cloud coordinates with the strain field, forming the basis for describing deformation continuity. Second, constitutive equation embedding: This function calculates the stress tensor based on the generalized Hooke's law. It introduces the physical properties of the material into the model, ensuring the physical correctness of the stress-strain response. Third, constitutive constraints: By calculating the difference between the stress derived from the displacement field through constitutive relations and the stress directly predicted by the network, it enforces that the predicted displacement and stress fields must satisfy constitutive relations, significantly improving the accuracy and physical consistency of stress prediction. Fourth, the construction of governing equations and loss functions: This function calculates the momentum balance equation of the stress field and, combined with constitutive constraints, forms the physical residual term.

[0040] During training, these residuals are computed across the point cloud across the entire computational domain. Boundary conditions are imposed by adding an additional boundary point cloud data loss term to the loss function. For example, on a fixed boundary point cloud, the loss function penalizes deviations of the network's predicted displacement from zero; on a stressed boundary point cloud, it penalizes deviations of the predicted stress from the preset load. Thus, the boundary conditions, along with the physical residual term, constitute the complete loss function using a specific point cloud dataset and its target values, guiding the network to satisfy all physical constraints.

[0041] To achieve stable and efficient training and ensure the physical plausibility of the prediction results, this invention employs a meticulously designed network architecture and training strategy. The main network adopts a fully connected architecture with four hidden layers, each containing one hundred neurons to balance model expressive power and computational cost. The training process uses the Adam optimizer to minimize the L2 norm loss function, which is a weighted sum of the residual terms of the control equations and the boundary condition data terms, thus jointly constraining the network output to simultaneously satisfy internal physical laws and external boundary conditions. This invention implements a phased intelligent training strategy: First, a first round of basic training is performed, i.e., for a specific load condition, starting with randomly initialized parameters, a sufficient number of iterative training iterations are executed to bring the network to convergence for that fixed geometry and load condition, thereby learning an accurate physical field mapping relationship; subsequently, variable fine-tuning is performed, i.e., while maintaining geometric invariance, the load magnitude is changed, and the converged model is used as initial parameters for a small number of iterations of fine-tuning training. Since the network already possesses prior knowledge of geometric deformation patterns, this stage can quickly adapt to new load levels. This "pre-training-fine-tuning" strategy enables the finally trained model to have the ability to make parametric predictions of the same geometric structure under different load conditions, achieving efficient and rapid parameter scanning analysis. It significantly avoids the huge computational overhead of retraining every time the conditions are changed in traditional methods, providing a powerful and efficient solution for the rapid design and performance analysis of gear systems.

[0042] Physical Information Neural Networks (PINNs) are an important method in scientific machine learning, aiming to incorporate physical laws into deep learning. Their mainstream implementation frameworks, such as DeepXDE built on TensorFlow or PyTorch, have a standard workflow as follows: Figure 3 As shown, the main steps are as follows. First, the computational domain and the partial differential equations (PDEs) describing the physical process need to be defined. Then, a large number of placement points are randomly collected within the computational domain, at its boundaries, and under initial conditions. The core step is to construct a fully connected feedforward neural network, whose input is the spatiotemporal coordinates (x, t), and whose output is the physical field u(x, t) to be solved. The training objective of this network is to minimize a composite loss function, which is typically composed of the weighted sum of squares of the PDE residuals, boundary conditions, and initial conditions. The optimization process uses algorithms such as Adam or L-BFGS, adjusting the network weights through backpropagation to force the network to approximate a solution that both fits the sparse data and satisfies the physical constraints.

[0043] This method introduces boundary-aware point cloud data and directly utilizes industry-standard geometric models from CAD / CAE, eliminating the reliance on analytical functions to describe complex boundaries and reducing preprocessing barriers and application costs. More importantly, this invention combines loss functions with a phased training strategy, deeply embedding physical constraints into the learning process. This ensures that the solution strictly satisfies physical laws / formulas and boundary conditions, significantly improving the physical rationality and numerical stability of the solution. Ultimately, this method achieves efficient and highly reliable simulation of complex structures such as gear systems while maintaining high accuracy, providing a reliable "physical engine" for digital twins and intelligent design.

[0044] like Figure 7 As shown, in a second aspect, embodiments of the present invention provide a gear point cloud static force simulation system based on PINN, including a target import and analysis module 100, a point and boundary determination module 200, and a neural network analysis module 300, wherein: The target import and analysis module 100 is used to acquire and import the target gear geometric model, perform sparse mesh fast finite element analysis on the target gear geometric model, and export the global mesh and boundary information. The point and boundary determination module 200 is used to build a point cloud model based on the global grid and boundary information, and to determine the boundary conditions; The neural network analysis module 300 is used to construct a PINN physical information neural network based on a point cloud model and boundary conditions, and to construct a loss function to perform multiple staged network trainings to obtain the prediction results of PINN, and output the prediction results as simulation results.

[0045] This system, through the collaboration of over 300 modules including a target import and analysis module 100, a point and boundary determination module 200, and a neural network analysis module, directly utilizes standard geometric models by introducing boundary-aware point cloud data. This eliminates the reliance on analytical functions to describe complex boundaries, reducing preprocessing barriers and application costs. More importantly, this invention employs a hybrid loss function and a phased training strategy, deeply embedding physical constraints into the learning process. This ensures that the solution strictly satisfies physical laws / formulas and boundary conditions, significantly improving the physical rationality and numerical stability of the solution. Ultimately, this invention achieves efficient and highly reliable simulation of complex structures such as gear systems while maintaining high accuracy, providing a reliable "physical engine" for digital twins and intelligent design.

[0046] like Figure 8 As shown, in a third aspect, embodiments of this application provide an electronic device including a memory 101 for storing one or more programs; and a processor 102. When the one or more programs are executed by the processor 102, they implement the methods described in any of the first aspects above.

[0047] The system also includes a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected directly or indirectly to each other to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used for signaling or data communication with other node devices.

[0048] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0049] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0050] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0051] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0052] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by processor 102, the computer program implements the methods described in any of the first aspects above. If the functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0054] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for simulating the static forces of gear point clouds based on PINN, characterized in that, Includes the following steps: Obtain and import the target gear geometry model, perform sparse mesh fast finite element analysis on the target gear geometry model, and export the global mesh and boundary information; A point cloud model is established based on the global grid and boundary information, and the boundary conditions are determined. A PINN physical information neural network is constructed based on a point cloud model and boundary conditions. A loss function is constructed to perform multiple staged network trainings to obtain the prediction results of PINN, and these prediction results are output as simulation results.

2. The method for static force simulation of gear point clouds based on PINN according to claim 1, characterized in that, The boundary conditions include various information such as boundary grid information, displacement, and stress.

3. The method for static force simulation of gear point clouds based on PINN according to claim 2, characterized in that, Also includes: Apply corresponding displacement constraints to the mesh on the fixed surface; Apply the corresponding stress conditions to the grid on the stressed surface.

4. The method for static force simulation of gear point clouds based on PINN according to claim 3, characterized in that, Also includes: Calculate the residuals of the governing equations, displacement boundary residuals, and force boundary residuals over the entire domain.

5. The method for static force simulation of gear point clouds based on PINN according to claim 4, characterized in that, The loss function is composed of a weighted sum of the residuals of the governing equations, the residuals of the displacement boundary, and the residuals of the force boundary.

6. The method for static force simulation of gear point cloud based on PINN according to claim 1, characterized in that, The method for obtaining and importing the target gear geometry model includes the following steps: Obtain the gear geometric model file provided by the user or use computer-aided design software to complete the parametric three-dimensional geometric modeling of the gear transmission system to obtain the target gear geometric model.

7. The method for static force simulation of gear point cloud based on PINN according to claim 1, characterized in that, It also includes the following steps: Modify the initial boundary conditions and model parameters, train the network again, and output the simulation results.

8. A static force simulation system for gear point clouds based on PINN, characterized in that, It includes a target import and analysis module, a point and boundary determination module, and a neural network analysis module, among which: The target import and analysis module is used to acquire and import the target gear geometric model, perform sparse mesh fast finite element analysis on the target gear geometric model, and export the global mesh and boundary information. The point and boundary determination module is used to build a point cloud model based on the global grid and boundary information, and to determine the boundary conditions; The neural network analysis module is used to construct the PINN physical information neural network based on the point cloud model and boundary conditions, and to construct a loss function to perform multiple staged network training to obtain the prediction results of PINN. The verification output module is used to determine whether the prediction result has reached the preset accuracy. If so, the prediction result is output as the simulation result.

9. An electronic device, characterized in that, include: Memory, used to store one or more programs; processor; When the one or more programs are executed by the processor, the method as described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method as described in any one of claims 1-7.

Images