Weld joint automatic identification and self-adaptive welding system and method for large non-standard parts

Abstract

The invention discloses a welding seam automatic identification and self-adaptive welding system and method.The welding seam automatic identification and self-adaptive welding system comprises a movable fixing frame, a welding positioner, a welding robot and a central control unit, and a global wide-area 3D scanner is arranged on the movable fixing frame and used for obtaining the overall point cloud of a workpiece; a structured light vision sensor is arranged on the welding robot and is used for finely scanning a welding seam area; the central control unit is in communication connection with the global wide-area 3D scanner, the welding robot, the structured light vision sensor and the welding power source and used for scheduling the sequence of global scanning and local fine scanning. According to the method, a multi-stage collaborative visual sensing framework of global coarse positioning and local fine recognition is adopted, the contradiction between a large range and high precision is solved, meanwhile, dynamic matching of welding parameters and equipment movement is achieved through finite element simulation and real-time monitoring data closed loop, and full-process automatic quality control is achieved in large weldments for the first time.

Description

Technical Field

[0001] This invention relates to a welding technology, and more particularly to an automatic weld seam identification and adaptive welding system and method for large non-standard parts. Background Technology

[0002] Currently, traditional manual welding and semi-automatic welding methods generally have the following problems when welding large structural components (such as engineering machinery frames): Non-standard: The workpieces are large in size, irregular in shape, few in batches, and have low repeatability, making it difficult to use traditional automated welding equipment designed for standardized workpieces.

[0003] Reliance on manual labor: It mainly relies on experienced welders to visually identify the weld seam location and perform manual operations, which is labor-intensive, has low production efficiency, and the welding quality is greatly affected by human factors.

[0004] Insufficient consistency in welding quality: Manual operation is significantly affected by skill level, resulting in large fluctuations in weld quality and a defect rate (such as porosity and undercut) as high as 15%-20%, and it is difficult to ensure the consistency of welds with complex shapes.

[0005] Low welding efficiency: The welding time for a single piece is more than 50% longer than that of automated methods, and the labor intensity is high. The average daily workload of a single workstation is only 1 / 3 of that of automated equipment.

[0006] Initial limitations of automation: Existing teach-and-playback welding robots require precise pre-programming and positioning, and adjusting the trajectory for non-standard parts takes a very long time. They cannot adapt to minor deformations, assembly errors, or thermal deformations of the workpiece.

[0007] Limitations of simple vision systems: Some systems use laser vision sensors for weld seam tracking, but they are usually limited to tracking, i.e., fine-tuning after the general path is known, and lack initial seeking ability. They are also insufficient for handling complex weld seams (such as staggered weld seams and irregular bevels). Summary of the Invention

[0008] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a reasonable design, highly adaptive and efficient automatic weld seam identification and adaptive welding system and method for large non-standard parts.

[0009] The technical solution of this invention is: An automatic weld seam recognition and adaptive welding system for large non-standard parts includes a mobile fixed frame, a welding positioner, a welding robot, and a central control unit. The mobile fixed frame is equipped with a global wide-area 3D scanner to acquire the overall point cloud of the workpiece. The welding robot is equipped with a structured light vision sensor to perform fine scanning of the weld seam area. The central control unit is communicatively connected to the global wide-area 3D scanner, the welding robot, the structured light vision sensor, and the welding power supply. It is used to schedule the order of global scanning and local fine scanning. The central control unit receives detection signals from the global wide-area 3D scanner and the structured light vision sensor, uses a RandLA-Net deep learning semantic segmentation model to identify the weld seam type, start point, and end point from the overall point cloud, generates a welding path, and controls the welding robot to weld along the preset path. Simultaneously, based on the bevel size and root gap obtained from the fine scanning, it matches welding process parameters in real time to complete the overall welding.

[0010] Furthermore, the welding process parameters are: current of 150-300A, voltage of 20-30V, and welding speed of 20-50 cm / min.

[0011] Furthermore, the structured light vision sensor tracks the weld seam in real time at a frequency of 5-10Hz during the welding process and feeds back the deviation to the welding robot to achieve closed-loop control.

[0012] Furthermore: The global wide-area 3D scanner is a laser line scan sensor with a scanning field of view ≥2m×2m and a point cloud density ≥0.5 mm.

[0013] Furthermore, the process parameters include three sets of current-voltage-speed triplets corresponding to the root gap g≤1.0 mm, 1.0 mm<g≤2.0 mm, and g>2.0 mm. For every 0.1 mm increase in gap, the current increases by 5-8 A.

[0014] Furthermore: the mobile fixed frame is a gantry frame, overhead crane, or the first axis of a large robot, and the welding robot is an industrial robot with six or more degrees of freedom, possessing sufficient working range and flexibility.

[0015] An automatic weld seam identification and adaptive welding method for large non-standard parts, utilizing the aforementioned automatic weld seam identification and adaptive welding system, includes the following steps: S1. Loading and Positioning: Hoist the large non-standard parts to the work area and fix them on the welding positioner, while moving the mobile fixing frame above the welding positioner. S2. Global Scan and Preliminary Identification: The global wide-area 3D scanning unit is activated to scan the surface of the large non-standard part, obtain the overall point cloud of the workpiece, and the central control unit receives the data information to preliminarily identify the approximate location and sequence of multiple weld seams to be welded. S2. Path planning and ranking: The central control unit uses the RandLA-Net model to perform semantic segmentation on the overall point cloud to obtain weld seam point cloud clusters. The robot's collision-free movement path is generated by RANSAC fitting, and the path is ranked by weights of weld seam length, accessibility, and thermal deformation accumulation. S4. Local fine scanning and feature extraction: The welding robot moves to the starting point of the target weld according to the sorted path, and starts the structured light vision sensor at the end to perform fine scanning to obtain the bevel angle α, root gap g, and misalignment amount Δh. S5. Adaptive parameter matching: Based on the extracted features, with a gap width of 1.2±0.3mm, the final welding path and real-time process parameters are generated: current 220A, voltage 24.5V, and speed 35cm / min. S6. Welding Execution: The central control unit issues a command, and the welding robot performs welding. At the same time, the structured light vision sensor tracks in real time and feeds back the deviation to the robot for closed-loop control. S7 Welding complete: The system saves welding process parameters, actual trajectory, point cloud and images to form a digital weld file.

[0016] Further: In step S5, if g ≤ 1.0 mm, single-pass welding is selected, and the thin bevel group in the process parameter library is called: current 180-220 A, voltage 22-26 V, speed 30-45 cm / min; if 1.0 mm < g ≤ 2.0 mm, double-pass welding of root pass + cover pass is selected, with root pass parameters: current 160-200 A, voltage 21-25 V, speed 25-35 cm / min, and cover pass parameters: current 200-240 A, voltage 23-27 V, speed 30-40 cm / min; if g > 2.0 mm or Δh > 1 mm, root pass, fill pass, and cover pass are executed sequentially. The fill pass uses oscillating welding with an oscillation frequency of 2-4 Hz, an oscillation amplitude of 2-6 mm, and a dwell time of 0.1-0.3 s at both ends; the system generates the final TCP trajectory based on the above matching results and sends the current, voltage, speed, and oscillation mode to the welding power source.

[0017] Further: In step S6, the welding robot initiates the arc according to the preset trajectory. The structured light vision sensor measures the actual gap g′ at the root in real time at a frequency of 5-10Hz and compares it with the set value g in step S4: when |g′-g|≤0.3 mm, the original parameters are maintained; when |g′-g|>0.3 mm, the welding speed v′=v×(g / g′) is corrected online, and the current ±10 A and voltage ±1 V are adjusted synchronously to achieve adaptive follow-up.

[0018] Further: In step S3, taking the center of the welding robot's base as the starting point, the obstacle avoidance area is the global point cloud convex hull expanded by 100mm from the start point to the end point of the weld.

[0019] The beneficial effects of the invention are: 1. This invention adopts a multi-level collaborative visual sensing architecture of global coarse positioning and local fine recognition, which solves the contradiction between large range and high precision. At the same time, through finite element simulation and real-time monitoring data closed loop, it realizes dynamic matching between welding parameters and equipment movement, and achieves full-process automated quality control for the first time in large welded parts.

[0020] 2. This invention applies deep learning semantic segmentation algorithms to the 3D point cloud recognition of non-standard weld seams, especially for extracting features of complex and irregular weld seams.

[0021] 3. This invention adopts a deep integration and data interaction method among the vision system, robot system and welding system. In particular, the adaptive control strategy that directly generates welding process parameters based on real-time visual features does not require precision fixtures and a large amount of pre-programming, and can quickly adapt to welding tasks of large non-standard parts of different shapes and sizes.

[0022] 4. This invention eliminates quality fluctuations caused by workpiece processing errors, assembly errors, and thermal deformation through visual sensing and adaptive control, ensuring good weld formation consistency.

[0023] 5. This invention achieves full automation from identification to welding, significantly reducing production preparation and actual operation time, and lowering the demand for highly skilled workers.

[0024] 6. This invention upgrades from "invisible" teaching and reproduction to "understandable" active recognition, solving the core pain point of non-standard parts automation. Furthermore, the entire process data can be recorded and traced, providing basic data support for intelligent manufacturing and digital factories.

[0025] 7. The invention has a complex weld identification time of <100ms, which meets the production line cycle requirements; and the welding time for a single large welded part is shortened from 24h to 15h, increasing annual production capacity by more than 30%; at the same time, the first-pass yield rate of welds is increased from 80% for manual welding to over 98%, and the proportion of X-ray flaw detection level III and above reaches 95%.

[0026] 8. This invention integrates multi-scale features with industrial camera calibration technology to solve the problem of weld seam recognition accuracy for large components at long distances and with a large field of view. It reduces the recognition error by 60% compared with traditional visual algorithms, is easy to promote and implement, and has good economic benefits. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of an automatic weld seam recognition and adaptive welding system for large non-standard parts. Detailed Implementation

[0028] Example: See Figure 1 In the picture, An automatic weld seam recognition and adaptive welding system for large non-standard parts includes a mobile fixed frame, a welding positioner, a welding robot, and a central control unit. The mobile fixed frame is equipped with a global wide-area 3D scanner to acquire the overall point cloud of the workpiece. The welding robot is equipped with a structured light vision sensor to perform fine scanning of the weld seam area. The central control unit is communicatively connected to the global wide-area 3D scanner, the welding robot, the structured light vision sensor, and the welding power supply. It is used to schedule the order of global scanning and local fine scanning. The central control unit receives detection signals from the global wide-area 3D scanner and the structured light vision sensor, uses a RandLA-Net deep learning semantic segmentation model to identify the weld seam type, start point, and end point from the overall point cloud, generates a welding path, and controls the welding robot to weld along the preset path. Simultaneously, based on the bevel size and root gap obtained from the fine scanning, it matches welding process parameters in real time to complete the overall welding.

[0029] Preferred option: Welding process parameters: current 150-300A, voltage 20-30V, welding speed 20-50cm / min.

[0030] Preferred solution: The structured light vision sensor tracks the weld seam in real time at a frequency of 5-10Hz during the welding process and feeds back the deviation to the welding robot to achieve closed-loop control.

[0031] Preferred solution: The global wide-area 3D scanner is a laser line scan sensor with a scanning field of view ≥2m×2m and a point cloud density ≥0.5 mm.

[0032] Preferred solution: The process parameters include three sets of current-voltage-speed triplets corresponding to the root gap g≤1.0 mm, 1.0 mm<g≤2.0 mm, and g>2.0 mm. For every 0.1 mm increase in gap, the current increases by 5-8A.

[0033] Preferred solution: The mobile fixed frame is a gantry frame, overhead crane, or the first axis of a large robot, and the welding robot is an industrial robot with six or more degrees of freedom, possessing sufficient working range and flexibility.

[0034] An automatic weld seam identification and adaptive welding method for large non-standard parts includes the following steps: S1. Loading and Positioning: Hoist the large non-standard parts to the work area and fix them on the welding positioner, while moving the mobile fixing frame above the welding positioner. S2. Global Scan and Preliminary Identification: The global wide-area 3D scanning unit is activated to scan the surface of the large non-standard part, obtain the overall point cloud of the workpiece, and the central control unit receives the data information to preliminarily identify the approximate location and sequence of multiple weld seams to be welded. S2. Path planning and ranking: The central control unit uses the RandLA-Net model to perform semantic segmentation on the overall point cloud to obtain weld seam point cloud clusters. The robot's collision-free movement path is generated by RANSAC fitting, and the path is ranked by weights of weld seam length, accessibility, and thermal deformation accumulation. S4. Local fine scanning and feature extraction: The welding robot moves to the starting point of the target weld according to the sorted path, and activates the structured light vision sensor at the end to perform fine scanning to obtain the bevel angle α, root gap g, and misalignment Δh. S5. Adaptive parameter matching: Based on the extracted features, with a gap width of 1.2±0.3mm, the final welding path and real-time process parameters are generated: current 220A, voltage 24.5V, and speed 35cm / min. S6. Welding Execution: The central control unit issues instructions, the welding robot performs welding, and the structured light vision sensor tracks in real time and feeds back the deviation to the robot for closed-loop control. S7 Welding complete: The system saves welding process parameters, actual trajectory, point cloud and images to form a digital weld file.

[0035] Preferred solution: In step S5, if g ≤ 1.0 mm, single-pass welding is selected, and the thin bevel group in the process parameter library is called: current 180-220 A, voltage 22-26 V, speed 30-45 cm / min; if 1.0 mm < g ≤ 2.0 mm, double-pass welding of root pass + cover pass is selected, with root pass parameters: current 160-200 A, voltage 21-25 V, speed 25-35 cm / min, and cover pass parameters: current 200-240 A, voltage 23-27 V, speed 30-40 cm / min; if g > 2.0 mm or Δh > 1 mm, the root pass, fill pass, and cover pass are executed sequentially. The fill pass uses oscillating welding with an oscillation frequency of 2-4 Hz, an oscillation amplitude of 2-6 mm, and a dwell time of 0.1-0.3 s at both ends; the system generates the final TCP trajectory based on the above matching results and sends the current, voltage, speed, and oscillation mode to the welding power source; Preferred solution: In step S6, the welding robot initiates the arc according to the preset trajectory. The structured light vision sensor measures the actual gap g′ at the root in real time at a frequency of 5-10Hz and compares it with the set value g in step S4: when |g′-g|≤0.3 mm, the original parameters are maintained; when |g′-g|>0.3 mm, the welding speed v′=v×(g / g′) is corrected online, and the current ±10 A and voltage ±1 V are adjusted synchronously to achieve adaptive follow-up.

[0036] Preferred solution: In step S3, the starting point is the center of the welding robot's base, and the obstacle avoidance area is the global point cloud convex hull extended by 100mm from the start point to the end point of the weld.

[0037] The fully automated process of this invention consists of: global scanning -> preliminary identification and planning -> robot localization -> local fine scanning -> feature extraction and parameter adaptation -> welding execution, forming a complete closed loop.

[0038] Central Control Unit: As the system's brain, it integrates all the aforementioned modules. At its core is an industrial PC that runs the robot control system, vision processing software, and path planning algorithms. It is responsible for scheduling the work of global and local vision sensors, processing data, generating work instructions, and controlling the robot and welding power source to collaboratively complete the task.

[0039] Large workpiece positioning system: For ultra-large workpieces, a positioner or locator can be used in coordination with a robot to ensure that the weld is always in the optimal welding position (PA / PB position).

[0040] Welding power source and wire feeding mechanism: A high-performance digital welding power source integrated with the robot control system, capable of receiving instructions from the path planning module and adjusting output parameters in real time.

[0041] Point cloud processing and weld feature extraction unit: This unit stitches, denoises, and segments point cloud data from global and local scans. It utilizes artificial intelligence algorithms (such as deep learning neural networks) to perform semantic segmentation of the point cloud, automatically identifying weld types (such as V-groove, fillet weld, and lap weld), start points, end points, and trajectories.

[0042] Adaptive Welding Path Generation: Based on the extracted weld features, an optimized robotic welding path (TCP trajectory) is automatically generated. Simultaneously, based on real-time detected parameters such as bevel size and gap, welding process parameters (such as current, voltage, welding speed, and oscillation mode) are adaptively generated and matched to a process parameter library.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications made based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. An automatic weld seam recognition and adaptive welding system for large non-standard parts, comprising a mobile fixed frame, a welding positioner, a welding robot, and a central control unit, characterized in that: The mobile fixed frame is equipped with a global wide-area 3D scanner to acquire the overall point cloud of the workpiece. The welding robot is equipped with a structured light vision sensor to perform fine scanning of the weld area. The central control unit is communicatively connected to the global wide-area 3D scanner, the welding robot, the structured light vision sensor, and the welding power supply. It is used to schedule the order of global scanning and local fine scanning. The central control unit receives the detection signals from the global wide-area 3D scanner and the structured light vision sensor, uses the RandLA-Net deep learning semantic segmentation model to identify the weld type, start point, and end point from the overall point cloud, generates a welding path, and controls the welding robot to weld according to the preset path. At the same time, based on the bevel size and root gap obtained from the fine scanning, the welding process parameters are matched in real time to complete the overall welding.

2. The automatic weld seam identification and adaptive welding system for large non-standard parts according to claim 1, characterized in that: The welding process parameters are: current 150-300A, voltage 20-30V, and welding speed 20-50 cm / min.

3. The automatic weld seam identification and adaptive welding system for large non-standard parts according to claim 1, characterized in that: The structured light vision sensor tracks the weld seam in real time at a frequency of 5-10Hz during the welding process and feeds back the deviation to the welding robot to achieve closed-loop control.

4. The automatic weld seam identification and adaptive welding system for large non-standard parts according to claim 1, characterized in that: The global wide-area 3D scanner is a laser line scan sensor with a scanning field of view ≥2m×2m and a point cloud density ≥0.5 mm.

5. The automatic weld seam identification and adaptive welding system for large non-standard parts according to claim 1, characterized in that: The process parameters include three sets of current-voltage-speed triplets corresponding to the root gaps g≤1.0 mm, 1.0 mm<g≤2.0 mm, and g>2.0 mm. For every 0.1 mm increase in gap, the current increases by 5-8 A.

6. The automatic weld seam identification and adaptive welding system for large non-standard parts according to claim 1, characterized in that: The mobile fixed frame is a gantry frame, overhead crane, or the first axis of a large robot, and the welding robot is an industrial robot with six or more degrees of freedom, possessing sufficient working range and flexibility.

7. A method for automatic weld seam identification and adaptive welding of large non-standard parts using the automatic weld seam identification and adaptive welding system of any one of claims 1-6, comprising the following steps: S1. Loading and Positioning: Hoist the large non-standard parts to the working area and fix them on the welding positioner, while moving the mobile fixing frame above the welding positioner. S2. Global Scan and Preliminary Identification: The global wide-area 3D scanning unit is activated to scan the surface of the large non-standard part, obtain the overall point cloud of the workpiece, and the central control unit receives the data information to preliminarily identify the approximate location and sequence of multiple weld seams to be welded. S2. Path planning and ranking: The central control unit uses the RandLA-Net model to perform semantic segmentation on the overall point cloud to obtain weld seam point cloud clusters. The robot's collision-free movement path is generated by RANSAC fitting, and the path is ranked by weights of weld seam length, accessibility, and thermal deformation accumulation. S4. Local fine scanning and feature extraction: The welding robot moves to the starting point of the target weld according to the sorted path, and starts the structured light vision sensor at the end to perform fine scanning to obtain the bevel angle α, root gap g, and misalignment amount Δh. S5. Adaptive parameter matching: Based on the extracted features, with a gap width of 1.2±0.3mm, the final welding path and real-time process parameters are generated: current 220A, voltage 24.5V, and speed 35cm / min. S6. Welding Execution: The central control unit issues a command, and the welding robot performs welding. At the same time, the structured light vision sensor tracks in real time and feeds back the deviation to the robot for closed-loop control. S7 Welding complete: The system saves welding process parameters, actual trajectory, point cloud and images to form a digital weld file.

8. The automatic weld seam identification and adaptive welding method for large non-standard parts according to claim 7, characterized in that: in In step S5, if g ≤ 1.0 mm, single-pass welding is selected, and the thin bevel group in the process parameter library is called: current 180-220A, voltage 22-26 V, speed 30-45 cm / min; if 1.0 mm < g ≤ 2.0 mm, double-pass welding of root pass + cover pass is selected, with root pass parameters: current 160-200 A, voltage 21-25 V, speed 25-35 cm / min, and cover pass parameters: current 200-240 A, voltage 23-27 V, speed 30-40 cm / min; if g > 2.0 mm or Δh > 1 mm, root pass, fill pass, and cover pass are executed sequentially. The fill pass uses oscillating welding with an oscillation frequency of 2-4 Hz, an oscillation amplitude of 2-6 mm, and a dwell time of 0.1-0.3 s at both ends. The system generates the final TCP trajectory based on the above matching results and sends the current, voltage, speed, and oscillation mode to the welding power source.

9. The automatic weld seam identification and adaptive welding method for large non-standard parts according to claim 7, characterized in that: in In step S6, the welding robot initiates the arc according to the preset trajectory. The structured light vision sensor measures the actual gap g′ at the root in real time at a frequency of 5-10 Hz and compares it with the set value g in step S4: when |g′-g|≤0.3 mm, the original parameters are maintained; when |g′-g|>0.3 mm, the welding speed v′=v×(g / g′) is corrected online, and the current ±10 A and voltage ±1 V are adjusted synchronously to achieve adaptive follow-up.

10. The automatic weld seam identification and adaptive welding method for large non-standard parts according to claim 7, characterized in that: in In step S3, the starting point is the center of the welding robot's base, and the obstacle avoidance area is the global point cloud convex hull expanded by 100mm from the start point to the end point of the weld.

Images