An Adaptive Method and System for Weld Inspection and Grinding Parameters Based on 3D Vision

Abstract

This invention proposes a 3D vision-based adaptive method and system for weld seam inspection and grinding parameters. The method includes: S1, calibration stage: configuring a position mapping table, grinding parameter table, and detection threshold for each weld seam ID; S2, runtime stage, including: S21, image acquisition and ROI matching; S22, weld seam type determination; S23, dynamic matching of current grinding parameters; S24, execution of grinding according to the final determined grinding parameters; S25, reporting weld seam data before and after grinding. The system is used to implement the above method. This invention, through a separate architecture for calibration and runtime, and adaptive geometric modeling for weld seam types, achieves dynamic matching and automatic optimization of grinding parameters, avoiding over-grinding and under-grinding; simultaneously, it automatically reports data before and after grinding, forming a closed-loop quality traceability system. The entire grinding process requires no manual intervention, meeting the requirements of intelligent manufacturing digital management.

Description

Technical Field

[0001] This invention relates to the field of automated grinding technology for industrial robots, specifically to a method and system for weld seam inspection and grinding parameter adaptation based on 3D vision. Background Technology

[0002] In the field of automated grinding using industrial robots, weld seam removal is a common process. However, due to factors such as workpiece positioning errors, weld seam size fluctuations, and poor material consistency, traditional grinding methods with fixed parameters cannot guarantee stable quality: over-grinding will damage the base material, while under-grinding requires repeated rework, which seriously affects production efficiency and product qualification rate.

[0003] While vision-based weld inspection and grinding methods have some applications in the current technology, most are only used for simple flatness inspection. They have failed to form a closed-loop control from inspection to parameter optimization, and lack targeted processing for different weld types (joints, corner joints).

[0004] To address the aforementioned issues, current vision-based methods for setting weld inspection and grinding parameters are mainly categorized into three types, but each has its own significant drawbacks: The first type is based on traditional image processing and structured light vision methods. After extracting the weld contour through edge detection or structured light projection, parameters are set according to human experience. However, it is sensitive to lighting and reflection and has difficulty in handling the three-dimensional geometric features of complex curved surfaces such as fillet welds. It still remains in the open-loop mode of "measurement-execution".

[0005] The second category is weld detection methods based on deep learning, which use convolutional neural networks to segment weld areas or extract point cloud features. However, their performance is limited by the quality of training data and they only focus on improving recognition accuracy, lacking dynamic optimization of grinding process parameters.

[0006] The third category is grinding methods based on fixed parameter tables or force control. After determining the fixed parameters through trial and error, these methods have been used continuously. Although some solutions introduce force control technology to compensate for contact force fluctuations or integrate visual perception to adjust parameters, their core perception indicators are mostly texture roughness and surface gloss rather than the three-dimensional geometric features of the weld. The process parameters still need to be preset manually, and a true closed-loop optimization of visual measurement and grinding parameters has not been achieved.

[0007] In summary, existing technologies have shortcomings such as reliance on manual experience for parameter setting, lack of closed-loop optimization mechanisms, poor adaptability to weld types, and inability to support adaptive parameters for multiple grinding processes. Summary of the Invention

[0008] In order to solve the technical problems existing in the background art, the present invention proposes a method and system for adaptive welding seam detection and grinding parameters based on 3D vision.

[0009] This invention proposes a 3D vision-based adaptive method for weld seam inspection and grinding parameters, comprising the following steps: S1. Calibration stage: Configure a position mapping table for each weld ID to record the correspondence between the end pose of the grinding robot when taking pictures and the ROI parameters of the weld, a grinding parameter table sorted in ascending order of maximum allowable height, and various required detection thresholds, and store them in memory with the project ID as the key; S2, the runtime phase, includes the following steps: S21. According to the current pose of the grinding robot, match the position mapping table to obtain the corresponding ROI parameters; and collect the 2D grayscale image and 3D point cloud of the weld area; S22. Determine the weld type based on the spatial distribution characteristics of the point cloud within the ROI, and calculate the weld height and offset. S23. Dynamically match the current grinding parameters from the grinding parameter table based on the weld height calculated in step S22; S24. Control the grinding robot to perform grinding with the final determined grinding parameters and weld offset; S25. Report the weld data before and after grinding to form a closed loop for quality traceability.

[0010] Preferably, during the grinding process, a maintenance request counter is set for each weld. When the number of requests for the same weld exceeds a preset threshold, the grinding parameters are automatically upgraded to the corresponding parameters at a higher level in the parameter table, and the number of requests is forcibly set to the same preset threshold.

[0011] Preferably, the preset threshold for the number of requests for the same weld seam is 2. When the number of requests for the same weld seam exceeds 2, the grinding parameters are automatically upgraded to the next level in the parameter table, and the number of requests for this time is forcibly set to 2.

[0012] Preferably, the grinding parameters in the grinding parameter table include at least pressure, speed, and number of grinding cycles; when upgrading the grinding parameters, pressure and speed should be upgraded simultaneously, or only the number of grinding cycles should be upgraded.

[0013] Preferably, in step S21, the ROI parameters include position, size, angle, and detection direction.

[0014] Preferably, in step S21, the detection threshold includes: reference point width, segmentation distance, positive and negative thresholds, and fillet weld height range.

[0015] Preferably, in step S22, the criteria for determining the weld type are as follows: if the point cloud within the ROI is distributed near a single plane and the height change is less than a set threshold, it is determined to be a splicing weld; if the point cloud within the ROI presents an arc-shaped distribution or has a curved surface, it is determined to be a fillet weld.

[0016] Preferably, if it is a spliced ​​weld, a reference plane is fitted and a height difference diagram is generated; then the average value of the maximum height of each segment is calculated along the weld direction as the weld height and offset. If it is a fillet weld, fit a cylindrical surface and calculate the signed distance; then calculate the maximum value of the distance in each segment along the weld direction as the weld height and offset.

[0017] Preferably, in step S23, the matching rule for the grinding parameters is: take the first parameter that satisfies "the weld height is less than the maximum allowable height", and if there is no such parameter, take the last parameter.

[0018] Preferably, in step S25, the weld data before and after grinding includes: weld height, defect area, grinding parameters used, number of requests, and timestamp.

[0019] Preferably, the defect area is obtained by performing positive and negative threshold binarization on the height difference map and then performing contour analysis.

[0020] This invention proposes a 3D vision-based adaptive system for weld seam inspection and grinding parameters. This system is used to implement the aforementioned 3D vision-based adaptive method for weld seam inspection and grinding parameters, and includes: a calibration module, a data acquisition module, a processing module, a parameter adaptation module, and a data interaction module, wherein: The calibration module is used to configure and store the location mapping table, grinding parameter table and detection threshold for each weld ID; The acquisition module is used to acquire images of the current pose of the grinding robot and the weld area; The processing module is electrically connected to the acquisition module and the calibration module respectively. It is used to match the ROI from the position mapping table according to the pose of the grinding robot, determine the weld type, and calculate the weld height and offset. The parameter adaptive module is electrically connected to the calibration module and the processing module respectively, and is used to dynamically match the corresponding grinding parameters from the grinding parameter table according to the calculated weld height. The data interaction module is connected to the parameter adaptation module to send grinding parameters to the grinding robot or PLC, and to report the weld data before and after grinding to the MES or local storage.

[0021] Preferably, the parameter adaptive module also maintains a request counter for each weld and automatically upgrades the parameter level when the number of requests exceeds the pre-approval threshold, and forces the current number of requests to be equal to the pre-approval threshold.

[0022] Preferably, it also includes a concurrency management module, which is used to isolate data containers and counters of different projects by using project ID as the key, to protect data access by using read-write locks and mutex locks, and to submit time-consuming tasks to the thread pool for asynchronous execution.

[0023] This invention avoids over-grinding and under-grinding by separating calibration and operation, determining weld type (joint welds: planar fitting or fillet welds: cylindrical fitting), and dynamically matching and automatically optimizing grinding parameters. Simultaneously, it automatically reports data before and after grinding, forming a closed-loop quality traceability system that facilitates production monitoring and quality analysis. The entire grinding process requires no manual intervention, meeting the requirements of intelligent manufacturing and digital management. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the ROI and strip-shaped area of ​​the splicing weld; Figure 2 A point cloud diagram of a fillet weld; Figure 3 This is a schematic diagram of the ROI and strip-shaped area of ​​the fillet weld; Figure 4 This is a schematic diagram of a 3D vision-based adaptive system for weld inspection and grinding parameters. Detailed Implementation

[0025] Example 1:

[0026] This embodiment proposes a 3D vision-based adaptive method for weld seam detection and grinding parameters, applicable to robotic automated grinding scenarios for spliced ​​welds and fillet welds. The method is divided into a calibration phase and an operation phase, specifically including the following steps: S1, Calibration Phase Before the project officially goes into operation, operators complete the calibration through offline programming or teaching to configure the following three types of parameters for each weld ID: Position mapping table: Records the correspondence between the end pose of the grinding robot when taking pictures and the parameters (position, size, angle, detection direction) of the weld ROI (region of interest).

[0027] Grinding parameter table: Several sets of parameters are arranged in ascending order of maximum allowable height. Each set includes pressure (N), speed (mm / s), and number of grinding passes. For example: Level 1: Maximum allowable height 0.5mm, pressure 20N, speed 50mm / s, number of cycles 1; Second gear: maximum allowable height 1.0mm, pressure 30N, speed 40mm / s, number of cycles 2; Level 3: Maximum allowable height 1.5mm, pressure 40N, speed 30mm / s, number of cycles 3.

[0028] Detection thresholds include reference point width, segmentation distance, positive and negative thresholds, and fillet weld height range.

[0029] All of the above parameters are stored in the system memory with "Project ID" as the key for quick retrieval at runtime.

[0030] S2, the operational phase, specifically includes the following steps: S21. First, match the position mapping table according to the current pose of the grinding robot to obtain the corresponding ROI parameters; then, acquire the 2D grayscale image and 3D point cloud of the weld area. The specific steps are as follows: System initialization: The grinding robot sends an initialization request to clear the weld data container and request counter under the specified project ID, in preparation for receiving new tasks.

[0031] Image Acquisition and ROI Matching: After the grinding robot moves to the shooting point, it sends a shooting request. The system matches the current pose of the grinding robot to the position mapping table, automatically obtaining the corresponding ROI parameters to achieve precise alignment between the shooting position and the calibration data. Subsequently, the 3D vision system simultaneously acquires 2D grayscale images and 3D point clouds of the weld area, and performs preprocessing such as filtering and noise reduction.

[0032] S22. Determine the weld type based on the spatial distribution characteristics of the point cloud within the ROI, and calculate the weld height: If the point cloud is distributed near an approximately single plane and the height variation is less than a set threshold (e.g., 0.3 mm), it is determined to be a splicing weld. In this case, as shown in Figure 1, the point cloud of the strip region within the ROI is extracted, a reference plane is fitted, the signed distance from all points within the ROI to the reference plane is calculated, and a height difference map is generated. Then, the ROI is divided into several segments along the weld direction (each segment is, for example, 5 mm in length), the average of the maximum height within each segment is calculated as the height of the weld segment, and the weld offset is calculated.

[0033] If the point cloud exhibits an arc-shaped distribution or has a clearly curved surface, it is determined to be a fillet weld. In this case, if... Figure 2-3 As shown, high-point clouds above a set threshold within the strip region of the ROI are extracted as candidate points. A cylindrical surface is fitted using the least squares method or the RANSAC algorithm, and the signed distance from all points within the ROI to this cylindrical surface is calculated. Then, similarly, the area is divided into segments along the weld direction, and the maximum distance within each segment is calculated as the weld height. The weld offset is also calculated.

[0034] S23. Based on the calculated weld height, find the matching parameter from the grinding parameter table. The matching rule is: take the first parameter that satisfies "weld height is less than the maximum allowable height". If no such parameter is found, take the last parameter.

[0035] For example, when the calculated weld height is 1.2 mm: Level 1 (maximum 0.5mm): Not met Level 2 (maximum 1.0mm): Not met. Level 3 (maximum 1.5mm): Meets the requirements → Matches the parameters for Level 3 (pressure 40N, speed 30mm / s, number of cycles 3).

[0036] S24. The finalized grinding parameters (pressure, speed, number of passes) and the weld offset calculated in step S22 (for trajectory fine-tuning) are encapsulated according to the agreed communication protocol and sent to the grinding robot or PLC controller. The grinding robot performs the grinding action according to the received parameters.

[0037] As a further aspect of this embodiment, during the grinding process, a request counter is maintained for each weld (using project ID + weld ID as the joint key), with an initial value of 0. The counter is incremented by 1 after each grinding request (including taking photos, matching, and sending parameters). When the number of requests for the same weld exceeds a preset threshold (e.g., the threshold is 2 times), a parameter upgrade mechanism is triggered: the grinding parameters are automatically upgraded to the next level in the parameter table (e.g., from level 2 to level 3); simultaneously, the number of requests is forcibly set to the same preset threshold 2 (i.e., when the threshold is 2, the number of requests is forcibly set to 2), ensuring that the same weld still retains the possibility of another upgrade after the parameter upgrade, avoiding skipping reasonable intermediate parameters due to the counter accumulating infinitely (e.g., jumping directly from 3 times to 10 times), resulting in over-grinding or parameter waste.

[0038] For example: A weld is initially matched with the second level of parameters. After the first and second grinding, the weld height is still too high. On the third request, the threshold (2 times) is exceeded, and the parameter is automatically upgraded to the third level, and the counter is reset to 2. If the quality requirements are still not met after grinding at the third level, the upgrade will be triggered again on the fourth request (because the current counter is 2, and it becomes 3 after the second request, which still exceeds the threshold), and the upgrade will continue to the fourth level.

[0039] In practical work, when parameters are upgraded, you can choose to upgrade the pressure and speed in sync (incrementing according to the table), or only upgrade the number of polishing cycles (for example, from 2 to 3).

[0040] S25. Data Reporting and Quality Traceability: After the grinding robot completes the grinding process, it sends a request for inspection results. It also packages the weld data before and after grinding, reports it to the MES system via HTTP, saves it to a local CSV file, and stores the inspection image with defect annotations. Specifically, the reported data includes: weld height (measured value before and after grinding), defect area (obtained by binarizing the height difference image with positive and negative thresholds and then performing contour analysis), grinding parameters used (pressure, speed, number of passes, parameter level), as well as the number of requests, timestamp, weld ID, project ID, etc.

[0041] These data form a complete quality traceability loop, which can be used for subsequent process analysis and optimization.

[0042] Example 2:

[0043] This embodiment provides a 3D vision-based adaptive system for weld seam inspection and grinding parameters, used to implement the method described in Embodiment 1. The system includes the following modules: Calibration Module: This module is responsible for configuring and storing a location mapping table, a grinding parameter table (at least including pressure, speed, and grinding times), and various detection thresholds for each weld ID during the project configuration phase. All data is stored in an in-memory database using the project ID as the key.

[0044] Acquisition Module: This module is used to acquire images of the current pose of the grinding robot and the weld area.

[0045] Processing Module: This module is electrically connected to both the acquisition module and the calibration module. During operation, this module matches the ROI parameters based on the current pose of the grinding robot and distinguishes between spliced ​​welds and fillet welds based on the 2D grayscale image and 3D point cloud of the weld area acquired by the acquisition module. If it is a spliced ​​weld, the weld height and offset are calculated using planar fitting; if it is a fillet weld, the weld height and offset are calculated using cylindrical fitting.

[0046] Parameter Adaptive Module: This module is electrically connected to the calibration module and the processing module respectively. It is responsible for dynamically matching the best parameters from the grinding parameter table based on the weld height calculated by the processing module (matching rule: take the first parameter item that satisfies weld height < maximum allowable height); and maintaining an independent request counter for each weld (project ID + weld ID); when the number of requests exceeds the preset threshold, it automatically upgrades the parameters to the next level and forces the current number of requests to be set to the preset threshold.

[0047] Data interaction module: This module is connected to the parameter adaptation module and is used to send the grinding parameters and weld offset output by the parameter adaptation module to the grinding robot or PLC controller; and to report the weld data before and after grinding (height, defect area, parameters, number of requests, timestamp, etc.) to the MES system or save it to local storage (CSV file + labeled image).

[0048] Concurrency Management Module: To support simultaneous operation of multiple robots or multiple projects, this module implements the following concurrency control mechanisms: Using the project ID as the key, isolate the weld data containers and counters of different projects so that they do not interfere with each other; Read-write locks are used to protect concurrent reading of configuration parameters (allowing multiple robots to read simultaneously). Use mutex locks to protect write operations on the data container (to prevent data conflicts); All time-consuming tasks (such as image processing, parameter calculation, and network upload) are submitted to the thread pool for asynchronous execution to ensure that the communication thread responds quickly.

[0049] This invention, through its architecture that separates calibration and runtime, its adaptive geometric modeling based on weld type, and its count-based secondary grinding parameter upgrade mechanism, offers the following advantages compared to existing technologies: 1. Improve grinding quality and consistency: During the calibration stage, all weld-related parameters (position mapping table, pressure and speed table, detection threshold, etc.) are configured at once and stored with the project ID as the key. They can be directly called during runtime, avoiding repeated manual intervention for debugging and greatly improving system response efficiency and consistency between workpieces.

[0050] 2. Improved accuracy of grinding parameter matching: By automatically determining the weld type (joint weld: planar fitting or fillet weld: cylindrical fitting), accurate three-dimensional reconstruction of complex weld morphology is achieved, which effectively improves the accuracy of grinding parameter matching and avoids over-grinding (damage to the base material) or under-grinding (rework required) caused by fixed parameters.

[0051] 3. Improved production efficiency: An independent request counter is maintained for each weld. When the number of grinding requests for the same weld exceeds a preset threshold, the parameters are automatically upgraded to the next level, and the current request count is forcibly set to the threshold. This mechanism solves the problems of unreasonable parameters and frequent manual intervention during multiple grinding processes for high-residue welds, reducing the number of grinding operations and shortening the cycle time per piece while ensuring quality.

[0052] 4. Enhanced adaptability and digital management capabilities: This invention can adapt to both splice welds and fillet welds simultaneously, without the need for manual switching or reprogramming, thus enhancing the system's versatility and engineering practicality.

[0053] 5. Form a complete quality traceability closed loop: By automatically reporting weld data (height, defect area, usage parameters, timestamp, etc.) before and after grinding to the MES system and saving local images and CSV files, a complete quality traceability closed loop is formed, providing a digital foundation for process optimization and production monitoring, which meets the requirements of intelligent manufacturing for full-process digital management.

[0054] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for adaptive parameter detection and grinding of welds based on 3D vision, characterized in that, Includes the following steps: S1. Calibration stage: Configure a position mapping table for each weld ID to record the correspondence between the end pose of the grinding robot when taking pictures and the ROI parameters of the weld, a grinding parameter table sorted in ascending order of maximum allowable height, and various required detection thresholds, and store them in memory with the project ID as the key; S2, the runtime phase, includes the following steps: S21. According to the current pose of the grinding robot, match the position mapping table to obtain the corresponding ROI parameters; and collect the 2D grayscale image and 3D point cloud of the weld area; S22. Determine the weld type based on the spatial distribution characteristics of the point cloud within the ROI, and calculate the weld height and offset. S23. Dynamically match the current grinding parameters from the grinding parameter table based on the weld height calculated in step S22; S24. Control the grinding robot to perform grinding with the final determined grinding parameters and weld offset; S25. Report the weld data before and after grinding to form a closed loop for quality traceability.

2. The adaptive method for weld seam inspection and grinding parameters based on 3D vision according to claim 1, characterized in that, During the grinding process, a request counter is maintained for each weld. When the number of requests for the same weld exceeds the preset threshold, the grinding parameters are automatically upgraded to the corresponding parameters at a higher level in the parameter table, and the number of requests is forcibly set to the same as the preset threshold. Preferably, the preset threshold for the number of requests for the same weld seam is 2. When the number of requests for the same weld seam exceeds 2, the grinding parameters are automatically upgraded to the next level in the parameter table, and the number of requests for this time is forcibly set to 2.

3. The adaptive system for weld seam inspection and grinding parameters based on 3D vision according to claim 2, characterized in that, The grinding parameters in the grinding parameter table should include at least pressure, speed, and number of grinding cycles. When upgrading grinding parameters, pressure and speed should be upgraded simultaneously, or only the number of grinding cycles should be upgraded.

4. The adaptive method for weld seam inspection and grinding parameters based on 3D vision according to claim 1, characterized in that, In step S21, the ROI parameters include position, size, angle, and detection direction; Preferably, in step S21, the detection threshold includes: reference point width, segmentation distance, positive and negative thresholds, and fillet weld height range.

5. The adaptive method for weld seam inspection and grinding parameters based on 3D vision according to claim 1, characterized in that, In step S22, the criteria for determining the weld type are as follows: if the point cloud within the ROI is distributed near a single plane and the height change is less than the set threshold, it is determined to be a splicing weld; if the point cloud within the ROI presents an arc-shaped distribution or has a curved surface, it is determined to be a fillet weld. Preferably, if it is a spliced ​​weld, a reference plane is fitted and a height difference diagram is generated; then the average value of the maximum height of each segment is calculated along the weld direction as the weld height and offset. If it is a fillet weld, fit a cylindrical surface and calculate the signed distance; then calculate the maximum value of the distance in each segment along the weld direction as the weld height and offset.

6. The adaptive method for weld seam inspection and grinding parameters based on 3D vision according to claim 1, characterized in that, In step S23, the matching rule for grinding parameters is: take the first parameter that satisfies "weld height is less than the maximum allowable height"; if there is no such parameter, take the last parameter.

7. The adaptive method for weld seam inspection and grinding parameters based on 3D vision according to claim 1, characterized in that, In step S25, the weld data before and after grinding includes: weld height, defect area, grinding parameters used, number of requests, and timestamp.

8. The adaptive method for weld seam inspection and grinding parameters based on 3D vision according to claim 7, characterized in that, The defect area is obtained by binarizing the height difference map with positive and negative thresholds and performing contour analysis.

9. A 3D vision-based adaptive system for weld inspection and grinding parameters, characterized in that, This system is used to implement the 3D vision-based weld inspection and grinding parameter adaptive method according to any one of claims 1-8, comprising: a calibration module, an acquisition module, a processing module, a parameter adaptive module, and a data interaction module, wherein: The calibration module is used to configure and store the location mapping table, grinding parameter table and detection threshold for each weld ID; The acquisition module is used to acquire images of the current pose of the grinding robot and the weld area; The processing module is electrically connected to the acquisition module and the calibration module respectively. It is used to match the ROI from the position mapping table according to the pose of the grinding robot, determine the weld type, and calculate the weld height and offset. The parameter adaptive module is electrically connected to the calibration module and the processing module respectively, and is used to dynamically match the corresponding grinding parameters from the grinding parameter table according to the calculated weld height. The data interaction module is connected to the parameter adaptation module to send grinding parameters to the grinding robot or PLC, and to report the weld data before and after grinding to the MES or local storage.

10. The adaptive system for weld inspection and grinding parameters based on 3D vision according to claim 9, characterized in that, The parameter adaptive module also maintains a request counter for each weld and automatically upgrades the parameter level when the number of requests exceeds the pre-review threshold, and forces the current number of requests to be equal to the pre-review threshold. Preferably, it also includes a concurrency management module; the concurrency management module is used to isolate data containers and counters of different projects by using project ID as the key, to protect data access by using read-write locks and mutex locks, and to submit time-consuming tasks to the thread pool for asynchronous execution.

Images