GPU parallel-based material damage simulation method, electronic equipment and medium
By discretizing the material structure into subdomains and restricting neighborhood search, the problems of GPU memory limitations and global search complexity are solved, enabling efficient and accurate large-scale near-field dynamic damage simulation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-11
- Publication Date
- 2026-03-27
AI Technical Summary
Existing GPU-parallel near-field dynamic damage simulation methods suffer from low computational efficiency and insufficient accuracy when dealing with large-scale material damage, due to limitations in single-GPU memory capacity and global neighborhood search complexity.
The overall material structure is discretized into multiple subdomains, and the affiliation between material points and subdomains is managed through a specific data structure. Neighborhood search is restricted to calculations within the subdomains, and data is stored and transmitted in subdomain order to reduce global search.
Efficient and accurate large-scale near-field dynamic damage simulation was achieved under limited single-GPU memory conditions, improving computational efficiency and accuracy.
Smart Images
Figure CN121744640A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer simulation technology, specifically to a GPU-based parallel material damage simulation method, electronic device, and medium. Background Technology
[0002] Numerical simulation of material damage and failure processes is a crucial aspect of engineering design and safety assessment. To accurately simulate complex discontinuous mechanical behaviors such as crack initiation and propagation, peridynamics (PD) theory has emerged. This theory employs a nonlocal integral form, avoiding the dependence on continuity inherent in traditional continuum mechanics, and is particularly suitable for simulating brittle fracture and impact damage. However, with increasingly complex simulation models and a surge in particle numbers, the computational load has grown exponentially. To improve computational efficiency, leveraging the massively parallel computing capabilities of graphics processing units (GPUs) for acceleration has become an important technological direction in this field.
[0003] However, existing GPU-parallel near-field dynamics damage simulation methods still face significant bottlenecks in practice. GPU memory capacity has a physical limit, while the total particle data, neighborhood information, and intermediate variables required for large-scale simulations often far exceed this limit. This makes single global data loading impossible, limiting the scale of simulable physics. Although there are solutions to reduce storage by decreasing data precision, this may sacrifice computational stability and simulation accuracy. Existing solutions that attempt to expand total memory through multi-GPU collaboration inevitably introduce expensive hardware costs and complex load balancing and communication synchronization issues.
[0004] In summary, existing technologies struggle to perform near-field dynamics simulations of large-scale material damage under limited single-GPU memory conditions. Furthermore, the highly complex global neighborhood search severely restricts simulation efficiency. Therefore, overcoming the memory bottleneck and significantly reducing computational complexity on ordinary hardware platforms to achieve efficient and high-precision large-scale near-field dynamics damage simulation has become a pressing technical challenge in this field. Summary of the Invention
[0005] In view of this, it is necessary to provide a GPU-parallel material damage simulation method, electronic device and medium to solve the technical problems of low simulation efficiency and low accuracy in the current method when performing near-field dynamic simulation of large-scale material damage and failure, which is difficult to complete the calculation under the limited single GPU memory conditions. The global high-complexity neighborhood search also severely restricts the simulation efficiency.
[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a GPU-parallel material damage simulation method, comprising: Discretize the overall material structure to be simulated into multiple material points; Based on the GPU memory capacity, the overall computational domain containing the material points is dynamically divided into multiple subdomains to obtain the division result. Each subdomain includes an internal computational region and a boundary region extending outward from the computational region by at least one near-field range. Based on the partitioning results, a data structure for managing the affiliation relationship between material points and subdomains is constructed, and this data structure is used to convert the storage of information of each material point from the overall structure order to the subdomain order. Iterative calculations are performed sequentially by subdomain. For the current subdomain, the material point information required for the calculation of the current subdomain is transmitted from the host to the GPU device. The GPU device performs a calculation of the time step based on the material point information in the current subdomain. The neighborhood search operation is restricted to the data range of the current subdomain. After the calculation is completed, the updated material point information is transmitted back to the host from the GPU device. After the entire calculation is completed, all material point information is converted from being stored in subdomain order back to being stored in the overall structural order, and the final material point information is output as the damage simulation result.
[0007] In one possible implementation, the construction of the data structure for managing the affiliation of matter points with subdomains includes: Generate a header pointer array to index the storage starting position of each subdomain computation area; Generate a linked list array to record the sequential relationship of material points within the same subdomain computational region in the form of a linked list.
[0008] In one possible implementation, the dynamic division of the overall computational domain containing the material points into multiple subdomains based on the GPU memory capacity yields the following division results: Obtain the available video memory capacity of the GPU; Based on the available video memory capacity and the amount of video memory required to store information of a single material point, the maximum number of material points that a single subdomain is allowed to contain is calculated and determined. The overall computational domain is partitioned to form multiple subdomains, with the constraint of not exceeding the maximum number of material points, to obtain the partitioning result.
[0009] In one possible implementation, the step of transmitting the material point information required for the current subdomain calculation from the host to the GPU device includes: Transmit all material point information within the computational region of the current subdomain; If the computation to be performed involves neighborhood interactions, then information about the material points within the boundary region of the current subdomain is transmitted simultaneously.
[0010] In one possible implementation, prior to the step of converting the storage of material point information from the overall structural order to the subdomain order using this data structure, the GPU-parallel material damage simulation method further includes: In the host's memory, a total storage space is allocated for storing the material point information of all subdomains in subdomain order. The capacity of the total storage space is determined based on the total number of subdomains, the maximum number of material points in a single subdomain, and the number of bytes per unit data. In the GPU device's video memory, a single storage space is allocated for storing information about material points in a single subdomain. The capacity of the single storage space is determined based on the maximum number of material points in a single subdomain and the number of bytes per unit data.
[0011] In one possible implementation, the calculation of a time step based on the material point information within the current subdomain on the GPU device side includes: Within the data range of the current subdomain, perform a neighborhood search for each material point and establish a neighborhood list; Based on the neighborhood list, bond force calculation and motion equation integration are performed to update the kinematic state of the material points; The damage status of the material points is updated based on the updated kinematic state and the preset damage criteria.
[0012] In one possible implementation, the updated material point information includes updated material point displacement, velocity, and damage state.
[0013] In one possible implementation, the computational regions of the subdomains are arranged continuously and without overlap in space and collectively cover the overall computational domain, wherein a material point belongs to the computational region of a subdomain.
[0014] In a second aspect, the present invention also provides an electronic device, including a memory and a processor, wherein, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the GPU-based parallel material damage simulation method described in any of the above implementations.
[0015] Thirdly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the GPU-based parallel material damage simulation method described in any of the above implementations.
[0016] The beneficial effects of this invention are as follows: The GPU-parallel material damage simulation method provided by this invention discretizes the material structure and dynamically divides it into multiple subdomains containing computational regions and extended boundary regions based on GPU memory. Then, a specific data structure is used to reorganize the particle information storage order. During computation, only the data required by the current subdomain is scheduled to the GPU in batches, and local neighborhood searches and other near-field dynamics calculations are performed within that subdomain. Finally, the results are reorganized and output. This allows large-scale simulation tasks to be completed within the limited memory of a single GPU. Simultaneously, by reducing the global search to a series of local searches, the amount of unnecessary computation is significantly reduced. This effectively solves the problems of computational limitations due to insufficient GPU memory and low computational efficiency caused by global neighborhood searches in existing technologies, improving the efficiency and accuracy of large-scale material damage simulation under ordinary hardware conditions. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 A schematic flowchart of an embodiment of the GPU-parallel material damage simulation method provided by the present invention; Figure 2 For the present invention Figure 1 A schematic diagram of an embodiment of S103; Figure 3 For the present invention Figure 1 A schematic diagram of an embodiment of S102; Figure 4 For the present invention Figure 3 A schematic diagram of an embodiment of S102; Figure 5 A schematic flowchart of another embodiment of the GPU-parallel material damage simulation method provided by the present invention; Figure 6 For the present invention Figure 1 A schematic diagram of an embodiment of S104; Figure 7 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0020] In the description of the embodiments of the present invention, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.
[0021] The terms "first," "second," etc., used in the embodiments of this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature.
[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0023] Before demonstrating the embodiments, the following terms will be explained.
[0024] Peridynamics (PD) is a nonlocal continuum mechanics theory that uses integral equations to describe the interactions between material points. It is particularly suitable for simulating discontinuous mechanical problems such as the spontaneous initiation and propagation of cracks, and avoids the mathematical singularities faced by traditional theories at cracks.
[0025] Material Point: Also known as a particle, it is the basic computational unit obtained by spatial discretization of a continuous medium in a near-field dynamics model. Each material point carries physical information such as mass, position, displacement, velocity, and damage.
[0026] Near-field range (Horizon): The radius of a finite spatial region within which a point of matter can interact with other points of matter (usually denoted as Horizon). δ The interaction is limited to this range, which reflects the "nonlocal" nature of the model.
[0027] A bond is a virtual interaction unit that connects two material points located within each other's near-field range. Bond breaking is used to characterize the damage and failure process of materials.
[0028] Damage: In near-field dynamics, it usually refers to the proportion of all bonds broken at a point in a material. It is used to quantify the degree of local damage to a material and ranges from 0 (intact) to 1 (completely destroyed).
[0029] GPU Parallelism: Utilizes the highly parallel many-core architecture of graphics processing units to decompose computational tasks into a large number of threads that execute simultaneously, thereby accelerating computationally intensive tasks such as large-scale neighborhood searches and force calculations in near-field dynamics.
[0030] Host: This usually refers to the main part of the computer, which contains the central processing unit and system memory. It is responsible for program flow control, data preparation, and communication with devices.
[0031] Device: This usually refers to the graphics processor and its video memory, which is responsible for executing highly parallel computing kernels.
[0032] Subdomain: A local computational unit obtained by dynamically dividing the overall computational domain to overcome the limitations of video memory. Each subdomain is a basic data block that is used for batch-based independent computation.
[0033] Boundary Region: A transitional region formed by extending at least one "near-field range" outward from the subdomain computational region. Information about matter points within this region is introduced to ensure data integrity when performing "neighborhood searches" on matter points within the computational region.
[0034] Neighborhood Search: A key step in near-field dynamics calculations, which involves finding all other material points within its "near-field range" for each material point to determine the set of its interaction bonds.
[0035] Head Pointer Array: An indexed array where each element corresponds to a subfield, recording the starting storage position of the material point data of that subfield's calculation region in the reconstructed array.
[0036] Linked List Array: A storage array used in conjunction with a "head pointer array" to record the index of each material point belonging to the same subdomain computational region in the form of a linked list, thereby establishing a complete order of all material points within that subdomain.
[0037] Time Step: In iterative calculations, the physical time interval represented by each state update. Its value must satisfy numerical stability conditions. This invention provides a GPU-parallel material damage simulation method, apparatus, electronic device, and storage medium. The technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0038] Figure 1 A schematic diagram of an embodiment of the GPU-parallel material damage simulation method provided by the present invention is shown below. Figure 1 As shown, the GPU-parallel material damage simulation method includes: S101. Discretize the overall material structure to be simulated into multiple material points; S102. Based on the GPU memory capacity, the overall computational domain containing material points is dynamically divided into multiple subdomains to obtain the division result. Each subdomain includes an internal computational region and a boundary region extending outward from the computational region by at least one near-field range. S103. Based on the partitioning results, construct a data structure for managing the affiliation relationship between material points and subdomains, and use this data structure to convert the storage of information of each material point from the overall structure order to the subdomain order. S104. Perform iterative calculations in subdomain order. For the current subdomain, transmit the material point information required for the calculation of the current subdomain from the host to the GPU device. On the GPU device, perform a calculation of the time step based on the material point information in the current subdomain. The neighborhood search operation is restricted to the data range of the current subdomain. After the calculation is completed, transmit the updated material point information from the GPU device back to the host. S105. After the entire calculation is completed, convert all material point information from subdomain order storage back to overall structural order storage, and output the final material point information as the damage simulation result.
[0039] It should be noted that the data structure in this embodiment converts the storage of information of each material point from the overall structure order to the subdomain order. Alternatively, this data structure can be used to convert the storage of information of each material point from the subdomain order to the overall structure order. In practical applications, it can be flexibly set according to the requirements, and no specific limitation is made here.
[0040] The neighborhood search may not necessarily be performed during the time iteration, but it may also be performed before the time iteration.
[0041] In some embodiments of the present invention, the computational regions of the subdomains are arranged continuously and without overlap in space and together cover the overall computational domain, wherein a material point belongs to the computational region of a subdomain.
[0042] In some embodiments of the present invention, the updated material point information includes updated material point displacement, velocity, and damage status, etc.
[0043] In some embodiments of the present invention, such as Figure 2 As shown, step S103, which constructs a data structure for managing the affiliation of matter points and subdomains, includes: S201. Generate a head pointer array to index the storage start position of each subdomain computation area; S202. Generate a linked list array to record the sequential relationship of each material point within the same subdomain calculation area in the form of a linked list.
[0044] In this embodiment, firstly, a head pointer array is generated. The length of this array is equal to the total number of subdomains. Next, a linked list array is generated. The length of this array is the same as the total number of material points in the overall structure, and its function is to record the sequential relationships between material points belonging to the same subdomain computational region in the form of a static linked list.
[0045] Specifically, if a substance point A should be followed by substance point B, the value at the corresponding position of A in the linked list array is set to the global number of B, thus forming a linked structure; each linked list ends with a specific identifier (such as -1). This pair of arrays constitutes the core indexing mechanism for data reorganization. It is worth noting that, for complete management of subdomain information, a boundary array is usually also generated to separately record the substance point numbers contained in the boundary region of each subdomain, enabling efficient data filtering and transmission when needed.
[0046] To illustrate more clearly, consider a simulation model divided into four subdomains, with global matter point numbers ranging from 0 to 999. The head pointer array might be recorded as [0, 300, 600, 850], indicating that particle data for subdomain 1's computational region starts from position 0 of the new array, subdomain 2 from position 300, and so on. A linked list array manages the order within each subdomain: for example, if subdomain 1's computational region contains matter points with global numbers 5, 23, and 101, and their storage order is 5, 23, 101, then the linked list array stores 23 at position 5, 101 at position 23, and the end marker at position 101. By traversing the head pointer array to locate the starting point and then tracing the linked list array, the complete data for all matter points in that subdomain can be quickly and completely collected and transferred.
[0047] This embodiment explicitly transforms the implicit spatial hierarchy between material points into an index relationship that can be efficiently queried and manipulated by the computer. It achieves fast and error-free conversion between two storage modes: overall order and subdomain order. This allows subsequent transmission between the host and device to only the compact data block of the current subdomain, significantly reducing redundant data copying overhead. Simultaneously, it provides accurate data boundaries for subsequent neighborhood searches, reducing computational complexity. This also helps reduce GPU memory usage, thereby improving overall computational efficiency.
[0048] In some embodiments of the present invention, such as Figure 3 As shown, step S102 dynamically divides the overall computational domain containing the material points into multiple subdomains based on the GPU memory capacity, resulting in the following division results: S301. Obtain the available video memory capacity of the GPU; S302. Based on the available video memory capacity and the amount of video memory required to store information about a single material point, calculate and determine the maximum number of material points that a single subdomain is allowed to contain. S303. With the constraint of not exceeding the maximum number of material points, the overall computational domain is partitioned to form multiple subdomains, and the partitioning result is obtained.
[0049] In this embodiment, firstly, step S301 obtains the available video memory capacity of the GPU. In actual implementation, the total physical video memory of the GPU can be queried through the system interface, and system overhead is reserved according to the current operating environment to determine the net available video memory for computation. Next, step S302 calculates and determines the maximum number of material points that a single subdomain can contain based on the available video memory capacity and the amount of video memory required to store all information of a single material point (including coordinates, displacement, velocity, damage state, etc.). The specific calculation logic is: maximum number of particles = (e × available video memory) / overhead of storing information of a single particle, where e is a reduction factor less than 1 (e.g., 0.7 to 0.9) to reserve system overhead such as video memory alignment and kernel function execution, ensuring that the reserved video memory space is sufficient. Finally, step S303 uses the calculated maximum number of material points as a hard constraint to partition the geometric space of the overall computation domain. When partitioning, it is necessary to ensure that the computational regions of each subdomain are continuous, non-overlapping, and completely cover the overall structure, and expand outward to form boundary regions, ultimately forming multiple subdomains and obtaining the partitioning results.
[0050] This embodiment establishes a universal mechanism that directly and quantitatively transforms physical hardware limitations (GPU memory capacity) into algorithm control parameters (subdomain size). This ensures that, regardless of the simulation model's scale, the total data volume of a single subdomain will not exceed the GPU's memory capacity, thus fundamentally avoiding memory overflow errors caused by excessive data volume. Simultaneously, the hard-constraint-guided partitioning allows for precise allocation of a fixed-size GPU memory space to each subdomain on the device side in subsequent steps, simplifying memory management and improving data copying efficiency.
[0051] In some embodiments of the present invention, such as Figure 4 As shown, step S102 transmits the material point information required for the current subdomain calculation from the host to the GPU device, including: S401. Transmit all material point information within the computational region of the current subdomain; S402. If the computation to be performed involves neighborhood interactions, the material point information within the boundary region of the current subdomain is also transmitted.
[0052] Specifically, in step S401, the material points within the current subdomain are the targets whose states need to be updated in the current time step, therefore, it is necessary to transmit all material point information within the computational region of the current subdomain. Step S402 further determines the conditional branch: if the computational task to be executed involves neighborhood interactions, then it is necessary to transmit the material point information within the boundary region of the current subdomain simultaneously; otherwise, it is not necessary to transmit. This implementation is based on the computational principle of near-field dynamics: operations such as neighborhood search and bond force calculation require obtaining information on all neighbors within the near-field range of the material points within the computational region of the subdomain, and these neighbors may be located within the boundary region; conversely, for some calculations that only rely on the historical state of the material points themselves (such as only updating their damage markers), their neighbor data is not required.
[0053] In this embodiment, the selective transfer strategy aims to further reduce video memory overhead.
[0054] To illustrate this, consider the following example: Suppose that two core computational kernels need to be executed sequentially within an iteration over a time step. The first kernel performs bond force calculations, and the second kernel updates the motion state. Before executing the first kernel, because the calculation involves neighborhood interactions, the host needs to transmit the coordinates and displacement information of all material points in the current subdomain's computational region and boundary regions to the device. After the first kernel completes its calculation, before starting the second kernel, the motion state of the material points is based solely on their updated force states, thus eliminating the need to transmit material point information from the boundary regions.
[0055] This embodiment effectively reduces the amount of data copied from the host to the device each time by avoiding the transmission of particle information in the boundary region when it is not needed, thereby reducing transmission latency and bus bandwidth pressure. At the same time, less data transmission means that the space occupied in the device's video memory for storing temporary data of the current subdomain is reduced. This allows each subdomain to accommodate a larger core computing area with a fixed video memory capacity, thereby improving the overall utilization efficiency of hardware resources and the problem scale of a single calculation, and improving the execution efficiency of simulation tasks under limited video memory constraints.
[0056] In some embodiments of the present invention, such as Figure 5 As shown, before the step of converting the storage of each material point information from the overall structural order to the subdomain order using this data structure, this GPU-parallel material damage simulation method also includes: S501. In the host's memory, allocate a total storage space for storing the material point information of all subdomains in subdomain order. The capacity of the total storage space is determined based on the total number of subdomains, the maximum number of material points in a single subdomain, and the number of bytes per unit data. S502. In the GPU device's video memory, allocate a single storage space for storing information about a single subdomain of material points. The capacity of the single storage space is determined based on the maximum number of material points in a single subdomain and the number of bytes per unit data.
[0057] In this embodiment, before performing the data reorganization step, the method further includes a storage space pre-allocation step to establish an efficient and controllable memory environment for subsequent allocation calculations.
[0058] Specifically, in step S501, a contiguous block of memory is pre-allocated on the host machine to store the information of all material points in the subdomains in subdomain order. Its capacity is calculated using the following formula: Total storage capacity = Total number of subdomains × Maximum number of material points in a single subdomain × Number of bytes per unit data. This method ensures that there is a fixed-capacity repository on the host machine sufficient to hold all the data from the entire simulation model after it has been reorganized in the new format.
[0059] Simultaneously, in step S502, a single-use storage space is allocated in the GPU's video memory to store information about matter points in a single subdomain. Its capacity is determined by the following formula: Single-use storage space capacity = Maximum number of matter points in a single subdomain × Number of bytes per unit data. This allocation contrasts sharply with the large arrays on the host side; it provides a temporary cache only for the computation of a single subdomain, and its size is directly constrained by the maximum number of matter points in the aforementioned dynamic partitioning step.
[0060] This embodiment establishes a fixed mapping between a "large host-side cache pool" and a "small device-side workspace." The fixed-size device-side workspace strictly matches the upper limit of the dynamically partitioned constraints, fundamentally preventing memory overflow that might occur during a single computation. Furthermore, the fixed-size memory workspace allows for highly optimized memory access patterns in GPU kernel functions, which is beneficial for improving the execution efficiency of the computing cores.
[0061] In some embodiments of the present invention, such as Figure 6 As shown, step S104, which calculates the time step on the GPU device based on the material point information within the current subdomain, includes: S601. Within the data range of the current subdomain, perform a neighborhood search for each material point and establish a neighborhood list; S602. Calculate bond forces and integrate motion equations based on neighborhood lists to update the kinematic state of material points; S603. Update the damage status of the material points based on the updated kinematic state and the preset damage criteria.
[0062] This embodiment employs a numerical solution framework based on near-field dynamics, but the data field of view for all operations is strictly limited to the current subdomain.
[0063] In step S601, a neighborhood search is first performed for each matter point within the data range of the current subdomain. This step iterates through all other matter points in the subdomain, calculates their distances to the target point, and filters out matter points that fall within a preset "near-field range," ultimately creating a neighborhood list for each matter point. This list forms the basis for all subsequent interaction calculations.
[0064] Then, in step S602, the bond forces of the near-field dynamics core are calculated based on the neighborhood list established in step S601. Specifically, according to the preset constitutive model, the bond forces between point pairs are calculated, thereby updating the kinematic states such as displacement and velocity of each material point.
[0065] In step S603, based on the kinematic state updated in step S602 (based on bond elongation), a preset damage criterion (e.g., determining that the bond is broken when the bond elongation exceeds a certain critical value) is applied to make a judgment, and the macroscopic damage state of the material point is updated accordingly, that is, the proportion of broken bonds among all its bonds is counted.
[0066] This embodiment decomposes the global, highly complex neighborhood search and interaction calculations in traditional near-field dynamics simulations into a series of independent, significantly reduced-scale local subdomain calculations. Step S601 performs the search only within the current subdomain, avoiding the enormous computational overhead of traversing all particles in the domain, achieving a speedup of thousands of times. Secondly, because the calculations are strictly limited to the subdomain data blocks already transferred to GPU memory, extremely high data locality and cache hit rate are ensured, thus fully leveraging the parallel computing performance of the GPU. This decomposition-based computation mode transforms the simulation of ultra-large-scale problems into the sequential processing of multiple small-to-medium-scale, GPU-capable subdomain problems in hardware. This overcomes the absolute limitation of single-GPU memory capacity on simulation scale while maintaining computational accuracy, achieving a significant improvement in overall computational efficiency.
[0067] like Figure 7 As shown, the present invention also provides an electronic device 700. The electronic device 700 includes a processor 701, a memory 702, and a display 703. Figure 7 Only some components of the electronic device 700 are shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead.
[0068] In some embodiments, processor 701 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 702 or process data, such as the GPU-based parallel material damage simulation method of the present invention.
[0069] In some embodiments, processor 701 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 701 may be local or remote. In some embodiments, processor 701 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.
[0070] In some embodiments, memory 702 may be an internal storage unit of electronic device 700, such as a hard disk or memory of electronic device 700. In other embodiments, memory 702 may also be an external storage device of electronic device 700, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 700.
[0071] Furthermore, the memory 702 may include both internal storage units of the electronic device 700 and external storage devices. The memory 702 is used to store application software and various types of data installed on the electronic device 700.
[0072] In some embodiments, display 703 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 703 is used to display information from electronic device 700 and to display a visual user interface. Components 701-703 of electronic device 700 communicate with each other via a system bus.
[0073] In one embodiment, when processor 701 executes a GPU-parallel material damage simulation program in memory 702, the following steps can be performed: Discretize the overall material structure to be simulated into multiple material points; Based on the GPU memory capacity, the overall computational domain containing the material points is dynamically divided into multiple subdomains to obtain the division result. Each subdomain includes an internal computational region and a boundary region extending outward from the computational region by at least one near-field range. Based on the partitioning results, a data structure is constructed to manage the affiliation relationship between material points and subdomains. This data structure is then used to convert the storage of material point information from the overall structure order to the subdomain order. Iterative calculations are performed sequentially by subdomain. For the current subdomain, the material point information required for the calculation of the current subdomain is transmitted from the host to the GPU device. The GPU device performs a calculation of the time step based on the material point information in the current subdomain. The neighborhood search operation is restricted to the data range of the current subdomain. After the calculation is completed, the updated material point information is transmitted back to the host from the GPU device. After the entire calculation is completed, all material point information is converted from being stored in subdomain order back to being stored in the overall structural order, and the final material point information is output as the damage simulation result.
[0074] It should be understood that when the processor 701 executes the GPU-based parallel material damage simulation program in the memory 702, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.
[0075] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 700 mentioned. Electronic device 700 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 700 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).
[0076] Accordingly, this application also provides a computer-readable storage medium for storing computer-readable programs or instructions. When the programs or instructions are executed by a processor, they can implement the steps or functions of the GPU-based parallel material damage simulation methods provided in the above-described method embodiments.
[0077] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.
[0078] The above provides a detailed description of the GPU-based parallel material damage simulation method, apparatus, electronic device, and storage medium provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.
Claims
1. A GPU-parallel material damage simulation method, characterized in that, include: Discretize the overall material structure to be simulated into multiple material points; Based on the GPU memory capacity, the overall computational domain containing the material points is dynamically divided into multiple subdomains to obtain the division result. Each subdomain includes an internal computational region and a boundary region extending outward from the computational region by at least one near-field range. Based on the partitioning results, a data structure for managing the affiliation relationship between material points and subdomains is constructed, and this data structure is used to convert the storage of information of each material point from the overall structure order to the subdomain order. Iterative calculations are performed sequentially by subdomain. For the current subdomain, the material point information required for the calculation of the current subdomain is transmitted from the host to the GPU device. The GPU device performs a calculation of the time step based on the material point information in the current subdomain. The neighborhood search operation is restricted to the data range of the current subdomain. After the calculation is completed, the updated material point information is transmitted back to the host from the GPU device. After the entire calculation is completed, all material point information is converted from being stored in subdomain order back to being stored in the overall structural order, and the final material point information is output as the damage simulation result.
2. The GPU-parallel material damage simulation method according to claim 1, characterized in that, The data structure constructed for managing the affiliation relationship between matter points and subdomains includes: Generate a header pointer array to index the storage starting position of each subdomain computation area; Generate a linked list array to record the sequential relationship of material points within the same subdomain computational region in the form of a linked list.
3. The GPU-parallel material damage simulation method according to claim 1, characterized in that, The process of dynamically dividing the overall computational domain containing the material points into multiple subdomains based on the GPU memory capacity yields the following division results: Obtain the available video memory capacity of the GPU; Based on the available video memory capacity and the amount of video memory required to store information of a single material point, the maximum number of material points that a single subdomain is allowed to contain is calculated and determined. The overall computational domain is partitioned to form multiple subdomains, with the constraint of not exceeding the maximum number of material points, to obtain the partitioning result.
4. The GPU-parallel material damage simulation method according to claim 1, characterized in that, The step of transmitting the material point information required for the current subdomain calculation from the host to the GPU device includes: Transmit all material point information within the computational region of the current subdomain; If the computation to be performed involves neighborhood interactions, then information about the material points within the boundary region of the current subdomain is transmitted simultaneously.
5. The GPU-parallel material damage simulation method according to claim 1, characterized in that, Before the step of converting the storage of material point information from the overall structural order to the subdomain order using this data structure, the GPU-parallel material damage simulation method further includes: In the host's memory, a total storage space is allocated for storing the material point information of all subdomains in subdomain order. The capacity of the total storage space is determined based on the total number of subdomains, the maximum number of material points in a single subdomain, and the number of bytes per unit data. In the GPU device's video memory, a single storage space is allocated for storing information about material points in a single subdomain. The capacity of the single storage space is determined based on the maximum number of material points in a single subdomain and the number of bytes per unit data.
6. The GPU-parallel material damage simulation method according to claim 1, characterized in that, The calculation of a time step based on the material point information in the current subdomain on the GPU device includes: Within the data range of the current subdomain, perform a neighborhood search for each material point and establish a neighborhood list; Based on the neighborhood list, bond force calculation and motion equation integration are performed to update the kinematic state of the material points; The damage status of the material points is updated based on the updated kinematic state and the preset damage criteria.
7. The GPU-parallel material damage simulation method according to any one of claims 1 to 6, characterized in that, The updated material point information includes the updated material point displacement, velocity, and damage status.
8. The GPU-parallel material damage simulation method according to any one of claims 1 to 6, characterized in that, The computational regions of the subdomains are arranged continuously and without overlap in space, and together cover the overall computational domain, wherein a material point belongs to the computational region of a subdomain.
9. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the GPU-based parallel material damage simulation method according to any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps in the GPU-based parallel material damage simulation method according to any one of claims 1 to 8.