A large complex structure sensitivity calculation method, system, device and medium

Abstract

The application discloses a large complex structure sensitivity calculation method, system, device and medium, and relates to the technical field of structural parameter design. The method comprises the following steps: determining a target component in an original model through interactive selection, and obtaining an initial state; performing boundary identification on the target component, and extracting the mechanical environment of boundary nodes; generating a reference submodel based on the target component, and applying the mechanical environment to the reference submodel; selecting perturbation design parameters in the reference submodel, and generating corresponding derivative submodels; submitting the reference submodel and all the derivative submodels to a solver for batch calculation, and automatically obtaining a result file generated by each model calculation; extracting key performance indicators from each result file, and calculating the sensitivity of each design parameter to the performance indicators. The application can solve the problem that a structural engineer needs to quickly evaluate the influence degree of design variables on global performance response in the product optimization process, thereby efficiently guiding the structural optimization design.

Description

Technical Field

[0001] This invention relates to the field of structural parameter design technology, and in particular to a method, system, device and medium for calculating the sensitivity of large and complex structures. Background Technology

[0002] With the increasing demands of modern industry for product performance, lightweighting, and reliability, the optimal design of large and complex structures (such as aerospace vehicles, automobile bodies, and heavy machinery) has become a core aspect of engineering development. Structural optimization typically involves complex coupling relationships between numerous design variables (such as plate and shell thickness, beam section properties, and material parameters) and various performance responses (such as stress, displacement, and frequency). In this process, sensitivity analysis, as a crucial foundation for optimization design, is used to quantify the impact of changes in design variables on key performance indicators, thereby identifying the parameters most sensitive to structural behavior, guiding the optimization direction, improving convergence efficiency, and reducing computational costs.

[0003] However, directly calculating the sensitivity of design parameters based on the global finite element model in large and complex structures faces significant challenges in computational efficiency and resource consumption. Due to the massive scale of the structure and the huge number of elements and degrees of freedom, the global model needs to be resolved for each parameter perturbation, resulting in excessively long individual sensitivity calculations and making it difficult to apply efficiently in engineering practice. This is especially true when parameter evaluation of multiple local components or detailed features is required, where the full model analysis method becomes cumbersome and uneconomical. Furthermore, traditional methods typically rely on manual intervention for sub-model extraction, boundary condition processing, and result extraction, which is cumbersome, error-prone, and difficult to integrate and automate.

[0004] While existing finite element analysis (FEM) software offers some sensitivity analysis capabilities, it typically relies on global model perturbations, resulting in high computational resource consumption and insufficient flexibility and detail in assessing responses to changes in local parameters. Particularly when dealing with assemblies composed of multiple components, the lack of efficient and automated multi-scale analysis mechanisms hinders the rapid and effective transfer of macroscopic model analysis results to local sub-models for sensitivity quantification, severely limiting its application in the optimization design of complex structures. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device and medium for calculating the sensitivity of large and complex structures, in order to solve or improve at least one of the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for calculating the sensitivity of large and complex structures, comprising: The target component in the original model is determined through interactive selection, and the initial state is obtained; Boundary identification is performed on the target component, and the mechanical environment of the boundary nodes is extracted; A baseline sub-model is generated based on the target component, and the mechanical environment is applied to the baseline sub-model to reproduce the force state in the original model; Based on the stress state, perturbation design parameters are selected in the baseline sub-model, and a derived sub-model corresponding to the perturbation design parameters is generated. The baseline sub-model and all derived sub-models are submitted to the solver for batch calculation. The solver monitors and waits for all calculation tasks to complete, and then automatically obtains the result file generated by each model calculation. Key performance indicators are extracted from each of the result files. Based on the calculation formula: Sensitivity = (Stress change / Original stress) / (Perturbation design parameter change), the sensitivity of each design parameter to the performance indicators is calculated, and the target component and its corresponding information are stored in a structured data file.

[0007] Optionally, the step of determining the target component in the original model through interactive selection and obtaining the initial state specifically involves: In the finite element post-processing environment, an interactive selection interface is provided, allowing users to interactively select target components through UI operations, and automatically acquire and store the initial key performance indicators of the target components by setting scripts.

[0008] Optionally, the step of performing boundary identification on the target component and extracting the mechanical environment of the boundary nodes specifically includes: The target component is identified from the overall model, and the boundary nodes connecting the target component to other components in the model are automatically identified. The displacement or load data of the boundary nodes in the initial static analysis results are automatically queried and extracted.

[0009] Optionally, the step of generating a reference sub-model based on the target component and applying the mechanical environment to the reference sub-model to reproduce the force state in the original model specifically includes: The CAE environment automatically switches from post-processing mode to pre-processing mode; based on the target component, an independent reference sub-model containing only the target component and related properties and materials is generated, and the displacement or load data in the mechanical environment is applied to the reference sub-model as boundary conditions to reproduce the stress state in the original model.

[0010] Optionally, the step of selecting perturbation design parameters in the baseline sub-model based on the stress state and generating a derived sub-model corresponding to the perturbation design parameters specifically includes: The design parameters in the baseline sub-model are perturbed in a programmed manner, and a corresponding derived sub-model is generated for each perturbed parameter, which integrates the component name, boundary load and original stress.

[0011] This invention also provides a sensitivity calculation system for large and complex structures, comprising: The component selection and state acquisition unit is used to determine the target component in the original model through interactive selection and to acquire the initial state. A boundary recognition and mechanical environment extraction unit is used to perform boundary recognition on the target component and extract the mechanical environment of the boundary nodes; The stress state reproduction unit is used to generate a reference sub-model based on the target component and apply the mechanical environment to the reference sub-model to reproduce the stress state in the original model; The data perturbation unit is used to select perturbation design parameters in the reference sub-model based on the force state, and generate a derived sub-model corresponding to the perturbation design parameters. The batch calculation unit is used to submit the baseline sub-model and all derived sub-models to the solver for batch calculation, monitor and wait for all calculation tasks to complete, and then automatically obtain the result file generated by each model calculation. The sensitivity calculation unit is used to extract key performance indicators from each of the result files, calculate the sensitivity of each design parameter to the performance indicators according to the calculation formula: sensitivity = (stress change / original stress) / (perturbation design parameter change), and store the target component and corresponding information in a structured data file.

[0012] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to perform the sensitivity calculation method for large and complex structures described above.

[0013] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the sensitivity calculation method for large and complex structures as described above.

[0014] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This invention discloses a method, system, device, and medium for calculating the sensitivity of large and complex structures. The method includes: interactively selecting and determining target components in the original model and obtaining the initial state; identifying the boundaries of the target components and extracting the mechanical environment of the boundary nodes; generating a baseline sub-model based on the target components and applying the mechanical environment to the baseline sub-model; selecting perturbation design parameters in the baseline sub-model and generating corresponding derived sub-models; submitting the baseline sub-model and all derived sub-models to a solver for batch calculation and automatically obtaining the result files generated by each model calculation; extracting key performance indicators from each result file and calculating the sensitivity of each design parameter to the performance indicators. This invention can solve the problem that structural engineers need to quickly assess the impact of design variables on the global performance response during product optimization, thereby efficiently guiding structural optimization design. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0016] Figure 1 This is a flowchart illustrating the sensitivity calculation method for large and complex structures in this embodiment; Figure 2 This is a schematic diagram of a linearly distributed load applied to a second-order Lagrange element node in this embodiment; Figure 3 This is a contour map showing the axial stress distribution in this embodiment; Figure 4 This is a schematic diagram of stress redistribution in this embodiment; Figure 5 This is a schematic diagram of the hierarchical structure of the objects being analyzed in this embodiment; Figure 6 This is a bar chart showing the sensitivity calculation results in this embodiment. Detailed Implementation

[0017] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The purpose of this invention is to provide a method, system, device and medium for calculating the sensitivity of large and complex structures, in order to solve or improve at least one of the above-mentioned technical problems.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, this invention provides a method for calculating the sensitivity of large and complex structures, executed in a HyperWorks environment where initial static analysis has been completed and results obtained. The specific steps include: Step 1: Interactive selection and initial state acquisition of the target component In the post-processing (HyperView) environment, an interactive selection interface is provided to receive one or more components as analysis objects. Component selection is achieved by calling the Interactive_Selection API. Initial key performance indicators of the analysis objects, such as maximum stress values, are automatically acquired and stored via scripts. As a specific implementation, the specific processing steps include: In the method described in this scheme, determining the target component to obtain the initial state operation is the first step of the multi-scale sub-model method, namely, local problem definition and state capture.

[0021] First, an interactive decomposition of the problem domain is performed. User clicks divide the global model topologically into two parts: the local subdomain (the substructure the user is interested in) and the external domain (the rest of the model that is ignored). Then, boundary state variable capture is performed, which is the process of capturing the boundary conditions of the local subdomain from the solution of the global model. The boundary node displacements or loads extracted from the mechanical environment serve as the mechanical interface between the submodel and the external domain; the benchmark response is the captured initial maximum stress, representing the performance benchmark of the local subdomain under undisturbed conditions.

[0022] Step 2: Analyze the automatic identification of boundaries and extraction of the mechanical environment. The components selected in Step 1 are identified from the overall model as the analysis objects. The `find_node` API is used to automatically identify the boundary nodes connecting the analysis objects to other components in the model, and the `get_node_dis_value` API is used to automatically query and extract the displacement or load data of these boundary nodes from the initial static analysis results. This step aims to extract boundary conditions, and the specific processing procedure is as follows: In the method described in this scheme, boundary conditions are extracted using a multi-scale sub-model approach, based on Saint-Venant's principle: far from the load application region, stress distribution depends only on the resultant force and is independent of the specific distribution of the load. In finite element discretization problems, if the stress is set as a continuous boundary load, but the stress is always applied to the mesh nodes, such as... Figure 2 The linear distributed load is shown, and how it is applied to the nodes of a second-order Lagrange element with side length L.

[0023] The same nodal load can correspond to countless load distributions, but the finite element problem will always have the same solution. According to Saint-Venant's principle, such loads should essentially produce the same stress field as long as they are spaced a certain distance apart. Since the size of the load redistribution region is based on the element surface, the linear dimension is essentially a single element layer within the structure. Therefore, the solution at the outermost layer of the element may not match the actual load, while the solutions at the inner layers will be identical.

[0024] For example: A rectangular plate is subjected to a boundary load with an exponentially distributed stress distribution, and calculations are performed using a refined mesh. The contour plot of the axial stress distribution is shown below. Figure 3 As shown.

[0025] The study examines the difference between the stress distribution obtained above and that obtained using a coarsened mesh. The axial stress error at different scales is shown in the figure below. The finer the mesh, the smaller the error; after the first element layer, the error decreases rapidly. Figure 4 In fact, this is a combined effect of mesh convergence and stress redistribution conforming to Saint-Venant's principle.

[0026] According to Saint-Venant's principle, if the boundary condition extraction location in the global model is sufficiently far from the analysis location in the local model, its influence can be ignored. Based on simulation examples, when analyzing stress concentration around a hole, the truncation boundary of the sub-model should be at least 3 to 5 times the hole diameter from the hole edge; if it is at a plate connection, it should be 3 to 5 times the plate thickness away from the connection. The selection of the sub-model boundary itself requires engineering judgment. If the boundary is too close, Saint-Venant's principle does not hold, leading to boundary effects contaminating the analysis results; if the boundary is too far, the sub-model size becomes too large, losing the efficiency advantage of multi-scale analysis. This scheme describes the sub-model method, which, based on existing global model analysis results, performs more detailed finite element analysis on the local model. The finite element mesh scale meets the above requirements, allowing the extraction of displacement boundary conditions from the global model and their application to the local model.

[0027] Step 3: Generation of the baseline sub-model and application of boundary conditions The set_page_client API enables automatic switching from post-processing (HyperView) mode to pre-processing (HyperMesh) mode in the CAE environment. Based on the component selected in step one, an independent reference sub-model containing only that component and its related properties and materials is generated. The boundary node displacement or load data extracted in step two are applied as boundary conditions to the reference sub-model to reproduce its stress state in the original model.

[0028] As a specific implementation method, in the method described in this scheme, if a sub-model is to be analyzed, in order to ensure that the calculation results of the sub-model are consistent with those in the "global model", the correct boundary conditions must be applied to the truncated boundary. In finite element theory, the classification and application methods of boundary conditions originate from the process of deriving the "weak form" (integral equation) from the "strong form" (differential equation).

[0029] For example, the equilibrium equations for a one-dimensional rod are in strong form as follows: in, Indicates the distance along the axis of the member; The main variable to be solved is the member in... Axial displacement at the location; Indicates axial strain; For elastic modulus, Let be the cross-sectional area of ​​the rod. Indicates the tensile stiffness of the rod; This refers to the external force acting per unit length.

[0030] Based on Galerkin's method, a weight function v(x) is introduced, and a weighted integral is performed on the strong form: in, The total length of the rod; The weight function.

[0031] Reduce the order of the derivative of the principal variable u by integration by parts: in, It indicates a virtual response.

[0032] Substituting back to the original expression yields the weak form: Boundary terms: in, Indicates the axial force in the rod; This indicates the axial force at the left and right ends of the member. This represents the weight function values ​​of the rod at the left and right ends.

[0033] The natural boundary condition (force P) appears in the boundary integral term and acts directly on the load vector F when assembling the finite element equations; the forced boundary condition (displacement u) is not reflected in the integral, but is applied as an external constraint on the solution space.

[0034] Accordingly, in sub-model techniques, there are two main strategies for applying boundary conditions. The displacement-based method extracts the displacements of all nodes on the truncation boundary from the global model and applies them as forced boundary conditions to the sub-model. This is the most commonly used and robust method because displacement is a fundamental solution variable in finite element analysis, and its continuity and accuracy are usually well guaranteed at the element boundaries. The stress-based method extracts the stress or nodal forces on the truncation boundary from the global model and applies them as "natural boundary conditions" to the boundaries of the sub-model. Stress is the derivative of displacement, which typically results in a "one-order reduction" in accuracy in finite element calculations and may be discontinuous at the element boundaries.

[0035] In the method described in this scheme, the original external forces (natural boundary conditions) need to be removed to ensure that the sub-model is driven entirely by displacements (forced boundary conditions). This ensures that the deformation compatibility of the sub-model at the truncation boundary is consistent with the global model, and the displacement values ​​in the global model are applied to the sub-model as non-zero forced displacements. The method described in this scheme directly imposes constraints on the main variables, i.e., on the governing equations. Modifications were made. Among them... It is the overall stiffness matrix of the system; It is the system's displacement vector; It is the load vector of the system.

[0036] Commonly used constraint application methods in linear eigenvalue buckling analysis include the penalty function method and the Lagrange multiplier method.

[0037] The penalty function method is mainly used in commercial software to transform constrained optimization problems into unconstrained optimization problems: First, a very large number C (penalty factor, for example, 10) is selected. 10 ~10 20 Secondly, the diagonal elements of the K matrix are modified to... Finally, modify the corresponding element of F. This method is easily implemented through programming and is an approximate method. The Lagrange multiplier method is a theory for handling multi-point constrained problems. It introduces the Lagrange multiplier λ, adding the constraint equations as unknowns to the system of equations, forming a matrix equation: in, These are Lagrange multipliers, corresponding to constraint forces; The coefficient matrix representing the constraint equations; This represents the value on the right-hand side of the constraint equation. The equation satisfies the constraints precisely, but increases the equation size.

[0038] The method described in this scheme uses the Lagrange multiplier method to apply constraints and combines it with the Lanczos method in buckling analysis calculations.

[0039] Step 4: Programmatic perturbation of design parameters and generation of derived sub-models A procedural, minor modification (e.g., increasing the shell element thickness by 1%) is made to one or more design parameters in the baseline sub-model; a corresponding derived sub-model (.fem calculation file) is generated for each perturbated parameter. The entire sub-model generation and modification process is implemented by calling the export_part function, which integrates information such as component name, boundary load, and original stress.

[0040] In this specific implementation, sensitivity in this step refers to the gradient or derivative of the system response (or performance index) with respect to changes in design parameters. Commonly used design variables P in engineering include plate / shell thickness, material elastic modulus, and beam section moment of inertia; performance indices R include maximum stress, maximum displacement, natural frequency, and buckling factor. The mathematical expression for sensitivity is: By comparing the sensitivity of the performance indicators of design variables, the parameters that have the greatest impact on structural performance can be identified, allowing for the greatest performance improvement with the fewest modifications, thereby guiding the optimization design.

[0041] There are three main methods for calculating sensitivity: Finite difference method. By making a small perturbation to the design parameters (e.g., +1%), and then resolving the model, the derivative is approximated by the change in response. The sensitivity can be approximated using the forward difference formula: in, It is the perturbation step size of the design variables; This is the system response under the original design parameters. This is the system response obtained through reanalysis after parameter changes. This method is simple to implement, applicable to any "black box" solver, and easy to automate. However, it is computationally expensive (requiring an additional solution for each variable), and its accuracy depends on the perturbation step size ΔP.

[0042] Analytical methods. Internally, there are two main approaches—the direct method and the adjoint method—which differ in efficiency and applicable scenarios. The direct method is based on the equilibrium equations of the finite element system. Analytical differentiation of the design variable P yields: in, It is the system's stiffness matrix; It is the system's displacement vector; This is the system's load vector. Solve for... The displacement sensitivity is calculated first, and then the sensitivity of the response R is calculated. The adjoint method introduces an adjoint vector Ψ, and the adjoint equation is solved. Each response R only requires solving the adjoint equation once, and the cost of calculating sensitivity is almost independent of the number of design variables.

[0043] Semi-analytical method. A compromise that analytically calculates the derivative of the stiffness matrix K with respect to design variables while retaining the structure of the finite difference method. It is more accurate than the finite difference method and easier to implement than the analytical method.

[0044] The method described in this scheme uses the finite difference method based on the multi-scale sub-model approach to describe sensitivity. This method not only retains the advantages of the finite difference method in terms of ease of automation and batch solving, but also avoids the problem of excessively long single sensitivity calculation time due to the large model size and many degrees of freedom.

[0045] In the method described in this scheme, sensitivity calculation employs a manager-worker parallel computing model to address the easily parallelizable task of multivariate sensitivity calculation. The TCL script acts as the manager, first decomposing the selected target component into multiple independent sensitivity calculation tasks. Then, each task is asynchronously submitted to a solver process via the operating system—a coarse-grained parallel strategy based on multiple processes. After all tasks are submitted, the manager establishes a synchronization barrier through a polling mechanism, waiting for all worker processes to complete before uniformly collecting and summarizing the results to generate an analysis report.

[0046] The method described in this scheme can be used to analyze the sensitivity of the buckling factor λ to the design variable P. That is, for the characteristic equation Taking the derivative with respect to P, the analytical solution for the sensitivity of the i-th buckling factor λi is: Where φi is the buckling mode and λi is the buckling factor, both of which are obtained from the LBA solution; K / P is the derivative of the stiffness matrix K with respect to P; KG / P is the geometric stiffness matrix K G The derivative with respect to P. Although more complex than the finite difference method, it is extremely computationally efficient.

[0047] Step 5: Automated Batch Solving and Result Acquisition The system automatically submits the baseline sub-model generated in step three and all derived sub-models generated in step four to the solver (OptiStruct) for batch computation; it monitors and waits for all computation tasks to complete, and then automatically retrieves the result file generated by each model computation.

[0048] Step Six: Sensitivity Calculation and Result Quantification Output Automatically extract key performance indicators (e.g., maximum stress) from each result file; calculate the sensitivity value of each design parameter to the performance indicator based on the preset sensitivity calculation formula (stress change / original stress) / (perturbation design parameter change); sort the calculated sensitivity values ​​of each parameter, obtain the component name through the get_part_name API, and output them along with other information to a structured data file (e.g., a CSV file).

[0049] As can be seen from the above steps, this invention not only automates the entire process from model processing and batch solving to result extraction, significantly reducing manual operations and time costs for engineers, but also enables rapid assessment of the impact of multiple design parameters on structural performance, helping engineers quickly identify sensitive variables and providing direction for structural optimization. Furthermore, it seamlessly integrates the pre-processing, solving, and post-processing stages of CAE analysis, providing a complete and efficient solution.

[0050] As a specific implementation, aircraft component A is taken as the overall analysis object, and the design variables of its subordinate components a and b are taken as the sensitivity calculation objects. Component a contains design variables 1-5, and component b contains design variables 6-10. The hierarchical structure is as follows: Figure 5 As shown in the figure. The sensitivity histograms for the 10 design variables a and b obtained according to the calculation method of this embodiment are shown in the figure. Figure 6 As shown.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0052] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for calculating the sensitivity of large and complex structures, characterized in that, include: The target component in the original model is determined through interactive selection, and the initial state is obtained; Boundary identification is performed on the target component, and the mechanical environment of the boundary nodes is extracted; A baseline sub-model is generated based on the target component, and the mechanical environment is applied to the baseline sub-model to reproduce the force state in the original model; Based on the stress state, perturbation design parameters are selected in the baseline sub-model, and a derived sub-model corresponding to the perturbation design parameters is generated. The baseline sub-model and all derived sub-models are submitted to the solver for batch calculation. The solver monitors and waits for all calculation tasks to complete, and then automatically obtains the result file generated by each model calculation. Key performance indicators are extracted from each of the result files. Based on the calculation formula: Sensitivity = (Stress change / Original stress) / (Perturbation design parameter change), the sensitivity of each design parameter to the performance indicators is calculated, and the target component and its corresponding information are stored in a structured data file.

2. The sensitivity calculation method for large and complex structures according to claim 1, characterized in that, The process of interactively selecting and determining the target component in the original model and obtaining the initial state specifically involves: In the finite element post-processing environment, an interactive selection interface is provided, allowing users to interactively select target components through UI operations, and automatically acquire and store the initial key performance indicators of the target components by setting scripts.

3. The sensitivity calculation method for large and complex structures according to claim 1, characterized in that, The step of boundary identification of the target component and extraction of the mechanical environment of the boundary nodes specifically includes: The target component is identified from the overall model, and the boundary nodes connecting the target component to other components in the model are automatically identified. The displacement or load data of the boundary nodes in the initial static analysis results are automatically queried and extracted.

4. The sensitivity calculation method for large and complex structures according to claim 1, characterized in that, The process of generating a baseline sub-model based on the target component and applying the mechanical environment to the baseline sub-model to reproduce the force state in the original model specifically includes: The CAE environment automatically switches from post-processing mode to pre-processing mode; based on the target component, an independent reference sub-model containing only the target component and related properties and materials is generated, and the displacement or load data in the mechanical environment is applied to the reference sub-model as boundary conditions to reproduce the stress state in the original model.

5. The method for calculating the sensitivity of large and complex structures according to claim 1, characterized in that, Based on the stress state, selecting perturbation design parameters in the baseline sub-model and generating derived sub-models corresponding to the perturbation design parameters specifically includes: The design parameters in the baseline sub-model are perturbed in a programmed manner, and a corresponding derived sub-model is generated for each perturbed parameter, which integrates the component name, boundary load and original stress.

6. A sensitivity calculation system for large and complex structures, characterized in that, include: The component selection and state acquisition unit is used to determine the target component in the original model through interactive selection and to acquire the initial state; A boundary recognition and mechanical environment extraction unit is used to perform boundary recognition on the target component and extract the mechanical environment of the boundary nodes; The stress state reproduction unit is used to generate a reference sub-model based on the target component and apply the mechanical environment to the reference sub-model to reproduce the stress state in the original model. The data perturbation unit is used to select perturbation design parameters in the reference sub-model based on the force state, and generate a derived sub-model corresponding to the perturbation design parameters. The batch calculation unit is used to submit the baseline sub-model and all derived sub-models to the solver for batch calculation, monitor and wait for all calculation tasks to complete, and then automatically obtain the result file generated by each model calculation. The sensitivity calculation unit is used to extract key performance indicators from each of the result files, calculate the sensitivity of each design parameter to the performance indicators according to the calculation formula: sensitivity = (stress change / original stress) / (perturbation design parameter change), and store the target component and corresponding information in a structured data file.

7. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store a computer program, and the processor running the computer program to enable the electronic device to perform the sensitivity calculation method for large and complex structures according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the sensitivity calculation method for large and complex structures as described in any one of claims 1-5.

Images