Discrete element modularized large-scale rapid modeling method and system

Abstract

The invention discloses a discrete element modular large-scale rapid modeling method and system, and the method comprises the steps: building conventional small-scale discrete element unit modules in different shapes in advance, and building a neural network model for the small-scale discrete element unit modules; numerical simulation results of small-scale discrete element unit modules under different boundary conditions of the multi-physics field are obtained, a data set is established, a neural network model is pre-trained through the data set, the input of the neural network model is the boundary conditions, and the output of the neural network model is the numerical simulation results; and building a large-scale discrete element model through the plurality of small-scale discrete element unit modules, solving by using a pre-trained neural network model to replace iterative calculation of the small-scale discrete element unit modules, and obtaining a final numerical simulation result of the large-scale discrete element model through solving. According to the method, the problem that the current large-scale discrete element simulation needs to consume a large amount of computing resources is solved, the cost of the computing resources is preposed, and a user can build a large-scale discrete element model only by utilizing a trained module to perform simple local computing.

Description

Technical Field

[0001] The present invention relates to the technical field of discrete element modeling, and in particular to a discrete element modular large-scale rapid modeling method and system. Background Art

[0002] Discrete element modeling is a numerical method used to simulate the behavior of solid particle systems. It predicts the macroscopic properties and dynamic behavior of particle systems by treating solid particles as independent individuals and modeling the interactions between them. It has a wide range of applications in many fields. Thanks to the high-speed parallel calculation of large matrices using graphics processing units (GPUs), today's discrete element software can model, calculate, and post-process large-scale models with up to millions of particles. However, large discrete element models still require long calculation times and professional-grade GPU computing resources, as well as low accuracy. Summary of the invention

[0003] The present application provides a discrete element modular large-scale rapid modeling method and system to solve the problems in the prior art that large-scale discrete element modeling requires a long computing time and computing resources of professional-grade GPUs, as well as low accuracy.

[0004] According to the first aspect, an embodiment provides a discrete element modular large-scale rapid modeling method, the method comprising:

[0005] Pre-constructing conventional small-scale discrete element unit modules of different shapes, and constructing a neural network model for the small-scale discrete element unit modules, obtaining numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multi-physical fields and establishing a data set, and using the data set to pre-train the neural network model, the input of the neural network model is the boundary condition, and the output is the numerical simulation result;

[0006] A large-scale discrete element model is constructed by using a plurality of the small-scale discrete element unit modules, and a pre-trained neural network model is used to solve the iterative calculation of the small-scale discrete element unit modules, so as to obtain the final numerical simulation result of the large-scale discrete element model through the solution.

[0007] Furthermore, a neural network model is constructed for the discrete element unit module, specifically including:

[0008] The neural network model adopts a large model based on the Transformer algorithm.

[0009] Furthermore, the numerical simulation results of the discrete element unit module under different boundary conditions of multi-physics fields are obtained, including:

[0010] A server cluster is used to perform numerical simulation of a large-scale discrete element unit module under different boundary conditions in multiple physical fields to obtain numerical simulation results. The physical fields include stress field, strain field, seepage field, temperature field and electromagnetic field.

[0011] Furthermore, the numerical simulation results of the discrete element unit module under different boundary conditions of multiple physical fields are obtained, which specifically includes:

[0012] Through a series of experiments, an experimental database is established to serve as calibration data for discrete element simulation.

[0013] Furthermore, the neural network model is pre-trained using the data set, specifically including:

[0014] The data set is divided into a training set, a test set and a validation set, and the neural network model is trained, tested and validated.

[0015] Furthermore, a large-scale discrete element model is constructed by using a plurality of the small-scale discrete element unit modules, specifically including:

[0016] A local refined discrete element model is constructed separately for key parts, and a large-scale discrete element model is constructed by combining multiple small-scale discrete element unit modules with regular shapes and local refined discrete element models with special shapes.

[0017] Furthermore, the final numerical simulation results of the large-scale discrete element model are obtained by solving, including:

[0018] According to the continuity conditions between discrete element unit modules, the output results of the neural network model are transferred between discrete element unit modules, and the final numerical simulation results are obtained. The continuity conditions include geometric continuity conditions, boundary condition continuity conditions and physical field continuity conditions.

[0019] According to the second aspect, an embodiment provides a discrete element modular large-scale rapid modeling system, the system comprising:

[0020] A unit module construction and training module is used to pre-construct conventional small-scale discrete element unit modules of different shapes, and to construct a neural network model for the small-scale discrete element unit modules, to obtain numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multiple physical fields and to establish a data set, and to pre-train the neural network model using the data set, wherein the input of the neural network model is the boundary conditions, and the output is the numerical simulation results;

[0021] The large-scale discrete element model construction module is used to build a large-scale discrete element model through multiple small-scale discrete element unit modules, and use a pre-trained neural network model to solve instead of the iterative calculation of the small-scale discrete element unit modules, so as to obtain the final numerical simulation result of the large-scale discrete element model through the solution.

[0022] According to a third aspect, an embodiment provides an electronic device, characterized in that the device includes: a processor and a memory;

[0023] The memory is used to store one or more program instructions;

[0024] The processor is used to run one or more program instructions to execute the steps of a discrete element modular large-scale rapid modeling method as described in any of the above items.

[0025] According to the fourth aspect, an embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a discrete element modular large-scale rapid modeling method as described in any one of the above items are implemented.

[0026] The present application provides a discrete element modular large-scale rapid modeling method and system, pre-constructs conventional small-scale discrete element unit modules of different shapes, and constructs a neural network model for the small-scale discrete element unit modules, obtains the numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multiple physical fields and establishes a data set, and uses the data set to pre-train the neural network model, the input of the neural network model is the boundary condition, and the output is the numerical simulation result; a large-scale discrete element model is built through a plurality of the small-scale discrete element unit modules, and the pre-trained neural network model is used to solve the iterative calculation of the small-scale discrete element unit module, and the final numerical simulation result of the large-scale discrete element model is obtained by solving. The present invention adopts a "divide and conquer" strategy to split the large-scale engineering simulation discrete element model of millions of units, and solves the split discrete element unit modules through a neural network model, which can greatly improve the computational efficiency, so that the discrete element method can be better applied in engineering scale. The present invention solves the pain point that large-scale discrete element simulations today require a large amount of computing resources. It puts the cost of computing resources in front of the user, and the user only needs to use the trained modules to perform simple local calculations to build a large-scale discrete element model. Based on the above advantages, the user's calculation time is greatly shortened, further expanding the application of discrete element technology on an engineering scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flowchart of a discrete element modular large-scale rapid modeling method provided by one embodiment of the present invention;

[0028] Figure 2 A specific implementation flow chart of a discrete element modular large-scale rapid modeling method provided by one embodiment of the present invention;

[0029] Figure 3 A schematic diagram of neural network model training in a discrete element modular large-scale rapid modeling method provided by one embodiment of the present invention;

[0030] Figure 4 A numerical simulation result of parameter calibration in a discrete element modular large-scale rapid modeling method provided by an embodiment of the present invention;

[0031] Figure 5 An application example of a discrete element modular large-scale rapid modeling method provided by an embodiment of the present invention;

[0032] Figure 6 A schematic diagram of the logical structure of a discrete element modular large-scale rapid modeling system provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0034] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0035] The first embodiment of the present invention provides a discrete element modular large-scale rapid modeling method, which is combined with Figure 1 and Figure 2 Provide detailed explanation.

[0036] like Figure 1As shown, in step S100, small-scale discrete element unit modules of conventional different shapes are pre-constructed, and a neural network model is constructed for the small-scale discrete element unit modules, the numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multiple physical fields are obtained and a data set is established, and the neural network model is pre-trained using the data set, and the input of the neural network model is the boundary conditions, and the output is the numerical simulation results.

[0037] Specifically, in this embodiment, for the interior of large blocks in the model, modules of specific conventional shapes (such as cuboids and ellipsoids) pre-trained by neural network algorithms are used to replace millions of discrete element units. These modules can be regarded as neural network models trained by a training set composed of a series of discrete element simulations calibrated by real experiments, with the input being the boundary conditions and the output being the changes in the physical field inside the module.

[0038] In this embodiment, a server cluster is first used to perform numerical simulation of large-scale discrete element unit modules under different boundary conditions of multiple physical fields. The physical fields that need to be calculated and recorded include but are not limited to stress fields, strain fields, seepage fields, temperature fields, electromagnetic fields, etc. The data set composed of the above numerical simulation results is split into a training set, a validation set, and a test set, and a suitable neural network model is selected for training. Based on the complexity of the physical field and the magnitude of the input and output data, this embodiment intends to use a large model based on the Transformer algorithm for fitting. By using the transform module that comes with the torchvision toolkit in the python resource library to train and calculate the simulated data, the hyperparameter setting is continuously tested according to the unit module size, the number of unit body particles, and the number of physical field couplings to obtain the optimal hyperparameter value, and finally a transform neural network that can directly solve the numerical solution by specifying the boundary conditions without discrete element iterative calculations is obtained, such as Figure 3 shown.

[0039] In this embodiment, a test database is also established through a series of experiments to be used as calibration data for discrete element simulation. According to the use requirements of the object to be modeled, a multi-level numerical test of multiple physical field factors is established, and then the significance level of the physical field influencing factors is analyzed. Based on the above analysis, the test parameter table is set and the discrete unit model is solved. The specific example is as follows:

[0040] Firstly, conventional triaxial test data of a series of common materials (including artificial materials and geotechnical materials) under different states (mainly water content state, other states need to be considered for specific special materials) are collected as calibration data for discrete element simulation.

[0041] For example, for the discrete element linear elastic contact model, the particle connection parameters are the normal stiffness (K n), tangential stiffness (K s ), fracture displacement (X b ), initial shear resistance (F 0 ) and friction coefficient (μ). The above five microscopic material parameters can be obtained from the material macroscopic parameters through parameter training. The macroscopic parameters of the material include Young's modulus (E), Poisson's ratio (v), compressive strength (C u ), tensile strength (T u ) and the internal friction coefficient (u i ). The process of material calibration is as follows: First, the approximate relationship between the macroscopic parameters and the microscopic parameters of the particles is obtained according to the conversion formula. After that, the parameters are calibrated and calibrated through triaxial tests. The specific process is: ① Obtain the initial confining parameters according to the macro-micro conversion formula; ② Perform discrete element triaxial numerical calculations, and compare the Young's modulus, compressive strength, and tensile strength obtained by numerical simulation with the macroscopic parameters based on the calculation results. If the comparison result is greater than the set error range, the material microparameters are updated; ③ Repeat the above process until the difference between the material numerical simulation results and the macroscopic parameter set values ​​meets the requirements, and the calibration process ends. In addition, for urgently needed data, experiments need to be carried out as a supplement; the material parameter calibration test was carried out using loess mechanical parameters under 100Kpa confining pressure and different water contents. The numerical simulation results are shown in Figure 4 .

[0042] like Figure 1 As shown, in step S200, a large-scale discrete element model is constructed by using a plurality of the small-scale discrete element unit modules, and a pre-trained neural network model is used to solve the iterative calculation of the small-scale discrete element unit modules, and the final numerical simulation result of the large-scale discrete element model is obtained by solving the solution.

[0043] Specifically, in this embodiment, a local refined discrete element model is constructed for key parts separately, and a large-scale discrete element model is constructed by using a plurality of small-scale discrete element unit modules with conventional shapes and a local refined discrete element model with special shapes. Figure 5 As shown in the figure, by constructing large and small modules and directly performing discrete element modeling of key parts, the amount of calculation and thus the calculation time can be greatly reduced.

[0044] After obtaining a series of discrete element modules of different sizes and shapes, the geometric continuity conditions, boundary condition continuity conditions and physical field continuity conditions should be satisfied between the discrete element modules. According to the continuity conditions between the discrete element modules, the output results of the neural network model are transferred between the discrete element modules. The boundary conditions-discrete element module, discrete element module-discrete element module, boundary conditions-overall model and physical field equations together constitute the equation group of the model. Finally, the finite difference method is used to obtain the numerical solution of the above equation group.

[0045] The discrete element modular large-scale rapid modeling method of this embodiment refers to the idea of ​​"appropriate details" in the mesh division of the element modeling. Aiming at the weakness that large discrete element models need to consume more computing resources and time, the "divide and conquer" strategy is adopted to split the large-scale engineering simulation discrete element model of millions of units, and the discrete element unit modules after splitting are solved by neural networks, which greatly improves the computing efficiency and enables the discrete element method to be better applied in the engineering scale. Finally, before the actual packaging into application software, a series of tests and front-end software design are required to ensure the reliability and ease of use of the final software.

[0046] Corresponding to the above-disclosed discrete element modular large-scale rapid modeling method, the embodiment of the present invention also discloses a discrete element modular large-scale rapid modeling system, such as Figure 6 As shown, it specifically includes:

[0047] A unit module construction and training module is used to pre-construct conventional small-scale discrete element unit modules of different shapes, and to construct a neural network model for the small-scale discrete element unit modules, to obtain numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multiple physical fields and to establish a data set, and to pre-train the neural network model using the data set, wherein the input of the neural network model is the boundary conditions, and the output is the numerical simulation results;

[0048] The large-scale discrete element model construction module is used to build a large-scale discrete element model through multiple small-scale discrete element unit modules, and use a pre-trained neural network model to solve instead of the iterative calculation of the small-scale discrete element unit modules, so as to obtain the final numerical simulation result of the large-scale discrete element model through the solution.

[0049] It should be noted that for the detailed description of a discrete element modular large-scale rapid modeling system provided in an embodiment of the present invention, reference can be made to the relevant description of a discrete element modular large-scale rapid modeling method provided in an embodiment of the present application, which will not be repeated here.

[0050] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.

Claims

1. A discrete element modular large-scale rapid modeling method, characterized in that: The method comprises: Pre-constructing conventional small-scale discrete element unit modules of different shapes, and constructing a neural network model for the small-scale discrete element unit modules, obtaining numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multi-physical fields and establishing a data set, and using the data set to pre-train the neural network model, the input of the neural network model is the boundary condition, and the output is the numerical simulation result; A large-scale discrete element model is constructed by using a plurality of the small-scale discrete element unit modules, and a pre-trained neural network model is used to solve the iterative calculation of the small-scale discrete element unit modules, so as to obtain the final numerical simulation result of the large-scale discrete element model through the solution.

2. A discrete element modular large-scale rapid modeling method as claimed in claim 1, characterized in that: Constructing a neural network model for the discrete element unit module specifically includes: The neural network model adopts a large model based on the Transformer algorithm.

3. A discrete element modular large-scale rapid modeling method as claimed in claim 1, characterized in that: Obtain numerical simulation results of discrete element modules under different boundary conditions of multi-physics fields, including: A server cluster is used to perform numerical simulation of a large-scale discrete element unit module under different boundary conditions in multiple physical fields to obtain numerical simulation results. The physical fields include stress field, strain field, seepage field, temperature field and electromagnetic field.

4. A discrete element modular large-scale rapid modeling method as claimed in claim 1, characterized in that: Obtain numerical simulation results of discrete element modules under different boundary conditions of multi-physics fields, including: Through a series of experiments, an experimental database is established to serve as calibration data for discrete element simulation.

5. The discrete element modular large-scale rapid modeling method according to claim 1, characterized in that: Pre-training the neural network model using the data set specifically includes: The data set is divided into a training set, a test set and a validation set, and the neural network model is trained, tested and validated.

6. A discrete element modular large-scale rapid modeling method as claimed in claim 1, characterized in that: Building a large-scale discrete element model by using a plurality of the small-scale discrete element unit modules specifically includes: A local refined discrete element model is constructed separately for key parts, and a large-scale discrete element model is constructed by combining multiple small-scale discrete element unit modules with regular shapes and local refined discrete element models with special shapes.

7. A discrete element modular large-scale rapid modeling method as claimed in claim 1, characterized in that: The final numerical simulation results of the large-scale discrete element model are obtained by solving, including: According to the continuity conditions between discrete element unit modules, the output results of the neural network model are transferred between discrete element unit modules, and the final numerical simulation results are obtained. The continuity conditions include geometric continuity conditions, boundary condition continuity conditions and physical field continuity conditions.

8. A discrete element modular large-scale rapid modeling system, characterized in that: The system comprises: A unit module construction and training module is used to pre-construct conventional small-scale discrete element unit modules of different shapes, and to construct a neural network model for the small-scale discrete element unit modules, to obtain numerical simulation results of the small-scale discrete element unit modules under different boundary conditions of multiple physical fields and to establish a data set, and to pre-train the neural network model using the data set, wherein the input of the neural network model is the boundary conditions, and the output is the numerical simulation results; The large-scale discrete element model construction module is used to build a large-scale discrete element model through multiple small-scale discrete element unit modules, and use a pre-trained neural network model to solve instead of the iterative calculation of the small-scale discrete element unit modules, so as to obtain the final numerical simulation result of the large-scale discrete element model through the solution.

9. An electronic device, characterized in that: The device comprises: a processor and a memory; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the steps of a discrete element modular large-scale rapid modeling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a discrete element modular large-scale rapid modeling method as described in any one of claims 1 to 7 are implemented.

Images