Butt welding back weld width prediction and control method and system

Abstract

The invention provides a butt welding back weld width prediction method which comprises the following steps: collecting molten pool image data under the same groove angle and audio data during welding, converting the collected audio data into two-dimensional data, and taking the two-dimensional data, the molten pool image data and the back weld width as input values; and constructing a regression model for predicting the back weld width by using the molten pool image and the audio. The invention further provides a butt welding back weld width control method, the output value of the regression model serves as the input value of the PID algorithm, and the PID algorithm is used for regulating and controlling the welding current in the welding process. The method has high precision and anti-interference capability, real-time control over the width of the back welding seam can be achieved, and the welding quality of butt welding is guaranteed.

Description

Technical Field

[0001] The invention relates to the field of welding, and in particular to the field of back weld width prediction and control. Background Art

[0002] In order to ensure the safety of welding products, it is necessary to conduct certain inspections and evaluations on the appearance quality of the weld after welding. At present, this type of inspection still mainly relies on professionals with inspection experience to achieve visual inspection or use measuring tools such as weld inspection rulers. However, due to the lag of manual inspection, it is difficult to control the actual forming of the weld in real time. If repairs are performed, certain labor costs and economic losses will be incurred. In addition, during the welding process, the weld width and excess height have an important influence on the quality and strength of the welding. The weld width on the back of the weld is one of the intuitive manifestations of the penetration state. In butt welding, the back weld is formed by the front weld penetrating the butt gap of the workpiece, which is closely related to the formation of the front weld.

[0003] The back width of the weld is an important basis for judging the welding quality. How to accurately monitor and adjust the back weld width in real time is an effective way to improve the welding quality. Summary of the invention

[0004] An object of the present invention is to provide a method for predicting the width of the back weld of a butt weld.

[0005] To achieve the above-mentioned purpose, a method for predicting the back weld width of butt welding includes the following steps: collecting molten pool image data and audio data during welding at the same groove angle, as well as the back weld width at the corresponding moments of the molten pool image data and the audio data; converting the collected audio data into two-dimensional data, making the molten pool image data and the back weld width as input values ​​together, and constructing a regression model for predicting the back weld width using molten pool images and audio.

[0006] In one or more embodiments, the collected audio data is subjected to short-time Fourier transform to generate a spectrogram.

[0007] In one or more embodiments, a back weld is obtained using a three-dimensional scanner; the height value perpendicular to the bottom direction of the workpiece is used as a grayscale value, and the three-dimensional data obtained by the three-dimensional scanner is converted into a two-dimensional grayscale image; the contour of the back weld is obtained using threshold segmentation; and a continuous back weld length-width curve is obtained by fitting.

[0008] In one or more embodiments, the length of the back weld at the corresponding moment of each image is calculated, and the back weld width corresponding to the front molten pool in each image is obtained according to the length-width curve.

[0009] In one or more embodiments, the steps of constructing a regression model are: dividing the image data into blocks in a Patch Partition module and flattening it in the channel dimension; extracting the features of the image layer by layer through multiple continuous hierarchical structures; normalizing and pooling the feature data, and obtaining the backside melt width prediction value through a fully connected layer.

[0010] Another object of the present invention is to provide a method for controlling the width of the back weld of butt welding, using the above-mentioned method for predicting the width of the back weld of butt welding to obtain a regression model, using the output value of the regression model as the input value of the PID algorithm, and using the PID algorithm to regulate the welding current of the welding process.

[0011] In one or more embodiments, the regression model is used to obtain multiple groups of back weld width prediction values, and an average value of the multiple groups of back weld width prediction values ​​is used as an input value of the PID algorithm.

[0012] In one or more embodiments, the welding current is regulated by adjusting a control coefficient of a PID algorithm.

[0013] Another object of the present invention is to provide a butt welding back weld width control system, which includes an image acquisition system, an audio acquisition system, a welding system and a control module, the image acquisition system includes a molten pool camera; the audio acquisition system includes a microphone and an acquisition card; the welding system includes a welding gun and a welding robot, the welding robot is connected to the welding gun, the molten pool camera and the microphone are arranged on the welding gun and move with the welding gun; the control module is controlled to send a welding current signal to the welding system through the above method.

[0014] The welding gun is at an angle of 60° to the welding direction.

[0015] The above-mentioned prediction and control method is based on the image data and audio data obtained by the molten pool camera and the microphone as dual inputs, respectively, to construct a regression model for the prediction value of the back weld width of the butt weld. In particular, by adding the two-dimensional data converted from the audio data as input values ​​together with the image data, and by processing and fusing them separately, an accurate regression model is obtained, which can significantly improve the anti-interference ability and accuracy of the model, avoid the problem of accuracy affected by errors in the image data, and realize real-time and accurate prediction of the back weld width of the butt weld; and further based on the predicted value, the welding current is adjusted through the PID control algorithm to realize real-time control of the back weld width and ensure the welding quality of the butt weld. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The above and other features, properties and advantages of the present invention will become more apparent through the following description in conjunction with the accompanying drawings and embodiments, in which:

[0017] Figure 1 This is the device diagram of the butt welding back weld width monitoring and control system;

[0018] Figure 2 Schematic diagram of the workpiece used in the butt welding experiment;

[0019] Figure 3 It is three-dimensional point cloud data;

[0020] Figure 4 is a two-dimensional grayscale image;

[0021] Figure 5 is the profile of the back weld;

[0022] Figure 6 is a continuous back weld length-width curve;

[0023] Figure 7 This is the regression network structure diagram of the backside melt width;

[0024] Figure 8 This is the flow chart of the Patch Merging model;

[0025] Fig. 9 It is a structural diagram of two consecutive Swin Transformer modules;

[0026] Fig.10 It is the curve diagram of the actual measured value and the model predicted value of the back weld width of butt welding;

[0027] Fig.11 This is a comparison chart of the control effect of the back weld width of the butt weld with a groove angle of 30°;

[0028] Fig.12 This is a comparison chart of the control effect of the back weld width of the butt weld with a groove angle of 45°;

[0029] Fig.13 This is a comparison chart of the control effect of the back weld width of the butt weld when the groove angle suddenly changes from 45° to 30°;

[0030] Fig.14 This is a comparison chart of the control effect of the back weld width of butt welds with groove angles gradually changing from 30° to 45°. DETAILED DESCRIPTION

[0031] The present invention is further described below in conjunction with specific embodiments and drawings. More details are elaborated in the following description to facilitate a full understanding of the present invention. However, the present invention can obviously be implemented in a variety of other ways different from the description herein. Those skilled in the art can make similar generalizations and deductions based on actual application situations without violating the connotation of the present invention. Therefore, the protection scope of the present invention should not be limited by the content of this specific embodiment.

[0032] It should be noted that these and other subsequent drawings are only examples and are not drawn to scale, and should not be used to limit the actual scope of protection required by the present invention.

[0033] The prediction method disclosed in the present invention utilizes a molten pool camera to collect the molten pool image of the front side of the molten pool during the welding process, and simultaneously utilizes a microphone to collect the audio signal during the welding process, constructs a regression model of the two sets of data and the back weld width, and adjusts the welding current through a PID fuzzy control algorithm to achieve real-time control of the back weld width, thereby ensuring the welding quality of butt welding.

[0034] When acquiring data, during the GMAW welding process, the strong arc light and high temperature radiation often affect the imaging effect of the molten pool in the image, interfere with the acquisition of detailed information, and make it difficult to analyze the molten pool morphological parameters. Therefore, it is necessary to optimize the visual sensor and extract high-level abstract information to build a backside melt width regression model.

[0035] Based on this, the method for predicting the back weld width of butt welding described in the present invention includes the following steps: collecting molten pool image data and audio data during welding at the same groove angle, as well as the back weld width at the corresponding moment of the molten pool image data and audio data; converting the collected audio data into two-dimensional data, and using it together with the molten pool image data and the back weld width as input values, to construct a regression model for predicting the back weld width using molten pool images and audio.

[0036] Specifically, the method uses a butt weld backside weld width monitoring and control system such as Figure 1 As shown, it includes an image acquisition system 101, an audio acquisition system 102, a welding system 103 and a control module 104. The control part in the welding system is directly connected to the welding robot through a computer, and sends an ideal welding current value to modify its internal parameters. The visual system includes but is not limited to a molten pool camera and a lens with a focal length of 35mm.

[0037] In some embodiments, the molten pool camera is fixed on the welding gun, maintained at a 60° angle to the welding direction, and moves with the welding gun, so that the molten pool is always in the same position in the image, which can reduce the interference of the background environment and focus on the morphological changes of the molten pool during welding. The acoustic system includes a microphone and an acquisition card. The microphone is fixed on the welding gun. In order to prevent the two modules from blocking the influence, they are placed symmetrically relative to the welding gun, preferably at a 60° angle to the welding direction, and move with the welding gun.

[0038] After the above-mentioned visual system is built, continuous molten pool images are acquired using an acquisition frame rate including but not limited to 400Hz. To ensure high-speed acquisition, it is preferred to crop the image size to 500×700 in the acquisition program. In view of the jitter problem of the follow-up equipment during the welding process, a 0-50 offset is randomly added to the horizontal and vertical directions of the original image to make the model more robust. Continuous multiple frames of images (for example, three frames) are input into the model as a video sequence.

[0039] After the above acoustic system is built, the audio signal is acquired using a sampling frequency including but not limited to 51200Hz. Fourier transform can convert complex time and space signals into the frequency domain, but for some non-stationary signals, such as welding sound signals, their frequency domain characteristics change with time, so short-time Fourier transform (STFT) is used to understand the main frequency characteristics of the sound in the local time period. Short-time Fourier transform regards the non-stationary process as the superposition of a series of short-time stationary signals, and the short-time nature can be achieved by windowing in time.

[0040] The formula for short-time Fourier transform is as follows: For discrete signal s, a window function W is used to slide along the direction of s(n) for Fourier transform. In some embodiments, for the acoustic signal collected by the system, the window length is set to 128 and the step length is set to 32, and several 64*64 spectrograms are generated using STFT according to the welding time.

[0041] In order to obtain different penetration states of butt welds at different groove angles, the back side weld width data of butt welds are obtained, and the welding experiments for making the back side weld width data set of butt welds are divided into 4 groups. The experimental parameters are shown in Table 1. In each group, the data of two welds are selected as the training set, and the data of the other weld is selected as the test set, see Table 1.

[0042] Table 1 Experimental parameters

[0043]

[0044]

[0045] The workpieces used in the butt welding experiment are as follows Figure 2 As shown, it is composed of two 304 stainless steel plates with a groove angle of 30° or 45° butt-jointed, with a thickness of 5mm and a blunt edge thickness of 1mm at the groove.

[0046] In order to ensure the accuracy of the back weld width, a handheld 3D scanner was used to scan the bottom of the workpiece after the welding experiment was completed to obtain the following Figure 3 The 3D point cloud data shown in the figure shows that the protruding part in the middle is the back weld formed by the molten pool penetration in the butt welding experiment.

[0047] In order to obtain the weld width value conveniently, the height value perpendicular to the bottom of the workpiece in the point cloud data is used as the gray value, so that the three-dimensional data of the bottom of the workpiece is converted into Figure 4 After that, the appropriate area is intercepted, and the edge features of the back weld are enhanced using grayscale value stretching, Scharr operator and other methods. The contour of the back weld is obtained using threshold segmentation. The results are shown in Figure 5 Finally, according to the number of non-zero pixel values ​​in each column, the coordinates of each pixel location, and the actual world size of a single pixel, the continuous back weld length-width curve is obtained by fitting through Matlab software, as shown in Figure 6 shown.

[0048] In the process of making the data set, it is necessary to ensure that the front image of the molten pool at each moment corresponds to the back weld width value at that moment. Taking the welding start time as the standard point, arc light appears in the front molten pool image at this time, and the back weld length-width curve appears non-zero value. According to the known welding speed and camera acquisition frame rate, the length of the back weld at the corresponding moment of each image can be calculated, and the back weld width corresponding to the front molten pool in each image can be calculated according to the length-width curve.

[0049] according to Predict the back weld width, where n represents the number of images collected by the camera from this moment to the moment of welding, f represents the frame rate of camera image acquisition, v represents the welding speed of the welding robot, and φ is the function of the back weld length-width curve fitting in the butt welding experiment. Based on the above method, each set of video audio data and the corresponding back weld width value data set are prepared. The audio is acquired synchronously with the image data throughout the experiment and is also used as the input of the regression model, which is equally important as the image data. With the dual input of audio and image data, the anti-interference ability of the model can be improved. For example, the image may be affected by the arc light and errors may occur. By adding audio to construct a multimodal model, errors in the results can be avoided and the accuracy of the model can be further improved.

[0050] After step S1 is completed, step S2 is performed to build a regression model based on Swin Transformer to predict the back weld width of the butt weld.

[0051] The overall network structure is as follows Figure 7 As shown in the figure, the image is first input into the Patch Partition module for partitioning, that is, each 4×4 adjacent pixels is a Patch and flattened in the channel dimension; followed by four consecutive hierarchical structure Stages, which are similar to convolutional neural networks; finally, the predicted backside melt width data is obtained through the LayerNorm layer, the pooling layer, and the fully connected layer.

[0052] The four consecutive hierarchical structure stages include hierarchical structure Stage 1, hierarchical structure Stage 2, hierarchical structure Stage 3 and hierarchical structure Stage 4, wherein the hierarchical structure Stage 1 includes a Linear Embedding layer and a Swin Transformer module; the hierarchical structure Stage 2, the hierarchical structure Stage 3 and the hierarchical structure Stage 4 all include a Patch Merging module and a Swin Transformer module.

[0053] The Patch Merging module process is as follows Figure 8 As shown in the figure, for 2×2 adjacent pixels in the feature map, they are separated into 4 feature maps according to their positions and concatenated in the depth dimension. After that, the entire layer is normalized, and then a fully connected layer is used to perform linear changes in the depth dimension, so that the depth becomes twice the original. Through the Patch Merging structure, the network can successively downsample the feature map by 4 times, 8 times, 16 times, etc., which plays a role similar to the pooling layer in the convolutional network to increase the receptive field of the convolution kernel, generate features of multiple sizes, and help special analysis of task targets in the image.

[0054] The Swin Transformer module structure is as follows Fig. 9 As shown in the figure, the W-MSA (Windows Multi-head Self-Attention) module and the SW-MSA (Shifted Windows Multi-Head Self-Attention) module are used in two consecutive units respectively. The W-MSA module divides the entire feature map into multiple windows and performs self-attention calculations in each window instead of directly calculating the entire feature map, which saves a lot of computing power. The total attention calculation amount of W-MSA is Ω(W-MSA)=4hwC 2 +2M 2 hwC, where h, w, and C are the height, width, and depth of the feature map input by the Euclidean module, respectively, and M represents the size of the window. The SW-MSA module solves the problem that information between different windows cannot be communicated by offsetting the window in W-MSA to the right and downward by M / 2 pixels, thus ensuring the global attention of the feature map.

[0055] The back weld width regression model is used to predict the back weld width after the butt welding experiment. The prediction results are as follows: Fig.10As shown in the figure, considering the visualization of prediction accuracy and generalization performance, the MAE (Mean Absolute Error) and RMSE (Root Mean Squared Error) methods are used to evaluate the prediction results of the back weld width regression model. In the test set, the MAE reaches 0.4678mm and the RMSE reaches 0.3517mm. The fitting results of the actual measured values ​​meet the use needs and the accuracy meets the requirements. Therefore, the regression model is used in the subsequent control system to obtain the real-time back weld width value, and the butt welding experiment is carried out to verify it.

[0056] Combined with the introduction of the above prediction method, we can also understand a method for controlling the width of the back weld of butt welding: use a regression model to obtain multiple groups of back weld width prediction values, and use the average value of multiple groups of back weld width prediction values ​​as the input value of the PID algorithm, thereby regulating the welding current and controlling the width of the back weld.

[0057] Due to its advantages of ease of use, good robustness and high reliability, PID (Proportional Integral Differential) control algorithm is widely used in the control field of industrial production. Fuzzy control is suitable for complex situations of nonlinearity and time-varying in the welding process. By adjusting the PID coefficient, it can cope with the changes in welding parameters and achieve better back weld width control effect.

[0058] The PID control algorithm is: Where u(t) represents the controller output signal, e(t) = r(t) - c(t), represents the deviation between the target value r(t) and the input value c(t), and K p Represents the proportionality coefficient, T i represents the integration time constant, T d represents the differential time constant, K i =K p / T i , represents the integral coefficient, K d =K p ×T d , represents the differential coefficient.

[0059] The proportional link in the PID control algorithm can quickly reflect the deviation of the controlled object and play a major control role. The integral link is mainly used to eliminate the static error in the system, that is, the difference between the target value and the measured value of the output when the system control process tends to be stable. The differential link mainly introduces a correction signal through the rate of change of the deviation signal to reduce the adjustment time of the system. Through the combination of the three algorithms, PID can effectively correct the deviation of the controlled object and make it reach a stable target value.

[0060] In one embodiment, the target value r(t) of the back weld width is determined, the above regression model is used to predict the real-time back weld width, and multiple times, such as the average value of the three most recent weld widths, are selected as the input value c(t), and the deviation e(t) and the deviation change rate e are calculated. c (t), through the control coefficient K p , K i , K d The fuzzy control enables the PID controller to regulate the welding current of the welding robot more quickly and stably, and obtain the back weld width close to the target value.

[0061] Different embodiments of the method and test results are described below.

[0062] The first embodiment is based on the GMAW (gas metal arc welding) process, adopts a non-pulsed DC transmission mode, and verifies the effectiveness and generalization ability of the back weld width monitoring and control system by conducting butt welding experiments on workpieces with different groove angles. The parameters of the butt welding experiment are shown in Table 2. Based on the continuous adjustment of the results during the experiment, the initial PID control coefficient K in this experiment p =25, K i =0.1, K d =0.

[0063] Table 2 Welding parameters

[0064]

[0065] The experiment in the first embodiment was conducted on a workpiece with a groove angle of 30° and a starting welding current of 105A for butt welding. To compare the effect of the back weld width control system, the first group did not use the weld width control system, and the second group used the weld width control system, wherein the target value of the back weld width was set to 3mm. The experimental results are shown in FIG. Fig.11 shown.

[0066] Fig.12 The upper middle part shows the front groove of the butt joint of the 30° groove workpiece, the back penetration effect of the first group without control, and the back penetration effect of the second group with control. It is easy to see from the figure that the first group of welds without the control system did not achieve penetration due to the small initial welding current, and its back weld width was always 0mm. The second group of welds using the control system immediately reached the target value of 3mm for the back weld width at the beginning, and then remained stable near the target value, with MAE reaching 0.2603mm and RMSE reaching 0.3673mm with the target value.

[0067] On a workpiece with a groove angle of 45°, a control experiment was conducted with an initial welding current of 130A. The results are as follows: Fig.12shown.

[0068] Fig.13 The control effect of the back weld width is similar to that at 30°. When the control system is not used, the weld of the first group is not fully melted and the back weld width is 0. After the weld width of the second group approaches the target, it fluctuates around the target value. Due to heat accumulation, the average weld width value gradually increases. The MAE with the target value reaches 0.2620mm, and the RMSE reaches 0.3612mm.

[0069] Two groups of butt welding experiments with 30° and 45° grooves showed that the back weld width control system worked well at a constant angle.

[0070] In order to verify the generalization ability of the back weld width control system, the second embodiment increases the complexity of the welding environment by changing the groove angle. A butt welding experiment was conducted on a steel plate with a groove angle suddenly changed from 45° to 30°. The back weld width results are shown in Figure 2. Fig.13 shown.

[0071] In the first group of butt welding experiments without the control system, the weld was not fully melted at the 45° groove. When the groove angle suddenly changed to 30°, the steel in the groove became thinner, and the weld was instantly fully melted, and the average width exceeded 2mm. In the second group of butt welding experiments using the control system, the weld width at the 45° groove section was always close to the target value. After the angle suddenly changed, the weld width on the back side increased significantly due to the influence of heat accumulation and steel thickness, but gradually fell back to the target value under the adjustment of the control system. Due to the short length of the workpiece, the effect was not obvious. The MAE of the second group and the target value reached 0.9739mm, and the RMSE reached 1.1383mm.

[0072] The butt welding experiment was conducted using a workpiece with a groove angle that gradually changed from 30° to 45°, which was a relatively gentle change in the welding environment. The experimental results are as follows Fig.14 shown.

[0073] from Fig.14 It can be seen that although the back weld of the first group of butt welding experiments without the control system was fully melted, its width fluctuated around 2 mm, far from the target value of 3 mm. After the second group used the control strategy, its back weld width stabilized around the target value after two large fluctuations, and its MAE with the target value reached 0.3860 mm and RMSE reached 0.5076 mm.

[0074] Through the above two test results, the back weld width control method of butt welding disclosed in the present invention is stable at the target value for workpieces with a groove of 30° to 45° under the designed control system, which effectively enhances the welding formation and quality.

[0075] Combined with the introduction of the above method, we can also understand a control system, such as Figure 1 As shown, the system includes an image acquisition system 101 , an audio acquisition system 102 , a welding system 103 and a control module 104 .

[0076] The image acquisition system 101 includes a molten pool camera, and the audio acquisition system 102 includes a microphone and an acquisition card; the welding system 103 includes a welding gun 1031 and a welding robot 1032, the welding robot 1032 is connected to the welding gun 1031, the molten pool camera and the microphone are arranged on the welding gun 1031 and move along the welding direction A with the welding gun; the control module 104 is used to send a welding current signal to the welding system through the above method, so as to realize real-time control of the back weld width during butt welding.

[0077] For workpieces with constant groove angles, the back weld width using the above control system reaches the target value in the initial stage of welding and stabilizes around the target value; for workpieces with variable groove angles, the back weld width using the above control system gradually approaches the target value and tends to stabilize after short-term fluctuations and adjustments.

[0078] The present application uses specific words to describe the embodiments of the present application. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or multiple times in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0079] Although the present invention is disclosed as above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting the width of the back weld of a butt weld, characterized in that: The steps include: Collecting molten pool image data and audio data during welding at the same groove angle, as well as the back weld width at the corresponding moments of the molten pool image data and the audio data; The collected audio data is converted into two-dimensional data, and used together with the molten pool image data and the back weld width as input values ​​to construct a regression model for predicting the back weld width using the molten pool image and audio.

2. The method for predicting the width of the butt weld back side weld according to claim 1, characterized in that: The collected audio data is subjected to short-time Fourier transform to generate a spectrogram.

3. The method for predicting the width of the butt weld back side weld according to claim 1, characterized in that: The back weld is obtained by using a 3D scanner; the height value perpendicular to the bottom of the workpiece is used as the grayscale value, and the 3D data obtained by the 3D scanner is converted into a 2D grayscale image; the contour of the back weld is obtained by using threshold segmentation; and a continuous back weld length-width curve is obtained by fitting.

4. The method for predicting the width of the butt weld back side weld according to claim 3, characterized in that: The length of the back weld at the corresponding moment of each image is calculated, and the back weld width corresponding to the front molten pool in each image is obtained according to the length-width curve.

5. The method for predicting the width of the butt weld back side weld according to claim 1, characterized in that: The steps to build a regression model are; Divide the image data into blocks and flatten them in the channel dimension; The features of the image are extracted layer by layer through multiple continuous hierarchical structures; The feature data is normalized and pooled, and the backside melt width prediction value is obtained through the fully connected layer.

6. A method for controlling the width of the back weld of a butt weld, characterized in that: The regression model is obtained by using the method for predicting the width of the butt weld back side weld as described in claims 1-5, The output value of the regression model is used as the input value of the PID algorithm, and the PID algorithm is used to regulate the welding current of the welding process.

7. The method for controlling the width of the butt weld back side weld according to claim 6, characterized in that: The regression model is used to obtain multiple groups of back weld width prediction values, and the average value of the multiple groups of back weld width prediction values ​​is used as the input value of the PID algorithm.

8. The method for controlling the width of the butt weld back side weld according to claim 7, characterized in that: The welding current is adjusted by adjusting the control coefficient of the PID algorithm.

9. Butt welding back weld width control system, characterized in that: The system includes: Image acquisition system, including melt pool camera; Audio acquisition system, including microphone and acquisition card; A welding system, comprising a welding gun and a welding robot, wherein the welding robot is connected to the welding gun, and the molten pool camera and the microphone are arranged on the welding gun and move with the welding gun; A control module, wherein the control module is used to send a welding current signal to the welding system through the method according to any one of claims 6-8.

10. The butt welding back side weld width control system according to claim 9, wherein the welding gun forms an angle of 60° with the welding direction.

Images