Pipe bending forming springback prediction method and device, equipment and medium

Abstract

The invention provides a pipe bending forming springback prediction method and device, equipment and a medium, and relates to the technical field of data processing. The method comprises the following steps: firstly, acquiring a first number group of sample pipe bending forming data; secondly, redundancy screening of pipe characteristic quantities is conducted on the first number group of sample pipe bending forming data, and a first number group of screened pipe bending forming data is formed; then, based on each group of screening pipe bending forming data and the corresponding springback angle label, a springback angle prediction model is constructed, and based on each group of screening pipe bending forming data and the corresponding springback radius label, a springback radius prediction model is constructed; and finally, by using the rebound angle prediction model and the rebound radius prediction model, determining target process parameter data of which the rebound angle meets a rebound angle threshold condition and the rebound radius meets a rebound radius threshold condition. Based on the above content, the problem of low prediction precision or high prediction cost in the prior art can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to a method, apparatus, equipment, and medium for predicting the springback of pipe bending forming. Background Technology

[0002] Thin-walled tubing, characterized by its lightweight, high strength and toughness, and high precision, is widely used in aerospace, shipbuilding, and automotive industries. To achieve the transmission of media and energy within limited space, three-dimensional bending forming of the tubing is necessary. During the bending forming process of thin-walled tubing, defects such as wrinkling, cracking, and springback are particularly prone to occur. Springback is one of the major defects, manifesting as a rebound in the bending radius and angle, especially during cold bending (room temperature bending), where it is particularly pronounced. This not only affects the processing accuracy and efficiency of the tubing but also forces assembly during subsequent processes, thus impacting service performance. With increasingly stringent quality and precision requirements for thin-walled bent tubing, precise control of the bending springback is essential. Since tubing bending is a complex elastoplastic forming process, incorporating material nonlinearity, geometric nonlinearity, and boundary condition nonlinearity, bending springback is influenced by the intrinsic properties of the tubing, its geometric dimensions, process parameters, and conditional parameters. Traditional trial-and-error methods obtain process parameters for controlling bending springback through combined bending experiments. However, this approach wastes significant time and resources, as each material and specification change requires a separate experiment. Furthermore, even if a control method is found for a particular type of pipe, fluctuations in material properties and slight changes in pipe geometry can cause springback fluctuations, affecting the quality of the bent pipe fittings. In contrast, analytical methods and finite element simulations can effectively reduce costs and improve efficiency in controlling the bending springback of thin-walled pipes. However, these technologies currently have the following drawbacks: 1) Traditional analytical methods struggle to handle the coupling effects of material nonlinearity, geometric nonlinearity, and boundary condition nonlinearity, resulting in relatively low prediction accuracy; 2) Finite element simulation calculations have high time costs per operation, leading to relatively high prediction costs. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, apparatus, equipment and medium for predicting the springback of pipe bending forming, so as to improve the problems of low prediction accuracy or high prediction cost in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution: A method for predicting the springback of pipes during bending and forming includes: Obtain a first set of sample pipe bending forming data, wherein each set of sample pipe bending forming data includes the value of at least one process parameter and the values ​​of multiple pipe characteristic quantities, and each set of sample pipe bending forming data corresponds to a springback angle label and a springback radius label. Redundancy screening of pipe feature quantities is performed on the first set of sample pipe bending forming data to form the first set of screened pipe bending forming data, wherein the number of pipe feature quantities included in each set of screened pipe bending forming data is less than or equal to the number of pipe feature quantities included in each set of sample pipe bending forming data. At least based on each set of the selected pipe bending forming data and the corresponding springback angle label, a springback angle prediction model is constructed, and at least based on each set of the selected pipe bending forming data and the corresponding springback radius label, a springback radius prediction model is constructed. Using the springback angle prediction model and the springback radius prediction model, a leading edge search is performed under the pipe material characteristics of the target pipe to determine the target process parameter data that satisfy both the springback angle threshold condition and the springback radius threshold condition.

[0005] In a preferred embodiment of this application, in the above-described pipe bending springback prediction method, the step of performing redundancy filtering of pipe feature quantities on the first set of sample pipe bending data to form a first set of filtered pipe bending data includes: The first redundancy screening of pipe bending forming data in the first group of sample pipes is performed to form intermediate pipe bending forming data in the first group. The first redundancy screening refers to the screening based on the correlation between pipe features. The pipe bending and forming data in the first quantity group are subjected to a second redundancy screening of pipe feature quantities to form a first quantity group of screened pipe bending and forming data. The second redundancy screening refers to the screening based on the correlation importance between pipe feature quantities and labels.

[0006] In a preferred embodiment of this application, the step of performing a first redundancy screening of pipe feature quantities on the first set of sample pipe bending forming data to form a first set of intermediate pipe bending forming data in the above-mentioned pipe bending forming method includes: For each pair of pipe characteristics, the correlation coefficient between the two pipe characteristics is determined based on the values ​​of the two pipe characteristics in the bending forming data of the first group of sample pipes. Based on the correlation coefficient between each pair of pipe characteristics, the first redundancy screening of pipe characteristics corresponding to the first set of sample pipe bending and forming data is performed to form the first set of intermediate pipe bending and forming data.

[0007] In a preferred embodiment of this application, in the aforementioned pipe bending springback prediction method, the step of performing a first redundancy screening of multiple pipe feature quantities corresponding to the first set of sample pipe bending forming data based on the correlation coefficient between each pair of pipe feature quantities to form the first set of intermediate pipe bending forming data includes: For each pair of pipe material characteristics, determine whether the absolute value of the correlation coefficient between the two pipe material characteristics is greater than a predetermined correlation coefficient threshold. For each pair of pipe material features whose absolute value of the correlation coefficient is greater than the correlation coefficient threshold, a first machine learning model and a second machine learning model are established with the first pipe material feature as the input and the springback angle and springback radius as the outputs. A third machine learning model and a fourth machine learning model are established with the second pipe material feature as the input and the springback angle and springback radius as the outputs. Using the values ​​of the first type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, the first machine learning model is trained and the corresponding first training error is determined. Using the values ​​of the first type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, the second machine learning model is trained and the corresponding second training error is determined. Using the values ​​of the second type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, the third machine learning model is trained and the corresponding third training error is determined. Using the values ​​of the second type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, the fourth machine learning model is trained and the corresponding fourth training error is determined. Based on the magnitude relationship between the first training error and the third training error, the corresponding two types of pipe feature quantities are screened out to form a first set of intermediate pipe bending forming data corresponding to the springback angle label. Based on the magnitude relationship between the second training error and the fourth training error, the corresponding two types of pipe feature quantities are screened out to form a first set of intermediate pipe bending forming data corresponding to the springback radius label.

[0008] In a preferred embodiment of this application, in the above-mentioned pipe bending springback prediction method, the step of performing a second redundancy screening of pipe characteristic quantities on the intermediate pipe bending data of the first quantity group to form the first quantity group of screened pipe bending data includes: For each pipe feature included in the first set of intermediate pipe bending forming data, a fifth machine learning model and a sixth machine learning model are established, with each pipe feature other than the first set of intermediate pipe bending forming data as input and springback angle and springback radius as output. The fifth machine learning model is trained using the value of each pipe feature input in the first set of intermediate pipe bending forming data and the corresponding springback angle label, and the corresponding fifth training error is determined. The sixth machine learning model is trained using the value of each pipe feature input in the first set of intermediate pipe bending forming data and the corresponding springback radius label, and the corresponding sixth training error is determined. In each type of pipe feature included in the first set of intermediate pipe bending forming data, the pipe feature with the minimum value of the corresponding fifth training error is screened out to form the first set of screened pipe bending forming data corresponding to the springback angle label. In each type of pipe feature included in the first set of intermediate pipe bending forming data, the pipe feature with the minimum value of the corresponding sixth training error is screened out to form the first set of screened pipe bending forming data corresponding to the springback radius label.

[0009] In a preferred embodiment of this application, in the above-described pipe bending forming springback prediction method, the steps of constructing a springback angle prediction model based at least on each set of the selected pipe bending forming data and the corresponding springback angle label, and constructing a springback radius prediction model based at least on each set of the selected pipe bending forming data and the corresponding springback radius label, include: Based on the process parameter expansion method, the first group of screened pipe bending forming data corresponding to the springback angle label is expanded to form a second group of screened pipe bending forming data corresponding to the springback angle label. Also, based on the process parameter expansion method, the first group of screened pipe bending forming data corresponding to the springback radius label is expanded to form a second group of screened pipe bending forming data corresponding to the springback radius label. In this case, the value of the same process parameter is the same in the first group of sample pipe bending forming data. A springback angle prediction model is trained by filtering pipe bending forming data using a second set of data corresponding to the springback angle label, and a springback radius prediction model is trained by filtering pipe bending forming data using a second set of data corresponding to the springback radius label.

[0010] In a preferred embodiment of this application, the steps of using a second set of data corresponding to the springback angle label to filter pipe bending forming data and training a springback angle prediction model, and using a second set of data corresponding to the springback radius label to filter pipe bending forming data and training a springback radius prediction model, include: Using process parameters and pipe characteristics as inputs and springback angle as output, multiple initial springback angle models are constructed. Furthermore, pipe bending forming data are screened using a second set of numbers corresponding to springback angle labels. Each of the initial springback angle models is trained to form multiple trained springback angle models. The trained springback angle model with the smallest error is determined as the springback angle prediction model. Using process parameters and pipe characteristics as inputs and springback radius as output, multiple initial springback radius models are constructed. Furthermore, pipe bending forming data are screened using a second set of data corresponding to the springback radius label. Each of the initial springback radius models is trained to form multiple trained springback radius models. The trained springback radius model with the smallest error is determined as the springback radius prediction model.

[0011] This application also provides a pipe bending forming springback prediction device, comprising: The data acquisition module is used to acquire a first number of sets of sample pipe bending forming data, wherein each set of sample pipe bending forming data includes the value of at least one process parameter and the value of multiple pipe characteristic quantities, and each set of sample pipe bending forming data corresponds to a springback angle label and a springback radius label. The feature quantity filtering module is used to perform redundant filtering of pipe feature quantities on the first number of sample pipe bending and forming data to form the first number of filtered pipe bending and forming data, wherein the number of pipe feature quantities included in each group of filtered pipe bending and forming data is less than or equal to the number of pipe feature quantities included in each group of sample pipe bending and forming data. The prediction model building module is used to build a springback angle prediction model based on at least each set of the selected pipe bending forming data and the corresponding springback angle label, and to build a springback radius prediction model based on at least each set of the selected pipe bending forming data and the corresponding springback radius label. The parameter determination module is used to perform a frontal surface search under the pipe characteristics of the target pipe using the springback angle prediction model and the springback radius prediction model, and to determine the target process parameter data that satisfy the springback angle threshold condition and the springback radius threshold condition.

[0012] Based on the above, this application also provides an electronic device, including: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-described method for predicting the springback of pipe bending.

[0013] Based on the above, this application also provides a computer-readable storage medium storing a computer program that, when executed, performs each step of the above-described method for predicting the springback of pipe bending.

[0014] The pipe bending forming springback prediction method, apparatus, equipment, and medium provided in this application firstly acquire a first set of sample pipe bending forming data; secondly, perform redundancy screening of pipe feature quantities on the first set of sample pipe bending forming data to form a first set of screened pipe bending forming data; then, based on each set of screened pipe bending forming data and the corresponding springback angle label, construct a springback angle prediction model, and based on each set of screened pipe bending forming data and the corresponding springback radius label, construct a springback radius prediction model; finally, using the springback angle prediction model and the springback radius prediction model, determine the target process parameter data where both the springback angle and springback radius meet the springback radius threshold conditions. Based on the above, on the one hand, the redundancy screening of pipe feature quantities can improve the problem of model overfitting to a certain extent, thereby improving the prediction accuracy; on the other hand, the joint determination using the springback angle prediction model and the springback radius prediction model results in higher reliability of the determined target process parameter data. Furthermore, since the model can be used directly every time it is determined, it can effectively reduce computational costs compared to schemes that require finite element simulation calculations for each iteration. Therefore, it can improve the problems of low prediction accuracy or high prediction costs in existing technologies. Attached Figure Description

[0015] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0016] Figure 1 This is a structural block diagram of an electronic device provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart illustrating the pipe bending springback prediction method provided in the embodiments of this application.

[0018] Figure 3 A flowchart illustrating the machine learning strategy for predicting and controlling springback during bending of thin-walled circular tubing, as provided in this application embodiment.

[0019] Figure 4 This is a schematic diagram of the recursive elimination and screening results of the key feature quantity of the rebound angle provided in the embodiments of this application.

[0020] Figure 5 This is a schematic diagram of the recursive elimination screening results of the key feature quantity of rebound radius provided in the embodiments of this application.

[0021] Figure 6 A schematic diagram of the process optimization dataset and Pareto front provided in the embodiments of this application.

[0022] Figure 7 This is a block diagram of the pipe bending and springback prediction device provided in the embodiments of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0025] like Figure 1 As shown in the figure, an embodiment of this application provides an electronic device. The electronic device may include a memory, a processor, and a tube bending springback prediction device.

[0026] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The pipe bending springback prediction device includes at least one software functional module stored in the memory in the form of software or firmware. The processor is used to execute executable computer programs stored in the memory, such as the software functional modules and computer programs included in the pipe bending springback prediction device, to implement the pipe bending springback prediction method provided in the embodiments of this application.

[0027] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0028] Optionally, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0029] Understandable. Figure 1 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices.

[0030] Combination Figure 2 This application also provides a method for predicting the springback of tubing during bending forming, applicable to the aforementioned electronic device. The method steps defined in the process of the method for predicting the springback of tubing during bending forming can be implemented by the electronic device.

[0031] The following will be about Figure 2 The specific process shown will be explained in detail.

[0032] Step S110: Obtain the bending forming data of the first number of sample tubes.

[0033] In this embodiment, the electronic device can acquire a first set of sample tube bending forming data. Each set of sample tube bending forming data includes the value of at least one process parameter and the values ​​of multiple tube characteristic quantities. For example, the first set of sample tube bending forming data includes the value of at least one process parameter and the values ​​of multiple tube characteristic quantities, and the second set of sample tube bending forming data also includes the value of at least one process parameter and the values ​​of multiple tube characteristic quantities. Each set of sample tube bending forming data corresponds to a springback angle label and a springback radius label, both of which are determined by actual measurement. Furthermore, it should be noted that the process parameter can refer to one or more of the following: nominal bending angle, relative bending radius, friction coefficient between the mandrel and the inner surface of the tube, and friction coefficient between the mold and the outer surface of the tube. The tube characteristic quantities can be one or more of the following: room temperature tensile mechanical properties, forming properties, geometric dimensions, and microstructure characteristics. Each type of tube characteristic quantity can also include one or more specific tube characteristic quantities.

[0034] Step S120: Redundancy screening of pipe feature quantities is performed on the first group of sample pipe bending forming data to form the first group of screened pipe bending forming data.

[0035] In this embodiment of the application, after obtaining the first set of sample pipe bending and forming data, the electronic device can perform redundancy screening of the pipe feature quantities in the first set of sample pipe bending and forming data to form a first set of screened pipe bending and forming data. The number of pipe feature quantities included in each set of screened pipe bending and forming data is less than or equal to the number of pipe feature quantities included in each set of sample pipe bending and forming data. That is, if there are redundant pipe feature quantities among the various pipe feature quantities included in the sample pipe bending and forming data, these redundant pipe feature quantities can be removed, retaining only the non-redundant pipe feature quantities.

[0036] Step S130: Based on at least each set of the selected pipe bending forming data and the corresponding springback angle label, construct a springback angle prediction model; and based on at least each set of the selected pipe bending forming data and the corresponding springback radius label, construct a springback radius prediction model.

[0037] In this embodiment, after forming the first number of sets of screened pipe bending and forming data, the electronic device can construct a springback angle prediction model based at least on each set of screened pipe bending and forming data and the corresponding springback angle label, and at least on each set of screened pipe bending and forming data and the corresponding springback radius label. That is, the springback angle prediction model can be constructed solely based on each set of screened pipe bending and forming data and the corresponding springback angle label, or it can be constructed by combining other data with each set of screened pipe bending and forming data and the corresponding springback angle label. Similarly, the springback radius prediction model can be constructed solely based on each set of screened pipe bending and forming data and the corresponding springback radius label, or it can be constructed by combining other data with each set of screened pipe bending and forming data and the corresponding springback radius label.

[0038] Step S140: Using the springback angle prediction model and the springback radius prediction model, a leading edge search is performed under the pipe material characteristics of the target pipe to determine the target process parameter data that satisfy the springback angle threshold condition and the springback radius threshold condition.

[0039] In this embodiment, after obtaining the springback angle prediction model and the springback radius prediction model, the electronic device can use these models to perform a leading edge search under the pipe characteristics of the target pipe, determining the target process parameter data where both the springback angle and springback radius meet the springback angle threshold condition and the springback radius threshold condition. In other words, for the target pipe, which has fixed pipe characteristics, the process parameters are continuously adjusted, and predictions are made using the springback angle prediction model and the springback radius prediction model respectively, until the predicted springback angle and springback radius meet the springback radius threshold condition. Then, the corresponding process parameters are determined as the target process parameter data. It should be noted that the target process parameter data may include a single value of the process parameter, or multiple values ​​of the process parameter, such as a range.

[0040] Based on the above, on the one hand, the redundant screening of pipe characteristic quantities can mitigate the problem of model overfitting to some extent, thereby improving prediction accuracy. On the other hand, the joint determination using springback angle and springback radius prediction models results in more reliable target process parameter data. Furthermore, since the model can be directly used every time it is determined, the computational cost can be effectively reduced compared to schemes that require finite element simulation calculations for each iteration. Therefore, this approach can address the problems of low prediction accuracy or high prediction cost in existing technologies.

[0041] Firstly, regarding step S110, it should be noted that the specific method for obtaining the bending forming data of the first set of sample pipes is not limited and can be selected according to actual needs.

[0042] For example, in an alternative implementation, the first set of sample tube bending forming data can be obtained either from relevant literature or manuals, or through experiments. Furthermore, the first set of data can be determined comprehensively based on the range of tube characteristic variations and the step size, i.e., determining the minimum sample size. It should also be noted that the first set of sample tube bending forming data needs to be obtained under fixed process parameters to ensure that key characteristic quantities are not missed during screening, and that other characteristic quantities have a certain range; otherwise, if the data variation range is too small, key characteristic quantities cannot be screened out. That is, the values ​​of the same process parameter are the same in the first set of sample tube bending forming data. For example, the value of the friction coefficient between the mold and the outer surface of the tube in the first set of sample tube bending forming data is the same as the value of the friction coefficient between the mold and the outer surface of the tube in the second set of sample tube bending forming data.

[0043] Additionally, exemplarily, each set of sample tube bending forming data may include 13 characteristic quantities in 4 categories: room temperature tensile mechanical properties, forming properties, geometric dimensions, and microstructure, denoted by fij (i represents the characteristic quantity category, and j represents the specific characteristic quantity under that category). f11-f15 represent the room temperature tensile mechanical properties of the tube, namely tensile strength (f11), yield strength (f12), elastic modulus (f13), elongation (f14), and Poisson's ratio (f15). f21-f22 represent the various forming properties of the tube, namely thickness anisotropy index (f21) and work hardening index (f22). f31-f32 represent the geometric dimensions of the tube, namely true outer diameter (f31) and true wall thickness (f32). f41-f44 represent the microstructure of the tube, namely grain size grade (f41), average grain size (f42), number of common slip systems (f43), and maximum orientation density of ODF (f44). The initial number of characteristic quantities is n1. However, the extraction of feature quantities here is not limited to the above-mentioned limited performance indicators, and other performance indicators for pipe bending and forming can be added.

[0044] Secondly, regarding step S120, it should be noted that the specific method for redundant screening of pipe feature quantities in the first group of sample pipe bending forming data is not limited and can be selected according to actual needs.

[0045] For example, in an alternative implementation, in order to ensure the sufficiency of the redundancy screening, that is, to effectively remove redundant pipe feature quantities, the above step S120 may further include steps S121 and S122, wherein the specific contents of each step are as follows.

[0046] Step S121: Perform a first redundancy screening of pipe feature quantities on the first quantity group of sample pipe bending forming data to form the first quantity group of intermediate pipe bending forming data.

[0047] In this embodiment, the first set of sample pipe bending forming data can be subjected to a first redundancy screening of pipe feature quantities to form intermediate pipe bending forming data in the first set. The first redundancy screening refers to filtering based on the correlation between pipe feature quantities. For example, if the correlation between two pipe feature quantities is very high, one of them can be retained.

[0048] Step S122: Perform a second redundancy screening of pipe feature quantities on the intermediate pipe bending and forming data of the first quantity group to form the first quantity group of screened pipe bending and forming data.

[0049] In this embodiment of the application, after obtaining the first set of intermediate pipe bending and forming data, a second redundancy screening of pipe feature quantities can be performed on the first set of intermediate pipe bending and forming data to form a first set of screened pipe bending and forming data. The second redundancy screening refers to screening based on the correlation importance between pipe feature quantities and labels. This allows for the removal of pipe feature quantities that have a relatively small impact on the results (labels) to a certain extent.

[0050] It is understood that in step S121 above, the specific method of performing the first redundancy screening of pipe feature quantities on the first number of sample pipe bending forming data is not limited. For example, in an alternative embodiment, in order to ensure that effective screening can be performed based on the correlation between pipe feature quantities, step S121 above may further include steps S121a and S121b, wherein the specific contents of each step are as follows.

[0051] Step S121a: For each of the two pipe material characteristics, based on the values ​​of the two pipe material characteristics in the bending forming data of the first group of sample pipe materials, determine the correlation coefficient between the two pipe material characteristics.

[0052] In this embodiment, for each pair of pipe characteristics, a correlation coefficient can be determined based on the values ​​of the two pipe characteristics in the bending forming data of the first set of sample pipes. For example, the Pearson correlation coefficient r can be calculated using the following formula: ; Among them, a i and b i These are the values ​​of the two pipe material characteristics in the i-th set of data, a m and b m It is the average value of the two pipe material characteristics a and b across all data.

[0053] Step S121b: Based on the correlation coefficient between each pair of pipe material features, perform a first redundancy screening of the pipe material features corresponding to the first set of sample pipe bending and forming data to form the first set of intermediate pipe bending and forming data.

[0054] In this embodiment, after obtaining the correlation coefficient between two pipe material characteristics, a first redundancy screening of pipe material characteristics can be performed on the multiple pipe material characteristics corresponding to the first set of sample pipe bending forming data based on the correlation coefficient between each pair of pipe material characteristics, forming the first set of intermediate pipe bending forming data. Wherein, the larger the absolute value of r, the stronger the characteristic a... i and b iThe stronger the linear correlation between them, the more likely it is that one of the pipe features with a smaller impact on output performance can be removed by comparison, leaving n2 pipe features remaining.

[0055] It is understood that in step S121b above, the specific method of forming the first set of intermediate tube bending forming data is not limited. For example, in an alternative embodiment, in order to effectively remove a tube feature quantity that has a small impact on output performance, the specific contents of each of the above steps b1, b2, b3, b4 and b5 are as follows.

[0056] Step b1: For each pair of pipe material characteristics, determine whether the absolute value of the correlation coefficient between the two pipe material characteristics is greater than a predetermined correlation coefficient threshold.

[0057] In this embodiment of the application, for each pair of pipe material characteristics, the absolute value of the correlation coefficient between the two pipe material characteristics can be determined to be greater than a predetermined correlation coefficient threshold. For example, when the absolute value of the correlation coefficient is greater than the correlation coefficient threshold, it indicates that the correlation between the corresponding two pipe material characteristics is high; when the absolute value of the correlation coefficient is not greater than the correlation coefficient threshold, it indicates that the correlation between the corresponding two pipe material characteristics is low. For example, the correlation coefficient threshold can be 0.9.

[0058] Step b2: For each pair of pipe material features whose absolute value of the correlation coefficient is greater than the correlation coefficient threshold, a first machine learning model and a second machine learning model are established with the first pipe material feature as input and the springback angle and springback radius as outputs. A third machine learning model and a fourth machine learning model are established with the second pipe material feature as input and the springback angle and springback radius as outputs.

[0059] In this embodiment, for each pair of pipe material features whose absolute value of the correlation coefficient is greater than the correlation coefficient threshold (i.e., two pipe material features with high correlation), a first machine learning model and a second machine learning model are established using the first pipe material feature as input and the springback angle and springback radius as outputs, respectively. A third machine learning model and a fourth machine learning model are established using the second pipe material feature as input and the springback angle and springback radius as outputs, respectively. It should be noted that the specific types of the first, second, third, and fourth machine learning models are not limited and can be selected according to requirements. For example, the first and third machine learning models can be of the same type, both used to predict the springback angle. The second and fourth machine learning models can also be of the same type, both used to predict the springback radius.

[0060] Step b3: Using the value of the first type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, train the first machine learning model and determine the corresponding first training error; and using the value of the first type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, train the second machine learning model and determine the corresponding second training error.

[0061] In this embodiment of the application, after determining the first machine learning model and the second machine learning model, the first machine learning model can be trained using the value of the first type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, and the corresponding first training error (i.e., the difference between the predicted springback angle and the springback angle label) can be determined. The second machine learning model can be trained using the value of the first type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, and the corresponding second training error (i.e., the difference between the predicted springback radius and the springback radius label) can be determined.

[0062] Step b4: Using the values ​​of the second type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, train the third machine learning model and determine the corresponding third training error. Using the values ​​of the second type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, train the fourth machine learning model and determine the corresponding fourth training error.

[0063] In this embodiment of the application, after determining the third machine learning model and the fourth machine learning model, the third machine learning model can be trained using the value of the second type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, and the corresponding third training error (i.e., the difference between the predicted springback angle and the springback angle label) can be determined. The fourth machine learning model can be trained using the value of the second type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, and the corresponding fourth training error (i.e., the difference between the predicted springback radius and the springback radius label) can be determined.

[0064] Step b5: Based on the magnitude relationship between the first training error and the third training error, the corresponding two types of pipe feature quantities are screened out to form a first set of intermediate pipe bending forming data corresponding to the springback angle label; and based on the magnitude relationship between the second training error and the fourth training error, the corresponding two types of pipe feature quantities are screened out to form a first set of intermediate pipe bending forming data corresponding to the springback radius label.

[0065] In this embodiment, after obtaining the first training error, second training error, third training error, and fourth training error, two corresponding pipe feature quantities can be filtered out based on the magnitude relationship between the first training error and the third training error to form a first set of intermediate pipe bending forming data corresponding to the springback angle label. Similarly, two corresponding pipe feature quantities can be filtered out based on the magnitude relationship between the second training error and the fourth training error to form a first set of intermediate pipe bending forming data corresponding to the springback radius label. For example, pipe feature quantities with larger training errors can be filtered out. For instance, if the first training error is greater than the third training error, for the springback angle, the first type of pipe feature quantity can be filtered out to form a first set of intermediate pipe bending forming data corresponding to the springback angle label. If the second training error is less than the fourth training error, for the springback radius, the second type of pipe feature quantity can be filtered out to form a first set of intermediate pipe bending forming data corresponding to the springback radius label.

[0066] It is understood that in step S122 above, the specific method of performing the second redundancy screening of pipe feature quantities on the intermediate pipe bending forming data of the first quantity group is not limited. For example, in an alternative embodiment, in order to ensure the reliability of the second redundancy screening, step S122 above may further include steps S122a, S122b and S122c, wherein the specific contents of each step are as follows.

[0067] Step S122a: For each type of pipe feature included in the first set of intermediate pipe bending forming data, a fifth machine learning model and a sixth machine learning model are established, with each type of pipe feature other than the specified pipe feature as input and the springback angle and springback radius as outputs. The fifth machine learning model is trained using the values ​​of each type of pipe feature input from the first set of intermediate pipe bending forming data and the corresponding springback angle label, and the corresponding fifth training error is determined. Similarly, the sixth machine learning model is trained using the values ​​of each type of pipe feature input from the first set of intermediate pipe bending forming data and the corresponding springback radius label, and the corresponding sixth training error is determined.

[0068] In this embodiment, for each pipe feature included in the first set of intermediate pipe bending forming data, each pipe feature other than the specified pipe feature is used as input, and the springback angle and springback radius are used as outputs. A fifth machine learning model (e.g., for springback angle prediction) and a sixth machine learning model (e.g., for springback radius prediction) are established. The fifth machine learning model is trained using the value of each pipe feature input from the first set of intermediate pipe bending forming data and the corresponding springback angle label, and the corresponding fifth training error (i.e., the difference between the predicted springback angle and the springback angle label) is determined. The sixth machine learning model is trained using the value of each pipe feature input from the first set of intermediate pipe bending forming data and the corresponding springback radius label, and the corresponding sixth training error (i.e., the difference between the predicted springback radius and the springback radius label) is determined.

[0069] Step S122b: In each type of pipe feature included in the first set of intermediate pipe bending forming data, the pipe feature with the minimum value of the corresponding fifth training error is screened out to form the first set of screened pipe bending forming data corresponding to the springback angle label.

[0070] In this embodiment, after obtaining the fifth training error, the pipe feature with the minimum value of the corresponding fifth training error can be filtered out from each type of pipe feature included in the first set of intermediate pipe bending forming data, forming a first set of filtered pipe bending forming data corresponding to the springback angle label. That is, since the corresponding pipe feature does not participate in the model's processing, the minimum error indicates that the corresponding pipe feature has a small influence on the result, and therefore, it can be filtered out.

[0071] Step S122c: In each type of pipe feature included in the first set of intermediate pipe bending forming data, the pipe feature with the minimum value of the corresponding sixth training error is screened out to form the first set of screened pipe bending forming data corresponding to the springback radius label.

[0072] In this embodiment of the application, after obtaining the sixth training error, the pipe feature quantity with the minimum value of the corresponding sixth training error can be screened out from each type of pipe feature quantity included in the first set of intermediate pipe bending forming data, so as to form the first set of screened pipe bending forming data corresponding to the springback radius label.

[0073] It should be noted that steps S122a, S122b, and S122c above can be performed iteratively to determine the effective removal of redundant pipe feature quantities. For example, recursive elimination can be performed, such as sequentially taking out one pipe feature quantity, inputting the remaining n²-1 pipe feature quantities, and outputting the springback change value (including two output quantities, namely the springback angle). rebound radius Where θ0 is the nominal bending angle, θ1 is the actual bending angle, R0 is the nominal bending radius, and R1 is the actual bending radius, an input-output machine learning model is established, and n² model errors can be obtained. The error calculation formula can be: ; ; Here, "pre" represents the predicted result, and "exp" represents the corresponding label. The pipe feature quantity corresponding to the minimum error in the model error is removed, leaving n²-1 pipe feature quantities. This process is repeated in the next round of recursive elimination until the minimum model error increases instead of decreasing, at which point the recursive elimination stops. Thus, the number of remaining pipe feature quantities for the springback angle is n³, and the number of remaining pipe feature quantities for the springback radius is n⁴.

[0074] Thirdly, regarding step S130, it should be noted that the specific methods for constructing the rebound angle prediction model and the rebound radius prediction model are not restricted and can be selected according to actual needs.

[0075] For example, in an alternative implementation, in order to ensure that the constructed rebound angle prediction model and rebound radius prediction model have high prediction accuracy, the above step S130 may further include steps S131 and S132, wherein the specific contents of each step are as follows.

[0076] Step S131: Based on the process parameter expansion method, the first group of screened pipe bending forming data corresponding to the springback angle label is expanded to form a second group of screened pipe bending forming data corresponding to the springback angle label; and based on the process parameter expansion method, the first group of screened pipe bending forming data corresponding to the springback radius label is expanded to form a second group of screened pipe bending forming data corresponding to the springback radius label.

[0077] In this embodiment, the first set of screened pipe bending forming data corresponding to the springback angle label can be expanded based on process parameter expansion to form a second set of screened pipe bending forming data corresponding to the springback angle label. Similarly, the first set of screened pipe bending forming data corresponding to the springback radius label can be expanded based on process parameter expansion to form a second set of screened pipe bending forming data corresponding to the springback radius label. In the first set of sample pipe bending forming data, the value of the same process parameter is the same. This allows for the supplementation of data sets with different process parameters (i.e., the values ​​of the same process parameter differ in the supplemented data sets), which can be determined through literature or orthogonal experimental methods.

[0078] Step S132: Use the second set of data corresponding to the springback angle label to filter the pipe bending forming data and train a springback angle prediction model; and use the second set of data corresponding to the springback radius label to filter the pipe bending forming data and train a springback radius prediction model.

[0079] In this embodiment of the application, after obtaining the second set of screened pipe bending forming data, a springback angle prediction model can be trained using the second set of screened pipe bending forming data corresponding to the springback angle label, and a springback radius prediction model can be trained using the second set of screened pipe bending forming data corresponding to the springback radius label, i.e., training is performed separately.

[0080] It is understood that in step S132 above, the specific method of training the rebound angle prediction model and the rebound radius prediction model is not limited. For example, in an alternative implementation, in order to ensure that the training of the rebound angle prediction model and the rebound radius prediction model has high prediction accuracy, step S132 above may further include steps S132a and S132b, wherein the specific contents of each step are as follows.

[0081] Step S132a: Using process parameters and pipe characteristics as inputs and springback angle as output, construct multiple initial springback angle models. Then, use the second group of numbers corresponding to the springback angle label to filter the pipe bending forming data, train each initial springback angle model to form multiple trained springback angle models, and determine the trained springback angle model with the smallest error as the springback angle prediction model.

[0082] In this embodiment, multiple initial springback angle models (such as ANN, SVR, LIN, KNN, etc.) can be constructed using process parameters and pipe characteristics as inputs and springback angle as output. Furthermore, pipe bending forming data can be filtered using a second set of data corresponding to the springback angle labels. Each initial springback angle model is then trained to form multiple trained springback angle models. The trained springback angle model with the smallest error is then determined as the springback angle prediction model. Step S132b involves constructing multiple initial springback radius models using process parameters and pipe characteristics as inputs and springback radius as output. Then, pipe bending forming data is filtered using a second set of data corresponding to the springback radius label. Each initial springback radius model is trained to form multiple trained springback radius models. The trained springback radius model with the smallest error is determined as the springback radius prediction model.

[0083] In this embodiment, multiple initial springback radius models (such as ANN, SVR, LIN, KNN, etc.) can be constructed using process parameters and pipe characteristics as inputs and springback radius as output. Furthermore, pipe bending forming data can be filtered using a second set of numbers corresponding to the springback radius labels, and each of the initial springback radius models can be trained to form multiple trained springback radius models. The trained springback radius model with the smallest error is then determined as the springback radius prediction model.

[0084] It should be noted that during training, one-quarter of the data sample (i.e., the second set of pipe bending forming data) can be randomly selected as the test set, and the remaining three-quarters as the training set. Commonly used 50-fold, 80-fold, or 100-fold cross-validation methods can be applied to the training set to optimize the model parameters. Furthermore, to reduce the fluctuations in model predictions caused by differences in the dataset, the training and test sets can be randomly divided repeatedly, t times. The average of the validation errors from these t iterations is taken as the model error to evaluate the model's prediction accuracy. The validation error can be calculated using the following formula. ; Where, p pre p represents the predicted output metric. exp The output index (i.e., label) of the representative experiment is calculated according to the above formula, and the model with the smallest verification error can be selected as the preferred model.

[0085] Fourthly, regarding step S140, it should be noted that the specific method for determining the target process parameter data that satisfies both the springback angle and springback radius threshold conditions is not limited and can be selected according to actual needs.

[0086] For example, in an alternative implementation, since the rebound angle prediction model and the rebound radius prediction model have two output indicators... and This requires simultaneously predicting and controlling both the bending springback angle and the bending springback radius. Therefore, a Pareto front based on a non-dominated sorting genetic algorithm can be used for optimization. The selection of the Pareto front requires a fixed input of the pipe's material characteristics. Then, the front is searched by adjusting process parameters. These adjustments must conform to the adjustable step size of the process. For example, if the bending angle step size for CNC pipe bending is 0.1°, then the bending angle step size cannot be less than 0.1° when performing Pareto front optimization. The initial population size and maximum number of generations can be defined by the user, satisfying the following threshold calculation formula. ; ; In the formula, This refers to the maximum springback angle threshold required for pipe bending technology. It is the maximum springback radius threshold required by bending technology.

[0087] Based on this, a machine learning model is established, inputting the pipe characteristics of the material, performing Pareto front search, and then, according to the process parameters satisfying the above threshold calculation formula, calculating the error between the predicted and actual bending springback index, thus obtaining an experimental verification data set. If the experimental verification error exceeds the set threshold... and The experimental validation set data can be used to re-guide machine learning so that the error meets the requirements. If the error threshold is met, the prediction model is then solidified.

[0088] To facilitate understanding of the above-mentioned method for predicting springback during pipe bending, this application also provides a specific application example, the details of which are as follows: Combination Figure 3 Taking the CNC bending forming of 5A02 aluminum alloy tubes as an example, this paper conducts machine learning for prediction and control of bending springback based on small sample data, specifically including the following steps: S1: Screening of key features for pipe materials. This mainly includes: data collection → feature extraction → machine learning feature screening to determine the minimum core key features.

[0089] Data were collected using an experimental method. The nominal geometric dimensions of the pipes included Φ6×1mm, Φ10×1mm, Φ16×1mm, and Φ38×1mm. Fixed process parameters were: nominal bending angle 90°, relative bending radius 2, friction coefficient between the mandrel and the inner surface of the pipe 0.2, and friction coefficient between the mold and the outer surface of the pipe 0.2.

[0090] Feature quantities were extracted from the above-mentioned pipes, including 13 features in 4 categories: room temperature tensile mechanical properties, formability, geometric dimensions, and microstructure. These features were represented by f. ij (where i represents the feature category and j represents the specific feature within that category) is used to represent it. 11 -f 15 The room temperature tensile mechanical properties of the pipe are represented by tensile strength (f) 11 ), yield strength (f 12 ), elastic modulus (f 13 ), elongation (f) 14 Poisson's ratio (f) 15 ), f 21 -f 22 The various forming properties of the pipe are represented by the thickness anisotropy index (f). 21 ), work hardening index (f 22 ), f 31 -f 32 The representative pipe geometric dimensions are the actual outer diameter (f) 31 True wall thickness (f) 32 ), f 41 -f 44 The microstructure characteristics of the pipe are represented by the grain size grade (f). 41 ), average grain size (f 42 ), number of common slip systems (f 43 ), ODF maximum orientation density (f 44 The initial number of features is n1.

[0091] Specifically, by heat-treating the pipe to change the room temperature tensile strength from 165 MPa to 225 MPa in 15 MPa increments, other indicators will also change accordingly. A total of n1 = 13 characteristic quantities are involved. The springback angle after bending is obtained through CNC bending. rebound radius A total of 78 datasets were collected. Machine learning was used to screen key features, starting with correlation screening to quickly identify redundant factors. Specifically, the Pearson correlation coefficient r was calculated using the following formula: (1) In equation (1), a i and b i Let a be the value of two certain features in the i-th group of data (i=78). m and b m It is the average of features a and b across all data.

[0092] The larger the absolute value of r, the stronger the characteristic quantity ai and b i The stronger the linear correlation between them, the better. When setting it up, use one of the features as input, ... and For each output, a machine learning model is built, and the feature with the larger error in the two models is removed. The remaining feature of the pipe corresponding to the springback angle is: f. 12 f 13 f 14 f 21 f 22 f 31 f 32 f 42 f 44 Number of features The rebound radius corresponds to the remaining characteristic quantity of the pipe as: f 12 f 13 f 14 f 15 f 21 f 22 f 31 f 32 f 42 f 43 Number of features 10.

[0093] Then, recursive elimination is performed, taking out one feature value in sequence, inputting the remaining n²-1 feature values, and outputting the rebound change value (including two output values, namely the rebound angle). rebound radius Where θ0 is the nominal bending angle, θ1 is the actual bending angle, R0 is the nominal bending radius, and R1 is the actual bending radius, an input-output machine learning model is established, and n² model errors can be obtained. The error calculation formula is as follows: (2) (3) Remove the feature corresponding to the minimum error in the model error, leaving n²-1 features. Then perform another round of recursive elimination until the minimum model error increases instead of decreasing, at which point the recursive elimination stops. Figure 4 and Figure 5 After performing recursive elimination in sequence, based on the model error results for the rebound angle and rebound radius, from... Figure 4 and Figure 5 The remaining key feature quantities can be determined to be n3=6 and n4=7, respectively, corresponding to f 12 f 14 f 21 f 31 f 32 f 42 and f12 f 14 f 15 f 21 f 22 f 31 f 32 f 43 .

[0094] S2: Construction of a machine learning prediction model for key features and process parameters. Based on the 78 samples in S1, the model was expanded through experiments, adding data sets under different process parameters, increasing the number of data sets to 108. The process parameters included nominal bending angle, relative bending radius, friction coefficient between the mandrel and the inner surface of the tube, and friction coefficient between the mold and the outer surface of the tube. The bending angle ranged from 30° to 150° with a step size of at least 20°. The relative bending radius included values ​​of 2, 2.5, and 3, and the friction coefficient ranged from 0.1 to 0.3 with a step size of at least 0.1.

[0095] A machine learning model is constructed by overlaying process parameters on key features of the pipe material. The key features of the material and the process parameters are used as inputs, and the springback angle is output. and rebound radius Machine learning models were established separately. Here, ANN, SVR, LIN, and KNN were used to construct models respectively. Based on the principle of minimizing prediction error, SVR was selected as the rebound angle prediction model, and KNN was selected as the rebound radius prediction model.

[0096] To further optimize the model, one-quarter of the data sample is randomly selected as the test set, and the remaining three-quarters are used as the training set. The training set is then optimized using the common 10-fold cross-validation method. To reduce the fluctuation in model predictions caused by differences in the dataset, the training and test sets are randomly divided repeatedly in S2, with 108 repetitions. The average of the 108 validation errors is taken as the model error to evaluate the model's prediction accuracy. The validation error is calculated using the following formula. (4) In the above formula, p pre p represents the predicted output metric. exp The output index of the experiment is calculated according to the above formula, and the model with the smallest verification error is the preferred model.

[0097] S3: Multi-objective optimized bending springback control strategy.

[0098] S3 has two output indicators and To simultaneously predict and control the bending springback angle and bending springback radius, a Pareto front based on a non-dominated sorting genetic algorithm can be used for optimization. The key features of the material are fixed as input, and the Pareto front search is performed by adjusting process parameters. The process parameter adjustments must conform to the adjustable step size of the process. For example, if the nominal bending angle of a CNC pipe bend is 60°, and the step size is set to 0.1°, the search is performed within the range of 60-65°. The bending radius is a fixed value of 2.5, and the friction coefficients are 0.1 and 0.3, respectively. The initial population size is 200, and the maximum number of generations is 200, satisfying the following threshold calculation formula. (5) (6) In the formula, The maximum springback angle threshold required for pipe bending technology is 0.5°. The maximum springback radius threshold for bending technology is 0.8. Figure 6 The Pareto front surface under the above conditions is given. Process parameters that meet the requirements of springback angle and springback radius can be selected from the figure, namely the actual bending angle and friction coefficient value, for example, the actual bending angle.

[0099] S4: Experimental verification.

[0100] A pipe with a nominal size of Φ20×1mm was selected, and key characteristic quantities of the pipe were obtained through performance testing. Under the premise of a nominal bending angle of 80° and a relative bending radius of 3, the predicted springback angle was obtained. , True rebound angle , The prediction errors are respectively and The condition is less than the set threshold. and Requirements.

[0101] Input the Pareto front surface from the model in S2 and calculate it based on S3 to obtain recommended process parameters. The actual bending angle should be 83.2°, and the friction coefficient should be controlled at 0.15. Conduct a CNC bending experiment and measure the springback angle. , It meets the requirements of a maximum rebound angle threshold of 0.5° and a maximum rebound radius threshold of 0.8°.

[0102] Compared with existing methods, the above-mentioned method yields a machine learning model with high prediction accuracy and strong generalization ability. It also introduces the screening of key features of pipe materials to effectively remove redundant features and avoid overfitting of the model. In addition, compared with the previous method that only considered the prediction of bending springback angle, it establishes the prediction of bending springback radius. Combining these two indicators, a multi-objective optimization bending springback control strategy is given, which can effectively and quickly optimize the range of bending process control parameters based on the performance values ​​of pipe materials.

[0103] Furthermore, it should be noted that, according to the inventors' research, some conventional machine learning methods, if only limited process parameters are used as input, such as key bending process parameters, are not comprehensive enough. If material properties are considered, but known material properties are used as input, the prediction model has a relatively large error when the data sample is limited. Other performance indicators of materials affecting bending springback are not considered, and there is a lack of correlation redundancy screening, which easily leads to model overfitting. Moreover, the solution provided by the embodiments of this application can also improve the problem of relatively low prediction reliability caused by conventional solutions that only consider the bending angle as the output and do not consider the change in springback control of the bending radius.

[0104] Combination Figure 7 This application also provides a tube bending springback prediction device applicable to the aforementioned electronic equipment. The tube bending springback prediction device may include a data acquisition module, a feature quantity screening module, a prediction model construction module, and a parameter determination module.

[0105] The data acquisition module is used to acquire a first number of sets of sample pipe bending forming data, wherein each set of sample pipe bending forming data includes the value of at least one process parameter and the values ​​of multiple pipe characteristic quantities, and each set of sample pipe bending forming data corresponds to a springback angle label and a springback radius label. In this embodiment of the application, the data acquisition module can be used to perform... Figure 2 The relevant content regarding the data acquisition module in step S110 shown can be found in the preceding description of step S110.

[0106] The feature quantity filtering module is used to perform redundancy filtering of pipe feature quantities on the first set of sample pipe bending forming data to form a first set of filtered pipe bending forming data. The number of pipe feature quantities included in each set of filtered pipe bending forming data is less than or equal to the number of pipe feature quantities included in each set of sample pipe bending forming data. In this embodiment, the feature quantity filtering module can be used to perform... Figure 2 The relevant content regarding the feature quantity screening module in step S120 shown can be found in the previous description of step S120.

[0107] The prediction model construction module is used to construct a springback angle prediction model based at least on each set of the selected pipe bending forming data and the corresponding springback angle label, and to construct a springback radius prediction model based at least on each set of the selected pipe bending forming data and the corresponding springback radius label. In this embodiment, the prediction model construction module can be used to execute... Figure 2 The relevant content regarding the prediction model construction module in step S130 shown can be found in the previous description of step S130.

[0108] The parameter determination module is used to perform a leading edge search under the pipe characteristics of the target pipe using the springback angle prediction model and the springback radius prediction model, to determine the target process parameter data where the springback angle and springback radius both meet the springback angle threshold condition. In this embodiment, the parameter determination module can be used to execute... Figure 2 The relevant content regarding the parameter determination module in step S140 shown can be found in the preceding description of step S140.

[0109] In this embodiment of the application, corresponding to the above-described method for predicting the springback of tubing during bending forming applied to the electronic device, a computer-readable storage medium is also provided. This computer-readable storage medium stores a computer program that, when executed, performs each step of the method for predicting the springback of tubing during bending forming. The steps executed by the aforementioned computer program during operation will not be described in detail here, but can be found in the preceding explanation of the method for predicting the springback of tubing during bending forming.

[0110] In summary, the pipe bending forming springback prediction method, apparatus, equipment, and medium provided in this application firstly acquire a first set of sample pipe bending forming data; secondly, perform redundant screening of pipe feature quantities on the first set of sample pipe bending forming data to form a first set of screened pipe bending forming data; then, based on each set of screened pipe bending forming data and the corresponding springback angle label, construct a springback angle prediction model, and based on each set of screened pipe bending forming data and the corresponding springback radius label, construct a springback radius prediction model; finally, using the springback angle prediction model and the springback radius prediction model, determine the target process parameter data where both the springback angle and springback radius meet the springback radius threshold conditions. Based on the above, on the one hand, the redundant screening of pipe feature quantities can improve the problem of model overfitting to a certain extent, thereby improving the prediction accuracy; on the other hand, the joint determination using the springback angle prediction model and the springback radius prediction model results in higher reliability of the determined target process parameter data. Furthermore, since the model can be used directly every time it is determined, it can effectively reduce computational costs compared to schemes that require finite element simulation calculations for each iteration. Therefore, it can improve the problems of low prediction accuracy or high prediction costs in existing technologies.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0112] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0113] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, electronic device, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0114] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting the springback of pipes during bending and forming, characterized in that, include: Obtain a first set of sample pipe bending forming data, wherein each set of sample pipe bending forming data includes the value of at least one process parameter and the values ​​of multiple pipe characteristic quantities, and each set of sample pipe bending forming data corresponds to a springback angle label and a springback radius label. Redundancy screening of pipe feature quantities is performed on the first set of sample pipe bending forming data to form the first set of screened pipe bending forming data, wherein the number of pipe feature quantities included in each set of screened pipe bending forming data is less than or equal to the number of pipe feature quantities included in each set of sample pipe bending forming data. At least based on each set of the selected pipe bending forming data and the corresponding springback angle label, a springback angle prediction model is constructed, and at least based on each set of the selected pipe bending forming data and the corresponding springback radius label, a springback radius prediction model is constructed. Using the springback angle prediction model and the springback radius prediction model, a leading edge search is performed under the pipe material characteristics of the target pipe to determine the target process parameter data that satisfy both the springback angle threshold condition and the springback radius threshold condition.

2. The method for predicting springback during pipe bending as described in claim 1, characterized in that, The step of performing redundancy screening of pipe feature quantities on the first set of sample pipe bending forming data to form the first set of screened pipe bending forming data includes: The first redundancy screening of pipe bending forming data in the first group of sample pipes is performed to form intermediate pipe bending forming data in the first group. The first redundancy screening refers to the screening based on the correlation between pipe features. The pipe bending and forming data in the first quantity group are subjected to a second redundancy screening of pipe feature quantities to form a first quantity group of screened pipe bending and forming data. The second redundancy screening refers to the screening based on the correlation importance between pipe feature quantities and labels.

3. The method for predicting springback during pipe bending as described in claim 2, characterized in that, The step of performing a first redundancy screening of pipe feature quantities on the first group of sample pipe bending forming data to form the first group of intermediate pipe bending forming data includes: For each pair of pipe characteristics, the correlation coefficient between the two pipe characteristics is determined based on the values ​​of the two pipe characteristics in the bending forming data of the first group of sample pipes. Based on the correlation coefficient between each pair of pipe characteristics, the first redundancy screening of pipe characteristics corresponding to the first set of sample pipe bending and forming data is performed to form the first set of intermediate pipe bending and forming data.

4. The method for predicting springback during pipe bending as described in claim 3, characterized in that, The step of performing a first redundancy screening of pipe feature values ​​corresponding to the first set of sample pipe bending forming data based on the correlation coefficient between each pair of pipe feature values ​​to form the first set of intermediate pipe bending forming data includes: For each pair of pipe material characteristics, determine whether the absolute value of the correlation coefficient between the two pipe material characteristics is greater than a predetermined correlation coefficient threshold. For each pair of pipe material features whose absolute value of the correlation coefficient is greater than the correlation coefficient threshold, a first machine learning model and a second machine learning model are established with the first pipe material feature as the input and the springback angle and springback radius as the outputs. A third machine learning model and a fourth machine learning model are established with the second pipe material feature as the input and the springback angle and springback radius as the outputs. Using the values ​​of the first type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, the first machine learning model is trained and the corresponding first training error is determined. Using the values ​​of the first type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, the second machine learning model is trained and the corresponding second training error is determined. Using the values ​​of the second type of pipe feature quantity and the corresponding springback angle label in the first set of sample pipe bending forming data, the third machine learning model is trained and the corresponding third training error is determined. Using the values ​​of the second type of pipe feature quantity and the corresponding springback radius label in the first set of sample pipe bending forming data, the fourth machine learning model is trained and the corresponding fourth training error is determined. Based on the magnitude relationship between the first training error and the third training error, the corresponding two types of pipe feature quantities are screened out to form a first set of intermediate pipe bending forming data corresponding to the springback angle label. Based on the magnitude relationship between the second training error and the fourth training error, the corresponding two types of pipe feature quantities are screened out to form a first set of intermediate pipe bending forming data corresponding to the springback radius label.

5. The method for predicting springback during pipe bending as described in claim 2, characterized in that, The step of performing a second redundancy screening of pipe feature quantities on the intermediate pipe bending and forming data of the first quantity group to form the first quantity group of screened pipe bending and forming data includes: For each pipe feature included in the first set of intermediate pipe bending forming data, a fifth machine learning model and a sixth machine learning model are established, with each pipe feature other than the first set of intermediate pipe bending forming data as input and springback angle and springback radius as output. The fifth machine learning model is trained using the value of each pipe feature input in the first set of intermediate pipe bending forming data and the corresponding springback angle label, and the corresponding fifth training error is determined. The sixth machine learning model is trained using the value of each pipe feature input in the first set of intermediate pipe bending forming data and the corresponding springback radius label, and the corresponding sixth training error is determined. In each type of pipe feature included in the first set of intermediate pipe bending forming data, the pipe feature with the minimum value of the corresponding fifth training error is screened out to form the first set of screened pipe bending forming data corresponding to the springback angle label. In each type of pipe feature included in the first set of intermediate pipe bending forming data, the pipe feature with the minimum value of the corresponding sixth training error is screened out to form the first set of screened pipe bending forming data corresponding to the springback radius label.

6. The method for predicting springback in pipe bending forming according to any one of claims 1-5, characterized in that, The steps of constructing a springback angle prediction model based at least on each set of the selected pipe bending forming data and the corresponding springback angle label, and constructing a springback radius prediction model based at least on each set of the selected pipe bending forming data and the corresponding springback radius label, include: Based on the process parameter expansion method, the first group of screened pipe bending forming data corresponding to the springback angle label is expanded to form a second group of screened pipe bending forming data corresponding to the springback angle label. Also, based on the process parameter expansion method, the first group of screened pipe bending forming data corresponding to the springback radius label is expanded to form a second group of screened pipe bending forming data corresponding to the springback radius label. In this case, the value of the same process parameter is the same in the first group of sample pipe bending forming data. A springback angle prediction model is trained by filtering pipe bending forming data using a second set of data corresponding to the springback angle label, and a springback radius prediction model is trained by filtering pipe bending forming data using a second set of data corresponding to the springback radius label.

7. The method for predicting springback during pipe bending as described in claim 6, characterized in that, The steps of using a second set of data corresponding to the springback angle label to filter pipe bending forming data and training a springback angle prediction model, and using a second set of data corresponding to the springback radius label to filter pipe bending forming data and training a springback radius prediction model, include: Using process parameters and pipe characteristics as inputs and springback angle as output, multiple initial springback angle models are constructed. Furthermore, pipe bending forming data are screened using a second set of numbers corresponding to springback angle labels. Each of the initial springback angle models is trained to form multiple trained springback angle models. The trained springback angle model with the smallest error is determined as the springback angle prediction model. Using process parameters and pipe characteristics as inputs and springback radius as output, multiple initial springback radius models are constructed. Furthermore, pipe bending forming data are screened using a second set of data corresponding to the springback radius label. Each of the initial springback radius models is trained to form multiple trained springback radius models. The trained springback radius model with the smallest error is determined as the springback radius prediction model.

8. A device for predicting the springback of pipes during bending and forming, characterized in that, include: The data acquisition module is used to acquire a first number of sets of sample pipe bending forming data, wherein each set of sample pipe bending forming data includes the value of at least one process parameter and the value of multiple pipe characteristic quantities, and each set of sample pipe bending forming data corresponds to a springback angle label and a springback radius label. The feature quantity filtering module is used to perform redundant filtering of pipe feature quantities on the first number of sample pipe bending and forming data to form the first number of filtered pipe bending and forming data, wherein the number of pipe feature quantities included in each group of filtered pipe bending and forming data is less than or equal to the number of pipe feature quantities included in each group of sample pipe bending and forming data. The prediction model building module is used to build a springback angle prediction model based on at least each set of the selected pipe bending forming data and the corresponding springback angle label, and to build a springback radius prediction model based on at least each set of the selected pipe bending forming data and the corresponding springback radius label. The parameter determination module is used to perform a frontal surface search under the pipe characteristics of the target pipe using the springback angle prediction model and the springback radius prediction model, and to determine the target process parameter data that satisfy the springback angle threshold condition and the springback radius threshold condition.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor connected to the memory is used to execute the computer program stored in the memory to implement the pipe bending springback prediction method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium contains a computer program that, when executed, performs the pipe bending springback prediction method according to any one of claims 1-7.

Images