A scale-reduced near-field finite element calculation acceleration method and implementation method for structural fracture analysis
By decomposing large structures into substructures and using neural networks to train equivalent shape functions, the problem of insufficient computational efficiency of near-field dynamics in large-scale structures is solved, achieving efficient order reduction calculation and accurate fracture analysis.
Patent Information
- Application Number
- CN202610648385.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-12
- Publication Date
- 2026-06-23
AI Technical Summary
Existing numerical methods for near-field dynamics are computationally inefficient and costly when dealing with large-scale structures, making them difficult to apply directly to large-scale engineering problems, and they also incur significant overhead from repeated calculations.
The large structure is divided into several substructures, and a neural network method is introduced to train the parameters of the substructures. The order reduction calculation is achieved by constructing the equivalent shape function of the coarse-scale unit. Combined with the adaptive update mechanism, the computational complexity is reduced and the efficiency is improved.
It significantly improves the computational efficiency of peridynamics in large-scale simulations, making it applicable to larger-scale and more complex engineering problems, reducing computational degrees of freedom and costs, while maintaining an accurate description of fracture behavior.
Smart Images

Figure CN122263547A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computational mechanics, specifically relating to a scale-reduced near-field finite element method for structural fracture analysis and its implementation. Background Technology
[0002] Brittle failure is a significant form of structural failure in engineering mechanics, characterized by sudden structural failure without significant plastic deformation, often leading to catastrophic consequences. Therefore, preventing such brittle failure is crucial for ensuring structural safety in modern engineering fields such as civil and mechanical engineering.
[0003] In recent years, the international academic community has proposed and developed a novel nonlocal continuum mechanics theory—perifield dynamics—to address the problem of structural and material failure under load. This theory posits that materials are composed of numerous material points containing material information. Each material point not only carries its own material information but also interacts with other material points within a certain range of its neighborhood. Perifield dynamics establishes its model based on the fundamental idea of nonlocality, replacing the differential equations in traditional continuum mechanics with spatial integral differential equations. This allows for a unified mechanical description of both continuous and discontinuous domains, overcoming the limitations of traditional theories in handling discontinuous problems. It can naturally simulate the entire process of a structure from crack initiation to propagation and eventual complete failure. This is mainly due to the fact that the constitutive model of perifield dynamics directly includes descriptions of damage and fracture, eliminating the need for additional crack initiation criteria or pre-defined crack paths; therefore, discontinuities can spontaneously form during simulation. This theory has been successfully applied to mechanical problems involving significant discontinuous deformation effects in structures, such as static and dynamic crack propagation and impact failure. The proposal of peri-field dynamics is not only a useful supplement to the traditional continuum mechanics theory, but also provides new research ideas and methods for solving the classic mechanical problem of solid failure.
[0004] However, compared with classical local theory, the nonlocality of peridynamics leads to higher computational costs, which poses a challenge to its numerical solution and limits its widespread application in practical engineering. To improve the solution efficiency of peridynamic numerical models, several acceleration methods have been proposed, mainly including the following four common processing strategies:
[0005] (1) Realization of peri-field dynamics based on high-performance computing
[0006] Leveraging rapidly advancing computer hardware and parallel computing technology, numerous independent operations in near-field dynamics numerical calculations can be distributed across multiple processors for parallel execution, significantly improving computational efficiency, particularly for large-scale numerical problems. However, the data communication and synchronization between parallel tasks greatly influence the speedup ratio; improper handling can slow down computation and, more seriously, introduce errors into the results. Furthermore, parallel acceleration only speeds up computation; it doesn't truly reduce the computational load. When dealing with large-scale structural problems, unsolvable situations may still arise.
[0007] (2) Realization of peri-field dynamics based on coupling modeling
[0008] In this method, coupled modeling between peridynamics and classical continuum mechanics is a crucial approach to improving computational efficiency. The method employs peridynamic models in critical regions where cracks may occur to accurately capture crack initiation and propagation, while non-critical regions are described using the more computationally efficient classical continuum theory. By establishing constraints between different models, coupling between different theoretical frameworks is achieved, thereby significantly reducing the overall computational scale while maintaining computational accuracy in critical regions. This "regional coupling" strategy effectively balances computational accuracy and efficiency, providing feasibility for applying peridynamics to the failure analysis of large-scale practical engineering structures.
[0009] (3) Implementation of near-field dynamics accelerated by neural networks
[0010] Deep learning neural networks, as powerful computational tools, have been gradually introduced into the field of computational mechanics, providing new acceleration pathways for the numerical implementation of peridynamics. Among them, the physical information neural network, as a specific framework, has shown good results on multiple problems by directly constructing the solution space using a feedforward neural network with spatial and temporal coordinates as input. The innovation of this framework lies in embedding physical constraints such as governing equations into the loss function through automatic differentiation techniques, and in being able to handle partial differential equations or parameter identification tasks using inverse problem-solving strategies. However, neural network-based acceleration methods for peridynamics still face several challenges. On the one hand, the constructed neural networks usually do not have complete generalization ability. Since the governing equations of peridynamics change during material damage, models trained for a specific problem are often difficult to directly generalize to other problems. On the other hand, to achieve a reasonable balance between computational efficiency and prediction accuracy, the problem scale that existing methods can handle is still limited, making it difficult to directly apply them to large-scale complex engineering scenarios.
[0011] (4) Introducing a peri-field dynamics model within the traditional finite element framework
[0012] Patent ZL202110885638.8 proposes a near-field finite element method (PeriFEM). Its main idea is to introduce a near-field dynamic model into the traditional finite element framework. By constructing near-field elements, the numerical solution of nonlocal fracture problems can be realized, thereby solving the problem that the near-field dynamic model is difficult to implement on existing finite element software platforms.
[0013] Compared to the three methods mentioned above, the core advantage of method (4) lies in its engineering practicality and platform compatibility. Although method (1) can accelerate the solution, it does not reduce the amount of computation, and improper parallel task decomposition and data synchronization may slow down the speed or even lead to errors. Although method (2) can balance accuracy and efficiency, it faces technical challenges such as spatial scale mismatch, time step asynchrony and interface force transmission error. Method (3) is limited by insufficient generalization ability and problem scale limitation, making it difficult to be directly applied to large-scale complex engineering scenarios. In contrast, PeriFEM introduces a peri-field dynamics model into the traditional finite element framework and constructs peri-field elements (PE), incorporating the solution of nonlocal fracture problems into the standard paradigm of the finite element method. This allows engineers to directly perform peri-field dynamics calculations on widely used commercial finite element software platforms such as ABAQUS and ANSYS without developing an independent solver or rebuilding the numerical framework. Under the same mesh roughness, PeriFEM is easier to obtain correct results than the traditional PD meshless method, achieving effective simulation. This method not only utilizes the mature mathematical foundation and software platform advantages of the finite element method, but also overcomes the shortcomings of traditional finite element methods that require continuous mesh reconstruction when simulating crack propagation, thus becoming an important bridge for near-field dynamics to move towards engineering applications.
[0014] However, method (4) still has the following main problems:
[0015] ① Large computational scale and high computational cost: Since the near-field dynamics model is essentially a nonlocal model, each element needs to interact with multiple elements in its neighborhood, resulting in a computational complexity significantly higher than that of the traditional finite element method. In the case of fine mesh discretization, the system's degrees of freedom are enormous, leading to low computational efficiency;
[0016] ② Difficult to apply directly to large-scale engineering problems: When the structure size is large or the crack propagation path needs to be finely characterized, a high-resolution mesh must be used, which further increases the computational burden and limits the application of this method in large-scale engineering problems;
[0017] ③ The overhead of repeated calculations is large: During the fracture evolution process, as the bond state is constantly updated, the stiffness matrix or related local quantities need to be calculated repeatedly, resulting in a long overall calculation time.
[0018] Therefore, it is necessary to propose a near-field finite element method for structural fracture analysis that can solve the above problems. Summary of the Invention
[0019] The purpose of this invention is to overcome the shortcomings of existing numerical calculation methods for perifield dynamics in terms of computational efficiency when dealing with large-scale structures. Based on the concept of structural decomposition, this invention breaks down large structures into several substructures (i.e., coarse-scale units) and introduces a neural network method to train the parameters of the substructures to obtain a surrogate model. This model generates the shape function parameters of the substructures in real time, achieving effective order reduction and accelerated computation of the original problem, thereby significantly improving the computational efficiency of perifield dynamics in large-scale simulations.
[0020] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0021] A scale-reduced near-field finite element method for structural fracture analysis to accelerate computation includes the following steps:
[0022] (A1) Target Structure Perform geometric modeling, and set the material properties (Young's modulus and Poisson's ratio), boundary conditions, load conditions, near-field dynamic parameters, coarse-scale element construction rules, and computational control parameters of the target structure; among them, the computational control parameters include the analysis problem type, maximum loading step or maximum time step, and loading step size or time step size;
[0023] Then, by performing finite element decomposition according to the original problem, we obtain... There are finite elements, each finite element is denoted as a fine-scale element, the mesh formed by all the fine-scale elements is denoted as a fine-scale mesh, and the computational unit formed by the interaction between any two fine-scale elements is denoted as a fine-scale near-field element.
[0024] Based on spatial neighborhood relationships, several fine-scale units are grouped together and denoted as coarse-scale units. The grid formed by all coarse-scale units is denoted as a coarse-scale grid.
[0025] (A2) Obtain the equivalent shape function for each coarse-scale unit. , , It is the identity matrix. This refers to the portion that actually needs to be calculated;
[0026] (A3) Calculate the global stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh. ;
[0027] (A4) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid. ;
[0028] (A5) Solve for the global nodal displacements of the coarse-scale mesh. ;
[0029] (A6) Solve for the global nodal displacements of the fine-scale mesh. ;
[0030] (A7) Determine whether the current loading step or time step has reached the maximum loading step or time step requirement ("loading step" is used for quasi-static problems, and "time step" is used for dynamic problems). If yes, proceed to step (A9); otherwise, proceed to step (A8).
[0031] (A8) Based on the global nodal displacement of the fine-scale mesh Update the bond states of all fine-scale near-field units;
[0032] If a break occurs, return to step (A2), that is, return to the coarse-scale cell in the corresponding coarse-scale mesh and re-obtain the equivalent shape function of the coarse-scale cell;
[0033] If no broken keys are generated and the original problem is a quasi-static problem, then add a loading step and return to step (A4).
[0034] If no key breaks occur and the original problem is a dynamic problem, then add a time step and return to step (A4).
[0035] (A9) Global nodal displacement based on fine-scale mesh Output target structure The displacement cloud map is obtained, and the target structure is calculated and output based on the bond break information. The equivalent damage cloud map.
[0036] While existing technologies also include methods that divide the research area into coarse grid cells and further subdivide each coarse grid cell into fine grid cells to form a multi-scale grid, these methods typically employ a top-down hierarchical partitioning strategy. That is, a coarse grid is first constructed, and then finer subdivisions are made within the coarse grid, resulting in a clear hierarchical nesting relationship between the coarse and fine grids.
[0037] The segmentation method of the present invention differs fundamentally from the aforementioned prior art, specifically in that:
[0038] The segmentation method is different: from "top-down" to "bottom-up". This invention first segments the target structure. A uniform fine-scale mesh is generated using the traditional finite element method (corresponding to the fine-scale elements in step A1). Then, based on spatial neighborhood relationships, several fine-scale elements are combined according to certain rules to construct coarse-scale elements. That is, the coarse-scale elements of this invention are not obtained by meshing the target structure, but are directly formed by the aggregation of fine-scale elements.
[0039] The construction of coarse-scale units is based on different principles: in the prior art, the division of coarse and fine meshes is usually based on geometric division rules or preset mesh hierarchy relationships, while in this invention, the construction of coarse-scale units is based on the set relationship of fine-scale units, and can be combined with the near-field interaction range (such as neighborhood relationship) for organization, thereby better adapting to the nonlocal characteristics of near-field dynamics.
[0040] Unlike the coupling method of subsequent calculation methods, existing multi-scale mesh methods mainly serve local refinement calculations or multi-scale coupled analysis, while the meshing method of this invention serves the "scale-reduced near-field finite element method". Its core purpose is to construct equivalent shape functions through the mapping from fine-scale elements to coarse-scale elements, thereby reducing the order of computational degrees of freedom.
[0041] The adoption of the above-mentioned segmentation method, which differs from existing technologies, is mainly based on the following considerations:
[0042] • Ensure the consistency and reversibility of the order reduction process: Directly aggregating fine-scale units into coarse-scale units can establish a clear degree-of-freedom mapping relationship, which facilitates the construction of equivalent shape functions and realizes the recovery from coarse-scale results to fine-scale results;
[0043] • Better adapt to the nonlocal characteristics of peridynamics: There are neighborhood interactions between units in peridynamics. Using a partitioning method based on fine-scale unit sets is beneficial to retain fine-scale nonlocal information at the coarse-scale level.
[0044] • Improve computational efficiency and reduce modeling complexity: There is no need to build separate coarse meshes and then refine them, avoiding the complexity of multi-layer mesh generation and management, while providing a unified data structure foundation for subsequent order reduction calculations and neural network modeling.
[0045] Therefore, the meshing method of the present invention differs from the prior art in both its construction method and the computational objectives it serves, and provides a foundation for achieving efficient scale-reduced near-field finite element analysis.
[0046] like Figure 1 As shown, the present invention first performs finite element decomposition on the original problem to obtain... The problem is that when there are a large number of fine-scale elements, the numerical solution to this problem will require a huge amount of computation, and it may even become unmanageable. Therefore, based on spatial neighborhood relationships, several fine-scale elements are grouped together as a coarse-scale element, and the mesh formed by all the coarse-scale elements is called the coarse-scale mesh. Finally, after the above hierarchical transformation, the order of the original problem can be reduced. For example, solid nodes (internal nodes) in the coarse-scale element will be eliminated, so the order of the original problem will be reduced, and the amount of computation required will be reduced.
[0047] As a preferred technical solution:
[0048] As described above, in the scale-reduced near-field finite element method for structural fracture analysis, in step (A2), the equivalent shape function of each coarse-scale element... The actual part that needs to be calculated It was obtained through calculation, and the calculation steps are as follows:
[0049] (A21) Calculate the near-field finite element stiffness of coarse-scale elements. The formula is as follows:
[0050] ;
[0051] ;
[0052] ;
[0053] ;
[0054] In the formula, This represents the total number of fine-scale near-field elements within the coarse-scale element;
[0055] This is the nodal degree of freedom transformation matrix for the fine-scale near-field elements within the coarse-scale element;
[0056] is the element stiffness matrix of the fine-scale near-field element in the coarse-scale element;
[0057] and These represent two fine-scale units that constitute the fine-scale near-field unit;
[0058] Let be the shape function difference matrix of the fine-scale near-field elements in the coarse-scale element, where x and x represent points within two fine-scale cells that constitute a fine-scale near-field cell, respectively;
[0059] It is the micromodulus matrix;
[0060] and Represent and the volume of x;
[0061] and These are the shape function matrices of the two fine-scale units that constitute the fine-scale near-field unit;
[0062] For near-field dynamic bonds;
[0063] for The micromodulus coefficient;
[0064] for At time step (When dealing with quasi-static problems) Indicates loading step, in dynamic problems This represents the state at time step (here, time step is used uniformly), and its value is either 0 or 1. express Fracture, 1 indicates Not broken;
[0065] and for The components of a vector;
[0066] (A22) Near-field finite element stiffness of coarse-scale elements Expanding on the expression, the result is as follows:
[0067] ;
[0068] In the formula, , , and They are respectively The sub-blocks, where subscripts b and i represent the boundary node and internal node of the coarse-scale element, respectively, are... ;
[0069] Let be the weighting function, representing the number of times a fine-scale element appears in a coarse-scale element. When calculating the equivalent shape function of each coarse-scale element, the weights of overlapping elements need to be considered. For example, if a fine-scale element appears in two coarse-scale elements, then the stiffness of that fine-scale element needs to be corrected when calculating its stiffness. ;
[0070] (A23) Calculate the equivalent shape function for each coarse-scale element. The actual part that needs to be calculated The formula is as follows:
[0071] .
[0072] As described above, in the scale-reduced near-field finite element method for structural fracture analysis, in step (A2), the equivalent shape function of each coarse-scale element... The actual part that needs to be calculated It is obtained through prediction. The specific process is as follows: first, determine the bond states of all fine-scale near-field units in each coarse-scale unit; then, input these states into the trained neural network, which outputs the equivalent form function of the coarse-scale unit. The actual part that needs to be calculated ;
[0073] The training steps for a neural network are as follows:
[0074] (B1) Obtain coarse-scale units from the sample structure according to the method in step (A1). The sample structure and the target structure can be the same or different. The basic parameters of each coarse-scale unit include the size of the coarse-scale unit, the size of the fine-scale unit, the arrangement of the fine-scale units, the distribution of Poisson's ratio, and the distribution of elastic modulus.
[0075] (B2) Cracks of different positions and lengths are randomly arranged in each coarse-scale unit to randomly change the bond state of all fine-scale near-field units in each coarse-scale unit, and the bond state of all fine-scale near-field units in all coarse-scale units is normalized.
[0076] (B3) Obtain the equivalent shape function of each coarse-scale unit through calculation. The actual part that needs to be calculated ;
[0077] (B4) Determine the hyperparameters of the neural network, including the number of layers, the number of neurons in each layer, the type of activation function, the learning rate, the loss function, the batch size, the period, etc.
[0078] (B5) Taking the bond states of all fine-scale near-field units in each coarse-scale unit after normalization as input, and the equivalent shape function of each coarse-scale unit as input. The actual part that needs to be calculated To provide theoretical output, the weight parameters and bias parameters of the neural network are continuously updated iteratively until the termination condition is met, which completes the training of the neural network. The resulting trained neural network can be reused in different overall geometry and boundary condition problems within the same coarse-scale unit construction rules and material parameters.
[0079] As described above, in the scale-reduced near-field finite element method for structural fracture analysis, step (A3) involves calculating the global stiffness matrix of the coarse-scale mesh. The formula used is as follows:
[0080] ;
[0081] ;
[0082] In the formula, The condensation stiffness matrix for each coarse-scale element;
[0083] The symbol represents the condensed stiffness matrix of all coarse-scale elements. Assembly;
[0084] The near-field finite element stiffness is represented by the coarse-scale element.
[0085] As described above, in a scale-reduced near-field finite element method for structural fracture analysis, in step (A4), the global load vector of the coarse-scale mesh... The calculation formula is as follows:
[0086] ;
[0087] ;
[0088] ;
[0089] In the formula, This represents the overall load vector for the fine-scale mesh.
[0090] The symbol represents the total load vector of the fine-scale mesh. The number of degrees of freedom nodes corresponding to the coarse-scale grid are extracted and assembled.
[0091] This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh;
[0092] For the element load vector of the fine-scale mesh;
[0093] It is a fine-scale unit;
[0094] The shape function matrix of the fine-scale unit. Points within a fine-scale unit;
[0095] For external force field at point External force vector at the location;
[0096] for The volume.
[0097] As described above, in a scale-reduced near-field finite element method for structural fracture analysis, step (A5) involves solving for the global nodal displacements of the coarse-scale mesh if the original problem is quasi-static. The formula used is as follows:
[0098] ;
[0099] If the original problem is dynamic, then solve for the global nodal displacements of the coarse-scale mesh. The formula used is as follows:
[0100] ;
[0101] ;
[0102] ;
[0103] ;
[0104] ;
[0105] In the formula, The mass matrix of the coarse-scale grid;
[0106] This represents the overall nodal acceleration of the coarse-scale grid at time step t.
[0107] This represents the total nodal displacement of the coarse-scale grid at time step t.
[0108] This represents the total load vector of the coarse-scale grid at time step t.
[0109] The equivalent mass matrix for coarse-scale units;
[0110] The symbol represents the equivalent mass matrix of all coarse-scale units. Assembly;
[0111] This represents the overall mass matrix of all fine-scale cells within the coarse-scale cell;
[0112] This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh;
[0113] The mass matrix of the fine-scale unit;
[0114] It is a fine-scale unit;
[0115] The shape function matrix of the fine-scale unit. Points within a fine-scale unit;
[0116] for The volumetric mass density;
[0117] for The volume.
[0118] As described above, in the scale-reduced near-field finite element method for structural fracture analysis, step (A6) involves solving for the global nodal displacements of the fine-scale mesh. The formula used is as follows:
[0119]
[0120] ;
[0121] ;
[0122] In the formula, This represents the nodal displacement of the fine-scale element within the coarse-scale element;
[0123] The symbol represents the nodal displacement of all coarse-scale elements in the fine-scale elements. Assembly;
[0124] This refers to the displacement of the boundary nodes in the coarse-scale element;
[0125] The symbol represents the global nodal displacement from the coarse-scale grid. Extracting boundary node displacements from coarse-scale elements .
[0126] As described above, in the scale-reduced near-field finite element method for structural fracture analysis, step (A9) requires that if the original problem is dynamic, the acceleration of the fine-scale mesh should also be output. Acceleration of fine-scale mesh acceleration from coarse-scale grid It was obtained through a one-step downscaling calculation.
[0127] The implementation method of the scale-reduced near-field finite element method for structural fracture analysis, as described above, in commercial finite element software includes the following steps:
[0128] (E1) In commercial finite element software and its preprocessing module or secondary development interface, establish the geometric model of the target structure, set the material properties, boundary conditions, load conditions, near-field dynamic parameters and calculation control parameters, and perform fine-scale meshing on the geometric model to obtain fine-scale elements and fine-scale meshes. The calculation unit formed by the interaction between any two fine-scale elements is called a fine-scale near-field element.
[0129] Subsequently, coarse-scale cells are constructed based on the fine-scale grid to form a coarse-scale grid;
[0130] (E2) Obtain the equivalent shape function for each coarse-scale unit;
[0131] (E3) The process of calculating the overall stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh is written into the user element or user subroutine of the commercial finite element software, and the calculation of the overall stiffness matrix of the coarse-scale mesh is completed by the commercial finite element software.
[0132] (E4) Calculate the total load vector of the coarse-scale mesh based on the coarse-scale mesh, and complete the numerical solution in a commercial finite element solver to obtain the total nodal displacement of the coarse-scale mesh;
[0133] (E5) Recover the total nodal displacements of the fine-scale mesh based on the total nodal displacements of the coarse-scale mesh;
[0134] (E6) Determine whether the current loading step or time step has met the preset requirements, and update the bond state of the fine-scale near-field element according to the overall node displacement of the fine-scale mesh; if there are new broken bonds, return to step (E2) and reacquire the equivalent shape function of the affected coarse-scale element; if there are no new broken bonds and the preset loading step or time step requirements have not been met, proceed to the next loading step or the next time step to continue the calculation, i.e., return to step (E4).
[0135] (E7) When the current loading step or time step reaches the preset requirements, the displacement cloud map, equivalent damage cloud map and crack propagation process of the target structure are output in the post-processing module of the commercial finite element software.
[0136] The above-described method for accelerating near-field finite element analysis of structural fracture analysis, using scale-reduced near-field finite element methods, is implemented in material microstructure fracture simulation analysis based on X-ray micro-CT data, and includes the following steps:
[0137] (F1) Obtain X-ray micro-CT scan data of material samples and reconstruct the images of X-ray micro-CT scan data to obtain two-dimensional or three-dimensional image data of material microstructure.
[0138] Among them, two-dimensional or three-dimensional image data of material microstructure are used to characterize the pores, inclusions, cracks, different material phases or other microstructural features inside the material;
[0139] (F2) Based on two-dimensional or three-dimensional image data of material microstructure, establish a geometric model of the target material microstructure, and identify material phases, pore regions, inclusion regions or initial defect regions according to the two-dimensional or three-dimensional image data of material microstructure, and assign corresponding material properties to different regions according to the identification results;
[0140] (F3) Set the computational conditions required for the simulation analysis of material microstructure fracture, and construct fine-scale and coarse-scale meshes based on the geometric model;
[0141] The calculation conditions include material properties, boundary conditions, load conditions, near-field dynamic parameters, bond breakage criteria, maximum loading step or maximum time step, and loading step size or time step size.
[0142] The geometric model is divided into several fine-scale elements, and all the fine-scale elements constitute a fine-scale mesh. The computational unit formed by the interaction between any two fine-scale elements is called a fine-scale near-field element.
[0143] Subsequently, based on spatial neighborhood relationships, several fine-scale units are grouped into coarse-scale units, and all coarse-scale units constitute a coarse-scale grid.
[0144] (F4) Obtain the equivalent shape function for each coarse-scale unit;
[0145] (F5) Calculate the overall stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh;
[0146] (F6) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid;
[0147] (F7) Under the given load conditions or boundary conditions, solve for the total nodal displacement of the coarse-scale mesh, and solve for the total nodal displacement of the fine-scale mesh based on the total nodal displacement of the coarse-scale mesh.
[0148] (F8) Determine and update the bond state of the fine-scale near-field elements based on the total nodal displacement of the fine-scale mesh;
[0149] If a break occurs, return to the corresponding coarse-scale unit and re-obtain the equivalent shape function of that coarse-scale unit, i.e., return to step (F4).
[0150] If no broken bonds are generated and the current analysis is a quasi-static fracture analysis, then add a loading step and return to step (F6).
[0151] If no broken bonds are generated and the current analysis is a dynamic fracture analysis, then increase the time step and return to step (F6).
[0152] (F9) When the current loading step or time step reaches the preset requirements, output the simulation analysis results of material microstructure fracture;
[0153] The simulation analysis results include displacement cloud maps of the material microstructure, equivalent damage cloud maps, crack initiation locations, crack propagation paths, and the analysis results of the influence of pores, inclusions, or initial defects on the material fracture behavior.
[0154] The above-described method for accelerating near-field finite element analysis of structural fracture analysis, using a scale-reduced near-field finite element method, is implemented in the dynamic damage simulation analysis of structural impact / explosion, comprising the following steps:
[0155] (G1) Establish the geometric model of the target structure to be analyzed and set the calculation conditions required for dynamic failure simulation analysis;
[0156] The target structures include perforated structures, notched structures, plate structures, beam structures, shell structures, or other engineering structures that are subjected to impact loads or explosive loads.
[0157] The calculation conditions include material properties, bulk density, boundary conditions, impact or explosive load conditions, load time history, time step, total calculation time, near-field dynamic parameters, and bond breakage criteria;
[0158] Impact loads or explosive loads include internal pressure loads, impact velocity boundary loads, impact force loads, pressure time history loads, or equivalent explosive impact loads;
[0159] (G2) The geometric model is divided into finite element subdivisions to obtain several fine-scale elements. All fine-scale elements constitute a fine-scale mesh. The computational element formed by the interaction between any two fine-scale elements is called a fine-scale near-field element.
[0160] (G3) Based on spatial neighborhood relationships, several fine-scale units are grouped into coarse-scale units, and all coarse-scale units constitute a coarse-scale grid;
[0161] This step is used to convert some degrees of freedom in a fine-scale mesh into equivalent degrees of freedom in a coarse-scale mesh, thereby reducing the computational scale of dynamic problems;
[0162] (G4) Obtain the equivalent shape function for each coarse-scale unit;
[0163] (G5) Calculate the global stiffness matrix and global mass matrix of the coarse-scale mesh.
[0164] (G6) Calculate the overall load vector for the coarse-scale mesh;
[0165] Among them, the overall stiffness matrix is used to characterize the dynamic stiffness response of the structure, the overall mass matrix is used to characterize the inertial effect of the structure, and the overall load vector is used to characterize the effect of impact load or explosive load at the current time step.
[0166] (G7) Under the time-stepping framework, solve for the global nodal displacement and global nodal acceleration of the coarse-scale mesh at the current time step; wherein, the global nodal displacement and global nodal acceleration are obtained by solving the dynamic equations of the coarse-scale mesh;
[0167] (G8) Based on the total nodal displacement of the coarse-scale mesh, recover the total nodal displacement of the fine-scale mesh, and update the bond state of the fine-scale near-field elements based on the total nodal displacement of the fine-scale mesh.
[0168] Specifically, when a bond in a fine-scale near-field cell satisfies the bond breakage criterion, the bond state is updated to the broken state.
[0169] (G9) Determine whether the current time step has reached the preset maximum time step or the preset total calculation time;
[0170] If the preset maximum time step or preset total computation time is not reached, determine whether there are any new broken bonds based on the bond state of the fine-scale near-field unit.
[0171] If a new break occurs, return to step (G4), reacquire the equivalent shape function of the affected coarse-scale element, and continue with subsequent calculations;
[0172] If no new key breaks are added, proceed to the next time step and return to step (G6).
[0173] If the preset maximum time step or preset total calculation time is reached, proceed to step (G10).
[0174] (G10) Outputs the dynamic damage simulation analysis results of the target structure under impact load or explosive load;
[0175] The results of dynamic failure simulation analysis include displacement contour maps, acceleration contour maps, equivalent damage contour maps, crack initiation locations, crack propagation paths, failure ranges, and fragmentation modes of the target structure at each time step.
[0176] Invention principle:
[0177] This invention can effectively solve the problems existing in the current near-field finite element method because:
[0178] 1. Reducing computational degrees of freedom through scale reduction: This invention divides the fine-scale mesh into several coarse-scale units. By constructing the equivalent shape function of the coarse-scale units, the fine-scale degrees of freedom are mapped to the coarse-scale space, thereby significantly reducing the number of degrees of freedom participating in the overall solution and reducing the computational scale.
[0179] 2. Preserving fine-scale information through equivalent shape functions: Equivalent shape functions can reflect the fine-scale near-field interactions and fracture states within coarse-scale elements, enabling accurate descriptions of structural fracture behavior while performing order-reduction calculations.
[0180] 3. Replace high-cost local computation with neural networks: Offline training is performed on coarse-scale units to establish the mapping relationship between equivalent shape functions and their feature parameters. During the computation process, prediction is performed directly through neural networks, avoiding repeated solutions to local substructure problems, thereby further reducing computation time.
[0181] 4. Accuracy is guaranteed through an adaptive update mechanism: During the fracture process, when the state of the fine-scale bond changes, only the equivalent shape function of the affected coarse-scale unit is updated, thus achieving a balance between computational efficiency and computational accuracy.
[0182] Therefore, this invention, through the technical approach of "scale reduction + neural network acceleration", effectively reduces computational complexity and cost while maintaining the fracture simulation capability of the near-field finite element method, enabling the method to be applied to larger-scale and more complex engineering problems, and has significant technological progress.
[0183] The method of this invention is implemented by computer and applied to the process of structural damage analysis and fracture simulation. It involves specific engineering technical steps such as structural geometric modeling, mesh generation, stiffness matrix assembly, load application and numerical solution. It also uses neural networks to predict and accelerate key parameters in the engineering calculation process. The overall solution directly affects the mechanical response analysis of engineering structures and belongs to the engineering technical method in the field of computational mechanics, rather than a simple mathematical calculation method or abstract algorithm.
[0184] Beneficial effects:
[0185] This invention further introduces the concept of scale reduction and neural network acceleration strategy based on the existing near-field finite element method, and has the following substantial technical differences:
[0186] 1. Introduction of multi-scale order reduction mechanism: This invention constructs a hierarchical structure of "fine-scale grid - coarse-scale grid" to transform the large-scale near-field interaction calculations originally performed at the fine scale into equivalent calculations at the coarse-scale grid, thereby significantly reducing the scale of the system's degrees of freedom;
[0187] 2. An equivalent shape function construction method is proposed: Based on the bond state of the fine-scale near-field unit inside the coarse-scale unit, the equivalent shape function of the coarse-scale unit is constructed to realize the effective mapping of fine-scale information to the coarse-scale space, thereby completing the order reduction while ensuring the ability to describe fracture behavior.
[0188] 3. Acceleration by introducing a neural network surrogate model: To address the problem of high computational cost of equivalent shape functions, this invention constructs training samples independent of specific structures and uses a neural network to predict the equivalent shape functions of coarse-scale units, thereby avoiding repeated solutions to local substructure problems and further improving computational efficiency.
[0189] Based on the above differences, the present invention has at least the following technical advancements compared to the existing near-field finite element method:
[0190] 1. From "computable" to "efficient": While maintaining the fracture simulation capability of the near-field finite element method, the computational complexity is significantly reduced, making it applicable to large-scale structural problems;
[0191] 2. Significantly reduce computational degrees of freedom and computational cost: By reducing the order, the original high-dimensional nonlocal problem is transformed into a low-dimensional equivalent problem;
[0192] 3. Improve the engineering applicability and scalability of the method: By combining neural networks to achieve rapid prediction, the method has the potential to be widely applied in complex engineering scenarios;
[0193] 4. Provides a new technical path for large-scale engineering applications of near-field dynamics.
[0194] Therefore, this invention not only differs from the existing near-field finite element method in terms of technical means, but also achieves significant improvements in computational efficiency and engineering practicality.
[0195] This invention discloses a scale-reduced near-field finite element method for structural fracture analysis, which enables efficient scale reduction calculations for large-scale structural problems. It transforms large-scale problems that are difficult to solve directly into solvable problems and significantly improves the computation speed of solvable problems. By training the substructure in a problem-independent manner, the resulting neural network possesses excellent generalization ability and can be directly applied to fracture problems under different geometric and boundary conditions without repeated training, greatly improving applicability and computational efficiency. Furthermore, by leveraging substructure dimensionality reduction and neural network prediction, the solution time for linear equations can be significantly reduced, achieving high efficiency in the near-field dynamics analysis process while ensuring computational accuracy.
[0196] The scale-reduced near-field finite element method of the present invention can decompose a large-scale structure and reduce the order of the substructures without losing information values. The resulting reduced-order structure can still recover the displacement information on the fine-scale mesh. Attached Figure Description
[0197] Figure 1 This is a schematic diagram illustrating the transformation between fine-scale units and meshes and coarse-scale units and meshes in this invention;
[0198] Figure 2 This is a schematic diagram of random sample generation with cracks according to the present invention; in the figure, a is a schematic diagram of sample generation when the crack initiation point is located outside the substructure, b is a schematic diagram of sample generation when the crack initiation point is located inside the substructure, l is the size of the substructure, l c c is the length of the crack. x c y The x and y coordinates represent the crack initiation points, respectively. The angle between the crack and the x-axis;
[0199] Figure 3 This is a schematic diagram illustrating the normalization of bond states in fine-scale near-field units according to the present invention.
[0200] Figure 4 This invention provides the equivalent shape function for outputting coarse-scale elements. The actual part that needs to be calculated A schematic diagram of a neural network; in the diagram, PE represents the normalized decimal number of the bond of each fine-scale near-field unit (actually...). Figure 3 (Result after normalization) Subscript This represents the total number of fine-scale near-field units in a substructure;
[0201] Figure 5This is a schematic diagram of the quasi-static calculation results of the double-slotted plate structure in Embodiment 2 of the present invention; in the figure, (a) is a schematic diagram of the geometric dimensions and load conditions of the plate; (b) is a cloud map of the final damage result, 15×15 represents the substructure dimensions used in this example, 160 represents the length of the slot, and 675 represents the length and width of the plate;
[0202] Figure 6 The figure shows the calculation results of the dynamic fracture of the perforated plate structure under high internal pressure in Embodiment 4 of the present invention; in the figure, (a) is a schematic diagram of the geometric dimensions and load conditions of the plate; (b) is a cloud map of the final damage result, 15×15 represents the substructure size used in this example, 45 represents the side length of the central square hole, and 1203 represents the length and width of the plate.
[0203] Figure 7 This is a schematic diagram showing the geometric dimensions and load conditions of the two-dimensional flat plate structure in Embodiment 5 of the present invention;
[0204] Figure 8 The equivalent damage cloud map at time step t=20 is obtained by using the method in Example 5 and the original PeriFEM method; in the figure, (a) is the equivalent damage cloud map obtained by using the original PeriFEM method, (b) is the equivalent damage cloud map obtained by using 10×10 coarse-scale elements, and (c) is the equivalent damage cloud map obtained by using 15×15 coarse-scale elements.
[0205] Figure 9 The diagram shows the computation time obtained using the method in Example 5 and the original PeriFEM method on the same hardware platform. In the figure, (a) is the computation time obtained using the original PeriFEM method, (b) is the computation time obtained using 10×10 coarse-scale elements, and (c) is the computation time obtained using 15×15 coarse-scale elements. Detailed Implementation
[0206] The present invention will be further described below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
[0207] Example 1
[0208] A scale-reduced near-field finite element method for structural fracture analysis is proposed, comprising the following steps:
[0209] (A1) such as Figure 1 As shown, for the target structure Perform geometric modeling, and set the material properties, boundary conditions, load conditions, near-field dynamic parameters, coarse-scale element construction rules, and computational control parameters of the target structure; among them, the computational control parameters include the analysis problem type, maximum loading step or maximum time step, and loading step size or time step size;
[0210] Then, following the original problem (quasi-static problem), a finite element subdivision was performed to obtain... There are finite elements, each finite element is denoted as a fine-scale element, the mesh formed by all the fine-scale elements is denoted as a fine-scale mesh, and the computational unit formed by the interaction between any two fine-scale elements is denoted as a fine-scale near-field element.
[0211] Based on spatial neighborhood relationships, several fine-scale units are grouped together and denoted as coarse-scale units. The grid formed by all coarse-scale units is denoted as a coarse-scale grid.
[0212] (A2) Obtain the equivalent shape function for each coarse-scale unit. , , It is the identity matrix. This refers to the portion that actually needs to be calculated;
[0213] Equivalent form function of each coarse-scale unit The actual part that needs to be calculated It is obtained through prediction. The specific process is as follows: first, determine the bond states of all fine-scale near-field units in each coarse-scale unit; then, input these states into the trained neural network, which outputs the equivalent form function of the coarse-scale unit. The actual part that needs to be calculated ;
[0214] The training steps for a neural network are as follows:
[0215] (B1) Obtain coarse-scale units from the sample structure according to the method in step (A1). The sample structure and the target structure can be the same or different. The basic parameters of each coarse-scale unit include the size of the coarse-scale unit, the size of the fine-scale unit, the arrangement of the fine-scale units, the distribution of Poisson's ratio, and the distribution of elastic modulus.
[0216] (B2) Cracks of different locations and lengths are randomly arranged within each coarse-scale unit (e.g., Figure 2 As shown, a substructure of size l × l is embedded in the center of a square region with a side length of 2l; firstly, in the region... The initiation point of the crack is randomly selected within the interior. And randomly assign the angle between the crack and the x-axis. and crack length Depending on the starting point, crack direction, and length, several relative relationships between the crack and its substructure can be obtained, such as... Figure 2 As shown in Figure a, the relationship between cracks and substructures can be categorized into three types: ① The crack and substructure do not intersect, in which case all bonds within the substructure remain intact; ② One end of the crack is located inside the substructure while the other end is located outside it, corresponding to the case where the substructure contains the crack tip; ③ The crack completely penetrates the substructure, thus dividing the substructure into two parts that do not interact with each other. Figure 2 As shown in Figure b, when the crack initiation point is located in a specific region near the substructure, two relationships will also occur: ② the substructure contains the crack tip and ④ the crack is completely contained within the substructure. This involves randomly changing the bond states of all fine-scale near-field elements in each coarse-scale element, and normalizing the bond states of all fine-scale near-field elements in all coarse-scale elements (e.g., ...). Figure 3 (as shown)
[0217] (B3) Obtain the equivalent shape function of each coarse-scale unit through calculation. The actual part that needs to be calculated ;
[0218] (B4) Determine the hyperparameters of the neural network, including the number of layers, the number of neurons in each layer, the type of activation function, the learning rate, the loss function, the batch size, the period, etc.
[0219] (B5) For example Figure 4 As shown, the bond states of all fine-scale near-field units in each coarse-scale unit after normalization are used as input, and the equivalent shape function of each coarse-scale unit is used as input. The actual part that needs to be calculated To provide theoretical output, the weight parameters and bias parameters of the neural network are continuously updated iteratively until the termination condition is met, which completes the training of the neural network. The resulting trained neural network can be reused in different overall geometry and boundary condition problems within the same coarse-scale unit construction rules and material parameters.
[0220] (A3) Calculate the global stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh. The formula used is as follows:
[0221] ;
[0222] ;
[0223] In the formula, The condensation stiffness matrix for each coarse-scale element;
[0224] The symbol represents the condensed stiffness matrix of all coarse-scale elements. Assembly;
[0225] The near-field finite element stiffness of the coarse-scale element;
[0226] (A4) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid. The calculation formula is as follows:
[0227] ;
[0228] ;
[0229] ;
[0230] In the formula, This represents the overall load vector for the fine-scale mesh.
[0231] The symbol represents the total load vector of the fine-scale mesh. The number of degrees of freedom nodes corresponding to the coarse-scale grid are extracted and assembled.
[0232] This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh;
[0233] For the element load vector of the fine-scale mesh;
[0234] It is a fine-scale unit;
[0235] The shape function matrix of the fine-scale unit. Points within a fine-scale unit;
[0236] For external force field at point External force vector at the location;
[0237] for Volume;
[0238] (A5) Solve for the global nodal displacements of the coarse-scale mesh. The formula used is as follows:
[0239] ;
[0240] (A6) Solve for the global nodal displacements of the fine-scale mesh. The formula used is as follows:
[0241]
[0242] ;
[0243] ;
[0244] In the formula, This represents the nodal displacement of the fine-scale element within the coarse-scale element;
[0245] The symbol represents the nodal displacement of all coarse-scale elements in the fine-scale elements. Assembly;
[0246] This refers to the displacement of the boundary nodes in the coarse-scale element;
[0247] The symbol represents the global nodal displacement from the coarse-scale grid. Extracting boundary node displacements from coarse-scale elements ;
[0248] (A7) Determine whether the current loading step has reached the maximum loading step requirement. If yes, proceed to step (A9); otherwise, proceed to step (A8).
[0249] (A8) Based on the global nodal displacement of the fine-scale mesh Update the bond states of all fine-scale near-field units;
[0250] If a broken bond occurs, return to step (A2);
[0251] If no broken key is generated, add a loading step and return to step (A4).
[0252] (A9) Global nodal displacement based on fine-scale mesh Output target structure The displacement cloud map is obtained, and the target structure is calculated and output based on the bond break information. The equivalent damage cloud map.
[0253] Example 2
[0254] Specific application of Example 1, wherein:
[0255] In step (A1), the quasi-static problem is set up as follows: Consider a two-dimensional flat plate problem with pre-fabricated slots on both sides (double-slotted flat plate), whose geometric dimensions and preset maximum load conditions are as follows. Figure 5 As shown in figure a, the preset maximum load condition is horizontal displacement. Vertical displacement The loading process is divided into 100 loading steps, with each loading step having a length of [missing information]. , Young's modulus and Poisson's ratio were set to... and ;
[0256] Step (A1) yields 455,625 fine-scale elements and 3,136 coarse-scale elements;
[0257] In steps (A1) and (B1), the number of fine-scale elements in the length direction of the coarse-scale element and the number of fine-scale elements in the width direction are both 15.
[0258] The final equivalent damage cloud map is as follows: Figure 5 As shown in b.
[0259] The literature (A Hybrid Local / Nonlocal Continuum Mechanics Modeling and Simulation of Fracture in Brittle Materials. Comput Model Eng Sci. 2019;121(2):399–423.) simulates the static failure problem of a two-dimensional plate, and the literature (Experimental and numerical study on the behavior of concrete subjected to biaxial tension and shear. Advanced Cement Based Materials. 1993; 1(1) 22-37.) is the corresponding experimental work. The experimental results are consistent with the calculation results of this invention in terms of crack propagation path and failure mode, thus verifying the effectiveness of the method of this invention.
[0260] Example 3
[0261] A scale-reduced near-field finite element method for structural fracture analysis is proposed, comprising the following steps:
[0262] (A1) such as Figure 1 As shown, for the target structure Perform geometric modeling, and set the material properties, boundary conditions, load conditions, near-field dynamic parameters, coarse-scale element construction rules, and computational control parameters of the target structure; among them, the computational control parameters include the analysis problem type, maximum loading step or maximum time step, and loading step size or time step size;
[0263] Then, following the original problem (dynamic problem), a finite element decomposition was performed to obtain... There are finite elements, each finite element is denoted as a fine-scale element, the mesh formed by all the fine-scale elements is denoted as a fine-scale mesh, and the computational unit formed by the interaction between any two fine-scale elements is denoted as a fine-scale near-field element.
[0264] Based on spatial neighborhood relationships, several fine-scale units are grouped together and denoted as coarse-scale units. The grid formed by all coarse-scale units is denoted as a coarse-scale grid.
[0265] (A2) Obtain the equivalent shape function for each coarse-scale unit. , , It is the identity matrix. This refers to the portion that actually needs to be calculated;
[0266] Equivalent form function of each coarse-scale unit The actual part that needs to be calculated It is obtained through prediction. The specific process is as follows: first, determine the bond states of all fine-scale near-field units in each coarse-scale unit; then, input these states into the trained neural network, which outputs the equivalent form function of the coarse-scale unit. The actual part that needs to be calculated ;
[0267] The training steps for a neural network are as follows:
[0268] (B1) Obtain coarse-scale units from the sample structure according to the method in step (A1). The sample structure and the target structure can be the same or different. The basic parameters of each coarse-scale unit include the size of the coarse-scale unit, the size of the fine-scale unit, the arrangement of the fine-scale units, the distribution of Poisson's ratio, and the distribution of elastic modulus.
[0269] (B2) Cracks of different positions and lengths are randomly arranged in each coarse-scale unit to randomly change the bond state of all fine-scale near-field units in each coarse-scale unit, and the bond state of all fine-scale near-field units in all coarse-scale units is normalized.
[0270] (B3) Obtain the equivalent shape function of each coarse-scale unit through calculation. The actual part that needs to be calculated ;
[0271] (B4) Determine the hyperparameters of the neural network, including the number of layers, the number of neurons in each layer, the type of activation function, the learning rate, the loss function, the batch size, the period, etc.
[0272] (B5) Taking the bond states of all fine-scale near-field units in each coarse-scale unit after normalization as input, and the equivalent shape function of each coarse-scale unit as input. The actual part that needs to be calculated To provide theoretical output, the weight parameters and bias parameters of the neural network are continuously updated iteratively until the termination condition is met, which completes the training of the neural network. The resulting trained neural network can be reused in different overall geometry and boundary condition problems within the same coarse-scale unit construction rules and material parameters.
[0273] (A3) Calculate the global stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh. The formula used is as follows:
[0274] ;
[0275] ;
[0276] In the formula, The condensation stiffness matrix for each coarse-scale element;
[0277] The symbol represents the condensed stiffness matrix of all coarse-scale elements. Assembly;
[0278] The near-field finite element stiffness of the coarse-scale element;
[0279] (A4) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid. The calculation formula is as follows:
[0280] ;
[0281] ;
[0282] ;
[0283] In the formula, This represents the overall load vector for the fine-scale mesh.
[0284] The symbol represents the total load vector of the fine-scale mesh. The number of degrees of freedom nodes corresponding to the coarse-scale grid are extracted and assembled.
[0285] This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh;
[0286] For the element load vector of the fine-scale mesh;
[0287] It is a fine-scale unit;
[0288] The shape function matrix of the fine-scale unit. Points within a fine-scale unit;
[0289] For external force field at point External force vector at the location;
[0290] for Volume;
[0291] (A5) Solve for the global nodal displacements of the coarse-scale mesh. The formula used is as follows:
[0292] ;
[0293] ;
[0294] ;
[0295] ;
[0296] ;
[0297] In the formula, The mass matrix of the coarse-scale grid;
[0298] This represents the overall nodal acceleration of the coarse-scale grid at time step t.
[0299] This represents the total nodal displacement of the coarse-scale grid at time step t.
[0300] This represents the total load vector of the coarse-scale grid at time step t.
[0301] The equivalent mass matrix for coarse-scale units;
[0302] The symbol represents the equivalent mass matrix of all coarse-scale units. Assembly;
[0303] This represents the overall mass matrix of all fine-scale cells within the coarse-scale cell;
[0304] This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh;
[0305] The mass matrix of the fine-scale unit;
[0306] It is a fine-scale unit;
[0307] The shape function matrix of the fine-scale unit. Points within a fine-scale unit;
[0308] for The volumetric mass density;
[0309] for Volume;
[0310] (A6) Solve for the global nodal displacements of the fine-scale mesh. The formula used is as follows:
[0311]
[0312] ;
[0313] ;
[0314] In the formula, This represents the nodal displacement of the fine-scale element within the coarse-scale element;
[0315] The symbol represents the nodal displacement of all coarse-scale elements in the fine-scale elements. Assembly;
[0316] This refers to the displacement of the boundary nodes in the coarse-scale element;
[0317] The symbol represents the global nodal displacement from the coarse-scale grid. Extracting boundary node displacements from coarse-scale elements ;
[0318] (A7) Determine whether the current time step has reached the maximum time step requirement. If yes, proceed to step (A9); otherwise, proceed to step (A8).
[0319] (A8) Based on the global nodal displacement of the fine-scale mesh Update the bond states of all fine-scale near-field units;
[0320] If a broken bond occurs, return to step (A2);
[0321] If no broken key is generated, increase the time step and return to step (A4).
[0322] (A9) Global nodal displacement based on fine-scale mesh Output target structure The displacement cloud map is obtained, and the target structure is calculated and output based on the bond break information. The equivalent damage cloud map should be output. In addition, the acceleration of the fine-scale mesh should also be output. Acceleration of fine-scale mesh acceleration from coarse-scale grid It was obtained through a one-step downscaling calculation.
[0323] Example 4
[0324] Specific application of Example 3, wherein:
[0325] In step (A1), the dynamic problem setting is as follows: Consider a two-dimensional plate problem with a rectangular notch in the middle (the plate containing the rectangular notch), whose geometric dimensions and preset maximum load conditions are as follows. Figure 6 As shown in Figure a, internal pressure is applied at the central notch. A total of 20,000 time steps were performed. In the first 1,000 time steps, the internal pressure was divided into 1,000 parts and applied gradually. The volumetric mass density in the dynamic problem was set to ρ=1000. Young's modulus and Poisson's ratio were set to... and ;
[0326] In steps (A1) and (B1), the number of fine-scale elements in the length direction of the coarse-scale element and the number of fine-scale elements in the width direction are both 15.
[0327] Step (A1) yields 1,447,209 fine-scale elements and 10,000 coarse-scale elements;
[0328] The final equivalent damage cloud map is as follows: Figure 6 As shown in b.
[0329] Although this embodiment is a numerical construct example, its failure mode is similar to the experimental results of concrete slab impact failure in existing literature, such as the literature (Peridynamic modeling and simulations on concrete dynamic failure and penetration subjected to impact loadings, Engineering Fracture Mechanics, Volume 259, 2022, 108135.), which can be used to indirectly verify the rationality of the method of the present invention in dynamic fracture problems.
[0330] Example 5
[0331] Specific application of Example 3, wherein:
[0332] In step (A1), the dynamic problem setting is as follows: Consider a two-dimensional plate tensile dynamic fracture problem containing five random vertical pre-fabricated cracks, with the following geometric dimensions and load conditions: Figure 7 As shown; the plate is 759 mm long and 255 mm wide. Tensile loads of T = 3400 Pa are applied to both sides of the plate throughout the calculation. The bulk density in the dynamic problem is set to ρ = 1000. Young's modulus and Poisson's ratio are set to... and The time step was ∆t=0.002, the total calculation time was 20, and a total of 10,000 time steps were performed; among them, five vertical prefabricated cracks were randomly distributed inside the plate, such as... Figure 7 As shown, the center coordinates of crack ① are (76.7, 164.3), and the crack length is 97.3; the center coordinates of crack ② are (184.1, 83.5), and the crack length is 111.2; the center coordinates of crack ③ are (338.2, 165.1), and the crack length is 85.7; the center coordinates of crack ④ are (521.3, 90.3), and the crack length is 102.7; the center coordinates of crack ⑤ are (652.8, 148.5), and the crack length is 93.5.
[0333] Step (A1) yields 193,545 fine-scale elements;
[0334] If the number of fine-scale units in the length direction and the number of fine-scale units in the width direction of the coarse-scale unit are both 10 in steps (A1) and (B1), 3888 coarse-scale units are obtained.
[0335] If the number of fine-scale units in the length direction and the number of fine-scale units in the width direction of the coarse-scale unit are both 15 in steps (A1) and (B1), 1323 coarse-scale units are obtained.
[0336] To illustrate the computational accuracy and efficiency of the method of the present invention, the results obtained in Example 5 are compared with the computational results of the original PeriFEM method, and the results are as follows:
[0337] Original PeriFEM method: The original PeriFEM method in the patent application with publication number CN113705040A was used to calculate the same two-dimensional plate tensile dynamic fracture problem. The calculation region was discretized into 193545 (i.e. 759×255) fine-scale elements.
[0338] The equivalent damage cloud maps obtained by the two methods at time step t = 20 are compared, and the results are as follows: Figure 8As shown in the figure, in this dynamic fracture problem, the propagation and reflection of stress waves near the crack can cause the crack to deflect and propagate. The original PeriFEM method, the method of the present invention using 10×10 coarse-scale elements, and the method of the present invention using 15×15 coarse-scale elements can all capture the crack propagation and crack deflection phenomena. Moreover, the crack propagation path obtained by the method of the present invention is basically consistent with the result obtained by the original PeriFEM method, indicating that the method of the present invention can still maintain the effective prediction ability of the dynamic crack propagation process while reducing the number of computational elements.
[0339] To ensure the comparability of computation time comparisons, the original PeriFEM method, the method of this invention using 10×10 coarse-scale cells, and the method of this invention using 15×15 coarse-scale cells were all performed on the same hardware platform. In this embodiment, the hardware platform used for computation time testing includes an Intel Core™ i7-12700F processor, 64GB of memory, and an NVIDIA RTX 3070 graphics card with 8GB of video memory. This hardware platform is only used to illustrate the test conditions of the example in this embodiment and does not constitute a limitation on the implementation of this invention.
[0340] The above-mentioned computation time is compared to, for example Figure 9 As shown, the total computation time of the original PeriFEM method is 527 minutes, while the total computation time of the present invention using 10×10 coarse-scale elements is 299 minutes, and the total computation time using 15×15 coarse-scale elements is 255 minutes. Compared with the original PeriFEM method, the total computation time of the present invention is reduced by 43.3% when using 10×10 coarse-scale elements and by 51.6% when using 15×15 coarse-scale elements. It can be seen that the method of the present invention can significantly reduce the number of computational elements and the total computation time while maintaining similar dynamic fracture simulation results to the original PeriFEM method, proving the effectiveness and computational efficiency advantages of the present invention in dynamic fracture problems.
[0341] The method of this invention is not an independent mathematical calculation method, but a calculation method for analyzing the failure of specific engineering structures. Its technical solution has been combined with specific application scenarios in Embodiment 2 (quasi-static problem), Embodiments 4 and 5 (dynamic problem).
[0342] Specifically:
[0343] Example 2 corresponds to a quasi-static failure analysis scenario (such as tensile failure of a double-slotted plate). Its calculation process includes structural modeling, mesh generation, load application, fracture evolution solution and result output, which reflects the application process of the method of the present invention in actual engineering structural analysis.
[0344] Example 4 corresponds to the dynamic failure analysis scenario of structural impact / explosion (such as the dynamic fracture problem of a plate with a notch under internal pressure), and Example 5 corresponds to the dynamic tensile failure analysis scenario of structural (such as the dynamic cracking problem of a plate with a crack under tension). The calculation process includes the construction of dynamic equations, time step solution, fracture update and result output, which reflects the application of the method of the present invention in dynamic engineering problems.
[0345] Examples 6-7
[0346] A scale-reduced near-field finite element method for structural fracture analysis is proposed, which is basically the same as in Examples 1 and 3, except that in step (A2), the equivalent shape function of each coarse-scale element is... The actual part that needs to be calculated It was obtained through calculation, and the calculation steps are as follows:
[0347] (A21) Calculate the near-field finite element stiffness of coarse-scale elements. The formula is as follows:
[0348] ;
[0349] ;
[0350] ;
[0351] ;
[0352] In the formula, This represents the total number of fine-scale near-field elements within the coarse-scale element;
[0353] This is the nodal degree of freedom transformation matrix for the fine-scale near-field elements within the coarse-scale element;
[0354] is the element stiffness matrix of the fine-scale near-field element in the coarse-scale element;
[0355] and These represent two fine-scale units that constitute the fine-scale near-field unit;
[0356] Let be the shape function difference matrix of the fine-scale near-field elements in the coarse-scale element, where x and x represent points within two fine-scale cells that constitute a fine-scale near-field cell, respectively;
[0357] It is the micromodulus matrix;
[0358] and Represent and the volume of x;
[0359] and These are the shape function matrices of the two fine-scale units that constitute the fine-scale near-field unit;
[0360] For near-field dynamic bonds;
[0361] for The micromodulus coefficient;
[0362] for At time step The state at time t is either 0 or 1, where express Fracture, 1 indicates Not broken;
[0363] and for The components of a vector;
[0364] (A22) Near-field finite element stiffness of coarse-scale elements Expanding on the expression, the result is as follows:
[0365] ;
[0366] In the formula, , , and They are respectively The sub-blocks, where subscripts b and i represent the boundary node and internal node of the coarse-scale element, respectively, are... ;
[0367] Here, is the weighting function, representing the number of times a fine-scale cell appears in a coarse-scale cell;
[0368] (A23) Calculate the equivalent shape function for each coarse-scale element. The actual part that needs to be calculated The formula is as follows:
[0369] .
[0370] To illustrate that this invention can be used to solve the fracture response of engineering structures under different working conditions, the methods of Examples 1 and 3 are used as core calculation modules and embedded into specific engineering analysis processes to be associated with different application scenarios. The specific process is as follows:
[0371] Example 8
[0372] Example 1 describes the implementation of a scale-reduced near-field finite element method for structural fracture analysis in commercial finite element software. The steps are as follows:
[0373] (E1) In commercial finite element software and its preprocessing module or secondary development interface, establish the geometric model of the target structure, set the material properties, boundary conditions, load conditions, near-field dynamic parameters and calculation control parameters, and perform fine-scale meshing on the geometric model to obtain fine-scale elements and fine-scale meshes. The calculation unit formed by the interaction between any two fine-scale elements is called a fine-scale near-field element.
[0374] Subsequently, coarse-scale cells are constructed based on the fine-scale grid to form a coarse-scale grid;
[0375] (E2) Obtain the equivalent shape function for each coarse-scale unit;
[0376] (E3) The process of calculating the overall stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh is written into the user element or user subroutine of the commercial finite element software, and the calculation of the overall stiffness matrix of the coarse-scale mesh is completed by the commercial finite element software.
[0377] (E4) Calculate the total load vector of the coarse-scale mesh based on the coarse-scale mesh, and complete the numerical solution in a commercial finite element solver to obtain the total nodal displacement of the coarse-scale mesh;
[0378] (E5) Recover the total nodal displacements of the fine-scale mesh based on the total nodal displacements of the coarse-scale mesh;
[0379] (E6) Determine whether the current loading step or time step has met the preset requirements, and update the bond state of the fine-scale near-field element according to the overall node displacement of the fine-scale mesh; if there are new broken bonds, return to step (E2) and reacquire the equivalent shape function of the affected coarse-scale element; if there are no new broken bonds and the preset loading step or time step requirements have not been met, proceed to the next loading step or the next time step to continue the calculation, i.e., return to step (E4).
[0380] (E7) When the current loading step or time step reaches the preset requirements, the displacement cloud map, equivalent damage cloud map and crack propagation process of the target structure are output in the post-processing module of the commercial finite element software.
[0381] The above method yields the following analytical results:
[0382] Displacement field, strain field, and stress distribution of the structure during loading;
[0383] The location of crack initiation, propagation path, and penetration process;
[0384] Structural damage evolution process and damage distribution cloud map;
[0385] The variation law of structural fracture response under different loading conditions (such as displacement loading, force loading, etc.);
[0386] Quantitative analysis results of fracture behavior in key areas (such as stress concentration areas and notch areas);
[0387] The overall load-bearing capacity degradation process and failure mode identification results of the structure;
[0388] Results can be used for engineering evaluation (such as safety margin analysis, ultimate load estimation, etc.).
[0389] A comparison of the above application method with traditional finite element methods and existing near-field finite element methods reveals that the present invention has the following advantages:
[0390] Compatible with existing commercial finite element software systems: The method of this invention can be embedded into existing finite element platforms through user-defined elements or user subroutines without changing the original solver structure, and has good engineering adaptability;
[0391] Achieving unified simulation of crack initiation and propagation: Compared with the traditional finite element method that relies on preset crack paths or additional fracture criteria, this invention can naturally describe the entire process of crack initiation and propagation, improving the physical rationality of fracture simulation.
[0392] Significantly reduce computational scale and cost: By using the scale reduction method, the large-scale nonlocal computation originally performed at the fine scale is transformed into an equivalent computation at the coarse scale, effectively reducing the number of degrees of freedom;
[0393] Improve computational efficiency: By using neural networks to predict equivalent shape functions, the problem of repeatedly solving local substructure problems can be avoided, thereby significantly shortening the overall computation time;
[0394] Enhanced ability to analyze complex structures: capable of handling porous structures, notched structures, and complex geometric models, making it more applicable in fracture analysis of engineering structures;
[0395] Enhance engineering analysis and design capabilities: While maintaining computational accuracy, transform fracture analysis from a "high-cost analysis tool" into a "routine analysis method that can be used for engineering design and optimization".
[0396] Example 9
[0397] Example 1 describes a method for accelerating near-field finite element analysis based on scale reduction for structural fracture analysis, implemented in material microstructure fracture simulation analysis using X-ray micro-CT data. The steps are as follows:
[0398] (F1) Obtain X-ray micro-CT scan data of material samples and reconstruct the images of X-ray micro-CT scan data to obtain two-dimensional or three-dimensional image data of material microstructure.
[0399] Among them, two-dimensional or three-dimensional image data of material microstructure are used to characterize the pores, inclusions, cracks, different material phases or other microstructural features inside the material;
[0400] (F2) Based on two-dimensional or three-dimensional image data of material microstructure, establish a geometric model of the target material microstructure, and identify material phases, pore regions, inclusion regions or initial defect regions according to the two-dimensional or three-dimensional image data of material microstructure, and assign corresponding material properties to different regions according to the identification results;
[0401] (F3) Set the computational conditions required for the simulation analysis of material microstructure fracture, and construct fine-scale and coarse-scale meshes based on the geometric model;
[0402] The calculation conditions include material properties, boundary conditions, load conditions, near-field dynamic parameters, bond breakage criteria, maximum loading step or maximum time step, and loading step size or time step size.
[0403] The geometric model is divided into several fine-scale elements, and all the fine-scale elements constitute a fine-scale mesh. The computational unit formed by the interaction between any two fine-scale elements is called a fine-scale near-field element.
[0404] Subsequently, based on spatial neighborhood relationships, several fine-scale units are grouped into coarse-scale units, and all coarse-scale units constitute a coarse-scale grid.
[0405] (F4) Obtain the equivalent shape function for each coarse-scale unit;
[0406] (F5) Calculate the overall stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh;
[0407] (F6) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid;
[0408] (F7) Under the given load conditions or boundary conditions, solve for the total nodal displacement of the coarse-scale mesh, and solve for the total nodal displacement of the fine-scale mesh based on the total nodal displacement of the coarse-scale mesh.
[0409] (F8) Determine and update the bond state of the fine-scale near-field elements based on the total nodal displacement of the fine-scale mesh;
[0410] If a break occurs, return to the corresponding coarse-scale unit and re-obtain the equivalent shape function of that coarse-scale unit, i.e., return to step (F4).
[0411] If no broken bonds are generated and the current analysis is a quasi-static fracture analysis, then add a loading step and return to step (F6).
[0412] If no broken bonds are generated and the current analysis is a dynamic fracture analysis, then increase the time step and return to step (F6).
[0413] (F9) When the current loading step or time step reaches the preset requirements, output the simulation analysis results of material microstructure fracture;
[0414] The simulation analysis results include displacement cloud maps of the material microstructure, equivalent damage cloud maps, crack initiation locations, crack propagation paths, and the analysis results of the influence of pores, inclusions, or initial defects on the material fracture behavior.
[0415] The above method yields the following analytical results:
[0416] Crack initiation location and propagation path at the microstructure scale of materials;
[0417] Internal damage evolution process and damage distribution cloud map of the material;
[0418] Displacement field and strain response of materials under external load;
[0419] Quantitative analysis results of the influence of microstructural defects (such as pores, inclusions, etc.) on fracture behavior.
[0420] A comparison of the above application method with traditional fine-scale numerical simulation methods based on CT data reveals that the present invention has the following advantages:
[0421] Significantly reduce computational scale: By using a scale reduction method, the high-resolution CT model is transformed into a coarse-scale computational problem, effectively reducing the computational degrees of freedom;
[0422] Preserving microstructure information and fracture characteristics: Using equivalent shape functions to retain fine-scale near-field interactions and damage information, achieving a balance between accuracy and efficiency;
[0423] Improve computational efficiency: By predicting equivalent shape functions through neural networks, the problem of repeatedly solving local substructure problems is avoided, significantly reducing computation time;
[0424] Enhanced engineering applicability: It transforms fracture analysis based on real CT microstructures from "computable" to "efficiently computable," making it valuable for engineering applications and promotion.
[0425] Example 10
[0426] Example 3 describes the implementation of a scale-reduced near-field finite element method for structural fracture analysis in the simulation analysis of dynamic damage caused by structural impact / explosion. The steps are as follows:
[0427] (G1) Establish the geometric model of the target structure to be analyzed and set the calculation conditions required for dynamic failure simulation analysis;
[0428] The target structures include perforated structures, notched structures, plate structures, beam structures, shell structures, or other engineering structures that are subjected to impact loads or explosive loads.
[0429] The calculation conditions include material properties, bulk density, boundary conditions, impact or explosive load conditions, load time history, time step, total calculation time, near-field dynamic parameters, and bond breakage criteria;
[0430] Impact loads or explosive loads include internal pressure loads, impact velocity boundary loads, impact force loads, pressure time history loads, or equivalent explosive impact loads;
[0431] (G2) The geometric model is divided into finite element subdivisions to obtain several fine-scale elements. All fine-scale elements constitute a fine-scale mesh. The computational element formed by the interaction between any two fine-scale elements is called a fine-scale near-field element.
[0432] (G3) Based on spatial neighborhood relationships, several fine-scale units are grouped into coarse-scale units, and all coarse-scale units constitute a coarse-scale grid;
[0433] This step is used to convert some degrees of freedom in a fine-scale mesh into equivalent degrees of freedom in a coarse-scale mesh, thereby reducing the computational scale of dynamic problems;
[0434] (G4) Obtain the equivalent shape function for each coarse-scale unit;
[0435] (G5) Calculate the global stiffness matrix and global mass matrix of the coarse-scale mesh;
[0436] (G6) Calculate the overall load vector for the coarse-scale mesh;
[0437] Among them, the overall stiffness matrix is used to characterize the dynamic stiffness response of the structure, the overall mass matrix is used to characterize the inertial effect of the structure, and the overall load vector is used to characterize the effect of impact load or explosive load at the current time step.
[0438] (G7) Under the time-stepping framework, solve for the global nodal displacement and global nodal acceleration of the coarse-scale mesh at the current time step; wherein, the global nodal displacement and global nodal acceleration are obtained by solving the dynamic equations of the coarse-scale mesh;
[0439] (G8) Based on the total nodal displacement of the coarse-scale mesh, recover the total nodal displacement of the fine-scale mesh, and update the bond state of the fine-scale near-field elements based on the total nodal displacement of the fine-scale mesh.
[0440] Specifically, when a bond in a fine-scale near-field cell satisfies the bond breakage criterion, the bond state is updated to the broken state.
[0441] (G9) Determine whether the current time step has reached the preset maximum time step or the preset total calculation time;
[0442] If the preset maximum time step or preset total computation time is not reached, determine whether there are any new broken bonds based on the bond state of the fine-scale near-field unit.
[0443] If a new break occurs, return to step (G4), reacquire the equivalent shape function of the affected coarse-scale element, and continue with subsequent calculations;
[0444] If no new key breaks are added, proceed to the next time step and return to step (G6).
[0445] If the preset maximum time step or preset total calculation time is reached, proceed to step (G10).
[0446] (G10) Outputs the dynamic damage simulation analysis results of the target structure under impact load or explosive load;
[0447] The results of dynamic failure simulation analysis include displacement contour maps, acceleration contour maps, equivalent damage contour maps, crack initiation locations, crack propagation paths, failure ranges, and fragmentation modes of the target structure at each time step.
[0448] The above method yields the following analytical results:
[0449] Dynamic displacement field and acceleration response of a structure under impact or explosion;
[0450] Crack initiation time, propagation path, and penetration process;
[0451] Structural damage evolution process and damage distribution cloud map;
[0452] Structural fragmentation mode and fragment distribution characteristics (such as fracture area, damage range, etc.);
[0453] Comparative analysis results of the structural impact resistance performance under different load conditions.
[0454] A comparison of the above application method with existing near-field dynamics methods reveals the following advantages of this invention:
[0455] It can naturally simulate the dynamic fracture process: without the need to preset crack paths or additional fracture criteria, it can achieve unified simulation of crack initiation and propagation;
[0456] Significantly reduce the computational cost of large-scale dynamic problems: By reducing the system's degrees of freedom through scale reduction methods, large-scale impact / explosion problems that were originally difficult to solve can be transformed into solvable problems;
[0457] Improve computational efficiency: By using neural networks to predict equivalent shape functions, high-cost local calculations are avoided at each time step, significantly accelerating the dynamic simulation process;
[0458] Suitable for complex structures and complex load conditions: It can handle dynamic problems with porous structures, notched structures and complex boundary conditions, and has good engineering applicability;
[0459] Achieving a balance between accuracy and efficiency: While maintaining the ability to simulate near-field dynamic fractures, a significant improvement in computational efficiency is achieved.
Claims
1. A scale-reduced near-field finite element method for accelerating structural fracture analysis, characterized in that, Includes the following steps: (A1) Target Structure Perform geometric modeling, and set the material properties, boundary conditions, load conditions, near-field dynamic parameters, coarse-scale element construction rules, and computational control parameters of the target structure; among them, the computational control parameters include the analysis problem type, maximum loading step or maximum time step, and loading step size or time step size; Then, by performing finite element decomposition according to the original problem, we obtain... There are finite elements, each finite element is denoted as a fine-scale element, the mesh formed by all the fine-scale elements is denoted as a fine-scale mesh, and the computational unit formed by the interaction between any two fine-scale elements is denoted as a fine-scale near-field element. Based on spatial neighborhood relationships, several fine-scale units are grouped together and denoted as coarse-scale units. The grid formed by all coarse-scale units is denoted as a coarse-scale grid. (A2) Obtain the equivalent shape function for each coarse-scale unit. , , It is the identity matrix. This refers to the portion that actually needs to be calculated; (A3) Calculate the global stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh. ; (A4) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid. ; (A5) Solve for the global nodal displacements of the coarse-scale mesh. ; (A6) Solve for the global nodal displacements of the fine-scale mesh. ; (A7) Determine whether the current loading step or time step has reached the maximum loading step or time step requirement. If yes, proceed to step (A9); otherwise, proceed to step (A8). (A8) Based on the global nodal displacement of the fine-scale mesh Update the bond states of all fine-scale near-field units; If a broken bond occurs, return to step (A2); If no broken keys are generated and the original problem is a quasi-static problem, then add a loading step and return to step (A4). If no key breaks occur and the original problem is a dynamic problem, then add a time step and return to step (A4). (A9) Global nodal displacement based on fine-scale mesh Output target structure The displacement cloud map is obtained, and the target structure is calculated and output based on the bond break information. The equivalent damage cloud map.
2. The method for accelerating near-field finite element analysis of structural fracture analysis according to claim 1, characterized in that, In step (A2), the equivalent shape function of each coarse-scale unit The actual part that needs to be calculated It was obtained through calculation, and the calculation steps are as follows: (A21) Calculate the near-field finite element stiffness of coarse-scale elements. The formula is as follows: ; ; ; ; In the formula, This represents the total number of fine-scale near-field elements within the coarse-scale element; This is the nodal degree of freedom transformation matrix for the fine-scale near-field elements within the coarse-scale element; is the element stiffness matrix of the fine-scale near-field element in the coarse-scale element; and These represent two fine-scale units that constitute the fine-scale near-field unit; Let be the shape function difference matrix of the fine-scale near-field elements in the coarse-scale element, where x and x represent points within two fine-scale cells that constitute a fine-scale near-field cell, respectively; It is the micromodulus matrix; and Represent and the volume of x; and These are the shape function matrices of the two fine-scale units that constitute the fine-scale near-field unit; For near-field dynamic bonds; for The micromodulus coefficient; for At time step The state at time t is either 0 or 1, where express Fracture, 1 indicates Not broken; and for The components of a vector; (A22) Near-field finite element stiffness of coarse-scale elements Expanding on the expression, the result is as follows: ; In the formula, , , and They are respectively The sub-blocks, where subscripts b and i represent the boundary node and internal node of the coarse-scale element, respectively, are... ; Here, is the weighting function, representing the number of times a fine-scale cell appears in a coarse-scale cell; (A23) Calculate the equivalent shape function for each coarse-scale element. The actual part that needs to be calculated The formula is as follows: 。 3. The method for accelerating near-field finite element analysis of structural fracture analysis according to claim 1, characterized in that, In step (A2), the equivalent shape function of each coarse-scale unit The actual part that needs to be calculated It is obtained through prediction. The specific process is as follows: first, determine the bond states of all fine-scale near-field units in each coarse-scale unit; then, input these states into the trained neural network, which outputs the equivalent form function of the coarse-scale unit. The actual part that needs to be calculated ; The training steps for a neural network are as follows: (B1) Obtain coarse-scale units from the sample structure using the method in step (A1); (B2) Cracks of different positions and lengths are randomly arranged in each coarse-scale unit to randomly change the bond state of all fine-scale near-field units in each coarse-scale unit, and the bond state of all fine-scale near-field units in all coarse-scale units is normalized. (B3) Obtain the equivalent shape function of each coarse-scale unit through calculation. The actual part that needs to be calculated ; (B4) Determine the hyperparameters of the neural network; (B5) Taking the bond states of all fine-scale near-field units in each coarse-scale unit after normalization as input, and the equivalent shape function of each coarse-scale unit as input. The actual part that needs to be calculated To achieve the theoretical output, the weight and bias parameters of the neural network are continuously updated iteratively until the termination condition is met, thus completing the training of the neural network.
4. The method for accelerating near-field finite element analysis of structural fracture analysis by reducing the scale as described in claim 1, characterized in that, In step (A3), the global stiffness matrix of the coarse-scale mesh is calculated. The formula used is as follows: ; ; In the formula, The condensation stiffness matrix for each coarse-scale element; The symbol represents the condensed stiffness matrix of all coarse-scale elements. Assembly; The near-field finite element stiffness is represented by the coarse-scale element.
5. The method for accelerating near-field finite element analysis of structural fracture analysis by reducing the scale as described in claim 1, characterized in that, In step (A4), the global load vector of the coarse-scale mesh The calculation formula is as follows: ; ; ; In the formula, This represents the overall load vector for the fine-scale mesh. The symbol represents the total load vector of the fine-scale mesh. The number of degrees of freedom nodes corresponding to the coarse-scale grid are extracted and assembled. This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh; For the element load vector of the fine-scale mesh; It is a fine-scale unit; The shape function matrix of the fine-scale unit. Points within a fine-scale unit; For external force field at point External force vector at the location; for The volume.
6. The method for accelerating near-field finite element analysis of structural fracture analysis by reducing the scale as described in claim 1, characterized in that, In step (A5), if the original problem is a quasi-static problem, then solve for the global nodal displacements of the coarse-scale mesh. The formula used is as follows: ; If the original problem is dynamic, then solve for the global nodal displacements of the coarse-scale mesh. The formula used is as follows: ; ; ; ; ; In the formula, The mass matrix of the coarse-scale grid; This represents the overall nodal acceleration of the coarse-scale grid at time step t. This represents the total nodal displacement of the coarse-scale grid at time step t. This represents the total load vector of the coarse-scale grid at time step t. The equivalent mass matrix for coarse-scale units; The symbol represents the equivalent mass matrix of all coarse-scale units. Assembly; This represents the overall mass matrix of all fine-scale cells within the coarse-scale cell; This is the transformation matrix for the nodal degrees of freedom of the fine-scale mesh; The mass matrix of the fine-scale unit; It is a fine-scale unit; The shape function matrix of the fine-scale unit. Points within a fine-scale unit; for The volumetric mass density; for The volume.
7. The method for accelerating near-field finite element analysis of structural fracture analysis by reducing the scale as described in claim 1, characterized in that, In step (A6), the global nodal displacements of the fine-scale mesh are solved. The formula used is as follows: ; ; ; In the formula, This represents the nodal displacement of the fine-scale element within the coarse-scale element; The symbol represents the nodal displacement of all coarse-scale elements in the fine-scale elements. Assembly; This refers to the displacement of the boundary nodes in the coarse-scale element; The symbol represents the global nodal displacement from the coarse-scale grid. Extracting boundary node displacements from coarse-scale elements .
8. The method for accelerating near-field finite element analysis of structural fracture analysis by reducing the scale as described in claim 1, characterized in that, In step (A9), if the original problem is a dynamic problem, the acceleration of the fine-scale mesh should also be output. Acceleration of fine-scale mesh acceleration from coarse-scale grid It was obtained through a one-step downscaling calculation.
9. The method for implementing the scale-reduced near-field finite element method for structural fracture analysis as described in claim 1 in commercial finite element software, characterized in that, Includes the following steps: (E1) In commercial finite element software and its preprocessing module or secondary development interface, establish the geometric model of the target structure, set the material properties, boundary conditions, load conditions, near-field dynamic parameters and calculation control parameters, and perform fine-scale meshing on the geometric model to obtain fine-scale elements and fine-scale meshes. The calculation unit formed by the interaction between any two fine-scale elements is called a fine-scale near-field element. Subsequently, coarse-scale cells are constructed based on the fine-scale grid to form a coarse-scale grid; (E2) Obtain the equivalent shape function for each coarse-scale unit; (E3) The process of calculating the overall stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh is written into the user-defined element or user subroutine of the commercial finite element software, and the calculation of the overall stiffness matrix of the coarse-scale mesh is completed by the commercial finite element software. (E4) Calculate the total load vector of the coarse-scale mesh based on the coarse-scale mesh, and complete the numerical solution in a commercial finite element solver to obtain the total nodal displacement of the coarse-scale mesh; (E5) Recover the total nodal displacements of the fine-scale mesh based on the total nodal displacements of the coarse-scale mesh; (E6) Determine whether the current loading step or time step has met the preset requirements, and update the bond state of the fine-scale near-field element according to the overall node displacement of the fine-scale mesh; if there are new broken bonds, return to step (E2) and reacquire the equivalent shape function of the affected coarse-scale element; if there are no new broken bonds and the preset loading step or time step requirements have not been met, proceed to the next loading step or the next time step to continue the calculation, i.e., return to step (E4). (E7) When the current loading step or time step reaches the preset requirements, the displacement cloud map, equivalent damage cloud map and crack propagation process of the target structure are output in the post-processing module of the commercial finite element software.
10. The method for implementing the scale-reduced near-field finite element method for structural fracture analysis as described in claim 1 in the simulation analysis of material microstructure fracture based on X-ray micro-CT data, characterized in that... Includes the following steps: (F1) Obtain X-ray micro-CT scan data of material samples and reconstruct the images of X-ray micro-CT scan data to obtain two-dimensional or three-dimensional image data of material microstructure. (F2) Based on two-dimensional or three-dimensional image data of material microstructure, establish a geometric model of the target material microstructure, and identify material phases, pore regions, inclusion regions or initial defect regions according to the two-dimensional or three-dimensional image data of material microstructure, and assign corresponding material properties to different regions according to the identification results; (F3) Set the computational conditions required for the simulation analysis of material microstructure fracture, and construct fine-scale and coarse-scale meshes based on the geometric model; The calculation conditions include material properties, boundary conditions, load conditions, near-field dynamic parameters, bond breakage criteria, maximum loading step or maximum time step, and loading step size or time step size. The geometric model is divided into several fine-scale elements, and all the fine-scale elements constitute a fine-scale mesh. The computational unit formed by the interaction between any two fine-scale elements is called a fine-scale near-field element. Subsequently, based on spatial neighborhood relationships, several fine-scale units are grouped into coarse-scale units, and all coarse-scale units constitute a coarse-scale grid. (F4) Obtain the equivalent shape function for each coarse-scale unit; (F5) Calculate the overall stiffness matrix of the coarse-scale mesh based on the coarse-scale mesh; (F6) Calculate the overall load vector of the coarse-scale grid based on the coarse-scale grid; (F7) Under the given load conditions or boundary conditions, solve for the total nodal displacement of the coarse-scale mesh, and solve for the total nodal displacement of the fine-scale mesh based on the total nodal displacement of the coarse-scale mesh. (F8) Determine and update the bond state of the fine-scale near-field elements based on the total nodal displacement of the fine-scale mesh; If a break occurs, return to the corresponding coarse-scale unit and re-obtain the equivalent shape function of that coarse-scale unit, i.e., return to step (F4). If no broken bonds are generated and the current analysis is a quasi-static fracture analysis, then add a loading step and return to step (F6). If no broken bonds are generated and the current analysis is a dynamic fracture analysis, then increase the time step and return to step (F6). (F9) When the current loading step or time step reaches the preset requirements, output the simulation analysis results of material microstructure fracture; The simulation analysis results include displacement cloud maps of the material microstructure, equivalent damage cloud maps, crack initiation locations, crack propagation paths, and the analysis results of the influence of pores, inclusions, or initial defects on the material fracture behavior.
11. The method for implementing the scale-reduced near-field finite element calculation acceleration method for structural fracture analysis as described in claim 1 in the simulation analysis of dynamic damage from structural impact / explosion, characterized in that, Includes the following steps: (G1) Establish the geometric model of the target structure to be analyzed and set the calculation conditions required for dynamic failure simulation analysis; The target structures include perforated structures, notched structures, plate structures, beam structures, shell structures, or other engineering structures that are subjected to impact loads or explosive loads. The calculation conditions include material properties, bulk density, boundary conditions, impact or explosive load conditions, load time history, time step, total calculation time, near-field dynamic parameters, and bond breakage criteria; Impact loads or explosive loads include internal pressure loads, impact velocity boundary loads, impact force loads, pressure time history loads, or equivalent explosive impact loads; (G2) The geometric model is divided into finite element subdivisions to obtain several fine-scale elements. All fine-scale elements constitute a fine-scale mesh. The computational element formed by the interaction between any two fine-scale elements is called a fine-scale near-field element. (G3) Based on spatial neighborhood relationships, several fine-scale units are grouped into coarse-scale units, and all coarse-scale units constitute a coarse-scale grid; (G4) Obtain the equivalent shape function for each coarse-scale unit; (G5) Calculate the global stiffness matrix and global mass matrix of the coarse-scale mesh; (G6) Calculate the overall load vector for the coarse-scale mesh; (G7) Under the time-stepping framework, solve for the global nodal displacement and global nodal acceleration of the coarse-scale mesh at the current time step; (G8) Based on the total nodal displacement of the coarse-scale mesh, recover the total nodal displacement of the fine-scale mesh, and update the bond state of the fine-scale near-field elements based on the total nodal displacement of the fine-scale mesh. Specifically, when a bond in a fine-scale near-field cell satisfies the bond breakage criterion, the bond state is updated to the broken state. (G9) Determine whether the current time step has reached the preset maximum time step or the preset total calculation time; If the preset maximum time step or preset total computation time is not reached, determine whether there are any new broken bonds based on the bond state of the fine-scale near-field unit. If a new break occurs, return to step (G4), reacquire the equivalent shape function of the affected coarse-scale element, and continue with subsequent calculations; If no new key breaks are added, proceed to the next time step and return to step (G6). If the preset maximum time step or preset total calculation time is reached, proceed to step (G10). (G10) Outputs the dynamic damage simulation analysis results of the target structure under impact load or explosive load; The results of dynamic failure simulation analysis include displacement contour maps, acceleration contour maps, equivalent damage contour maps, crack initiation locations, crack propagation paths, failure ranges, and fragmentation modes of the target structure at each time step.
Citation Information
Patent Citations
Near-field finite element method for structural damage analysis and implementation method of near-field finite element method in commercial software
CN113705040A
