A method and system for determining grid structure parameters

By optimizing the solution of the linear equations for the parameters of the space frame structure using a block-to-pipeline parallel method, the problem of slow computation speed in existing technologies is solved, enabling more efficient simulation of space frame structures and improving engineering efficiency.

CN114595547BActive Publication Date: 2024-11-01PEKING UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202011430455.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-07
Publication Date
2024-11-01
Estimated Expiration
2040-12-07

AI Technical Summary

Technical Problem

In the simulation of existing space frame structures, the calculation speed of solving the linear equations is slow, which cannot fully utilize the parallel computing capabilities of multi-core CPUs, resulting in low engineering efficiency.

Method used

A block-to-pipeline parallel approach is adopted, which optimizes the solution process of linear equations through block parallel and pipeline parallel stages. This includes designing a block task pool, a ancestor node elimination schedule, and a pipeline task pool to improve parallel computing efficiency.

Benefits of technology

It significantly improves the calculation speed of space frame structure parameters, shortens the time ratio of the simulation process, and improves the efficiency of engineering applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114595547B_ABST
    Figure CN114595547B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for determining parameters of a grid structure. The method comprises: mechanically abstracting the initial parameters of the grid structure and establishing a CAE model; meshing the CAE model to obtain a unit stiffness matrix and node forces; determining the overall stiffness matrix and overall forces based on the unit stiffness matrix and the node forces; constructing a linear equation group based on the overall stiffness matrix and the overall forces; solving the linear equation group using a block-to-pipeline parallel method to obtain a displacement; the block-to-pipeline parallel method includes a block parallel stage and a pipeline parallel stage; determining the stress and strain of the structure based on the displacement through an interpolation relationship; and calibrating the initial parameters according to the stress and strain relationship. The present invention is applied to the simulation of grid structures, and solving the linear equation group using a block-to-pipeline parallel method can improve the calculation speed, shorten the time proportion of the simulation process in engineering applications, and improve engineering efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of grid structures, and in particular to a method and system for determining grid structure parameters. Background Art

[0002] The current simulation process of grid structures generally consists of three steps: users provide models, computers perform calculations, and output results for post-processing. The core of the computer calculation part and the part that takes up the most computing time and computing resources is solving the linear equations. The direct method is a widely used method for solving linear equations. However, the direct method solution part in current computing programs is mostly serial programs or parallel programs with insufficient parallelism, which cannot fully adapt to the development trend of multi-core CPUs. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for determining grid structure parameters, which are applied to the simulation of grid structures, can improve the calculation speed of structural parameters, shorten the time proportion of the simulation process in engineering applications, and improve engineering efficiency.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for determining grid structure parameters, comprising:

[0006] Conduct mechanical abstraction on the initial parameters of the grid structure and establish a CAE model;

[0007] Meshing the CAE model to obtain a unit stiffness matrix and nodal forces;

[0008] Determine the overall stiffness matrix and overall force based on the unit stiffness matrix and the node force;

[0009] Constructing a system of linear equations based on the global stiffness matrix and the global force;

[0010] The linear equations are solved by using a block-to-pipeline parallel method to obtain displacements; the block-to-pipeline parallel method includes a block parallel stage and a pipeline parallel stage;

[0011] Based on the displacement, determining stress and strain conditions of the structure through interpolation relationships;

[0012] The initial parameters are calibrated according to the stress and strain relationship.

[0013] Optionally, the block parallel stage specifically includes:

[0014] Design the block task pool taskpool_dm, ancestor node elimination schedule ancestor_pro, pipeline elimination chain, pipeline task pool taskpool_pp, pipeline task waiting pool taskpool_rest_pp and multiple locks;

[0015] The program obtains nodes from the block task pool taskpool_dm for each thread of each ancestor node until the block task pool taskpool_dm is empty. After obtaining the nodes, the execution block of the block task pool taskpool_dm is self-eliminated to make the block task pool taskpool_dm a tool block.

[0016] If the node received belongs to class A, it will start looping through the ancestor nodes after completing the block self-elimination, trying to obtain control. If the acquisition is successful, it will execute the block mutual elimination and update the elimination progress table ancestor_pro of the current ancestor node. Otherwise, it will continue to loop until all ancestor nodes have completed the block mutual elimination and updated;

[0017] If the received node belongs to class B, there is no need to complete the above block mutual elimination part of the loop, but in the process of block self-elimination, once a row completes self-elimination, the row that completes self-elimination will be added to the corresponding elimination chain of the target row belonging to class C; if it is determined in the process whether the target row has met the conditions for entering the pipeline task pool taskpool_pp or the pipeline waiting task pool taskpool_pp_rest, if so, the target row will be added to the corresponding task pool of the two;

[0018] When the block task pool is empty, the threads start the pipeline parallel phase.

[0019] Optionally, the pipeline parallel stage specifically includes:

[0020] The thread takes a preset number of rows from taskpool_pp and adds them to the local task pool mytaskpool until all rows belonging to class C have been taken; loops mytaskpool, and for each row, uses tool rows to eliminate the row according to its elimination chain; if the row has been eliminated by all tool rows, performs self-elimination of the row and adds it to the elimination chain of the corresponding target row; when adding, checks whether the target row meets the conditions for entering the pipeline task pool or the pipeline waiting area task pool. If so, moves the target row out of mytaskpool.

[0021] Optionally, if a row wants to enter the pipeline task pool taskpool_pp, the entry conditions are: the row has never entered the pipeline task pool before; the node to which the row belongs has been completed by all the child nodes belonging to class A.

[0022] If a row wants to enter the pipeline waiting task pool taskpool_rest_pp, the entry conditions are: the row has never entered the pipeline task pool and the pipeline waiting task pool before; the elimination chain of the row is full; the node to which the row belongs has not yet completed block mutual elimination by all child nodes belonging to class A.

[0023] The present invention also provides a grid structure parameter determination system, comprising:

[0024] Modeling module, used to perform mechanical abstraction on the initial parameters of the grid structure and establish a CAE model;

[0025] A partitioning module is used to perform mesh partitioning on the CAE model to obtain a unit stiffness matrix and a node force;

[0026] A global stiffness matrix and global force determination module, used to determine the global stiffness matrix and the global force based on the unit stiffness matrix and the node force;

[0027] An equation group construction module, used for constructing a linear equation group based on the overall stiffness matrix and the overall force;

[0028] A solving module, used for solving the linear equations by adopting a block-to-pipeline parallel method to obtain a displacement;

[0029] A stress and strain determination module, used to determine the stress and strain of the structure through an interpolation relationship based on the displacement;

[0030] The calibration module is used to calibrate the initial parameters according to the relationship between stress and strain.

[0031] Optionally, the linear equations are as follows:

[0032] Kx=F

[0033] Where K is the global stiffness matrix, F represents the global force, and x represents the displacement.

[0034] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0035] The present invention discloses a method and system for determining parameters of a grid structure, the method comprising: mechanically abstracting the initial parameters of the grid structure and establishing a CAE model; meshing the CAE model to obtain a unit stiffness matrix and a node force; determining an overall stiffness matrix and an overall force based on the unit stiffness matrix and the node force; constructing a linear equation group based on the overall stiffness matrix and the overall force; solving the linear equation group using a block-to-pipeline parallel method to obtain a displacement; the block-to-pipeline parallel method includes a block parallel stage and a pipeline parallel stage; determining the stress and strain of the structure based on the displacement through an interpolation relationship; and calibrating the initial parameters according to the stress and strain relationship. The present invention is applied to the simulation of grid structures, and solving the linear equation group using a block-to-pipeline parallel method can improve the calculation speed, shorten the time proportion of the simulation process in engineering applications, and improve engineering efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0037] Figure 1 This is a flow chart of a method for determining grid structure parameters according to an embodiment of the present invention;

[0038] Figure 2 Schematic diagram of block division

[0039] Figure 3 This is a binary tree and layered diagram. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The purpose of the present invention is to provide a method and system for determining grid structure parameters, which are applied to the simulation of grid structures, can improve the calculation speed of structural parameters, shorten the time proportion of the simulation process in engineering applications, and improve engineering efficiency.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] like Figure 1 As shown, a method for determining grid structure parameters includes:

[0044] Step 101: Perform mechanical abstraction on the initial parameters of the grid structure and establish a CAE model.

[0045] Step 102: Meshing the CAE model to obtain unit stiffness matrix and node forces.

[0046] Step 103: Based on the unit stiffness matrix and the node force, determine the overall stiffness matrix and the overall force.

[0047] Step 104: construct a linear equation system based on the overall stiffness matrix and the overall force.

[0048] Step 105: adopting a block-to-pipeline parallel method to solve the linear equation group and obtain displacement; the block-to-pipeline parallel method includes a block parallel stage and a pipeline parallel stage.

[0049] Step 106: Based on the displacement, determine the stress and strain of the structure through an interpolation relationship.

[0050] Step 107: Calibrate the initial parameters according to the relationship between stress and strain.

[0051] The stress and strain of the structure are determined through interpolation relationships, and based on this, it is checked whether the currently selected parameters meet the safety and economic requirements. If not, it is necessary to reselect the material properties (including cross-sectional dimensions, material type, etc.) and calculate again until they meet the requirements.

[0052] The solution in step 105 is the one that occupies the most computing resources and takes the longest time in the calculation process. Therefore, the present invention adopts a block-to-pipeline parallel method to solve the problem, which can significantly improve the calculation speed. The method is described in detail below:

[0053] This method is based on the traditional finite element direct method solution architecture. After symbolic decomposition is completed, this method is executed. In order to use storage space more efficiently and speed up the process, this method combines block parallelism with pipeline parallelism. In the early stage of numerical decomposition, block parallelism is used to increase the parallel granularity and reduce the length of pipeline parallel elimination chain. In the later stage, pipeline parallelism is used to increase parallelism and thread extensibility. Therefore, the method can be divided into the following steps: 1. Pre-processing stage; 2. Preparation stage for parallel task design; 3. Parallel block elimination stage; 4. Parallel pipeline elimination stage.

[0054] Step 1: Pre-processing stage of parallel solution technology.

[0055] Prepare the data file and provide it to the program. The program executes the following data segments in sequence: data file reading, compression, reordering, and symbol decomposition.

[0056] For data files, the code provided by this program supports matrices in Rutherford-Boeing format and Harwell-Boeing format, where the parameter MXTYPE in the third row of the data header is required to be RSA (Real Symmetric Assembled). If there is a right-hand term, the parameter RHSTYP in the fifth row needs to be read in with an F prefix (full storage) and saved in CSR (compact sparse row) format. Data in other formats can be read in by the user, but must be saved as a sparse matrix stored in CSR format.

[0057] The compression stage uses cell storage technology to compress the original CSR format storage into CELL format storage. This format can effectively reduce the storage volume of indicators. The storage of JA and JU can be reduced to 1 / 9 or even 1 / 36 of the original CSR format. In addition to saving memory, it can also bring better data locality and improve the computing efficiency of the basic calculation of the parallel part.

[0058] The purpose of reordering is to reduce the number of filler elements, so this step can also be called filler element optimization. The implementation of this step can greatly reduce the floating point operations of decomposition. The reordering technology of this program uses the METIS_4.0.3 library published by Karggis Lab for reordering. Its core is to use the multiple partitioning algorithm and the minimum degree algorithm to select the insulators in turn, and divide the part without the insulators into left and right subtrees to solve the problem by divide and conquer.

[0059] The symbolic decomposition stage is a necessary stage for solving the sparse matrix direct method. The number and position of elements are determined through symbolic analysis, which facilitates the subsequent numerical decomposition.

[0060] Step 2: Preparation phase for parallel elimination.

[0061] For a clear description, some terms are first defined.

[0062] Co-cancellation element: In LDLT decomposition, a row needs to be doubled by the previous row, which is called the co-cancellation element of row A to row B.

[0063] Self-elimination: After a row A is eliminated by all the previous rows, it needs to be multiplied by itself, which is called self-elimination.

[0064] Tool row and target row: According to the LDLT elimination relationship, if row A eliminates row B, row B is called the target row (of row A) and row A is the tool row. Once the target row is eliminated, it becomes the tool row (of the corresponding row).

[0065] Tool blocks, quasi-tool blocks and target blocks: Using the graph decomposition of METIS, the matrix of the linear equations can be reordered as Figure 2 The node topology of the matrix shown is equivalent to a binary tree, where each isolator is the root node, and the two sub-blocks correspond to the left and right subtrees respectively. Starting from the root of the tree, the numbering starts from zero from the root to the leaf node, and the node number can be simply used to refer to each sub-block. It can be analogous to the previous name. If block A is a subtree of block B, then block A is called the tool block of block B, and block B is the target block of block A. The tool block needs to perform block mutual elimination on all its target blocks; the target block can perform block self-elimination after all tool blocks have performed block mutual elimination on it. After block self-elimination, it becomes a tool block and starts to eliminate the remaining blocks. The block that has been completed mutual elimination by all tool blocks is called a quasi-tool block, and the quasi-tool block becomes a tool block after completing block self-elimination.

[0066] For PCs, computing usually does not exceed 16 nodes. To balance tasks and maximize granularity, you can choose layers 1 and 2 to meet the requirements of

[0067]

[0068] Therefore, according to the specified number of threads np, two key layers can be calculated, and the nodes can be divided into three categories according to the layers:

[0069] Class A: nodes that perform block self-cancellation and block mutual cancellation;

[0070] Class B: nodes that only perform block self-elimination;

[0071] Class C: Nodes eliminated by water flow.

[0072] In the decomposition tree, the nodes of layer [layer1, layer2) belong to node class A, the nodes of layer layer2 belong to node class B, and the nodes of layer (0, layer2) belong to node class C.

[0073] The following is the details of step 2.

[0074] Preparation phase for block parallel elimination:

[0075] Design a block task pool taskpool_dm: It stores the quasi-tool block nodes belonging to class A and class B in the binary tree. Initially, it is a leaf node of layer1.

[0076] Design the ancestor node elimination progress table ancestor_pro to record the progress of each ancestor node being eliminated by its child blocks. Initially, it is designed to be the number of class A child nodes it has. Every time a class A child node completes the block mutual elimination, it decreases by one. When it decreases to zero, it becomes a quasi-tool block; for class A and class B nodes, after becoming a quasi-tool block, they are added to the task pool; for class C nodes, becoming a quasi-tool block means that the rows in it will no longer compete with the "block" for data, and can start pipeline elimination.

[0077] Design pipeline row elimination chain: The elimination chain is stored in CSR form. Only rows belonging to class C nodes have this elimination chain, and the length of the elimination chain of row j is a kj However, due to the block design, only the non-zero elements of the rows in class C need to be stored. Under this design, the amount of storage can be reduced.

[0078] Design a pipeline task pool taskpool_pp: which stores the row numbers of nodes belonging to class C in ascending order. The condition for entering taskpool_pp is that the row belongs to a node of class C and the elimination progress of the node is zero.

[0079] Design of pipeline task waiting pool: taskpool_rest_pp: Because the time to check whether a row enters the pipeline pool is when the elimination chain of the row is updated, if the elimination chain of a row is full but still in the block mutual elimination stage, it cannot enter the pipeline task pool taskpool_pp, otherwise data competition will occur and the result will be distorted. Therefore, taskpool_rest_pp is designed to store rows whose elimination chains are full but do not meet the conditions for entering taskpool_pp.

[0080] Individual locks designed to avoid data conflicts.

[0081] Step 3: Parallel block elimination phase.

[0082] For block parallel elimination, the program collects nodes from the block task pool taskpool_dm for each thread of each ancestor node until the block task pool taskpool_dm is empty. After receiving the node, perform block self-elimination on it to make it a tool block. If the node received belongs to class A, after completing the block self-elimination, it starts to loop through its ancestor nodes (i.e., its corresponding target block) and try to obtain its control. If the acquisition is successful, the block mutual elimination is executed and the elimination schedule table ancestor_pro of the ancestor node is updated (if the node inode belongs to class A and class B, the corresponding element ancestor_pro[inode] of its ancestor node elimination schedule as an ancestor node is 0, then it is added to the block task pool taskpool_dm), otherwise the loop continues until all ancestor nodes are completed and updated by them. If the node received belongs to class B, there is no need to complete the above block mutual elimination part of the cycle, but in the process of block self-elimination, once a row completes the self-elimination of the row, the row (already a tool row) is added to the corresponding elimination chain of the target row belonging to class C; if the process determines whether the target row has met the conditions for entering the pipeline task pool taskpool_pp or the pipeline waiting task pool taskpool_pp_rest, if so, it will be added to the corresponding task pool of the two. Because the ancestor node of class B must belong to class C, there is no need to join the block task pool taskpool_dm.

[0083] If a row wants to enter the pipeline task pool taskpool_pp, the entry conditions are: 1. The row has never entered the pipeline task pool before; 2. The node to which the row belongs has been completed by all its child nodes belonging to class A.

[0084] If a row wants to enter the pipeline waiting task pool taskpool_rest_pp, the entry conditions are: 1. The row has never entered the pipeline task pool and the pipeline waiting task pool before; 2. The elimination chain of the row is full; 3. The node to which the row belongs has not yet completed the block mutual elimination by all its child nodes belonging to class A.

[0085] When the block task pool is empty, the thread enters step 4 and starts the pipeline stage.

[0086] In the program, a sharding acceleration design is implemented for this process. Each child node is divided into many parts. After each part is executed, the above cycle is repeated to the corresponding shard of the ancestor node to alleviate the waiting degree caused by data conflicts.

[0087] Step 4: Parallel Pipeline Elimination Stage

[0088] like Figure 3As shown in the figure, in the partition tree, the nodes of layer (0, layer2) belong to node class C. The rows in the nodes of this class do not enter the task pool as a whole in the form of blocks, but enter the task pool in a pipelined parallel manner when they meet the entry conditions. It is worth mentioning that for most matrices with practical significance, whether circuits or building structures, the number of rows of sub-nodes completed in block parallel will account for the vast majority of the number of matrix rows. Since a large number of tool rows have completed the elimination of class C rows in the block mutual elimination meta-task of node class A, the length of the elimination chain of each row in this algorithm is much smaller than the length of the elimination chain of pure pipeline parallel. At the same time, because the rows in class A and class B do not need to be eliminated through pipeline parallelism, the total length of the pipeline task pool will be greatly reduced, thereby saving the overhead of managing the pipeline task pool.

[0089] In addition, through the results of the block division, the pipeline task pool taskpool_pp can be bucketed. Through the idea of ​​bucket sorting, the pipeline task pool can be divided into many buckets. There is a first-level size relationship between the buckets, and the buckets are sorted by themselves. This method reduces the total length of the queue to be sorted, further saving the overhead of managing the pipeline task pool.

[0090] The specific process of pipeline elimination is as follows: the thread takes a certain number of rows from taskpool_pp and adds them to the local task pool mytaskpool until all rows belonging to class C have been taken. Loop through mytaskpool, and for each row, use tool rows to eliminate it according to its elimination chain. If it has been eliminated by all tool rows (i.e., the elimination chain is full), it will be self-eliminated and added to the elimination chain of the corresponding target row. When adding, check whether the target row meets the conditions for entering the pipeline task pool or the pipeline waiting area task pool. Move the row out of mytaskpool.

[0091] After completing the LDLT decomposition, a general triangular matrix back-substitution solver is used to sequentially solve the overall displacement vector x of the structure, that is, to convert the full matrix Ax=B into the solution of three triangular matrices and diagonal matrices in sequence, LTz=F, Dy=z, Lx=y.

[0092] The advantages of the method of the present invention include: (1) improving the situation where the parallelism of pure block elimination decreases when approaching the top level, and a large number of threads are idling; (2) avoiding the management overhead caused by the excessive length of the pure pipeline task pool; (3) avoiding the situation where each row has an elimination chain and the total length of the elimination chain is almost the total length of the JU, thereby reducing memory overhead; (4) avoiding the situation where the granularity is small and the performance is wasted due to frequent access to the task pool, thereby increasing computing efficiency; (5) avoiding the situation where each row has a lock, which increases memory and computing overhead.

[0093] Let's start with the detailed description:

[0094] (1) Improve the situation that the parallelism of pure block elimination decreases when approaching the top level, and a large number of threads are idling. This problem is alleviated by introducing pipeline parallelism (step 4), which is also the most essential innovation of the present invention.

[0095] (2) Avoid the management overhead caused by the excessive length of the pure pipeline task pool. For pure pipeline parallel elimination, a task pool equal to the number of squares is required to manage the tasks, but the task pool must be ordered, so the management of the task pool will waste a lot of time. By introducing step 3, a large number of rows can be completed in step 3 through block parallel elimination before pipeline parallelization, without entering the pipeline task pool, thus greatly reducing the overhead of task pool management.

[0096] (3) Avoid the situation where each row has an elimination chain and the total length of the elimination chain is almost the total length of JU, reducing memory overhead: Similar to (2), by introducing step 3, block parallel elimination is performed before pipeline parallelization, so that a large number of rows can be completed in step 3 in the form of block mutual elimination, and there is no need to eliminate through pipeline parallelization, so there is no need for elimination chains, thereby reducing memory consumption.

[0097] (4) Avoid the situation where the granularity is small and the performance waste is caused by frequent access to the task pool, thereby increasing computing efficiency; in step (4), taking more than one row in sequence is equivalent to grading the task pool, thereby avoiding the performance waste caused by frequent access to the task pool.

[0098] (5) Avoid the situation where each row has a lock, which increases memory and computing overhead.

[0099] The present invention also provides a grid structure parameter determination system, comprising:

[0100] The modeling module is used to perform mechanical abstraction on the initial parameters of the grid structure and establish a CAE model.

[0101] The partitioning module is used to perform mesh partitioning on the CAE model to obtain the unit stiffness matrix and the node force.

[0102] The overall stiffness matrix and overall force determination module is used to determine the overall stiffness matrix and overall force based on the unit stiffness matrix and the node force.

[0103] The equation group construction module is used to construct a linear equation group based on the overall stiffness matrix and the overall force. The linear equation group is as follows:

[0104] Kx=F

[0105] Where K is the global stiffness matrix, F represents the global force, and x represents the displacement.

[0106] The solution module is used to solve the linear equations by adopting a block-to-pipeline parallel method to obtain the displacement.

[0107] The stress and strain determination module is used to determine the stress and strain of the structure through an interpolation relationship based on the displacement.

[0108] The calibration module is used to calibrate the initial parameters according to the relationship between stress and strain.

[0109] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0110] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for determining grid structure parameters, characterized in that: include: Conduct mechanical abstraction on the initial parameters of the grid structure and establish a CAE model; Meshing the CAE model to obtain a unit stiffness matrix and nodal forces; Determine the overall stiffness matrix and overall force based on the unit stiffness matrix and the node force; Constructing a system of linear equations based on the global stiffness matrix and the global force; The linear equations are solved by adopting a block-to-pipeline parallel method to obtain displacement; The block-to-pipeline parallel method includes a block parallel stage and a pipeline parallel stage; Based on the displacement, determining stress and strain conditions of the structure through interpolation relationships; Calibrate the initial parameters based on the stress and strain relationship The block parallel stage specifically includes: Design the block task pool taskpool_dm, ancestor node elimination schedule ancestor_pro, pipeline elimination chain, pipeline task pool taskpool_pp, pipeline task waiting pool taskpool_rest_pp and multiple locks; The program obtains nodes from the block task pool taskpool_dm for each thread of each ancestor node until the block task pool taskpool_dm is empty. After obtaining the nodes, the execution block of the block task pool taskpool_dm is self-eliminated to make the block task pool taskpool_dm a tool block. If the node received belongs to class A, it will start looping through the ancestor nodes after completing the block self-elimination, trying to obtain control. If the acquisition is successful, it will execute the block mutual elimination and update the elimination progress table ancestor_pro of the current ancestor node. Otherwise, it will continue to loop until all ancestor nodes have completed the block mutual elimination and updated; If the received node belongs to class B, there is no need to complete the above block mutual elimination part of the loop, but in the process of block self-elimination, once a row completes self-elimination, the row that completes self-elimination will be added to the corresponding elimination chain of the target row belonging to class C; if it is determined in the process whether the target row has met the conditions for entering the pipeline task pool taskpool_pp or the pipeline waiting task pool taskpool_pp_rest, if so, the target row will be added to the corresponding task pool of the two; When the block task pool is empty, the threads start the pipeline parallel phase.

2. The method for determining grid structure parameters according to claim 1, characterized in that: The pipeline parallel stage specifically includes: The thread takes a preset number of rows from taskpool_pp and adds them to the local task pool mytaskpool until all rows belonging to class C have been taken; loops mytaskpool, and for each row, uses tool rows to eliminate the row according to its elimination chain; if the row has been eliminated by all tool rows, performs self-elimination of the row and adds it to the elimination chain of the corresponding target row; when adding, checks whether the target row meets the conditions for entering the pipeline task pool or the pipeline waiting area task pool. If so, moves the target row out of mytaskpool.

3. The method for determining grid structure parameters according to claim 2, characterized in that: If a row wants to enter the pipeline task pool taskpool_pp, the entry conditions are: the row has never entered the pipeline task pool before; the node to which the row belongs has been completed by all the child nodes belonging to class A. If a row wants to enter the pipeline waiting task pool taskpool_rest_pp, the entry conditions are: the row has never entered the pipeline task pool and the pipeline waiting task pool before; the elimination chain of the row is full; the node to which the row belongs has not yet completed block mutual elimination by all child nodes belonging to class A.

4. A grid structure parameter determination system, characterized in that: include: Modeling module, used to perform mechanical abstraction on the initial parameters of the grid structure and establish a CAE model; A partitioning module is used to perform mesh partitioning on the CAE model to obtain a unit stiffness matrix and a node force; A global stiffness matrix and global force determination module, used to determine the global stiffness matrix and the global force based on the unit stiffness matrix and the node force; An equation group construction module, used for constructing a linear equation group based on the overall stiffness matrix and the overall force; A solving module, used for solving the linear equations by adopting a block-to-pipeline parallel method to obtain a displacement; A stress and strain determination module, used to determine the stress and strain of the structure through an interpolation relationship based on the displacement; A calibration module, used for calibrating the initial parameters according to the relationship between stress and strain; The block parallel stage specifically includes: Design the block task pool taskpool_dm, ancestor node elimination schedule ancestor_pro, pipeline elimination chain, pipeline task pool taskpool_pp, pipeline task waiting pool taskpool_rest_pp and multiple locks; The program obtains nodes from the block task pool taskpool_dm for each thread of each ancestor node until the block task pool taskpool_dm is empty. After obtaining the nodes, the execution block of the block task pool taskpool_dm is self-eliminated to make the block task pool taskpool_dm a tool block. If the node received belongs to class A, it will start looping through the ancestor nodes after completing the block self-elimination, trying to obtain control. If the acquisition is successful, it will execute the block mutual elimination and update the elimination progress table ancestor_pro of the current ancestor node. Otherwise, it will continue to loop until all ancestor nodes have completed the block mutual elimination and updated; If the received node belongs to class B, there is no need to complete the above block mutual elimination part of the loop, but in the process of block self-elimination, once a row completes self-elimination, the row that completes self-elimination will be added to the corresponding elimination chain of the target row belonging to class C; if it is determined in the process whether the target row has met the conditions for entering the pipeline task pool taskpool_pp or the pipeline waiting task pool taskpool_pp_rest, if so, the target row will be added to the corresponding task pool of the two; When the block task pool is empty, the threads start the pipeline parallel phase.

5. The grid structure parameter determination system according to claim 4, characterized in that: The linear equations are as follows: Kx=F Where K is the global stiffness matrix, F represents the global force, and x represents the displacement.

Citation Information

Patent Citations

  • Method for storing and generating stiffness matrix for finite-element analysis in metal bulk plastic forming

    CN102184298A

  • Constant geometric method-based composite thin-walled structure inherent frequency design method

    CN108763658A

  • A full-band antenna array and radome integrated simulation method

    CN109726439A

  • Systems, methods and, media for simulating deformations of nonlinear elastic bodies

    US20190012831A1

  • Systems and methods for reducing memory traffic and power consumption in a processing environment by solving a system of linear equations

    US8417755B1