Near-field dynamics matrix-free finite element method and system based on partial assembly strategy
By decomposing the finite element stiffness matrix into a continuous product of multiple operators and solving iteratively, the problems of low computational efficiency and difficulty in solving the assembly matrix in traditional methods for complex geometric problems are solved. This enables efficient and low-storage near-field dynamics model simulation, which is suitable for complex materials and large-scale simulations.
Patent Information
- Application Number
- CN202511155411.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-18
- Publication Date
- 2025-11-18
AI Technical Summary
Existing technologies are computationally inefficient when dealing with complex geometric problems, making it difficult to meet the requirements of practical engineering. Furthermore, traditional finite element methods involve a huge amount of computation to assemble the stiffness matrix and have poor sparsity in the solution. Existing matrix-free methods have slow convergence speed and high computational cost, making it difficult to achieve efficient parallel computing.
A near-field dynamics matrix-free finite element method based on a partial assembly strategy is adopted. By decomposing the finite element stiffness matrix into a continuous product of multiple operators, the element stiffness matrix is implicitly represented, and the stiffness matrix is calculated and restored in real time. It is then solved iteratively, avoiding explicit assembly of the global stiffness matrix. High-performance parallel computing is achieved by utilizing the computing power of SIMD devices such as GPUs.
It achieves efficient simulation of arbitrary nonlocal mechanical problems, reduces storage requirements, supports fatigue fracture simulation of materials of arbitrary scale, improves computational efficiency and scalability, and overcomes the communication bottlenecks and memory bandwidth limitations of traditional methods.
Smart Images

Figure CN120974837A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the field of near-field dynamics, and particularly relates to a near-field dynamics matrix-free finite element method and system based on a partial assembly strategy. BACKGROUND
[0002] Near-field dynamics is a mechanical model based on non-local action principle, which can effectively capture long-range interaction and multi-scale effect in the system, and has unique advantages in the simulation of complex mechanical behaviors such as material fracture and damage. This technology has key application value in the fields of industrial detection, material science, explosion dynamics, geotechnical engineering, etc. For example, the fatigue damage prediction of aero-engine blades (GB / T 30598-2014) can be early warned through numerical simulation; the delamination failure prediction of composite materials (such as carbon fiber reinforced polymers, CFRP); the simulation of instantaneous pressure pulse processes such as structure fragmentation, shock wave propagation, and explosive crack propagation under explosive load; the progressive failure and landslide prediction of non-continuous deformation problems such as rock mass fracture, joint slip, and slope instability.
[0003] The prior art mainly includes:
[0004] 1. Meshless method: The meshless method has natural advantages in handling large deformation and moving boundary problems, and the particles can be freely moved and redistributed, the calculation is relatively simple, and in some cases, there is no need to solve large linear equations. The convergence speed of this method is slow, more particles are needed to achieve similar accuracy to the finite element method, and the calculation cost will increase sharply with the increase of particles. The theoretical basis is relatively weak, and there is no perfect theoretical framework, and the reliability and predictability are low.
[0005] 2. Finite difference method: The finite difference method is efficient, simple and direct in handling regular geometric shapes and simple boundary conditions, and is easy to implement. This method is difficult to handle problems with complex geometric shapes, and is difficult to meet the requirements of actual engineering.
[0006] 3. Finite element method: The finite element method can accurately handle complex geometric problem domains through flexible mesh division, support adaptive mesh refinement, have strict mathematical theory support, and can approximate complex solutions by selecting appropriate basis functions (such as DG elements), and have higher accuracy and stability. The calculation of the stiffness matrix of this method is very large, and the solution of algebraic equations lacks sparsity and is difficult to solve. SUMMARY
[0007] In view of the deficiencies of the above prior art, the present application provides a near-field dynamics matrix-free finite element method based on a partial assembly strategy, which replaces the traditional assembly matrix-based strategy with a matrix-free method to realize the finite element numerical simulation of the near-field dynamics model.
[0008] According to an aspect of the present specification, a near-field dynamic non-matrix finite element method based on a partial assembly strategy is provided, comprising:
[0009] Based on the decomposition of the near-field dynamic operator, the finite element stiffness matrix is decomposed into a continuous product of multiple operators;
[0010] The element stiffness matrix is implicitly expressed as a function of several parameters, and the restoration is calculated in real time when needed, realizing the partial assembly of the element stiffness matrix;
[0011] The partially assembled element stiffness matrix is iteratively solved, and in the iterative solving process, the element stiffness matrix and the element vector are directly calculated, and the calculation results are accumulated back to the global degree of freedom.
[0012] As a further technical solution, the finite element stiffness matrix is decomposed into a continuous product of multiple operators:
[0013]
[0014] wherein, is a distributed parallel part, is a shared memory parallel part, is a base function part, is an integral quadrature part.
[0015] As a further technical solution, the operator is a matrix corresponding to the th sub-grid , responsible for mapping the global degree of freedom to the local degree of freedom, and the elements are defined as follows:
[0016]
[0017] The transpose operator is responsible for mapping the local degree of freedom back to the global degree of freedom.
[0018] As a further technical solution, the operator is responsible for mapping the local degree of freedom to the element degree of freedom, and the sub-grid contains elements, and the number of degrees of freedom of each element is ), the operator is a matrix, and the elements are defined as follows:
[0019]
[0020] wherein for a subgrid total number of degrees of freedom.
[0021] As a further technical solution, the method further comprises:
[0022] defining a mapping representing a mapping from element degrees of freedom to global degrees of freedom, where the index set represents a global degree of freedom index set on the mesh , is the total number of degrees of freedom, represents an index set of element degrees of freedom, where the mapping is isomorphic to the operator .
[0023] As a further technical solution, the element stiffness matrix is implicitly represented as a function of several parameters, and the restoration is calculated in real time when needed, realizing partial assembly of the element stiffness matrix, including:
[0024] mapping the element degrees of freedom on the element to the global degrees of freedom;
[0025] Given an element and integration points , define the shape function as where is the index of the basis function at the global degree of freedom, represents the global degree of freedom of the vertex of the element whose element degree of freedom is .
[0026] Given an element pair , define its connection matrix to describe the topological relationship of the two elements at the element degree of freedom, and project the value of the shape function on the element to the element .
[0027] Given an element pair , obtain the tensor form of the element stiffness matrix according to the integration point set thereof.
[0028] As a further technical solution, the element stiffness matrix after partial assembly is iteratively solved, including:
[0029] Calculate the element vector , project the global vector to the element degree of freedom;
[0030] Calculate the element stiffness matrix-vector product ;
[0031] Calculate The calculation result is accumulated back to the global degree of freedom.
[0032] According to an aspect of the present application, a partial assembly strategy based near-field dynamic matrix-free finite element system is provided, comprising:
[0033] A first main module is configured to decompose the finite element stiffness matrix into a continuous product of multiple operators based on near-field dynamic operator decomposition;
[0034] A second main module is configured to implicitly represent the element stiffness matrix as a function of several parameters, and calculate the restoration in real time when needed, so as to realize partial assembly of the element stiffness matrix;
[0035] A third main module is configured to iteratively solve the element stiffness matrix after partial assembly, and directly calculate the element stiffness matrix and the element vector during the iterative solving process, and accumulate the calculation result back to the global degree of freedom.
[0036] According to an aspect of the present application, a computing device is provided, comprising a memory and a processor, the memory stores program instructions executed by the processor, and the processor invokes the program instructions to perform the partial assembly strategy based near-field dynamic matrix-free finite element method.
[0037] According to an aspect of the present application, a non-transitory computer readable storage medium is provided, which stores computer instructions, and the computer instructions make the computer execute the partial assembly strategy based near-field dynamic matrix-free finite element method.
[0038] The present application realizes finite element numerical simulation of the near-field dynamic model through operator decomposition of the finite element stiffness matrix of the near-field dynamics, partial assembly strategy of the linear system and matrix-free solving of the algebraic system, and has the following advantages compared with the prior art:
[0039] 1. Universality: suitable for any mechanical problem based on non-local action principle, not limited to specific constitutive equation;
[0040] 2. High performance: can fully exert the computing capacity of the SIMD device such as GPU;
[0041] 3. Low storage requirement: low requirement for the video memory bandwidth and video memory capacity;
[0042] 4. High scalability: supports any scale multi-card GPU cluster, and can process any scale material fatigue fracture simulation. BRIEF DESCRIPTION OF DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings used in the embodiments or prior art description. Obviously, the drawings described below are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained based on these drawings without creative labor.
[0044] Figure 1 A flowchart of a partial assembly strategy-based near-field dynamic no-matrix finite element method provided by the embodiments of the present application is shown.
[0045] Figure 2 A near-field dynamic operator decomposition data flow graph provided by the embodiments of the present application is shown.
[0046] Figure 3 A comparison diagram of a traditional finite element method and a no-matrix finite element method provided by the embodiments of the present application is shown.
[0047] Figure 4 A matrix-vector product algorithm diagram based on the EbE strategy provided by the embodiments of the present application is shown. DETAILED DESCRIPTION
[0048] The terms "comprising" and "having" and any variations thereof in the specification and claims of the present application and the above drawings are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to the steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.
[0049] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application. In addition, the technical features in each embodiment or single embodiment provided by the present application can be combined with each other to form new technical solutions, and the combination is not restricted by the order of steps and / or structure composition mode, but should be based on the implementation by those skilled in the art. When the combination of technical solutions contradicts each other or cannot be implemented, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope of the present application.
[0050] The embodiments of the present application provide a partial assembly strategy-based near-field dynamic no-matrix finite element method, as shown in Figure 1As shown, firstly, based on the near-field dynamics operator decomposition, the finite element stiffness matrix is decomposed into a continuous product of multiple operators; then, the element stiffness matrix is implicitly represented as a function of several parameters, and the calculation is performed in real time when needed to partially assemble the element stiffness matrix; then, the partially assembled element stiffness matrix is iteratively solved. During the iterative solution process, the element stiffness matrix and element vectors are directly calculated, and the calculation results are accumulated back to the global degrees of freedom.
[0051] This invention comprises three main stages: finite element stiffness matrix operator decomposition of near-field dynamics, partial assembly strategy for linear systems, and matrix-free solution for algebraic systems. It solves the problem of low efficiency of traditional finite element methods based on assembly matrices, overcomes the communication limitations between cluster nodes in large-scale parallelism, breaks through the bandwidth bottleneck of existing matrix-free methods on GPUs, and improves the computational efficiency and parallelism of the algorithm on ultra-large-scale GPU clusters.
[0052] In the near-field dynamics operator decomposition, let Indicates the area Global mesh partitioning, Depend on Composed of subgrids, the first Each subgrid is denoted as Global grid The degrees of freedom defined on the surface are called global degrees of freedom, such as the values of the finite element solution at global nodes, while the degrees of freedom on the submesh are called local degrees of freedom. In addition, the degrees of freedom on the element (such as triangles, tetrahedrons, etc.) are called element degrees of freedom, which usually depend on the geometry of the element and the basis functions.
[0053] like Figure 2 As shown, the finite element stiffness matrix The assembly and solution can be decomposed into a series of successive products of operators, whose actions are cascaded and correspond to the distributed parallel part. Shared memory parallel part Basis function part and the integral quadrature part :
[0054]
[0055] Operator It corresponds to the first Subgrid one A matrix, whose elements are defined as follows:
[0056]
[0057] For the Subgrid Let the global degrees of freedom be numbered as follows: The local degrees of freedom are numbered as , For subgrid The total number of degrees of freedom. Operator It is The matrix, which represents the operators of all subgrids. Combining them yields global operators. :
[0058]
[0059] Its transpose calculation It is responsible for mapping local degrees of freedom back to global degrees of freedom. In a distributed computing environment, and The implementation typically relies on parallel programming tools such as MPI (Message Passing Interface) to transfer data between different computing nodes, thereby efficiently assembling and solving the finite element stiffness matrix.
[0060] Operator It is responsible for mapping local degrees of freedom to element degrees of freedom. Let the local mesh be... Include There are units, and the number of degrees of freedom for each unit is . For the first Unit: Let the local degrees of freedom be numbered as... Let the unit degree of freedom be numbered as Operator It is The matrix has its elements defined as follows:
[0061]
[0062] here, Indicates the first The first unit The unit degree of freedom corresponds to the local degree of freedom of the first unit degree of freedom. One degree of freedom.
[0063] Combine all units of the operators as follows: Local operators can be obtained :
[0064]
[0065] in It is The matrix. Operator transpose Responsible for mapping unit degrees of freedom back to local degrees of freedom.
[0066] In shared memory parallel computing, and The implementation usually relies on multi-threaded programming tools such as OpenMP to distribute computing tasks among multiple threads, thereby efficiently assembling and solving finite element stiffness matrices.
[0067] Assume that the local grid The first Unit There are Quadrature points And define Base function , is a × Matrix, whose elements are defined as follows:
[0068]
[0069] Where, The gradient value of the first Base function on the first Quadrature point In the first unit .
[0070] Let D(m,k) be a diagonal matrix, whose i-th diagonal element represents the weight of the i-th integral point And the unit Jacobian determinant , that is:
[0071]
[0072] Further, block-diagonal splice and :
[0073]
[0074] The unit stiffness matrix on the local grid It can be represented as a block-diagonal matrix:
[0075]
[0076] Similarly, the unit stiffness matrix of all units on the global grid can be represented as:
[0077]
[0078] Where,
[0079]
[0080] respectively called shape function matrix and node integration matrix.
[0081] In distributed computing environment, given a distributed grid, the global grid is first partitioned into multiple sub-grids (e.g. by Metis or other graph partitioning libraries), then each sub-grid is assigned to different distributed computing nodes.
[0082] Operator Responsible for mapping between global grid degrees of freedom and local grid degrees of freedom, corresponding to MPI level parallel computation in programming. MPI is used for data communication between different processors or computing nodes, and only necessary data can be exchanged between nodes, so as to efficiently complete the parallel assembly and solution of global stiffness matrix.
[0083] Operator and Used to establish the mapping between local grid degrees of freedom and element degrees of freedom, corresponding to shared memory level parallel computation. After receiving the sub-grid, the computing node will further divide the element calculation task on the sub-grid into multiple thread processing. Since these threads share the same memory address space, the shared memory parallel method can be used to efficiently process the calculation task at this level. Load balancing and orderly access to shared resources become the optimization focus.
[0084] Shape function matrix , and node integration matrix The calculation of the corresponding vector instruction level parallel (such as AVX SSE instruction of x86 processor). The parallel optimization at this level can further improve the calculation efficiency, especially in large-scale finite element calculation problems.
[0085] In the partial assembly strategy of linear system, based on operator decomposition theory, the element stiffness matrix of approach dynamics model is expressed as:
[0086]
[0087] where, denotes the element, denotes the element weight set of the point on the element, p and represent the integration point and weight respectively, , represent the integration point and Jacobian matrix determinant of integration coordinate transformation, , represent the integration point and The basis function values on, The second-order infinitesimal modulus tensor of the linearized constitutive model at the integration point and The value, , These represent the product points in the test function set. and The basis function values on.
[0088] In a fully assembled matrix-free strategy, the element stiffness matrix is pre-computed and explicitly stored in memory. This method is relatively simple to implement, but it consumes a significant amount of GPU bandwidth. For example, a... The multiplication of dense matrices and vectors will involve Second memory access and Sub-floating-point operations. Considering that the arithmetic capability of a GPU is far greater than the bandwidth of its video memory, the performance bottleneck of this step is memory bandwidth. To address this, this invention proposes a matrix-free strategy for partial assembly. In this strategy, the element stiffness matrix is no longer stored as a dense matrix, but is implicitly represented as a function of several parameters, which is calculated and reconstructed in real time when needed.
[0089] Define mapping This represents the mapping from unit degrees of freedom to global degrees of freedom. The index set... Represents a grid The global degree of freedom index set on, The total degrees of freedom, This is a set of indices representing the degrees of freedom of a unit; let's assume for now that all units are of the same type. It's not difficult to see the mapping... with operators Isomorphism allows the unit degrees of freedom on a unit to be mapped to global degrees of freedom.
[0090] Given unit and integration points Define the shape function as in Denotes basis functions. It is the index of the basis function in the global degrees of freedom. Indicated in unit The middle element has degrees of freedom of The global degrees of freedom of the vertices.
[0091] Given unit pair Define its connection matrix :
[0092]
[0093] Contact Matrix The topological relationship between two elements at the element freedom degree is described, and the shape function values on the element can be projected to the element .
[0094] Given a pair of elements , whose integration point sets are denoted as , the tensor form of the element stiffness matrix can be obtained:
[0095]
[0096] where the corresponding shape functions are as follows:
[0097]
[0098] and represent the Jacobian matrix determinant of the integration coordinate transformation on , respectively. Then the total stiffness matrix can be expressed as:
[0099]
[0100] where
[0101]
[0102] So far, the complete operator decomposition process of the bond-based peridynamic (BPD) model has been given. It is not difficult to find that the of the classical Laplace equation is a block diagonal matrix, while the of the non-local equation is a sparse block matrix. This further explains why the NbN and DbD strategies are not suitable for solving non-local problems. According to the different levels of operator decomposition, three different finite element solving strategies can be given:
[0103] Traditional fully assembly method (TFA): Calculate the total stiffness matrix , and hand over to the linear equation system solver.
[0104] Fully assembly matrix-free method (FAMF): Calculate the set of element stiffness matrices , and directly hand it over to the iterative solver for solution, where the dense matrix-vector multiplication is performed.
[0105] Partial assembly matrix-free method (PAMF): Only saves the necessary information for generating the element stiffness matrix, such as the element's global degree of freedom index. Integral points and Integration rules and coefficients It is then directly passed to the iterative solver, which calculates the product of the element stiffness matrix and the element vector at each step of the iteration process.
[0106] In matrix-free methods for solving algebraic systems, the matrix-free finite element method (FEM) is a type of numerical computation method developed based on the traditional FEM. It aims to avoid explicitly constructing and storing large matrices such as the global stiffness matrix. Traditional FEM requires assembling the global stiffness matrix during the solution process, while the matrix-free FEM skips this step and directly calculates the matrix-vector product using the element stiffness matrix, such as... Figure 3 As shown, the matrix-free method avoids the data contention and load balancing issues involved in multi-threaded parallel assembly of the total stiffness matrix, making it easier to implement parallel computing and adaptive algorithms, and exhibiting better parallelism in large-scale computations. In a distributed computing environment, this method can easily distribute computational tasks to different computing nodes, thereby improving computational efficiency. In algorithms such as adaptive mesh refinement, it can more flexibly handle mesh changes and local refinement, improving computational accuracy and efficiency.
[0107] The continuous dynamic equations are discretized into finite element equations. By introducing a trial function space, the solutions to the equations can be expressed as a linear combination of basis functions, thus yielding a global system of algebraic equations:
[0108]
[0109] Among them, matrix The global stiffness matrix. For unknown solution vectors, Let be the load vector. Let the set of element stiffness matrices be... According to the finite element operator decomposition theory, the global stiffness matrix can be expressed as:
[0110]
[0111] In the iterative solution process, the sparse matrix is calculated each time. with vector When multiplying, the calculation process can be broken down into three steps: First, calculate the unit vector. The global vector Project onto the element degrees of freedom; then, calculate the product of the dense matrix (i.e., the element stiffness matrix) and the vector. Finally, calculate The calculation results are then summed back to the global degrees of freedom, such as... Figure 4 As shown.
[0112] The highly parallel nature of matrix-vector multiplication makes it well-suited for implementation on Single Instruction Multiple Data (SIMD) devices such as GPUs. Current GPU matrix-free solvers primarily employ three parallel strategies: Node-by-Node (NbN), Degree-of-Freedom-by-Degree-of-Freedom (DbD), and Element-by-Element (EbE). The EbE strategy unfolds parallel computation at the element level, with each thread independently processing element stiffness moments. The computations between threads are relatively independent with minimal differences in computational complexity. The EbE strategy offers a moderate computational granularity, ensuring a certain level of parallelism without introducing excessive thread management overhead due to overly fine granularity. The computation of each element depends only on information within that element, typically requiring no cross-element data exchange, thus minimizing load balancing issues and making it more suitable for implementing matrix-free finite element methods for nonlocal problems.
[0113] The implementation of the various embodiments of the present invention is based on programmed processing through a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a matrix-free finite element system for near-field dynamics based on a partial assembly strategy. This system is used to execute a matrix-free finite element method for near-field dynamics based on a partial assembly strategy from the above method embodiments.
[0114] The system includes: a first main module, used to decompose the finite element stiffness matrix into a continuous product of multiple operators based on near-field dynamics operator decomposition; a second main module, used to implicitly represent the element stiffness matrix as a function of several parameters, and to calculate and restore it in real time when needed, thereby realizing partial assembly of the element stiffness matrix; and a third main module, used to iteratively solve the partially assembled element stiffness matrix, directly calculating the element stiffness matrix and element vectors during the iterative solution process, and accumulating the calculation results back to the global degrees of freedom.
[0115] This invention provides a matrix-free finite element system for near-field dynamics based on a partial assembly strategy. Addressing the shortcomings of existing near-field dynamics calculations, it employs several modules described above and replaces the traditional matrix-based strategy with a matrix-free method to achieve finite element numerical simulation of near-field dynamics models.
[0116] It should be noted that the system embodiments provided by the present application are used to implement the methods in the above method embodiments, and are also used to implement the methods in other method embodiments provided by the present application, the difference is only that the corresponding function modules are set, and the principle is basically the same as that of the above system embodiments provided by the present application. As long as the person skilled in the art improves the modules in the above system embodiments on the basis of the above system embodiments, refers to the specific technical solutions in other method embodiments, obtains the corresponding technical means by combining technical features, and the technical solutions composed of these technical means, on the premise of ensuring the practicability of the technical solutions, the corresponding system class embodiments are obtained, which are used to implement the methods in other method class embodiments.
[0117] Based on the same inventive concept as the above embodiments, the present application embodiment also provides a computing device comprising a memory and a processor, the memory storing program instructions executed by the processor, and the processor invoking the program instructions to execute the partial assembly strategy-based near-field dynamic no-matrix finite element method.
[0118] When the computing device executes the partial assembly strategy-based near-field dynamic no-matrix finite element method, the slow convergence speed, poor accuracy, high computational cost, lack of perfect theory, and low reliability and predictability of the meshless method are overcome; the finite difference method is difficult to handle problems with complex geometric shapes, which can meet the complex requirements of actual engineering, and also avoids the communication cost of directly assembling the total stiffness matrix of the finite element method, and accelerates the solution of the algebraic equation system.
[0119] In the embodiments of the present application, the memory can be a non-volatile memory such as a hard disk drive (HDD) or a solid-state drive (SSD), etc., and can also be a volatile memory such as a random-access memory (RAM). The memory can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures and capable of being accessed by a computer, but is not limited to this. The memory in the embodiments of the present application can also be a circuit or any other device capable of realizing a storage function, used to store program instructions and / or data.
[0120] In an embodiment of the present application, a processor can be a general purpose processor, a digital signal processor, an application specific integrated circuit, a field programmable gate array or other programmable logic device, discrete gate or transistor logic, discrete hardware components, can implement or perform the methods, steps and logical block diagrams disclosed in the embodiments of the present application. The general purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in conjunction with the embodiments of the present application can be directly embodied for hardware processor execution, or be executed by a combination of hardware and software modules in the processor.
[0121] According to an aspect of the present specification, a non-transitory computer-readable storage medium storing computer instructions is provided, the computer instructions causing the computer to perform the partial assembly strategy-based near-field dynamic no-matrix finite element method.
[0122] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) having computer usable program code embodied therein.
[0123] The present application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (systems) and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, a special purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce an apparatus for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in the flow or multiple flows and / or blocks Figure 1 The apparatus for implementing the functions specified in the flow or multiple flows and / or blocks.
[0124] These computer program instructions can also be stored in a computer readable memory that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce a product including instruction apparatus, which implements the functions specified in the flow Figure 1 The functions specified in the flow or multiple flows and / or blocks Figure 1 The apparatus for implementing the functions specified in the flow or multiple flows and / or blocks.
[0125] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer implemented process, so that the instructions executed on the computer or other programmable data processing devices provide a process for implementing the flowchart Figure 1 one flow or multiple flows and / or the functions specified in one block or multiple blocks. Figure 1 one flow or multiple flows and / or the functions specified in one block or multiple blocks.
[0126] In summary of the above embodiments, the key technical innovation of the present application is:
[0127] 1. Multi-level parallel framework based on operator decomposition: in a distributed computing environment, tasks are efficiently allocated and executed on distributed nodes, multiple GPUs within the nodes, and stream processing units on each GPU, achieving inter-group load balancing.
[0128] 2. Iterative solution strategy based on matrix-free method: the matrix-free method is used to replace the traditional finite element solution framework based on assembling the total stiffness matrix, overcoming the inter-group communication difficulty bottleneck in the traditional method in large-scale numerical simulation.
[0129] 3. Element stiffness matrix assembly strategy using partial assembly strategy: compressing the storage of element stiffness matrix, while reducing the memory bandwidth pressure, achieving better parallel performance.
[0130] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present application.
Claims
1. A matrix-free finite element method for near-field dynamics based on a partial assembly strategy, characterized in that, include: Based on near-field dynamics operator decomposition, the finite element stiffness matrix is decomposed into a continuous product of multiple operators; The element stiffness matrix is implicitly represented as a function of several parameters, and can be calculated and restored in real time when needed, thus realizing partial assembly of the element stiffness matrix. The stiffness matrix of the partially assembled elements is solved iteratively. During the iterative solution process, the element stiffness matrix and element vectors are directly calculated, and the calculation results are accumulated back to the global degrees of freedom.
2. The matrix-free finite element method for near-field dynamics based on a partial assembly strategy according to claim 1, characterized in that, finite element stiffness matrix Decomposing it into a continuous product of multiple operators is: in, For the distributed parallel part, For the shared memory parallel part, For the basis function part, This is the integral part.
3. The matrix-free finite element method for near-field dynamics based on a partial assembly strategy according to claim 2, characterized in that, Operator It corresponds to the first Subgrid One A matrix, responsible for mapping global degrees of freedom to local degrees of freedom, has its elements defined as follows: Its transpose operator It is responsible for mapping local degrees of freedom back to global degrees of freedom.
4. The matrix-free finite element method for near-field dynamics based on a partial assembly strategy according to claim 2, characterized in that, Operator Responsible for mapping local degrees of freedom to element degrees of freedom, with sub-mesh as an example. Include There are units, and the number of degrees of freedom for each unit is . ), operator It is The matrix has the following element definitions: in For subgrid The total number of degrees of freedom.
5. The matrix-free finite element method for near-field dynamics based on a partial assembly strategy according to claim 1, characterized in that, The method further includes: Define mapping This represents the mapping from unit degrees of freedom to global degrees of freedom, where the index set... Represents a grid The global degree of freedom index set on, The total degrees of freedom, The index set representing the unit degrees of freedom, where the mapping with operators Isomorphism.
6. The matrix-free finite element method for near-field dynamics based on a partial assembly strategy according to claim 5, characterized in that, The element stiffness matrix is implicitly represented as a function of several parameters, and its reconstruction is performed in real time when needed, achieving partial assembly of the element stiffness matrix, including: Map the unit degrees of freedom on the unit to the global degrees of freedom; Given unit and integration points Define the shape function as in , is the index of the basis function in the global degrees of freedom. Indicated in unit The middle element has degrees of freedom. The global degrees of freedom of the vertices; Given unit pair Define its connection matrix Describe the topological relationship between two elements under element degrees of freedom, and represent the shape functions in the element. The value on is projected onto the cell. superior; Given unit pair The tensor form of the element stiffness matrix is obtained from its set of integration points.
7. The matrix-free finite element method for near-field dynamics based on a partial assembly strategy according to claim 2, characterized in that, The stiffness matrix of the partially assembled elements is solved iteratively, including: Calculate unit vector , will global vector Projected onto the unit degree of freedom; Calculate the element stiffness matrix-vector product ; calculate The calculation results are then added back to the global degrees of freedom.
8. A matrix-free finite element system for near-field dynamics based on a partial assembly strategy, characterized in that, include: The first main module is used to decompose the finite element stiffness matrix into a continuous product of multiple operators based on near-field dynamics operator decomposition. The second main module is used to implicitly represent the element stiffness matrix as a function of several parameters, and to calculate and restore it in real time when needed, so as to realize the partial assembly of the element stiffness matrix. The third main module is used to iteratively solve the stiffness matrix of the partially assembled elements. During the iterative solution process, the element stiffness matrix and element vectors are directly calculated, and the calculation results are accumulated back to the global degrees of freedom.
9. A computing device, characterized in that, It includes a memory and a processor, the memory storing program instructions to be executed by the processor, the processor invoking the program instructions to execute the near-field dynamics matrix-free finite element method based on a partial assembly strategy as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the near-field dynamics matrix-free finite element method based on a partial assembly strategy as described in any one of claims 1 to 7.
