An adaptive anti-deformation corrugated plate weld seam extraction method and system

Abstract

The application relates to the technical field of welding automation, and provides a corrugated plate welding seam extraction method and system which are self-adaptive and anti-deformation, the method comprising the following steps: acquiring a three-dimensional point cloud of a corrugated plate surface; segmenting two adjacent geometric surfaces to be welded; selecting a reference surface and a projection surface; calculating the normal distance of each point of the projection surface to the reference surface, screening effective projection points reflecting the real welding seam boundary based on a set threshold value; projecting the effective projection points to the reference surface and performing fitting to obtain a welding seam track. The application automatically excludes local deformation interference through a threshold screening mechanism, directly extracts the welding seam from the real point cloud with high precision, and overcomes the defect that the prior art relies on an ideal geometric model and causes a large error.

Description

Technical Field

[0001] This invention belongs to the field of welding automation technology, specifically relating to an adaptive, deformation-resistant method and system for extracting corrugated plate welds. Background Technology

[0002] Corrugated steel sheets are widely used in chemical containers, ship structures, and shipping containers due to their high strength and lightweight properties. Welding is a key process in the forming of corrugated steel sheet components, and accurate extraction of weld locations is a prerequisite for subsequent automated welding path planning. In existing technologies, mainstream weld extraction methods are typically based on the assumption of ideal geometric surfaces, that is, assuming that the constituent planes or curved surfaces of the corrugated steel sheet are absolutely regular. These methods first divide the ideal geometric surface and then obtain the theoretical location of the weld through intersection of theoretical geometry.

[0003] However, in actual industrial production, corrugated sheet workpieces are inevitably affected by various factors such as rolling processes, transportation collisions, and clamping stress, resulting in localized depressions, bulges, or warping deformations that cannot meet the assumption of ideal geometry. The theoretical weld location calculated based on this assumption often deviates significantly from the actual weld location on the deformed corrugated sheet, typically by 1-3 mm. This deviation directly leads to errors in welding robot path planning, causing inaccurate welding torch alignment and resulting in welding defects such as incomplete fusion, undercut, and weld beads, seriously affecting welding quality and structural safety.

[0004] To improve weld seam extraction accuracy, some studies have focused on using vision or laser sensors to acquire workpiece information. For example, patent CN101559512B discloses a method for detecting and controlling the welding trajectory of butt welds on flat plates based on laser ranging. This method uses a laser sensor to scan the weld seam contour laterally and employs a B-spline algorithm for trajectory fitting. However, this method is mainly applicable to butt welds on flat plates and does not consider complex curved surface structures such as corrugated plates with periodic undulations and local deformations. Therefore, its algorithm is difficult to directly transfer and guarantee accuracy.

[0005] Patent CN112122842A discloses a Delta welding robot system based on laser vision, applied to the automatic welding of corrugated plates. It acquires point clouds using a line laser sensor and extracts the weld path using an adjacent circular algorithm. While this algorithm is effective for regular corrugations, for corrugated plates with local deformation, the calculation of the circular path and common tangent is easily affected by deformation points, causing the extracted weld center position to deviate from the actual position, resulting in insufficient robustness.

[0006] Patent CN120644849A discloses an intelligent welding method and system for the curved surface of container corrugated plates, which performs welding through 3D scanning, surface reconstruction, and intelligent path planning. This solution focuses on global optimization and closed-loop control of the welding process. Although it mentions point cloud processing and surface reconstruction, its core lies in the path planning algorithm and real-time defect feedback adjustment. It does not reveal a core method specifically designed for high-precision extraction of corrugated plate weld positions that can fundamentally overcome the influence of local deformation.

[0007] Therefore, existing technologies lack a method that can effectively adapt to the true geometry of corrugated plates (including local deformation) and accurately extract weld positions directly from high-precision point cloud data. This is a technical bottleneck restricting high-quality automated welding of corrugated plates. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention aims to provide an adaptive, deformation-resistant method for extracting corrugated plate welds, thereby solving the technical problem that existing technologies, based on ideal geometric assumptions, cannot adapt to the actual deformation of corrugated plates, resulting in low weld extraction accuracy.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] In a first aspect, the present invention proposes an adaptive, deformation-resistant method for extracting corrugated plate welds, comprising the following steps:

[0011] S1. Obtain three-dimensional point cloud data of the corrugated plate surface, wherein the point cloud data includes geometric deviation information caused by workpiece deformation;

[0012] S2. Segment the two adjacent geometric surfaces to be welded from the point cloud data, and denote them as geometric surface A and geometric surface B;

[0013] S3. Determine the spatial intersection relationship between geometric surface A and geometric surface B, and select one of the geometric surfaces as the reference surface and the other as the projection surface;

[0014] S4. Calculate the normal distance from each sampling point on the projection surface to the reference surface, and according to the preset distance threshold T used to distinguish between weld boundary points and local deformation points, select a set of valid projection points whose normal distance is not greater than the threshold T from the sampling points.

[0015] S5. Project the points in the set of effective projection points onto the reference plane, and fit the projected points to obtain the weld trajectory.

[0016] Furthermore, in step S1, point cloud data is acquired using a laser scanner or depth camera, and the density of the point cloud data is not less than 5 points / mm. 2 The ranging error is no greater than 0.08mm.

[0017] Furthermore, step S2 employs a region growing segmentation algorithm, wherein the conditions for determining whether two points belong to the same geometric surface include: the angle between the normal vectors of the two points does not exceed 3°-8°, and the Euclidean distance between the two points does not exceed 1mm-3mm.

[0018] Furthermore, in step S4, the distance threshold T ranges from 0.5 mm to 3.0 mm.

[0019] Furthermore, the distance threshold T is adaptively determined based on at least one parameter among the material, thickness, corrugation height, and processing accuracy requirements of the corrugated plate.

[0020] Furthermore, the adaptive determination includes: querying a pre-stored parameter-threshold matching database based on the input workpiece parameters to output a recommended distance threshold T.

[0021] Furthermore, in step S4, before or after filtering valid projection points, a step of denoising the sampling points on the projection surface or the set of valid projection points is also included.

[0022] Furthermore, in step S5, the RANSAC algorithm is used for linear fitting, or the least squares method is used for polynomial curve fitting.

[0023] Secondly, this invention proposes an adaptive corrugated plate weld extraction system to implement the aforementioned adaptive deformation-resistant corrugated plate weld extraction method, comprising:

[0024] The data acquisition module is used to acquire the three-dimensional point cloud data;

[0025] The data processing module is used to perform point cloud segmentation, normal distance calculation, effective point selection based on threshold T, and trajectory fitting.

[0026] The threshold setting module is used to provide functions for setting, recommending, or adjusting the distance threshold T; and

[0027] The result output module is used to output the weld trajectory.

[0028] Furthermore, the data acquisition module includes a high-precision tooling fixture with a repeatability of no more than 0.05 mm; and / or, the threshold setting module includes a human-computer interaction interface for displaying the impact of threshold adjustment on the number of effective projection points and the preview trajectory.

[0029] The beneficial effects of this invention are as follows: Existing technologies extract weld seams based on regular geometric models (such as drawing circles), which implicitly assume that the corrugated plate has a regular shape. This invention completely abandons this assumption and directly processes high-density real point cloud data containing all local deformation information. The normal distance from each point on the projection surface to the reference surface is calculated; this distance physically represents the convex height of the point relative to the reference surface. By setting a reasonable distance threshold T, points whose height is within the normal weld seam fluctuation range (≤T) are determined as valid projection points reflecting the true intersection trend, while points with abnormal heights (>T) due to severe local depressions or convexities are discarded as deformation noise. This filters out local deformation interference caused by rolling, transportation, clamping, etc., from the data source, ensuring that the data points used for subsequent fitting are pure and accurate. Compared to existing technologies that rely on ideal models or regular algorithms and are severely affected by deformation, this invention can stably control the weld seam position extraction error from 1-3mm to within ±0.5mm, achieving a qualitative leap in accuracy and meeting the stringent requirements of high-end automated welding. In the geometric surface segmentation stage, by setting thresholds for the included normal vector angle (3°-8°) and Euclidean distance (1mm-3mm), the region growing algorithm can flexibly adapt to real surfaces with varying flatness and curvature, ensuring accurate segmentation of adjacent surfaces even in the presence of deformation. The threshold T is not a fixed value but is related to the workpiece's physical properties (material, thickness) and process requirements (machining accuracy). The system incorporates a parameter-threshold matching database based on extensive process experiments. After the user inputs the workpiece specifications, the system automatically recommends the optimal T value through a lookup table. Simultaneously, this module provides an interactive interface, allowing users to fine-tune the process based on real-time previews. This enables the same system to adaptively handle corrugated sheet workpieces of different thicknesses, corrugation shapes, and deformation degrees without reprogramming or adjusting the core algorithm logic. It possesses both experience-based automatic decision-making capabilities and retains the flexibility of manual fine-tuning, greatly improving the method's versatility, ease of use, and portability across different production lines. The entire extraction process requires no manual intervention in complex geometric analysis and parameter trial and error. Operators only need to clamp the workpiece and confirm the parameters, and the system can automatically complete high-precision extraction, significantly improving efficiency compared to traditional manual measurement or semi-automatic methods. Since the results are entirely determined by the algorithm and parameters, the influence of human experience differences is eliminated, ensuring high consistency and traceability of weld seam extraction results in mass production. Attached Figure Description

[0030] Figure 1 This is a schematic flowchart of the adaptive deformation-resistant corrugated plate weld extraction method of the present invention.

[0031] Figure 2 A schematic diagram of a point cloud data acquisition scenario for a corrugated board.

[0032] Figure 3This is a schematic diagram showing how a point cloud is separated into two adjacent geometric surfaces (A and B) after processing by a region growing segmentation algorithm.

[0033] Figure 4 This diagram illustrates the selection of valid projection points, showing how points on the projection surface are classified as valid projection points (highlighted) or invalid points based on a comparison of their normal distance with a threshold T.

[0034] Figure 5 This is a schematic diagram of the final weld contour curve obtained by curve fitting after projecting the effective projection points onto the reference plane.

[0035] Figure 6 This is a schematic diagram showing the module composition and connection relationship of the adaptive corrugated plate weld seam extraction system of the present invention.

[0036] Figure 7 This is a detailed flowchart illustrating the algorithm execution process within the data processing module. Detailed Implementation

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] like Figure 1 As shown, the adaptive deformation-resistant corrugated plate weld extraction method of the present invention uses high-density point cloud data of the corrugated plate surface as the processing object. It does not rely on ideal geometric assumptions throughout the process. Each step clearly defines the operating specifications, parameter ranges, judgment criteria, and data processing logic to ensure reproducibility and scalability. Specifically, it includes the following steps:

[0039] Step 1: Data Acquisition. Obtain 3D point cloud data of the corrugated plate surface. The point cloud data includes geometric deviation information caused by workpiece deformation.

[0040] The core objective of this step is to collect high-density, high-precision point cloud data containing the true geometry of the corrugated plate (including local deformations) to provide a reliable data foundation for subsequent processing.

[0041] Specifically, this includes step 1.1: Preliminary preparations:

[0042] Workpiece fixing: High-precision tooling fixtures are used to fix the corrugated plate workpieces on the working platform. The repeatability of the fixtures is ≤0.05mm, ensuring that the workpieces do not loosen or vibrate during the scanning process (vibration displacement ≤0.02mm).

[0043] Surface cleaning: Wipe the surface of the corrugated plate with anhydrous ethanol to remove oil, rust, dust and other impurities, so as to avoid obstructing the scanning path or causing abnormal point cloud reflection intensity.

[0044] Environmental control: Avoid direct exposure to strong light (light intensity ≤ 500 lux) to reduce the impact of environmental factors on scanning accuracy.

[0045] Step 1.2: Equipment Selection and Parameter Setting:

[0046] Equipment selection: Select an industrial-grade 2D / 3D laser scanner or depth camera. The core performance indicators must meet the following requirements: point cloud density ≥ 5 points / mm², ranging error ≤ 0.08mm.

[0047] Parameter configuration: Adjust the scanning range according to the size of the corrugated plate to ensure complete coverage of the area to be welded and adjacent geometric surfaces (edge ​​redundancy scanning width ≥ 50mm); set the scanning resolution to 0.1mm and the reflection intensity threshold to 100 (for corrugated plates made of metal); enable the multi-angle stitching mode, plan 3-5 scanning perspectives, and ensure the perspective overlap rate is ≥ 30% to avoid point cloud loss caused by a single perspective.

[0048] Step 1.3: Scan Execution and Data Preprocessing:

[0049] Scanning operation: Start the laser scanner or depth camera and scan or take pictures sequentially according to the preset angle.

[0050] Data stitching: The ICP (Iterative Closest Point) registration algorithm is used to stitch and fuse point clouds from multiple perspectives. The registration error is ≤0.08mm, ensuring that the stitched point cloud is free from misalignment and overlap.

[0051] Data cleansing: Remove duplicate points (duplicate point criteria: spatial distance between two points ≤ 0.05mm) and isolated noise points (points with a distance ≥ 0.5mm from their adjacent points), and retain valid point cloud data.

[0052] Data format conversion: Convert the purified point cloud data into PLY or PCD format, including the 3D coordinates (X, Y, Z) and normal vector of each point. The information on reflection intensity provides complete data support for subsequent geometric surface segmentation.

[0053] Step 1.4: Data Transmission and Storage:

[0054] The processed point cloud data is transmitted to the data processing unit via a gigabit Ethernet interface at a transmission rate of ≥1000Mbps, ensuring no data loss and no delay.

[0055] The system automatically stores the original point cloud data, stitching results, and cleaned data, and archives them according to the naming rule of "workpiece number-production date-scanning time" for easy traceability and review.

[0056] Step 2: Geometric Surface Segmentation. Two adjacent geometric surfaces to be welded are segmented from the point cloud data, denoted as geometric surface A and geometric surface B.

[0057] The core objective of this step is to accurately segment the two adjacent geometric surfaces (geometric surface A and geometric surface B) to be welded from high-density point cloud data, laying the foundation for subsequent intersection relationship analysis and projection calculation. Specifically, it includes the following steps:

[0058] Step 2.1: Algorithm Selection:

[0059] The region growing segmentation algorithm is selected, which is superior to traditional gray-scale threshold segmentation and edge detection segmentation algorithms. Its core advantage lies in its ability to cluster based on the geometric features (normal vector, spatial distance) of point clouds, which can adapt to the segmentation requirements of deformable surfaces and curved surfaces, and has high segmentation accuracy and strong robustness.

[0060] Step 2.2: Key parameter settings (quantifiable and adjustable):

[0061] Seed point selection: The "automatic recognition + manual verification" mode is adopted. The two adjacent geometric surfaces of the area to be welded are automatically identified in the point cloud data. 3-5 seed points are evenly selected for each geometric surface (the seed points should be located in the center area of ​​the surface and avoid the edges or areas with severe deformation). If the automatic recognition effect is not good, manual marking of seed points is supported.

[0062] The threshold for the included angle of the normal vector is 3°-8°. This range can filter out more than 90% of the data on the same geometric surface. It can be adjusted according to the geometric shape of the corrugated plate: 3°-5° for planar corrugated plates and 5°-8° for curved corrugated plates. It is used to determine whether two points belong to the same geometric surface (if the included angle of the normal vectors of two points is less than or equal to the threshold, they are determined to be the same type of point).

[0063] Euclidean distance threshold: 1mm-3mm. This range can correctly classify points to geometric surfaces. It can be adjusted according to the point cloud density: 2mm-3mm when the point cloud density is ≥8 points / mm², and 3mm-5mm when the point cloud density is 5-8 points / mm². It is used to determine whether the spatial distance between two points meets the clustering requirements of the same surface (if the spatial distance between two points is ≤ the threshold, they are included in the same surface).

[0064] Minimum number of cluster points: ≥500 points, to avoid clustering noisy points into invalid geometric surfaces.

[0065] Step 2.3: Execution Segmentation and Optimization:

[0066] The region growing algorithm is started, and clustering is spread outward from the seed point. All point cloud data are traversed, and points that satisfy "the angle between normal vectors ≤ the threshold and the Euclidean distance ≤ the threshold" are included in the same geometric surface until no points that meet the conditions can be added.

[0067] After segmentation, two independent point cloud subsets are obtained—geometric surface A (surface to be welded 1) and geometric surface B (surface to be welded 2), which are automatically calculated by the system.

[0068] Standard deviation of normal vector, percentage of unassigned points (unassigned points refer to points that are not clustered to any geometric surface).

[0069] Segmentation result verification and optimization: If the proportion of unassigned points is >0.5%, or if there is oversegmentation of geometric surfaces (one surface is split into multiple subsets) or undersegmentation (two surfaces are not completely separated), then adjust the threshold of the angle between normal vectors or the threshold of Euclidean distance, and re-execute the segmentation operation until the requirements of "unassigned point proportion ≤0.5%, no oversegmentation, and no undersegmentation" are met.

[0070] Step 2.4 Segmentation result output:

[0071] Output the point cloud data of the segmented geometric surfaces A and B (preserving 3D coordinates and normal vector information), and display them through a 3D visualization interface. Different geometric surfaces are marked with different colors, making it easy for operators to intuitively confirm the segmentation effect.

[0072] Step 3: Intersection analysis.

[0073] The core objective of this step is to analyze the spatial relationship between geometric surfaces A and B, determine a reasonable projection reference plane and projection direction, and ensure the accuracy of subsequent projection calculations. Specifically, this includes:

[0074] Step 3.1: Geometric feature calculation:

[0075] Normal vector statistics: PCA (Principal Component Analysis) was used to calculate the normal vectors of the point cloud data of geometric surfaces A and B, respectively, to obtain the average normal vector of each geometric surface. , And the standard deviation of the normal vector (which reflects the smoothness of the geometric surface; the smaller the standard deviation, the smoother the surface).

[0076] Spatial Angle Calculation: Calculate the spatial angle θ between the two planes based on their average normal vectors. The formula is as follows: θ must satisfy 10°-170°: if θ≤10° or θ≥170°, it is judged as an approximately parallel plane, and the geometric plane segmentation result needs to be re-checked (there may be segmentation errors).

[0077] Step 3.2: Rules for determining the projection reference plane and projection direction:

[0078] The principle for selecting the reference plane is "plane priority, coplanar priority", that is, the geometric plane with the smaller standard deviation of the normal vector and the coplanarity of all welds is selected as the reference plane, and the other geometric plane is selected as the projection plane.

[0079] Projection direction determination: The projection direction is the direction of the average normal vector of the reference plane. This ensures that the projection lines are perpendicular to the reference plane, minimizing accuracy loss during the projection process.

[0080] Step 3.3: Parameter Recording and Display:

[0081] The system automatically records the markings of the datum plane and the projection plane, as well as the average normal vector of the datum plane. The spatial angle θ between the two planes is determined, and the projection direction is marked in the visualization interface to provide clear parameter basis for subsequent projection calculations.

[0082] Step 4: Threshold Filtering

[0083] This step is one of the core innovations of this invention. By setting a reasonable distance threshold, it filters out effective projection points in the projection surface (such as geometric surface A) that can reflect the true boundary of the weld, and eliminates invalid data caused by local deformation (i.e., eliminates deformation interference data), thus ensuring the accuracy of weld extraction from the source. Specifically, it includes:

[0084] Step 4.1: Logic and formula for calculating normal distance:

[0085] Calculate each point in the projection plane (such as geometric plane A). Normal distance relative to a reference plane (such as geometric plane B) The physical meaning of this distance is a point The vertical projection distance to the reference plane is calculated using the following formula:

[0086]

[0087] in, The coordinates of any point on the reference plane (such as geometric plane B) can be selected as the center point of the reference plane or any seed point. The average normal vector of the reference plane; during the calculation process, the system automatically traverses all points on the projection plane and calculates them one by one. It also generates a histogram of normal distance distribution to visually display the distance distribution characteristics.

[0088] Step 4.2: Method for setting the distance threshold T (quantization, adaptive):

[0089] The threshold T is the selection criterion for valid projection points. Its core function is to distinguish between "points in raised areas that can reflect the true boundary of the weld" and "invalid points in areas of local deformation and depression". The setting principle is to "cover the true boundary of the weld and eliminate deformation interference". The specific setting method is as follows:

[0090] Step 4.2.1: Basic Threshold Range: Divide the corrugated board into three ranges based on the processing accuracy requirements:

[0091] High-precision workpieces (machining error allowable range ±0.5mm, such as corrugated plates for chemical pressure vessels and core ship components): T=0.5mm-1.0mm.

[0092] For workpieces with ordinary precision (machining error allowable range ±1.0mm, such as ordinary steel structural parts and corrugated plates for general equipment shells): T=1.0mm-2.0mm.

[0093] For workpieces under special working conditions (with large deformation, measured deformation ≤3mm, such as corrugated plates for large steel structures): T=2.0mm-3.0mm (the accuracy of subsequent fitting needs to be improved simultaneously).

[0094] Step 4.2.2: Automatic Threshold Recommendation: The system has a built-in database containing threshold matching data for corrugated plates of different materials, specifications, and machining precision. These data are set based on empirical values. After the user inputs workpiece parameters (material, thickness, corrugation height, machining precision), the system automatically recommends the optimal threshold T by looking up a table. For example, if the user selects "Material: Q235B, Thickness: 8mm, Corrugation Height: 30mm, Machining Precision: Normal", the system automatically matches the closest threshold data.

[0095] Step 4.2.3: Manual fine-tuning function: Supports users to manually fine-tune the threshold T via the touch screen, with an adjustment step of 0.01mm. During the fine-tuning process, the number and proportion of effective projection points are displayed in real time, making it easy for users to intuitively judge the rationality of the threshold.

[0096] Step 4.3: Filtering of valid projection points:

[0097] Filtering rule: Traverse all points on the projection plane and filter those that meet the criteria. Points ≤ T are considered "valid projection points." These points are located on the projection plane within a region that is higher than the reference plane but within the normal error range, accurately reflecting the true intersection boundary between the projection plane and the reference plane; normal distances are excluded. Points with a value of >T are invalid data, mostly representing localized deformations or depressions. These points can blur the weld outline and interfere with extraction accuracy.

[0098] Step 4.4: Output of Filtering Results:

[0099] Output point cloud data of valid projection points (preserving 3D coordinates and normal distance information), and overlay it on the original point cloud model in the visualization interface. Valid projection points are marked with a highlight color to facilitate operators in confirming the filtering effect.

[0100] Step 5: Effective projection and fitting, output weld position.

[0101] The core objective of this step is to accurately project the effective projection points onto the reference plane. Through noise reduction and curve fitting, a contour curve that reflects the true shape of the weld is generated, thus obtaining the weld trajectory. Specifically, this includes the following steps:

[0102] Step 5.1: Effective Projection Calculation (Precise Projection):

[0103] The filtered valid projection points are aligned along the normal direction of the reference plane. Project vertically onto the reference plane, ensuring the projected point falls precisely on the reference plane without offset; Projected point The formula for calculating the coordinates is as follows:

[0104]

[0105]

[0106]

[0107] in, , , For valid projection points The original coordinates, for Normal distance to the reference plane, , , The average normal vector of the reference plane; after projection, a set of projected points is generated. And record the coordinate information of each projection point.

[0108] Step 5.2: Denoising the projected point cloud (removing noise interference):

[0109] A small number of isolated noise points may be introduced during the projection process (such as environmental noise during scanning or misjudged points during screening). These need to be removed through denoising to ensure the accuracy of the fitted curve. The RANSAC (Random Sample Consensus) algorithm is used for denoising. This algorithm can effectively separate interior points (points that conform to the weld contour features) from exterior points (noise points). The algorithm parameters are set as follows: number of iterations: ≥1000 times to ensure that all possible combinations of interior points are fully traversed.

[0110] Interior point threshold: 0.1mm, meaning that points that are ≤0.1mm away from the fitted straight line / curve are considered interior points.

[0111] Confidence level: ≥99%, ensuring the reliability of the denoising results.

[0112] After denoising is completed, the proportion of noise points (number of noise points / total number of projection points) should be counted. The proportion of noise points should be ≤1.5%. If the proportion exceeds this, the projection calculation or filtering steps should be checked again, and systematic errors should be eliminated before denoising is performed again.

[0113] Step 5.3: Weld contour fitting (precise modeling):

[0114] Based on the weld morphology (straight or curved) of the corrugated plate, select an appropriate fitting algorithm to ensure that the fitted curve can accurately reflect the true morphology of the weld.

[0115] Straight weld (intersection of adjacent planes of a planar corrugated plate, the weld is a straight line): The RANSAC fitting algorithm is used on the projected 3D planar point cloud, and the fitting goal is to find a straight line. (Represented by a point P on the line and a direction vector v), such that the sum of the squares of the distances from all interior points to the line is minimized; the fitting error must be ≤0.08mm (the fitting error is defined as the average distance from all interior points to the fitted line); where the formula for the distance from a point to a line is defined as: Where: P is a point on the line, v is the direction vector of the line, and Q is a point outside the line.

[0116] Curved welds (intersection of curved corrugated plates or deformed surfaces, with the weld being a curve): A cubic polynomial fitting algorithm is used on the projected 3D planar point cloud, with the fitting objective being to find a cubic polynomial curve. (or spatial cubic polynomial curve) The fitting error must be ≤0.1mm (the fitting error is defined as the average distance from all interior points to the fitted curve).

[0117] The formula for the distance from a point to a surface is defined as follows: Where X is a point outside the curve, and O is the point on the curve closest to X.

[0118] During the fitting process, the system calculates the fitting error in real time. If the error exceeds the allowable range, the fitting algorithm parameters (such as the polynomial order) are adjusted or noise is removed again until the error meets the standard.

[0119] Step 5.4: Output of fitting results:

[0120] Output the fitted weld contour curve C, record the mathematical model parameters of the curve (such as the slope and intercept of the straight line, the coefficients of the polynomial), the fitting error, the number of interior points, etc., and overlay them on the reference plane model in the visualization interface, highlighting curve C with a bright color, such as... Figure 5 As shown, this allows operators to visually view the fitting results.

[0121] Based on the same inventive concept, this invention also proposes an adaptive corrugated plate weld extraction system to implement the above method, including: a data acquisition module, a data processing module, a threshold setting module, and a result output module.

[0122] The data acquisition module serves as the high-precision data input terminal. Its core components include: a 2D / 3D laser scanner or depth camera, a gigabit Ethernet transmission unit, high-precision tooling fixtures, and a worktable. Its core functions include: rapidly and stably acquiring high-density, high-precision point cloud data from the corrugated plate surface; ensuring workpiece positioning accuracy through tooling fixtures; and ensuring data is transmitted to the data processing module without loss or delay through the high-speed transmission unit, providing reliable raw data for subsequent processing.

[0123] The data processing module, as the core computing unit of the system, consists of: an industrial-grade processor, high-performance memory, a high-speed solid-state drive, and a dedicated algorithm package.

[0124] The dedicated algorithm package is developed using a computer programming language and includes functional modules such as point cloud preprocessing (stitching, denoising), region growing and segmentation, normal distance calculation, thresholding, RANSAC denoising, and polynomial fitting. The calling logic is as follows: Figure 7 As shown.

[0125] It supports adjusting core algorithm parameters (such as segmentation threshold, distance threshold T, and fitting error threshold) through the threshold setting module to adapt to different working conditions. After receiving point cloud data from the data acquisition module, it automatically triggers the algorithm execution process without manual intervention, realizing automated calculation from data input to weld curve sampling point output.

[0126] The threshold setting module serves as a human-computer interaction unit. Its core component is the data interaction interface.

[0127] The core functions include: parameter input, threshold recommendation and adjustment, status display, and fault interaction.

[0128] Specifically, it supports users to input basic parameters of the corrugated board (material, thickness, corrugation height, dimensions), processing accuracy requirements, deformation estimation, and other information.

[0129] The system can automatically recommend a distance threshold T based on user-input parameters, while also supporting manual fine-tuning (adjustment step size of 0.01mm). During fine-tuning, the system displays the number and proportion of effective projection points and the preliminary fitting effect in real time. It can also display the system's operating status (standby, scanning, processing, completed, fault), current execution step, core parameter values, extraction accuracy indicators, and other information in real time, facilitating real-time monitoring by operators.

[0130] When the system malfunctions, the display screen shows the cause of the malfunction, troubleshooting suggestions, and handling steps, and users can confirm the malfunction handling results by pressing buttons.

[0131] The results output module serves as the data display and export terminal. Its core components include: a high-definition industrial display screen, a large-capacity storage unit, multiple data interfaces, and a report generation unit.

[0132] Among them, the high-definition industrial display screen is a 27-inch high-definition display screen with a resolution of 2560×1440 and a brightness of ≥500cd / m², which makes it easy for operators to intuitively view the location of the weld.

[0133] The large-capacity storage unit is a 2TB SSD solid-state drive, used for long-term storage of raw point cloud data, intermediate processing data, final weld data and extraction reports, which are archived in categories of "workpiece number-date-time" and have a storage period of ≥5 years.

[0134] The system features multiple data interfaces, including Gigabit Ethernet, USB 3.0, RS485, and HDMI, supporting data export, external device connection (such as printers and USB flash drives), and seamless integration with welding robot control systems. Core functions include visualization, data export, report generation, and data traceability. Specifically, it displays the original point cloud of the corrugated plate, the segmented geometric surfaces, effective projection points, and weld contour curves in real-time in 3D model form, supporting model rotation, scaling, and translation for easy multi-angle viewing. It supports exporting the final weld data to four standard formats: DXF, STL, IGES, and PLY, with an export speed ≥10MB / s, which can be directly imported into the welding robot control system. It automatically generates weld extraction quality reports, including basic workpiece information, equipment parameters, extraction steps, core parameters, accuracy indicators (fitting error, overlap, extraction error), and running time, supporting PDF saving and printing. It supports querying historical extraction data and reports by workpiece number, date, batch, etc., facilitating quality traceability and process optimization.

[0135] This invention abandons the assumption of ideal geometric surfaces and, based on real point cloud data of corrugated plates, combines threshold filtering and projection mechanisms to eliminate systematic deviations caused by deformation at their source. The following actual test data shows that the weld position extraction error of this invention can be strictly controlled within ±0.5mm, far superior to the 1-3mm deviation level of existing technologies, fully meeting the high-precision operation requirements of welding robots.

[0136]

[0137] This invention employs a parameterized and standardized extraction process, eliminating the need to develop new algorithms for corrugated plates of specific specifications or deformation amounts. It is thickness-adaptable, handling corrugated plates of varying thicknesses from 2 to 20 mm. It is compatible with various corrugation patterns, including sine waves, trapezoidal waves, and arc waves. It can tolerate localized deformations of 0.5-3 mm without requiring manual adjustment of the core algorithm. It is applicable to corrugated plate welding in multiple fields such as chemical engineering, shipbuilding, steel structures, and automotive manufacturing, demonstrating exceptional versatility.

[0138] The integrated system of this invention achieves full automation from data acquisition to weld output, requiring no manual intervention. The processing time for a single workpiece is ≤5 seconds, a 140-fold improvement over the existing technology's 10 minutes / workpiece, significantly shortening the production cycle. No professional algorithmic knowledge or geometric modeling experience is required from operators; only basic workpiece parameters need to be input to start the extraction process, reducing labor costs and operational barriers. The extraction results are highly consistent, unaffected by differences in operator experience, ensuring stable welding quality in mass production.

[0139] The hardware and software algorithms are packaged into a dedicated algorithm package, ensuring high stability. It supports seamless integration with mainstream welding robot control systems, eliminating the need for large-scale modifications to existing production lines. The operation process is simple and intuitive, conforming to the usage habits of workshop operators and facilitating rapid deployment and application. All selected products are mature industrial-grade products, ensuring controllable procurement costs and convenient maintenance.

[0140] At the algorithm level, the algorithm package can be upgraded to add features such as AI adaptive threshold models and simultaneous extraction of multi-faceted weld seams. At the hardware level, it can be compatible with higher-precision scanners (scanning accuracy ≥0.05mm) or depth cameras and faster processors to further improve extraction accuracy and efficiency. At the functional level, it can be expanded to include mobile monitoring, remote operation and maintenance, and data network management.

[0141] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for extracting weld seams from corrugated plates with adaptive deformation resistance, characterized in that, Includes the following steps: S1. Obtain three-dimensional point cloud data of the corrugated plate surface, wherein the point cloud data includes geometric deviation information caused by workpiece deformation; S2. Segment the two adjacent geometric surfaces to be welded from the point cloud data, and denote them as geometric surface A and geometric surface B; S3. Determine the spatial intersection relationship between geometric surface A and geometric surface B, and select one of the geometric surfaces as the reference surface and the other as the projection surface; S4. Calculate the normal distance from each sampling point on the projection surface to the reference surface, and select a set of valid projection points whose normal distance is not greater than the threshold T from the sampling points according to a preset distance threshold T used to distinguish weld boundary points from local deformation points; in step S4, the distance threshold T ranges from 0.5mm to 3.0mm; the distance threshold T is adaptively determined according to at least one parameter among the material, thickness, corrugation height, and processing accuracy requirements of the corrugated plate; the adaptive determination includes: querying a pre-stored parameter-threshold matching database according to the input workpiece parameters to output the recommended distance threshold T; in step S4, before or after selecting valid projection points, a step of denoising the sampling points on the projection surface or the set of valid projection points is also included; Calculate the normal distance of each point in the projection plane relative to the reference plane. The physical meaning of this distance is the perpendicular projection distance from the point to the reference plane; the threshold T is the screening criterion for valid projection points, which distinguishes between "points in the raised area that can reflect the true boundary of the weld" and "invalid points in the local deformation or depression area"; all points on the projection plane are traversed, and those that meet the criteria are selected. Points ≤ T are considered "valid projection points." These points are located on the projection plane above the reference plane but within the normal error range, accurately reflecting the true intersection boundary between the projection plane and the reference plane; normal distances are excluded. Points with a value greater than T are invalid data representing localized deformation or depression areas. S5. Project the points in the set of effective projection points onto the reference plane, and fit the projected points to obtain the weld trajectory.

2. The adaptive deformation-resistant corrugated plate weld extraction method according to claim 1, characterized in that, In step S1, point cloud data is acquired by a laser scanner or depth camera. The density of the point cloud data is not less than 5 points / mm², and the ranging error is not greater than 0.08mm.

3. The adaptive deformation-resistant corrugated plate weld extraction method according to claim 1, characterized in that, Step S2 employs a region growing segmentation algorithm, wherein the conditions for determining whether two points belong to the same geometric surface include: the angle between the normal vectors of the two points does not exceed 3°-8°, and the Euclidean distance between the two points does not exceed 1mm-3mm.

4. The adaptive deformation-resistant corrugated plate weld extraction method according to claim 1, characterized in that, In step S5, the RANSAC algorithm is used for linear fitting, or the least squares method is used for polynomial curve fitting.

5. An adaptive corrugated plate weld extraction system for implementing the method as described in any one of claims 1-4, characterized in that, include: The data acquisition module is used to acquire the three-dimensional point cloud data; The data processing module is used to perform point cloud segmentation, normal distance calculation, effective point selection based on threshold T, and trajectory fitting. The threshold setting module is used to provide functions for setting, recommending, or adjusting the distance threshold T; as well as The result output module is used to output the weld trajectory.

6. The adaptive corrugated plate weld extraction system according to claim 5, characterized in that, The data acquisition module includes a high-precision tooling fixture with a repeatability of no more than 0.05 mm; and / or, the threshold setting module includes a human-machine interface for displaying the impact of threshold adjustment on the number of effective projection points and the preview trajectory.

Images