Multi-process integrated control system of short tube intelligent group welding robot

Abstract

The application discloses a short pipe intelligent group welding robot multi-process integrated control system and belongs to the technical field of welding robots, which comprises a visual identification module, a motion control module, a parameter compensation module, a central control module and a data conversion and abnormality identification module. The visual identification module is used for collecting short pipe welding image data at different stages of the welding process of the welding robot, including short pipe images before, during and after welding, and transmitting the collected data into the central control module. The motion control module comprises a motion controller, a motor driver and an execution motor, and controls the action of the welding robot by receiving control instructions from the central control module. The short pipe intelligent group welding robot multi-process integrated control system can quickly identify welding abnormalities in the working process of the welding robot, and adjust parameters in different welding stages to optimize the actual welding process of the welding robot and improve the welding quality.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of welding robots, in particular to a multi-process integrated control system of a short pipe intelligent assembly welding robot. BACKGROUND

[0002] In the field of modern industrial manufacturing, short pipe welding, as a key process in the pipeline engineering and mechanical manufacturing industries, directly affects the product performance and service life. With the increasing demand for automation and intelligentization in the manufacturing industry, traditional manual welding methods have been unable to meet the requirements of large-scale and high-precision production due to low efficiency, poor quality stability, and high labor intensity.

[0003] Although short pipe assembly welding by welding robots can improve welding efficiency, due to the small size and large curvature of short pipes, defects such as uneven weld width, misalignment, porosity, and poor continuity may occur during welding. The existing multi-process integrated control system of welding robots lacks real-time detection capability for minor welding defects. Moreover, due to the uneven appearance of welds, it is difficult to quickly locate welding abnormalities and analyze the causes. In addition, different welding parameters during the welding process of welding robots are coupled, and there is a lack of targeted adjustment strategies for welding parameters in different welding stages under welding abnormalities, resulting in the inability to timely correct welding defects and low practicality and functionality. SUMMARY

[0004] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a multi-process integrated control system of a short pipe intelligent assembly welding robot, which improves the detection method and processing method to solve the above technical problems.

[0005] To achieve the above purpose, the present application adopts the following technical solutions:

[0006] The multi-process integrated control system of the short pipe intelligent assembly welding robot includes a visual recognition module, a motion control module, a parameter compensation module, a central control module, and a data conversion and abnormality recognition module.

[0007] The visual recognition module uses an industrial camera to collect short pipe welding image data at different stages of the welding process of the welding robot, including pre-welding, welding, and post-welding short pipe images, and transmits the collected data into the central control module.

[0008] The motion control module includes a motion controller, a motor driver, and an execution motor, which receives control instructions from the central control module to control the action of the welding robot.

[0009] The central control module transmits control instructions into the motion control module based on the input welding program, receives parameter compensation content of the parameter compensation module, performs real-time iterative replacement on the welding parameters in the welding program, receives short pipe welding image data in the visual recognition module, pre-processes the image and transmits it to the data conversion and abnormality recognition module;

[0010] The data conversion and abnormality recognition module performs line conversion processing on the short pipe welding image data obtained in different welding stages of the welding process, identifies short pipe welding abnormalities based on the processed short pipe image data after welding, analyzes the causes of the abnormalities in combination with the image line data of different welding stages, and transmits data to the parameter compensation module.

[0011] The parameter compensation module identifies the parameters to be adjusted based on the short pipe abnormality data obtained by the data conversion and abnormality recognition module, adjusts the parameters to be adjusted based on the historical data large model, and outputs the final parameter compensation result according to the short pipe image data after parameter adjustment.

[0012] Further, the data conversion and abnormality recognition module performs line conversion processing on the short pipe welding image data obtained in different welding stages of the welding process, identifies short pipe welding abnormalities based on the processed short pipe image data after welding, analyzes the causes of the abnormalities in combination with the image line data of different welding stages, and transmits data to the parameter compensation module, including the following steps:

[0013] For the short pipe welding image data obtained in different welding stages of the welding process, edge points in the short pipe welding image data are detected by an edge detection algorithm, the edge points are converted into continuous lines, and features are extracted from the processed short pipe image line data after welding, including weld width, weld length, weld continuity, and weld edge roughness.

[0014] Based on the features in the short pipe image line data after welding, welding abnormalities are identified, the image line data of different welding stages are associated, the time and space relationship of the line data is established, and the causes of the abnormalities are analyzed.

[0015] Further, for the short pipe welding image data obtained in different welding stages of the welding process, edge points in the short pipe welding image data are detected by an edge detection algorithm, the edge points are converted into continuous lines, and features are extracted from the processed short pipe image line data after welding, including weld width, weld length, weld continuity, and weld edge roughness, including the following steps:

[0016] Edge detection algorithms are used to identify edge points in short pipe welding image data obtained at different welding stages during the welding process, including the following steps:

[0017] The Sobel operator is used to calculate the image at... and gradient in direction and The algorithm eliminates edge blurring through nonmaximum suppression and identifies strong edge points through dual threshold detection. The algorithm formula is as follows:

[0018] ;

[0019] ;

[0020] in, Images of short pipe welding at different stages, preprocessed by the central control module. This represents a convolution operation, which further calculates the gradient magnitude and gradient direction, where the gradient magnitude... gradient direction ;

[0021] Identifying strong edge points through dual-threshold detection involves steps based on thresholds. and For gradient magnitude When a comparison is performed, > This represents the current strong edge point, when < This indicates that the current pixel is a non-edge point. ≤ ≤ When a pixel is connected to a strong edge point, it is determined to be a strong edge point; when a pixel is connected to a non-edge point, it is determined to be a non-edge point.

[0022] The Hough transform is used to detect straight lines in an image, connecting edge points into continuous lines, including:

[0023] By analyzing each strong edge point in the image According to discretization Value range calculation obtained Value, of which Further traverse the accumulator array Find points that are greater than a preset threshold; these are the parameters of the detected line. For each line that meets the condition... By reverse deduction, the representation of a straight line in the Cartesian coordinate system can be obtained;

[0024] For each straight line detected by the Hough transform, determine the two endpoints. and , calculate the angle of each straight line , for two straight lines and , calculate the angle difference , while calculating the minimum endpoint distance ;

[0025] Based on the distance threshold and the angle difference threshold, when and meet respectively less than the distance threshold and the angle difference threshold, then the straight lines and are merged into a new straight line.

[0026] Further, the features extracted from the processed post-welding short tube image line data include weld width, weld length, weld continuity, and weld edge roughness, including the following steps:

[0027] Based on the line obtained by Hough transform, the distance between the two boundary lines perpendicular to the weld direction is calculated by determining the boundary line of the weld, and the average value is taken as the final weld width after multiple calculations at different positions of the weld.

[0028] Based on the line obtained by Hough transform, the total distance between all pixel points on the line is calculated along the weld line, and the Euclidean distance between adjacent pixel points is accumulated to obtain the weld length.

[0029] Based on the line obtained by Hough transform, the number of lines is counted as a weld continuity indicator, and when the number of lines exceeds the set threshold, it represents that the current welding weld continuity is poor.

[0030] The distance variance of each pixel point on the edge line and the pixel points in its neighborhood is calculated as the weld edge roughness, and the greater the variance, the rougher the weld edge.

[0031] Further, based on the features in the post-welding short tube image line data, the welding abnormalities are identified, the image line data of different welding stages are associated, the space-time relationship of the line data is established, and the abnormal reasons are analyzed, including the following steps:

[0032] Based on the standard or experience of normal welding, set a threshold for the normal range of each feature, including weld width, weld length, weld continuity, and weld edge roughness, and mark the features outside the threshold range as abnormal features.

[0033] Add a time label to each image and corresponding line data of each welding stage, record their order in the welding process, associate the line data of adjacent welding stages according to the time sequence, judge whether it belongs to different stages of the same weld by comparing the endpoint position and angle difference of the weld line of adjacent stages, judge whether it belongs to different stages of the same weld by comparing the endpoint distance and angle difference of the weld line of adjacent stages, the welding stage includes before welding, during welding and after welding;

[0034] According to the feature combination and space-time variation of the welding abnormality, the welding abnormality reason is output according to the preset rule.

[0035] Further, the short pipe abnormal data obtained by the data transformation and abnormality identification module is combined with the abnormal reason identification to be adjusted parameter, the to-be-adjusted parameter is adjusted based on the historical data large model, and the final parameter compensation result is output according to the short pipe image data after parameter adjustment, including the following steps:

[0036] The short pipe abnormal data is obtained from the data transformation and abnormality identification module, including the abnormal values of weld width, weld length, weld continuity and weld edge roughness, and the corresponding welding stage and related process parameters, wherein the related process parameters are the process parameter values of the input welding program in different welding stages;

[0037] A parameter adjustment large model is established by collecting historical welding data, and parameter adjustment values are generated by combining welding process parameters involved in different abnormal reasons to perform targeted parameter compensation on the current welding program.

[0038] Further, the parameter adjustment large model is established by collecting historical welding data, and parameter adjustment values are generated by combining welding process parameters involved in different abnormal reasons to perform targeted parameter compensation on the current welding program, including the following steps:

[0039] A large amount of historical welding data is collected, including short pipe image data, corresponding process parameters and welding quality evaluation results of normal and different abnormal reason welding conditions, features related to welding quality and process parameters are extracted from the historical data, and a parameter adjustment large model is trained through a neural network;

[0040] The process parameters are input, the welding quality features are output, the mapping relationship between the process parameters and the welding quality is learned by training the model, the initial adjustment range of the to-be-adjusted parameter is set based on the mean and standard deviation of the parameter adjustment historical data under different abnormal reasons based on the 3σ principle, the to-be-adjusted parameter value is input into the trained parameter adjustment large model, the welding quality feature value that may be obtained after adjustment is predicted, and the deviation of the predicted welding quality feature value from the normal range is calculated, including the following steps:

[0041] According to the lower limit of the normal range , upper limit and predicted value , calculate the deviation :

[0042] When < , ;

[0043] When > , ;

[0044] When ≤ ≤ , ;

[0045] Based on the gradient descent method, combined with the deviation adjust the parameters , when > 0 and the deviation is positive, then , when the deviation is negative, then , where represents the learning rate, represents the deviation The gradient of , repeat the above adjustment steps until the predicted welding quality feature value enters the normal range, or the maximum number of iterations is reached;

[0046] Based on the abnormality transmitted by the data transformation and abnormality identification module, the parameter adjustment large model is output, and the welding quality feature value enters the central control module, and the welding program is adjusted and fed back to the motion control module.

[0047] Further, the central control module, based on the input welding program, transmits control instructions into the motion control module, receives the parameter compensation content of the parameter compensation module, and performs real-time iteration replacement on the welding parameters in the welding program, receives the short pipe welding image data in the visual recognition module, and pre-processes the image and transmits it to the data transformation and abnormality identification module, including the following steps:

[0048] Receive the parameter optimization content sent by the parameter compensation module through the communication interface, including the welding parameter name that needs to be adjusted and the adjusted parameter value, find the welding parameter that needs to be adjusted in the data structure of the welding program, and replace the parameter value with the parameter value provided by the parameter compensation module;

[0049] Receive the short pipe welding image data in the visual recognition module for preprocessing, including grayscale processing and denoising.

[0050] Compared with the prior art, the present application has the beneficial effects that:

[0051] 1. In the present application, by line conversion of the identified image of the butt welding of the short pipe in the welding robot working process, the current welding process of the welding robot can be quickly identified for abnormalities, and the causes of the abnormalities can be analyzed based on different welding stages, so that the abnormal parameters can be adjusted accordingly, the purpose of real-time adjustment of the welding of the welding robot is achieved, and the quality of the butt welding of the short pipe group is improved.

[0052] 2. In the present application, by real-time identification of abnormalities in the welding process, including problems such as uneven weld width and poor continuity, welding defects can be found in time, and by targeted adjustment of abnormal parameters, deviations in the welding process can be immediately corrected to avoid further expansion of defects and ensure that the weld quality meets the standard requirements, thereby improving the overall quality of the butt welding of the short pipe group, and by correlating the welds in different stages of welding, the causes of different welding faults and the involved parameters can be comprehensively analyzed, and the welding program of the welding robot can be adjusted accordingly.

[0053] 3. In the present application, through the closed-loop system of the parameter compensation module, the central control module and the data conversion and abnormality identification module, the welding abnormalities can be quickly identified, the parameters can be quickly determined, and the parameters can be adjusted and compensated, and through the real-time feedback and adjustment mechanism, deviations in the welding process can be corrected in time to avoid further expansion of defects and ensure that the welding product quality of the welding robot always meets the standard requirements, thereby enhancing the practicality. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The figure is a block diagram of the multi-process integrated control system of the short pipe intelligent butt welding robot of the present application. DETAILED DESCRIPTION

[0055] The technical solutions of the present application will be described below in conjunction with the embodiments, obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0056] As shown in Figure 1 The multi-process integrated control system of the short pipe intelligent butt welding robot of the present application embodiment includes a visual identification module, a motion control module, a parameter compensation module, a central control module and a data conversion and abnormality identification module.

[0057] The visual identification module is used to collect short pipe welding image data at different stages of the welding robot welding process through an industrial camera, including short pipe images before, during and after welding, and transmits the collected data into the central control module.

[0058] It should be noted that according to the size of the short pipe and the detection accuracy requirement, the industrial camera with appropriate resolution is selected, and generally a high-resolution camera, i.e. an industrial camera with more than 5 million pixels, is used to ensure that the details of the short pipe can be clearly captured.

[0059] The motion control module includes a motion controller, a motor driver and an execution motor, which controls the action of the welding robot by receiving the control instruction of the central control module.

[0060] It should be noted that the motion controller includes multi-axis motion control, which can control multiple joint axes of the robot at the same time to realize complex motion trajectories, and supports multiple interpolation methods such as linear interpolation and circular interpolation to realize smooth motion trajectories. According to the input path point and speed requirement, the motion parameters of each axis are calculated, and the motor driver needs to select a motor driver with appropriate power according to the load and motion requirement of the robot joint axis.

[0061] The central control module transmits the control instruction into the motion control module based on the input welding program, receives the parameter compensation content of the parameter compensation module, and iteratively replaces the welding parameters in the welding program in real time, receives the short pipe welding image data in the visual recognition module, pre-processes the image and transmits it to the data conversion and abnormality recognition module, including the following steps:

[0062] The parameter optimization content sent by the parameter compensation module is received through the communication interface, including the welding parameter name to be adjusted and the adjusted parameter value. The welding parameter to be adjusted is found in the data structure of the welding program, and the parameter value is replaced with the parameter value provided by the parameter compensation module;

[0063] The short pipe welding image data in the visual recognition module is received for pre-processing, including grayscale processing and denoising.

[0064] It should be noted that the grayscale processing of the short pipe welding image can reduce the color information in the image, which is convenient for subsequent line conversion of the image, and the formula is:

[0065] ;

[0066] In the above formula, is the gray value, represents the red color in the original image, represents the green color in the original image, The short tube welding image data received is denoised to further improve the quality and clarity of the image. The denoising method is mean filtering. For each pixel point in the short tube welding image, the average value of all pixel points in the neighborhood of the pixel point is taken as the new value of the pixel point.

[0067] In this embodiment, the data transformation and anomaly identification module processes the short tube welding image data obtained in different welding stages of the welding process by line transformation, identifies the short tube welding anomaly based on the processed short tube image data after welding, obtains short tube anomaly data, analyzes the causes of the anomaly in combination with the image line data of different welding stages, and transmits the data to the parameter compensation module, including the following steps:

[0068] The edge points in the short tube welding image data are detected by an edge detection algorithm, the edge points are converted into continuous lines, and features are extracted from the processed short tube image line data after welding, including weld width, weld length, weld continuity, and weld edge roughness, including the following steps:

[0069] The edge points in the short tube welding image data obtained in different welding stages of the welding process are identified using an edge detection algorithm, including the following steps:

[0070] The Sobel operator is used to calculate the gradients and of the image in the and directions, respectively, the edge blur is eliminated by non-maximum suppression, and strong edge points are identified by double-threshold detection, and the algorithm formula is:

[0071] ;

[0072] ;

[0073] wherein, represents the short tube welding image of different welding stages preprocessed by the central control module, represents a convolution operation, and the gradient amplitude and gradient direction are further calculated, wherein the gradient amplitude , and the gradient direction ;

[0074] It should be noted that non-maximum suppression is applied to the gradient magnitude in the gradient direction, retaining only local maxima and eliminating edge blurring. Specifically, the gradient magnitude of each pixel is compared with the gradient magnitudes of its two adjacent pixels in the gradient direction. If the gradient magnitude of the pixel is not the largest, it is set to 0. By using an edge detection algorithm, edge points in the obtained image are identified. By judging strong edge points, it is easier to perform line processing on the edge points in the subsequent process, which reduces the amount of computation and improves the response speed in the actual short pipe intelligent welding process.

[0075] Identifying strong edge points through dual-threshold detection involves steps based on thresholds. and For gradient magnitude When a comparison is performed, > This represents the current strong edge point, when < This indicates that the current pixel is a non-edge point. ≤ ≤ When a pixel is connected to a strong edge point, it is determined to be a strong edge point; when a pixel is connected to a non-edge point, it is determined to be a non-edge point.

[0076] The Hough transform is used to detect straight lines in an image, connecting edge points into continuous lines, including:

[0077] By analyzing each strong edge point in the image According to discretization Value range calculation obtained Value, of which Further traverse the accumulator array Find points that are greater than a preset threshold; these are the parameters of the detected line. For each line that meets the condition... By working backwards, we can obtain the representation of a straight line in the Cartesian coordinate system;

[0078] It should be noted that, It is the perpendicular distance from the origin to the line. This is the angle between the perpendicular line and the positive x-axis, expressed in radians or degrees, typically ranging from 0 to π. The preset threshold needs to be manually adjusted based on image complexity and noise levels, such as to determine whether true straight lines are missed or false noise is detected. The parameters of the detected straight line in polar coordinates;

[0079] For each straight line detected by the Hough transform, determine the two endpoints. and Calculate the angle of each straight line. For two straight lines and Calculate the angle difference Simultaneously calculate and obtain the minimum endpoint distance. ;

[0080] It should be noted that the minimum distance between the endpoints That is, calculate the straight lines separately. Two endpoints to Distance and straight line Two endpoints to The minimum distance between the endpoints is taken as the minimum distance between the endpoints. By measuring the spatial proximity of the two lines, it is shown that the endpoints of the two lines are almost touching or overlapping, and it is determined whether the two lines are the same line. By merging the lines that pass the determination, the accuracy of the system in actual use is improved.

[0081] Based on the distance threshold and the angle difference threshold, when and If both the distance threshold and the angle difference threshold are met simultaneously, then the straight line will be... and They merge into a new straight line.

[0082] It should be noted that the distance threshold and angle difference threshold require performing a Hough transform on a batch of short pipe welding images to obtain a large number of straight lines. The distance and angle difference between each pair of straight lines are calculated, and the thresholds are set based on a combination of the mean and standard deviation using an empirical method. A large number of straight line segments are detected by performing a Hough transform on a representative batch of short pipe welding images as raw data. The distance and angle difference between all possible pairs of straight lines are then calculated. and This yields the distance distribution and angle difference distribution, and the threshold is obtained by combining the mean and standard deviation with the 3sigma principle.

[0083] Features are extracted from the processed post-weld short pipe image line data, including weld width, weld length, weld continuity, and weld edge roughness, including the following steps:

[0084] Based on the lines obtained by Hough transform, the boundary lines of the weld are determined, the distance between the two boundary lines in the direction perpendicular to the weld is calculated, multiple calculations are performed at different positions of the weld, and then the average value is taken as the final weld width.

[0085] Based on the lines obtained by Hough transform, the total distance between all pixels on the weld line is calculated, and the weld length is obtained by accumulating the Euclidean distance between adjacent pixels.

[0086] Based on the lines obtained by Hough transform, the number of lines is counted as an indicator of weld continuity. When the number of lines exceeds a set threshold, it indicates that the current weld continuity is poor.

[0087] It should be noted that the more the number of lines, the more the breakpoints of the weld exist, and the threshold needs to be set based on the current welding index parameter combined with the experience method;

[0088] For example, the vertical distance between the boundary lines is measured at the middle, left end and right end of the weld, respectively, to obtain A1, A2 and A3 as 12.5px, 11.8px and 13.2px, respectively, and the average weld width is (12.5px+11.8px+13.2px) / 3=12.5px, combined with the calibration scale 1px=0.1mm, the current average weld width is calculated to be 1.25mm, which is within the normal range of 1-1.5mm, representing that the current weld is normal. The response speed in actual use can be improved by calculating the simplified lines, so as to achieve the effect of timely response and rapid judgment.

[0089] The distance variance of each pixel point on the edge line and the pixel points in its neighborhood is calculated as the welding edge roughness. The larger the variance, the rougher the weld edge.

[0090] It should be noted that the specific calculation method of the welding edge roughness is that for each pixel point on the edge line, the pixel points in its neighborhood are taken, and the distance of these pixel points to the point is calculated, so as to calculate the distance variance.

[0091] Based on the features in the short tube image line data after welding, welding abnormalities are identified, the image line data of different welding stages are associated, the space-time relationship of the line data is established, and the abnormal reasons are analyzed, including the following steps:

[0092] Based on the standard or experience of normal welding, a threshold of normal range is set for each feature, including weld width, weld length, weld continuity, and weld edge roughness. Features not within the threshold range are marked as abnormal features;

[0093] A time label is added to each image and corresponding line data of each welding stage to record the sequence in the welding process. The line data of adjacent welding stages are associated according to the time sequence, the end point position and angle difference of the weld line of adjacent stages are compared, and it is judged whether it belongs to different stages of the same weld. The weld with an end point distance and an angle difference less than the distance threshold and the angle difference threshold is determined as different stages of the same weld. The welding stages include before welding, during welding and after welding;

[0094] It should be noted that the values of the distance threshold and the angle difference threshold are consistent with the distance threshold and the angle difference threshold when determining whether it is the same straight line.

[0095] According to the feature combination and space-time variation of the welding anomaly, the welding anomaly reason is output according to the preset rule.

[0096] It should be noted that by using data mining algorithms, including decision trees, association rule mining, etc., a large amount of welding data is analyzed to find the potential relationship between abnormal features and abnormal reasons, and a preset rule is established, which includes "if the same weld seam suddenly narrows in different stages and the weld edge roughness increases, it is caused by too small welding current", "if the same weld seam has poor weld continuity and insufficient weld length in different stages, it is caused by too fast welding speed or welding interruption", etc. The final abnormal reason is output based on the abnormal data of different stages of the current weld.

[0097] The parameter compensation module, based on the short pipe abnormal data obtained by the data transformation and abnormality identification module, identifies the parameters to be adjusted based on the abnormal reasons, adjusts the parameters to be adjusted based on the historical data large model, and outputs the final parameter compensation result based on the short pipe image data after parameter adjustment, including the following steps:

[0098] The short pipe abnormal data is obtained from the data transformation and abnormality identification module, including the abnormal values of weld width, weld length, weld continuity, weld edge roughness, and the corresponding welding stages and related process parameters, wherein the related process parameters are the process parameter values of the recorded welding program in different welding stages;

[0099] A parameter adjustment large model is established by collecting historical welding data, and parameter adjustment values are generated based on the welding process parameters involved in different abnormal reasons, to perform targeted parameter compensation on the current welding program, including the following steps:

[0100] A large amount of historical welding data is collected, including short pipe image data, corresponding process parameters and welding quality evaluation results of normal and different abnormal reason welding conditions, and features related to welding quality and process parameters are extracted from the historical data. The parameter adjustment large model is trained through neural network.

[0101] The process parameters are input, and the welding quality features are output. The mapping relationship between the process parameters and the welding quality is learned by training the model. Based on the mean and standard deviation of the parameter adjustment historical data under different abnormal reasons, the initial adjustment range is set for the parameter to be adjusted based on the 3σ principle. The current parameter to be adjusted is input into the trained parameter adjustment large model, the welding quality feature value that may be obtained after adjustment is predicted, and the deviation of the predicted welding quality feature value from the normal range is calculated, including the following steps:

[0102] According to the lower limit , the upper limit and the predicted value , the deviation is calculated :

[0103] When < , ;

[0104] When > , ;

[0105] When ≤ ≤ , ;

[0106] Based on the gradient descent method, the deviation is adjusted to the front to be adjusted parameter , when > 0 and the deviation is positive, then , when the deviation is negative, then , wherein represents the learning rate, represents the deviation about The gradient is repeated until the predicted welding quality feature value enters the normal range, or the maximum number of iterations is reached.

[0107] It should be noted that if the new welding quality feature meets the normal range requirement, the final adjusted parameter is output as the final parameter compensation result, if it does not meet the requirement, the initial adjustment range or learning rate may need to be adjusted again, and the parameter adjustment iteration is performed again until a satisfactory result is obtained.

[0108] Based on the data conversion and abnormality identification module transmission, the abnormality is output through the parameter adjustment large model, and the welding quality feature value enters the central control module, and the welding program is adjusted and fed back to the motion control module.

[0109] The short pipe intelligent group welding robot multi-process integrated control system, when in use, the line conversion is carried out on the recognition image of the short pipe butt welding in the working process of the welding robot, so as to quickly identify the abnormality in the welding process of the current welding robot, and analyze the abnormal reason based on different welding stages, so as to adjust the abnormal parameter, and achieve the purpose of real-time adjustment of the welding of the welding robot, and improve the quality of the short pipe group welding.

[0110] Through the closed loop system of the parameter compensation module, the central control module and the data conversion and abnormality identification module, the welding abnormality is quickly identified, the parameters are quickly determined, and the parameters are adjusted and compensated, through the real-time feedback and adjustment mechanism, the deviation in the welding process can be corrected in time, the defects are avoided from being further expanded, the welding robot welding product quality is ensured to always meet the standard requirements, and the practicality is enhanced.

[0111] In the embodiments of the present application, it should be understood that the disclosed device, apparatus and method can be implemented in other ways. For example, the above-described device embodiments are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, another division manner can be used. The modules described as separate components can or can not be physically separate. The components displayed as modules can or can not be physical units. That is, they can be located in one place or distributed on a plurality of network units. According to actual needs, some or all of the modules can be selected to achieve the purpose of the method of the embodiments.

[0112] The above embodiments are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A multi-process integrated control system for a short-pipe intelligent welding robot, characterized in that: It includes a visual recognition module, a motion control module, a parameter compensation module, a central control module, and a data conversion and anomaly recognition module; The visual recognition module uses an industrial camera to collect short pipe welding image data at different stages of the welding robot welding process, including short pipe images before welding, during welding and after welding, and transmits the collected data to the central control module. The motion control module includes a motion controller, a motor driver, and an actuator motor. It controls the welding robot's movements by receiving control commands from the central control module. The central control module, based on the input welding program, transmits control commands to the motion control module, receives parameter compensation content from the parameter compensation module, performs real-time iterative replacement of welding parameters in the welding program, receives short pipe welding image data from the visual recognition module, preprocesses the images, and transmits them to the data conversion and anomaly recognition module. The data conversion and anomaly identification module performs line conversion processing on the short pipe welding image data obtained at different welding stages during the welding process. Based on the processed short pipe welding image data, it identifies short pipe welding anomalies to obtain short pipe anomaly data. Combining the image line data at different welding stages, it analyzes the causes of anomalies and transmits the data to the parameter compensation module. The parameter compensation module, based on the short tube anomaly data obtained by the data conversion and anomaly identification module, identifies the parameters to be adjusted by combining the anomaly cause identification, adjusts the parameters to be adjusted based on the historical data large model, and outputs the final parameter compensation result based on the short tube image data after parameter adjustment, including the following steps: Short pipe anomaly data is obtained from the data conversion and anomaly identification module, including anomaly values ​​of weld width, weld length, weld continuity, and weld edge roughness, as well as the corresponding welding stage and related process parameters. The related process parameters are the process parameter values ​​of the welding program entered at different welding stages. By collecting historical welding data to establish a large-scale parameter adjustment model, and combining the welding process parameters involved in different anomalies, parameter adjustment values ​​are generated to perform targeted parameter compensation for the current welding program, including the following steps: Collect a large amount of historical welding data, including short pipe image data of normal and different abnormal welding conditions, corresponding process parameters and welding quality assessment results. Extract features related to welding quality and process parameters from historical data, and train a large model with parameter adjustment through neural network. Using process parameters as input and welding quality characteristics as output, the training model learns the mapping relationship between process parameters and welding quality. Based on the historical data mean and standard deviation of parameters under different abnormal causes, the model sets an initial adjustment range for the parameters to be adjusted based on the 3σ principle. The current value of the parameter to be adjusted is input into the trained parameter adjustment model to predict the possible welding quality characteristic values ​​after adjustment. The deviation between the predicted welding quality characteristic values ​​and the normal range is calculated, including the following steps: According to the lower limit of the normal range Upper limit and predicted values Calculate the deviation : when < hour, ; when > hour, ; when ≤ ≤ hour, ; Based on gradient descent, combined with bias For the parameters to be adjusted Make adjustments when When >0 and the deviation is positive, then When the deviation is negative, then ,in Represents the learning rate. Representative bias about The gradient is then adjusted, and the above adjustment steps are repeated until the predicted welding quality feature value enters the normal range or the maximum number of iterations is reached. Based on the anomalies transmitted by the data conversion and anomaly identification module, the welding quality characteristic values ​​are output to the central control module through parameter adjustment of the large model, and the welding program parameters are adjusted and fed back to the motion control module.

2. The multi-process integrated control system for the short pipe intelligent assembly welding robot according to claim 1, characterized in that: The data conversion and anomaly identification module performs line conversion processing on the short pipe welding image data obtained at different welding stages during the welding process. Based on the processed post-weld short pipe image data, it identifies short pipe welding anomalies to obtain short pipe anomaly data. Combining the image line data from different welding stages, it analyzes the causes of the anomalies and transmits the data to the parameter compensation module. This includes the following steps: For short pipe welding image data obtained at different welding stages during the welding process, edge points in the short pipe welding image data are detected by an edge detection algorithm, the edge points are converted into continuous lines, and features are extracted from the processed welded short pipe image line data, including weld width, weld length, weld continuity, and weld edge roughness. Based on the features in the image line data of the short pipe after welding, welding anomalies are identified, image line data of different welding stages are correlated, the spatiotemporal relationship of the line data is established, and the causes of anomalies are analyzed.

3. The multi-process integrated control system for the short pipe intelligent assembly welding robot according to claim 2, characterized in that: The method involves taking short pipe welding image data obtained at different welding stages during the welding process, detecting edge points in the short pipe welding image data using an edge detection algorithm, converting the edge points into continuous lines, and extracting features from the processed welded short pipe image line data, including weld width, weld length, weld continuity, and weld edge roughness, including the following steps: Edge detection algorithms are used to identify edge points in short pipe welding image data obtained at different welding stages during the welding process, including the following steps: The Sobel operator is used to calculate the image at... and gradient in direction and The algorithm eliminates edge blurring through nonmaximum suppression and identifies strong edge points through dual threshold detection. The algorithm formula is as follows: ； ； in, Images of short pipe welding at different stages, preprocessed by the central control module. This represents a convolution operation, which further calculates the gradient magnitude and gradient direction, where the gradient magnitude... gradient direction ; Identifying strong edge points through dual-threshold detection involves steps based on thresholds. and For gradient magnitude When a comparison is performed, > This indicates that the current pixel is a strong edge point. < This indicates that the current pixel is a non-edge point. ≤ ≤ When a pixel is connected to a strong edge point, it is determined to be a strong edge point; when a pixel is connected to a non-edge point, it is determined to be a non-edge point. The Hough transform is used to detect straight lines in an image, connecting edge points into continuous lines, including: By analyzing each strong edge point in the image According to discretization Value range calculation obtained Value, of which , It is the perpendicular distance from the origin to the line. It is the angle between the line containing the perpendicular distance from the origin to the line and the positive x-axis, ranging from 0 to π. Further iterations are made by traversing the accumulator array. Find points that are greater than a preset threshold; these are the parameters of the detected line. For each line that meets the condition... By working backwards, we can obtain the representation of a straight line in the Cartesian coordinate system; For each straight line detected by the Hough transform, determine the two endpoints. and Calculate the angle of each straight line. For two straight lines and Calculate the angle difference Simultaneously calculate and obtain the minimum endpoint distance. ; Based on the distance threshold and the angle difference threshold, when and If both the distance threshold and the angle difference threshold are met simultaneously, then the straight line will be... and They merge into a new straight line.

4. The multi-process integrated control system for the short pipe intelligent assembly welding robot according to claim 3, characterized in that: The step of extracting features from the processed post-weld short pipe image line data, including weld width, weld length, weld continuity, and weld edge roughness, includes the following steps: Based on the lines obtained by Hough transform, the boundary lines of the weld are determined, the distance between the two boundary lines in the direction perpendicular to the weld is calculated, multiple calculations are performed at different positions of the weld, and then the average value is taken as the final weld width. Based on the lines obtained by Hough transform, the total distance between all pixels on the weld line is calculated, and the weld length is obtained by accumulating the Euclidean distance between adjacent pixels. Based on the lines obtained by Hough transform, the number of lines is counted as an indicator of weld continuity. When the number of lines exceeds a set threshold, it indicates that the current weld continuity is poor. The roughness of the weld edge is determined by calculating the variance of the distance between each pixel on the edge line and its neighboring pixels. The larger the variance, the rougher the weld edge.

5. The multi-process integrated control system for the short pipe intelligent assembly welding robot according to claim 3, characterized in that: The method of identifying welding anomalies based on features in the post-weld short pipe image line data, associating image line data from different welding stages to establish spatiotemporal relationships among the line data, and analyzing the causes of anomalies includes the following steps: Based on the standards or experience of normal welding, a threshold for the normal range is set for each feature, including weld width, weld length, weld continuity, and weld edge roughness. Features that are not within the threshold range are marked as abnormal features. Add time stamps to the images and corresponding line data of each welding stage to record their sequence in the welding process. Associate the line data of adjacent welding stages according to the time sequence. By comparing the endpoint positions and angle differences of weld lines in adjacent stages, determine whether they belong to different stages of the same weld. Welds with endpoint distances and angle differences less than distance thresholds and angle difference thresholds respectively are determined to be different stages of the same weld. Welding stages include before welding, during welding, and after welding. Based on the characteristic combination and spatiotemporal changes of welding anomalies, the cause of the welding anomaly is output according to preset rules.

6. The multi-process integrated control system for the short pipe intelligent assembly welding robot according to claim 3, characterized in that: The central control module, based on the input welding program, transmits control commands to the motion control module, receives parameter compensation content from the parameter compensation module, iteratively replaces welding parameters in the welding program in real time, receives short pipe welding image data from the visual recognition module, preprocesses the images, and transmits them to the data conversion and anomaly recognition module, including the following steps: The system receives parameter optimization information from the parameter compensation module via the communication interface, including the names of the welding parameters that need to be adjusted and the adjusted parameter values. It then searches for the welding parameters that need to be adjusted in the data structure of the welding program and replaces the parameter values ​​with those provided by the parameter compensation module. The short pipe welding image data received from the visual recognition module is preprocessed, including grayscale processing and noise reduction processing.

Images