A method and system for detecting surface defects of a cathode copper plate

Abstract

The application provides a kind of cathode copper plate surface defect detection method and system, it is related to metal surface quality detection technical field.The cathode copper plate surface defect detection method includes: using mobile 3D line laser scanner to obtain intensity map and height map simultaneously, convert height map into point cloud, and combine two-dimensional positioning, single scanning line independent vibration compensation, local reference surface fitting, spatial clustering and geometric measurement, realize cathode copper plate single surface defect detection, measurement, grade determination and network display.The application uses each scanning profile to independently fit baseline to correct Z-direction common mode displacement, combined with local reference surface fitting, the measurement result directly corresponds to the real geometric deviation of the defect relative to the nearby normal copper plate surface, improves the accuracy of particle height, area and volume measurement.

Description

Technical Field

[0001] This invention relates to the field of metal surface quality inspection technology, and also to the field of industrial machine vision and three-dimensional measurement technology, and in particular to a method and system for detecting surface defects of cathode copper plates based on two-dimensional positioning and three-dimensional vibration compensation. Background Technology

[0002] Cathode copper plates are an important product in the copper smelting and electrolytic refining process, and their surface quality affects subsequent processing, transportation, and quality grading. In industrial production, it is usually necessary to automatically inspect one side of the cathode copper plate to identify different types of defects such as abnormal surface texture, protrusions, depressions, edge abnormalities, and attachments, and to measure the height, size, area, or volume of defects with obvious three-dimensional morphology.

[0003] Existing two-dimensional machine vision inspection methods typically utilize texture, brightness, and edge features from grayscale images, intensity images, or reflectance images for defect localization and classification, offering advantages such as fast processing speed and strong defect category recognition capabilities. However, two-dimensional images cannot directly represent the true height and depth of defects relative to the normal surface. Furthermore, reflections, color differences, or slight texture variations on the metal surface may create image features similar to defects, making it difficult to perform geometric measurements and severity determination independently.

[0004] Three-dimensional line laser inspection can obtain height or depth information of the measured surface and further generate a three-dimensional point cloud to describe the spatial morphology of defects. However, the cathode copper plate may still experience slight vibrations along the Z-direction perpendicular to the plate surface within a fixed station, and the scanner may also generate micro-vibrations or attitude fluctuations as it moves along the vertical guide rail. Since the three-dimensional data of the entire plate is formed by stitching together multiple scanning contours acquired at different times, the above relative motion will cause height misalignment between contours, thereby affecting the fitting of local reference surfaces, the extraction of outliers, and the measurement of defect height, area, and volume.

[0005] Furthermore, different cathode copper plates may exhibit overall height differences, local tilting, warping, and slow changes in plate shape. Using a uniform standard plane or a global plane of the entire plate as the measurement benchmark can easily confuse the actual plate shape changes with local defects. Therefore, a detection scheme is needed that can first locate candidate defect regions using two-dimensional images, perform Z-axis vibration compensation on each scan contour within a local range, and establish a local reference surface based on the normal plate surface around the defect, so as to take into account defect category identification, three-dimensional geometric measurement, and on-site automated applications. Summary of the Invention

[0006] The technical problems to be solved by this invention include: first, how to form a stable and traceable closed loop of automatic scanning and data processing after the cathode copper plate arrives at the inspection station; second, how to establish a unified data association between the intensity map and the height map collected by the same 3D line laser scanner, and to achieve positioning based on the intensity map and then accurate measurement based on the local point cloud after the height map is converted into a point cloud; third, how to eliminate the influence of slight relative Z-axis vibration, scan contour misalignment and lateral slight tilt on three-dimensional measurement during the scanning process; fourth, how to accurately measure the true height, area, size and volume of defects with obvious geometric undulations relative to the nearby normal plate surface when there are different copper plate heights, local tilts, warps or slow plate shape changes; and fifth, how to use point cloud geometric information to cross-validate the intensity map detection results, reduce misjudgments caused by only texture or reflection changes, and publish the results to the field terminal via the Web.

[0007] To address the aforementioned technical problems, this invention proposes a method based on strength Figure 2 The cathode copper plate surface defect detection scheme, which is mainly based on local three-dimensional point cloud measurement guided by two-dimensional positioning, uses a mobile 3D line laser scanner to simultaneously acquire intensity map and height map, converts the height map into point cloud, and combines two-dimensional positioning, independent vibration compensation of single scan line, local reference surface fitting, spatial clustering and geometric measurement to realize single surface defect detection, measurement, classification and network display of cathode copper plates.

[0008] The terms used in this invention are explained as follows:

[0009] Intensity map: A two-dimensional image formed by arranging the reflection intensity values ​​obtained by a 3D line laser scanner during continuous scanning according to the scan contour number and the sampling sequence number within the contour. It is used to characterize the grayscale, reflection intensity, and texture changes of the measured surface. The rows and columns in the intensity map represent discrete sampling indices, and their pixel values ​​represent the reflection intensity at the corresponding locations.

[0010] Height map: A two-dimensional image formed by arranging height values ​​obtained by a 3D line laser scanner during continuous scanning according to the scan contour sequence number and the sampling sequence number within the contour. In terms of storage structure, the height map is a two-dimensional matrix determined by row and column indices, with each valid pixel storing a height value corresponding to its sampling position. Because each two-dimensional sampling position carries a single height information, the height map is often referred to as 2.5D data. The rows and columns in the height map are still discrete sampling indices, not the X and Y physical coordinates in the point cloud.

[0011] Point cloud: A data set consisting of multiple discrete points with three-dimensional physical coordinates. Based on the calibration parameters of the 3D line laser scanner, the arrangement of the scanned contours, and the sampling interval along the scanning direction, the effective pixels in the height map are converted into point cloud points with X, Y, and Z coordinates. In this invention, an organized point cloud is preferably used, meaning that the point cloud points maintain a correspondence with the row and column indices of the height map.

[0012] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based spatial clustering algorithm used to identify one or more clusters of defective points with arbitrary shapes in a set of outliers, and to identify low-density isolated points as noise points.

[0013] The technical solution of this invention is divided into a device layer, a system layer, and a method layer: the device layer is responsible for position detection, motion scanning, position feedback, and data acquisition; the system layer is responsible for data reception, point cloud generation, algorithm processing, result storage, and web publishing; the method layer is responsible for two-dimensional positioning, two-dimensional-three-dimensional mapping, single scan line independent vibration compensation, local reference surface fitting, fixed threshold screening, spatial clustering, geometric measurement, and two-dimensional-three-dimensional mutual verification judgment.

[0014] In a first aspect, the present invention provides a method for detecting surface defects on a cathode copper plate, comprising the following steps:

[0015] After the photoelectric positioning sensor detects that the cathode copper plate has entered the preset threshold range at the inspection station, the scanning start condition is generated. The 3D line laser scanner, mounted on a vertical guide rail slide, scans along the height direction of the copper plate, simultaneously acquiring intensity and height maps generated by the same device. The industrial control computer converts the height map into a point cloud based on calibration parameters and the arrangement of the scan contours, and locates the defect candidate area based on the intensity map. Based on the two-dimensional candidate area, a defect core area, buffer zone, and peripheral reference area are constructed from the point cloud. For each scan contour covered by the defect candidate area, a normal surface reference is selected from the scan contour itself. The system independently fits local baselines to correct the Z-axis common-mode displacement and lateral micro-tilt of the scan profile; combines multiple compensated scan profiles into a local point cloud; fits a local reference plane or local reference surface using the point cloud of the outer reference area; calculates the signed distance from each point in the local point cloud to the local reference surface; extracts outliers based on a preset fixed geometric threshold; performs spatial clustering on the outliers; measures the height, size, area, and volume of the resulting defect point clusters; verifies, confirms, and determines the severity of the detection based on the intensity map detection category and the point cloud geometric results; and saves the detection results and publishes them via a web service.

[0016] In a second aspect, the present invention provides a cathode copper plate surface defect detection system to implement the steps of the method described in the first aspect.

[0017] The system includes a detection device, as well as a data acquisition module, a point cloud generation module, a two-dimensional positioning module, a two-dimensional-three-dimensional correlation module, a single scan line independent vibration compensation module, a local reference surface fitting module, an anomaly extraction module, a spatial clustering module, a geometric measurement module, a two-dimensional-three-dimensional mutual verification module, a result storage module, and a Web service module running on an industrial control computer.

[0018] The data acquisition module takes as input the position signal, scanning status, and camera output stream, and outputs an intensity map I(u,k) and a height map H(u,k) arranged in the acquisition order. The point cloud generation module takes as input the height map, scanning direction sampling information, and calibration parameters, and outputs a local or full-board point cloud P(u,k)=(x,y,z). The 2D positioning module takes as input the intensity map and outputs defect category, confidence level, and 2D candidate bounding boxes. The 2D-3D correlation module takes as input the 2D candidate bounding boxes, equipment area parameters, and point cloud index relationships, and outputs the core area, buffer area, and peripheral reference area. The single-scanline independent vibration compensation module takes as input the local point cloud and three-region information, and outputs the vibration-compensated local point cloud. The local reference surface fitting module takes as input the peripheral reference area. The point cloud is output as a local planar or curved surface model; the anomaly extraction module takes the signed distance from a point to a reference surface as input and outputs a set of positive or negative anomalies; the spatial clustering module takes the set of anomalies as input and outputs a cluster of defect points; the geometric measurement module takes the cluster of defect points and the sampling interval of the point cloud as input and outputs height, length, width, projected area, and volume; the 2D-3D cross-verification module takes the 2D category and 3D geometric quantities as input and outputs the defect confirmation result, severity, and confidence level; the Web service module takes the detection result, result image, and local point cloud as input and outputs a visual page for browser terminals.

[0019] Thirdly, the present invention provides a cathode copper plate surface defect detection device to implement the steps of the method described in the first aspect.

[0020] The device includes: a detection mechanism body, a cathode copper plate fixing mechanism, a photoelectric positioning sensor, a 3D line laser scanner, a vertical guide rail, a scanner mounting slide, a drive unit, a position feedback unit, a limit protection unit, a control cabinet, and an industrial control computer installed in the control cabinet.

[0021] The cathode copper plate fixing mechanism is used to hold the cathode copper plate to be tested at the detection position, allowing slight relative vibration in the Z-direction during scanning. The 3D line laser scanner is mounted on the slide table, and the laser line covers the effective detection width of the copper plate along the width direction. The vertical guide rail is used to drive the 3D line laser scanner to scan from top to bottom along the height direction of the copper plate. The photoelectric positioning sensor is used to detect that the cathode copper plate has entered the preset detection threshold range and output a positioning signal. The position feedback unit is used to output at least one of the slide table position, speed, start position status, and end position status to the industrial control computer. The industrial control computer is used to receive the intensity map and height map output by the 3D line laser scanner, convert the height map into a point cloud, and execute the defect detection algorithm. The industrial control computer is further deployed with a Web service module for publishing the detection results to the browser terminal.

[0022] Compared with the prior art, the present invention can achieve the following beneficial technical effects:

[0023] First, a complete detection loop is formed by positioning sensors, motion scanning, position feedback, industrial computer processing, and Web display, so that the algorithm and hardware technology features work together to form an organic whole.

[0024] Secondly, this invention uses intensity maps and height maps obtained synchronously by the same 3D line laser scanner for correlation processing. Both adopt a two-dimensional image storage format based on row and column indexing, which facilitates the establishment of candidate region correspondence between source data. After the height map is calibrated and converted into a point cloud, geometric compensation and measurement are performed using the X, Y, and Z physical coordinates of the point cloud.

[0025] Furthermore, the core feature of this invention is "independently fitting a baseline for each scan profile to correct the Z-axis common-mode displacement," which enables the compensation parameters to adapt to the relative Z-axis vibration at different scan times. After fitting with a local reference surface, the measurement results directly correspond to the true geometric deviation of the defect relative to the nearby normal copper plate surface, thereby improving the accuracy of particle height, area, and volume measurements.

[0026] Furthermore, by combining fixed threshold filtering with spatial clustering, this invention can extract one or more defect point clusters while eliminating noise points; especially in dense particle scenes, it can output both the geometric values ​​of individual clusters and the overall abnormal area. Density clustering has good adaptability to clusters of arbitrary shapes and noise points, and is therefore suitable as a preferred implementation.

[0027] Finally, the 2D-3D cross-verification mechanism enables this invention to possess both 2D classification and 3D measurement capabilities. 2D localization is used to narrow down the point cloud processing range, eliminate baseline fitting contamination areas, and provide initial categories; 3D results are used to identify protrusion-type defects, reduce false detections caused solely by reflection or texture, and can be published to multiple terminals on-site via a web interface, improving on-site efficiency. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method for detecting surface defects on a cathode copper plate, as disclosed in this invention. It illustrates the entire process from data acquisition, conversion, defect location, extraction, and calculation.

[0029] Figure 2 This is a timing diagram for detection; it illustrates the time relationship between arrival trigger, scan start, continuous acquisition, scan end, algorithm processing, and web publishing.

[0030] Figure 3 This is a structural diagram of a cathode copper plate surface defect detection system disclosed in this invention; it illustrates the composition relationship between the detection device, industrial control computer, web server and browser terminal.

[0031] Figure 4 This diagram illustrates the relationship between data flow and interfaces, showing the data flow of arrival signals, position feedback, intensity maps, height maps, point clouds, detection results, and the web page. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Steps S1, S2… in the described embodiments of the present invention do not limit the scope of execution of the present invention; the various models, simulation environments, and software described in the present invention are not considered as the only limiting methods of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0033] In this invention, computer device / equipment / system refers to a related entity applied to a computer, such as hardware, a combination of hardware and software, software, or software in execution. More specifically, for example, software includes, but is not limited to, a process running on a processor, a processor, an object, executable software, an execution thread, a program, and / or a computer. Furthermore, an application program or script running on a server, and the server itself, can also be software. One or more software programs may be in an execution process and / or thread, and the software may be localized on one computer and / or distributed across two or more computers, and may be run on various computer-readable media.

[0034] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0035] In a first aspect, the present invention provides an embodiment of a method for detecting surface defects in a cathode copper plate, such as... Figure 1 As shown.

[0036] The overall process is as follows:

[0037] Receive the arrival signal of the cathode copper plate at the detection station;

[0038] The 3D line laser scanner is controlled to perform line scanning along the height direction of the cathode copper plate, collecting intensity and height values. After the scanning is completed, an intensity map and a height map of the cathode copper plate surface are formed.

[0039] A three-dimensional point cloud is generated based on the height map and the calibration parameters of the 3D line laser scanner.

[0040] Defect localization is performed on the intensity map to obtain at least one two-dimensional defect candidate region;

[0041] Based on the data association between the two-dimensional defect candidate region and the three-dimensional point cloud, the corresponding local region of the three-dimensional point cloud is determined, and the defect core region, buffer zone and peripheral reference region are constructed.

[0042] For each scan profile covered by the local area of ​​the three-dimensional point cloud, reference points are selected from the non-defect core area and non-buffer normal surface points of the corresponding scan profile itself, and the local baseline of the scan profile is independently fitted to correct the Z-axis common mode displacement and lateral tilt of the scan profile.

[0043] The independently corrected scan contours are combined into a local point cloud, and the local reference plane or local reference surface is fitted using the outer reference area.

[0044] Calculate the signed distance from each point in the local point cloud to the local reference plane or local reference surface;

[0045] Filter out anomalies based on a preset fixed geometric threshold;

[0046] Spatial clustering of outliers yields at least one cluster of defect points;

[0047] Calculate the defect height, length, width, projected area, and / or volume based on the defect point cluster;

[0048] The system combines two-dimensional defect categories and three-dimensional geometric measurement results to complete defect confirmation, severity determination, and result output.

[0049] like Figure 2 As shown, the specific process can be as follows:

[0050] S1, Scan Trigger and Data Acquisition;

[0051] The cathode copper plate is transported to the inspection station and held within the inspection window by a fixing mechanism. The photoelectric positioning sensor outputs a positioning signal after the copper plate enters a preset distance threshold range. The industrial control computer receives this signal and initiates the scanning task. The industrial control computer first checks whether the slide is at the scanning start position and whether the scanner is in the acquisition ready state. Then, it controls the vertical guide rail to drive the 3D line laser scanner to move from top to bottom along the height direction of the copper plate, simultaneously acquiring intensity and height values ​​during the movement. Multiple scan contours are arranged according to the acquisition sequence to form an intensity map and a height map. Upon reaching the scanning termination position, acquisition stops, and the guide rail returns to its initial position. Algorithm processing and guide rail return can be executed in parallel.

[0052] Both the intensity map and the height map are stored in a two-dimensional format. Their row indices correspond to the scan contours arranged in chronological order, and their column indices correspond to the sampling positions arranged along the laser line direction within the same scan contour. The pixel values ​​of the intensity map are the reflection intensity values ​​at the corresponding sampling positions, used to characterize the grayscale, reflection intensity, and texture changes of the cathode copper plate surface. The pixel values ​​of the height map are the height values ​​at the same or related sampling positions, used to characterize the geometric undulations of the cathode copper plate surface.

[0053] In online 3D acquisition mode, the 3D line laser scanner forms a scanning profile with each acquisition and continuously acquires data at a preset fixed profile acquisition frequency; multiple scanning profiles are arranged in the acquisition order to form an intensity map and a height map. Within the effective detection area, the guide rail motion mechanism drives the scanner to move at a constant speed along the height direction of the cathode copper plate, so each row in the image corresponds to a discrete scanning position in the height direction of the copper plate.

[0054] Optionally, the scanner can use free-run acquisition with position feedback correlation to correct the scanning direction sampling interval or scanning contour coordinates; it can also use external line triggering or encoder triggering after the equipment is upgraded.

[0055] In one embodiment of the present invention, the original acquired data includes an intensity map I(u,k) and a height map H(u,k), and optionally receives position or velocity information of the guide rail motion mechanism. Here, u is the sampling index along the laser line direction, i.e., the width direction X of the cathode copper plate, and k is the scanning contour number along the height direction of the cathode copper plate, i.e., the vertical scanning direction Y; the pixel values ​​of I(u,k) are the reflection intensity values ​​at the corresponding sampling positions, and the pixel values ​​of H(u,k) are the height values ​​at the corresponding sampling positions. Both the intensity map and the height map are two-dimensional matrices determined by u and k indices in their storage structure; the height map is often referred to as 2.5D data because each two-dimensional sampling position corresponds to a height value, but it does not yet include the X, Y, Z three-dimensional physical coordinates in point cloud form.

[0056] Let v be the preset scanning speed of the guide rail within the effective detection area, and let v be the contour acquisition frequency of the 3D line laser scanner. The nominal sampling interval along the height direction of the cathode copper plate is then... The physical coordinates of the k-th scan contour in the Y direction are: ,in To determine the effective starting position for scanning, the height map is converted into a point cloud based on the calibration parameters of the 3D line laser scanner.

[0057] ;

[0058] In the above expression, u is the sampling index along the width direction X of the cathode copper plate, k is the scan profile number along the height direction Y of the cathode copper plate, and H(u,k) is the height value of the corresponding sampling position in the height map. Let k be the physical position of the k-th scan profile along the height direction of the cathode copper plate. The coordinate transformation function is determined by the calibration parameters of the 3D line laser scanner. A point cloud is a data set composed of multiple discrete points with three-dimensional physical coordinates; in this embodiment, the point cloud maintains the correspondence with the (u,k) index, thus it is an organized point cloud. X represents the physical coordinate along the width direction of the copper plate, Y represents the physical coordinate along the height direction of the copper plate, and Z represents the height coordinate perpendicular to the surface of the copper plate; the specific positive direction is determined by the device calibration coordinate system.

[0059] In this context, X, Y, and Z are physical coordinates in millimeters. It should be noted that both the intensity map and height map use row and column numbers as data indices; the X, Y, and Z coordinates in the point cloud are physical coordinates calculated using calibration parameters and the sampling relationship of the scanning direction. Therefore, the row and column indices in the image and the X and Y coordinates in the point cloud should not be directly equated in meaning or value, but rather a correspondence should be established through organized indexing relationships or calibration mapping.

[0060] S2, Two-dimensional positioning and two-dimensional-three-dimensional association;

[0061] The industrial computer performs defect detection on the intensity map and obtains two-dimensional candidate boxes:

[0062] ;

[0063] Where j is the index of the two-dimensional candidate region. and Let be the coordinates of the top-left and bottom-right corners of the j-th two-dimensional candidate box in the intensity map, respectively. For defect category identification, This represents the confidence level output by the target detection model; u represents the horizontal index of the intensity map, and v represents the vertical index of the intensity map.

[0064] The two-dimensional positioning module preferably employs a target detection model; alternatively, an instance segmentation model or a regular image algorithm may also be used. For defect categories expected to have obvious geometric undulations, such as round-headed particles, irregularly shaped particles, crystalline copper, protruding foreign objects, and thick edges, the process further proceeds to the three-dimensional measurement flow; for defect categories that mainly manifest as texture or grayscale changes, only two-dimensional judgment may be retained or three-dimensional judgment may be used as an auxiliary verification.

[0065] The 2D-3D correlation module determines the 3D local point cloud region corresponding to the candidate box based on the 2D candidate box in the intensity map, the row and column index relationship between the intensity map and the height map, and the organized index relationship between the height map and the point cloud, and constructs three regions: the defect core region. Buffer and peripheral reference area The core region is the point cloud region corresponding to the two-dimensional candidate box mapping; the buffer zone is located outside the core region and is used to cover the defect edges that are not completely enclosed by the two-dimensional candidate box, the small index deviation that may exist between the intensity map and the height map, and the part of the actual three-dimensional boundary of the defect that exceeds the two-dimensional apparent boundary, thereby preventing defect points from entering the normal surface reference point set; the outer reference region is located outside the buffer zone and is used for vibration compensation and local reference surface fitting.

[0066] Preferably, the two-dimensional frame is expanded outward in millimeters in both the horizontal and scanning directions to form a buffer zone, so as to reduce the impact of mapping error on the purity of the reference area.

[0067] S3, Independent vibration compensation for a single scan line;

[0068] This step is one of the core aspects of this invention. Compensation is performed independently for each scan profile that falls into the local region. By estimating the local baseline of each scan profile separately, the compensation parameters can adapt to the relative Z-axis vibration at different scan times, reducing the height misalignment between scan profiles, improving the continuity of the local point cloud, and enhancing the measurement stability of defect height, area, and volume.

[0069] Let the original point on the k-th scan contour be... In this scanned profile, areas belonging to the core of the defect are first eliminated. and buffer The points are then processed by removing edge points, invalid points, outliers, and significantly abnormal points to obtain the reference point set. .based on Fit the local baseline of the profile; use a linear model when the plate surface is locally approximately flat:

[0070] ;

[0071] in, The slope parameter of the local baseline of the k-th scan profile is used to characterize the slight tilt of the profile in the width direction of the cathode copper plate. The intercept parameter characterizes the overall common offset of the profile along the Z-direction. When the cathode copper plate or scanner undergoes relative displacement in the Z-direction, the normal surface reference points within the same profile will rise or fall as a whole, and the fitted value will reflect this. Consequently, the common offset and lateral tilt are offset by subtracting the local baseline from the original height, while the local height difference of the defect relative to the nearby normal plate surface is preserved.

[0072] The term "characterization" or "compensation" as used in this specification means that the Z-axis common offset and lateral tilt are not directly measured by a single sensor, but are obtained by fitting a normal surface reference point as part of the local baseline model parameters, and are compensated by calculating the difference between the original height and the local baseline.

[0073] The local baseline model can be selected based on the fitting residuals. Preferably, a first-order linear model is first used to fit the reference point and its root mean square residual or a preset quantile value of the absolute residual is calculated. When the residual is not greater than a preset flatness threshold, the current contour is determined to be approximately flat within the reference range, and the first-order model is used. When the residual of the first-order model exceeds the flatness threshold, and the ratio of the residual of the second-order model to the first-order model reaches a preset ratio threshold, the lateral curvature is determined to be significant, and the second-order model is used. If the residuals of both models exceed the corresponding reliability thresholds, the contour is marked as a contour with insufficient reference surface or low confidence.

[0074] If the lateral curvature of the board surface is significant, a secondary model can be used:

[0075] ;

[0076] In the quadratic model, is the curvature coefficient, used to characterize the degree of second-order curvature of the scan profile along the width direction of the cathode copper plate; This is the first-order slope coefficient, used to characterize lateral tilt; The intercept coefficient characterizes the Z-axis common offset of the profile. The parameters are derived from the set of reference points. Obtained through least squares fitting, robust fitting, or random sampling consistency fitting.

[0077] Then, the corrected height residuals of each point on the profile relative to the local baseline are calculated: consistent with the linear model, the original height is subtracted from the fitted baseline to offset common offsets and lateral tilt.

[0078] ;

[0079] The corrected points are represented as:

[0080] ;

[0081] Because each contour is independently fitted based on its own normal surface reference point, even if the scan length is large or some contours have strong local anomalies, they will not affect each other. This strategy is suitable for the relative height measurement scenario targeted by this invention: it measures the offset of the defect relative to the normal plate surface near the contour, rather than the height of the entire plate relative to a uniform absolute plane. The compensation here includes not only slight Z-axis jitter of the copper plate, but also common-mode offset of the scan contour caused by micro-vibration of the guide rail, micro-motion of the camera mounting posture, etc.

[0082] Optionally, when the number of valid reference points for a certain scan profile is insufficient or the reference points are only distributed on one side, within a preset neighborhood window centered on that profile, the scan profiles that are closest to and have a reliable fit in front of and behind it are selected respectively, and the baseline model parameters are linearly interpolated according to the profile number distance to obtain the compensation parameters for the current profile; if there is only a reliable neighboring profile on one side, the parameters of that neighboring profile can be used as the initial value, and refitted in combination with the remaining valid reference points of the current profile. When reliable compensation parameters cannot be obtained, a low confidence mark is set for the measurement results of the current profile or the corresponding defect.

[0083] Optionally, a robust fit can also be used for baseline fitting.

[0084] S4, Local reference surface fitting;

[0085] Multiple independently compensated scan contours are combined into a local point cloud. Then, in the outer reference area The local reference plane or local reference surface is fitted. A local reference plane is preferred.

[0086] AX + BY + CZ + D = 0;

[0087] Candidate points The signed distance to the local reference plane is:

[0088] ;

[0089] Where A, B, and C are the components of the local reference plane normal vector in the X, Y, and Z directions, and D is the plane constant term; The i-th candidate point in the local point cloud to be tested; Candidate points Signed perpendicular distance to the local reference plane, in millimeters. Denominator Used to normalize the plane normal vector. By unifying the direction of the local reference plane normal vector, protrusions that are far from the normal plate surface and face the scanner are assigned positive distances, while depressions that are lower than the normal plate surface are assigned negative distances.

[0090] If there is slow bending in a local area of ​​the board surface, a local surface Z=f(X,Y) can be used for fitting. This step is used to eliminate the overall height difference, local tilt, warping and slow board shape changes between different copper plates, so that the height measured later is expressed as the geometric value of the defect relative to the normal board surface nearby, rather than the value relative to the absolute coordinate origin of the equipment.

[0091] Optionally, the local reference surface can be a plane, a quadratic surface, or a moving least squares surface.

[0092] Optionally, outlier removal, hole filling, and noise filtering can be added before fitting the local reference surface.

[0093] S5. Fixed threshold filtering and spatial clustering;

[0094] After establishing the local reference surface, outlier points are extracted based on a preset fixed geometric threshold. For protrusion-type defects, the following is extracted:

[0095] ;

[0096] in This is a fixed height threshold in millimeters. For dent-type defects, optionally extract:

[0097] ;

[0098] in This is a fixed indentation threshold measured in millimeters.

[0099] Optionally, the above thresholds are all preset by process standards and no adaptive threshold strategy is adopted.

[0100] Spatial clustering of the anomaly set yields at least one defect point cluster. In a preferred embodiment, DBSCAN is used to extract the point clusters, and the neighborhood radius parameter is... Given in millimeters, the minimum number of points is an integer; to avoid the influence of differences in axis scales, clustering can be performed on the plate projection coordinates (X,Y), and... Used as an additional measurement dimension.

[0101] DBSCAN is a preferred implementation because it can identify clusters of arbitrary shapes and distinguish noise points in spatial data containing noise points, but the present invention is not limited to this algorithm. The spatial clustering can also employ Euclidean clustering, region growing, or organized point cloud connected component analysis.

[0102] Optionally, after spatial clustering, a two-dimensional contour reconstruction of the point cluster boundaries can be performed to improve area estimation.

[0103] S6. Geometric measurement and two-dimensional-three-dimensional mutual verification;

[0104] For each cluster of defect points Calculate the following geometric quantities: robust height ,length ,width Projected area ,volume and location center Credibility marker The three-dimensional measurement results are obtained:

[0105] ;

[0106] Among these, the higher quantile value is preferred over the maximum value of a single point as the height:

[0107] ;

[0108] The projected area is calculated based on the sampling interval:

[0109] ;

[0110] in, For defect clusters The number of effective sampling units in the projected grid of the board surface, or equivalently, the number of effective points used for area statistics after removing duplicate projections; and These represent the point cloud sampling intervals in the width and height directions of the cathode copper plate, respectively. When there are irregular samplings in the point cloud projection onto the plate surface, the point cloud can be rasterized or a two-dimensional contour can be reconstructed before statistical analysis.

[0111] Volume is approximated by integrating by height:

[0112] ;

[0113] in, and The sampling interval for point clouds is measured in millimeters.

[0114] The 2D-3D cross-verification module outputs the final result based on the 2D defect category and 3D geometry. For defect categories expected to have obvious protrusions, if a candidate region is detected in 2D and a valid cluster of protruding points exists in 3D, the defect is confirmed and classified according to a fixed threshold. If a candidate region is detected in 2D but a valid cluster of protruding points does not form in 3D, and the local point cloud is complete, the reference area is sufficient, and the measurement is reliable, the confidence level of the 2D result is reduced or it is marked as a texture anomaly. If the local point cloud is incomplete or the reference points are insufficient, the 2D result is retained and marked as pending verification. For categories primarily based on texture features, the 3D result can be used as supplementary information rather than a rejection condition.

[0115] S7, Web display and results output;

[0116] The web display does not alter the core detection algorithm; it merely serves as a result publishing interface. After completing the detection of a single cathode copper plate, the industrial control computer's algorithm module encapsulates the defect category, 2D location, 3D geometry, severity, confidence level, result image path, and local point cloud path into a unified result data structure, using the detection task identifier or copper plate identifier as an index, and writes it to the result database or file system. The web service module reads the aforementioned result data structure and visualization resources based on the same identifier, generating a page or interface response for browser terminals to query and display. Thus, a clear data transfer relationship is established between the acquisition task, algorithm results, stored records, and the web page through a unified identifier and standardized data fields.

[0117] In a second aspect, the present invention provides a cathode copper plate surface defect detection system to implement the steps of the method described in the first aspect.

[0118] like Figure 3 As shown, the system includes a detection device, an industrial control computer, and a web server; wherein:

[0119] The detection device includes a positioning sensor, a 3D line laser scanner, a vertical guide rail, a drive unit, a position feedback unit, and a cathode copper plate fixing mechanism.

[0120] The positioning sensor is used to detect when the cathode copper plate enters the detection station and outputs a positioning signal.

[0121] The vertical guide rail and drive unit are used to drive the 3D line laser scanner to scan along the height direction of the cathode copper plate;

[0122] The position feedback unit is used to output the position or speed information of the 3D line laser scanner during the scanning process;

[0123] The 3D line laser scanner is used to output intensity and height maps;

[0124] The industrial control computer is communicatively connected to the positioning sensor, the 3D line laser scanner, and the position feedback unit. The industrial control computer includes a data acquisition module, a point cloud generation module, a two-dimensional positioning module, a two-dimensional-three-dimensional association module, a single scan line independent vibration compensation module, a local reference surface fitting module, an anomaly point extraction module, a spatial clustering module, a geometric measurement module, and a two-dimensional-three-dimensional mutual verification module.

[0125] The web server is connected to the industrial control computer and is used to receive the detection results and publish the defect display page to the browser terminal.

[0126] The single-scan-line independent vibration compensation module is configured to: fit a local baseline for each scan profile within a local area based on the normal surface reference point of the scan profile itself, and correct the Z-direction common-mode displacement.

[0127] Preferably, the relationship between key data flows and interfaces is as follows: Figure 4 As shown, it includes:

[0128] Sensor-control interface relationship: The photoelectric position sensor outputs a level position signal to the scanning control logic; the position feedback unit outputs a position or speed signal to the industrial control computer; the 3D line laser scanner outputs intensity map data stream, height map data stream, and optional metadata to the industrial control computer; the industrial control computer converts the height map into a point cloud and outputs the defect result structure and visualization layer to the Web service module; the Web service module outputs HTML pages, result images, and interface data to the browser terminal.

[0129] Thirdly, the present invention provides a cathode copper plate surface defect detection device to implement the steps of the method described in the first aspect.

[0130] The device includes the following components:

[0131] The main body of the testing institution;

[0132] Photoelectric positioning sensor installed on the main body of the testing mechanism;

[0133] 3D line laser scanner mounted on a vertical guide rail slide;

[0134] A drive unit for driving the slide table to move along the vertical guide rail;

[0135] A position feedback unit used to output the position or speed information of the slide table;

[0136] Industrial control computers and network interfaces are installed in the control cabinet;

[0137] The photoelectric positioning sensor is connected to the industrial control computer and is used to trigger a scanning task when the cathode copper plate enters the preset detection threshold range.

[0138] The 3D line laser scanner is used to acquire intensity and height maps of a single surface of the cathode copper plate during the movement of the slide table.

[0139] The industrial control computer is used to convert the height map into a point cloud, and combine it with the intensity map to perform two-dimensional positioning, single-scanline independent vibration compensation, local reference surface fitting, spatial clustering and geometric measurement;

[0140] The network interface is used to publish the detection results to a web server or to provide a web service page directly from the industrial control computer.

[0141] Example 1:

[0142] After the cathode copper plate enters the inspection station, the photoelectric positioning sensor outputs a positioning signal; the industrial control computer controls the vertical guide rail to drive the 3D line laser scanner to complete the scanning of the entire plate from top to bottom; during the scanning, intensity map and height map are collected simultaneously, and the industrial control computer converts the height map into a point cloud according to the calibration parameters, and completes defect detection and Web result publication in parallel during the guide rail return process.

[0143] Example 2:

[0144] For round-headed particle candidates detected in 2D, the industrial control computer maps the 2D bounding box to the local point cloud, selects the left and right normal plate surface points of each scan contour covering the area to fit the local baseline, and independently corrects the Z-direction common mode displacement; after fitting the local plane in the outer reference area, positive outlier points are extracted, spatial clustering is used to obtain particle point clusters, and the height, area and volume of the particles are output.

[0145] Example 3:

[0146] For a scene containing multiple adjacent particles in a 2D detection frame, the industrial control computer performs fixed threshold filtering and spatial clustering on the same local point cloud to separate multiple convex point clusters; it outputs the geometric quantities of each point cluster and calculates the total abnormal area for overall determination of dense particle scenes.

[0147] Example 4:

[0148] For particle candidates caused by high reflectivity texture in a 2D image, if the local point cloud does not form a sufficiently continuous cluster of convex points above a fixed threshold after independent compensation by a single scan line and fitting with a local reference surface, the 2D-3D mutual verification module will downgrade the candidate to texture abnormality or pending verification, thereby reducing false detections without actual convexities.

[0149] The embodiments described above are merely examples of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for detecting surface defects in a cathode copper plate, characterized in that, Includes the following steps: S1. Control the laser line scanner to move along a direction parallel to the copper plate and simultaneously perform line-by-line scanning to obtain the intensity map and height map of the copper plate; then obtain the three-dimensional coordinate point cloud of the copper plate according to the scanner calibration parameters; S2. Perform defect detection on the intensity map to obtain the defect category and determine the core area directly corresponding to the defect, the buffer zone of the defect edge, and the outer reference area. S3. Remove defect points and abnormal points from the single-row scan results to obtain a set of reference points; independently perform straight line or curve fitting on the set of reference points for the row to obtain a local baseline; subtract the local baseline from the original scan profile to offset common offset and tilt, obtain a compensation profile, and determine the local height difference of the defective part in the row relative to the normal plate surface. S4. In the local area containing the defect, combine multiple compensation contours and use the data that falls into the outer reference area to fit to obtain a reference surface. Calculate the directional distance from each point in the defect core area and buffer zone to the reference surface. S5. Extract outlier points based on the directional distance, and then perform spatial clustering to obtain defect point clusters; S6. Calculate the geometric quantities of each defect point cluster, and verify them with the defect category obtained in step S2 using intensity maps and 3D point clouds to obtain the defect discrimination result. S7. Output the defect identification results.

2. The method according to claim 1, characterized in that, In step S1 Intensity maps are used to characterize the grayscale, reflectance intensity, and texture variations of a copper plate surface; Height maps are used to characterize the geometric undulations of the copper plate surface; The multiple scan profiles obtained by line-by-line scanning are arranged in the acquisition order to form intensity maps and height maps; The point cloud is obtained by performing coordinate transformation based on the single-line sampling index, scan line number, and sampling point height value of the line-by-line scan.

3. The method according to claim 1, characterized in that, In step S2 Defect detection methods include: using object detection models, using instance segmentation models, or using regular image algorithms; The buffer zone is located outside the core area and is used to cover defect edges that are not fully contained within the core area, index discrepancies between the intensity map and the height map, and the portion of the actual three-dimensional boundary of the defect that extends beyond the two-dimensional apparent boundary.

4. The method according to claim 1, characterized in that, In step S3 Defect points include points located in the core area and the buffer zone; Anomalies include: edge points, invalid points, outliers, and large anomalies.

5. The method according to claim 4, characterized in that, Fitting methods include: First, a linear model is used to fit the reference point set, and the root mean square residual or the preset quantile value of the absolute residual is calculated. When the residual is not greater than the preset flatness threshold, it is determined that the current contour is approximately flat within the reference range, and a linear model is used for fitting. When the residual of the primary model exceeds the flatness threshold, and the reduction ratio of the residual of the secondary model relative to the primary model reaches a preset ratio threshold, it is determined that the lateral bending is obvious, and the secondary model is adopted. If the residuals of both models exceed the corresponding reliability threshold, the profile is marked as insufficient reference surface or low confidence profile.

6. The method according to claim 1, characterized in that, In step S3 When the number of reference points is insufficient for fitting, select the single-line scan result that is closest to and has a reliable fit in front of and behind it, and perform linear interpolation according to the line distance to obtain the compensation parameters of the current reference point. If only one reliable neighbor scan result exists, then use that neighbor scan result as the initial value and refit it in combination with the remaining valid reference points in the current row; When reliable compensation parameters are unavailable, set a low confidence flag for the current row or the corresponding defect measurement results.

7. The method according to claim 1, characterized in that, In step S4, the reference surface includes a reference plane or a reference curved surface. The normal vector of the reference surface is normalized and its direction is unified so that the protrusions facing the scanner side correspond to positive distances and the depressions correspond to negative distances.

8. The method according to claim 1, characterized in that, In step S5, an anomaly threshold is preset by the process standard, DBSCAN is used to extract point clusters of arbitrary shapes, and the directional distance is used as an additional measurement dimension.

9. The method according to claim 1, characterized in that, In step S6, the geometric quantities of the defect point cluster and the corresponding calculation methods include: Height: The highest quantile value of directional distance in a statistical defect cluster; Projected area: calculated using the number of effective sampling units and the sampling interval; Volume: Integral based on height on the projected area.

10. The method according to claim 1, characterized in that, In step S6, the mutual verification method includes: If the intensity map detects a candidate region and the 3D point cloud has an effective cluster of protruding points, it is confirmed as a highly anomalous surface-attached particle defect. If the intensity map detects a candidate region but the 3D point cloud does not form an effective cluster of convex points, and the local point cloud is complete, the reference area is sufficient, and the measurement is reliable, then the confidence level of the 2D result should be reduced or it should be marked as a texture anomaly. If the local point cloud is incomplete or there are insufficient reference points, retain the two-dimensional result and mark it as pending review; For defect categories that are primarily characterized by texture features, geometric quantities are used as supplementary explanations.

11. The method according to claim 1, characterized in that, In step S7, the output defect discrimination results include: defect category, two-dimensional location, three-dimensional geometric quantity, defect level, confidence level, result map path, and local point cloud path.

12. A cathode copper plate surface defect detection system, used to perform the steps of the method according to any one of claims 1-11, characterized in that, The system includes a detection device, an industrial control computer, and a web server; among which: The detection device includes a positioning sensor, a 3D line laser scanner, a vertical guide rail, a drive unit, a position feedback unit, and a cathode copper plate fixing mechanism; The positioning sensor is used to detect when the cathode copper plate enters the detection station and outputs a positioning signal. The vertical guide rail and drive unit are used to drive the 3D line laser scanner to scan along the height direction of the cathode copper plate; The position feedback unit is used to output the position or speed information of the 3D line laser scanner during the scanning process; The 3D line laser scanner is used to output intensity and height maps; The industrial control computer is communicatively connected to the positioning sensor, the 3D line laser scanner, and the position feedback unit. The industrial control computer includes a data acquisition module, a point cloud generation module, a two-dimensional positioning module, a two-dimensional-three-dimensional association module, a single scan line independent vibration compensation module, a local reference surface fitting module, an anomaly point extraction module, a spatial clustering module, a geometric measurement module, and a two-dimensional-three-dimensional mutual verification module. The web server is connected to the industrial control computer and is used to receive the detection results and publish the defect display page to the browser terminal. The single-scan-line independent vibration compensation module is configured to: fit a local baseline for each scan profile within a local area based on the normal surface reference point of the scan profile itself, and correct the Z-direction common-mode displacement.

13. A cathode copper plate surface defect detection device, used to perform the steps of the method according to any one of claims 1-11, characterized in that, The device includes the following components: The main body of the testing institution; Photoelectric positioning sensor installed on the main body of the testing mechanism; 3D line laser scanner mounted on a vertical guide rail slide; A drive unit for driving the slide table to move along the vertical guide rail; A position feedback unit used to output the position or speed information of the slide table; Industrial control computers and network interfaces are installed in the control cabinet; The photoelectric positioning sensor is connected to the industrial control computer and is used to trigger a scanning task when the cathode copper plate enters the preset detection threshold range. The 3D line laser scanner is used to acquire intensity and height maps of a single surface of the cathode copper plate during the movement of the slide table. The industrial control computer is used to convert the height map into a point cloud, and combine it with the intensity map to perform two-dimensional positioning, single-scanline independent vibration compensation, local reference surface fitting, spatial clustering and geometric measurement; The network interface is used to publish the detection results to a web server or to provide a web service page directly from the industrial control computer.

Images