Multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning

By establishing a refined model of a single bolt connection and a machine learning model to obtain the equivalent load, the problem of large computational load and low efficiency in the deformation analysis of multi-bolt connection structures is solved, and efficient deformation prediction and assembly optimization are achieved.

CN122333637APending Publication Date: 2026-07-03NORTHWESTERN POLYTECHNICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202610453565.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-04-08
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies are computationally intensive and inefficient in the deformation analysis of multi-nail connected structural assemblies, making it difficult to accurately characterize contact nonlinearity, local plastic deformation, and friction effects, resulting in overall profile deviations of the assembly.

Method used

By establishing a detailed model of a single bolt connection and obtaining the equivalent load, a deformation analysis model for multi-bolt connections is constructed. A machine learning model is then used to learn the mapping relationship between load boundary conditions and deformation response, thereby predicting the deformation of multi-bolt structural connections.

Benefits of technology

It significantly reduces the computational cost of full-detail finite element modeling, achieving a balance between computational efficiency and engineering accuracy, and provides technical support for the optimization of assembly processes and design iteration of large-size multi-bolt structures.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122333637A_ABST
    Figure CN122333637A_ABST
Patent Text Reader

Abstract

This invention discloses a deformation analysis method for multi-bolt structural connections based on displacement field equivalence and machine learning, relating to the field of advanced aircraft connection technology, and aimed at solving the technical problems of high computational load and low efficiency in existing technologies. The method includes: establishing a refined model of a single bolt connection process; analyzing the deformation of the connected components during the connection process based on the refined model; obtaining the equivalent load based on the deformation displacement field; constructing a multi-bolt connection deformation analysis model and creating connection elements within the analysis model; simulating the overall component by inputting the equivalent load to obtain a sample dataset; training a machine learning model using the sample dataset and learning the mapping relationship between load boundary conditions and deformation response from the simulation samples; and predicting the deformation of the multi-bolt structural connection based on the trained machine learning model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of advanced aircraft connection technology, and more specifically, to a method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning. Background Technology

[0002] The circumferential seam area of ​​the aircraft fuselage, as the core area for fuselage section docking, is not only a critical path for transmitting overall loads but also a weak link in ensuring aerodynamic shape accuracy and structural integrity. Studying the connection deformation behavior in this area is of great significance for ensuring the continuity of load transmission, optimizing assembly processes, and ultimately achieving high-precision fuselage docking. However, this area has a large structural dimension and relatively small thickness. During assembly, local nonlinear deformations (such as warping) caused by individual bolt connections are easily transmitted and amplified along the circumferential and longitudinal directions in the flexible panels, ultimately leading to overall profile deviations in the assembly. This mechanical transmission mechanism, where local nonlinearity leads to global cumulative errors, is a core technical challenge in achieving high-efficiency, high-precision assembly.

[0003] From the current research perspective, the mainstream methods for analyzing the deformation of multi-bolt connection structures include analytical methods and full-scale refined finite element simulation. On the one hand, the simplified analytical method usually equates the bolt connection to a spring element or a simplified contact model, which can significantly reduce the modeling and computational complexity. However, its assumptions are often too idealistic, making it difficult to accurately characterize the complex mechanical behaviors that exist in the actual assembly process, such as contact nonlinearity, local plastic deformation, friction effects, and uneven distribution of preload. On the other hand, although full-scale refined finite element simulation can theoretically obtain results closer to reality by establishing detailed contact pairs for each bolt connection, applying real preload, and performing nonlinear solutions, its computational cost is extremely high. The tens of thousands of contact nonlinearities in the model and the stability problems of iterative solutions make a single simulation extremely time-consuming. It is evident that the existing technologies suffer from high computational load and low computational efficiency. Summary of the Invention

[0004] The purpose of this invention is to provide a deformation analysis method for multi-nail structure connections based on displacement field equivalence and machine learning, to solve the technical problems of high computational load and low efficiency in existing technologies. In view of this, this invention is achieved through the following scheme.

[0005] In a first aspect, the present invention provides a method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning, comprising: A detailed model of a single bolt connection process is established. Based on this detailed model, the deformation of the connected parts during the connection process is analyzed, and the equivalent load is obtained from the deformation displacement field. A deformation analysis model for multi-bolt connections is constructed, and connection elements are created in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; a sample dataset is obtained by simulating the overall component by inputting the equivalent load; A machine learning model is trained using the sample dataset, and the mapping relationship between load boundary conditions and deformation response is learned from the simulation samples. Deformation prediction of multi-nail structure connections is performed based on the trained machine learning model.

[0006] Compared with existing technologies, the deformation analysis method for multi-bolt structural connections based on displacement field equivalence and machine learning in this invention establishes a refined model of the single bolt connection process, analyzes the deformation of the connected components during the connection process based on this refined model, and obtains the equivalent load based on the deformation displacement field; then, a multi-bolt connection deformation analysis model is constructed, and connection elements are created in the analysis model; the node information of each connection element corresponds to the single bolt connection structure; by inputting the equivalent load into the overall component for simulation calculation, a sample dataset is obtained; further, the sample dataset is used to train a machine learning model, and the mapping relationship between load boundary conditions and deformation response is learned from the simulation samples, and the deformation prediction of multi-bolt structural connections is performed based on the trained machine learning model. Based on the above technical solution of this invention, the local displacement field is extracted through refined finite element analysis of single-bolt connections, and then transformed into the displacement boundary conditions of the overall structure through an equivalent transfer method, thereby significantly reducing the high computational cost of full-detail finite element modeling; using this equivalent method to generate multi-condition datasets, a machine learning proxy model is trained, and a rapid prediction of overall deformation is achieved while balancing computational efficiency and engineering accuracy, providing technical support for the optimization of assembly processes and design iteration of large-size multi-bolt structures. The technical solution of the present invention solves the technical problems of large computational load and low efficiency in the prior art.

[0007] Furthermore, in the multi-nail structure connection deformation analysis method of the present invention based on displacement field equivalence and machine learning, the step of analyzing the deformation of the connected components during the connection process according to the refined model includes: Let the bolt diameter be D, and the thicknesses of the connected components, the fuselage panel and the strip plate, be respectively... and With the center of the connecting hole as the origin, construct a structure with a radius of R and a height of... A cylindrical region; The cylindrical region contains nodes that are distributed in a ring around the connecting hole.

[0008] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, obtaining the equivalent load based on the deformation displacement field includes: The fuselage adopts a cylindrical structure. A global Cartesian coordinate system is established with the fuselage axis as the Y-axis. The direction of the bolts is determined based on the center coordinates of the connecting holes on the outer surface of the wall panel and the inner surface of the belt plate. Determine any node within the cylindrical region and obtain the distance from that node to the center line of the hole shaft; Based on the different positions of the node from the center of the bore shaft, the node is divided radially into... k The axial and radial displacements of all nodes in the i-th group are averaged and taken as the effective equivalent load.

[0009] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, after averaging the axial and radial displacements of all nodes in the i-th group as the effective equivalent load, the load corresponding to the entire cylindrical region includes axial load and radial load.

[0010] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, the step of creating connection units in the analysis model includes: The analysis model is divided into multiple connection units; Determine the hole axis direction based on the inlet and outlet of each bolt hole; A local cylindrical coordinate system with the hole axis as the Z-axis is established at the center of each connecting unit, and a cylindrical region is established in each connecting unit; Each connection unit is the same size as a single analysis model.

[0011] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, the sample dataset includes a sample dataset covering different connection sequences and structural deformations.

[0012] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, during the process of predicting the deformation of the multi-nail structure connection: The sequence of bolt connections after assembly is used as the load boundary condition input into the model, and the model outputs the overall deformation response result of the structure.

[0013] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, after the machine learning model is trained, it includes: K-fold cross-validation is used to evaluate generalization ability. The dataset is divided into k mutually exclusive subsets. The model is retrained using k-1 subsets as training data and the remaining subset is used as validation data to evaluate performance. This process is repeated k times to ensure that each sample is validated once. The mean and variance of the validation error for each fold are calculated.

[0014] Furthermore, in the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning of the present invention, the machine learning model includes: The machine learning models include support vector regression, nearest neighbor, and random forest models.

[0015] Secondly, the present invention provides a deformation analysis system for multi-nail structure connections based on displacement field equivalence and machine learning, comprising: The equivalent load acquisition module is used to establish a detailed model of a single bolt connection process, analyze the deformation of the connected parts during the connection process based on the detailed model, and obtain the equivalent load based on the deformation displacement field. The dataset acquisition module is used to construct a deformation analysis model for multi-bolt connections and create connection elements in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; and the sample dataset is obtained by simulating the overall component by inputting the equivalent load. The model training module uses sample datasets to train machine learning models and learns the mapping relationship between load boundary conditions and deformation response from simulation samples. The deformation prediction module is used to predict the deformation of multi-nail structure connections based on the trained machine learning model.

[0016] Compared with the prior art, the beneficial effects of the multi-nail structure connection deformation analysis system based on displacement field equivalence and machine learning of the present invention are the same as the beneficial effects of the multi-nail structure connection deformation analysis method based on displacement field equivalence and machine learning described in the above technical solution, and will not be repeated here. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a method for analyzing the deformation of a multi-nail structure connection according to the present invention. Figure 2 This is a schematic diagram of the geometric dimensions of the single-nail connection model component in Embodiment 4 of the present invention; Figure 3 This is a schematic diagram of the displacement field equivalent process in Embodiment 4 of the present invention; Figure 4 This is a schematic diagram of the finite element model of the multi-nail connection in Embodiment 4 of the present invention; Figure 5 This is a schematic diagram of the multi-nails connection sequence in Embodiment 4 of the present invention; Figure 6 This is a schematic diagram comparing the magnitude of the Z-direction displacement of the measurement point in Embodiment 4 of the present invention.

[0018] Figure label: exist Figure 4 In the diagram, A is the connecting area, B is the fuselage panel, and C is the strip panel. Detailed Implementation

[0019] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0020] It should be noted that when a component is referred to as being "fixed to" or "set on" another component, it can be directly on or indirectly on that other component. When a component is referred to as being "connected to" another component, it can be directly connected to or indirectly connected to that other component.

[0021] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. "Several" means one or more, unless otherwise explicitly specified.

[0022] From the current research perspective, the mainstream methods for analyzing the deformation of multi-bolt connection structures include analytical methods and full-scale refined finite element simulation. On the one hand, the simplified analytical method usually equates the bolt connection to a spring element or a simplified contact model, which can significantly reduce the modeling and computational complexity. However, its assumptions are often too idealistic, making it difficult to accurately characterize the complex mechanical behaviors that exist in the actual assembly process, such as contact nonlinearity, local plastic deformation, friction effects, and uneven distribution of preload. On the other hand, although full-scale refined finite element simulation can theoretically obtain results closer to reality by establishing detailed contact pairs for each bolt connection, applying real preload, and performing nonlinear solutions, its computational cost is extremely high. The tens of thousands of contact nonlinearities in the model and the stability problems of iterative solutions make a single simulation extremely time-consuming. It is evident that the existing technologies suffer from high computational load and low computational efficiency.

[0023] To address the aforementioned technical problems, this invention provides a method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning, comprising: A detailed model of a single bolt connection process is established. Based on this detailed model, the deformation of the connected parts during the connection process is analyzed, and the equivalent load is obtained from the deformation displacement field. A deformation analysis model for multi-bolt connections is constructed, and connection elements are created in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; a sample dataset is obtained by simulating the overall component by inputting the equivalent load; A machine learning model is trained using the sample dataset, and the mapping relationship between load boundary conditions and deformation response is learned from the simulation samples. Deformation prediction of multi-nail structure connections is performed based on the trained machine learning model.

[0024] In the above-mentioned technical solution, the deformation analysis method for multi-bolt structural connections based on displacement field equivalence and machine learning of the present invention establishes a detailed model of a single bolt connection process, analyzes the deformation of the connected components during the connection process based on this detailed model, and obtains the equivalent load based on the deformation displacement field; then, a multi-bolt connection deformation analysis model is constructed, and connection elements are created in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; by inputting the equivalent load into the overall component for simulation calculation, a sample dataset is obtained; further, the sample dataset is used to train a machine learning model, and the mapping relationship between load boundary conditions and deformation response is learned from the simulation samples, and the deformation prediction of multi-bolt structural connections is performed based on the trained machine learning model. Based on the above-mentioned technical solution of the present invention, the local displacement field is extracted through detailed finite element analysis of a single-bolt connection, and then transformed into the displacement boundary conditions of the overall structure through an equivalent transfer method, thereby significantly reducing the high computational cost of full-detail finite element modeling; using this equivalent method to generate a multi-condition dataset, train a machine learning proxy model, and achieve rapid prediction of overall deformation while balancing computational efficiency and engineering accuracy, providing technical support for the optimization of assembly process and design iteration of large-size multi-bolt structures. The technical solution of the present invention solves the technical problems of large computational load and low efficiency in the prior art.

[0025] To better understand the present invention, the following specific embodiments further illustrate the content of the present invention, but the content of the present invention is not limited to the following embodiments.

[0026] Example 1 This embodiment provides a method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning, including: Step 1: Establish a detailed model of a single bolt connection process, analyze the deformation of the connected parts during the connection process based on the detailed model, and obtain the equivalent load based on the deformation displacement field; Step 2: Construct a deformation analysis model for multi-bolt connections and create connection elements in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; obtain a sample dataset by simulating the overall component by inputting the equivalent load; Step 3: Train a machine learning model using the sample dataset and learn the mapping relationship between load boundary conditions and deformation response from simulation samples; Step 4: Predict the deformation of the multi-nailing structure connection based on the trained machine learning model.

[0027] Example 2 This embodiment provides a method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning, including: Step 1: Establish a detailed model of a single bolt connection process, analyze the deformation of the connected parts during the connection process based on the detailed model, and obtain the equivalent load based on the deformation displacement field.

[0028] In this step, the deformation of the connected parts during the connection process is analyzed based on the refined model, including: Let the bolt diameter be D, and the thicknesses of the connected components, the fuselage panel and the strip plate, be respectively... and With the center of the connecting hole as the origin, construct a structure with a radius of R and a height of... A cylindrical region; The cylindrical region contains nodes that are distributed in a ring around the connecting hole.

[0029] In this step, the equivalent load is obtained based on the deformation displacement field, including: The fuselage adopts a cylindrical structure. A global Cartesian coordinate system is established with the fuselage axis as the Y-axis. The direction of the bolts is determined based on the center coordinates of the connecting holes on the outer surface of the wall panel and the inner surface of the belt plate. Determine any node within the cylindrical region and obtain the distance from that node to the center line of the hole shaft; Based on the different positions of the node from the center of the bore shaft, the node is divided radially into... k The axial and radial displacements of all nodes in the i-th group are averaged and taken as the effective equivalent load.

[0030] In this step, the axial and radial displacements of all nodes in the i-th group are averaged and taken as the effective equivalent load. The load corresponding to the entire cylindrical region includes axial load and radial load.

[0031] Step 2: Construct a deformation analysis model for multi-bolt connections and create connection elements in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; obtain a sample dataset by simulating the overall component by inputting the equivalent load; In this step, connection units are created in the analysis model, including: The analysis model is broken down into multiple connected units; Determine the hole axis direction based on the inlet and outlet of each bolt hole; A local cylindrical coordinate system with the hole axis as the Z-axis is established at the center of each connecting unit, and a cylindrical region is established in each connecting unit; Each connection unit is the same size as a single analysis model.

[0032] In this step, the sample dataset includes sample datasets covering different connection orders and structural deformations.

[0033] Step 3: Train the machine learning model using the sample dataset and learn the mapping relationship between load boundary conditions and deformation response from the simulation samples; the machine learning model includes support vector regression model, nearest neighbor model and random forest model.

[0034] Step 4: Predict the deformation of the multi-nailing structure connection based on the trained machine learning model.

[0035] In this step, after the machine learning model has been trained, the following steps are included: K-fold cross-validation is used to evaluate generalization ability. The dataset is divided into k mutually exclusive subsets. The model is retrained using k-1 subsets as training data and the remaining subset is used as validation data to evaluate performance. This process is repeated k times to ensure that each sample is validated once. The mean and variance of the validation error for each fold are calculated.

[0036] In this step, during the deformation prediction of multi-nail structure connections: the sequence of bolt connections after assembly is used as the load boundary condition input into the model, and the model outputs the overall deformation response result of the structure.

[0037] Example 3 This embodiment provides a method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning, including:

[0038] A detailed model of a single bolt connection process is established. Based on this detailed model, the deformation of the connected components during the connection process is analyzed, and the equivalent load is obtained from the deformation displacement field. A multi-bolt connection deformation analysis model is constructed, and connection elements are created in the analysis model. The node information of each connection element corresponds to the single-bolt connection structure. By inputting the equivalent load into the overall component for simulation calculation, a sample dataset is obtained. A machine learning model is trained using the sample dataset, and the mapping relationship between load boundary conditions and deformation response is learned from the simulation samples. The deformation prediction of multi-bolt structure connections is performed based on the trained machine learning model.

[0039] Specifically, in this embodiment, the refined model includes two layers of aluminum-lithium alloy plates, countersunk high-strength bolts and nuts. The dimensions of the two aluminum-lithium alloy plates are 24mm × 24mm × 2mm. The threads of the bolts and nuts are ignored, and the geometric dimensions are modeled based on the actual dimensions of the HST11AG5-3 titanium alloy high-strength bolt and the HST1070-5 aluminum alloy high-strength nut. Furthermore, the stress-strain constitutive relationship of the aluminum-lithium alloy and titanium alloy materials is established using the Johnson-Cook constitutive model. The elastic part can be expressed as: The plastic part can be represented as ;in, Indicates stress, Indicates the elastic modulus. Indicates strain, Indicates the initial yield stress. Indicates the strain hardening coefficient. This indicates the hardening index.

[0040] Furthermore, the various parts of the above-mentioned detailed model adopt "face-to-face" contact, and the contact interface is divided into normal and tangential interactions. All parts use reduced integral hexahedral elements C3D8R. In order to improve the accuracy of the model, a finer mesh is used in the area around the hole. During bolt installation, all degrees of freedom of the bottom of the lower aluminum-lithium alloy plate are restricted, the five degrees of freedom of the bolt except for the Z-direction movement are restricted, and a pin displacement load is applied. After installation, a nut model is added, and a preload is applied to the bolt using the section method.

[0041] Furthermore, to further analyze the deformation of the connected components and calculate the subsequent equivalent load, it is necessary to group the nodes around the holes of the connected components according to their spatial positions; assuming the diameter of the high-lock bolt is D, and the thicknesses of the fuselage panel and the strip plate of the connected components are respectively... and With the center of the connecting hole as the origin, construct a structure with a radius of R (R>D) and a height of... cylindrical area The nodes contained in this cylindrical region will be distributed in a ring around the connecting hole. Since the interference of the connected parts is not significantly different along the bolt axis in bolted connections, when dividing the nodes in the cylindrical region, there is no need to perform axial grouping, only radial grouping.

[0042] Furthermore, since aircraft fuselages often employ a cylindrical segment structure, a global Cartesian coordinate system is established using the fuselage axis as the Y-axis. The coordinates of the hole centers on the outer surface of the panel and the inner surface of the strip can be obtained from the geometric model of the fuselage docking area. and The direction of the bolt hole can be represented as: ; In the cylindrical region any node The coordinates are The distance from this point to the center line of the hole shaft can be expressed as: ; Based on the different positions of the node from the center of the hole shaft, the node can be divided radially into k Groups, denoted as Node_Set The specific division method is as follows: ; in, Representing different nodes, Represents a set of nodes. Indicates the bolt diameter. This indicates the distance from the corresponding node to the center line of the hole shaft. This represents the radial increment, which is determined based on the bolt diameter, the radius of the cylindrical region, and the mesh size.

[0043] Furthermore, a local cylindrical coordinate system is established with the center of the bolt hole as the origin and the axis direction as the Z-axis, and the coordinates are extracted. The axial and radial displacements of all nodes within the node will The average axial and radial displacements of all nodes are denoted as follows: and And use it as an equivalent load for the entire cylindrical node region Node_Set The corresponding loads are: Axial load Load_Axial and radial load Load_Radial .

[0044] Furthermore, based on the actual structure of the fuselage docking area, a multi-bolt connection simulation geometric model was constructed. This model was then divided into multiple connection units, each with dimensions consistent with the single-bolt connection simulation model. The positional information of each connection hole was clearly defined, assuming the center coordinates of each bolt hole inlet are as follows: The coordinates of the exit center are Then the direction of the hole axis is Furthermore, to facilitate the application of equivalent loads, it is necessary to center each connection unit. Created at the location Let Cylindrical_i be the local cylindrical coordinate system along the Z-axis, and let Node_Set be created in each connected element. .

[0045] Furthermore, when applying the equivalent load, each connection element corresponds to an analysis step, with the equivalent load from the previous step being passed to the next step. iTaking an equivalent element as an example, the corresponding equivalent load is applied based on the cylindrical coordinate system Cylindrical_i, and the load includes the axial load Load_Axial. and radial load Load_Radial The corresponding application is applied to the node set Node_Set That is, the node region Corresponding axial load and radial load .

[0046] Furthermore, in this embodiment, the material parameters, contact properties, and mesh type of the multi-nail connection simulation geometry model are consistent with those of the single-nail connection simulation model. Due to the larger size of the multi-nail structure, the overall mesh size needs to be set according to the geometry model, but the area around the holes still uses a fine mesh, consistent with the single-nail simulation model. Finally, based on the actual assembly connection constraints, corresponding boundary conditions are added to the simulation model.

[0047] Furthermore, this embodiment utilizes a dataset to train a Support Vector Regression (SVR) model, learning the high-dimensional nonlinear mapping relationship between load boundary conditions and deformation response from high-fidelity simulation samples, thereby establishing a more efficient assembly deformation prediction model; specifically: Support Vector Regression (SVR) is a powerful machine learning algorithm developed based on the principle of structural risk minimization in statistical learning theory. It originated from the successful application of Support Vector Machines (SVMs) proposed by Vapnik et al. in classification problems. By introducing an ε-insensitive loss function, its superior performance is extended to the field of regression analysis, thus solving complex nonlinear regression problems. Given a... n Training dataset of samples ,in, For the input vector, This corresponds to the output value. The core objective of SVR is to find a regression function. This allows the function to approximate all training samples with an accuracy of ε.

[0048] Furthermore, we define the ε-insensitive loss function for standard SVR: ; in, Let y be the predicted value of the objective function, and y be the true value. This function represents the objective expression... The difference between the actual value y and the true value y is less than or equal to If so, then the value of the loss function is equal to 0.

[0049] Furthermore, the core of support vector machine regression is to find an optimal surface that minimizes the error of all original data points from the sought classification surface. Additionally, considering the potential for error, two slack variables are introduced. and The original optimization problem can then be expressed as: ; The constraints are: ; in, Represents the weight vector. Indicates the bias term. Represents slack variables. C Represents the regularization parameter. C The larger the value, the greater the training error. The larger the sample penalty function, the better. This represents the upper limit of the error of the objective function. The smaller the value, the smaller the error of the objective function. This represents the feature mapping function.

[0050] Furthermore, to solve the above problem more efficiently, the original problem is transformed into a dual form, and the Larger function is introduced to solve the dual problem. The corresponding Larger function is: ; Derive the dual problem using KKT conditions: ; The constraints are: ; It is the inner product after mapping the samples to a high-dimensional feature space. Because the feature space has a high dimension, or even infinite dimension, direct calculation is not possible. Difficult. To avoid this obstacle, we can imagine a function like this: ; This function Known as the "kernel function," the choice of kernel function determines the quality of the feature space, further affecting the performance of the support vector machine. Commonly used kernel functions include: Linear kernel: ; Polynomial kernel: ; RBF core (most commonly used): ; Sigmoid core: ; The solution can be found by solving the dual problem. Value, usually and The value of 0 will be non-zero. and The corresponding samples are called support vectors. The value can be calculated using any support vector. b The value is determined by the condition, but in practice, all samples that meet the conditions are often selected for the solution. b Then take the average value, and then b It can be calculated using the following formula: ; The final regression function is: ; In the above expression , For Lagrange multipliers, For kernel function, The upper limit of the objective function error. For the set of support vectors ,b This is a bias term.

[0051] Furthermore, a displacement field equivalent model is used to provide a dataset for the SVR model. The dataset is normalized and divided into training and testing sets. Based on the characteristics of the dataset, the RBF kernel is selected as the kernel function of the model, and the kernel function matrix is ​​calculated. When using the RBF kernel, the initial parameters of the model will include , and Where C is the regularization parameter, The upper limit of the objective function error. Set the RBF kernel bandwidth and initial parameters before model training.

[0052] The SVR model training uses the SMO algorithm to solve the dual problem, meaning that each optimization only optimizes the dual problem. With two Lagrange multipliers and the rest fixed, the solution is obtained. Then, the calculated support vectors for each group are... b The final regression function is obtained by summing the values ​​and taking the average.

[0053] After the SVR model is trained, it is used k Folded cross-validation is used to evaluate generalization ability, and the dataset is divided into... k A mutually exclusive subset, in turn, is defined by one of them. k-1 The model is retrained using one subset as training data, and its performance is evaluated using the remaining subset as validation data. This process is repeated k times to ensure that each sample is validated once. By calculating the mean and variance of the validation errors at each fold, an unbiased estimate of the model's generalization error is obtained, quantifying the stability of the model's performance.

[0054] Example 4 Firstly, this embodiment provides a deformation analysis method for multi-nail structure connections based on displacement field equivalence and machine learning. In this example, a region of approximately 144mm × 72mm is selected in the circumferential seam connection area between the middle and aft fuselage of an aircraft. A detailed model of the single-nail connection is used to obtain the local deformation field, which is then equivalently transferred to the simplified overall structure as displacement boundary conditions. Furthermore, a dataset is generated using the equivalence method, a support vector regression model is trained, and a lightweight proxy model capable of quickly predicting overall deformation is constructed. The specific implementation process is as follows: Figure 1 As shown.

[0055] Furthermore, based on the actual connection characteristics of the aircraft fuselage docking area, and taking a single bolt connection as the research object, a simulation model of the aluminum-lithium alloy countersunk bolt connection process is established. The connection process includes two parts: bolt installation and bolt pre-tightening; specifically: In this embodiment, the simulation model includes two layers of aluminum-lithium alloy plates, countersunk high-strength bolts and nuts. The dimensions of the two aluminum-lithium alloy plates are 24mm × 24mm × 2mm. The threads of the bolts and nuts are ignored. The geometric dimensions are modeled based on the actual dimensions of the HST11AG5-3 titanium alloy high-strength bolt and the HST1070-5 aluminum alloy high-strength nut. The model's geometric dimensions are as follows: Figure 2 As shown in the figure, the upper and lower plates in the model are made of aluminum-lithium alloy, and the bolt material is Ti-6Al-4V. Specific parameters are shown in Table 1.

[0056] Table 1 Material parameters of the single-nail connection model

[0057]

[0058] Furthermore, there are numerous contact interfaces between the components in the model, primarily existing between the upper and lower plates, between the lower plate and the upper surface of the nut, between the outer surface of the bolt and the hole wall, and between the outer surface of the bolt and the nut. Contact pairs include... Figure 3 As shown in the figure. All contacts adopt the discrete surface-to-surface contact type, the contact property is set to finite slip, the normal behavior between each contact pair is set to hard contact to avoid mesh penetration, and the tangential behavior is set to penalty friction. The setting of the tangential friction coefficient is based on relevant literature. The specific friction coefficients between interfaces are shown in Table 2.

[0059] Table 2 Contact Surface and Tangential Friction Coefficients

[0060] Furthermore, all components utilize reduced integral hexahedral elements C3D8R. Since the equivalent method is based on nodal displacements, a more refined mesh is required in the area near the hole. The aluminum-lithium alloy plate is divided into sections with a radius of 2R centered on the hole. A 0.3mm mesh is used within the selected area, and a 1mm mesh is used for the remaining areas.

[0061] Furthermore, during bolt installation, all degrees of freedom at the bottom of the lower aluminum-lithium alloy plate are restricted, and the bolt is coupled to a central reference point. The reference point is restricted in five degrees of freedom except for movement in the Z direction, and a pin displacement load is applied. The displacement load is along the negative Z-axis and has a magnitude of 12mm to ensure that the bolt can be fully pressed into the aluminum-lithium alloy plate.

[0062] Further, after installation, a pre-tightening simulation is performed. The deformed aluminum-lithium alloy plate and bolt models, along with their residual stress fields, are imported. Simultaneously, a nut model is constructed, and binding constraints are established for the remaining bolts. Since the threaded portion is simplified in the model, the section method is used to apply the pre-tightening force. The HST1070-5 high-lock nut has a designed breaking torque of 2.0 N·m. The corresponding pre-tightening force is calculated to be 2421 N using the following formula: ;in, Indicates preload. Indicates the tightening torque. d The diameter of the bolt. K This is the torque coefficient.

[0063] Furthermore, in this embodiment, the simulation results of the single-nail connection are processed to analyze the deformation of the aluminum-lithium alloy plate during the connection process. The nodes are divided into corresponding regions, the deformation of the nodes in each region is extracted, and the equivalent load is calculated. Specifically: With the center of the connecting hole as the origin, construct a cylindrical region with a radius of R (R=3mm) and a height of 4mm. The nodes in this region will be distributed in a ring around the connecting hole. Since the mesh size near the hole is approximately 0.3mm, this region contains 3 sets of nodes in the radial direction, denoted as follows from the inside out: .

[0064] Furthermore, in the set of nodes In the process, 3×10 nodes were selected, and the spatial coordinates of these nodes were extracted. The axial and radial displacements of each node were calculated (see Table 3). Comparison of the data revealed that each node... The data within the range showed no significant difference. The average axial and radial displacements of all nodes are denoted as follows: and And this is used as the equivalent load. The loads corresponding to the entire cylindrical node region Node_Set are Load_Axial and Load_Radial ,in , .

[0065] Table 3 Axial and radial displacements of nodes

[0066] Furthermore, in this embodiment, a 144mm×72mm connection area is selected as the research object in the docking area between the middle fuselage and the rear fuselage of the aircraft. A simulation model of the connection process in this area is constructed based on the displacement field equivalent analysis method, and the corresponding nodal areas are established and equivalent loads are applied.

[0067] Furthermore, taking the docking of the mid-fuselage and the mid-rear fuselage as an example, this area is entirely connected by bolts, and the connecting components include corner plates, connecting plates, and fuselage wall panels. The mid-fuselage and the mid-rear fuselage are connected by overlapping plates, while the corner plates connect the mid-rear fuselage to the bulkhead. All connecting holes are connected by interference bolts. This example focuses on the connection process between the plate and the mid-fuselage.

[0068] Furthermore, based on the actual connection conditions, establish, for example... Figure 4 The finite element model shown includes two components: the mid-fuselage panel and the strip plate. In the actual connection, the strip plate contains six rows of bolt holes, while the mid-fuselage and mid-rear fuselage panels each contain three rows. Since this example studies the connection between the strip plate and the mid-fuselage, only three rows of bolt holes are retained on the strip plate.

[0069] Furthermore, both the mid-fuselage panel and the strip are made of aluminum-lithium alloy with parameters consistent with the single-nail connection simulation model. The normal contact behavior of both is set to hard contact, and the tangential contact behavior is set to penalized friction with a friction coefficient of 0.3. The mesh type uses reduced integral hexahedral elements C3D8R, with a mesh size of 0.3 mm in the perforated area and 1 mm in other locations.

[0070] Furthermore, the connection between the fuselage panel and the strip includes nine connection holes, according to... Figure 5 The connection area is divided into nine connection units, numbered 1 to 9, each unit measuring 24×24mm. To facilitate the application of equivalent loads, a local cylindrical coordinate system is constructed with the center of each connection unit as the origin and the axis of the connection hole as the Z-axis, denoted sequentially as Cylindrical_1 to Cylindrical_9.

[0071] Furthermore, the equivalent load Load_Axial of each connection element is calculated using the nodal displacement field of a single bolt connection process. and Load_Radial In multi-nail connection simulations, loads are applied to the corresponding nodes in a single-nail structure as equivalently as possible. Therefore, a cylindrical region with radius R (R=3mm) and height 4mm is constructed with the center of the connection hole as the origin. This region contains three sets of nodes, denoted as Set_Node_w1, Set_Node_w2, and Set_Node_w3.

[0072] Furthermore, to verify the accuracy of the equivalent method in predicting deformation of multi-nail connections, 12 measurement points were set on the mid-fuselage panel, as shown in the figures below. Figure 5 As shown. Following the commonly used "center method" connection sequence, the multi-nail connection process is simulated, with the specific connection sequence being 5-2-8-6-4-3-9-1-7. After the simulation, the Z-direction displacement of the measurement points is extracted. Simultaneously, a refined simulation model of the selected area is constructed, measurement points are set at the same locations, and the Z-direction displacement of the measurement points is compared with the equivalent model. The results are shown below. Figure 6 As shown.

[0073] Furthermore, in this embodiment, a multi-pin connection analysis model is used to quickly generate structural deformation datasets under different connection orders. The dataset is then used to train an SVR regression model, and the model performance is evaluated through cross-validation. Specifically: In multi-bolt structures, different connection sequences can significantly affect structural deformation. Therefore, the connection sequence of the nine bolts is used as the input variable, and the deformation at the measurement points is used as the output. Commonly used connection sequences in engineering include the edge method and the center method. Twenty connection sequences from each method were randomly selected, and a total of 20 sets of samples were obtained through simulation.

[0074] Furthermore, in order to avoid numerical instability or convergence difficulties during the training of the SVR model due to excessively large differences in magnitude, it is necessary to first normalize all sample data, and select 16 sets from the 20 sets of sample data as the training set, and the other four sets as test samples.

[0075] Furthermore, input the training set To address the problem of linear inseparability of the original data, an RBF kernel is used to map the data to a high-dimensional space, and the kernel matrix is ​​calculated based on the RBF kernel function. ; Furthermore, set the model hyperparameters. , and Where C is the regularization parameter, The upper limit of the objective function error. Given the RBF kernel bandwidth, solve the following convex quadratic programming dual problem to obtain the Lagrange multipliers. and nonzero and The corresponding samples are the support vectors.

[0076] ; Furthermore, based on the Lagrange multipliers obtained in the previous step, the model bias term is calculated using the support vector set. b Define the support vector index set: ; Then bias b The calculation formula is: ; The final regression function is: ; Furthermore, the connection order of the four test samples was input into the SVR regression model to obtain the average structural deformation, and then the validity was verified. To evaluate the accuracy of the model, among which For predicted values, These are simulation values. The model prediction results are shown in Table 4. The validity of all four sets of data in the test set exceeded 90%, indicating that the SVR model has high accuracy in predicting connection deformation.

[0077] Table 4. SVR Model Prediction Results

[0078] Furthermore, based on this, cross-validation can be used to evaluate the generalization ability of the model. The 20 datasets are divided into 5 mutually exclusive subsets, each containing 4 sets of data. The model is retrained using 4 subsets as training data in turn, and the remaining 1 subset is used as validation data to evaluate the performance. This process is repeated 5 times to ensure that each sample is validated once.

[0079] Secondly, this embodiment provides a deformation analysis system for multi-nail structure connections based on displacement field equivalence and machine learning, including: The equivalent load acquisition module is used to establish a detailed model of a single bolt connection process, analyze the deformation of the connected parts during the connection process based on the detailed model, and obtain the equivalent load based on the deformation displacement field. The dataset acquisition module is used to construct a deformation analysis model for multi-bolt connections and create connection elements in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; and the sample dataset is obtained by simulating the overall component by inputting the equivalent load. The model training module uses sample datasets to train machine learning models and learns the mapping relationship between load boundary conditions and deformation response from simulation samples. The deformation prediction module is used to predict the deformation of multi-nail structure connections based on the trained machine learning model.

[0080] In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for multi-nail structure connection deformation analysis based on displacement field equivalence and machine learning, characterized in that, include: A detailed model of a single bolt connection process is established. Based on this detailed model, the deformation of the connected parts during the connection process is analyzed, and the equivalent load is obtained from the deformation displacement field. A deformation analysis model for multi-bolt connections is constructed, and connection elements are created in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; a sample dataset is obtained by simulating the overall component by inputting the equivalent load; A machine learning model is trained using the sample dataset, and the mapping relationship between load boundary conditions and deformation response is learned from the simulation samples. Deformation prediction of multi-nail structure connections is performed based on the trained machine learning model.

2. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 1, characterized in that, The analysis of the deformation of the connected components during the connection process based on the refined model includes: Let the bolt diameter be D, and the thicknesses of the connected components, the fuselage panel and the strip plate, be respectively... and With the center of the connecting hole as the origin, construct a structure with a radius of R and a height of... A cylindrical region; The cylindrical region contains nodes that are distributed in a ring around the connecting hole.

3. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 2, characterized in that, The process of obtaining the equivalent load based on the deformation displacement field includes: The fuselage adopts a cylindrical structure. A global Cartesian coordinate system is established with the fuselage axis as the Y-axis. The direction of the bolts is determined based on the center coordinates of the connecting holes on the outer surface of the wall panel and the inner surface of the belt plate. Determine any node within the cylindrical region and obtain the distance from that node to the center line of the hole shaft; Based on the different positions of the node from the center of the bore shaft, the node is divided radially into... k The axial and radial displacements of all nodes in the i-th group are averaged and taken as the effective equivalent load.

4. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 3, characterized in that, After averaging the axial and radial displacements of all nodes in the i-th group as the effective equivalent load, the load corresponding to the entire cylindrical region includes both axial and radial loads.

5. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 1, characterized in that, Creating connection units in the analysis model includes: The analysis model is divided into multiple connection units; Determine the hole axis direction based on the inlet and outlet of each bolt hole; A local cylindrical coordinate system with the hole axis as the Z-axis is established at the center of each connecting unit, and a cylindrical region is established in each connecting unit; Each connection unit is the same size as a single analysis model.

6. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 1, characterized in that, The sample dataset includes sample datasets covering different connection orders and structural deformations.

7. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 1, characterized in that, During the process of predicting deformation of multi-nail structure connections: The sequence of bolt connections after assembly is used as the load boundary condition input into the model, and the model outputs the overall deformation response result of the structure.

8. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 1, characterized in that, After the machine learning model is trained, it includes: K-fold cross-validation is used to evaluate generalization ability. The dataset is divided into k mutually exclusive subsets. The model is retrained using k-1 subsets as training data and the remaining subset is used as validation data to evaluate performance. This process is repeated k times to ensure that each sample is validated once. The mean and variance of the validation error for each fold are calculated.

9. The method for analyzing the deformation of multi-nail structure connections based on displacement field equivalence and machine learning according to claim 1, characterized in that, The machine learning model includes: The machine learning models include support vector regression, nearest neighbor, and random forest models.

10. A deformation analysis system for multi-nail structure connections based on displacement field equivalence and machine learning, characterized in that, include: The equivalent load acquisition module is used to establish a detailed model of a single bolt connection process, analyze the deformation of the connected parts during the connection process based on the detailed model, and obtain the equivalent load based on the deformation displacement field. The dataset acquisition module is used to construct a deformation analysis model for multi-bolt connections and create connection elements in the analysis model; the node information of each connection element corresponds to the single-bolt connection structure; and the sample dataset is obtained by simulating the overall component by inputting the equivalent load. The model training module uses sample datasets to train machine learning models and learns the mapping relationship between load boundary conditions and deformation response from simulation samples. The deformation prediction module is used to predict the deformation of multi-nail structure connections based on the trained machine learning model.