Method and system for acquiring distribution characteristics of plate shape residual stress of metal plate

Abstract

The present application relates to a kind of metal plate plate shape residual stress distribution feature acquisition method and system, comprising: obtaining the plate shape data of metal plate sub-region and the distribution grade of its corresponding sub-region residual stress value;Residual stress distribution grade prediction network model based on convolutional neural network is constructed and it is trained;Collect the plate shape data of production site metal plate, divide sub-region plate shape data and input into the distribution grade prediction network model after training, output each sub-region residual stress distribution grade of metal plate and draw metal plate plate shape residual stress distribution feature map.The present application finds the relationship between metal plate plate shape feature and residual stress distribution by means of convolutional neural network model, can efficiently obtain plate residual stress distribution feature;Modeling is simple and easy to realize and use cost is low, meet the conditions of real-time acquisition used in industrial field.

Description

Technical Field

[0001] This invention relates to the field of metal processing and deep learning prediction of residual stress distribution, and in particular to a method and system for obtaining residual stress distribution characteristics of metal plates. Background Technology

[0002] Defects in metal sheets are caused by residual stress or uneven distribution of residual stress within the sheet. The main purpose of sheet straightening is to reduce and eliminate residual stress, thereby improving the sheet shape and flatness. Therefore, obtaining the distribution characteristics of residual stress in the sheet is particularly important for processes such as sheet straightening. Currently, methods for measuring residual stress in sheet metal can be divided into destructive and non-destructive methods based on whether they damage the sheet. Commonly used destructive methods include the pinhole method, strip cutting method, and layering method. Destructive methods damage the sheet and contradict the original purpose of straightening. Non-destructive methods refer to measuring residual stress in sheet metal without damaging its internal structure using existing modern physical technologies. Common methods include X-ray method, ultrasonic method, and indentation strain method. However, these methods have low detection efficiency and are difficult to measure directly in real time on the rolling line.

[0003] Therefore, researching new approaches to obtain the residual stress distribution characteristics of sheet materials is of vital importance for improving the processing quality of sheet material products. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for obtaining the residual stress distribution characteristics of metal plates, which can realize the real-time acquisition of the residual stress distribution characteristics of plates in industrial settings, solve the problems existing in the prior art, and improve the quality and efficiency of metal plate processing.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for obtaining the residual stress distribution characteristics of a metal plate, the method comprising:

[0007] Obtain the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress value of each corresponding sub-region;

[0008] Construct a network model for predicting the distribution level of residual stress in sub-regions of a plate based on a convolutional neural network;

[0009] The distribution level prediction network model is trained using the plate shape data of each sub-region and the corresponding distribution level dataset of residual stress in the sub-region;

[0010] Collect plate shape data of each sub-region of the metal plate shape at the production site and input them into the trained distribution level prediction network model to obtain the residual stress distribution level of each sub-region of the metal plate; draw the residual stress distribution feature map of the metal plate shape based on the residual stress distribution level of the plate shape sub-region.

[0011] Optionally, obtaining the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress values ​​of each corresponding sub-region specifically includes:

[0012] Industrial cameras are used to collect image data of the sheet metal, or structured light scanners are used to collect point cloud data matrix of the sheet metal plane height as the sheet shape data;

[0013] A sliding window is set up, and the data matrix within each sliding window is extracted by moving the sliding window to obtain multiple sub-regional plate-shaped data blocks of equal size.

[0014] The residual stress in the location area of ​​the metal plate corresponding to the sub-region plate shape data block is measured to obtain the residual stress distribution value of the sub-region.

[0015] Based on the preset residual stress distribution level corresponding to the range of residual stress distribution values ​​of the metal plate, the residual stress distribution values ​​of the sub-region are assigned residual stress distribution level values;

[0016] Optionally, the distribution level prediction network model includes a 5-layer structure;

[0017] The first layer consists of a convolutional layer with a kernel size of 3×3×16 and a stride of 1, and a max pooling layer.

[0018] The second layer includes one convolutional layer with a kernel size of 3×3×32 and a stride of 1, and four residual modules. Each residual module includes two convolutional layers with kernel size of 3×3×64 and a stride of 1.

[0019] The third layer includes one convolutional layer with a kernel size of 3×3×128 and a stride of 2, and four residual modules, each of which has two convolutional layers with a kernel size of 3×3×128 and a stride of 1.

[0020] The fourth layer includes one convolutional layer with a kernel size of 3×3×256 and a stride of 2, and three residual modules, each of which includes two convolutional layers with a kernel size of 3×3×256 and a stride of 1.

[0021] The 5th layer contains an average pooling layer and a fully connected layer, and outputs the result through a Softmax layer.

[0022] Optionally, each of the convolutional layers is connected to a batch normalization (BN) layer; the activation function used is the Leaky ReLU activation function.

[0023] Optionally, the expression for the Leaky ReLU activation function is:

[0024]

[0025] Among them, orange represents the input to the neuron. It is a very small constant.

[0026] Optionally, the loss function of the distribution level prediction network model is:

[0027]

[0028] Where ju represents a randomly distributed sample value, p represents the expected probability distribution of the sample ju, and q represents the actual probability distribution of the sample ju.

[0029] Optionally, the distribution level prediction network model is trained using the plate shape data of the sub-region and the distribution level data of the residual stress in the corresponding sub-region, specifically including:

[0030] The obtained plate shape data of the sub-region and the distribution level data of the residual stress of the corresponding sub-region are divided into training set, validation set and test set according to a preset ratio;

[0031] The distribution level prediction network model is trained using data from the training set and the validation set.

[0032] Determine if the current iteration count is the maximum iteration count;

[0033] If so, then save the model parameters of the current optimal distribution level prediction network model to obtain the trained distribution level prediction network model;

[0034] If not, return to step "select data from the training set and validation set to train the distribution level prediction network model" until the maximum number of iterations is reached to obtain the trained distribution level prediction network model;

[0035] The sub-region plate shape data in the test set are input into the trained distribution level prediction network model to obtain the predicted value of the residual stress distribution level of each sub-region of the metal plate in the test set.

[0036] The prediction accuracy is calculated based on the predicted value of the residual stress distribution level of each sub-region of the metal plate in the test set and the actual value of the residual stress distribution level of each sub-region of the metal plate.

[0037] Determine whether the prediction accuracy is greater than the preset accuracy.

[0038] If so, the residual stress in the plate-shaped sub-region is predicted using the currently trained distribution level prediction network model.

[0039] If not, increase the sample data in the training set or increase the value of the maximum number of iterations, and return to the step "select data from the training set and validation set to train the distribution level prediction network model".

[0040] Optionally, the metal plate shape data collected from the production site is divided into sub-regions and input into the trained distribution level prediction network model to obtain the residual stress distribution level of each sub-region of the metal plate; based on the residual stress distribution level of the sub-regions, a residual stress distribution feature map of the metal plate shape is drawn, specifically including:

[0041] The on-site shape data of the metal plate is obtained by using the industrial camera to collect image data of the plate or by using the structured light scanner to collect the plane height point cloud data matrix of the plate.

[0042] The data matrix within each sliding window is extracted from the on-site plate shape data matrix by moving the sliding window, resulting in multiple sub-region plate shape data blocks of equal size;

[0043] Each sub-region plate shape data block is input into the trained distribution level prediction network model to output the residual stress prediction distribution level value corresponding to the plate shape sub-region of the metal plate.

[0044] The residual stress distribution levels of each of the metal plate shape regions are arranged sequentially according to their corresponding metal plate region positions, and a residual stress distribution level map of the metal plate shape is drawn. The residual stress distribution level map is used to depict the residual stress distribution characteristics of the metal plate shape.

[0045] The present invention also provides a system for obtaining the residual stress distribution characteristics of a metal plate, the system comprising:

[0046] The sample data acquisition module is used to acquire the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress value of each corresponding sub-region.

[0047] The model building module is used to construct a network model for predicting the distribution level of plate residual stress based on a convolutional neural network.

[0048] The training module is used to train the distribution level prediction network model using the plate shape data of each sub-region of the metal plate and the corresponding residual stress distribution level data of the sub-region;

[0049] The prediction module is used to collect the plate shape data of each sub-region of the metal plate at the production site and input it into the trained distribution level prediction network model to obtain the residual stress distribution level of each sub-region of the metal plate and draw the residual stress distribution characteristics of the metal plate.

[0050] Optionally, the sample data acquisition module specifically includes:

[0051] The plate shape data acquisition submodule uses an industrial camera to collect plate image data or a structured light scanner to collect a plate plane height point cloud data matrix as the plate shape data of the metal plate;

[0052] The plate-shaped data sub-region division sub-module is used to set a sliding window, and extract the data matrix within each sliding window by moving the sliding window to obtain multiple sub-region plate-shaped data blocks of equal size.

[0053] The sub-region residual stress distribution value acquisition sub-module is used to measure the residual stress in the location area of ​​the metal plate corresponding to the sub-region plate shape data block, and obtain the sub-region residual stress distribution value.

[0054] The sub-module for determining the residual stress distribution level of a sub-region assigns a residual stress distribution level value to the residual stress distribution value of the sub-region based on the range of residual stress distribution values ​​of the metal plate and the corresponding preset residual stress distribution level.

[0055] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0056] This invention provides a method and system for obtaining the residual stress distribution characteristics of a metal plate, comprising: acquiring plate shape data of a sub-region of the metal plate and the distribution level of the residual stress values ​​of the corresponding sub-region; constructing a distribution level prediction network model of the residual stress distribution of the plate shape based on a convolutional neural network; training the distribution level prediction network model using the plate shape data of the sub-region and the distribution level dataset of the residual stress distribution of the corresponding sub-region; collecting plate shape data from the production site, dividing the plate shape data into sub-regions, and inputting the data into the trained distribution level prediction network model, outputting the residual stress distribution level of each sub-region of the metal plate; and drawing the residual stress distribution characteristics of the metal plate shape based on the residual stress distribution level of the sub-regions.

[0057] This invention utilizes a convolutional neural network model to find the relationship between the shape features of a metal plate and the distribution of residual stress in the plate. It can efficiently acquire the residual stress distribution features of the plate shape image, and the modeling is simple and easy to implement with low cost, meeting the conditions for real-time acquisition in industrial settings; thus improving the quality and efficiency of processing defective plates. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of a method for obtaining the residual stress distribution characteristics of a metal plate according to Embodiment 1 of the present invention;

[0060] Figure 2 This is a schematic diagram of the sub-region division of plate shape data and the corresponding residual stress distribution levels of the metal plate region provided in Embodiment 1 of the present invention;

[0061] Figure 3 This is a schematic diagram of the structure of the convolutional neural network model for residual stress distribution features provided in Embodiment 1 of the present invention;

[0062] Figure 4 This is a schematic diagram of the predicted plate residual stress distribution characteristics provided in Embodiment 1 of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The purpose of this invention is to provide a method and system for obtaining the residual stress distribution characteristics of metal plates, enabling real-time acquisition of these characteristics in industrial settings. A convolutional neural network model is designed to find the relationship between the metal plate shape characteristics and the residual stress distribution, establishing a predictive model for the relationship between the plate shape characteristics and the residual stress distribution.

[0065] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0066] Example 1

[0067] like Figure 1 As shown in the figure, this embodiment provides a method for obtaining the residual stress distribution characteristics of a metal plate, the method comprising:

[0068] Step S1: Obtain the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress value of each corresponding sub-region, in order to create a sample dataset;

[0069] Step S1 specifically includes:

[0070] Industrial cameras are used to collect image data of the sheet metal, or structured light scanners are used to collect point cloud data matrix of the sheet metal plane height as the sheet shape data;

[0071] Set up a sliding window, and extract the data matrix of the plate-shaped data from left to right and top to bottom through the sliding window to obtain multiple sub-region plate-shaped data blocks of equal size.

[0072] The residual stress at the center of the metal plate sub-region corresponding to the sub-region plate shape data block is measured to obtain the residual stress distribution value of the sub-region.

[0073] Based on the preset residual stress distribution level corresponding to the range of residual stress distribution values ​​of the metal plate, the residual stress distribution values ​​of the sub-region are assigned residual stress distribution level values; wherein, the preset residual stress distribution level is pre-set based on the existing residual stress distribution values ​​and is directly applied when performing distribution level analysis later.

[0074] The distribution level value of residual stress in the metal plate sub-region corresponding to each sub-region plate shape data block is used as the label value for network training of that sub-region plate shape data block, and a sample dataset is created.

[0075] like Figure 2 The image shows a section of a metal sheet, such as... Figure 2 (a) shows the plate shape data collected in this area (a to i represent the image pixel values ​​or unevenness point cloud data values ​​of the plate surface collected), shown through a sliding window as follows: Figure 2 In (a), the window (size 2×2) in the shaded area is used to slide from left to right and from top to bottom with a step size of 1 to extract the plate shape data within the window, which can yield 4 sub-region plate shape data blocks.

[0076] like Figure 2 As shown in (b), F1-F4 represent the residual stress distribution values ​​of the corresponding sub-regions within the selected metal sheet area; according to the range of values ​​for the divided residual stress distribution levels, the distribution levels D1-D4 corresponding to F1-F4 are obtained respectively; the label values ​​of the sub-region plate shape data blocks corresponding to F1-F4 are as follows: Figure 2 As shown in (c).

[0077] Step S2: Construct a network model for predicting the distribution level of residual stress in sub-regions of the plate based on a convolutional neural network;

[0078] The distribution level prediction network model includes a 5-layer structure; as follows: Figure 3 As shown.

[0079] The first layer consists of a convolutional layer with a kernel size of 3×3×16 and a stride of 1, and a max pooling layer.

[0080] The second layer includes one convolutional layer with a kernel size of 3×3×32 and a stride of 1, and four residual modules. Each residual module includes two convolutional layers with kernel size of 3×3×64 and a stride of 1.

[0081] The third layer includes one convolutional layer with a kernel size of 3×3×128 and a stride of 2, and four residual modules, each of which has two convolutional layers with a kernel size of 3×3×128 and a stride of 1.

[0082] The fourth layer includes one convolutional layer with a kernel size of 3×3×256 and a stride of 2, and three residual modules, each of which includes two convolutional layers with a kernel size of 3×3×256 and a stride of 1.

[0083] The 5th layer contains an average pooling layer and a fully connected layer, and outputs the result through a Softmax layer.

[0084] Each convolutional layer is connected to a batch normalization (BN) layer; the activation function used is the LeakyRELU activation function, whose expression is:

[0085]

[0086] Among them, orange represents the input to the neuron. It is a very small constant.

[0087] The cross-entropy function is used as the loss function, and its calculation formula is as follows:

[0088]

[0089] Where ju represents a randomly distributed sample value, p represents the expected probability distribution of the sample ju, and q represents the actual probability distribution of the sample ju.

[0090] Step S3: Train the distribution level prediction network model using the plate shape data of each sub-region and the distribution level data of the residual stress in the corresponding sub-region.

[0091] The sample dataset obtained in step S1 is used to train and validate the sub-region residual stress distribution level prediction network model;

[0092] When training the model, you can set the maximum number of iterations to 100 and the learning rate to 0.00001. These values ​​can also be adjusted arbitrarily according to actual needs.

[0093] Step S3 specifically includes:

[0094] Step S31: The sample dataset is divided into training set, validation set and test set according to the preset ratio of 80%, 10% and 10%;

[0095] Step S32: Select data from the training set and validation set to train the distribution level prediction network model;

[0096] Step S33: Determine if the current iteration number is the maximum iteration number;

[0097] If so, then save the model parameters of the current optimal distribution level prediction network model to obtain the trained distribution level prediction network model;

[0098] If not, return to step S32 until the maximum number of iterations is reached to obtain the trained distribution level prediction network model;

[0099] Step S34: Input the sub-region plate shape data of the test set into the trained distribution level prediction network model to obtain the predicted value of the residual stress distribution level of each sub-region of the metal plate in the test set;

[0100] Step S35: Calculate the prediction accuracy based on the predicted value of the residual stress distribution level of each sub-region of the test set metal plate and the actual value of the residual stress distribution level of each sub-region of the metal plate;

[0101] Step S36: Determine whether the prediction accuracy is greater than the preset accuracy; the preset accuracy can be selected as 80%, or can be set arbitrarily according to requirements.

[0102] If so, the residual stress in the plate-shaped sub-region is predicted using the currently trained distribution level prediction network model.

[0103] If not, increase the data in the training set or increase the value of the maximum number of iterations, and return to step S32.

[0104] Step S4: Collect metal plate shape data from the production site, divide the metal plate shape data into sub-regions, and input them into the trained distribution level prediction network model to output the residual stress distribution level of each sub-region of the metal plate; draw the residual stress distribution feature map of the metal plate shape based on the residual stress distribution level of the sub-regions.

[0105] Step S4 specifically includes:

[0106] Step S41: Use an industrial camera to collect image data of the sheet metal or use a structured light scanner to collect a point cloud data matrix of the sheet metal plane height to obtain the on-site sheet shape data of the metal sheet.

[0107] Step S42: Extract the data matrix within each sliding window of the field plate shape data by moving the set sliding window to obtain multiple sub-region plate shape data blocks of equal size;

[0108] Step S43: Input the plate shape data block of each sub-region into the trained distribution level prediction network model to output the residual stress prediction distribution level value corresponding to the plate shape sub-region of the metal plate;

[0109] Step S44: The residual stress distribution levels of each of the aforementioned metal plate shapes are arranged sequentially according to their corresponding metal plate region positions. A residual stress distribution level map of the metal plate shapes at the production site is then drawn using a heat map format, such as... Figure 4 As shown, the residual stress distribution characteristics of the metal plate are depicted by the residual stress distribution level diagram.

[0110] In this embodiment, firstly, the plate shape data of the metal plate sub-regions and the distribution level of the residual stress values ​​of the corresponding sub-regions are obtained to create a sample dataset; a plate shape sub-region residual stress distribution level prediction network model based on convolutional neural network is constructed, and the sample dataset is divided into training set and test set for training and verification of the residual stress distribution level prediction network model; plate shape data from the production site are collected, sub-region plate shape data are divided and input into the trained distribution level prediction network model, the residual stress distribution level of each sub-region of the metal plate is output, and the plate shape residual stress distribution characteristics of the metal plate are plotted.

[0111] This invention enables real-time acquisition of residual stress distribution characteristics of metal plates on the production site, which can be used to optimize processing techniques such as metal plate straightening, and improve the quality and efficiency of straightening and processing defective plates.

[0112] Example 2

[0113] This embodiment provides a system for obtaining the residual stress distribution characteristics of a metal plate, the system comprising:

[0114] The sample data acquisition module M1 acquires the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress value of each corresponding sub-region.

[0115] The model building module M2 is used to construct a network model for predicting the distribution level of residual stress in sub-regions of the plate based on a convolutional neural network.

[0116] Training module M3 is used to train the distribution level prediction network model using the plate shape data of each sub-region and the distribution level data of the residual stress in the corresponding sub-region;

[0117] The prediction module M4 is used to collect the plate shape data of each sub-region of the metal plate at the production site and input it into the trained distribution level prediction network model, output the residual stress distribution level of each sub-region of the metal plate, and draw the residual stress distribution feature map of the metal plate.

[0118] The sample data acquisition module M1 specifically includes:

[0119] The plate shape data acquisition submodule uses an industrial camera to collect plate image data or a structured light scanner to collect a plate plane height point cloud data matrix as the plate shape data of the metal plate;

[0120] The plate-shaped data sub-region division sub-module is used to set a sliding window, and extract the data matrix within each sliding window by moving the sliding window to obtain multiple sub-region plate-shaped data blocks of equal size;

[0121] The sub-region residual stress distribution value acquisition sub-module is used to measure the residual stress in the location area of ​​the metal plate corresponding to the sub-region plate shape data block, and obtain the sub-region residual stress distribution value.

[0122] The sub-module for determining the residual stress distribution level of a sub-region is used to assign a residual stress distribution level value to the residual stress distribution value of the sub-region based on the range of residual stress distribution values ​​of the metal plate and the corresponding preset residual stress distribution level.

[0123] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0124] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for obtaining the residual stress distribution characteristics of a metal plate, characterized in that, The method includes: Obtain the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress value of each corresponding sub-region; Construct a network model for predicting the distribution level of residual stress in sub-regions of a plate based on a convolutional neural network; The distribution level prediction network model is trained using the plate shape data of each sub-region and the distribution level data of the residual stress in the corresponding sub-region; Collect sheet shape data of each sub-region of the metal sheet shape at the production site and input them into the trained distribution level prediction network model to output the residual stress distribution level of each sub-region of the metal sheet; draw the residual stress distribution feature map of the metal sheet shape based on the residual stress distribution level of the sub-region of the sheet shape. Specifically, obtaining the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress values ​​of each corresponding sub-region includes: A structured light scanner was used to collect point cloud data matrix of the planar height of the sheet metal as the sheet shape data; A sliding window is set up, and the data matrix of the plate-shaped data is extracted by moving the sliding window to obtain multiple sub-region plate-shaped data blocks of equal size; The residual stress in the location area of ​​the metal plate corresponding to the sub-region plate shape data block is measured to obtain the residual stress distribution value of the sub-region. Based on the preset residual stress distribution level corresponding to the range of residual stress distribution values ​​of the metal plate, the residual stress distribution values ​​of the sub-region are assigned residual stress distribution level values.

2. The method according to claim 1, characterized in that, The network model for predicting the distribution level of residual stress in sub-regions of the plate based on convolutional neural networks includes a 5-layer structure. The first layer includes a convolutional kernel of size 3. 3 16, a convolutional layer with a stride of 1 and a max pooling layer; The second layer includes one convolutional kernel with a value of 3. 3 32, a convolutional layer with a stride of 1 and 4 residual modules, each of the residual modules comprising 2 convolutional kernels of 3... 3 64, convolutional layers with a stride of 1; The third layer includes one convolutional kernel with a value of 3. 3 128, a convolutional layer with a stride of 2 and 4 residual modules, each of which has 2 convolutional kernels of 3. 3 128, convolutional layers with a stride of 1; The fourth layer includes one convolutional kernel with a value of 3. 3 256, a convolutional layer with a stride of 2 and 3 residual modules, each residual module comprising 2 convolutional kernels with a stride of 3. 3 256, convolutional layers with a stride of 1; The 5th layer contains an average pooling layer and a fully connected layer, and outputs the result through a Softmax layer.

3. The method according to claim 2, characterized in that, Each of the convolutional layers is connected to a batch normalized layer; the activation function used is the Leaky ReLU activation function.

4. The method according to claim 3, characterized in that, The expression for the Leaky ReLU activation function is: in, For the input of neurons, It is a constant.

5. The method according to claim 1, characterized in that, The loss function of the distribution level prediction network model is: in, For randomly distributed sample values, For the sample The expected probability distribution For the sample The actual probability distribution.

6. The method according to claim 1, characterized in that, The distribution level prediction network model is trained using the plate shape data of the sub-region and the distribution level data of the residual stress in the corresponding sub-region, specifically including: The obtained plate shape data of the sub-region and the distribution level data of the residual stress of the corresponding sub-region are divided into training set, validation set and test set according to a preset ratio; The distribution level prediction network model is trained using data from the training set and the validation set. Determine if the current iteration count is the maximum iteration count; If so, then save the model parameters of the current optimal distribution level prediction network model to obtain the trained distribution level prediction network model; If not, return to step "select data from the training set and the validation set to train the distribution level prediction network model" until the maximum number of iterations is reached to obtain the trained distribution level prediction network model; The sub-region plate shape data in the test set are input into the trained distribution level prediction network model to obtain the predicted value of the residual stress distribution level of each sub-region of the metal plate in the test set. The prediction accuracy is calculated based on the predicted value of the residual stress distribution level of each sub-region of the metal plate in the test set and the actual value of the residual stress distribution level of each sub-region of the metal plate. Determine whether the prediction accuracy is greater than the preset accuracy. If so, the residual stress in the plate-shaped sub-region is predicted using the currently trained distribution level prediction network model. If not, increase the sample data in the training set or increase the value of the maximum number of iterations, and return to the step "select data from the training set and the validation set to train the distribution level prediction network model".

7. The method according to claim 1, characterized in that, The process involves collecting sheet shape data from various sub-regions of the metal plate at the production site and inputting it into the trained distribution level prediction network model to obtain the residual stress distribution level of each sub-region of the metal plate. Based on the residual stress distribution level of the sheet shape sub-regions, a residual stress distribution feature map of the metal plate is drawn, specifically including: The structured light scanner is used to collect point cloud data matrix of the plate's planar height to obtain the on-site plate shape data of the metal plate; The data matrix of the on-site plate shape data is extracted by moving the sliding window to obtain multiple sub-region plate shape data blocks of equal size; Each sub-region plate shape data block is input into the trained distribution level prediction network model to output the residual stress prediction distribution level value corresponding to the plate shape sub-region of the metal plate. The residual stress distribution levels of each of the metal plate shape sub-regions are arranged sequentially according to their corresponding metal plate region positions, and a residual stress distribution level map of the metal plate shape is drawn. The residual stress distribution level map is used to depict the residual stress distribution characteristics of the metal plate shape.

8. A system for obtaining the residual stress distribution characteristics of a metal plate according to any one of claims 1 to 7, characterized in that, The system includes: The sample data acquisition module is used to acquire the plate shape data of each sub-region of the metal plate and the distribution level of the residual stress value of each corresponding sub-region. The model building module is used to construct a network model for predicting the distribution level of residual stress in sub-regions of sheet metal based on a convolutional neural network. The training module is used to train the distribution level prediction network model using the plate shape data of each sub-region and the distribution level data of the residual stress in the corresponding sub-region; The prediction module is used to collect the plate shape data of each sub-region of the metal plate at the production site and input it into the trained distribution level prediction network model, output the residual stress distribution level of each sub-region of the metal plate, and draw the residual stress distribution feature map of the metal plate. The sample data acquisition module specifically includes: The plate shape data acquisition submodule uses a structured light scanner to collect the plane height point cloud data matrix of the plate as the plate shape data of the metal plate; The plate-shaped data sub-region division sub-module is used to set a sliding window, and extract the data matrix within each sliding window by moving the sliding window to obtain multiple sub-region plate-shaped data blocks of equal size; The sub-region residual stress distribution value acquisition sub-module is used to measure the residual stress in the location area of ​​the metal plate corresponding to the sub-region plate shape data block, and obtain the sub-region residual stress distribution value. The sub-region residual stress distribution level determination sub-module is used to assign a residual stress distribution level value to the residual stress distribution value of the sub-region based on the preset residual stress distribution level corresponding to the range of residual stress distribution values ​​of the metal plate.

Images