Profile stretch bending deformation prediction method based on machine learning
By combining finite element simulation with machine learning, we have achieved rapid and accurate prediction of profile bending deformation, solving the problems of large computational load and complex parameter adjustment in traditional methods, and improving the efficiency and quality of aerospace manufacturing.
Patent Information
- Application Number
- CN202511643209.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-11
- Publication Date
- 2026-03-24
AI Technical Summary
Traditional methods for predicting the tensile and bending deformation of profiles rely on finite element simulation, which involves large computational loads and complex parameter adjustments, failing to meet the demands of the aerospace industry for efficient production.
By combining finite element simulation data with machine learning technology, and through standardized data processing and machine learning models, we can achieve rapid and accurate prediction of profile bending deformation, identify key process parameters, and optimize the process.
It improves the real-time performance and accuracy of deformation prediction, reduces computing resources and labor costs, reduces material waste, enhances product forming quality and production stability, and promotes the development of intelligent manufacturing.
Smart Images
Figure CN121725943A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of metal forming technology, specifically relating to a machine learning-based method for predicting the tensile bending deformation of profiles. It can be used for predicting the tensile bending deformation and optimizing process parameters of large-section titanium alloy and aluminum alloy profiles, thereby improving process optimization efficiency. It falls under the category of industrial automation and intelligent manufacturing technology. Background Technology
[0002] The aerospace industry demands extremely high dimensional accuracy and forming quality from structural components. Aluminum and titanium alloy profiles, as important lightweight structural materials, are widely used in aircraft frames and load-bearing components. Profile bending is one of the key processes in manufacturing complex curved structures.
[0003] Traditional deformation prediction methods mainly rely on finite element simulation, but high-precision simulation involves large computational loads and complex parameter adjustments, failing to meet the demands of efficient production. On the other hand, machine learning technology demonstrates superior performance in data-driven modeling and prediction, enabling rapid and accurate deformation prediction by learning from large amounts of simulation data. Therefore, this invention designs a machine learning-based method for predicting the tensile and bending deformation of profiles. Summary of the Invention
[0004] This invention aims to provide a method for predicting the deformation of aerospace profile bending members by combining finite element simulation data and machine learning technology. It systematically accumulates bending deformation data of aluminum alloy and titanium alloy profiles and standardizes the data based on component characteristics to facilitate the use of machine learning models for deformation prediction under complex working conditions, thereby improving the real-time performance and accuracy of the prediction. Taking the top-mounted bending of a T-section profile as an example, the specific technical solution is as follows:
[0005] A machine learning-based method for predicting profile bending deformation, characterized in that the method comprises:
[0006] S1: Core variable parameter analysis; S2: Finite element simulation data accumulation; S3: Coordinate system establishment and transformation; S4: Standardized section generation: based on actual profile sections. Based on this, multiply its cross-section by the magnification factor. ( ), to obtain an enlarged standardized cross section S5: Calculation of arc length of standardized profile components: based on the initial length of the profile. Based on, Multiply by the deformation coefficient ( ), to obtain the standardized profile arc length S6: Construction of standardized virtual profile components; S7: Mapping of finite element simulation results; S8: Machine learning model training: Establish a profile bending machine learning model, select 80% of the data in step S2 for machine learning training to achieve rapid prediction of the profile bending process; S9: Machine learning model verification: Verify the accuracy of the machine learning model with the remaining 20% of the data in step S2.
[0007] Furthermore, S1 specifically refers to:
[0008] Based on the structural characteristics of the machine tool, the core variable parameters in the stretch bending forming process are analyzed. Taking the top-type profile stretch bending machine as an example, the stretch bending deformation process includes three core parameters: pre-tension amount, top-lift amount, and supplementary tension amount. The horizontal movement of the left clamp 2 and the right clamp 4 corresponds to the parameter pre-tension amount, and the upward movement of the mold 1 corresponds to the parameter top-lift amount. After bending, the movement of the left clamp 2 and the right clamp 4 along the tangent direction of the mold corresponds to the parameter supplementary tension amount.
[0009] Furthermore, S2 specifically refers to:
[0010] Based on the single variable method, for the three parameters of pre-tension, top tension, and supplementary tension, each parameter is set with n (n≥3) level values. A finite element simulation scheme is designed, and finite element simulation is performed to obtain a large amount of high-precision deformation result dataset. The dataset includes data on material stress distribution, strain distribution, and displacement distribution.
[0011] Furthermore, S3 specifically refers to:
[0012] With the bending center of the profile as the origin of the X and Y coordinate system, the horizontal direction of the profile is X, the vertical symmetry plane of the profile is Y, the width direction of the profile is Z, and the width direction symmetry plane is Z-axis.
[0013] Then, the origin of the established X / Y / Z rectangular coordinate system is converted into the origin of the cylindrical coordinate system, and the X / Y / Z rectangular coordinate system is converted into the cylindrical coordinate system.
[0014] Furthermore, S6 specifically refers to:
[0015] Based on standardized cross section With standard profile arc length Standardized virtual profile components are obtained; the virtual profile components are meshed in a cylindrical coordinate system with a Z-axis mesh size of 1 mm, a radial mesh size of 1 mm, and an angle increment of 1°.
[0016] Furthermore, S7 specifically refers to:
[0017] Using a quadratic approximation method, the finite element simulation of profile bending is mapped onto a standardized virtual profile component mesh, including the mapping of stress distribution, strain distribution, and displacement. This allows for the acquisition of output results under different input conditions. The data of the nodes on the standardized mesh are solved using the surrounding finite element simulation data points. The input conditions serve as the input set, and the values on the standardized virtual profile component mesh serve as the output set.
[0018] The method of the present invention is a two-dimensional tensile-bending deformation prediction method for profiles with different cross-sections.
[0019] The advantages of this invention compared to the prior art are as follows:
[0020] The beneficial effects of this invention are as follows:
[0021] (1) Based on the standardized data processing method proposed in this invention, the tensile bending finite element simulation results under different process parameters can be standardized, providing data in a unified format for machine learning data processing.
[0022] (2) By combining finite element simulation data with machine learning technology, the present invention can quickly and accurately predict the bending deformation of profiles under complex working conditions, which significantly improves the accuracy of deformation results and meets the high precision requirements of structural components in the aerospace field.
[0023] (3) The present invention can identify key process parameters that affect forming quality by analyzing a large amount of simulation data, and provide optimization suggestions for actual production, helping process engineers to optimize the bending process and thus improve the forming quality of the product.
[0024] (4) Since the present invention can effectively reduce the computational resources and manpower costs required by traditional simulation methods, it reduces the investment in the research and development and production process, and at the same time reduces material waste by optimizing process parameters, thereby reducing the overall production cost.
[0025] (5) The method of the present invention is not only applicable to titanium alloy and aluminum alloy profiles, but can also be extended to the bending process of profiles of other material types. It has strong versatility and provides a flexible solution for the manufacturing of different materials.
[0026] (6) Real-time monitoring and feedback: Combining machine learning technology, the system can monitor the deformation during the bending process in real time and make adjustments based on real-time data feedback to ensure that the production process is always in the best process state and improve the overall production stability.
[0027] (7) Promote the development of intelligent manufacturing: This invention embodies the concept of industrial automation and intelligent manufacturing, promotes the intelligentization process of metal forming technology, helps to improve the technical level and competitiveness of the entire industry, and promotes the development of the aerospace manufacturing industry towards a more efficient and intelligent direction. Attached Figure Description
[0028] Figure 1 For: A schematic diagram of the assembly and motion parameters of each component before the profile is bent;
[0029] Figure 2 Here is a schematic diagram showing the position and motion parameters of each component after the profile is bent.
[0030] Figure 3 For: A three-dimensional axonometric view of the position of each component after the profile is bent;
[0031] Figure 4 For: Standardized schematic diagram of profile cross-sectional shape;
[0032] Figure 5 Here is a schematic diagram of a standardized data processing bar coordinate system (X and Y axes);
[0033] Figure 6 Here is a schematic diagram of a standardized data processing bar coordinate system (Z-axis).
[0034] Figure 7 For example: A three-dimensional axonometric schematic diagram of a standardized data processing cylindrical coordinate system;
[0035] Figure 8 Here is a schematic diagram of the simulation data standardization method. Detailed Implementation
[0036] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the following examples provide a more detailed description of the invention. It should be noted that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.
[0037] The present invention takes the top-mounted tension bending of a T-section profile as an example, and the specific technical solution is as follows:
[0038] S1: Core Variable Parameter Analysis: Based on the machine tool's structural characteristics, this section analyzes the core variable parameters during the stretch bending process. Taking an overhead profile stretch bending machine as an example... Figure 1As shown, the core components include mold 1, left clamp 2, profile 3, and right clamp 4. Left clamp 2 and right clamp 4 clamp the two ends of profile 3 respectively, and mold 1 is located on the lower side of the middle of the profile. During the profile bending process, left clamp 2 and right clamp 4 stretch to both sides, then mold 1 pushes upwards, and finally left clamp 2 and right clamp 4 stretch tangentially to both sides. Correspondingly, the bending deformation process mainly includes three core parameters: pre-tension, upward pushing amount, and supplementary tension. The horizontal movement of left clamp 2 and right clamp 4 corresponds to the pre-tension parameter, and the upward movement of mold 1 corresponds to the upward pushing parameter. Figure 2 and Figure 3 The image shows the state of the profile after bending. After bending, the left clamp 2 and the right clamp 4 move along the tangent of the mold by a corresponding parameter of the additional stretching amount.
[0039] S2: Finite Element Simulation Data Accumulation: Based on the single variable method, for the three parameters of pre-tension, top tension, and supplementary tension, each parameter is set with n (n≥3) level values. A finite element simulation scheme is designed, and finite element simulations are performed to obtain a large amount of high-precision deformation result dataset. The dataset includes material stress distribution, strain distribution, displacement distribution, and other data.
[0040] S3: Coordinate System Establishment and Transformation: The center of the profile bending circle is taken as the origin of the X and Y axes of the coordinate system. The horizontal direction of the profile is the X-axis, and the vertical symmetry plane of the profile is the Y-axis. Figure 5 As shown, the profile's width direction (Z-axis) has its symmetry plane along the width direction as the origin of the Z-axis. Figure 6 As shown. Then, the origin of the established X / Y / Z rectangular coordinate system is converted to the origin of the cylindrical coordinate system, and the X / Y / Z rectangular coordinate system is converted to the cylindrical coordinate system, as follows. Figure 5 As shown;
[0041] S4: Standardized Section Generation: Based on the actual profile section Based on this, multiply its cross-section by the magnification factor. ( ), to obtain an enlarged standardized cross section ,like Figure 4 As shown in the figure, section 5 is a schematic diagram of the enlarged standardized section;
[0042] S5: Standardized Profile Component Arc Length Calculation: Based on the initial length of the profile Based on, Multiply by the deformation coefficient ( ), to obtain the standardized profile arc length ;
[0043] S6: Construction of Standardized Virtual Profile Components: Based on Standardized Cross Sections With standard profile arc length This yields standardized virtual profile components. The virtual profile components are then meshed in a cylindrical coordinate system, with a Z-axis mesh size of 1 mm, a radial mesh size of 1 mm, and an angle increment of 1°. Figure 5 and Figure 7 As shown;
[0044] S7: Finite Element Simulation Result Mapping: Using a quadratic approximation method, the finite element simulation of the profile bending tension is mapped onto a standardized virtualized profile component mesh. This includes mapping of stress distribution, strain distribution, and displacement, thereby obtaining output results under different input conditions, such as... Figure 8 As shown, the data of the nodes on the standardized mesh are solved using the surrounding finite element simulation data points. The input conditions serve as the input set, and the values on the standardized, virtualized profile component mesh serve as the output set.
[0045] S8: Machine learning model training: Establish a profile bending machine learning model, select 80% of the data in step S2 for machine learning training, so as to achieve rapid prediction of the profile bending process.
[0046] S9: Machine learning model validation: Validate the accuracy of the machine learning model using the remaining 20% of the data from step S2.
[0047] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.
Claims
1. A method for predicting tensile bending deformation of profiles based on machine learning, characterized in that, The method is as follows: S1: Core variable parameter analysis; S2: Accumulation of finite element simulation data; S3: Coordinate system establishment and transformation; S4: Standardized Section Generation: Based on the actual profile section Based on this, multiply its cross-section by the magnification factor. ( ), to obtain an enlarged standardized cross section ; S5: Standardized Profile Component Arc Length Calculation: Based on the initial length of the profile Based on, Multiply by the deformation coefficient ( ), to obtain the standardized profile arc length ; S6: Standardized virtual profile component construction; S7: Finite element simulation result mapping; S8: Machine learning model training: Establish a profile bending machine learning model, select 80% of the data in step S2 for machine learning training, so as to achieve rapid prediction of the profile bending process; S9: Machine learning model validation: Validate the accuracy of the machine learning model using the remaining 20% of the data from step S2.
2. The method for predicting profile bending deformation based on machine learning according to claim 1, characterized in that, Specifically, S1 refers to: Based on the structural characteristics of the machine tool, the core variable parameters in the stretch bending forming process are analyzed. Taking the top-type profile stretch bending machine as an example, the stretch bending deformation process includes three core parameters: pre-tension amount, top-lift amount, and supplementary tension amount. The horizontal movement of the left clamp 2 and the right clamp 4 corresponds to the parameter pre-tension amount, and the upward movement of the mold 1 corresponds to the parameter top-lift amount. After bending, the movement of the left clamp 2 and the right clamp 4 along the tangent direction of the mold corresponds to the parameter supplementary tension amount.
3. The method for predicting profile bending deformation based on machine learning according to claim 1, characterized in that, Specifically, S2 is: Based on the single variable method, for the three parameters of pre-tension, top tension, and supplementary tension, each parameter is set with n (n≥3) level values. A finite element simulation scheme is designed, and finite element simulation is performed to obtain a large amount of high-precision deformation result dataset. The dataset includes data on material stress distribution, strain distribution, and displacement distribution.
4. The method for predicting profile bending deformation based on machine learning according to claim 1, characterized in that, Specifically, S3 is: With the bending center of the profile as the origin of the X and Y coordinate system, the horizontal direction of the profile is X, the vertical symmetry plane of the profile is Y, the width direction of the profile is Z, and the width direction symmetry plane is Z-axis. Then, the origin of the established X / Y / Z rectangular coordinate system is converted into the origin of the cylindrical coordinate system, and the X / Y / Z rectangular coordinate system is converted into the cylindrical coordinate system.
5. The method for predicting profile bending deformation based on machine learning according to claim 1, characterized in that, Specifically, S6 is: Based on standardized cross section With standard profile arc length Standardized virtual profile components are obtained; the virtual profile components are meshed in a cylindrical coordinate system with a Z-axis mesh size of 1 mm, a radial mesh size of 1 mm, and an angle increment of 1°.
6. The method for predicting profile bending deformation based on machine learning according to claim 1, characterized in that, Specifically, S7 refers to: Using a quadratic approximation method, the finite element simulation of profile bending is mapped onto a standardized virtual profile component mesh, including the mapping of stress distribution, strain distribution, and displacement. This allows for the acquisition of output results under different input conditions. The data of the nodes on the standardized mesh are solved using the surrounding finite element simulation data points. The input conditions serve as the input set, and the values on the standardized virtual profile component mesh serve as the output set.
7. A method for predicting profile bending deformation based on machine learning according to any one of claims 1 to 6, characterized in that, The method described is a two-dimensional tensile-bending deformation prediction method for profiles with different cross-sections.