A method and system for on-line detection of three-dimensional profile of laser weld of dissimilar thickness plate

Abstract

The application discloses a kind of online detection methods and systems of three-dimensional topography of dissimilar thickness plate laser weld, it is related to weld visual detection technical field, comprising: S1, collection dissimilar thickness plate profile detection data, and the pre-processing of dissimilar thickness plate profile detection data is executed;S2, constructs thick plate side local reference profile and thin plate side local reference profile, calculates double-reference coupling deviation value, determines weld topography action area, weld center position, thick plate side weld toe boundary position and thin plate side weld toe boundary position;S3, constructs three-dimensional topography reconstruction grid, extracts main recess connected region and calculates recess connected body volume value, recess connected length value and weld center trajectory swing value;S4, calculate topography abnormal evolution value, determine topography abnormal section and corresponding abnormal type, output dissimilar thickness plate laser weld online detection result.The problems that reference positioning is not accurate in the prior art dissimilar thickness plate laser weld online detection, leading to unstable three-dimensional topography anomaly identification and lack of reliable quantitative basis are solved.

Description

Technical Field

[0001] This invention relates to the field of weld visual inspection technology, specifically to an online method and system for three-dimensional morphology inspection of laser welds in plates of varying thicknesses. Background Technology

[0002] With the continuous development of laser welding, visual inspection, machine vision analysis, and 3D reconstruction technologies, online inspection of weld surface morphology has gradually evolved from simple two-dimensional image recognition to a comprehensive inspection method integrating structured light scanning, contour extraction, and 3D reconstruction. Current technologies typically use line lasers, industrial cameras, contour fitting, and feature point extraction to detect the weld surface forming state, weld toe position, contour height, and surface undulation features, thereby providing data support for welding quality evaluation.

[0003] For example, application CN121526993A discloses a method for visually inspecting the forming quality features of weld surfaces using three-dimensional reconstruction. This method includes: S1, a line laser emits line structured light to scan the weld at a uniform speed along the weld bead direction, while simultaneously using an industrial camera to acquire images of the line structured light stripes formed on the weld surface; S2, the acquired RGB images are converted frame by frame to grayscale images, and image distortion correction and image denoising are performed; S3, the Region of Interest (ROI) is located in the line structured light stripe image of the weld surface using a pixel grayscale distribution calculation method; S4, the ROI region of the weld line structured light stripe image is segmented; S5, the centerline of the ROI region of the weld line structured light stripe image is extracted at the sub-pixel level using a grayscale extreme value weighted centering method.

[0004] For example, application CN111738985B discloses a visual inspection method and system for weld contours, including: firstly, acquiring weld cross-sectional contour data; determining that no weld exists in the cross-section; if a weld exists, dividing the weld contour region into a feature point region and two weld toe point regions using a horizontal straight line l; calculating the coordinates of the feature points in the feature point region; based on the calculated weld feature point positions, the distance between the weld feature points and the center of the scanning device can be obtained, which is the distance the scanning device will move next; determining the two weld toe points in the two weld toe point regions using the farthest distance method; and obtaining the weld contour height based on the weld toe point positions; repeating the above process until the entire weld is tracked and scanned to obtain the entire weld contour and weld contour height.

[0005] However, most existing technologies focus on contour extraction, feature point recognition, and surface reconstruction in scenarios involving welds of equal thickness or near-symmetrical welds. They typically employ a unified cross-sectional benchmark, a single contour reference, or a fixed weld toe segmentation method to analyze the weld morphology. When the object of inspection is a laser weld of varying thickness, there are significant differences between the thick and thin plate sides in terms of base material steps, local slope differences, forming transitions, and contour offsets. If the inspection method based on a single benchmark or symmetry assumption is still used, it is easy to lead to inaccurate positioning of the weld morphology action zone, distortion of the three-dimensional reconstruction results, and unstable identification of abnormal sections, making it difficult to provide a reliable quantitative basis for the online inspection of laser welds of varying thickness.

[0006] Therefore, in order to address the above problems, there is an urgent need for an online method and system for detecting the three-dimensional morphology of laser weld seams in plates of varying thicknesses. Summary of the Invention

[0007] Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides an online method and system for detecting the three-dimensional morphology of laser welds in plates of varying thicknesses. This method solves the problems of inaccurate benchmark positioning in existing online detection of laser welds in plates of varying thicknesses, which leads to unstable identification of three-dimensional morphological anomalies and a lack of reliable quantitative evidence.

[0009] Technical solution

[0010] To achieve the above objectives, the present invention provides the following technical solution: an online detection method for the three-dimensional morphology of laser welds in plates of varying thicknesses, comprising: S1, constructing a workpiece coordinate system for the plate of varying thicknesses, synchronously acquiring contour detection data of the plate of varying thicknesses, performing preprocessing on the contour detection data of the plate of varying thicknesses, and outputting the preprocessed contour detection data of the plate of varying thicknesses; S2, identifying the base material transition zone based on the preprocessed contour detection data of the plate of varying thicknesses and constructing a local reference contour on the thick plate side and a local reference contour on the thin plate side, calculating the dual-reference coupling deviation value, and determining the weld morphology action area and weld center based on the dual-reference coupling deviation value. S3. Based on the weld morphology location results, construct a three-dimensional morphology reconstruction mesh, extract the main concave connected region, and calculate the volume value of the concave connected body, the concave connected length value, and the weld center trajectory swing value, outputting the spatial morphology feature data of the plate with varying thickness; S4. Based on the spatial morphology feature data of the plate with varying thickness, calculate the corresponding absolute difference value and perform dimensionless processing, calculate the morphology anomaly evolution value based on the dimensionless processing result, determine the morphology anomaly segment and the corresponding anomaly type, and output the online detection result of the laser weld of the plate with varying thickness.

[0011] Furthermore, the specific steps for constructing a workpiece coordinate system for the plate with varying thickness, simultaneously acquiring contour detection data of the plate with varying thickness, and performing preprocessing on the contour detection data are as follows: The weld extension direction of the plate with varying thickness is selected as the longitudinal reference direction; a direction perpendicular to the weld extension direction and located on the workpiece surface is selected as the transverse reference direction; and a direction perpendicular to both the transverse and longitudinal reference directions is selected as the height reference direction. The workpiece coordinate system is then constructed. Simultaneously, contour detection data of the plate with varying thickness is acquired, including laser emission time values, cumulative workpiece displacement values, transverse coordinate values ​​of the preceding base material contour points, height coordinate values ​​of the preceding base material contour points, transverse coordinate values ​​of the subsequent weld contour points, and the height coordinate values ​​of the subsequent weld wheel. The system outputs preprocessed profile detection data, including contour height coordinates, pre-detection interval values, post-detection interval values, detection head pitch angle, detection head roll angle, detection head lateral position, detection head longitudinal position, and detection head height position. For the acquired profile detection data of the varying thickness plate, a precise time protocol time synchronization algorithm is used to perform multi-source acquisition time alignment processing; a linear interpolation algorithm is used to perform profile sequence resampling processing under different sampling cycles; a median absolute deviation anomaly detection algorithm is used to perform profile outlier data identification processing; a sliding median filtering algorithm is used to perform local burr noise suppression processing; and a rigid body coordinate transformation algorithm is used to perform coordinate unification processing between the detection coordinate system and the workpiece coordinate system, outputting the preprocessed profile detection data of the varying thickness plate.

[0012] Further, the specific steps for identifying the transition zone of the base material and constructing local reference contours for the thick plate side and thin plate side based on the preprocessed profile detection data of the plate with varying thickness are as follows: Read the preprocessed profile detection data of the plate with varying thickness, arrange all detection records in ascending order of the cumulative displacement value of the workpiece travel, and use the cumulative displacement value of the workpiece travel as the longitudinal registration benchmark. Combined with the previous detection interval value and the subsequent detection interval value, convert the previous base material contour and the subsequent weld contour to the same longitudinal position corresponding to the laser action position, and then perform co-position section pairing processing so that each workpiece's cumulative displacement value corresponds to a set of previous base material contours and a set of subsequent weld contours. For each workpiece's cumulative displacement value corresponding to the previous base material contour, arrange the height coordinate values ​​of the previous base material contour points in ascending order of the lateral coordinate values, use a sliding window linear fitting algorithm to calculate the contour slope at each lateral position, and then perform differential calculation on the contour slopes at adjacent lateral positions. The calculation process involves defining the lateral segment where the absolute value of the slope difference exceeds the slope abrupt change threshold at N consecutive lateral sampling points as the base material transition zone. The corresponding front base material contours for the lateral segments on either side of the base material transition zone are then defined as the left and right base material contours, respectively. The average height of the front base material contour points corresponding to the left and right base material contours is compared, and the side with the larger average height is defined as the thicker plate side base material contour, while the side with the smaller average height is defined as the thinner plate side base material contour. The lateral coordinates and height coordinates of the front base material contour points corresponding to the thicker plate side base material contour, as well as those corresponding to the thinner plate side base material contour, are extracted. A least-squares linear fitting algorithm is used to construct the local reference contours for the thicker and thinner plates. The lateral position with the largest absolute value of the slope difference of the front base material contours within the base material transition zone is taken as the corresponding lateral position value of the segmented reference boundary.

[0013] Further, the dual-reference coupling deviation value is calculated, and the weld morphology action zone, weld center position, thick plate side weld toe boundary position, and thin plate side weld toe boundary position are determined based on the dual-reference coupling deviation value. The specific steps for outputting the weld morphology positioning result are as follows: The lateral coordinate value and height coordinate value of the post-weld contour point are correlated with the local reference contours on the thick plate side and the thin plate side. The absolute values ​​of the differences between the height values ​​of the post-weld contour and the corresponding height values ​​of the local reference contours on the thick plate side are taken, and the absolute values ​​of the differences between the height values ​​of the post-weld contour and the corresponding height values ​​of the local reference contours on the thin plate side are taken. The two absolute values ​​of the difference are multiplied and the square root is taken to obtain the dual-side coupling deviation term; the segmented reference edges are... The absolute value of the difference between the lateral position value corresponding to the boundary and the lateral position value corresponding to the subsequent weld contour is added to a constant to obtain the boundary distance constraint term. The double-base coupling deviation term is divided by the boundary distance constraint term to obtain the double-base coupling deviation value. The double-base coupling deviation value at each lateral position is read sequentially along the lateral direction. The lateral segment where the double-base coupling deviation value exceeds the action zone determination threshold at K consecutive lateral sampling points is determined as the weld morphology action zone. The lateral position with the largest double-base coupling deviation value in the weld morphology action zone is determined as the weld center position. The left boundary position of the weld morphology action zone is determined as the weld toe boundary position on the thick plate side. The right boundary position of the weld morphology action zone is determined as the weld toe boundary position on the thin plate side. The weld morphology positioning result is output.

[0014] Further, the specific steps for constructing a three-dimensional topography reconstruction mesh based on the weld topography positioning results are as follows: Read the weld topography positioning results and the lateral coordinates and height coordinates of the subsequent weld topography points corresponding to the cumulative displacement values ​​of each workpiece. Arrange all detection records in ascending order of the cumulative displacement values ​​of the workpiece, and determine the lateral segment between the weld toe boundary on the thick plate side and the weld toe boundary on the thin plate side as the weld topography reconstruction area. For the weld topography reconstruction area corresponding to each workpiece's cumulative displacement value, use a linear interpolation algorithm at a uniform lateral sampling interval to resample the cross-sectional contour formed by the lateral coordinates and height coordinates of the subsequent weld topography points. Use the cumulative displacement value of the workpiece as the longitudinal coordinate, the resampled lateral position as the lateral coordinate, and the corresponding height coordinate of the subsequent weld topography point as the height coordinate to construct cross-sectional mesh points. Perform longitudinal splicing on each cross-sectional mesh point according to the order of the cumulative displacement values ​​of the workpiece, and use a piecewise cubic spline interpolation algorithm to fill in the missing mesh points between adjacent cross-sectional mesh points to form a three-dimensional topography reconstruction mesh.

[0015] Further, the specific steps for extracting the main concave connected region and calculating the volume value, length value, and center trajectory swing value of the concave connected body, and outputting the spatial morphological feature data of the plate with varying thickness are as follows: On the 3D morphological reconstruction mesh, calculate the difference between the height coordinate value of the post-weld contour point and the corresponding local reference contour height value for each mesh point; mark mesh points with differences less than the concave determination threshold as concave mesh points; and use a 3D connected domain marking algorithm to perform connected domain segmentation on spatially adjacent concave mesh points, determining the connected domain containing the most concave mesh points as the main concave connected region; read the absolute value of the difference corresponding to each mesh point within the main concave connected region, multiply it by the area of ​​the corresponding mesh unit, and then perform accumulation. Add the volume value of the concave connected body; calculate the distance between the starting and ending positions of the main concave connected region in the direction of the cumulative displacement value of the workpiece to obtain the concave connected length value; read the transverse coordinate values ​​corresponding to the weld center position of each section in the order of the cumulative displacement value of the workpiece, construct the weld center trajectory fitting line using the least squares linear fitting algorithm, and calculate the absolute value of the transverse deviation between the transverse coordinate value corresponding to the weld center position of each section and the weld center trajectory fitting line, and determine the maximum value of the absolute value of the transverse deviation as the weld center trajectory swing value; extract the volume value of the concave connected body, the concave connected length value and the weld center trajectory swing value, and output the spatial morphology feature data of the plate with different thickness.

[0016] Further, the specific steps for calculating the corresponding absolute difference values ​​and performing dimensionless processing based on the spatial morphology feature data of plates with varying thicknesses are as follows: Read the spatial morphology feature data of plates with varying thicknesses, and calculate the absolute difference values ​​of the volume, length, and center trajectory oscillation values ​​of the concave connected body between the current detection record and the previous detection record, respectively, according to the order of the cumulative displacement values ​​of the workpiece. Using the minimum-maximum normalization algorithm, perform dimensionless processing on the volume, length, and center trajectory oscillation values ​​of the concave connected body, the volume difference, length difference, and center trajectory oscillation values ​​of the concave connected body, respectively, to obtain the normalized values ​​of volume, length, oscillation, volume difference, length difference, and oscillation difference.

[0017] Furthermore, the specific steps for calculating the morphological anomaly evolution value based on the dimensionless processing results are as follows: For the current detection record corresponding to the cumulative displacement value of each workpiece, multiply the volume normalized value and the length normalized value, add one, and then take the natural logarithm to obtain the current spatial depression intensity value; add the volume difference normalized value and the length difference normalized value, take the square root, and then add one to obtain the spatial depression evolution gain value; add the swing normalized value and the swing difference normalized value, add one, and then take the natural logarithm to obtain the trajectory swing response value; then multiply the current spatial depression intensity value and the spatial depression evolution gain value, and add them to the trajectory swing response value to obtain the morphological anomaly evolution value.

[0018] Further, the specific steps for determining the abnormal morphology sections and corresponding abnormality types, and outputting the online detection results of laser welds for plates of varying thicknesses, are as follows: The section where the cumulative displacement value of the workpiece travels for Q consecutive detection records exceeds the abnormality judgment threshold is identified as the abnormal morphology section. The starting and ending cumulative displacement values ​​of the workpiece travel corresponding to each abnormal morphology section are respectively determined as the starting and ending positions of the abnormal morphology section. For each abnormal morphology section, the peak value of the normalized volume, the peak value of the normalized length, and the peak value of the swing are extracted. The peak value of the dynamic normalized value is calculated, and the peak values ​​of the volume normalized value, length normalized value, and oscillation normalized value are compared. When the peak value of the volume normalized value is the largest, the corresponding abnormal morphology segment is identified as the depression-dominant abnormal segment. When the peak value of the length normalized value is the largest, the corresponding abnormal morphology segment is identified as the continuously expanding abnormal segment. When the peak value of the oscillation normalized value is the largest, the corresponding abnormal morphology segment is identified as the trajectory oscillation abnormal segment. The starting position, ending position, and type of the abnormal morphology segment are extracted, and the online detection results of the laser weld seam of the plate with varying thickness are output.

[0019] The second aspect of this invention provides an online three-dimensional morphology detection system for laser welds in plates of varying thicknesses, comprising: a contour perception processing module, a dual-reference positioning module, a three-dimensional morphology reconstruction module, and an anomaly evolution output module. The contour perception processing module is used to construct a workpiece coordinate system for the plate of varying thicknesses, synchronously acquire contour detection data of the plate of varying thicknesses, perform preprocessing on the contour detection data, and output preprocessed contour detection data of the plate of varying thicknesses. The dual-reference positioning module is used to identify the base material transition zone based on the preprocessed contour detection data of the plate of varying thicknesses and construct local reference contours on the thick plate side and thin plate side, calculate the dual-reference coupling deviation value, and determine the weld shape based on the dual-reference coupling deviation value. The system includes four modules: a weld morphology detection module and a 3D morphology reconstruction module. The 3D morphology detection module is used to construct a 3D morphology reconstruction mesh based on the weld morphology detection results, extract the main depression connected region, and calculate the volume value, length value, and center trajectory swing value of the depression connected body, outputting spatial morphology feature data of the plate with varying thickness. The anomaly evolution output module is used to calculate the corresponding absolute difference value based on the spatial morphology feature data of the plate with varying thickness and perform dimensionless processing. Based on the dimensionless processing result, the module calculates the morphology anomaly evolution value, determines the morphology anomaly segment and the corresponding anomaly type, and outputs the online detection results of the laser weld of the plate with varying thickness.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) A method and system for online detection of three-dimensional morphology of laser weld seam in plate with different thicknesses. By introducing a matching mechanism of the corresponding cross sections of the pre-construction base material contour and the post-construction weld seam contour, and constructing local reference contours on the thick plate side and the thin plate side respectively in the case of plate with different thicknesses, it can avoid mistakenly incorporating the thickness difference of the base material, the slope difference of the plate surface and the local steps into the unified reference, thereby improving the pertinence of the weld seam morphology action area positioning, and establishing a double-sided local reference constraint mechanism for the detection of weld seam in plate with different thicknesses that is different from the detection method of plate with equal thickness.

[0023] (2) A method and system for online detection of three-dimensional morphology of laser weld seam in plate with different thicknesses. By calculating the deviation value of dual reference coupling, the deviation relationship between the post-weld contour and the local reference contour of the thick plate side and the local reference contour of the thin plate side is included in the same judgment process. Combined with the transverse position value corresponding to the segmented reference boundary, the weld morphology action area, the weld center position and the weld toe boundary position on both sides are jointly determined. This can improve the stability of the extraction of the weld center and boundary of plate with different thicknesses and avoid misleading the positioning results due to unilateral deviation or local contour change.

[0024] (3) A method and system for online detection of three-dimensional morphology of laser weld seam of plate with different thickness, which constructs a three-dimensional morphology reconstruction mesh based on the weld seam morphology positioning result, and further extracts the main depression connected area, the volume value of the depression connected body, the depression connected length value and the center trajectory swing value of the weld seam, can expand the single cross section contour anomaly into a spatial anomaly characterization result that is continuously distributed along the weld seam direction, so that the weld seam anomaly identification is improved from planar contour judgment to three-dimensional spatial defect identification, and the characterization ability of continuous depression expansion and center trajectory instability is enhanced.

[0025] (4) A method and system for online detection of three-dimensional morphology of laser weld seam of plate with different thickness. By performing dimensionless processing on the volume value of the concave connected body, the length value of the concave connected body, the oscillation value of the center trajectory of the weld seam and its corresponding absolute difference, and constructing the morphological anomaly evolution value based on the dimensionless processing result, the current spatial anomaly intensity, anomaly expansion trend and trajectory oscillation response can be incorporated into the same anomaly judgment process, realizing the quantitative output of the morphological anomaly section and its type, thereby improving the comparability, interpretability and engineering application value of the online detection results of laser weld seam of plate with different thickness. Attached Figure Description

[0026] Figure 1 Flowchart of an online method for detecting the three-dimensional morphology of laser weld seams in plates of varying thicknesses;

[0027] Figure 2 A structural diagram of an online three-dimensional morphology detection system for laser weld seams in plates of varying thicknesses;

[0028] Figure 3 Coupled zone diagram of abnormal evolution of weld morphology in plates of varying thickness;

[0029] Figure 4 This is a spatial correlation diagram of abnormal cross-sections of welds in plates of varying thicknesses. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figures 1-4This invention provides a technical solution: an online detection method for the three-dimensional morphology of laser welds in plates of varying thicknesses, comprising: S1, constructing a workpiece coordinate system for the plate of varying thicknesses, synchronously acquiring contour detection data of the plate of varying thicknesses, performing preprocessing on the contour detection data of the plate of varying thicknesses, and outputting the preprocessed contour detection data of the plate of varying thicknesses; S2, identifying the base material transition zone based on the preprocessed contour detection data of the plate of varying thicknesses and constructing a local reference contour on the thick plate side and a local reference contour on the thin plate side, calculating the dual-reference coupling deviation value, and determining the weld morphology action area, weld center position, and thickness based on the dual-reference coupling deviation value. S3. Based on the weld toe boundary positions on the plate side and the thin plate side, output the weld morphology positioning results; S4. Based on the weld morphology positioning results, construct a three-dimensional morphology reconstruction mesh, extract the main concave connected region and calculate the concave connected volume value, concave connected length value and weld center trajectory swing value, output the spatial morphology feature data of the plate with varying thickness; S5. Based on the spatial morphology feature data of the plate with varying thickness, calculate the corresponding difference absolute value and perform dimensionless processing, calculate the morphology anomaly evolution value based on the dimensionless processing result, determine the morphology anomaly segment and the corresponding anomaly type, and output the online detection results of the laser weld of the plate with varying thickness.

[0032] Specifically, the steps for constructing a workpiece coordinate system for a plate with varying thickness, simultaneously acquiring contour detection data of the plate with varying thickness, and performing preprocessing on the contour detection data are as follows: The weld extension direction of the plate with varying thickness is selected as the longitudinal reference direction; a direction perpendicular to the weld extension direction and located on the workpiece surface is selected as the transverse reference direction; and a direction perpendicular to both the transverse and longitudinal reference directions is selected as the height reference direction. A workpiece coordinate system is then constructed. The projection point of the weld's initial sampling position onto the workpiece surface is used as the origin. A longitudinal coordinate axis is established using the longitudinal reference direction; a transverse coordinate axis is established using the transverse reference direction; and a height coordinate axis is established using the height reference direction. Inspection is completed using an installation calibration plate. Initial extrinsic parameter calibration from the measuring coordinate system to the workpiece coordinate system ensures that all subsequently acquired contour points have a unified spatial reference. Simultaneously, contour detection data for plates with varying thicknesses is acquired, including laser emission time values, cumulative workpiece displacement values, lateral coordinate values ​​of the preceding base material contour points, height coordinate values ​​of the preceding base material contour points, lateral coordinate values ​​of the subsequent weld contour points, height coordinate values ​​of the subsequent weld contour points, preceding detection interval values, subsequent detection interval values, detection head pitch angle values, detection head roll angle values, detection head lateral position values, detection head longitudinal position values, and detection head height position values. The laser emission time value is written in real-time by the laser trigger signal acquisition circuit during each emission, and the cumulative workpiece displacement... The displacement values ​​are continuously output by a displacement encoder fixed to the conveying mechanism. The lateral and height coordinates of the preceding base material contour points are calculated by a structured light contour acquisition device located in front of the laser action position, scanning the surface of the unmelted base material. The lateral and height coordinates of the following weld contour points are calculated by a structured light contour acquisition device located behind the laser action position, scanning the surface of the formed weld. The preceding detection interval is taken as the installation distance from the preceding scanning center to the laser action position in the longitudinal reference direction, and the following detection interval is taken as the installation distance from the following scanning center to the laser action position in the longitudinal reference direction. The detection head pitch angle and detection head roll angle are also included. The lateral, longitudinal, and height positions of the detection head are output in real time by an attitude sensor and linear displacement feedback device mounted on the moving mechanism of the detection head. The pre-detection interval and post-detection interval are used to uniformly convert the pre-sampled base material contour and post-sampled weld contour to the same longitudinal reference position corresponding to the laser action position. The pitch angle, roll angle, lateral, longitudinal, and height positions of the detection head are used to eliminate the influence of changes in the installation posture and movement deviation of the detection head on the coordinates of the contour points. For the collected contour detection data of plates with varying thicknesses, a precise time protocol time synchronization algorithm is used to perform multi-source acquisition time alignment processing.Specifically, using the laser emission time value as a unified time anchor point, the output timing of the displacement encoder, the sampling timing of the preceding base material contour, the sampling timing of the subsequent weld contour, and the feedback timing of the detection head attitude displacement are mapped to the same time reference. This ensures that the cumulative displacement value of the workpiece travel, the lateral coordinate value of the preceding base material contour point, the height coordinate value of the preceding base material contour point, the lateral coordinate value of the subsequent weld contour point, the height coordinate value of the subsequent weld contour point, the pitch angle value of the detection head, the roll angle value of the detection head, the lateral position value of the detection head, the longitudinal position value of the detection head, and the height position value of the detection head have a consistent timing relationship within the same sampling cycle. A linear interpolation algorithm is used to perform contour sequence resampling processing under different sampling cycles. Specifically, a unified displacement sampling time is formed using the cumulative displacement value of the workpiece travel. A sampling grid is used to interpolate the lateral coordinates, height coordinates, and subsequent weld contour points of the preceding and following materials onto a unified displacement sampling grid. This ensures a one-to-one correspondence between the preceding and following contour sequences at the same longitudinal position, preventing misalignment in subsequent transition zone identification due to inconsistent original sampling step distances. The sampling interval of the unified displacement sampling grid is 0.8 to 1.2 times the median of the cumulative displacement values ​​of adjacent original workpieces, preferably 1.0 times. This value is chosen to maintain the same order of the resampled contour point density as the original sampling density, avoiding excessively large sampling intervals that could lead to the loss of local weld toe transitions, and excessively small sampling intervals that could result in redundant interpolation point accumulation. The median absolute value is used. The deviation anomaly detection algorithm performs contour outlier data identification processing. Specifically, using the contour point height distribution within a local displacement window as a statistical sample, it calculates the deviation of the height coordinates of the preceding base material contour points and the subsequent weld contour point height coordinates relative to the local median. Sampling points with deviations exceeding a preset multiple threshold are marked as outliers, thereby filtering out isolated anomalies caused by spatter reflection, smoke obstruction, and momentary defocusing. The local displacement window length is 5 to 11 times the uniform displacement sampling interval, preferably 7 times; the preset multiple threshold is 2.5 to 4.0, preferably 3.0. The local displacement window length is chosen to ensure that the statistical sample simultaneously covers both the locally stable base material segment and the locally changing weld toe segment; the preset multiple threshold is chosen to suppress outliers. While eliminating spike noise, the true step and depression boundaries are preserved. A sliding median filtering algorithm is used to suppress local burr noise. Specifically, the median result within the continuous displacement window replaces the spike-like fluctuations near the identified outliers, ensuring the integrity of true geometric changes such as weld toe transitions, base material steps, and local depressions, thus preventing burrs from being misjudged as base material transitions during subsequent local reference contour construction. The length of the sliding median filtering window is 3 to 7 times the uniform displacement sampling interval, preferably 5 times, chosen to suppress high-frequency burrs while avoiding excessive flattening of the weld toe boundary slope. A rigid body coordinate transformation algorithm is used to unify the coordinates between the detection coordinate system and the workpiece coordinate system, outputting preprocessed profile detection data for the plate with varying thickness.Specifically, a rotation matrix and translation vector are constructed based on the pitch angle, roll angle, lateral position, longitudinal position, and height of the detection head. The lateral and height coordinates of the preceding base material contour points, the following coordinates are mapped from the detection coordinate system to the workpiece coordinate system. Then, longitudinal position compensation is performed by combining the preceding and following detection interval values, resulting in preprocessed profile detection data of the differential thickness plate expressed in a unified workpiece coordinate system. Among these, the longitudinal... During position compensation, the pre-detection interval value and post-detection interval value are directly used as the longitudinal position translation reference. The compensation direction is determined by the installation relationship between the detection head and the laser action position. After the above processing, the pre-detection base material contour reflects the true geometric boundary of the base material before laser action, and the post-detection weld contour reflects the true formed surface after laser action. The two can be directly compared under the same longitudinal reference position, providing a feasible data foundation for subsequent base material transition zone identification, construction of local reference contours on the thick plate side, construction of local reference contours on the thin plate side, and calculation of dual-reference coupling deviation values.

[0033] In this implementation scheme, by first constructing a workpiece coordinate system, and then forming unified preprocessed profile detection data for uneven thickness plates based on the laser emission time value, cumulative workpiece travel displacement value, lateral coordinate value of the preceding base material profile point, height coordinate value of the preceding base material profile point, lateral coordinate value of the following weld profile point, height coordinate value of the following weld profile point, preceding detection interval value, following detection interval value, detection head pitch angle value, detection head roll angle value, detection head lateral position value, detection head longitudinal position value, and detection head height position value, this invention enables the preceding base material profile and the following weld profile to have a directly comparable data basis under the same spatial reference. It can simultaneously maintain the integrity of real geometric changes such as base material steps, weld toe bends, and local depressions, and reduce the interference caused by acquisition timing deviation, sampling step difference, attitude disturbance, and local burr noise on subsequent base material transition zone identification and local reference profile construction, thereby improving the consistency, stability, and reliability of the profile data on which the subsequent dual-reference coupling deviation value calculation depends.

[0034] Specifically, the steps for identifying the transition zone of the base material and constructing local reference contours for the thick plate and thin plate sides based on the pre-processed profile detection data of the plate with varying thickness are as follows: Read the pre-processed profile detection data of the plate with varying thickness, arrange all detection records in ascending order of the cumulative displacement value of the workpiece travel, and use the cumulative displacement value of the workpiece travel as the longitudinal registration benchmark. Combined with the pre-detection interval value and the post-detection interval value, convert the pre-detection base material contour and the post-detection weld contour to the same longitudinal position corresponding to the laser action position. Then, perform co-position section pairing processing, so that each workpiece's cumulative displacement value corresponds to a set of pre-detection base material contours and a set of post-detection weld contours. The pre-detection interval value is used to translate the sampling position of the pre-detection base material contour along the longitudinal reference direction to the laser action position, and the post-detection interval value is used to translate the sampling position of the post-detection weld contour along the longitudinal reference direction to the laser action position. After position conversion, the pre-detection base material contour reflects the true geometric state of the base material before laser action, and the post-detection weld contour reflects the true geometric state of the formed surface after laser action. Both have a directly comparable spatial basis at the same longitudinal position, thus avoiding... Misalignment of the front and rear sampling centers caused the base material steps to be misjudged as weld anomalies. For the preceding base material contour corresponding to the cumulative displacement value of each workpiece's travel, the height coordinates of the preceding base material contour points were arranged from smallest to largest according to their lateral coordinate values. A sliding window linear fitting algorithm was used to calculate the contour slope at each lateral position. Then, the contour slope at adjacent lateral positions was calculated by difference. The lateral segment where the absolute value of the slope difference exceeded the slope mutation threshold at N consecutive lateral sampling points was determined as the base material transition zone. The preceding base material steps corresponding to the lateral segments on both sides of the base material transition zone were then identified. The profiles of the parent material are defined as the left and right parent material profiles, respectively. The sliding window linear fitting algorithm selects continuous sampling points centered on the current lateral position to fit a local straight line. The length of the fitting window is 5 to 11 times the uniform lateral sampling interval, preferably 7 times. This value is chosen to ensure that the window simultaneously covers locally stable parent material segments while maintaining sensitivity to the transitions of parent material steps. The profile slope is taken as the slope value of the fitted straight line. The difference in profile slope at adjacent lateral positions is used to characterize the intensity of local surface slope abrupt changes. The slope abrupt change threshold is 0.08 to 0.25, preferably 0.15, the value is chosen to ensure that the natural slope fluctuation of the gently sloping parent material is below the threshold, and the abrupt slope of the transition step between thick and thin plates is above the threshold; N is 3 to 7, preferably 5, the value is chosen to suppress the identification of pseudo-transition zones triggered by single-point noise, while ensuring that the real parent material transition boundary forms a stable response on multiple continuous transverse sampling points; compare the average height of the preceding parent material contour points corresponding to the left and right parent material contours, and determine the parent material contour on the side with the larger average height as the thicker plate side parent material contour, and the parent material contour on the side with the smaller average height as the thinner plate side parent material contour; where, height The mean value is calculated based on the height coordinates of all preceding base material contour points within each side's transverse section. The principle behind using the mean value for discrimination is that the overall height reference of the base material surface on the thicker side is higher than that on the thinner side. This allows the left-right spatial position to be mapped to the semantic position of the thicker side and the thinner side, avoiding confusion caused by directly defining the thicker and thinner sides based solely on the transverse left-right direction, which can lead to problems with clamping direction changes. The transverse coordinates and height coordinates of the preceding base material contour points corresponding to the thicker side base material contour are extracted separately, as are the transverse coordinates and height coordinates of the preceding base material contour points corresponding to the thinner side base material contour. Using the coordinate values ​​and height coordinate values ​​of the preceding base material contour points, a least-squares linear fitting algorithm is employed to construct local reference contours for the thick plate side and the thin plate side. The lateral position with the largest absolute value of the slope difference between the preceding base material contours within the base material transition zone is taken as the corresponding lateral position value of the segmented reference boundary. Specifically, the least-squares linear fitting algorithm fits all lateral coordinate values ​​and height coordinate values ​​within the thick plate side base material contour to obtain the intercept parameters and slope parameters of the local reference contour for the thick plate side. The same process is then performed on the thin plate side base material contour to obtain the intercept parameters and slope parameters of the local reference contour for the thin plate side. Rate parameters; the principle of linear fitting is that the local surface of the base material outside the transition zone approximately satisfies the stable plane section characteristics within a small range. Using the local reference profile, the intrinsic slope difference of the base material can be separated from the post-weld profile. The lateral position with the largest absolute value of the slope difference between the preceding base material profiles within the transition zone is taken as the corresponding lateral position value of the segmented reference boundary. This is based on the fact that this position corresponds to the point of maximum geometric abrupt change when transitioning from the thick plate side base material surface to the thin plate side base material surface. Using this point as the boundary datum for the dual-datum local reference profiles provides a stable boundary reference for subsequent calculation of the dual-datum coupling deviation value.

[0035] In this implementation scheme, by using the cumulative displacement value of the workpiece travel as a unified longitudinal registration benchmark, the outline of the preceding base material and the outline of the subsequent weld are converted to the same longitudinal position corresponding to the laser action position. Then, based on the identification results of the base material transition zone, the outlines of the thick plate side base material and the thin plate side base material are distinguished. Furthermore, the lateral position values ​​corresponding to the local reference outlines of the thick plate side, the local reference outlines of the thin plate side, and the segmented reference boundaries are constructed. This decouples the intrinsic step features, local slope difference features, and post-weld forming deviation features of the base material of different thicknesses under the same spatial benchmark. This avoids the erroneous inclusion of the inherent geometric differences at the junction of thick and thin plates into the weld anomaly characterization process. Thus, it provides a reference basis with direction discrimination capability, boundary constraint capability, and geometric separation capability for the subsequent calculation of the dual-benchmark coupling deviation value, improving the pertinence, stability, and reliability of the weld morphology action zone positioning results.

[0036] Specifically, the steps for calculating the dual-reference coupling deviation value, determining the weld morphology action zone, weld center position, thick plate side weld toe boundary position, and thin plate side weld toe boundary position based on the dual-reference coupling deviation value, and outputting the weld morphology positioning result are as follows: The lateral coordinate values ​​and height coordinate values ​​of the post-weld contour points are correlated with the local reference contours on the thick plate side and thin plate side in the same lateral position. Specifically, when the lateral position is correlated, the lateral coordinate value of the post-weld contour point is used as the current lateral sampling position, and reference height values ​​corresponding to the current lateral sampling position are extracted from the local reference contours on the thick plate side and thin plate side respectively. When the current lateral sampling position does not fall on the original fitted sampling point, the corresponding reference height value is directly calculated along the fitted line of the local reference contour, ensuring that each post-weld contour point corresponds to a unique height value for the local reference contour on the thick plate side and a unique height value for the local reference contour on the thin plate side, thus guaranteeing that the dual-reference coupling deviation value calculation is based on the same lateral position. The height values ​​of the post-weld contour points are correlated with the local reference contours on the thick plate side and thin plate side. The absolute value of the difference between the corresponding height values ​​of the contours is taken. The absolute value of the difference between the height value corresponding to the subsequent weld contour and the height value corresponding to the local reference contour on the thin plate side is also taken. The square root of the two absolute values ​​is then multiplied to obtain the bilateral coupling deviation term. The absolute value is used to eliminate the sign difference caused by the weld contour being above or below the local reference contour, so that the deviation amplitudes of both sides can participate in the coupling calculation at the same scale. The square root of the multiplication is used to produce a significant response when the subsequent weld contour deviates from the local reference contour on the thick plate side and the local reference contour on the thin plate side at the same time, making it difficult for a single-sided isolated deviation to form a high value, thus highlighting the common deviation feature of both sides. The absolute value of the difference between the lateral position value corresponding to the segmented reference boundary and the lateral position value corresponding to the subsequent weld contour is taken, and a constant 1 is added to obtain the boundary distance constraint term. The lateral position value corresponding to the segmented reference boundary represents the geometric boundary position between the local reference contour on the thick plate side and the local reference contour on the thin plate side, the absolute value of the difference represents the boundary distance between the current lateral sampling position and the boundary position, and the constant 1 is set to 1.0, a constant is used to avoid the denominator becoming zero when the current lateral sampling position coincides with the corresponding lateral position value of the segment reference boundary; the dual-reference coupling deviation term is divided by the boundary distance constraint term to obtain the dual-reference coupling deviation value; among which, the dual-reference coupling deviation value is more sensitive to the common deviation of both sides near the boundary position, and the response to local isolated deviation weakens when far from the boundary position. This computational mechanism can be used to distinguish the local disturbance outside the inherent step of the base material of the plate with varying thickness from the true contour deviation inside the weld morphology action zone; each lateral position is read sequentially along the lateral direction. The dual-reference coupling deviation value at the location is used to define the transverse segment where the deviation value exceeds the threshold for the action zone at K consecutive transverse sampling points as the action zone of the weld morphology. K is set to 3 to 7, preferably 5, to suppress false high-value segments caused by single-point abnormal spikes, while ensuring that the true weld morphology action zone forms a stable over-threshold response at multiple consecutive transverse sampling points. The action zone determination threshold is set to 0.05 to 0.30, preferably 0.12, to ensure that the deviation value exceeds the threshold for the action zone in the base metal stable zone and the slightly disturbed zone outside the step. If the baseline coupling deviation value is lower than the threshold for the action zone, the double baseline coupling deviation value generated by the weld contour significant deviation zone is higher than the threshold for the action zone. The lateral position with the largest double baseline coupling deviation value within the weld morphology action zone is determined as the weld center position. The weld center position corresponds to the position with the most significant common deviation on both sides, which is used to characterize the dominant center of the weld forming deviation in the current section. The left boundary position of the weld morphology action zone is determined as the weld toe boundary position on the thick plate side, and the right boundary position of the weld morphology action zone is determined as the weld toe boundary position on the thin plate side. The left boundary position is taken as the starting lateral position of the continuous over-threshold lateral segment, and the right boundary position is taken as the ending lateral position of the continuous over-threshold lateral segment. In this way, the weld toe boundary is defined as the position where the weld contour changes from significant deviation to non-significant deviation. The weld morphology positioning result is output. The weld morphology positioning result includes at least the weld morphology action zone, the weld center position, the weld toe boundary position on the thick plate side, and the weld toe boundary position on the thin plate side. It can be directly used for three-dimensional morphology reconstruction mesh construction, extraction of the main concave connected region, and calculation of the weld center trajectory swing value. .

[0037] The specific formula for calculating the dual-reference coupling deviation is as follows:

[0038] ;

[0039] In the formula, This represents the dual-reference coupling deviation value at the j-th lateral position corresponding to the cumulative displacement value of the same workpiece. This represents the height value corresponding to the profile of the subsequent weld at the j-th lateral position. This represents the height value corresponding to the local reference profile of the thick plate side at the j-th lateral position. This represents the height value corresponding to the local reference profile of the thin plate side at the j-th lateral position. This represents the lateral position value corresponding to the profile of the subsequent weld at the j-th lateral position. This indicates the lateral position value corresponding to the segment reference boundary.

[0040] In this implementation scheme, by simultaneously placing the post-weld contour within the dual constraints of the local reference contours on the thick plate side and the thin plate side at the same lateral position, and then combining the lateral position values ​​corresponding to the segmented reference boundaries to perform boundary-sensitive adjustment of the lateral position distribution, the weld contour deviation judgment in the visual inspection process no longer relies on a single-sided reference relationship or a single height difference response. Instead, it forms a dual-benchmark positioning basis for the geometric imbalance characteristics of the cross-section of plates with varying thicknesses. This can more effectively distinguish the influence of the inherent steps of the base material and the influence of the actual weld forming anomalies, improving the positioning targeting and boundary judgment stability of the weld morphology action area, weld center position, thick plate side weld toe boundary position, and thin plate side weld toe boundary position. This provides a reliable cross-sectional positioning basis for subsequent three-dimensional morphology reconstruction mesh construction, extraction of the main concave connected region, and calculation of the weld center trajectory swing value.

[0041] Specifically, the steps for constructing a 3D morphology reconstruction mesh based on the weld morphology positioning results are as follows: Read the weld morphology positioning results and the lateral coordinates and height coordinates of the subsequent weld contour points corresponding to the cumulative displacement values ​​of each workpiece. Arrange all detection records in ascending order of the cumulative displacement values ​​of the workpieces. Define the lateral segment between the weld toe boundary on the thick plate side and the weld toe boundary on the thin plate side as the weld morphology reconstruction area. The weld morphology reconstruction area is limited to the area between the weld toe boundary on the thick plate side and the weld toe boundary on the thin plate side. This is to constrain the lateral range participating in the 3D reconstruction within the weld morphology action area, avoiding the introduction of the stable base material segment outside the weld morphology action area into the reconstruction process. The subsequent 3D topography reconstruction mesh is designed to accurately reflect the actual spatial changes corresponding to weld formation undulations, local depressions, and center offsets. For the weld topography reconstruction area corresponding to the cumulative displacement value of each workpiece, a linear interpolation algorithm is used to resample the cross-sectional profile formed by the lateral coordinates and height coordinates of the subsequent weld contour points under a uniform lateral sampling interval. The cross-sectional mesh points are constructed using the cumulative displacement value of the workpiece as the longitudinal coordinate, the resampled lateral position as the lateral coordinate, and the corresponding height coordinate of the subsequent weld contour point as the height coordinate. The uniform lateral sampling interval is 0.8 to 1.2 times the median difference between the lateral coordinates of adjacent original subsequent weld contour points, preferably 1.The value of 0 times is chosen to ensure that the density of lateral points after resampling is of the same order as the original contour sampling density. This avoids the loss of local concave edges and weld toe turning positions due to excessively large lateral sampling intervals, and avoids excessively small lateral sampling intervals that lead to overly dense interpolation points and amplify local numerical jitter. During linear interpolation, the lateral coordinate values ​​of two adjacent original post-weld contour points are used as the boundary of the interpolation interval. Within the interpolation interval, the height coordinate values ​​of the post-weld contour points corresponding to each resampled lateral position are calculated according to a uniform lateral sampling interval. This ensures that the weld morphology reconstruction area corresponding to the cumulative displacement value of the same workpiece forms a discretized cross-sectional contour under a uniform lateral reference. The principle of using the cumulative displacement value of the workpiece travel as the longitudinal coordinate is to directly represent the front-to-back relationship of the cross-section with the actual travel distance of the workpiece along the weld direction, so that the subsequent longitudinal splicing results maintain the actual weld extension distance attribute; the principle of using the resampled lateral position as the lateral coordinate is to eliminate the impact of differences in the number of lateral sampling points and the distribution of lateral sampling positions between different inspection records on the consistency of the three-dimensional mesh; the principle of using the corresponding post-weld contour point height coordinate value as the height coordinate is to maintain the direct mapping relationship of the geometric undulation features of the weld surface in three-dimensional space; the longitudinal coordinates of each cross-section mesh point are processed according to the order of the cumulative displacement value of the workpiece travel. The mesh is stitched together, and a piecewise cubic spline interpolation algorithm is used to fill in the missing mesh points between adjacent cross-sections, forming a 3D topography reconstruction mesh. Specifically, during longitudinal stitching, the difference in the cumulative displacement of the workpiece travel corresponding to the contours of two adjacent cross-sections is used as the longitudinal spacing, and the lateral position corresponding to the unified lateral sampling interval is used as the lateral registration reference. The mesh points of the previous cross-section are connected to the mesh points of the next cross-section in the same lateral position sequence to form the initial mesh skeleton. When there are missing height values, sampling breakpoints, or local occlusion blanks at the same lateral position between two adjacent cross-sections, a piecewise cubic spline interpolation algorithm is used to fill in the missing mesh points in the longitudinal direction. The missing grid points between adjacent cross-sections ensure that the 3D mesh maintains continuous surface characteristics along the weld extension direction. The piecewise cubic spline interpolation algorithm is chosen because it can preserve the local undulation trend while ensuring the continuity of the first and second derivatives of the surface, making it suitable for expressing the spatial morphological changes of the weld surface, such as gradual bulges, continuous depressions, and local rises, along the continuous travel direction. The spacing between the cumulative displacement values ​​of the workpieces traveling in the longitudinal interpolation between adjacent cross-sections is preferably 1 times the uniform longitudinal sampling interval. The uniform longitudinal sampling interval is 0.8 to 1.2 times the median of the difference between the cumulative displacement values ​​of adjacent original workpieces, preferably 1.The value of 0 times is chosen to ensure that the longitudinal sampling density remains consistent with the density of the original detection record. This avoids the spatial defects being flattened due to excessively sparse longitudinal sampling spacing, and the interpolation redundancy caused by excessively dense longitudinal sampling spacing. After the above processing, the resulting 3D topography reconstruction mesh possesses a continuous longitudinal mesh structure, a unified transverse mesh structure, and the ability to represent true height undulations. This provides a directly usable spatial discretization basis for subsequent extraction of the main depression connectivity region, calculation of the volume value of the depression connectivity body, calculation of the depression connectivity length value, and calculation of the weld center trajectory oscillation value.

[0042] In this implementation scheme, by mapping the weld morphology reconstruction area defined by the weld morphology positioning results to a unified lateral sampling interval and a unified longitudinal registration relationship, the lateral coordinate values ​​and height coordinate values ​​of the subsequent weld contour points corresponding to the cumulative displacement values ​​of different workpieces during visual inspection can form a continuous three-dimensional representation within the same spatial discrete frame. This transforms the originally scattered cross-sectional contour information into a three-dimensional morphology reconstruction mesh with a stable spatial correspondence, thereby enhancing the spatial recognizability of local undulations, continuous depressions, and longitudinal extension anomalies on the weld surface. This improves the consistency, integrity, and reliability of the spatial foundation upon which subsequent main depression connectivity region extraction, depression connectivity volume value calculation, depression connectivity length value calculation, and weld center trajectory swing value calculation depend.

[0043] Specifically, the steps for extracting the main concave connected region and calculating the volume, length, and center trajectory swing values ​​of the concave connected body, and outputting the spatial morphological feature data of the plate with varying thickness are as follows: On the 3D morphological reconstruction mesh, calculate the difference between the height coordinates of the subsequent weld contour point and the corresponding local reference contour height value at each mesh point. Mark mesh points with differences less than the concave determination threshold as concave mesh points. The difference is obtained by subtracting the local reference contour height value from the height coordinates of the subsequent weld contour point corresponding to the current mesh point. A difference less than the concave determination threshold indicates that the subsequent weld contour at the current mesh point is lower than the local reference contour to a preset concavity level. This determination relationship can be used to identify the locally concave areas on the weld surface. The depression is separated from the overall reconstructed surface; the depression judgment threshold is set between -0.05 and -0.30, preferably -0.12. The value is chosen so that the height difference caused by slight surface fluctuations is higher than the depression judgment threshold, while the height difference within the actual depression boundary is lower than the depression judgment threshold, thereby suppressing the interference of local noise fluctuations on depression identification; and a three-dimensional connected component labeling algorithm is used to perform connected component segmentation on spatially adjacent depression grid points, and the connected component containing the most depression grid points is determined as the main depression connected region; wherein, spatial adjacency is determined based on the grid connection relationship of the current depression grid point in the horizontal direction, vertical direction, and diagonal adjacent direction as the connectivity basis, and the three-dimensional connected component labeling algorithm identifies the connectable grid points by traversing each grid point. The principle behind grouping concave mesh points into the same connected domain and using the connected domain containing the most concave mesh points as the main concave connected region is that real weld defects typically manifest as continuously expanding dominant concave regions in space, while isolated local concave points usually only form small-scale discrete connected domains. This allows for the differentiation between dominant defect regions and scattered pseudo-defect regions. The absolute value of the difference between each mesh point within the main concave connected region is read, multiplied by the area of ​​the corresponding mesh cell, and then accumulated to obtain the volume of the concave connected volume. The area of ​​the corresponding mesh cell is obtained by multiplying a uniform horizontal sampling interval by a uniform vertical sampling interval. The absolute value of the difference characterizes the concavity depth of the current concave mesh point relative to the local reference contour. The absolute value of the difference is then multiplied by the area of ​​the corresponding mesh cell. Multiplying the area of ​​each individual concave grid cell yields the volume contribution value of the concave area. Accumulating the volume contribution values ​​of all concave grid cells yields the volume of the concave connected volume, thus quantitatively characterizing the scale of the spatial concave area. The distance between the starting and ending positions of the main concave connected region along the direction of the cumulative workpiece displacement is calculated to obtain the concave connected length. The starting position corresponds to the position with the smallest cumulative workpiece displacement within the main concave connected region, and the ending position corresponds to the position with the largest cumulative workpiece displacement within the main concave connected region. The distance between the starting and ending positions characterizes the continuous expansion range of the main concave connected region along the weld extension direction, thereby reflecting the degree of longitudinal spread of the concave defect.The transverse coordinate values ​​corresponding to the weld center positions of each section are read sequentially according to the cumulative displacement values ​​of the workpiece. A least-squares linear fitting algorithm is used to construct a weld center trajectory fitting line, and the absolute value of the transverse deviation between the transverse coordinate values ​​corresponding to the weld center positions of each section and the weld center trajectory fitting line is calculated. The maximum value of the absolute value of the transverse deviation is determined as the weld center trajectory oscillation value. The weld center trajectory fitting line is constructed using the transverse coordinate values ​​of the weld center positions corresponding to the cumulative displacement values ​​of each workpiece as fitting samples. The least-squares linear fitting algorithm is used to extract the overall trend line of the weld center along the workpiece's travel direction, and the absolute value of the transverse deviation is used to characterize the current section's weld center position relative to the overall trend. The principle of determining the maximum value of the absolute value of the lateral deviation as the weld center trajectory oscillation value lies in using the maximum deviation amplitude to characterize the most significant oscillation intensity of the weld center trajectory within a local segment, thus providing trajectory instability characteristics for subsequent calculation of morphological anomaly evolution values. The volume value of the concave connected body, the length value of the concave connected body, and the weld center trajectory oscillation value are extracted and output as spatial morphological feature data of plates with varying thicknesses. This spatial morphological feature data is used to jointly characterize the scale of the weld space concave, the extension range of the weld space concave, and the deviation intensity of the weld center trajectory, enabling the subsequent anomaly judgment process to perform comprehensive identification based on spatial volume characteristics, longitudinal extension characteristics, and trajectory oscillation characteristics simultaneously.

[0044] In this implementation scheme, by unifying the local concave regions, longitudinal extension range, and weld center offset trajectory in the three-dimensional topography reconstruction mesh into the volume value of the concave connected body, the length value of the concave connected body, and the swing value of the weld center trajectory, the spatial anomalies of welds in plates of varying thicknesses no longer remain at the level of a single cross-sectional profile, but form a continuous quantitative expression for the three-dimensional topography instability characteristics. This can more effectively distinguish between local sporadic fluctuations and continuous defect expansion, improve the ability of the spatial topography feature data of plates of varying thicknesses to characterize the true degree of anomalies, and provide a reliable basis with spatial scale characteristics, longitudinal extension characteristics, and trajectory instability characteristics for subsequent calculation of topography anomaly evolution values.

[0045] Specifically, the steps for calculating the corresponding absolute difference based on the spatial morphology feature data of plates with varying thicknesses and performing dimensionless processing are as follows: Read the spatial morphology feature data of plates with varying thicknesses, and calculate the absolute difference in the volume of the concave connected body, the absolute difference in the length of the concave connected body, and the absolute difference in the oscillation value of the weld center trajectory between the current detection record and the previous detection record, according to the order of the cumulative displacement values ​​of the workpiece. The current detection record uses the spatial morphology feature data of the plate with varying thicknesses corresponding to the current cumulative displacement value of the workpiece, and the previous detection record uses the spatial morphology feature data of the plate with varying thicknesses immediately preceding the current detection record in the order of the cumulative displacement values ​​of the workpiece. The absolute difference in the volume of the concave connected body is calculated by subtracting the volume of the concave connected body corresponding to the current detection record. The absolute value of the volume of the concave connected body corresponding to the previous detection record is obtained by subtracting the concave connected length value corresponding to the previous detection record from the concave connected length value corresponding to the current detection record, and the absolute value of the difference in the weld center trajectory sway value is obtained by subtracting the weld center trajectory sway value corresponding to the previous detection record from the weld center trajectory sway value corresponding to the current detection record. These absolute differences are used to construct a measure of the intensity of change between adjacent detection records. The purpose is to explicitly introduce the spatial anomaly change trend corresponding to the cumulative displacement value of continuous workpiece movement into the subsequent calculation process of morphological anomaly evolution value, so that the current spatial anomaly level and the change amplitude of adjacent positions can participate in anomaly identification together in the same judgment link. For the first detection record in the sequence of cumulative displacement values ​​of the workpiece, the absolute value of the difference in the volume of the concave connected body, the absolute value of the difference in the length of the concave connected body, and the absolute value of the difference in the oscillation value of the weld center trajectory are all set to 0. This ensures that the first detection record has complete input in subsequent dimensionless processing and calculation of morphological anomaly evolution values. The minimum-maximum normalization algorithm is used to perform dimensionless processing on the volume of the concave connected body, the length of the concave connected body, the oscillation value of the weld center trajectory, the absolute value of the difference in the volume of the concave connected body, the absolute value of the difference in the length of the concave connected body, and the absolute value of the difference in the oscillation value of the weld center trajectory, respectively, to obtain the normalized values ​​for volume, length, oscillation, volume difference, and length difference. The normalized values ​​are: normalized value and swing difference value; among them, the minimum-maximum normalization algorithm uses the minimum value of a certain feature quantity in the same batch of detection records as the lower bound and the maximum value of a certain feature quantity in the same batch of detection records as the upper bound, mapping the corresponding feature quantity of each detection record to the range of 0 to 1; the volume normalized value is obtained by subtracting the minimum volume value of the concave connected body in all detection records from the volume value of the concave connected body corresponding to the current detection record, and then dividing by the difference between the maximum and minimum volume values ​​of the concave connected body in all detection records; the length normalized value is obtained by subtracting the minimum concave connected body length value in all detection records from the length value of the concave connected body corresponding to the current detection record, and then dividing by the difference between the maximum and minimum length values ​​of the concave connected body in all detection records.The normalized value for oscillation is obtained by subtracting the minimum oscillation value of the weld center trajectory from the oscillation value of the current inspection record, and then dividing by the difference between the maximum and minimum oscillation values ​​of the weld center trajectory from all inspection records. The normalized value for volume difference is obtained by subtracting the minimum absolute value of the volume difference of the concave connected body from the absolute value of the volume difference of the concave connected body from all inspection records, and then dividing by the difference between the maximum and minimum absolute values ​​of the volume difference of the concave connected body from all inspection records. The normalized value for length difference is obtained by subtracting the minimum absolute value of the length difference of the concave connected body from the absolute value of the length difference of the concave connected body from all inspection records, and then dividing by the difference between the maximum and minimum absolute values ​​of the length difference of the concave connected body from all inspection records. The normalized value for oscillation difference is obtained by subtracting the minimum absolute value of the oscillation value difference of the weld center trajectory from the absolute value of the oscillation value of the current inspection record, and then dividing by the difference between the maximum and minimum absolute values ​​of the length difference of the concave connected body from all inspection records. The difference between the maximum and minimum absolute values ​​of the weld center trajectory oscillation values ​​in all inspection records is used to obtain the value. When the maximum and minimum values ​​of a certain feature are the same in all inspection records, the normalization result of the corresponding inspection record is uniformly set to 0 to avoid the denominator being zero. The principle of using the minimum-maximum normalization algorithm is to uniformly convert the volume values, length values, weld center trajectory oscillation values, and corresponding absolute differences of concave connected bodies with different dimensions, numerical ranges, and variation amplitudes into comparable dimensionless quantities. This ensures that volume features, length features, trajectory features, and variation features maintain a uniform scale in subsequent natural logarithm calculations, square root calculations, multiplicative coupling calculations, and additive response calculations, avoiding the dominance of a single feature in the overall anomaly judgment result due to its large original numerical magnitude. After the above processing, the obtained normalized values ​​of volume, length, oscillation, volume difference, length difference, and oscillation difference can be directly used for subsequent calculation of morphological anomaly evolution values.

[0046] In this implementation scheme, the volume value of the concave connection body, the length value of the concave connection body, the oscillation value of the weld center trajectory, and their corresponding absolute difference values ​​are uniformly converted into normalized values ​​of volume, length, oscillation, volume difference, length difference, and oscillation difference. This enables different spatial morphological features to have a quantitative basis for direct comparison at the same scale, avoiding the bias of subsequent judgment results caused by differences in the magnitude of the original numerical values. This allows the spatial morphological feature data of plates with varying thicknesses to characterize both the current anomaly intensity and the variation range between adjacent detection records, thereby improving the comprehensive characterization ability of subsequent morphological anomaly evolution value calculations for the degree of spatial anomaly, anomaly expansion trend, and trajectory instability.

[0047] Specifically, the steps for calculating the morphological anomaly evolution value based on the dimensionless processing results are as follows: For the current detection record corresponding to the cumulative displacement value of each workpiece, multiply the volume normalized value by the length normalized value, add one, and then take the natural logarithm to obtain the current spatial depression intensity value; whereby the volume normalized value is used to characterize the scale of the spatial depression corresponding to the main depression connected area, and the length normalized value is used to characterize the longitudinal extension of the main depression connected area along the direction of the cumulative displacement value of the workpiece. The principle of multiplying the volume normalized value by the length normalized value is to generate a synchronous amplification response when the scale of the spatial depression increases and the longitudinal extension increases, making it difficult for local small-scale depressions to form a high-intensity judgment. The principle of adding one to the product result is to ensure that the input value of the natural logarithm function is always greater than zero. The principle of taking the natural logarithm again is to perform compression mapping on larger spatial concavity composite features, making the spatial concavity intensity value monotonically increase with the enhancement of spatial anomalies, while avoiding excessive dominance of the volume value of the maximally concave connected volume on the overall morphological anomaly evolution value. The spatial concavity evolution gain value is obtained by adding the volume difference normalized value and the length difference normalized value, taking the square root, and then adding one. The volume difference normalized value is used to characterize the change in the spatial concavity scale of the current detection record relative to the previous detection record, and the length difference normalized value is used to characterize the change in the spatial concavity scale of the current detection record relative to the previous detection record. The principle of recording the variation in longitudinal extension and adding the two is to unify the spatial concavity variation characteristics into the same evolutionary quantity. The principle of taking the square root of the sum is to perform a smoothing process on the abruptly increasing differential characteristics, so that the differential change remains sensitive to the amplification effect of subsequent morphological anomaly evolution values ​​without excessive amplification. The principle of adding one is to make the spatial concavity evolution gain value a base gain starting from 1, so that the current spatial concavity intensity value maintains its original weight when there is no significant change, and obtains additional gain when there is a significant change. The trajectory oscillation response value is obtained by adding the oscillation normalized value and the oscillation differential normalized value, adding one, and then taking the natural logarithm. Among them, the oscillation normalized value is used as... To characterize the absolute instability of the weld center trajectory oscillation value under the current detection record, the oscillation difference normalized value is used to characterize the change amplitude of the weld center trajectory oscillation value of the current detection record relative to the previous detection record. The principle of adding the two is to simultaneously consider the current offset intensity of the weld center trajectory and the change intensity of adjacent positions. The principle of adding one is to ensure that the input value of the natural logarithm function is always greater than zero. The principle of taking the natural logarithm is to ensure that the trajectory oscillation response value maintains a distinguishable response to small oscillation changes and maintains controlled growth to large oscillation changes. Then, the current spatial depression intensity value is multiplied by the spatial depression evolution gain value and added to the trajectory oscillation response value to obtain the morphological anomaly evolution value.The principle behind multiplying the current spatial depression intensity value with the spatial depression evolution gain value is to synchronously amplify the current spatial depression anomaly level when there is a significant evolution trend, thus creating a coupled response between the spatial anomaly intensity and the amplitude of spatial anomaly changes. The principle behind adding the trajectory oscillation response values ​​is to directly superimpose the weld center trajectory instability characteristics onto the spatial depression composite anomaly, enabling the morphological anomaly evolution value to simultaneously characterize the spatial scale characteristics, longitudinal evolution characteristics, and weld center trajectory offset characteristics of the main depression connectivity area. After the above calculations, the morphological anomaly evolution value shows a synchronously increasing response to the increase in spatial depression scale, the enhancement of spatial depression longitudinal expansion, and the intensification of weld center trajectory oscillation, and can be directly used for subsequent morphological anomaly segment identification and morphological anomaly segment type determination.

[0048] The specific formula for calculating the morphological anomaly evolution value is as follows:

[0049] ;

[0050] In the formula, This represents the morphological anomaly evolution value corresponding to the cumulative displacement value of the t-th workpiece. This represents the normalized volume value corresponding to the cumulative displacement of the t-th workpiece. This represents the normalized length value corresponding to the cumulative displacement of the t-th workpiece. This represents the normalized value of the oscillation corresponding to the cumulative displacement of the t-th workpiece. This represents the normalized volume difference between the cumulative displacement value of the t-th workpiece and the cumulative displacement value of the (t-1)-th workpiece. This represents the normalized length difference between the cumulative displacement value of the t-th workpiece and the cumulative displacement value of the (t-1)-th workpiece. This represents the normalized value of the swing difference between the cumulative displacement value of the t-th workpiece and the cumulative displacement value of the (t-1)-th workpiece.

[0051] In this implementation scheme, by incorporating the normalized values ​​of volume, length, sway, volume difference, length difference, and sway difference into the same calculation process for morphological anomaly evolution, the spatial morphological feature data of plates with varying thicknesses is no longer merely described as static features. Instead, it forms a unified quantitative result that simultaneously reflects the current level of spatial anomaly, the changing trends of adjacent detection records, and the degree of instability of the weld center trajectory. This improves the ability of morphological anomalies to characterize the continuous evolution process, enhances the distinguishability of anomaly levels between different detection records, and provides a more stable quantitative basis for subsequent identification of morphological anomaly segments and determination of morphological anomaly segment types.

[0052] In this embodiment, Table 1 is a data table of morphological anomaly evolution values, which lists exemplary values ​​used to illustrate the calculation process of morphological anomaly evolution values ​​under five consecutive detection records along the workpiece travel direction. Specifically: In detection record 1, the volume normalization value is 0.18, the length normalization value is 0.22, the swing normalization value is 0.10, the volume difference normalization value is 0.00, the length difference normalization value is 0.00, the swing difference normalization value is 0.00, and the morphological anomaly evolution value is 0.1341; In detection record 2, the volume normalization value is 0.30, the length normalization value is 0.35, the swing normalization value is 0.16, the volume difference normalization value is 0.12, the length difference normalization value is 0.13, the swing difference normalization value is 0.06, and the morphological anomaly evolution value is 0.3486; In detection record 3, the volume normalization value is 0.55, the length normalization value is 0.58, the swing normalization value is 0.28, and the volume difference... The normalized value is 0.25, the normalized value of length difference is 0.23, the normalized value of swing difference is 0.12, and the morphological anomaly evolution value is 0.8052; in detection record 4, the normalized value of volume is 0.72, the normalized value of length is 0.80, the normalized value of swing is 0.45, the normalized value of volume difference is 0.17, the normalized value of length difference is 0.22, the normalized value of swing difference is 0.17, and the morphological anomaly evolution value is 1.2214; in detection record 5, the normalized value of volume is 0.60, the normalized value of length is 0.75, the normalized value of swing is 0.38, the normalized value of volume difference is 0.12, the normalized value of length difference is 0.05, the normalized value of swing difference is 0.07, and the morphological anomaly evolution value is 0.8963. As shown in Table 1, among the five detection records continuously collected along the workpiece's travel direction, there are significant differences in the normalized values ​​of volume, length, oscillation, and corresponding differential normalized values ​​for each detection record. This leads to significant differences in the morphological anomaly evolution values ​​corresponding to each detection record. Among them, the morphological anomaly evolution value corresponding to detection record 4 is the largest, indicating that the spatial depression intensity, depression expansion trend, and weld center trajectory oscillation response of the weld section corresponding to this detection record have the strongest combined effect, and it is more likely to correspond to the section with the most significant morphological anomaly.

[0053] Table 1. Data on the evolution of morphological anomalies

[0054]

[0055] like Figure 3As shown in the figure, the horizontal axis represents the cumulative displacement value of the workpiece, the vertical axis represents the amplitude of the anomaly evolution coupling band, and the color intensity of the band indicates the magnitude of the anomaly evolution value at the corresponding location. Anomaly evolution coupling band is constructed along the direction of the cumulative displacement value of the workpiece. This band synchronously characterizes the combined effect of the current spatial depression intensity, spatial depression evolution gain, and trajectory oscillation response on the anomaly evolution value within the same strip-shaped region. As shown in the figure, as the cumulative displacement of the workpiece increases from 100 to 160, the bandwidth of the abnormal evolution coupling band gradually increases and the color of the band gradually deepens. This indicates that the spatial depression intensity, depression expansion trend, and weld center trajectory oscillation in the corresponding section are all enhanced, and the morphological anomaly evolution value continues to rise. Among them, the morphological anomaly evolution value corresponding to the cumulative displacement of the workpiece at 160 reaches the maximum, indicating that the weld section at this position is the section with the most significant morphological anomaly. When the cumulative displacement of the workpiece continues to increase to 180, the bandwidth and color intensity of the abnormal evolution coupling band decrease compared to 160, but are still significantly higher than the section from 100 to 140. This indicates that the anomaly does not appear instantaneously and in isolation, but rather forms a continuous evolution along the weld extension direction before showing a decay trend. Figure 3 It can intuitively characterize the enhancement, peak appearance and fall process of abnormal morphology of laser weld seams in plates of varying thickness along the workpiece travel direction, and can be used to locate and identify abnormal morphology sections.

[0056] like Figure 4 As shown in the figure, the horizontal axis represents the lateral position, the vertical axis represents the height coordinate value, the dashed line represents the segmented local reference contour, the solid line represents the post-weld contour, the light-colored filled area represents the projection segment of the main concave connected area on the current section, and the vertical dotted line represents the segmented reference boundary. The left side of the figure is the local reference area on the thick plate side, and the right side is the local reference area on the thin plate side. The position of the weld toe boundary on the thick plate side is marked in the local reference area on the thick plate side, and the position of the weld toe boundary on the thin plate side is marked in the local reference area on the thin plate side. The center position of the weld is marked inside the weld morphology effect area. As can be seen from the figure, the post-weld contour deviates significantly from the segmented local reference contour on both sides of the segmented reference boundary. The contour on the thick plate side is raised more significantly, while the thin plate side shows a concave section that is lower than the local reference contour. The concave section corresponds to the projection position of the main concave connected area on the cross section, indicating that the three-dimensional spatial concave defect has a clear landing point on the cross section. At the same time, the weld center is located between the weld toe boundary on the thick plate side and the weld toe boundary on the thin plate side, which can intuitively reflect the center distribution relationship inside the weld morphology effect area. Figure 4 It can intuitively characterize the cross-sectional profile deviation features of welds of different thickness plates at the abnormal peak position, the bilateral local reference constraint relationship, and the spatial correspondence between the main concave connected area and the abnormal section of the cross-section. It can be used to explain the technical process of performing anomaly identification based on the dual-reference positioning results and the three-dimensional topography reconstruction results.

[0057] Specifically, the steps for determining abnormal morphology sections and corresponding abnormality types, and outputting the online detection results of laser weld seams in plates of varying thickness are as follows: The section where the cumulative displacement value of the workpiece exceeds the abnormality judgment threshold for Q consecutive detection records is identified as an abnormal morphology section. The starting and ending cumulative displacement values ​​of the workpiece corresponding to each abnormal morphology section are respectively determined as the starting and ending positions of the abnormal morphology section. Here, Q ranges from 3 to 7, preferably 5. The value is chosen to ensure that the determination of abnormal morphology sections is based on the common threshold exceeding by consecutive detection records, avoiding direct triggering of abnormal section identification when a single detection record is affected by local reflection fluctuations, instantaneous attitude disturbances, or local interpolation errors. To ensure that genuine and continuous anomalies can form a stable response within a finite longitudinal span, the anomaly judgment threshold is set between 0.35 and 0.95, preferably 0.60. The threshold is chosen so that the anomaly evolution value corresponding to the stable forming section is lower than the anomaly judgment threshold, while the anomaly evolution value corresponding to the section with continuously increasing spatial depression and significant trajectory oscillation is higher than the anomaly judgment threshold. When the anomaly evolution value of Q consecutive detection records exceeds the anomaly judgment threshold, the cumulative displacement value of the workpiece corresponding to the first exceeding threshold detection record is determined as the starting position of the anomaly section, and the cumulative displacement value of the workpiece corresponding to the last consecutive exceeding threshold detection record is determined as the ending position of the anomaly section, ensuring that the anomaly section has clear start and end boundaries in the longitudinal direction. For each morphologically abnormal segment, the peak values ​​of volume normalization, length normalization, and oscillation normalization are extracted and compared. Specifically, the volume normalization peak value is the maximum value among all volume normalization values ​​within the current morphologically abnormal segment; the length normalization peak value is the maximum value among all length normalization values ​​within the current morphologically abnormal segment; and the oscillation normalization peak value is the maximum value among all oscillation normalization values ​​within the current morphologically abnormal segment. The principle behind peak comparison is to use the most significant spatial anomaly feature within the segment as the basis for anomaly type determination, focusing anomaly type identification on the dominant instability manifestation of the current segment, rather than being influenced by weak response detection within the segment. The measurement record is diluted; when the peak value of the volume normalized value is the largest, the corresponding morphological anomaly segment is identified as the depression-dominant anomaly segment; where the peak value of the volume normalized value is the largest, it indicates that the spatial depression scale corresponding to the main depression connected area is dominant in the current segment, and the current anomaly is mainly manifested as a significant increase in the local depression volume; when the peak value of the length normalized value is the largest, the corresponding morphological anomaly segment is identified as the continuously expanding anomaly segment; where the peak value of the length normalized value is the largest, it indicates that the extension range of the main depression connected area along the cumulative displacement value direction of the workpiece travel is dominant in the current segment, and the current anomaly is mainly manifested as the continuous expansion of the spatial depression along the weld direction; when the peak value of the oscillation normalized value is the largest, the corresponding morphological anomaly segment is identified as the trajectory oscillation anomaly segment;Among them, the largest peak value of the oscillation normalized value indicates that the oscillation value of the weld center trajectory dominates in the current segment, and the current anomaly is mainly manifested as a significant shift of the weld center position relative to the overall trajectory. When there are the same maximum values ​​among the peak values ​​of volume normalized value, length normalized value, and oscillation normalized value, the type of morphological anomaly segment is determined in the order of priority of oscillation normalized value peak value, followed by length normalized value peak value, and then volume normalized value peak value. The basis for the value selection is that the deviation of the weld center trajectory usually directly reflects the instability of the forming path, the continuous expansion anomaly usually reflects the longitudinal propagation of defects, and the depression-dominated anomaly usually reflects the local spatial volume anomaly. The starting position, ending position, and type of morphological anomaly segment are extracted, and the online detection results of laser welds of varying thickness plates are output. Among them, the online detection results of laser welds of varying thickness plates are used to characterize the current weld anomaly occurrence interval, anomaly spatial range, and anomaly dominant mechanism in the longitudinal direction, so that subsequent quality judgment, process backtracking, and handling decisions have directly callable quantitative results.

[0058] In this implementation scheme, by transforming the morphological anomaly evolution value into an online detection result of laser weld seams in plates of varying thickness with a clear start position, end position, and anomaly type, the abnormal response in the continuous detection record during the visual inspection process can be improved from a single-point over-threshold state to a segmented judgment result. This can more effectively distinguish between local instantaneous fluctuations and continuous abnormal processes, enhance the boundary clarity of morphological anomaly segments in the longitudinal range, and improve the discrimination between depression-dominated anomaly segments, continuously expanding anomaly segments, and trajectory swinging anomaly segments. As a result, the online detection results of laser weld seams in plates of varying thickness have stronger interpretability, traceability, and applicability.

[0059] like Figure 2As shown, the second aspect of the present invention provides an online three-dimensional morphology detection system for laser weld seams of plates with varying thicknesses, comprising: a contour perception processing module, a dual-reference positioning module, a three-dimensional morphology reconstruction module, and an anomaly evolution output module, wherein: the contour perception processing module is used to construct the workpiece coordinate system of the plate with varying thicknesses, synchronously acquire contour detection data of the plate with varying thicknesses, perform preprocessing on the contour detection data of the plate with varying thicknesses, and output the preprocessed contour detection data of the plate with varying thicknesses; the dual-reference positioning module is used to identify the transition zone of the base material based on the preprocessed contour detection data of the plate with varying thicknesses and construct a local reference contour on the thick plate side and a local reference contour on the thin plate side, calculate the dual-reference coupling deviation value, and determine the anomaly evolution output based on the dual-reference coupling deviation value. The system defines the weld morphology action zone, weld center position, weld toe boundary position on the thick plate side, and weld toe boundary position on the thin plate side, and outputs the weld morphology positioning results. The 3D morphology reconstruction module constructs a 3D morphology reconstruction mesh based on the weld morphology positioning results, extracts the main concave connected region, and calculates the concave connected volume value, concave connected length value, and weld center trajectory swing value, outputting spatial morphology feature data for plates of varying thicknesses. The anomaly evolution output module calculates the corresponding absolute difference value based on the spatial morphology feature data of plates of varying thicknesses and performs dimensionless processing. Based on the dimensionless processing result, it calculates the morphology anomaly evolution value, determines the morphology anomaly segment and corresponding anomaly type, and outputs the online detection results of laser welds on plates of varying thicknesses.

[0060] In this implementation scheme, by sequentially connecting the pre-processed contour detection data of differential thickness plates, weld morphology positioning results, spatial morphology feature data of differential thickness plates, and online detection results of differential thickness plate laser welds along the same technical link, the visual inspection process of differential thickness plate laser welds has a continuous closed-loop relationship from contour acquisition, dual-reference positioning, three-dimensional morphology characterization to abnormal section output. This enables the unified processing of base material geometric differences, post-weld spatial concavity features, and weld center trajectory instability features under the same online detection framework, thereby improving the data connectivity, judgment stability, and result consistency of the online detection process of differential thickness plate laser weld three-dimensional morphology.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0062] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for online detection of the three-dimensional morphology of laser weld seams in plates of varying thicknesses, characterized in that, Includes the following steps: S1, construct the workpiece coordinate system of the plate with varying thickness, synchronously collect the contour detection data of the plate with varying thickness, perform preprocessing on the contour detection data of the plate with varying thickness, and output the preprocessed contour detection data of the plate with varying thickness. S2. Based on the pre-processed profile detection data of the plate with different thicknesses, identify the transition zone of the base material and construct the local reference profiles of the thick plate side and the thin plate side. Calculate the dual-reference coupling deviation value. Based on the dual-reference coupling deviation value, determine the weld morphology action area, weld center position, weld toe boundary position of the thick plate side and weld toe boundary position of the thin plate side, and output the weld morphology positioning result. S3, based on the weld morphology positioning results, construct a three-dimensional morphology reconstruction mesh, extract the main depression connected region and calculate the volume value of the depression connected body, the depression connected length value and the weld center trajectory swing value, and output the spatial morphology feature data of the plate with different thickness; S4: Calculate the corresponding absolute difference based on the spatial morphological feature data of the plate with varying thickness and perform dimensionless processing. Calculate the morphological anomaly evolution value based on the dimensionless processing result, determine the morphological anomaly segment and the corresponding anomaly type, and output the online detection result of the laser weld of the plate with varying thickness.

2. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 1, characterized in that: The specific steps for constructing the workpiece coordinate system for the plate with varying thickness, simultaneously acquiring the plate contour detection data, and performing preprocessing on the plate contour detection data are as follows: The weld extension direction of the workpiece with varying thickness is selected as the longitudinal reference direction, the direction perpendicular to the weld extension direction and located on the workpiece surface is selected as the transverse reference direction, and the direction perpendicular to both the transverse and longitudinal reference directions is selected as the height reference direction to construct a workpiece coordinate system. Simultaneously, contour detection data of the workpiece with varying thickness is collected, including laser emission time value, cumulative displacement value of workpiece travel, transverse coordinate value of the previous base material contour point, height coordinate value of the previous base material contour point, transverse coordinate value of the subsequent weld contour point, height coordinate value of the subsequent weld contour point, previous detection interval value, subsequent detection interval value, detection head pitch angle value, detection head roll angle value, detection head transverse position value, detection head longitudinal position value, and detection head height position value. For the collected profile detection data of the plate with varying thickness, a precise time protocol time synchronization algorithm is used to perform multi-source acquisition time alignment processing; a linear interpolation algorithm is used to perform profile sequence resampling processing under different sampling cycles; a median absolute deviation anomaly detection algorithm is used to perform profile outlier data identification processing; a sliding median filtering algorithm is used to perform local burr noise suppression processing; and a rigid body coordinate transformation algorithm is used to perform coordinate unification processing between the detection coordinate system and the workpiece coordinate system, outputting the preprocessed profile detection data of the plate with varying thickness.

3. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 2, characterized in that: The specific steps for identifying the transition zone of the base material and constructing local reference profiles for the thick plate side and thin plate side based on the preprocessed profile detection data of the plate with varying thickness are as follows: Read the pre-processed profile detection data of the plate with different thicknesses, arrange all detection records in ascending order of the cumulative displacement value of the workpiece, and use the cumulative displacement value of the workpiece as the longitudinal registration benchmark. Combine the previous detection interval value and the subsequent detection interval value, convert the previous base material profile and the subsequent weld profile to the same longitudinal position corresponding to the laser action position, and then perform the same section pairing process so that each workpiece cumulative displacement value corresponds to a set of previous base material profiles and a set of subsequent weld profiles. For each workpiece's cumulative displacement value, the preceding base material contour is arranged in ascending order of its lateral coordinate values. A sliding window linear fitting algorithm is used to calculate the contour slope at each lateral position. Then, a difference calculation is performed on the contour slopes at adjacent lateral positions. The lateral segment where the absolute value of the slope difference exceeds the slope change threshold at N consecutive lateral sampling points is determined as the base material transition zone. The preceding base material contours corresponding to the lateral segments on both sides of the base material transition zone are determined as the left and right base material contours, respectively. The average height of the preceding base material contour points corresponding to the left and right base material contours is compared. The base material contour on the side with the larger average height is determined as the thick plate side base material contour, and the base material contour on the side with the smaller average height is determined as the thin plate side base material contour. The lateral coordinates and height coordinates of the preceding parent material contour points corresponding to the thick plate side parent material contour are extracted, as are the lateral coordinates and height coordinates of the preceding parent material contour points corresponding to the thin plate side parent material contour. The least squares linear fitting algorithm is used to construct the local reference contours of the thick plate side and the thin plate side. The lateral position with the largest absolute value of the slope difference of the preceding parent material contour in the parent material transition zone is taken as the lateral position value corresponding to the segmented reference boundary.

4. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 3, characterized in that: The specific steps for calculating the dual-reference coupling deviation value, determining the weld morphology action zone, weld center position, weld toe boundary position on the thick plate side, and weld toe boundary position on the thin plate side based on the dual-reference coupling deviation value, and outputting the weld morphology positioning result are as follows: The lateral coordinates and height coordinates of the post-weld contour points are correlated with the local reference contours on the thick and thin plates. The absolute values ​​of the differences between the height values ​​of the post-weld contours and the corresponding height values ​​of the local reference contours on the thick and thin plates are taken. The square root of the two absolute values ​​is then multiplied to obtain the bilateral coupling deviation term. The absolute value of the difference between the lateral position values ​​of the segmented reference boundaries and the lateral position values ​​of the post-weld contours is then added to a constant of one to obtain the boundary distance constraint term. Divide the two-sided coupling deviation term by the boundary distance constraint term to obtain the dual-baseline coupling deviation value; Read the dual-reference coupling deviation values ​​at each lateral position in lateral order. Determine the lateral segment where the dual-reference coupling deviation value exceeds the threshold of the action zone for K consecutive lateral sampling points as the action zone of weld morphology. Determine the lateral position with the largest dual-reference coupling deviation value in the action zone of weld morphology as the center position of weld. Determine the left boundary position of the action zone of weld morphology as the weld toe boundary position on the thick plate side. Determine the right boundary position of the action zone of weld morphology as the weld toe boundary position on the thin plate side. Output the weld morphology positioning result.

5. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 4, characterized in that: The specific steps for constructing a three-dimensional topography reconstruction mesh based on the weld topography location results are as follows: Read the weld morphology positioning results and the lateral coordinate values ​​and height coordinate values ​​of the post-weld contour points corresponding to the cumulative displacement values ​​of each workpiece. Arrange all detection records in ascending order of the cumulative displacement values ​​of the workpieces. Determine the lateral segment between the weld toe boundary position on the thick plate side and the weld toe boundary position on the thin plate side as the weld morphology reconstruction area. For the weld morphology reconstruction area corresponding to the cumulative displacement value of each workpiece, a linear interpolation algorithm is used to resample the cross-sectional contour formed by the lateral coordinates and height coordinates of the subsequent weld contour points under a uniform lateral sampling interval. The cumulative displacement value of the workpiece is used as the longitudinal coordinate, the resampled lateral position is used as the lateral coordinate, and the corresponding height coordinate of the subsequent weld contour point is used as the height coordinate to construct the cross-sectional grid points. The cross-sectional grid points are longitudinally spliced ​​according to the order of the cumulative displacement value of the workpiece, and the missing grid points between adjacent cross-sectional grid points are filled in using a piecewise cubic spline interpolation algorithm to form a three-dimensional morphology reconstruction grid.

6. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 5, characterized in that: The specific steps for extracting the main concave connected region, calculating the volume value of the concave connected body, the concave connected length value, and the weld center trajectory swing value, and outputting the spatial morphological feature data of the plate with varying thickness are as follows: On the 3D topography reconstruction mesh, the difference between the height coordinate value of the post-weld contour point and the corresponding local reference contour height value is calculated point by point. Grid points with a difference less than the depression judgment threshold are marked as depression grid points. A 3D connected component marking algorithm is used to perform connected component segmentation on spatially adjacent depression grid points. The connected component containing the most depression grid points is determined as the main depression connected region. The absolute value of the difference corresponding to each grid point in the main depression connected region is read, multiplied by the area of ​​the corresponding grid unit, and accumulated to obtain the volume value of the depression connected body. The distance between the starting position and the ending position of the main depression connected region in the direction of the cumulative displacement value of the workpiece is calculated to obtain the depression connected length value. Read the transverse coordinate values ​​corresponding to the weld center position of each section in the order of cumulative displacement values ​​of workpiece travel, construct the weld center trajectory fitting line using the least squares straight line fitting algorithm, and calculate the absolute value of the transverse deviation between the transverse coordinate value corresponding to the weld center position of each section and the weld center trajectory fitting line. The maximum value of the absolute value of the transverse deviation is determined as the weld center trajectory swing value. Extract the volume value of the concave connected body, the length value of the concave connected body, and the oscillation value of the weld center trajectory, and output them as spatial morphology feature data of plates with varying thicknesses.

7. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 6, characterized in that: The specific steps for calculating the corresponding absolute difference based on the spatial morphological feature data of plates with varying thicknesses and performing dimensionless processing are as follows: Read the spatial morphological feature data of plates with varying thicknesses, and calculate the absolute difference of the volume of the concave connected body, the absolute difference of the length of the concave connected body, and the absolute difference of the oscillation value of the weld center trajectory between the current detection record and the previous detection record, according to the order of the cumulative displacement value of the workpiece. The minimum-maximum normalization algorithm is used to perform dimensionless processing on the volume value of the concave connected body, the length value of the concave connected body, the oscillation value of the weld center trajectory, the absolute value of the difference between the volume values ​​of the concave connected body, the absolute value of the difference between the length values ​​of the concave connected body, and the absolute value of the difference between the oscillation values ​​of the weld center trajectory, respectively, to obtain the normalized values ​​of volume, length, oscillation, volume difference, length difference, and oscillation difference.

8. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 7, characterized in that: The specific steps for calculating the morphological anomaly evolution value based on the dimensionless processing result are as follows: For the current detection record corresponding to the cumulative displacement value of each workpiece, multiply the normalized volume value and the normalized length value, add one, and then take the natural logarithm to obtain the current spatial indentation intensity value. The spatial depression evolution gain value is obtained by adding the volume difference normalized value and the length difference normalized value, taking the square root, and then adding one. The trajectory oscillation response value is obtained by adding the oscillation normalized value and the oscillation difference normalized value, taking the natural logarithm, and then multiplying the current spatial depression intensity value with the spatial depression evolution gain value and adding it with the trajectory oscillation response value.

9. The method for online detection of three-dimensional morphology of laser weld seams in plates of varying thicknesses according to claim 8, characterized in that: The specific steps for determining the abnormal morphology sections and corresponding abnormality types, and outputting the online inspection results of laser weld seams in plates of varying thicknesses, are as follows: The workpiece travel cumulative displacement value segment where the morphological abnormality evolution value exceeds the abnormality judgment threshold for Q consecutive detection records is determined as the morphological abnormality segment, and the starting workpiece travel cumulative displacement value and the ending workpiece travel cumulative displacement value corresponding to each morphological abnormality segment are determined as the starting position and ending position of the morphological abnormality segment, respectively. For each morphologically abnormal segment, the peak value of the normalized volume value, the peak value of the normalized length value, and the peak value of the normalized oscillation value are extracted from the segment, and the peak values ​​of the normalized volume value, the peak value of the normalized length value, and the peak value of the normalized oscillation value are compared. When the peak value of the normalized volume value is the largest, the corresponding morphological anomaly segment is identified as the depression-dominant anomaly segment. When the peak value of the length normalization value is the largest, the corresponding abnormal morphology segment is determined as a continuously expanding abnormal segment; when the peak value of the swing normalization value is the largest, the corresponding abnormal morphology segment is determined as a trajectory swing abnormal segment. Extract the starting position, ending position, and type of abnormal morphology section, and output the online detection results of laser weld seams in plates of varying thicknesses.

10. An online three-dimensional morphology inspection system for laser weld seams in plates of varying thicknesses, characterized in that, include: The module includes a contour perception processing module, a dual-reference point localization module, a 3D shape reconstruction module, and an anomaly evolution output module, among which: The contour perception processing module is used to construct the workpiece coordinate system of the workpiece with varying thickness, synchronously collect contour detection data of the workpiece with varying thickness, perform preprocessing on the contour detection data of the workpiece with varying thickness, and output the preprocessed contour detection data of the workpiece with varying thickness. The dual-reference positioning module is used to identify the transition zone of the base material based on the pre-processed profile detection data of the plate with different thicknesses and to construct the local reference profiles of the thick plate side and the thin plate side, calculate the dual-reference coupling deviation value, determine the weld morphology action area, weld center position, weld toe boundary position on the thick plate side and weld toe boundary position on the thin plate side based on the dual-reference coupling deviation value, and output the weld morphology positioning result. The three-dimensional topography reconstruction module is used to construct a three-dimensional topography reconstruction mesh based on the weld topography positioning results, extract the main depression connected region and calculate the volume value of the depression connected body, the depression connected length value and the weld center trajectory swing value, and output the spatial topography feature data of the plate with different thicknesses. The anomaly evolution output module is used to calculate the corresponding absolute difference based on the spatial morphology feature data of the plate with varying thickness and perform dimensionless processing, calculate the morphology anomaly evolution value based on the dimensionless processing result, determine the morphology anomaly segment and the corresponding anomaly type, and output the online detection result of the laser weld of the plate with varying thickness.

Images