An AI algorithm-based pipeline full-position robot intelligent welding system and welding method

Abstract

This invention provides an AI-based intelligent robotic welding system and method for all-position pipeline welding. A laser vision scanning system scans the pipeline weld / groove. An AI intelligent model generation system generates a 3D model of the all-position weld based on the scan data. An AI intelligent process generation system intelligently reads the weld size information from the model, generates the welding torch path for all-position welding, optimizes the welding torch path, matches initial welding parameters, and assigns them to the intelligent welding robot. During welding execution, an AI intelligent process optimization system predicts the weld pool and determines whether the process is qualified. If the process is unqualified, the welding parameters are repeatedly adjusted; if the process is qualified, optimized process parameter values ​​are output. This invention achieves precise control of welding quality for all-position welding of small and medium-diameter pipelines, improving the overall quality of all-position pipeline welding. It intelligently adapts to the complex environment of all-position welding of small and medium-diameter pipelines and can respond to complex welding process changes in real time.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence welding technology, and particularly relates to an intelligent welding system and welding method for all-position robotic pipeline welding based on AI algorithms. Background Technology

[0002] All-position welding of pipelines involves welding in multiple positions, including flat, vertical, horizontal, and overhead welding. This process is challenging and requires high adaptability to complex environments. Currently, commonly used intelligent all-position welding methods typically employ dedicated welding machines or intelligent welding carriages, and are only suitable for large-diameter pipelines. Dedicated welding machines have a stable welding structure, operate along a fixed circular trajectory, and do not involve dynamic changes in the welding torch, resulting in a relatively stable welding process with low complexity, making them suitable for scenarios with low demands for intelligent welding. Intelligent welding carriages use magnetic adsorption to the outer wall of the pipeline, and their welding trajectory is also relatively stable, with minimal impact on the welding torch's movement. Intelligent welding of small and medium-diameter pipelines involves complex process variations and requires high-precision parameters. Currently, there are no all-position welding robots specifically designed for small-diameter pipelines. Existing welding robots for small and medium-diameter pipelines are mainly used for welding pipelines with rotating weld joints. When using robotic arms for all-position welding, the dynamic changes of the robotic arms are complex. Therefore, this invention researches robotic all-position intelligent welding of small and medium-diameter pipelines and develops an AI-based intelligent robotic welding system and method for all-position pipeline welding. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an AI-based intelligent robotic welding system and method for all-position pipeline welding, enabling precise all-position welding of small and medium-diameter pipelines.

[0004] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0005] A welding method for an AI-based all-position robotic intelligent welding system for pipelines includes the following steps:

[0006] Step 1: Move the assembled workpieces to the workpiece fixing support device and fix them in place, then adjust the position of the intelligent welding robot;

[0007] Step 2: Use a laser vision scanning system to perform a 360° scan of the pipe weld / bevel and output the scan results to the computer system;

[0008] Step 3: The AI ​​intelligent model generation system within the computer system generates a 3D model of the pipeline's all-position weld seam based on laser vision scanning data;

[0009] Step 4: The AI ​​intelligent process generation system in the computer system intelligently reads the weld size information based on the three-dimensional model of the all-position weld of the pipeline, automatically generates the welding gun path for all-position welding, and optimizes the welding gun path by rapidly expanding the random number algorithm through AI, using strategies and tree structures to generate welding path planning and initial welding process parameters, which are then assigned to two intelligent welding robots for welding.

[0010] Step 5: During the welding process, the AI ​​intelligent process optimization system in the computer system extracts AI model features based on the information collected by the laser vision scanning system, molten pool monitoring system, and welding parameter acquisition system. Based on the parameter adjustment algorithm of deep learning of the AI ​​model, it outputs the initial values ​​of preliminary process parameter optimization. Based on the static and dynamic feature data of the molten pool, the AI ​​core prediction model is used to predict the weld molten pool.

[0011] Then, based on the original molten pool image, the threshold setting for predicting the molten pool is performed. If the value output by the loss function of the prediction model is lower than the preset qualified value, the process is judged to be unqualified. If the process is judged to be unqualified, the AI ​​model parameter adjustment is repeated. If the process is judged to be qualified, the optimized value of the process parameters is output, and the execution instructions are sent to the two intelligent welding robots respectively.

[0012] Step 6: After the root pass welding is completed, the system intelligently calls the fill / cover welding process to complete the fill / cover welding.

[0013] Furthermore, in step 4, the inverse kinematics calculation of the intelligent welding robot's end effector uses the following formula:

[0014]

[0015] In the formula, This indicates the time of the end effector of the intelligent welding robot. The joint angle, , , , , , Indicates the coefficients for calculating joint motion; , , , , Indicates the movement time of the intelligent welding robot;

[0016] The inverse kinematics calculation formula for the end effector of the intelligent welding robot is optimized as follows:

[0017]

[0018] In the formula, For the optimized joint angle, This is the pseudo-inverse of the Jacobian matrix. For the target end pose, This is the current end pose.

[0019] Furthermore, in step 5, the prediction model is as follows:

[0020] Input: history Step-by-step molten pool state sequence , ;

[0021] Output: Future Step-by-step molten pool state sequence , , , ;

[0022] in, , , They represent history respectively , , The state of the molten pool at any given time; , , Representing the future , , The state of the molten pool at any given moment. Indicates the current time The state of the molten pool at each future time increment, based on the baseline. The length of the molten pool The width of the molten pool The area of ​​the molten pool;

[0023] Predictive head calculation: ;

[0024] in, This is the final output of the decoder. A model representing a linear transformation. The model represents the normalized representation layer;

[0025] Loss function (multi-objective regression):

[0026]

[0027] in, , , These are weighting coefficients for the length, width, and area dimensions, used to balance the contributions of different features to the loss function; For time step, The time step for prediction represents the number of steps to predict forward. for The actual value of the molten pool length at any given time. for The predicted value of the molten pool length at any given time. for The actual value of the molten pool width at any given time. for The predicted value of the melt pool width at any given time. for The true value of the molten pool area at any given time. for The predicted value of the molten pool area at any given time.

[0028] Further, step 3 includes:

[0029] The AI ​​intelligent model generation system is based on the laser data acquisition results of the laser vision scanning system. It uses semantic segmentation model algorithms to learn weld seam image recognition, filters, segments, inverts colors, and thins the laser lines, and then performs AI intelligent processing through sub-pixel algorithms or edge detection technology to accurately extract the pixel coordinates of the laser lines in the image. , ), and then ( , Convert to camera coordinate system coordinates ( , , ):

[0030]

[0031] In the formula, ( , () are the coordinates of the main point. , Focal length;

[0032] Then through rotation matrix Translation vector Transform the camera coordinates to the robot coordinate system:

[0033]

[0034] in,( , , () represents the coordinates of the translation variable;

[0035] Generate a sequence of three-dimensional feature point coordinates and form a three-dimensional model of the pipeline weld in all positions.

[0036] An AI-based intelligent welding system for pipelines in all positions, designed to implement the above-mentioned welding method, includes a workpiece fixing support device for fixing the workpiece to be welded, a gantry support system above the workpiece fixing support device, two intelligent welding robots symmetrically mounted on the gantry support system, and a laser vision scanning system, a molten pool monitoring system, and a welding torch mounted at the end of each intelligent welding robot.

[0037] It also includes an intelligent control cabinet, which is connected to a computer system, a gantry support system, an intelligent welding robot, a welding power source, a laser vision scanning system, a molten pool monitoring system, and a welding torch. Based on the control commands generated by the computer system, the control cabinet is sent to each execution system to realize intelligent all-position welding operation. The computer system contains an intelligent welding process database, an AI intelligent model generation system, an AI intelligent process generation system, and an AI intelligent process optimization system.

[0038] It also includes a welding parameter acquisition system and gas cylinders. The welding parameter acquisition system collects process parameters during the welding process and outputs the acquisition results to the computer system.

[0039] The present invention has the following beneficial effects:

[0040] It achieves precise control of welding quality for all-position welding of small and medium diameter pipes, and improves the quality of all-position welding of pipes; it intelligently adapts to the complex environment of all-position welding of small and medium diameter pipes and can respond to complex welding process changes in real time. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of an all-position robotic intelligent welding system for pipelines.

[0042] Figure 2 Schematic diagram of terminal accessories for intelligent welding robots;

[0043] Figure 3 A schematic diagram of the workflow for an AI intelligent model generation system;

[0044] Figure 4 A schematic diagram of the workflow of an AI-powered intelligent process generation system;

[0045] Figure 5 A schematic diagram of the workflow for an AI-powered intelligent process optimization system.

[0046] In the diagram: 1-Computer system; 2-Intelligent control cabinet; 3-Gantry support system; 4-Intelligent welding robot; 401-Laser vision scanning system; 402-Molten pool monitoring system; 403-Welding torch; 5-Welding power source; 6-Welding parameter acquisition system; 7-Workpiece fixing support device; 8-Workpiece; 9-Gas cylinder. Detailed Implementation

[0047] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0048] like Figure 1 As shown, the intelligent welding system for all-position robotic pipelines based on AI algorithms of the present invention includes a computer system 1, an intelligent control cabinet 2, a gantry support system 3, an intelligent welding robot 4, a DC argon arc welding power supply 5, a welding parameter acquisition system 6, a workpiece fixing support device 7, a gas cylinder 9, a laser vision scanning system 401, a molten pool monitoring system 402, and a welding torch 403.

[0049] like Figure 1 , 2 As shown, the intelligent control cabinet 2 is connected to the computer system 1, the gantry support system 3, the intelligent welding robot 4, the DC argon arc welding power supply 5, the welding torch 403, the weld laser vision scanning system 401, the molten pool monitoring system 402, etc. Based on the control commands generated by the computer system 1, the control cabinet is sent to each execution system to realize intelligent all-position welding operation and realize path planning and process control for all-position welding of pipelines.

[0050] like Figure 1 , 2 As shown, the intelligent welding robot 4 is mounted on the gantry support system 3 and can move up and down and horizontally on the gantry support system 3 via a motor drive for position adjustment. The end of the intelligent welding robot 4 is equipped with accessories such as a laser vision scanning system 401, a weld pool monitoring system 402, and a welding torch 403. The welding power supply 5 is connected to the welding torch 403, providing the welding current and welding voltage during the welding process. The welding parameter acquisition system 6 collects process parameters such as welding current, welding voltage, and gas flow rate during the welding process and outputs the acquisition results to the computer system 1. The laser vision scanning system 401 has the function of emitting and collecting infrared signals before welding, scanning the pipe weld / groove before welding and outputting the scanning results to the computer system 1. The weld pool monitoring system 402 records the weld pool image during the welding process and outputs the weld pool image to the computer system 1. The workpiece fixing support device 7 is used to prevent and fix the workpiece 8 to be welded; it is a conventional structural design.

[0051] Computer system 1 contains an intelligent welding process database, an AI intelligent model generation system, an AI intelligent process generation system, and an AI intelligent process optimization system. Computer system 1 is connected to intelligent control cabinet 2 and issues implementation instructions through intelligent control cabinet 2.

[0052] The intelligent welding process database stores parameters such as welding current, welding voltage, welding speed, gas flow rate, wire feed speed, left and right swing amplitude, and left and right dwell time for welding positions from 0° to 360°, which can be automatically retrieved by the AI ​​intelligent process generation system and the AI ​​intelligent process optimization system.

[0053] The AI ​​intelligent model generation system is based on the laser data acquisition results of the 401 laser vision scanning system. It uses semantic segmentation model algorithm to learn weld seam image recognition, filters, segments, inverts colors, and thins the laser lines, and then performs AI intelligent processing through sub-pixel algorithm or edge detection technology to accurately extract the pixel coordinates of the laser lines in the image.

[0054] The AI-powered intelligent process generation system is based on a 3D model of the pipe's all-position weld seam. It intelligently reads the weld seam size information and automatically generates the welding torch path for all-position welding. Through AI-driven rapid expansion of random number algorithms, it optimizes the adoption strategy and tree structure to perform intelligent analysis of the optimal path for the joint space planning of the intelligent welding robot 4.

[0055] The AI ​​intelligent process optimization system adopts a three-layer penetrating model architecture. During the welding process, it extracts AI model features based on information collected by the laser vision scanning system 401, the molten pool monitoring system 402, and the welding parameter acquisition system 6. Based on the parameter adjustment algorithm of deep learning of the AI ​​model, it outputs the initial values ​​of preliminary process parameter optimization. Based on the static features (material, specifications, bevel angle, joint type, feature vector) and dynamic features (process parameters, molten pool features, time series vector) of the molten pool, the AI ​​core prediction model is used to predict the weld molten pool.

[0056] The welding method of the AI-based all-position robot intelligent welding system for pipelines includes the following processes:

[0057] Step 1: Move the assembled workpiece 8 (i.e., pipe / fitting) onto the workpiece fixing support device 7 and fix it, and adjust the intelligent welding robot 4 to the appropriate position;

[0058] Step 2: Use the laser vision scanning system 401 to perform a 360° scan on the weld / bevel of the pipe / fitting and output the scan results to the computer system 1;

[0059] Step 3: The AI ​​intelligent model generation system generates a 3D model of the pipeline's all-position welds based on laser vision scanning data. The specific process is as follows:

[0060] like Figure 3As shown, the AI ​​intelligent model generation system is based on the laser data acquisition results of the laser vision scanning system 401. It uses a semantic segmentation model algorithm to learn weld seam image recognition, filters, segments, inverts colors, and thins the laser lines, and then performs AI intelligent processing through sub-pixel algorithms or edge detection technology to accurately extract the pixel coordinates of the laser lines in the image. , ), and then ( , Convert to camera coordinate system coordinates ( , , ):

[0061] In the formula, ( , () are the coordinates of the main point. , Focal length;

[0062] Then through rotation matrix Translation vector Transform the camera coordinates to the robot coordinate system:

[0063] in,( , , () represents the coordinates of the translation variable;

[0064] Generate a sequence of three-dimensional feature point coordinates and form a three-dimensional model of the pipeline weld in all positions.

[0065] Step 4: As Figure 4 As shown, the AI ​​intelligent process generation system, based on the three-dimensional model of the pipe all-position weld generated in step 3, intelligently reads the weld size information, automatically generates the welding gun path for all-position welding, optimizes the welding gun path by rapidly expanding the random number algorithm through AI, and performs intelligent analysis of the optimal path for joint space planning, generates welding path planning and initial welding process parameters, and assigns them to two intelligent welding robots 4 to start welding.

[0066] The formula used for calculating the inverse kinematics of the fourth end effector of the intelligent welding robot is as follows:

[0067]

[0068] In the formula, This indicates that the end effector of the intelligent welding robot 4 is in time. The joint angle, , , , , , Indicates the coefficients for calculating joint motion; , , , , This indicates the motion time of the intelligent welding robot.

[0069] The inverse kinematics calculation formula for the four-end effect of the intelligent welding robot is optimized as follows:

[0070]

[0071] In the formula, For the optimized joint angle, This is the pseudo-inverse of the Jacobian matrix. For the target end pose, This is the current end-effector pose;

[0072] The aforementioned end-effector inverse kinematics calculation is the calculation of the motion trajectory of the robotic arm based on the running trajectory of the end of the intelligent welding robot. It is a term for robot kinematics calculation, which plans a smooth motion trajectory of joint angles through polynomial fitting.

[0073] Step 5: As Figure 5 As shown, during the welding process, the AI ​​intelligent process optimization system extracts AI model features based on information collected by the laser vision scanning system 401, the molten pool monitoring system 402, and the welding parameter acquisition system 6. Based on the parameter adjustment algorithm of the AI ​​model's deep learning, it outputs initial values ​​for preliminary process parameter optimization. Based on the static features (material, specifications, bevel angle, joint type, feature vector) and dynamic features (process parameters, molten pool features, time-series vector) of the molten pool, the AI ​​core prediction model predicts the weld molten pool. The prediction model is as follows:

[0074] Input: history Step-by-step molten pool state sequence , ;

[0075] Output: Future Step-by-step molten pool state sequence , , , ;

[0076] in, , , They represent history respectively , , The state of the molten pool at any given time; , , Representing the future , , The state of the molten pool at any given moment. Indicates the current time The state of the molten pool at each future time increment, based on the baseline. The length of the molten pool The width of the molten pool The area is the molten pool area.

[0077] Predictive head calculation: ;

[0078] in, This is the final output of the decoder. A model representing a linear transformation. The model represents the normalized representation layer;

[0079] Loss function (multi-objective regression):

[0080]

[0081] in, , , The weighting coefficients for the three dimensions of length, width, and area are respectively. , , This is used to balance the contributions of different features to the loss function; For time step, The time step for prediction represents the number of steps to predict forward. for The actual value of the molten pool length at any given time. for The predicted value of the molten pool length at any given time. for The actual value of the molten pool width at any given time. for The predicted value of the melt pool width at any given time. for The true value of the molten pool area at any given time. for The predicted value of the molten pool area at any given time.

[0082] The threshold setting for predicting the molten pool is based on the original molten pool image. If the value output by the loss function is lower than the preset qualified value, it is judged as unqualified. If the process is judged as unqualified, the AI ​​model parameter adjustment is repeated. If it is judged as qualified, the optimized value of the process parameter is output for the actuator parameter adjustment, and the execution command is sent to the two intelligent welding robots 4 respectively.

[0083] Step 6: After the root pass welding is completed, the system intelligently calls the fill / cover welding process to complete the fill / cover welding.

[0084] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A welding method for an AI-based all-position robotic intelligent welding system for pipelines, characterized in that, The process includes the following: Step 1: Move the assembled workpiece (8) onto the workpiece fixing support device (7) and fix it, and adjust the position of the intelligent welding robot (4); Step 2: Use a laser vision scanning system (401) to perform a 360° scan on the pipe weld / bevel and output the scan results to the computer system (1). Step 3: The AI ​​intelligent model generation system in the computer system (1) generates a three-dimensional model of the pipeline weld in all positions based on the laser vision scanning data; Step 4: The AI ​​intelligent process generation system in the computer system (1) intelligently reads the weld size information based on the three-dimensional model of the pipe all-position weld, automatically generates the welding gun path for all-position welding, and optimizes the welding gun path by using AI to quickly expand the random number algorithm, adopting strategies and tree structures to generate welding path planning and initial welding process parameters, which are then allocated to two intelligent welding robots (4) for welding. Step 5: During the welding process, the AI ​​intelligent process optimization system in the computer system (1) extracts AI model features based on the information collected by the laser vision scanning system (401), the molten pool monitoring system (402), and the welding parameter acquisition system (6). Based on the parameter adjustment algorithm of deep learning of the AI ​​model, it outputs the initial value of the preliminary process parameter optimization. Based on the static and dynamic feature data of the molten pool, it uses the AI ​​core prediction model to predict the weld molten pool. Then, based on the original molten pool image, the threshold setting for predicting the molten pool is performed. If the value output by the loss function of the prediction model is lower than the preset qualified value, the process is judged to be unqualified. If the process is judged to be unqualified, the AI ​​model parameter adjustment is repeated. If the process is judged to be qualified, the optimized value of the process parameter is output, and the execution command is sent to the two intelligent welding robots (4). Step 6: After the root pass welding is completed, the system intelligently calls the fill / cover welding process to complete the fill / cover welding; In step 4, the inverse kinematics calculation of the end effector of the intelligent welding robot (4) is performed using the following formula: ； In the formula, This indicates that the end effector of the intelligent welding robot (4) is in time The joint angle, , , , , , Indicates the coefficients for calculating joint motion; , , , , This indicates the motion time of the intelligent welding robot (4); The inverse kinematics calculation formula for the end effector of the intelligent welding robot (4) is optimized as follows: ； In the formula, For the optimized joint angle, This is the pseudo-inverse of the Jacobian matrix. For the target end pose, This is the current end-effector pose; In step 5, the prediction model is as follows: Input: history Step-by-step molten pool state sequence , ; Output: Future Step-by-step molten pool state sequence , , , ; in, , , They represent history respectively , , The state of the molten pool at any given time; , , Representing the future , , The state of the molten pool at any given moment. Indicates the current time The state of the molten pool at each future time increment, based on the baseline. The length of the molten pool The width of the molten pool The area of ​​the molten pool; Predictive head calculation: ; in, This is the final output of the decoder. A model representing a linear transformation. The model represents the normalized representation layer; The loss function is as follows: ； in, , , These are weighting coefficients for the length, width, and area dimensions, used to balance the contributions of different features to the loss function; τ is the time step, and τ is the prediction time step, representing the number of steps to predict forward. for The actual value of the molten pool length at any given time. for The predicted value of the molten pool length at any given time. for The actual value of the molten pool width at any given time. for The predicted value of the melt pool width at any given time. for The true value of the molten pool area at any given time. for Predicted value of the molten pool area at any given time; Step 3 includes: The AI ​​intelligent model generation system is based on the laser data acquisition results of the laser vision scanning system (401). It uses a semantic segmentation model algorithm to learn weld seam image recognition, filters, segments, inverts colors, and thins the laser lines, and then performs AI intelligent processing through sub-pixel algorithms or edge detection technology to accurately extract the pixel coordinates of the laser lines in the image. , ), and then ( , Convert to camera coordinate system coordinates ( , , ): ； In the formula, ( , () are the coordinates of the main point. , Focal length; Then through rotation matrix Translation vector Transform the camera coordinates to the robot coordinate system: ； in,( , , () represents the coordinates of the translation variable; Generate a sequence of three-dimensional feature point coordinates and form a three-dimensional model of the pipeline weld in all positions.

2. An AI-based intelligent robotic welding system for pipe welding based on an AI algorithm, used to implement the welding method of claim 1, characterized in that, Includes a workpiece fixing support device (7) for fixing the workpiece (8) to be welded, a gantry support system (3) is set above the workpiece fixing support device (7), and two intelligent welding robots (4) are symmetrically installed on the gantry support system (3). The intelligent welding robots (4) are equipped with a laser vision scanning system (401), a molten pool monitoring system (402), and a welding torch (403) at their ends. It also includes an intelligent control cabinet (2), which is connected to a computer system (1), a gantry support system (3), an intelligent welding robot (4), a welding power source (5), a laser vision scanning system (401), a molten pool monitoring system (402), and a welding torch (403). Based on the control commands generated by the computer system (1), the commands are sent to each execution system to realize intelligent all-position welding operation. The computer system (1) contains an intelligent welding process database, an AI intelligent model generation system, an AI intelligent process generation system, and an AI intelligent process optimization system. It also includes a welding parameter acquisition system (6) and a gas cylinder (9). The welding parameter acquisition system (6) collects the process parameters during the welding process and outputs the collection results to the computer system (1).

Images