A three-dimensional underwater explosion acoustic-structure coupled MPI parallel computing method
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- OCEAN UNIV OF CHINA
- Filing Date
- 2026-07-10
- Publication Date
- 2026-08-07
AI Technical Summary
[0003]针对上述问题，本发明研发了一种三维水下爆炸声固耦合MPI并行计算方法，旨在解决传统声学有限元法在三维水下爆炸声固耦合数值计算中精度不足与计算效率低下等问题
1、本发明创新性地将MPI并行计算技术应用于三维大规模水下爆炸声固耦合问题的求解。本发明采用METIS算法对流体计算域进行区域分解，将其划分为多个计算子域，每个CPU进程承担对应子域的计算任务。在并行求解过程中，计算子域边界单元的数值通量传输由MPI非阻塞通信技术实现，而流体域与结构域之间的数据交互由声固耦合算法和MPI集体通信完成。通过调用多核或多节点计算资源并行求解，本发明有效缩短了程序的计算耗时；
Smart Images
Figure CN122528565A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of computer data processing and computing technology, and particularly relates to a three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method. Background Technology
[0002] The acoustic-structure interaction (ASI) physical processes of underwater explosions involve cutting-edge challenges such as ultra-high pressure, strong discontinuities, and multiphase doping, presenting problems such as transient impact load characteristics and strong nonlinear ASI effects. The dynamic response of structures under far-field underwater explosion loads involves the coupling between acoustic fluids and structures. Shock wave loads can cause structural deformation and damage; furthermore, the structural motion response can affect the pressure distribution and the evolution of cavitation regions in the flow field. To accurately calculate the impact loads generated by underwater explosions and the resulting ASI effects, it is often necessary to discretize the fluid domain into sufficiently fine computational units to capture the strong discontinuities during shock wave propagation and the dynamic response of the structural interfaces. However, most practical underwater explosion engineering problems involve three-dimensional models, leading to a sharp increase in the number of discrete units. The massive computational load significantly extends the computation time, severely limiting computational efficiency. Therefore, traditional serial algorithms based on acoustic finite element methods are no longer sufficient to balance computational accuracy and efficiency, and cannot accurately and efficiently simulate the ASI problem of underwater explosions. Summary of the Invention
[0003] To address the aforementioned problems, this invention develops a parallel computational method for three-dimensional underwater explosion acoustic-structure interaction (MPI), aiming to solve the problems of insufficient accuracy and low computational efficiency of the traditional acoustic finite element method in three-dimensional underwater explosion acoustic-structure interaction numerical calculations.
[0004] This invention provides a parallel computation method for acoustic-structure interaction (MPI) in three-dimensional underwater explosions, characterized by the following process: Step 1: The program reads the relevant data of the discrete flow and structure; Step 2: Use the METIS algorithm to perform a domain decomposition task on the discrete flow field to obtain the partition data of the discrete flow field elements and nodes; Step 3: Using the MPI parallel computing method, the partitioned data of discrete flow field units and nodes are allocated to different MPI processes; each MPI process obtains the corresponding partitioned data of discrete flow field units and nodes. Step 4: Calculate the underwater explosion impact load in the three-dimensional acoustic flow field using the local discontinuous Galerkin method. Each MPI process obtains the hydrodynamic pressure in its corresponding subdomain. Step 5: Based on the fluid dynamic pressure in the corresponding subdomain, the dynamic response physical quantities of the structural nodes are calculated using the explicit finite element method. The acoustic-structure coupling algorithm is used to transfer the fluid dynamic pressure at the acoustic-structure coupling interface to the structural nodes and the acceleration in the dynamic response physical quantities of the structural nodes to the flow field nodes, so as to obtain the numerical flux of the flow field element. Step 6: Before enabling MPI processes to perform data communication, the fluid dynamic pressure and numerical flux are classified and summarized. Each MPI process performs non-blocking communication. At the same time, the parallel I / O mechanism of MPI parallel computing is introduced to output the corresponding dynamic response numerical results of each process.
[0005] Preferably, the discrete flow field data mainly includes node number, node coordinates, and flow field unit nodes; the structural data mainly includes structural node number, node coordinates, and structural unit nodes.
[0006] Preferably, step 2 is specifically performed as follows: S21, the initial graph representing the connection relationship of the original computational domain units is subjected to multiple rounds of coarsening to generate a series of bottom-level graphs with similar topological structures but smaller and smaller scales; S22, At the level of the bottom graph, the multi-level K-way partitioning method is applied to perform partitioning to obtain the partitioning result of the bottom graph; S23, the segmentation results of the bottom layer map are mapped back to the original fine map step by step to obtain the segmentation region map of the original fine map.
[0007] Preferably, step 3 specifically involves the following process: Before the program begins to compute tasks, each process calls the MPI initialization function, the MPI process number acquisition function, and the MPI communication domain size function to establish a parallel execution environment. Each MPI process performs local renumbering of the cells and nodes within its subdomain to obtain the local cell and node numbers for each MPI process.
[0008] Preferably, step 4 specifically involves the following process: The flow field during the explosion is treated as an acoustic fluid, and the propagation characteristics of the underwater explosion impact load in the flow field are calculated using the second-order wave equation. Integration is performed in discrete tetrahedral elements within the computational domain, and the divergence theorem is used to obtain a semi-discrete scheme of the wave equation in the three-dimensional flow field based on the local discontinuous Galerkin method. The hydrodynamic pressure of the fluid in the corresponding subdomain is solved using the semi-discrete scheme of the wave equation.
[0009] Preferably, step 5 specifically involves the following process: S51, taking the hydrodynamic pressure on the acoustic-structure coupling surface as an external load, and calculating the external load of each node of the structure by integrating over the surface of the structure; S52, based on the material constitutive model, calculates the nodal internal forces of the structure under external loads, and uses Newton's second law to calculate the nodal accelerations of the structure. S53 uses the central difference method to update the time domain of the structural nodes, and obtains the velocity and displacement of the structural nodes at the next time step. S54 transmits the acceleration of the structural nodes on the acoustic-structure coupling surface to the fluid nodes, obtaining the numerical flux of the acoustic-structure coupling boundary conditions in the fluid domain at the next moment, including displacement, velocity, and acceleration.
[0010] Preferably, step 5 further includes the following process: Each MPI process calls the MPI send command to send the pressure load and numerical flux of the inter-process boundary cell to the neighboring MPI process. Each MPI process calls the MPI receive command to receive the numerical flux sent from the MPI sender. Each MPI process obtains the numerical flux of its neighboring subdomain boundary cell. Each MPI process invokes MPI collective communication to globally reduce and sum the fluid pressure of the same structural node by the relevant processes on the acoustic-structure coupling interface, and synchronously distributes the accumulated total fluid pressure load of each structural node to all processes, so that each MPI process obtains the load data of the structural node.
[0011] Preferably, step 6 specifically includes: For each boundary-adjacent MPI process, a transmission buffer is opened, and the target process number, transmission data volume, cell number, subdomain boundary cell pressure load and numerical flux are copied to the buffer. Each MPI process obtains the cell fluid pressure load and numerical flux information that needs to be transmitted. Each MPI process performs non-blocking data communication. The sending end of the MPI process calls the MPI_Isend function to send the packaged buffered data to a specific target process, while the receiving end calls the MPI_Irecv function to receive the buffered data from the sending end. After all MPI processes have completed non-blocking communication, each process unpacks the received data and precisely fills the received data at the subdomain boundary. After completing non-blocking communication, the MPI multi-process synchronization function is called to ensure that each process completes non-blocking communication. Each MPI process obtains the pressure load and numerical flux information of the boundary cells of adjacent MPI processes. By introducing the efficient parallel I / O mechanism of MPI parallel computing, each MPI process can output the pressure load, structural displacement, velocity and other dynamic response numerical results of the discrete flow field subdomain unit corresponding to each process without interfering with each other.
[0012] Preferably, after all MPI processes have completed their computational tasks, each MPI process calls the MPI termination function to exit the MPI parallel computing environment.
[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention innovatively applies MPI parallel computing technology to solve large-scale three-dimensional underwater explosion acoustic-structure interaction problems. The invention employs the METIS algorithm to decompose the fluid computational domain into multiple computational subdomains, with each CPU process undertaking the computational task for its corresponding subdomain. During parallel solving, the numerical flux transmission between the boundary cells of the computational subdomains is achieved using MPI non-blocking communication technology, while data interaction between the fluid domain and the structural domain is completed by the acoustic-structure interaction algorithm and MPI collective communication. By utilizing multi-core or multi-node computing resources for parallel solving, this invention effectively shortens the computation time of the program. 2. This invention optimizes the parallel computation method for acoustic-structure interaction (MPI) in three-dimensional large-scale underwater explosions. First, before data communication, the numerical fluxes of boundary cells destined for the same process are categorized and summarized, merging multiple small data transmissions into a single large block transmission, thereby reducing the number of data communication operations and communication time. Second, a non-blocking communication interface is used to perform data communication of the numerical fluxes of boundary cells, effectively reducing process waiting time and improving program computational efficiency. Finally, an efficient parallel I / O mechanism is employed, setting independent input and output channels for each process, improving the speed at which the program reads flow field cell nodes and writes cell pressure load data. Attached Figure Description
[0014] To more clearly illustrate the technical solutions of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the following description is only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0015] Figure 1 This is a flowchart of a three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method according to the present invention.
[0016] Figure 2 This is a mapping diagram of the master surface node to the fluid slave surface node of the structure of the present invention.
[0017] Figure 3 This is a mapping diagram of fluid flow from surface nodes to main surface nodes of the structure according to the present invention.
[0018] Figure 4 This is a schematic diagram of the parallel I / O process of the present invention.
[0019] Figure 5 This is a comparison diagram of the acoustic-structure interaction (MPI) parallel calculation method for three-dimensional underwater explosion in this invention and the displacement response of the spherical shell measuring point calculated by ABAQUS. Detailed Implementation
[0020] 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 specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. 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.
[0021] This invention employs the LDG method to capture the propagation characteristics of impact loads in a flow field, uses the explicit finite element method (EFEM) to solve the nonlinear dynamic response of the structure, and combines an acoustic-structure interaction algorithm to achieve data interaction between fluid domain pressure loads and structural domain motion boundary conditions. To address large-scale computational requirements, this invention introduces MPI parallel technology, significantly reducing computation time and improving computational efficiency by fully utilizing multi-core or multi-node computing resources for parallel solving. Comparison of the calculation results with those from ABAQUS software verifies the effectiveness and accuracy of the proposed method. Comparison of computation time under different CPU core counts verifies the acceleration performance and parallel efficiency of the proposed method. This invention proposes a three-dimensional underwater explosion acoustic-structure interaction MPI parallel computation method, such as... Figure 1 As shown: Step 1: The program reads the relevant data of the discrete flow and structure; Step 2: Based on the discrete flow field data read in Step 1, the METIS algorithm is used to perform a region decomposition task on the discrete flow field to obtain the partition data of discrete flow field units and nodes. Step 3: Based on the partitioned data of discrete flow field units and nodes obtained in Step 2, MPI parallel computing technology is introduced to allocate the partitioned data of discrete flow field units and nodes to different MPI processes. Each MPI process obtains the corresponding partitioned data of discrete flow field units and nodes. Step 4: Based on the partition data of the corresponding discrete flow field unit and node obtained by each MPI process in Step 3, the underwater explosion impact load in the three-dimensional acoustic flow field is calculated using the local discontinuous Galerkin method. Each MPI process obtains the hydrodynamic pressure in its corresponding subdomain. Step 5: Based on the structural unit and node data read in Step 1 and the fluid dynamic pressure in the corresponding subdomain calculated by each MPI process in Step 4, the dynamic response of the structure is calculated using the explicit finite element method to obtain the physical quantities of dynamic response of the structural node displacement, velocity, and acceleration; the acoustic-structure interaction algorithm is used to transfer the fluid dynamic pressure load at the acoustic-structure interaction interface to the structural node and the acceleration of the structural node to the flow field node. Step 6: Based on the discrete flow field elements and node partitioning data obtained by each MPI process in Step 3, the fluid dynamic pressure calculated by each MPI process in Step 4 within its corresponding subdomain, and the structural displacement, velocity, and acceleration physical quantities calculated in Step 5, the process proceeds as follows: First, before enabling data communication by the MPI processes, all boundary element pressure loads and numerical fluxes required to be sent to the same target process number are categorized and summarized. Second, each MPI process performs non-blocking communication to realize the data transmission of boundary element pressure loads and numerical fluxes between processes. Finally, an efficient parallel I / O mechanism for MPI parallel computing is introduced, allowing each MPI process to simultaneously output the pressure loads and structural displacement, velocity, and dynamic response numerical results of the discrete flow field subdomain elements corresponding to each process without interference. This enables the calculation of large-scale three-dimensional underwater explosion acoustic-structure interaction problems.
[0022] The specific implementation process of the present invention will be further described below with reference to specific embodiments.
[0023] Step 1: Read relevant data for the discrete flow field and structure. The program reads relevant data for the discrete flow field and structure. The discrete flow field data mainly includes node numbers, node coordinates, and the composition of flow field unit nodes; the structural data mainly includes structural node numbers, node coordinates, and the composition of structural unit nodes. This step provides basic data for subsequent flow field region division, MPI parallel calculation of flow field dynamic pressure, and structural response calculation.
[0024] Step 2: Based on the unit node file in the discrete flow field from Step 1, the METIS algorithm is used to perform a region decomposition task on the discrete flow field to obtain the partition data of the discrete flow field units and nodes.
[0025] S2-1. Graph Coarsening. The initial graph, which represents the connection relationships of the original computational domain units, undergoes multiple rounds of coarsening to generate a series of bottom-level graphs with similar topologies but progressively smaller scales.
[0026] S2-2, Coarse Graph Partitioning. At the level of the bottom-level graph, a multi-level K-way partitioning method is applied to perform partitioning, resulting in the partitioning of the bottom-level graph.
[0027] S2-3, Graph Refinement. The segmentation results of the bottom-level graph are mapped back to the original fine-scale graph level by level to obtain the partitioned region map of the original fine-scale graph.
[0028] Through the above-described region partitioning steps, the METIS algorithm can maintain load balancing across subdomains while minimizing communication overhead between processes.
[0029] Step 3: Based on the partitioned data of discrete flow field units and nodes obtained in Step 2, MPI parallel computing technology is introduced to allocate the partitioned data of discrete flow field units and nodes to different MPI processes. Each MPI process obtains the corresponding partitioned data of discrete flow field units and nodes.
[0030] S3-1. Before the program starts calculating the task, each process calls the MPI initialization function, the MPI process number acquisition function, and the MPI communication domain size function to establish a parallel execution environment.
[0031] S3-2. Each MPI process locally renumbers the cells and nodes within its subdomain to obtain the local cell and node numbers for each MPI process. This avoids redundant loading of global mesh information.
[0032] Step 4: Based on the partition data of the corresponding discrete flow field unit and node obtained by each MPI process in Step 3, use the LDG method to calculate the underwater explosion impact load in the three-dimensional acoustic flow field, and obtain the fluid dynamic pressure in its corresponding subdomain for each MPI process.
[0033] In far-field underwater explosions, the flow field can be considered as an acoustic fluid. The shock wave from an underwater explosion in a three-dimensional acoustic flow field satisfies a second-order wave equation, which can be expressed as: (1) in, p It represents the dynamic pressure of the fluid, and Δ represents the Laplace operator. This represents the second derivative of the fluid's dynamic pressure with respect to time. c It is the fluid in the flow field The speed of sound inside, . and These are the bulk modulus and density of the fluid, respectively.
[0034] Multiply both sides of formula (1) by the basis functions. , , and Then, the tetrahedral elements are discretized within the computational domain. K By performing integration and applying the divergence theorem, a semi-discrete scheme for the three-dimensional LDG wave equation can be obtained: (2) (3) (4) (5) in, , , as well as They are three-dimensional spatial discrete units. K Approximate solution in It is the unit normal vector of the tetrahedral element pointing outward from the boundary. , , as well as It is the numerical flux on the boundary of a discrete unit in a three-dimensional flow field.
[0035] Step 5: Based on the structural element and node data read in Step 1 and the fluid pressure in the corresponding subdomain calculated by the LDG method for each MPI process in Step 4, the dynamic response physical quantities of the structure are calculated using the explicit finite element method. An acoustic-structure interaction algorithm is used to transfer the fluid dynamic pressure load at the acoustic-structure interaction interface to the structural nodes, and the acceleration of the structural nodes to the flow field nodes, thus obtaining the numerical flux of the flow field elements.
[0036] S4-1. Use EFEM to calculate the physical quantities at the structural nodes. The motion state of each discrete node in the structure follows Newton's second law: (6) in, It is the concentrated quality of the nodes. It is structural acceleration. and These are the external forces acting on the structural nodes and the internal forces at the nodes caused by element stresses, respectively.
[0037] structural nodes n External forces The result can be obtained by integrating the hydrodynamic pressure projected onto the structural surface on the acoustic-structure coupling plane. (7) in, For structural wet surfaces, The dynamic pressure of the flow field on the wetted surface of the structure.
[0038] structural nodes n Internal force The calculation expression is: (8) in, It is the spatial derivative of the basis functions. Represents structural units. It is the element stress, which can be determined by the material's elastoplastic constitutive model and the velocity-strain relationship.
[0039] The central difference method is used to integrate the relevant physical quantities of the structural nodes in the time domain, and then the velocities and displacements of the structural nodes are explicitly updated. : (9) (10) (11) in, For the structure time step, superscript n Representing the n Each time step. Represents the displacement of structural nodes. Indicates the velocity of structural nodes. This represents the acceleration of structural nodes.
[0040] S4-2. The entire acoustic-structure interaction algorithm can be decomposed into two physical modules: calculating acoustic fluid pressure and structural dynamic response. In each time step, the entire calculation follows the following calculation process: (1) Fluid domain solution: The fluid module is calculated independently to calculate the fluid dynamic pressure at the acoustic-structure interaction interface at the current moment. (2) Load transfer: Through the acoustic-structure interaction interface, the fluid pressure at the interface is converted into the external load of the wet surface of the structure and accurately transferred to the structural module. (3) Structural domain calculation: After receiving the pressure load, the structural module calculates the displacement, velocity and acceleration of all nodes of the structure. (4) Variable feedback: The structural module feeds back the calculated normal acceleration of the wet surface of the structure to the fluid module as the boundary condition at the acoustic-structure interaction point of the fluid domain at the next moment, thereby updating the load state of the flow field.
[0041] To address the issue of inconsistent spatial discretization between the fluid and structural domain mesh nodes at the acoustic-solid interface, a fluid-structure mesh mapping scheme is adopted to transfer physical quantities between the fluid and structural domains.
[0042] At the acoustic-structure interaction interface, the structural domain mesh is defined as the master surface, and the fluid domain mesh is defined as the slave surface. The mesh nodes on the master surface of the structural domain are mapped to the slave surface of the fluid domain, as follows: Figure 2 As shown. Nodes on the main surface of the domain. x n Mapping to the fluid domain, we obtain node P from the surface. x n According to the mapping point P( x n The nodal physical quantity values within the fluid domain element can be obtained by interpolating the basis functions at node P( x n The value of ). Nodes on the principal surface of the domain. x n With fluid domain from surface mapping node P( x n Following dynamic pressure balance, (12) in, For mapping points The basis functions obtained by interpolation based on their location. Let be the unit normal vector pointing from the principal surface of the structural domain to the secondary surface of the fluid domain. This refers to the pressure at the surface nodes within the fluid domain. (Superscript) i Represents the mapping point The vertex of the fluid unit in which it is located. It is the external force exerted on the structural nodes on the principal surface of the structural domain.
[0043] Similarly, the fluid domain is mapped from the mesh nodes on the surface to the main surface of the structural domain, such as... Figure 3 As shown. The fluid domain extends from the nodes on the surface. x m Mapping to the principal surface of the structural domain yields node P( x m According to the mapping point P( x m The nodal physical quantities within the structural domain element can be obtained by interpolating the basis functions to obtain the nodal P( x m The value of ) represents the fluid domain from the nodes on the surface. x m Mapping node P(to the principal face of the structural domain) x m Following the condition of continuity of motion, (13) in, This is the acceleration of the principal nodes of the structural domain. (Superscript) i Represents the mapping point The vertex of the structural unit. It is a fluid domain from the surface fluid The acceleration at that point.
[0044] At the acoustic-structure interaction interface, the dynamic pressure and acceleration at any node of the fluid satisfy the following: (14) S4-3. Each MPI process calls the MPI send command to send the pressure load and numerical flux of the inter-process boundary cell to the adjacent MPI process. Each MPI process calls the MPI receive command to receive the numerical flux sent from the MPI sender. Each MPI process can obtain the pressure load and numerical flux of its adjacent subdomain boundary cell.
[0045] S4-4. Each MPI process calls the MPI collective communication to perform global reduction summation of the pressure loads on the same structural node by the relevant processes on the acoustic-structure coupling interface, and synchronously distributes the accumulated total load of each structural node to all processes. Each MPI process can obtain the load data of the structural node.
[0046] Step 6: Based on the partitioned data of the corresponding discrete flow field elements and nodes obtained by each MPI process in Step 3, the fluid dynamic pressure calculated by each MPI process in Step 4 within its corresponding subdomain, and the physical quantities of dynamic response such as structural displacement and velocity calculated in Step 5. First, before enabling data communication by the MPI processes, all boundary element pressure loads and numerical fluxes required to be sent to the same target process number are categorized and summarized. Second, each MPI process performs non-blocking communication to realize the data transmission of boundary element pressure loads and numerical fluxes between processes. Finally, an efficient parallel I / O mechanism for MPI parallel computing is introduced, enabling each MPI process to simultaneously output the pressure loads and dynamic response numerical results of the discrete flow field subdomain elements corresponding to each process without interference.
[0047] S6-1. Before data communication in MPI parallel computing, all boundary cell pressure loads and numerical fluxes required by the same target process number are categorized and summarized according to the adjacency relationships. For each boundary-adjacent MPI process, a transmission buffer is allocated, and the target process number, transmission data volume, cell number, and subdomain boundary cell numerical flux are copied into this buffer. Each MPI process can obtain the cell pressure load and numerical flux information that needs to be transmitted.
[0048] S6-2. Each MPI process performs non-blocking data communication. The sending end of the MPI process calls the MPI_Isend function to send the packaged buffered data to the specific target process, while the receiving end calls the MPI_Irecv function to receive the buffered data from the sending end. After all MPI processes have completed non-blocking communication, each process unpacks the received data and precisely fills the received data at the subdomain boundary. After completing non-blocking communication, the MPI multi-process synchronization function is called to ensure that each process completes non-blocking communication. Each MPI process can obtain the pressure load and numerical flux information of the boundary cells of adjacent MPI processes.
[0049] S6-3. Introducing the efficient parallel I / O mechanism of MPI parallel computing, each MPI process can simultaneously output the pressure load, structural displacement, velocity, and other dynamic response numerical results of the discrete flow field subdomain elements corresponding to each process without interference. Figure 4 As shown.
[0050] Based on the numerical results of pressure loads, structural displacements, velocities, and other dynamic responses of the corresponding discrete flow field subdomain elements output in step 3 (entering the MPI parallel computing environment) and step 6 (each process outputting its corresponding discrete flow field subdomain elements), each MPI process calls the MPI termination function to exit the MPI parallel computing environment after all MPI processes have completed their computational tasks.
[0051] To verify the effectiveness and parallel acceleration performance of the calculation method described in this invention, a typical example of the dynamic response process of a plane wave impacting an immersed spherical shell was simulated. The calculation parameters are as follows: the flow field calculation domain is [-2m, 3m]×[-2m, -2m]×[-2m, -2m], the plane wave impact load is 1.4MPa, the center of the spherical shell is located at the origin, the thickness is 0.02m, and the density is 7784.5kg / m³. 3 The elastic modulus is 206.8 MPa, and the Poisson's ratio is 0.3. First, the displacement response at point (0, 0, 1m) on the spherical shell was extracted. The calculation results of this invention were compared with those of ABAQUS software to verify the effectiveness of the method described in this invention. The results show that the calculation results of the method described in this invention agree well with the calculation results of ABAQUS software. Figure 5 As shown in the figure. This proves that the present invention can accurately solve the acoustic-structure interaction problem of underwater explosions.
[0052] Secondly, calculations were performed using different numbers of CPU cores (with the calculation time using 64 CPU cores as a reference) to test its acceleration performance and parallel efficiency. The results show that as the number of CPU cores increases, the speedup ratio exhibits a good linear growth trend, and it still maintains 93% parallel efficiency in a 256-core computing scenario, as shown in Table 1. This demonstrates that the present invention has good acceleration performance and parallel scalability.
[0053] Table 1. Speedup and parallel efficiency of the parallel computing method for acoustic-structure interaction (MPI) in three-dimensional underwater explosions.
[0054] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
[0055] While the specific embodiments of the present invention have been described above, they are not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.
Claims
1. A parallel computation method for acoustic-structure interaction (MPI) in three-dimensional underwater explosions, characterized in that, The process includes the following: Step 1: The program reads the relevant data of the discrete flow and structure; Step 2: Use the METIS algorithm to perform a domain decomposition task on the discrete flow field to obtain the partition data of the discrete flow field elements and nodes; Step 3: Using the MPI parallel computing method, the partitioned data of discrete flow field units and nodes are allocated to different MPI processes; each MPI process obtains the corresponding partitioned data of discrete flow field units and nodes. Step 4: Calculate the underwater explosion impact load in the three-dimensional acoustic flow field using the local discontinuous Galerkin method. Each MPI process obtains the hydrodynamic pressure in its corresponding subdomain. Step 5: Based on the fluid dynamic pressure in the corresponding subdomain, the dynamic response physical quantities of the structural nodes are calculated using the explicit finite element method. The acoustic-structure coupling algorithm is used to transfer the fluid dynamic pressure at the acoustic-structure coupling interface to the structural nodes and the acceleration in the dynamic response physical quantities of the structural nodes to the flow field nodes, so as to obtain the numerical flux of the flow field element. Step 6: Before enabling MPI processes to perform data communication, the fluid dynamic pressure and numerical flux are classified and summarized. Each MPI process performs non-blocking communication. At the same time, the parallel I / O mechanism of MPI parallel computing is introduced to output the corresponding dynamic response numerical results of each process.
2. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 1, characterized in that: Discrete flow field data mainly includes node numbers, node coordinates, and flow field element nodes; structural data mainly includes structural node numbers, node coordinates, and structural element nodes.
3. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 1, characterized in that: The specific process of step 2 is as follows: S21, the initial graph representing the connection relationship of the original computational domain units is subjected to multiple rounds of coarsening to generate a series of bottom-level graphs with similar topological structures but smaller and smaller scales; S22, At the level of the bottom graph, the multi-level K-way partitioning method is applied to perform partitioning to obtain the partitioning result of the bottom graph; S23, the segmentation results of the bottom layer map are mapped back to the original fine map step by step to obtain the segmentation region map of the original fine map.
4. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 1, characterized in that: The specific process of step 3 is as follows: Before the program begins to compute tasks, each process calls the MPI initialization function, the MPI process number acquisition function, and the MPI communication domain size function to establish a parallel execution environment. Each MPI process performs local renumbering of the cells and nodes within its subdomain to obtain the local cell and node numbers for each MPI process.
5. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 1, characterized in that: The specific process of step 4 is as follows: The flow field during the explosion is treated as an acoustic fluid, and the propagation characteristics of the underwater explosion impact load in the flow field are calculated using the second-order wave equation. Integration is performed in discrete tetrahedral elements within the computational domain, and the divergence theorem is used to obtain a semi-discrete scheme of the wave equation in the three-dimensional flow field based on the local discontinuous Galerkin method. The hydrodynamic pressure of the fluid in the corresponding subdomain is solved using the semi-discrete scheme of the wave equation.
6. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 1, characterized in that: The specific process of step 5 is as follows: S51, taking the hydrodynamic pressure on the acoustic-structure coupling surface as an external load, and calculating the external load of each node of the structure by integrating over the surface of the structure; S52, based on the material constitutive model, calculates the nodal internal forces of the structure under external loads, and uses Newton's second law to calculate the nodal accelerations of the structure. S53 uses the central difference method to update the time domain of the structural nodes, and obtains the velocity and displacement of the structural nodes at the next time step. S54 transmits the acceleration of the structural nodes on the acoustic-structure coupling surface to the fluid nodes, obtaining the numerical flux of the acoustic-structure coupling boundary conditions in the fluid domain at the next moment, including displacement, velocity, and acceleration.
7. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 6, characterized in that: Step 5 also The process includes the following: Each MPI process calls the MPI send command to send the numerical flux of the inter-process boundary cell to the neighboring MPI process, and each MPI process calls the MPI receive command to receive the numerical flux sent from the MPI sender; each MPI process obtains the pressure load and numerical flux of its neighboring subdomain boundary cell. Each MPI process invokes MPI collective communication to globally reduce and sum the fluid pressure of the same structural node by the relevant processes on the acoustic-structure coupling interface, and synchronously distributes the accumulated total fluid pressure load of each structural node to all processes, so that each MPI process obtains the load data of the structural node.
8. The three-dimensional underwater explosion acoustic-structure coupling MPI parallel computing method as described in claim 1, characterized in that: Step 6 specifically includes: For each boundary-adjacent MPI process, a transmission buffer is opened, and the target process number, transmission data volume, cell number, subdomain boundary cell pressure load and numerical flux are copied to the buffer. Each MPI process obtains the cell fluid pressure load and numerical flux information that needs to be transmitted. Each MPI process performs non-blocking data communication. The sending end of the MPI process calls the MPI_Isend function to send the packaged buffered data to a specific target process, while the receiving end calls the MPI_Irecv function to receive the buffered data from the sending end. After all MPI processes have completed non-blocking communication, each process unpacks the received data and accurately fills the received data at the subdomain boundary. After completing non-blocking communication, the MPI multi-process synchronization function is called to ensure that each process completes non-blocking communication. Each MPI process obtains the pressure load and numerical flux information of the boundary cells of the MPI processes adjacent to it.