Battery weld mark defect detection method and device based on 3D point cloud segmentation and medium

Abstract

The invention discloses a battery weld mark defect detection method and device based on 3D point cloud segmentation and a medium, and relates to the technical field of weld mark defect detection.The method comprises the steps that original point cloud data is organized in a matrix form, and a rapid index is established to achieve batch data reading and rapid region segmentation; after any rectangular area is segmented according to columns, track continuity of each column of sub-point clouds in the scanning direction is reserved, straight line fitting can be directly carried out to construct a local datum line, and the calculation amount is reduced; by calculating the distance deviation from the point cloud in each segmented region to the corresponding reference line, the abnormal point deviating from the reference line can be quickly positioned, only one matrix index conversion, one region segmentation and one straight line fitting operation are needed, the time-consuming nearest neighbor search, curved surface reconstruction or voxelization process in the traditional point cloud processing is avoided, and the processing efficiency is improved. Typical defect types such as broken welding, cracks and weld beading can be accurately identified, and the detection precision and efficiency are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of weld defect detection technology, specifically to a method, equipment, and medium for detecting battery weld defects based on 3D point cloud segmentation. Background Technology

[0002] With the development of the new energy vehicle industry, the requirements for batteries are becoming increasingly stringent. During battery production, after the battery cells are installed in a metal casing (such as an aluminum or steel casing), the battery cover must be sealed and welded to the casing to ensure airtightness and safety. Poor welding quality can seriously affect battery safety. To address this issue, a testing requirement has emerged for determining whether the weld lines on the metal casing are continuous and smooth. Currently, the main defects include: broken welds, cracks, and weld beads.

[0003] Currently, there are two main approaches to detecting weld defects:

[0004] 1. Deep learning based on 2D images requires collecting a large number of defect images, manually labeling the defect locations, and finally training the model. Since the battery surface is made of metal, reflections during photography can easily lead to false defect detections. Furthermore, while the camera focal length is fixed, weld defects do not always remain in focus, affecting image quality. Due to this unstable imaging effect, this method suffers from a high false detection rate.

[0005] 2. Neural network learning based on 3D point clouds requires first extracting point clouds at the defect location, constructing defect features through convolutional feature extraction, and finally combining self-attention mechanisms, multilayer perception, and category feature comparison learning to identify different types of welding defects. Although 3D point cloud data is used, which can provide accurate geometric information of the weld surface compared to 2D images, the network is too complex, with each layer having more dimensions than 2D. Compared to traditional weld marks, the weld marks around the battery are smaller, and the defects are also smaller. To better capture the defects, the resolution must be increased, which will increase the amount of point cloud data to nearly a million. Heavy algorithms such as global point cloud registration and feature descriptor calculation are used, which not only have certain requirements for equipment but also increase the time consumption.

[0006] 3. The biggest characteristic of point clouds acquired by traditional lidar is their disorder. Point clouds are not arranged in a certain order. The first and second points may be far apart, but the second and third points may be relatively close. Therefore, traditional lidar has low accuracy when acquiring point clouds. Summary of the Invention

[0007] Existing technologies suffer from high false detection rates, complex algorithms, low accuracy of acquired point clouds, and complex processing. This invention aims to provide a method, equipment, and medium for detecting battery weld defects based on 3D point cloud segmentation. By combining the characteristics of point cloud data acquired by a 3D line laser camera, the point cloud is rapidly segmented, requiring only fitting and distance calculation to quickly determine the point cloud location of the defect. This not only offers advantages in speed but also accurately identifies points at the defect location and can quickly distinguish between different types of defects such as broken welds, cracks, and weld beads, thereby improving detection accuracy.

[0008] This invention is achieved through the following technical solution:

[0009] The first aspect of this invention provides a method for detecting battery weld defects based on 3D point cloud segmentation, comprising the following specific steps:

[0010] A 3D line laser scanning device is used to scan the surface of the battery weld lines to obtain raw point cloud data containing at least one weld line; the raw point cloud data includes multiple weld line trajectory point clouds distributed along the scanning direction.

[0011] The raw point cloud data is arranged in an m x n matrix format;

[0012] Convert the physical coordinates of any point in the original point cloud data into its row and column numbers in the point cloud matrix;

[0013] Based on the row and column numbers, establish an index correspondence between the physical coordinates of the points and the original data storage structure;

[0014] Based on the index correspondence, the corresponding points are read in batches from the original point cloud data. Any rectangle in the original point cloud data is divided into regions by column to form a sub-point cloud for each column.

[0015] Obtain the point set obtained from the segmentation, perform line fitting on the point set obtained from the segmentation, and obtain the benchmark corresponding to each segmentation region;

[0016] Based on the benchmark comparison, the defect location and defect type of the point cloud in each segmented region are obtained.

[0017] Furthermore, the step of converting the physical coordinates of any point in the original point cloud data into its row and column indices in the point cloud matrix specifically includes:

[0018] The row number of any point in the original point cloud data is calculated as follows:

[0019] ;

[0020] The column number of any point in the original point cloud data is calculated as follows:

[0021] ;

[0022] in, The spacing between two adjacent lines. The spacing between two adjacent columns, and Let Px and Py be the minimum values ​​in the x and y directions, respectively, where Px represents the y-coordinate of the point and Py represents the y-coordinate of the point.

[0023] Furthermore, obtaining the point set obtained from the segmentation includes:

[0024] Differential color encoding is applied to the segmented point set to generate a visualization image of the weld scar point cloud segmentation:

[0025] Assign different visual identifier colors to point sets in different areas;

[0026] Point sets within the same area are rendered using the same identifier color;

[0027] Different sets of points in different regions are distinguished by color differences.

[0028] Furthermore, the step of performing line fitting on the segmented point set to obtain the benchmark corresponding to each segmented region specifically includes:

[0029] Obtain the point set after splitting by column, and calculate the mean point of all points in the point set;

[0030] Construct a centralized matrix X, where each row of the centralized matrix X is the coordinate difference between the original point and the mean point;

[0031] The covariance matrix C is calculated based on the centered matrix X;

[0032] Find the eigenvalues ​​of the covariance matrix C;

[0033] Select the largest eigenvalue to solve for the corresponding eigenvector;

[0034] Normalize the feature vectors to obtain the direction vector of the fitted line;

[0035] Using the mean point as a point on the straight line, the fitted straight line is determined as the benchmark for each segmented region.

[0036] Furthermore, obtaining the defect location and defect type of the point cloud within each segmented region specifically includes:

[0037] The distance from point P in the original point cloud data to the fitted line is determined based on the fitted line and its direction vector.

[0038] Based on the distance from point P in the original point cloud data to the fitted straight line, and a specified threshold, the location and type of defects in the point cloud are determined.

[0039] Furthermore, determining the defect type specifically includes:

[0040] If the distance between point P and the fitted line is greater than the specified threshold, then point P can be determined to be a point at the defect location.

[0041] If the distance between point P and the fitted line is within the specified threshold, then point P can be determined to be a point in the normal region.

[0042] Furthermore, after determining that point P is the point at the defect location, point P is projected onto the fitted line to obtain... By comparing the Z-coordinate values ​​of the two points, it can be determined whether point P is a concave or convex point:

[0043] If the Z coordinate of P is greater than the projection point If the Z-coordinate is given, then P is the point at the convexity;

[0044] If the Z-coordinate of P is less than the projection point If the Z-coordinate is given, then P is the point at the depression.

[0045] Furthermore, the step of projecting point P onto the fitted line yields... The specific steps are as follows:

[0046] point Points on the fitted line Moving a distance d along the direction of the fitted line yields:

[0047] ;

[0048] in, To project point P onto a point on the fitted line, Let be a point passed through by the straight line, and d be the directed distance. The direction vector for fitting the straight line.

[0049] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a battery weld defect detection method based on 3D point cloud segmentation.

[0050] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for detecting battery weld defects based on 3D point cloud segmentation.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] Based on the characteristics of the raw point cloud data collected by the 3D line laser scanning device, this invention leverages the structural advantages brought by the matrix arrangement of the point cloud. Since the raw point cloud data is organized in an m-row n-column matrix and a correspondence between physical coordinates and matrix indices is established, point cloud segmentation does not require complex spatial search or neighborhood query; batch data reading can be achieved simply by using row and column numbers.

[0053] After dividing an arbitrary rectangular region into columns, each column of sub-point cloud retains the trajectory continuity along the scanning direction. Based on this characteristic, a local baseline can be directly constructed by straight line fitting, reducing the amount of computation.

[0054] In the defect detection stage, abnormal points deviating from the baseline can be quickly located by calculating the distance deviation of the point cloud in each segmented region to the corresponding baseline. This method requires only one matrix index transformation, one region segmentation, and one line fitting operation, avoiding the time-consuming nearest neighbor search, surface reconstruction, or voxelization processes in traditional point cloud processing, thus having an advantage in computational speed.

[0055] Meanwhile, since the 3D line laser scanning device collects data along a fixed direction, the weld trace point cloud has a clear directionality and continuity, which enables the straight line reference to accurately reflect the ideal weld morphology. Thus, through distance threshold judgment and deviation pattern analysis, typical defect types such as weld breakage (continuity interruption), cracks (local depression deviation), and weld beads (local protrusion deviation) can be accurately identified, effectively improving detection accuracy and efficiency. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0057] Figure 1 This is a defect detection method in an embodiment of the present invention; Figure 2 This is a schematic diagram of point cloud segmentation in an embodiment of the present invention; Figure 3 This is a point cloud image of the normal region after being segmented by rows in an embodiment of the present invention; Figure 4 This is a point cloud image of the defective region in this embodiment of the invention after being segmented by rows; Figure 5 This is a point cloud image of the defective region in an embodiment of the present invention after being segmented by columns; Figure 6This is a point cloud image of the normal region after being segmented by columns in an embodiment of the present invention; Figure 7 This is a schematic diagram showing the deviation of the defect from the reference in an embodiment of the present invention; Figure 8 This is a baseline schematic diagram in an embodiment of the present invention; Figure 9 This is an example diagram of weld mark detection in an embodiment of the present invention; Figure 10 This is a defect image detected by weld marks in an embodiment of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0059] As one possible implementation method, such as Figure 1 As shown, this embodiment provides a battery weld defect detection method based on 3D point cloud segmentation, including:

[0060] A 3D line laser scanning device is used to scan the surface of the battery weld lines to obtain raw point cloud data containing at least one weld line; the raw point cloud data contains multiple weld line trajectory point clouds distributed along the scanning direction.

[0061] The raw point cloud data is arranged in an m x n matrix format;

[0062] Convert the physical coordinates of any point in the original point cloud data into its row and column numbers in the point cloud matrix;

[0063] Based on the row and column numbers, establish an index correspondence between the physical coordinates of the points and the original data storage structure;

[0064] Based on the index correspondence, the corresponding points are read in batches from the original point cloud data. Any rectangle in the original point cloud data is divided into regions by column to form a sub-point cloud for each column.

[0065] Obtain the point set obtained from the segmentation, perform line fitting on the point set obtained from the segmentation, and obtain the benchmark corresponding to each segmentation region;

[0066] Based on the benchmark comparison, the defect location and defect type of the point cloud in each segmented region are obtained.

[0067] In this embodiment, based on the characteristics of the raw point cloud data acquired by the 3D line laser scanning device, the present invention leverages the structural advantages brought by the matrix arrangement of the point cloud. Since the raw point cloud data is organized in an m-row n-column matrix and a correspondence between physical coordinates and matrix indices is established, point cloud segmentation does not require complex spatial searches or neighborhood queries; batch data reading can be achieved simply by using row and column numbers. After segmenting any rectangular region by column, each column of sub-point cloud retains the trajectory continuity along the scanning direction. Based on this characteristic, straight line fitting can be directly performed to construct local baselines, reducing the computational load. In the defect detection stage, by calculating the distance deviation between the point cloud in each segmented region and the corresponding baseline, abnormal points deviating from the baseline can be quickly located. This method requires only one matrix index transformation, one region segmentation, and one line fitting operation, avoiding the time-consuming nearest neighbor search, surface reconstruction, or voxelization processes in traditional point cloud processing, thus offering advantages in computational speed. Simultaneously, because the 3D line laser scanning device acquires data along a fixed direction, the weld trace point cloud has clear directionality and continuity, enabling the straight-line reference to accurately reflect the ideal weld morphology. Through distance threshold determination and deviation pattern analysis, it can accurately identify typical defect types such as weld breaks (continuity interruption), cracks (local depression deviation), and weld beads (local protrusion deviation), effectively improving detection accuracy and efficiency.

[0068] The specific implementation process of this embodiment is as follows:

[0069] Step 1: Use a 3D line laser scanning device to scan the surface of the battery weld lines to obtain raw point cloud data containing at least one weld line; the raw point cloud data contains multiple weld line trajectory point clouds distributed along the scanning direction.

[0070] Step two involves segmenting the original point cloud data. The segmented point cloud is shown below. Figure 2 As shown ( Figure 2 There are multiple point clouds above. For ease of display, only a portion is selected, and their spacing is increased. Different colors represent different regions, and point clouds with the same color belong to the same segmented region.

[0071] The point clouds acquired by 3D line laser scanning devices differ from those acquired by traditional LiDAR. The most significant characteristic of point clouds acquired by traditional LiDAR is their disorder; the points are not arranged in a specific order. The first and second points might be far apart, while the second and third points might be relatively close. However, 3D line laser scanning devices combine the advantages of 2D cameras, allowing for a one-to-one correspondence between the point cloud and the pixels in an image, thus solving the problem of point cloud disorder. This is because 3D line laser scanning devices actively emit lasers during data acquisition. The laser light hits the object's surface and reflects onto the scanner's photosensitive elements. These photosensitive elements are pre-divided into regions, each corresponding to a number. When a region receives a light signal, it generates a pixel and a point cloud corresponding to that number. For example, when region 5 receives a light signal, it generates pixel number 5 and point cloud number 5. Their positions are calculated based on the position of region 5 (this article mainly demonstrates the calculation method and therefore does not explain the hardware principles). A 3D line laser scanning device can be understood as building upon the capabilities of a traditional 2D camera, calculating the point cloud corresponding to each pixel based on sensor data. Therefore, extracting point clouds by row (or column) is equivalent to reading pixels by row (or column). We only need to find the index of pixels in the same row (or column), and then read them according to the found index to obtain the point cloud for that row (or column). Suppose the point cloud acquired by the 3D line laser scanning device has m rows and n columns, and is arranged by row by default (the first row of point clouds is numbered 1, 2, 3…n; the second row is n+1, n+2…2n). Then, dividing the point cloud by column, the index of each column of point clouds is:

[0072] First column: 1, 1+n, 1+2n…1+(m-1)*n;

[0073] Second column: 2, 2+n, 2+2n…2+(m-1)*n;

[0074] …;

[0075] Column k: n, 2n, 3n…m*n;

[0076] …;

[0077] Column n: n, 2n, 3n…m*n;

[0078] When segmenting the original point cloud data, it is necessary to convert the physical coordinates of any point in the original point cloud data into its row and column indices in the point cloud matrix. The calculation process is as follows: obtain the minimum values ​​of the spacing between two adjacent rows, the spacing between two adjacent columns, and the x and y directions. Assume the spacing between two adjacent rows is... The spacing between adjacent columns is The minimum values ​​in the x and y directions are respectively and ;

[0079] The row number of any point in the original point cloud data is:

[0080] ;

[0081] The column number of any point in the original point cloud data is:

[0082] ;

[0083] in, The spacing between two adjacent lines. The spacing between two adjacent columns, and These are the minimum values ​​in the x and y directions, respectively.

[0084] In some possible implementations, the type of fitting chosen varies depending on the segmentation method. If the point cloud is segmented by row, the shape of each region's point cloud is roughly arched, and only a curve model can be selected for fitting. If the number of points in the defective area is large, it will affect the baseline of the fitted data.

[0085] Typically, the point cloud after dividing a normal region into rows looks like this. Figure 3 As shown, the point cloud after segmenting the defective region by row is as follows: Figure 4 As shown, comparing the point clouds segmented by row reveals a significant difference between defective and normal regions. If a spline curve model is used for fitting, the defective portion will also be considered part of the curve. Since row-based segmentation lacks robustness, column-based segmentation is employed. Using this method, the point cloud within each region is mostly normal, and the entire region approximates a straight line, allowing for fitting with a linear model to obtain a baseline.

[0086] The point cloud after segmenting the defective region by column is as follows: Figure 5 As shown, the point cloud after dividing the normal region by columns is as follows: Figure 6 As shown, after segmenting the point cloud by column, regardless of whether there are defective parts, the whole is approximately a straight line. When using a straight line model for fitting, since the normal point cloud accounts for a larger proportion, defects will not affect the fitting result.

[0087] Point clouds segmented by row have defects; the baseline fitted by a straight-line model is as follows: Figure 7 The green part shows a significant overlap between the baseline and the original point cloud, with only the defective areas showing a clear deviation from the baseline. Therefore, in this embodiment, corresponding points are read in batches from the original point cloud data based on the index correspondence, and arbitrary rectangles in the original point cloud data are divided into regions by column.

[0088] Step 3: After segmenting the original point cloud data into arbitrary rectangles by column, select a suitable model for fitting to obtain the baseline for each region, as shown below. Figure 8 As shown.

[0089] Assuming the point set obtained after column partitioning is Q: {p1, p2…pS}, then the process of fitting a straight line is as follows:

[0090] First, calculate the mean point of all points: ;

[0091] Reconstruct matrix X, where each row of X represents the coordinate difference between the original point and the mean point:

[0092] X= ;

[0093] Multiplying the transpose of matrix X by matrix X yields the covariance matrix C: ;

[0094] Next, calculate the eigenvalues ​​of the covariance matrix C: ;

[0095] When expanded, it appears as follows:

[0096] det =0;

[0097] Since the eigenvalues ​​may have multiple solutions: , , Take the largest solution among them. Find the corresponding eigenvector V:

[0098] V= =0;

[0099] After determining the correspondence between x, y, and z, set the value of x to 1, and calculate the values ​​of y and z as y1 and z1, respectively. Finally, normalize the vector [1, y1, z1] to obtain... : = [1,y1,z1];

[0100] Received This refers to the direction of the fitted line. After determining the direction of the line, it is necessary to determine a point on the line to define the line itself. The mean point calculated above can be used directly. As a point on a straight line.

[0101] Based on the calculated direction of the line and the points on the line, the parametric equation of the line can be derived as follows:

[0102] ;

[0103] Where (A, B, C) is the unit direction of the line (denoted as ). (x', y', z') are the points through which the line passes, and are also the mean points calculated above. .

[0104] Step 4: Based on the benchmark comparison, obtain the defect location and defect type of the point cloud in each segmented region.

[0105] According to the formula for the distance from a point to a line, the distance from point P to the fitted line can be calculated as follows:

[0106]

[0107] in:

[0108]

[0109] If the distance D between point P and the fitted line is greater than the specified threshold, then point P can be determined to be a point at the defect location; otherwise, it is a point in the normal area.

[0110] And point P can be projected onto the fitted line to obtain... By comparing the Z-coordinate values ​​of the two points, it can be determined whether point P is a concave or convex point: if the Z-coordinate of P is greater than that of the projected point... If the Z-coordinate is given, then P is a point at the convex point; otherwise, it is a point at the concave point. It can be obtained from the mean point on the straight line. The distance d along the straight line is calculated as follows:

[0111] or ;

[0112] Where + or - depends on The sign is + for positive and - for negative.

[0113] like Figure 9 As shown, this is the original point cloud data, containing a total of 560,000 points. The defective area is the central depression and convex part, as shown... Figure 10 As shown, black represents the normal parts, grayish-white represents the protruding parts, and the whiter the color, the more protrusions there are; red and yellow represent the concave parts, and the yellower the color, the deeper the concavity.

[0114] Compared to traditional weld marks, the weld marks around the battery are smaller, and the defects are also smaller. To capture these defects more effectively, the resolution must be increased, which raises the point cloud data volume to nearly a million records. Using overly complex algorithms not only places high demands on the equipment but also increases processing time. This embodiment combines the characteristics of the point cloud data acquired by the camera. After quickly segmenting the point cloud, only fitting and distance calculation are needed to quickly calculate the point cloud at the defect location. This not only has advantages in speed but also in accuracy, accurately identifying the points at the defect location and distinguishing the type of defect.

[0115] As one possible implementation, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a battery weld defect detection method based on 3D point cloud segmentation.

[0116] As one possible implementation, this embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a battery weld defect detection method based on 3D point cloud segmentation.

[0117] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting battery weld defects based on 3D point cloud segmentation, characterized in that, The specific steps include the following: A 3D line laser scanning device is used to scan the surface of the battery weld lines to obtain raw point cloud data containing at least one weld line; the raw point cloud data includes multiple weld line trajectory point clouds distributed along the scanning direction. The raw point cloud data is arranged in an m x n matrix format; Convert the physical coordinates of any point in the original point cloud data into its row and column numbers in the point cloud matrix; Based on the row and column numbers, establish an index correspondence between the physical coordinates of the points and the original data storage structure; Based on the index correspondence, the corresponding points are read in batches from the original point cloud data. Any rectangle in the original point cloud data is divided into regions by column to form a sub-point cloud for each column. Obtain the point set obtained from the segmentation, perform line fitting on the point set obtained from the segmentation, and obtain the benchmark corresponding to each segmentation region; Based on the benchmark comparison, the defect location and defect type of the point cloud in each segmented region are obtained.

2. The battery weld defect detection method based on 3D point cloud segmentation according to claim 1, characterized in that, The process of converting the physical coordinates of any point in the original point cloud data into its row and column numbers in the point cloud matrix specifically includes: The row number of any point in the original point cloud data is calculated as follows: ； The column number of any point in the original point cloud data is calculated as follows: ； in, The spacing between two adjacent lines. The spacing between two adjacent columns, and Let Px and Py be the minimum values ​​in the x and y directions, respectively, where Px represents the y-coordinate of the point and Py represents the y-coordinate of the point.

3. The battery weld defect detection method based on 3D point cloud segmentation according to claim 1, characterized in that, The process of obtaining the segmented point set includes: Differential color encoding is applied to the segmented point set to generate a visualization image of the weld scar point cloud segmentation: Assign different visual identifier colors to point sets in different areas; Point sets within the same area are rendered using the same identifier color; Different sets of points in different regions are distinguished by color differences.

4. The battery weld defect detection method based on 3D point cloud segmentation according to claim 1, characterized in that, The step of fitting a straight line to the segmented point set to obtain the benchmark corresponding to each segmented region specifically includes: Obtain the point set after splitting by column, and calculate the mean point of all points in the point set; Construct a centralized matrix X, where each row of the centralized matrix X is the coordinate difference between the original point and the mean point; The covariance matrix C is calculated based on the centered matrix X; Find the eigenvalues ​​of the covariance matrix C; Select the largest eigenvalue to solve for the corresponding eigenvector; Normalize the feature vectors to obtain the direction vector of the fitted line; Using the mean point as a point on the straight line, the fitted straight line is determined as the benchmark for each segmented region.

5. The battery weld defect detection method based on 3D point cloud segmentation according to claim 1, characterized in that, The process of obtaining the defect location and defect type of the point cloud within each segmented region specifically includes: The distance from point P in the original point cloud data to the fitted line is determined based on the fitted line and its direction vector. Based on the distance from point P in the original point cloud data to the fitted straight line, and a specified threshold, the location and type of defects in the point cloud are determined.

6. The battery weld defect detection method based on 3D point cloud segmentation according to claim 5, characterized in that, Determining the defect type specifically includes: If the distance between point P and the fitted line is greater than the specified threshold, then point P can be determined to be a point at the defect location. If the distance between point P and the fitted line is within the specified threshold, then point P can be determined to be a point in the normal region.

7. The battery weld defect detection method based on 3D point cloud segmentation according to claim 6, characterized in that, After determining that point P is the defect location, project point P onto the fitted line to obtain... By comparing the Z-coordinate values ​​of the two points, it can be determined whether point P is a concave or convex point: If the Z coordinate of P is greater than the projection point If the Z-coordinate is given, then P is the point at the convexity; If the Z-coordinate of P is less than the projection point If the Z-coordinate is given, then P is the point at the depression.

8. The battery weld defect detection method based on 3D point cloud segmentation according to claim 7, characterized in that, The step of projecting point P onto the fitted line yields... The specific steps are as follows: point Points on the fitted line Moving a distance d along the direction of the fitted line yields: ； in, To project point P onto a point on the fitted line, Let be a point passed through by the straight line, and d be the directed distance. The direction vector for fitting the straight line.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the battery weld defect detection method based on 3D point cloud segmentation as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the battery weld defect detection method based on 3D point cloud segmentation as described in any one of claims 1 to 8.

Images