Local rebound radius-based thin-wall part rebound prediction characterization method and device, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of sheet metal part forming, and relates to a thin-wall part springback prediction characterization method and device based on a local springback radius, electronic equipment and a storage medium. The method comprises the following steps: calculating the local rebound radius of each point according to a thin-wall part finite element model or discrete point coordinates obtained by experimental measurement; calculating a structure change factor of each discrete point according to the structure characteristics of the thin-wall part; and training a neural network model by taking the local springback radius, the structure change factor and the forming process parameters as input characteristics to predict the local springback radius and other key parameters under a target working condition, and finally reconstructing the overall profile coordinate of the thin-walled part after springback through a geometrical relationship.

Description

Technical Field

[0001] This invention belongs to the field of sheet metal forming technology, and relates to a method, device, electronic device and storage medium for predicting and characterizing the springback of thin-walled parts based on the local springback radius. Background Technology

[0002] In the field of sheet metal forming and manufacturing, springback is a key factor affecting the final surface accuracy of components. Accurate springback prediction is crucial for improving product quality and shortening the R&D cycle. Currently, springback prediction mainly relies on the finite element method based on material constitutive models, but this method has high computational costs and cannot meet the needs of rapid design and optimization iteration.

[0003] In recent years, machine learning methods have been introduced into rebound prediction. However, existing data-driven models mostly rely on the absolute coordinates of component nodes as input or output features, resulting in a strong correlation between their predictive ability and the size and geometry of the training samples. When the size or structure of the object to be predicted exceeds the range of the training set, the model accuracy drops significantly, and its generalization ability is insufficient, severely limiting its widespread application in engineering. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned defects of the prior art and provide a springback prediction and characterization method for thin-walled parts based on local springback radius. This method effectively improves the generalization ability of the neural network springback prediction model for thin-walled parts of different sizes and structures by introducing local springback radius and structural change factor as core characterization parameters.

[0005] Technical solution: Firstly, a method for predicting and characterizing the springback of thin-walled components based on local springback radius is provided, the specific steps of which include: S1: Calculate the local springback radius of each point based on the coordinates of discrete points obtained from the finite element model of the thin-walled component or experimental measurements; S2: Calculate the structural variation factor for each discrete point based on the structural characteristics of the thin-walled component; S3: The local springback radius, structural change factor and forming process parameters are used as input features to train a neural network model to predict the local springback radius and other key parameters under the target working condition. Finally, the overall surface coordinates of the thin-walled part after springback are reconstructed through geometric relationships.

[0006] Furthermore, in step S1, the method for calculating the local rebound radius is as follows: for three consecutive adjacent discrete points, the local rebound radius of the middle point is calculated using the formula for the radius of the circumcircle of the triangle formed by these three points. The calculation formula is: ; Where (x1,y1), (x2,y2), and (x3,y3) are the coordinates of the three adjacent discrete points.

[0007] Furthermore, in step S2, the calculation of the structural variation factor considers the changes in structural parameters of adjacent regions along the positive and negative directions of the discrete point's path. These structural parameters include thickness or rib height. The structural variation factor ( The formula for calculating ) is: ; in, The structural parameters at this discrete point are... Structural parameters for adjacent regions in a specified direction, The distance from the discrete point to the adjacent region is the structural change factor, which has two directions: positive and negative, and is expressed as follows: and .

[0008] Furthermore, in step S3, the neural network model is a BP neural network.

[0009] Furthermore, the process parameters include one or more of the following: mold curvature radius, forming temperature, and holding time; and the structural parameters include one or more of the following: skin thickness, rib height, and rib thickness.

[0010] Furthermore, in step S3, the specific method for reconstructing the surface coordinates based on the predicted local springback radius is as follows: Based on the predicted local rebound radius Initial length between discrete points Calculate the central angle corresponding to this curve segment. Based on the central angle, the incremental displacement of the discrete point in the x and y directions is calculated using geometric trigonometric relationships. and By accumulating the incremental displacements, the coordinates of each discrete point after rebound are obtained. , This reconstructs the entire surface profile, and the calculation formulas are as follows: ; ; ; ; .

[0011] Secondly, a springback prediction and characterization device for thin-walled parts based on local springback radius is provided, the specific steps of which include: The calculation module is used to calculate the local springback radius of each point based on the coordinates of discrete points obtained from the finite element model or experimental measurements of the thin-walled component; and to calculate the structural change factor of each discrete point based on the structural characteristics of the thin-walled component. The training prediction module is used to train a neural network model by taking the local springback radius, structural change factor and forming process parameters as input features to predict the local springback radius and other key parameters under the target working condition. Finally, the overall surface coordinates of the thin-walled part after springback are reconstructed through geometric relationships.

[0012] Thirdly, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of the first aspects.

[0013] Fourthly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described in any one of the first aspects.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Excellent generalization ability: This invention uses the local rebound radius, an intrinsic physical quantity, to replace the absolute coordinates of nodes as the core representation. This fundamentally decouples the absolute size relationship between the prediction model and the training samples, enabling the trained model to accurately predict the rebound of components far exceeding the size range of the training set. This solves the industry problem of poor size generalization ability of existing data-driven models.

[0015] 2. Detailed description of complex structures: By introducing a structural change factor, this invention can effectively quantify the local impact of complex structural features such as abrupt thickness changes, stiffener boundaries, and free ends on springback, significantly improving the model's prediction accuracy and robustness for thin-walled components with complex structures.

[0016] 3. High versatility and broad application prospects: This characterization method does not depend on the constitutive model of a specific material. Its input and output characteristics are universal and can be widely used in springback prediction of various sheet metal forming processes such as creep aging forming, stamping, and bending, providing the industry with an efficient and universal springback prediction solution. Attached Figure Description

[0017] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0018] Figure 1 This is a schematic diagram of the overall process of the method of the present invention; Figure 2 This is a finite element model diagram of a typical reinforced wall panel structure; Figure 3This is a schematic diagram of the BP neural network structure used in an embodiment of the present invention; Figure 4 A schematic diagram of the geometric relationship for reconstructing model node coordinates; Figure 5a This is a comparison chart of the prediction and simulation results of the method of this patent for a half-length 760mm wall panel. The comparison includes the springback profile and the local springback radius and yield strength. Figure 5b This is a comparison chart of the prediction and simulation results of the method of this patent for a half-length 760mm wall panel. The comparison includes incremental displacement ∆x, ∆y and nodal position error. Figure 6a The image shows a comparison between the prediction and simulation results of a traditional coordinate model for a 760mm half-length wall panel. The comparison includes the springback profile and the local springback radius and yield strength. Figure 6b This is a comparison chart of the prediction and simulation results of a traditional coordinate model for a half-length 760mm wall panel. The comparison includes incremental displacements ∆x and ∆y, and nodal position errors. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described examples are merely some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] This invention provides a method for predicting and characterizing the springback of thin-walled components based on local springback radius, such as... Figure 1 As shown, the specific steps include: S1: Calculate the local springback radius of each point based on the coordinates of discrete points obtained from the finite element model of the thin-walled component or experimental measurements.

[0021] The method for calculating the local rebound radius is as follows: For three consecutive adjacent discrete points, the local rebound radius of the middle point is calculated using the formula for the radius of the circumcircle of the triangle formed by these three points. The calculation formula is: ; Where (x1,y1), (x2,y2), and (x3,y3) are the coordinates of the three adjacent discrete points.

[0022] S2: Calculate the structural variation factor for each discrete point based on the structural characteristics of the thin-walled component.

[0023] The calculation of the structural variation factor considers the changes in structural parameters of adjacent regions along the positive and negative directions of the path of the discrete point. These structural parameters include thickness or rib height. The structural variation factor ( The formula for calculating ) is: ; in, The structural parameters at this discrete point are... Structural parameters for adjacent regions in a specified direction, The distance from the discrete point to the adjacent region is the structural change factor, which has two directions: positive and negative, and is expressed as follows: and .

[0024] S3: The local springback radius, structural change factor and forming process parameters are used as input features to train a neural network model to predict the local springback radius and other key parameters under the target working condition. Finally, the overall surface coordinates of the thin-walled part after springback are reconstructed through geometric relationships.

[0025] The neural network model is a BP neural network.

[0026] The process parameters include one or more of the following: mold curvature radius, forming temperature, and holding time; the structural parameters include one or more of the following: skin thickness, rib height, and rib thickness.

[0027] The specific method for reconstructing the surface coordinates based on the predicted local springback radius is as follows: Based on the predicted local rebound radius Initial length between discrete points Calculate the central angle corresponding to this curve segment. Based on the central angle, the incremental displacement of the discrete point in the x and y directions is calculated using geometric trigonometric relationships. and By accumulating the incremental displacements, the coordinates of each discrete point after rebound are obtained. , This reconstructs the entire surface profile, and the calculation formulas are as follows: ; ; ; ; .

[0028] The present invention also provides a springback prediction and characterization device for thin-walled parts based on local springback radius, the specific steps of which include: The calculation module is used to calculate the local springback radius of each point based on the coordinates of discrete points obtained from the finite element model or experimental measurements of the thin-walled component; and to calculate the structural change factor of each discrete point based on the structural characteristics of the thin-walled component. The training prediction module is used to train a neural network model by taking the local springback radius, structural change factor and forming process parameters as input features to predict the local springback radius and other key parameters under the target working condition. Finally, the overall surface coordinates of the thin-walled part after springback are reconstructed through geometric relationships.

[0029] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in any one of the first aspects.

[0030] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any one of the first aspects.

[0031] Example Taking the creep aging forming process of 7B50 aluminum alloy single-curvature high-ribbed wall panel as an example, the springback prediction method of this invention is applied. This embodiment aims to specifically illustrate the implementation process and excellent effect of this method, but the application of this invention is not limited to this specific process or material.

[0032] S1: Conduct finite element simulations of typical reinforced wall panel structures to obtain sample data.

[0033] First, based on the characteristics of highly reinforced wall panels, finite element models of highly reinforced wall panels with different structural parameters and creep aging forming process parameters were established, such as... Figure 2As shown in the figure, this is a quarter-symmetric model. The model has a half-length of 580mm and a half-width of 400mm, using S4R shell elements with a mesh size of 5mm × 10mm. The structural parameters mainly include two typical structures: one is a variable-thickness skin panel, whose skin is composed of regions of different thicknesses (e.g., 2mm, 3mm, 5mm, 8mm); the other is a skin-rib structure panel, which includes a skin of uniform thickness and several ribs (rib thickness 2mm, height 25mm or 40mm). The process parameters include the mold curvature radius (650mm, 1000mm, 2000mm, 4000mm) and the aging time in the aging process (2h, 2.2h, 2.5h, 2.8h, 3h). By combining these parameters, a total of 294 simulation conditions are generated. We embedded a self-developed unified constitutive model subroutine for the two-stage creep aging of 7B50 aluminum alloy into the ABAQUS finite element software to simulate the forming and springback process. This accurately calculated the nodal coordinates, stress field, and material yield strength of the panel after springback, thereby obtaining a large number of diverse sample datasets, providing a solid foundation for subsequent neural network training.

[0034] S2: Extract finite metadata and calculate the local rebound radius.

[0035] From the finite element simulation results, the coordinate sequence of each node on the centerline (Z=0) of the rebounded panel skin is extracted. Using the formula for the radius of the circumcircle of a triangle in claim 2, the local rebound radius of each node is calculated. By traversing all internal nodes (except the first and last points), the local rebound radius distribution curve along the length of the panel can be obtained. This step transforms the spatial coordinate information of the nodes into a more physically meaningful and size-independent local curvature representation.

[0036] S3: Calculate the structural change factor.

[0037] For each node in S2, based on its location and the structural characteristics of the wall panel, the structural change factor in both the positive and negative directions of the path is calculated using the formula in claim 3. and For nodes at the free end boundary, by... Set it to a minimum value (e.g., 0.1 mm). The distance from the node to the free end is used to quantify the free end effect. This factor effectively captures the local mechanical effects at structural discontinuities.

[0038] S4: Training the neural network and predicting reconstruction.

[0039] S41: Model Building and Training. Building such a model... Figure 3The BP neural network model shown. Input layer features include: process parameters (mold radius, high-temperature holding time), structural parameters (structural thickness), and structural variation factors in two directions corresponding to each node. and The output layer targets the local rebound radius calculated by S2 and the material yield strength obtained from finite element simulation. The hidden layer of the network has 64 neurons, using the ReLU activation function, and the mean squared error (MSE) loss function. The 294 simulation cases were randomly divided into training, test, and validation sets at a ratio of 80%, 10%, and 10%, respectively. Both input and output data were preprocessed using the MinMaxScaler function for normalization to improve the stability and convergence efficiency of network training.

[0040] S42: Model Prediction and Validation. To validate the model's generalization ability, a large panel with a half-length of 760 mm was selected as the prediction object. This panel contains eight skin segments of different thicknesses (3.5 mm, 4 mm, 5 mm) and eight ribs of different heights (38 mm, 25 mm, 35 mm) and thicknesses (3.5 mm, 2 mm). Its size and structural complexity significantly exceed the range of the training set samples (half-length of 580 mm). The process parameters (mold radius 1600 mm, high-temperature time 2.6 h), structural parameters, and calculated structural change factors of this new panel were input into the trained model to predict the local springback radius and yield strength at each point.

[0041] S43: Surface coordinate reconstruction. Utilizing the local springback radius sequence predicted by the model { }, according to the geometric mapping relationship in claim 5 (the principle of which is as follows) Figure 4 As shown, the surface coordinates are reconstructed. Specifically, this is done based on the initial mesh length. and predicted radius Calculate the central angle Then calculate the coordinate increment. and Finally, the coordinates of all nodes after bounce are obtained by summing them up. , This allows for the complete reconstruction of the spring-loaded surface of the large wall panel.

[0042] S44: Effect Verification. Verification showed that the maximum error in predicting and reconstructing the node positions of this large wall panel using the method of this invention was 1.012 mm, with a relative error of only 3.24%. The predicted yield strength had an error of less than 1 MPa in the skin region and less than 5 MPa in the rib region. A comparison of the predicted springback profile, local springback radius, and strength with the finite element simulation results is shown below. Figure 5a and Figure 5bAs shown, the curves of both models exhibit largely consistent trends, indicating that the model possesses good prediction accuracy. In contrast, a conventional neural network model using pre-formation node coordinates as input demonstrates a relative error of up to 14.66% in node position under the same prediction task, and its intensity prediction shows significant deviations at distant points (maximum error reaching 65 MPa). Its comparison is as follows: Figure 6a and Figure 6b As shown in the figure, this result fully demonstrates that the rebound prediction and characterization method based on local rebound radius proposed in this invention has excellent performance and significant advantages in effectively decoupling the model's dependence on the size of training samples and improving the model's generalization ability.

[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for characterizing springback prediction of thin-walled parts based on local springback radius, characterized in that, The specific steps include: S1: calculating the local springback radius of each point according to the discrete point coordinates obtained from the finite element model of the thin-walled part or experimental measurement; S2: calculating the structure change factor of each discrete point according to the structural characteristics of the thin-walled part; S3: taking the local springback radius, the structure change factor and the forming process parameters as input features together, training a neural network model to predict the local springback radius and other key parameters under the target working condition, and finally reconstructing the overall profile coordinates of the thin-walled part after springback through geometric relationship.

2. The method of claim 1, wherein, In step S1, the calculation method of the local rebound radius is as follows: for three adjacent discrete points, the local rebound radius of the middle point is calculated by the radius formula of the circumscribed circle of the triangle formed by the three points, and the radius formula of the circumscribed circle is as follows: The calculation formula of the radius of the circumscribed circle is as follows: ； Wherein, (x1, y1), (x2, y2), (x3, y3) are the coordinates of the three adjacent discrete points.

3. The method of claim 1, wherein, In step S2, the calculation of the structural variation factor considers the changes in structural parameters of adjacent regions along the positive and negative directions of the discrete point's path. These structural parameters include thickness or rib height. The structural variation factor ( The formula for calculating ) is: ； wherein, is the structure parameter at the discrete point, is the structure parameter of the adjacent region in the specified direction, is the distance from the discrete point to the adjacent region, the structure change factor has two directions of positive and negative, respectively represented as and .

4. The method of claim 1, wherein, In step S3, the neural network model is a BP neural network.

5. The method of claim 1, wherein, The process parameters include one or more of the mold curvature radius, the forming temperature and the holding time, and the structure parameters include one or more of the skin thickness, the rib height and the rib thickness.

6. The method of claim 1, wherein, In step S3, the specific method of reconstructing the profile coordinates according to the predicted local springback radius is: According to the predicted local rebound radius And the initial length between discrete points , Calculate the central angle of the circle corresponding to the segment of the curve ; Based on the central angle, the incremental displacement of the discrete point in the x and y directions is calculated by geometric trigonometric relationship And ; By accumulating the incremental displacement, the coordinates of each discrete point after rebounding are obtained , ), So as to reconstruct the whole surface, and the calculation formula is as follows: ； ； ； ； 。 7. A local springback radius based thin-walled part springback prediction characterization apparatus, comprising: The specific steps include: The calculation module is configured to calculate the local springback radius of each point according to the discrete point coordinates obtained from the finite element model of the thin-walled part or experimental measurement, and calculate the structure change factor of each discrete point according to the structural characteristics of the thin-walled part; The training and prediction module is configured to take the local springback radius, the structure change factor and the forming process parameters as input features together, train a neural network model to predict the local springback radius and other key parameters under the target working condition, and finally reconstruct the overall profile coordinates of the thin-walled part after springback through geometric relationship.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the method of any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the method of any one of claims 1-6. The program is executed by the processor to implement the method of any one of claims 1-6.

Images