Welding porosity prediction method based on temperature field RGB gradient fusion

Abstract

The invention discloses a welding porosity prediction method based on temperature field RGB gradient fusion. According to the method, welding time sequence temperature field data and a deep learning image recognition technology are creatively combined, and the method is achieved through the following technical paths that firstly, the mapping relation between welding process parameters and the porosity is established, and temperature thermal cycle data are obtained through finite element simulation; secondly, performing compression and enhancement processing on the temperature data of the three feature points, and converting a time sequence temperature curve into R, G and B three-channel color matrixes; a color gradient image is generated through multi-source data fusion; finally, a deep neural network model is constructed, and imaging intelligent grading prediction of the welding porosity is achieved. Compared with a traditional physical experiment method, the method has the advantages that the experiment cost is greatly reduced through a simulation technology, and the space-time evolution characteristics of the temperature field can be effectively represented by adopting a chromatographic gradient coding technology. The technology provides a new digital solution for welding quality prediction, and has important value in the field of welding engineering application.

Description

Technical Field

[0001] The invention belongs to the technical field of welding, and in particular relates to a welding porosity prediction method based on temperature field RGB gradient fusion. Background Art

[0002] Welding technology is widely used in the automotive, aerospace, shipbuilding and energy industries. Its high efficiency and reliability in connecting various materials and structures make it an indispensable and important process in modern industrial manufacturing. However, due to the high energy input of the heat source, the heat-affected zone is large, and defects such as welding slag, deformation, pores and thermal cracks are easily generated by traditional welding technology. In particular, pore defects are the easiest to generate and difficult to control, which limits the further improvement of welding quality.

[0003] Especially for the welding of aluminum alloy materials, porosity defects remain a key problem that needs to be solved in industrial applications. These pores are mainly caused by gas capture or incomplete discharge in the molten pool, usually including hydrogen, air or oxide particles. The presence of pores will significantly reduce the mechanical properties of the welded joint, especially fatigue strength and tensile properties. In order to solve this problem, many experts and scholars have conducted in-depth research on the formation mechanism of pores and proposed some innovative methods for pore suppression, such as adding laser oscillation to improve the dynamic characteristics of the molten pool.

[0004] Although laser welding technology has made significant progress in aluminum alloy welding, the prediction of porosity defects remains a complex and challenging problem. With the help of high-performance computing technology, researchers have developed multi-physics numerical simulation models that combine processes such as heat conduction, fluid dynamics, and metal evaporation to simulate the gas behavior inside the molten pool. However, most of them can only simulate the pore generation mechanism and cannot accurately obtain the porosity under different process parameters. In addition, the fluid simulation model is complex and time-consuming. Therefore, it is very necessary to seek a simple technical method that can predict porosity only through temperature field, which can save a lot of time and cost for process personnel and has direct engineering practical value. Summary of the invention

[0005] In response to the above technical problems, the present invention provides a welding porosity prediction method based on temperature field RGB gradient fusion, which aims to convert the multi-point thermal cycle temperature field process into a computer-recognizable image by RGB chromatographic gradient fusion, so as to facilitate the construction of a deep learning model based on chromatographic gradient, thereby achieving the purpose of predicting welding porosity.

[0006] 1. The technical solution of the present invention is: a welding porosity prediction method based on temperature field RGB gradient fusion, characterized in that it includes the following steps:

[0007] A. Determine the real physical welding experiment plan to obtain the original modeling basic data, take different welding process parameters as input, and use porosity as the output target for excellent classification;

[0008] B. Perform finite element simulation verification based on the physical test data in step A, and extract temperature field data;

[0009] C. Processing the temperature field data extracted in step B, the specific steps are:

[0010] C1. Compress each set of temperature thermal cycle result data points. According to the temperature thermal cycle probe data extracted by the finite element simulation model, the data point set is compressed by normal distribution random sampling. The sampled data points S(x) obey the normal distribution random variable with mean μ and standard deviation σ, that is, S(x)~N(μ,σ). The mean μ and standard deviation σ are determined according to the overall temperature range distribution, and the probability density is:

[0011]

[0012] C2. Perform data enhancement processing on each group of compressed temperature data. The data enhancement process can add noise and shift the original data;

[0013] C3. Align the three different point data in each data set, and convert them into red, green and blue color matrix images with R, G and B channels respectively. The temperature data is based on the appropriate temperature range of the material welding process, such as [22℃-1000℃] for aluminum alloy, and align the data to the RGB color description range [0, 255], and then rearrange the data into a 15×15 matrix image;

[0014] C4, fuse each group of RGB channel images into a color matrix image. The color matrix image generated by the fusion of the three RGB channels covers the information of the thermal cycle of the three points and the temperature gradient information between the points;

[0015] D. Using each set of image data finally obtained in step C as a deep learning training set input parameter, using the corresponding porosity classification as a deep learning model output parameter, and predicting the porosity of the welding workpiece according to the deep learning model finally trained;

[0016] 2. The method according to claim 1, characterized in that, in step A, the original modeling basic data includes welding trajectory, welding power, welding speed, welding swing frequency, welding swing amplitude, and porosity calculated based on the cross section of the weld, wherein the welding trajectory includes but is not limited to the shapes of "0", "8", and "∞";

[0017] 3. The method according to claim 1, characterized in that, in step B, the specific steps of performing finite element simulation verification based on the physical test data in step A and extracting temperature field data are as follows:

[0018] B1. Carry out 3D modeling of the workpiece model and import it into the finite element analysis software;

[0019] B2. In the finite element simulation software, analyze and set the material parameters, initial conditions, boundary conditions, etc. of the welding model. The material properties include density, thermal expansion coefficient, Young's modulus, Poisson's rate, yield strength, thermal conductivity and specific heat, etc. The specific meshing type is based on the structure of the welding workpiece, and the appropriate tetrahedral mesh or hexahedral mesh is selected. The initial conditions and boundary conditions include initial ambient temperature, thermal convection heat transfer coefficient and constraint force conditions, etc.;

[0020] B3. Determine the heat source model and calibrate the finite element simulation heat source model. The heat sources include electron beam and laser, and the heat source models include Gaussian heat source model, double ellipsoid heat source model and hemispherical heat source model, etc. The heat source parameters include but are not limited to welding voltage, welding current, laser power, heat source moving speed, etc.;

[0021] B4. Extracting the temperature thermal cycle probe result data of three designated points for each group of finite element simulation models, where the three points have temperature gradient characteristics in the temperature field;

[0022] 4. The method as claimed in claim 1, characterized in that in the step D, the deep learning model is a model based on a CNN convolutional neural network, including but not limited to network models such as Alexnet, Resnet, and Googlenet. For the trained deep learning model, the RGB color matrix image in the step C4 under any welding process parameters is input, and the porosity under the corresponding welding process parameters can be graded and predicted.

[0023] The beneficial effects of the present invention are as follows: the present invention proposes a welding porosity prediction method based on temperature field RGB gradient fusion. For the first time, the temperature gradient curve is converted into an RGB color spectrum and applied in the welding field, and applied in actual engineering cases to solve the problem of welding porosity defects that are expensive to detect and difficult to predict, and to select low-cost, high-efficiency and high-quality welding process solutions for enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic flow chart of a welding porosity prediction method based on temperature field RGB gradient fusion in the present invention.

[0025] Figure 2 It is a schematic diagram of the welding method in the present invention.

[0026] Figure 3 It is a schematic diagram of the welding trajectory in the present invention.

[0027] Figure 4 It is a schematic diagram of the combined heat source in the present invention.

[0028] Figure 5 It is a schematic diagram of heat source verification in the present invention.

[0029] Figure 6 Schematic diagram of sampling points in the present invention.

[0030] Figure 7 It is the original data curve in the present invention.

[0031] Figure 8 It is the compressed data curve in the present invention.

[0032] Fig. 9 It is the RGB temperature gradient fusion image in the present invention.

[0033] Fig.10 It is a deep learning framework diagram in the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose and technical solution of the present invention clearer, the following is a detailed description with reference to specific examples. It should be understood that the specific implementation examples described herein are only used to explain the present invention, but are not limited to the present invention.

[0035] like Figure 1 The figure shows a schematic flow chart of the application of a welding porosity prediction method based on temperature field RGB gradient fusion in laser swing welding. The application of a welding porosity prediction method based on temperature field RGB gradient fusion in laser swing welding includes the following steps:

[0036] A. Determine the real physical welding experiment plan to obtain the original modeling basic data, take different welding process parameters as input, and use porosity as the output target for excellent classification;

[0037] B. Perform finite element simulation verification based on the physical test data in step A, and extract temperature field data;

[0038] C. Processing the temperature field data extracted in step B, the specific steps are:

[0039] C1. Compress each set of temperature thermal cycle result data points. According to the temperature thermal cycle probe data extracted by the finite element simulation model, the data point set is compressed by normal distribution random sampling. The sampled data points S(x) obey the normal distribution random variable with mean μ and standard deviation σ, that is, S(x)~N(μ,σ). The mean μ and standard deviation σ are determined according to the overall temperature range distribution, and the probability density is:

[0040]

[0041] C2. Perform data enhancement processing on each group of compressed temperature data. The data enhancement process can add noise and shift the original data;

[0042] C3. Align the three different point data in each data set, and convert them into red, green and blue color matrix images with R, G and B channels respectively. The temperature data is based on the appropriate temperature range of the material welding process, such as [22℃-1000℃] for aluminum alloy, and align the data to the RGB color description range [0, 255], and then rearrange the data into a 15×15 matrix image;

[0043] C4, fuse each group of RGB channel images into a color matrix image. The color matrix image generated by the fusion of the three RGB channels covers the information of the thermal cycle of the three points and the temperature gradient information between the points;

[0044] D. Each set of image data finally obtained in step C is used as a deep learning training set input parameter, and the corresponding porosity classification is used as a deep learning model output parameter. The porosity of the welding workpiece is predicted according to the deep learning model that is finally trained.

[0045] In step A, the original modeling basic data includes welding trajectory, welding power, welding speed, welding oscillation frequency, welding oscillation amplitude, and porosity calculated based on the weld cross section, wherein the welding trajectory includes but is not limited to "0", "8", and "∞" shapes.

[0046] The present invention takes the laser oscillating welding of flat plate butt joint as an example. The three-dimensional size of the workpiece, such as Figure 2 As shown, the overall length is 40mm, the width is 40mm, and the height is 8mm; the welding track is in the shape of "0", such as Figure 3 As shown; the porosity classification is carried out according to GB / T 22085.2-2008 "Guidelines for Quality Classification of Electron Beam and Laser Welded Joints Part 2: Aluminum and Aluminum Alloys" B / C / D classification, and the porosity index of >10% is increased to unqualified E grade.

[0047] Table 1 Basic data table

[0048] Serial number Track shape Laser power Welding speed Swing Amplitude Oscillation frequency Porosity Defect classification 1 ○ 4 1.8 1 100 37.3 E 2 ○ 4 1.8 1 200 28.7 E 3 ○ 4 1.8 1 300 22.3 E 4 ○ 4 1.8 2 100 27.4 E 5 ○ 4 1.8 2 200 10.4 D 6 ○ 4 1.8 2 250 5.2 C 7 ○ 4 1.8 2 300 0 B 8 ○ 4 1.8 3 300 0 B

[0049] In step B, the specific steps of performing finite element simulation verification based on the physical test data in step A and extracting temperature field data are as follows:

[0050] B1. Carry out 3D modeling of the workpiece model and import it into the finite element analysis software;

[0051] B2. In the finite element simulation software, analyze and set the material parameters, initial conditions, boundary conditions, etc. of the welding model. The material properties include density, thermal expansion coefficient, Young's modulus, Poisson's rate, yield strength, thermal conductivity and specific heat, etc. The specific meshing type is based on the structure of the welding workpiece, and the appropriate tetrahedral mesh or hexahedral mesh is selected. The initial conditions and boundary conditions include initial ambient temperature, thermal convection heat transfer coefficient and constraint force conditions, etc.;

[0052] In step B2, the material properties of the welding workpiece are specifically set as follows: the material is set to 5A06, the density is shown in Table 2, the thermal conductivity is shown in Table 3, and the specific heat is shown in Table 4.

[0053] Table 2 Density table

[0054] Temperature (C) 20 100 200 300 400 600 800 1200 1500 <![CDATA[Density (kg·m -3 )]]> 2660 2660 2550 2500 2350 2350 2300 2300 2300

[0055] Table 3 Thermal conductivity coefficient table

[0056] Temperature (C) 20 100 200 300 400 600 800 1200 1500 <![CDATA[Thermal conductivity (W·m -1 ·C -1 )]]> 118 121 126 130 138 118 116 122 128

[0057] Table 4 Specific heat table

[0058] Temperature (C) 20 100 200 300 400 600 800 1200 1500 <![CDATA[Specific heat (W·m -1 ·°C -1 )]]> 924 921 1005 1047 1089 1390 1200 1010 1005

[0059] Determine the meshing type as hexahedral unit meshing, the mesh size is between 0.2mm-1.4mm, and the local density around the weld is increased; the initial condition sets the initial ambient temperature to 22℃ and the heat convection coefficient to 20W / m 2 ℃.

[0060] B3. Determine the heat source model and calibrate the finite element simulation heat source model. The heat sources include electron beam and laser, and the heat source models include Gaussian heat source model, double ellipsoid heat source model and hemispherical heat source model, etc. The heat source parameters include but are not limited to welding voltage, welding current, laser power, heat source moving speed, etc.;

[0061] In step B3, the heat source is an oscillating laser, and the heat source model is selected as a combination model of a Gaussian surface heat source and a body heat source, such as Figure 4 As shown; and the heat source model is calibrated, such as Figure 5 As shown;

[0062]

[0063] Q η =Q m +Q t

[0064] B4. Extracting the temperature thermal cycle probe result data of three designated points for each group of finite element simulation models, where the three points have temperature gradient characteristics in the temperature field;

[0065] In step C, the temperature field data extracted in step B is processed, and the specific steps are:

[0066] C1. The coordinates of the temperature sampling points are as follows: Figure 6 As shown, each set of temperature thermal cycle result data points is compressed. According to the temperature thermal cycle probe data extracted by the finite element simulation model, the data point set is compressed by normal distribution random sampling. The sampled data points S(x) obey the normal distribution random variable with mean μ = 0.4 and standard deviation σ = 1000, that is, S(x) ~ N(μ,σ), and its probability density is:

[0067]

[0068] The compressed data uses normal distribution to sample the original data at the center, as shown in the schematic diagram Figure 7 As shown, the compressed data is Figure 8 shown.

[0069] C2. Perform data enhancement processing on each group of compressed temperature data. The data enhancement process adds noise and shifts the original data to expand the number of samples.

[0070] C3. Align the three different points in each data set, scale the compressed data of sampling points 1-3 from the temperature range [22-1000] to the RGB color spectrum range [0-255], and then rearrange the data into a 15×15 matrix image and convert it into red, green, and blue color matrix images with R, G, and B channels respectively. Among them, sampling point 1 corresponds to the R channel, sampling point 2 corresponds to the G channel, and sampling point 3 corresponds to the B channel;

[0071] C4, each group of RGB channel images is fused into a color matrix image. The color matrix image generated by the fusion of the three RGB channels covers the information of the three-point thermal cycle and the temperature gradient information between the points, such as Fig. 9 As shown;

[0072] In step D, deep learning training is performed by improving the Googlenet framework, as shown in the following figure: Fig.10As shown, for the trained deep learning model, the RGB color matrix image in step C4 under any welding process parameters is input, and the porosity under the corresponding welding process parameters can be graded and predicted.

[0073] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A welding porosity prediction method based on temperature field RGB gradient fusion, characterized in that: The following steps are involved: A. Determine the real physical welding experiment plan to obtain the original modeling basic data, take different welding process parameters as input, and use porosity as the output target for excellent classification; B. Perform finite element simulation verification based on the physical test data in step A, and extract temperature field data; C. Processing the temperature field data extracted in step B, the specific steps are: C1. Compress each set of temperature thermal cycle result data points. According to the temperature thermal cycle probe data extracted by the finite element simulation model, the data point set is compressed by normal distribution random sampling. The sampled data points S(x) obey the normal distribution random variable with mean μ and standard deviation σ, that is, S(x)~N(μ,σ). The mean μ and standard deviation σ are determined according to the overall temperature range distribution, and the probability density is: C2. Perform data enhancement processing on each group of compressed temperature data. The data enhancement process can add noise and shift the original data; C3. Align the three different point data in each data set, and convert them into red, green and blue color matrix images with R, G and B channels respectively. The temperature data is based on the appropriate temperature range of the material welding process, such as [22℃-1000℃] for aluminum alloy, and align the data to the RGB color description range [0, 255], and then rearrange the data into a 15×15 matrix image; C4, fuse each group of RGB channel images into a color matrix image. The color matrix image generated by the fusion of the three RGB channels covers the information of the thermal cycle of the three points and the temperature gradient information between the points; D. Each set of image data finally obtained in step C is used as a deep learning training set input parameter, and the corresponding porosity classification is used as a deep learning model output parameter. The porosity of the welding workpiece is predicted according to the deep learning model that is finally trained.

2. The method according to claim 1, characterized in that: In step A, the original modeling basic data includes welding trajectory, welding power, welding speed, welding oscillation frequency, welding oscillation amplitude, and porosity calculated based on the weld cross section, wherein the welding trajectory includes but is not limited to "0", "8", and "∞" shapes.

3. The method according to claim 1, characterized in that: In step B, the specific steps of performing finite element simulation verification based on the physical test data in step A and extracting temperature field data are as follows: B1. Carry out 3D modeling of the workpiece model and import it into the finite element analysis software; B2. In the finite element simulation software, analyze and set the material parameters, initial conditions, boundary conditions, etc. of the welding model. The material properties include density, thermal expansion coefficient, Young's modulus, Poisson's rate, yield strength, thermal conductivity and specific heat, etc. The specific meshing type is based on the structure of the welding workpiece, and the appropriate tetrahedral mesh or hexahedral mesh is selected. The initial conditions and boundary conditions include initial ambient temperature, thermal convection heat transfer coefficient and constraint force conditions, etc.; B3. Determine the heat source model and calibrate the finite element simulation heat source model. The heat sources include electron beam and laser, and the heat source models include Gaussian heat source model, double ellipsoid heat source model and hemispherical heat source model, etc. The heat source parameters include but are not limited to welding voltage, welding current, laser power, heat source moving speed, etc.; B4. For each group of finite element simulation models, the temperature thermal cycle probe result data of three designated points are extracted, and the three points have temperature gradient characteristics in the temperature field.

4. The method according to claim 1, characterized in that: In step D, the deep learning model is a model based on a CNN convolutional neural network, including but not limited to network models such as Alexnet, Resnet, and Googlenet. For the trained deep learning model, the RGB color matrix image in step C4 under any welding process parameters is input, and the porosity under the corresponding welding process parameters can be graded and predicted.

Images