Power battery laser welding defect detection method based on three-dimensional vision

Through three-dimensional vision technology, combined with radius filtering, random sampling consistency algorithm and curvature threshold adaptive algorithm, the problem of small and medium-sized defect detection of power battery laser welding is solved, and the precise segmentation of welds and intelligent detection of defects is realized, which improves the detection Accuracy and consistency.

CN119006402AInactive Publication Date: 2024-11-22HUBEI UNIV OF ARTS & SCI +1
View PDF 0 Cites 12 Cited by

Patent Information

Application Number
CN202411075482.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-25
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art is difficult to accurately detect small-size defects in laser welding of power batteries, and the texture characteristics of welds and weld pits are similar, resulting in high detection difficulty and poor consistency of manual detection.

Method used

Using a three-dimensional vision-based method, the noise reduction and coarse segmentation of point cloud data is performed through radius filtering and random sampling consistency algorithm, combined with the curvature threshold adaptive algorithm to achieve accurate segmentation of welds, and the improved region growth algorithm is used for defect segmentation. Classification and size calculation.

Benefits of technology

It realizes accurate segmentation and measurement of small and medium-sized defects in welds, improves the intelligent level of laser welding defect detection of power batteries, and improves the accuracy and consistency of detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006402A_ABST
    Figure CN119006402A_ABST
Patent Text Reader

Abstract

The invention discloses a power battery laser welding defect detection method based on three-dimensional vision, and the method comprises the steps: carrying out the noise reduction of obtained laser welding point cloud data through radius filtering, and carrying out the rough segmentation of the laser welding point cloud data after noise reduction through a random sampling consensus algorithm, accurately segmenting the laser welding point cloud data subjected to coarse segmentation by using a curvature threshold self-adaptive algorithm to obtain welding point cloud data; performing defect segmentation on the welding seam point cloud data through an improved region growing algorithm to obtain welding seam defect point cloud data; and on the basis of the welding seam defect point cloud data, through a preset welding seam defect classification and size calculation method, obtaining the type and size of the welding seam defect. Accurate segmentation and size measurement of small-size defects in welding seams can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of laser welding defect detection, and in particular, relates to a power battery laser welding defect detection method based on three-dimensional vision. Background Art

[0002] Laser welding technology is widely used in the welding and packaging of power battery top covers. The quality of welding directly affects the performance, service life and safety of the battery, so the post-welding quality inspection of the top cover is particularly important. Different welding conditions in laser welding may produce a variety of defects, resulting in reduced product quality. Weld nodules and weld pits are two common welding defects. Due to their similar texture characteristics and small defect sizes, they are difficult to detect. Defects are usually detected by manual visual inspection. Due to the influence of manual subjective judgment, the post-welding quality consistency is poor, and for small-sized welding defects, it is impossible to visually determine whether they are defects.

[0003] In recent years, domestic and foreign scholars have conducted research on welding quality inspection issues from the perspective of two-dimensional and three-dimensional visual inspection technologies. Among them, the two-dimensional visual inspection method is usually combined with deep learning to achieve defect detection tasks. Three-dimensional visual inspection technology has the advantages of simplicity, high precision, and non-contact acquisition, and can be used to describe the surface morphology of complex objects. Based on the current status of research on intelligent welding quality inspection at home and abroad, there is little research on the detection and measurement of defects in laser welding of power batteries. There are still difficulties in tasks such as how to accurately segment welds, identify tiny defects in welds, and achieve accurate measurement of dimensions. Summary of the invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a power battery laser welding defect detection method based on three-dimensional vision, which can realize the precise segmentation of small-sized defects in welds and their size measurement, and is of great significance for the intelligent detection of power battery laser welding defects.

[0005] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0006] In a first aspect, the present invention provides a method for detecting defects in power battery laser welding based on three-dimensional vision, comprising:

[0007] The acquired laser welding point cloud data is denoised by using radius filtering, the denoised laser welding point cloud data is roughly segmented by using a random sampling consensus algorithm, and the roughly segmented laser welding point cloud data is accurately segmented by using a curvature threshold adaptive algorithm to obtain weld point cloud data;

[0008] The weld point cloud data is segmented by improved region growing algorithm to obtain weld defect point cloud data;

[0009] Based on the weld defect point cloud data, the category and size of the weld defect are obtained through the preset weld defect classification and size calculation method.

[0010] Furthermore, the denoising of the acquired laser welding point cloud data by using radius filtering and the rough segmentation of the denoised laser welding point cloud data by using a random sampling consensus algorithm include:

[0011] Obtain laser welding point cloud data;

[0012] Radius filtering is used to reduce noise on the acquired laser welding point cloud data;

[0013] A random sampling consensus algorithm is used to fit multiple planes. Among all the fitted planes, the two planes with the largest number of internal points are selected as the two planes connected by the weld, and the intersection equation of the two planes is calculated;

[0014] A cylindrical area with a neighborhood radius of R is made with the intersection of the two planes as the central axis, and the laser welding point cloud data is divided into two parts inside and outside the neighborhood radius. The point cloud data outside the neighborhood is eliminated, and the point cloud data within the neighborhood is used as the rough segmentation result of the laser welding point cloud data; wherein, the neighborhood radius R of the cylindrical area is 2 to 3 times the width of the weld.

[0015] Furthermore, the curvature threshold adaptive algorithm is used to accurately segment the roughly segmented laser welding point cloud data to obtain weld point cloud data, including:

[0016] Calculate the angle between the normal vector of each point in the roughly segmented laser welding point cloud data and the normal vector of the fitting plane :

[0017] ;

[0018] in, is the normal vector of each point in the roughly segmented laser welding point cloud data, is the normal vector of the fitted plane , Indicates the total number of point clouds in the laser welding point cloud data after rough segmentation;

[0019] Calculate the angle threshold between the normal vector of each point in the roughly segmented laser welding point cloud data and the normal vector of the fitting plane :

[0020] ;

[0021] Calculate the curvature of each point in the roughly segmented laser welding point cloud data :

[0022] ;

[0023] in, and Represents a set of orthogonal curvatures at the current point;

[0024] The curvature threshold is obtained by using the curvature threshold adaptive calculation method. ;

[0025] By comparing the curvature threshold with the curvature of each point, the precise segmentation of the laser welding point cloud data after rough segmentation is achieved through the set point cloud segmentation conditions to obtain the weld point cloud data.

[0026] Furthermore, the curvature threshold is compared with the curvature of each point, and the laser welding point cloud data is accurately segmented through the set point cloud segmentation conditions to obtain the weld point cloud data, including:

[0027] According to the curvature of each point in the roughly segmented laser welding point cloud data , get the mean and standard deviation of the curvature of all points, and calculate the curvature threshold :

[0028] ;

[0029] in, is the mean value of the curvature, is the standard deviation of the curvature, is the adjustment coefficient;

[0030] According to the curvature threshold and the size of the curvature of each point, each point is divided into a high threshold point set H and a low threshold point set L, and the ratio of point set H to point set L is obtained. ;

[0031] like If the set point cloud segmentation conditions are met, the curvature threshold meets the requirements, and the high threshold point set H is used as the weld point cloud data;

[0032] like If the set point cloud segmentation condition is not met, the value of the coefficient K is iteratively adjusted until the set point cloud segmentation condition is met;

[0033] Wherein, the point cloud segmentation condition is:

[0034] ;

[0035] Where W is the weld width and R is the neighborhood radius.

[0036] Furthermore, the points are divided into a high threshold point set H and a low threshold point set L according to the curvature threshold and the size of the curvature of each point, including:

[0037] If the curvature of the current point is greater than the curvature threshold, the current point is divided into the high threshold point set H;

[0038] If the curvature of the current point is less than or equal to the curvature threshold, and the angle between the normal vector of the current point and the normal vector of the fitting plane is greater than the angle threshold, the current point is divided into the high threshold point set H, otherwise the current point is divided into the low threshold point set L.

[0039] Further, the high threshold point set H is used as weld point cloud data, including:

[0040] The point cloud denoising method of radius filtering is used to process the outliers in the high threshold point set. The outliers are eliminated by setting the minimum number of neighboring points and the filter radius, and the high threshold point set after filtering is used as the weld point cloud data.

[0041] Furthermore, the defect segmentation of the weld point cloud data by using the improved region growing algorithm to obtain the weld defect point cloud data includes:

[0042] Select a point in the weld point cloud data as the initial seed point for region growth;

[0043] Determine whether the points adjacent to the initial seed point meet the set regional growth conditions. If so, add the points that meet the regional growth conditions to the growth area, and use the points that meet the regional growth conditions as new seed points to continue growing until all points in the weld point cloud data are traversed;

[0044] The set regional growth conditions include:

[0045] S1, determine whether the angle between the normal vector of the current point and the normal vector of the fitting plane is less than the corresponding angle threshold, if so, execute step S3; otherwise, execute step S2;

[0046] S2, determine whether the Euclidean distance from the current point to the ideal surface is less than the corresponding distance threshold, if so, execute step S3;

[0047] S3. Determine whether the curvature of the current point is less than the curvature threshold. If so, mark the current point as a qualified point and use it as a new seed point.

[0048] Furthermore, the method for obtaining the Euclidean distance from the current point to the ideal surface includes:

[0049] Suppose the equation of a circle in a two-dimensional plane is , any point in the plane The coordinates of , then point Euclidean distance to the surface for:

[0050] ;

[0051] in, are the coordinates of the center of the circle, is the radius of the circle, is the Euclidean distance from the current point to the ideal surface.

[0052] Furthermore, the weld defects include weld pit defects and weld bump defects;

[0053] The weld defect classification method comprises:

[0054] Calculate the curvature values ​​of all points in the weld defect point cloud data, and sum and average the curvatures of all points to obtain the average curvature value; if the average curvature value is positive, it indicates a weld nodule defect; if the average curvature value is negative, it indicates a weld pit defect;

[0055] The method for calculating the size of a weld defect comprises: projecting the weld defect point cloud data onto a two-dimensional plane, and obtaining the maximum length, width and depth information of the defect by calculating the size of the projection area.

[0056] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for detecting defects in power battery laser welding based on three-dimensional vision as described in the first aspect is implemented.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] Based on the random sampling consensus algorithm and the curvature threshold adaptive algorithm, the accurate segmentation of the weld point cloud data in the laser welding point cloud data can be achieved. The Euclidean distance feature parameter is introduced into the regional growing algorithm to perform defect segmentation on the weld point cloud data, and the segmentation of the weld defect point cloud data is achieved. The accurate segmentation and size measurement of small and medium-sized defects in the weld are achieved, which is of great significance for the intelligent detection of laser welding defects in power batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The figure is a schematic flow chart of a method for detecting defects in laser welding of power batteries based on three-dimensional vision provided by an embodiment of the present invention;

[0060] Figure 2 The figure shows a schematic diagram of a weld rough segmentation process and results provided by an embodiment of the present invention;

[0061] Figure 3 FIG. 1 is a schematic diagram of a normal vector angle model provided by an embodiment of the present invention;

[0062] Figure 4 The figure is a schematic diagram of a curvature threshold adaptive segmentation process provided by an embodiment of the present invention;

[0063] Figure 5 The figure shows a schematic diagram of a precise segmentation process of a weld provided by an embodiment of the present invention;

[0064] Figure 6 The figure is a schematic diagram of the flow of an improved region growing algorithm provided by an embodiment of the present invention;

[0065] Figure 7 FIG. 1 is a schematic diagram of a Euclidean distance model provided by an embodiment of the present invention;

[0066] Figure 8 Shown is a schematic diagram of a laser welding defect detection platform provided by an embodiment of the present invention;

[0067] Fig. 9 The figure is a schematic diagram of battery defect data provided by an embodiment of the present invention;

[0068] Fig.10 FIG. 1 is a schematic diagram of a segmentation result provided by an embodiment of the present invention;

[0069] Fig.11 Schematic diagram of segmentation results using different algorithms provided in an embodiment of the present invention;

[0070] Fig.12 FIG. 1 is a schematic diagram of a weld nodule defect segmentation result provided by an embodiment of the present invention;

[0071] Fig.13 FIG. 1 is a schematic diagram of a weld pit defect segmentation result provided by an embodiment of the present invention;

[0072] Fig.14 The figure is a schematic diagram of calculating defect size provided by an embodiment of the present invention;

[0073] Fig.15 FIG. 4 is a schematic diagram of measurement error provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0074] The technical concept of the present invention is: using radius filtering to reduce noise on the acquired laser welding point cloud data, using a random sampling consensus algorithm to roughly segment the laser welding point cloud data after noise reduction, and using a curvature threshold adaptive algorithm to accurately segment the roughly segmented laser welding point cloud data to obtain weld point cloud data; using an improved regional growing algorithm to segment the weld point cloud data for defects to obtain weld defect point cloud data; based on the weld defect point cloud data, using a preset weld defect classification and size calculation method to obtain the type and size of the weld defect. It can achieve accurate segmentation of small and medium-sized defects in welds and their size measurement.

[0075] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present application and its application or use.

[0076] Example 1

[0077] In a first aspect, the present invention provides a method for detecting defects in power battery laser welding based on three-dimensional vision, comprising:

[0078] The acquired laser welding point cloud data is denoised by using radius filtering, the denoised laser welding point cloud data is roughly segmented by using a random sampling consensus algorithm, and the roughly segmented laser welding point cloud data is accurately segmented by using a curvature threshold adaptive algorithm to obtain weld point cloud data;

[0079] The weld point cloud data is segmented by improved region growing algorithm to obtain weld defect point cloud data;

[0080] Based on the weld defect point cloud data, the category and size of the weld defect are obtained through the preset weld defect classification and size calculation method.

[0081] The specific process is as follows Figure 1 As shown in the figure, it includes three parts: weld segmentation, defect detection and defect measurement. First, the collected original point cloud is denoised by radius filtering, and multiple planes are fitted by the RANSA algorithm to extract the intersection lines of the weld planes. The points in the intersection neighborhood are retained and the outliers are removed to achieve rough segmentation of the weld. Then, a curvature threshold adaptive algorithm is proposed to achieve accurate segmentation of the weld. Thirdly, the improved regional growing algorithm is used to segment the weld defects. Finally, the defect size calculation criterion is set, and the defect size calculation and classification are completed according to the three-dimensional feature information of the defect.

[0082] The amount of point cloud data obtained by laser scanner is large and contains a lot of noise. Directly processing the point cloud will blur the geometric features of the model, affect the accuracy of the point cloud's description of the surface shape of the object, and reduce the accuracy and efficiency of feature extraction. Extracting welds through point cloud segmentation technology and only analyzing the target area can improve the accuracy of defect judgment. In the weld segmentation part, the method of combining coarse and fine positioning of the weld is used to achieve accurate segmentation of the weld, which includes:

[0083] The purpose of weld rough positioning is to eliminate most of the point cloud data that are irrelevant to the weld area, so that the weld extracted by fine positioning is closer to reality, such as Figure 2 As shown, the entire workflow and processing results of weld rough positioning are demonstrated.

[0084] Firstly, the point cloud data is denoised by using radius filtering to eliminate the influence of outliers on the model. Then, the random sampling consensus algorithm (RANSAC) is used to fit multiple planes and obtain the expressions of each plane. Among all the fitted planes, the two planes with the largest number of inner points are selected as the two planes connected by the weld. The two planes connected by the weld and their corresponding plane equations are obtained, and the equation of the intersection line of the two planes is calculated. Finally, a coarse positioning and segmentation model of the weld is established to obtain a cylindrical area with a neighborhood radius of R as the central axis with the intersection line of the two planes as the central axis. The point cloud data is divided into two parts inside and outside the neighborhood radius. The points within the neighborhood, that is, the point cloud data in the cylindrical model, are recorded as the coarse positioning results of the weld, and the points outside the neighborhood are eliminated to achieve coarse positioning of the weld.

[0085] Through the above segmentation process, a large amount of irrelevant point cloud data is eliminated, and the rough positioning of the weld is completed. In order to further extract the weld and realize automatic and accurate segmentation of the weld, this embodiment proposes a curvature threshold adaptive calculation method and combines the angle threshold between each point and the plane model normal vector as a segmentation condition to realize the precise segmentation of the weld. The specific steps are as follows:

[0086] Firstly, the angle between the normal vector of each point cloud and its fitting plane normal vector after rough positioning processing is calculated, and the angle threshold of each point and its fitting plane normal vector is set; then, the curvature of each point cloud is calculated, and the curvature threshold is calculated by threshold adaptive calculation method; finally, the threshold of each point is compared with the calculated value, and the accurate segmentation of the weld is achieved by setting the point cloud segmentation conditions.

[0087] In the representation of point cloud morphology features, the normal vector is an important parameter. Usually, the angle between the normal vector of each point in the model and the normal vector of the fitting plane is different. The more uneven the point cloud surface is, the larger the angle is. Based on this feature, the weld area can be extracted. The model for calculating the point cloud normal vector and the point-surface angle is as follows: Figure 3 Shown

[0088] The normal vector of each point is , the normal vector of the plane model is , , Represents the total number of point clouds in the laser welding point cloud data after rough segmentation, then the calculated value of the angle between each point in the model and the plane normal vector for:

[0089] ;

[0090] The angle between each point and the normal vector of the plane model is obtained by the above formula After that, we need to calculate the angle threshold of the point-to-face vector , by comparing the angle threshold The angle between the normal vector of the point and the surface The size relationship is used as one of the criteria for determining whether the point belongs to the weld area. Angle threshold Size:

[0091] ;

[0092] Curvature is also an important geometric feature to describe the surface shape of the point cloud. The larger the curvature value, the greater the degree of curvature at that point. Since the curvature of the weld area is different from that of the plane area, by setting an appropriate curvature threshold , compare the curvature calculation value with the curvature threshold as another judgment condition to determine whether the point belongs to the weld area, and realize accurate segmentation of the weld. The setting has a great influence on the accuracy of weld segmentation. Using different curvature thresholds will result in different segmentation results. How to choose a suitable curvature threshold is particularly important.

[0093] Since the curvature value of the point cloud in the weld area is usually higher than that of the parent material, a curvature threshold adaptive calculation method is proposed based on this feature. This method avoids the traditional method of manually setting the threshold and can automatically achieve robust segmentation of the weld. The specific calculation process is as follows: Figure 4 shown.

[0094] First, count the curvature of all points and calculate their mean and standard deviation to get the curvature threshold :

[0095] ;

[0096] in, is the mean value of the curvature, is the standard deviation of the curvature, is the adjustment coefficient;

[0097] According to the current threshold, the point cloud can be divided into two parts: a high threshold point set H and a low threshold point set L. The high threshold point set is the weld point set, and the low threshold point set is the non-weld point set.

[0098] When the ratio of the number of points in the point set H and L calculated by the curvature threshold meets the point cloud segmentation condition, it can be used as the curvature threshold for this segmentation. If the condition is not met, the value of the coefficient K needs to be adjusted iteratively until the condition is met.

[0099] The point cloud segmentation condition is:

[0100] ;

[0101] Where W is the weld width and R is the neighborhood radius.

[0102] Through the above steps, we can get the calculated values ​​of the angle between the point and the surface vector. , Angle threshold , curvature calculation value and curvature threshold When the point , the point is saved in the weld point set In, if , then we need to further determine the relationship between the normal vector of the point and the angle between the fitting plane; if the point , then save the point in the weld point set Otherwise, save the point in the non-weld point set In this case, there are some discrete point clouds in the weld point set A that are distributed outside the weld body, and this type of outliers needs to be filtered out. According to the distribution characteristics of the outliers, the point cloud denoising method of radius filtering is used to process the outliers in the weld point set. The outliers can be removed by setting the minimum number of neighboring points and the filter radius. The result of the weld point cloud set after filtering is the final result of the weld segmentation, such as Figure 5 shown.

[0103] The region growing algorithm has good scalability. Its principle is to divide the region according to the similarity of the point cloud data features. By comparing the difference between the feature data of the seed point and the points within the radius of its surrounding neighborhood, the points with smaller differences, that is, points with similar properties, are classified into the same category of data to achieve regional segmentation. The traditional region growing algorithm only considers two types of feature data, curvature and normal vector angle. It is easy to produce under-segmentation problems for the extraction of weld area defects in this paper. Based on this problem, it is proposed to introduce a new condition that introduces the Euclidean distance feature parameter and the normal vector angle to serially judge whether the point can be used as a seed point to improve the segmentation effect. Specifically, Figure 6 As shown, and are the Euclidean distance from the point to the ideal surface and the distance threshold respectively.

[0104] The region growing algorithm can be simply understood as starting from an original point (seed point) to grow the region. Points with similar geometric properties to the original point (i.e., points that meet the growth conditions) are divided into the same region until all point cloud data are traversed once, and the region can be divided into two regions.

[0105] The specific region growing steps include:

[0106] (1) First, select a point in the entire point cloud data as the starting point of region growth, that is, the seed point.

[0107] (2) Determine whether a point satisfies the growth region condition according to the defined region growth condition, where the growth region condition is as follows: Figure 6 As shown, if the conditions are met, the point can be marked as a new seed point for region growing, if not, it will not be marked.

[0108] (3) Perform region growing: Starting from the initial seed point, check whether the neighboring points meet the growth conditions. The neighboring points are points within a certain distance from the seed point. If the conditions are met, add the point to the current region and continue growing it as a new seed point. This process is repeated until there are no new points that can be added to the current region.

[0109] (4) Deleting seed points: After the current seed point is processed, it is deleted from the candidate seed point list to avoid repeated processing.

[0110] (5) Repeat steps (2), (3), and (4) until no more seed points are added, that is, "Q = 0", and region growing ends.

[0111] The Euclidean distance from a point to an ideal surface can be used to describe the surface texture of a point cloud and the degree of irregularity of the point cloud. Specifically, it is the degree of change in the distance from a point to an ideal surface, that is, the change in the size of the Euclidean distance from a point to an ideal surface of a weld in space. The Euclidean distance from a point to an ideal surface in three-dimensional space can be simplified to a two-dimensional plane, such as Figure 7 Shown is the Euclidean distance model.

[0112] Suppose the equation of a circle in a two-dimensional plane is , any point in the plane The coordinates of , then point Euclidean distance to the surface for:

[0113] ;

[0114] in, are the coordinates of the center of the circle, is the radius of the circle, is the point-to-surface distance of the point. The larger the value, the farther the point deviates from the surface model, and the greater the degree of fluctuation on the point cloud surface. It can be used as a characteristic parameter to describe the change of the point cloud surface morphology.

[0115] For the two types of defects, welding pits and weld nodules, the concave-convex change trends of their surface shapes are obviously different, which can be expressed as the difference between the positive and negative curvature values ​​in geometric features. If the curvature is positive, it means that the area is convex, and if the curvature is negative, it means that the area is concave. Based on this feature, a defect classification method is designed. For the segmented defect area, the curvature values ​​of all points are calculated, and the curvatures of all points in the area are summed and averaged. The defects are classified according to the positive and negative average curvature values ​​of the area.

[0116] The defect point cloud in the three-dimensional space is projected onto a two-dimensional plane, and the maximum length, width and depth information of the defect are obtained by calculating the size of the projection area.

[0117] Example 2

[0118] Based on Example 1, this embodiment provides a specific application example to illustrate the power battery laser welding defect detection method based on three-dimensional vision of the embodiment. The method includes the following steps:

[0119] A battery laser welding defect detection platform was designed. The batteries with different surface conditions were placed on the workbench and the laser scanner was used to obtain the top cover laser welding point cloud data. The scanner has an X-direction resolution of 0.03mm and a Z-direction resolution of 0.006mm. Each contour contains 640 data points. Figure 8 shown.

[0120] The laser scanner is used to obtain the three-dimensional data of pits, bulges and normal weld surfaces. Fig. 9 As shown, (a) to (c) are pit defect batteries, pit welding defect nodes and pit defect point clouds, respectively; (d) to (f) are convex defect batteries, convex welding defects and convex defect point clouds, respectively; (g) to (i) are defect-free batteries, battery welds and weld point clouds, respectively.

[0121] This experiment was conducted under the environment of VS2019 and PCL1.11.1 for data processing and analysis. A group of regions processed by the coarse segmentation algorithm was selected, with a total of 156,474 point clouds. The weld width W = 2.5 mm was set, and the neighborhood radius R of the segmentation model was usually set to 2-3 times the weld width. In this paper, R = 7.5 mm was taken, and the adjustment coefficient K = 10. The optimal segmentation threshold was obtained by the curvature threshold adaptive segmentation algorithm. The segmented high threshold point set H and low threshold point set L, i.e., the number of point clouds of the weld and parent material regions, are 52298 and 104176 respectively. The segmentation results are shown in Fig.10 shown.

[0122] From the above figure, we can see that the proposed method can effectively segment the weld area. After performing the segmentation operation according to the optimal threshold, the main area of ​​the weld is completely segmented, and only a small amount of hole information exists at the edge of the weld, which effectively solves the problem that the holes in the main body of the weld affect the difficulty of subsequent defect identification.

[0123] In order to verify the weld segmentation performance of the proposed method, the accuracy (A), recall (R) and segmentation time T are used as the segmentation evaluation indicators of the two groups of defects. The proposed algorithm is compared with the RANSAC and DBSCAN segmentation algorithms. The weld segmentation effect is shown in the figure below. Fig.11 The evaluation results of different algorithms are shown in Table 1. The calculation criteria of A and R are as follows:

[0124] ;

[0125] ;

[0126] Among them, the positive sample is the weld point cloud, and the negative sample is the battery shell point cloud. TP means that the positive sample point is identified as a positive sample point, TN means that the negative sample point is identified as a negative sample point, FP means that the negative sample point is identified as a positive sample point, and FN means that the positive sample point is identified as a negative sample point.

[0127] Depend on Fig.11 It can be seen that both RANSAC and DBSCAN segmentation algorithms have under-segmentation problems for the weld segmentation task in this article, while the method in this embodiment can effectively extract the weld area and completely retain the weld and defect areas.

[0128]

[0129] Table 1

[0130] As can be seen from Table 1, the segmentation accuracy and efficiency of the algorithm in this embodiment are better than those of the other two algorithms. For weld pit defects, the segmentation accuracy is improved by 6.52% and 4.97% respectively compared with the RANSAC and DBSCAN algorithms, and the processing speed is improved by 1.8 times and 3 times respectively. The evaluation results of weld nodule defects obtained by the method in this paper are also better than those of the other two algorithms, which shows that the method in this paper has high accuracy and real-time performance.

[0131] In the weld defect detection task, weld pits and weld nodules have their own defect size definitions. When the measured defect size does not meet the standard, it can be judged as a defect. The defect size standard and measurement error requirements are shown in Table 2.

[0132]

[0133] Table 2

[0134] In the defect segmentation part, the Euclidean distance feature parameter is introduced to improve the region growing algorithm and segment the weld nodules and pits in the weld. Fig.12 , 13 As shown, (a) to (d) respectively represent the RGB image of weld nodules and weld pit defects, the elevation point cloud map, the defect segmentation results based on the region growing algorithm, and the defect segmentation results based on the improved region growing algorithm.

[0135] The specific location of the defect in the weld point cloud can be seen in the RGB image and the elevation point cloud image. The results of segmenting the two types of defects using the region growing algorithm are as follows: Fig.12(c) and 13 (c), where the point cloud that does not meet the segmentation conditions accounts for a large proportion in the defect area, which will reduce the number of point clouds in the defect area and affect the final defect segmentation accuracy. To solve this problem, this paper introduces the Euclidean distance feature parameter to improve the regional growing algorithm to reduce the proportion of point clouds that do not meet the segmentation conditions. Finally, the points that do not participate in the clustering are merged with the point cloud of the weld area to obtain the final defect segmentation result as shown in Fig.12 (d), as shown in 13 (d).

[0136] like Fig.14 The figure shows the method for measuring the size of weld pits and weld nodules. Fig.14 (a) is a color image of two defects. The length, width and depth information of the defects need to be measured to determine whether they meet the size requirements based on the defect detection requirements. Fig.14 (b) shows the defect area segmented by the improved region growing algorithm, showing the three-dimensional shape characteristics of the defect. Fig.14 (c) is the two-dimensional projection of the point cloud of the defect area. The point cloud in space is projected onto the XY and XZ planes to achieve defect size measurement on the two-dimensional plane.

[0137] In order to verify the defect segmentation accuracy of the algorithm in this paper, the FARO six-degree-of-freedom measuring arm with a measurement accuracy of 0.018mm is used to detect the length, width and depth information of the weld nodule and weld pit defect area, and the measurement results are compared with the defect area size results calculated after the algorithm is improved to analyze the measurement error. The defect size measurement results are shown in Table 3. The measurement error comparison is as follows: Fig.15 shown.

[0138]

[0139] Table 3

[0140] Depend on Fig.15 As can be seen from Table 3, the average measurement error of the defect area size measured by the traditional region growing algorithm is 0.181 mm, and the average measurement error of the defect area size measured by the proposed method is 0.041 mm, and the average measurement accuracy is improved by 77%. The defect size data measured by the improved algorithm is closer to the true value, and the maximum measurement error is 0.062 mm, which meets the defect detection requirements.

[0141] This example verifies the feasibility of the method in Example 1 and compares the processing results with other defect segmentation methods. The experimental results show that in the weld segmentation part, compared with the DBSCAN and RANSAC algorithms, the proposed method has higher segmentation efficiency and accuracy. In the defect segmentation part, the maximum measurement error of the defect size measured by the proposed method is 0.062mm, and the average measurement error is 0.041mm. Compared with the algorithm before the improvement, the average measurement accuracy is improved by 77%, which meets the requirements of laser welding defect detection.

[0142] Example 3

[0143] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the power battery laser welding defect detection method based on three-dimensional vision as described in Example 1 is implemented.

[0144] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0145] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0146] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0147] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0148] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A power battery laser welding defect detection method based on three-dimensional vision, It is characterized in that include: The acquired laser welding point cloud data is denoised by using radius filtering, the denoised laser welding point cloud data is roughly segmented by using a random sampling consensus algorithm, and the roughly segmented laser welding point cloud data is accurately segmented by using a curvature threshold adaptive algorithm to obtain weld point cloud data; The weld point cloud data is segmented by improved region growing algorithm to obtain weld defect point cloud data; Based on the weld defect point cloud data, the category and size of the weld defect are obtained through the preset weld defect classification and size calculation method.

2. According to the three-dimensional vision-based power battery laser welding defect detection method of claim 1, It is characterized in that The method of using radius filtering to reduce noise on the acquired laser welding point cloud data and using a random sampling consensus algorithm to roughly segment the noise-reduced laser welding point cloud data includes: Obtain laser welding point cloud data; Radius filtering is used to reduce noise on the acquired laser welding point cloud data; A random sampling consensus algorithm is used to fit multiple planes. Among all the fitted planes, the two planes with the largest number of internal points are selected as the two planes connected by the weld, and the intersection equation of the two planes is calculated; A cylindrical area is made with the intersection of the two planes as the central axis, and the laser welding point cloud data is divided into two parts inside and outside the cylindrical area. The point cloud data outside the cylindrical area is eliminated, and the point cloud data inside the cylindrical area is used as the rough segmentation result of the laser welding point cloud data; wherein the neighborhood radius of the cylindrical area is 2 to 3 times the width of the weld.

3. The power battery laser welding defect detection method based on three-dimensional vision according to claim 1, It is characterized in that The method of using a curvature threshold adaptive algorithm to accurately segment the roughly segmented laser welding point cloud data to obtain weld point cloud data includes: Calculate the angle between the normal vector of each point in the roughly segmented laser welding point cloud data and the normal vector of the fitting plane : ; in, is the normal vector of each point in the roughly segmented laser welding point cloud data, is the normal vector of the fitted plane, , Indicates the total number of point clouds in the laser welding point cloud data after rough segmentation; Calculate the angle threshold between the normal vector of each point in the roughly segmented laser welding point cloud data and the normal vector of the fitting plane : ; Calculate the curvature of each point in the roughly segmented laser welding point cloud data : ; in, and Represents a set of orthogonal curvatures at the current point; The curvature threshold is obtained by using the curvature threshold adaptive calculation method. ; By comparing the curvature threshold with the curvature of each point, the precise segmentation of the laser welding point cloud data after rough segmentation is achieved through the set point cloud segmentation conditions to obtain the weld point cloud data.

4. The power battery laser welding defect detection method based on three-dimensional vision according to claim 3, It is characterized in that The curvature threshold is compared with the curvature of each point, and the laser welding point cloud data is accurately segmented through the set point cloud segmentation conditions to obtain the weld point cloud data, including: According to the curvature of each point in the roughly segmented laser welding point cloud data , get the mean and standard deviation of the curvature of all points, and calculate the curvature threshold : ; in, is the mean value of the curvature, is the standard deviation of the curvature, is the adjustment coefficient; According to the curvature threshold and the size of the curvature of each point, each point is divided into a high threshold point set H and a low threshold point set L, and the ratio of point set H to point set L is obtained. ; like If the set point cloud segmentation conditions are met, the curvature threshold meets the requirements, and the high threshold point set H is used as the weld point cloud data; like If the set point cloud segmentation condition is not met, the value of the coefficient K is iteratively adjusted until the set point cloud segmentation condition is met; Wherein, the point cloud segmentation condition is: ; Where W is the weld width and R is the neighborhood radius.

5. The power battery laser welding defect detection method based on three-dimensional vision according to claim 4, It is characterized in that The method of dividing each point into a high threshold point set H and a low threshold point set L according to the curvature threshold and the curvature of each point includes: If the curvature of the current point is greater than the curvature threshold, the current point is divided into the high threshold point set H; If the curvature of the current point is less than or equal to the curvature threshold, and the angle between the normal vector of the current point and the normal vector of the fitting plane is greater than the angle threshold, the current point is divided into the high threshold point set H, otherwise the current point is divided into the low threshold point set L.

6. The power battery laser welding defect detection method based on three-dimensional vision according to claim 4, It is characterized in that The method of using the high threshold point set H as weld point cloud data includes: The point cloud denoising method of radius filtering is used to process the outliers in the high threshold point set. The outliers are eliminated by setting the minimum number of neighboring points and the filter radius, and the high threshold point set after filtering is used as the weld point cloud data.

7. The power battery laser welding defect detection method based on three-dimensional vision according to claim 1, It is characterized in that The defect segmentation of the weld point cloud data by using the improved region growing algorithm to obtain the weld defect point cloud data includes: Select a point in the weld point cloud data as the initial seed point for region growth; Determine whether the points adjacent to the initial seed point meet the set regional growth conditions. If so, add the points that meet the regional growth conditions to the growth area, and use the points that meet the regional growth conditions as new seed points to continue growing until all points in the weld point cloud data are traversed; The set regional growth conditions include: S1, determine whether the angle between the normal vector of the current point and the normal vector of the fitting plane is less than the corresponding angle threshold, if so, execute step S3; otherwise, execute step S2; S2, determine whether the Euclidean distance from the current point to the ideal surface is less than the corresponding distance threshold, if so, execute step S3; S3. Determine whether the curvature of the current point is less than the curvature threshold. If so, mark the current point as a qualified point and use it as a new seed point.

8. The power battery laser welding defect detection method based on three-dimensional vision according to claim 7, It is characterized in that The method for obtaining the Euclidean distance from the current point to the ideal surface includes: Suppose the equation of a circle in a two-dimensional plane is , any point in the plane The coordinates of , then point Euclidean distance to the surface for: ; in, are the coordinates of the center of the circle, is the radius of the circle, is the Euclidean distance from the current point to the ideal surface.

9. The power battery laser welding defect detection method based on three-dimensional vision according to claim 1, It is characterized in that The weld defects include weld pit defects and weld bump defects; The weld defect classification method comprises: Calculate the curvature values ​​of all points in the weld defect point cloud data, and sum and average the curvatures of all points to obtain the average curvature value; if the average curvature value is positive, it indicates a weld nodule defect; if the average curvature value is negative, it indicates a weld pit defect; The method for calculating the size of a weld defect comprises: projecting the weld defect point cloud data onto a two-dimensional plane, and obtaining the maximum length, width and depth information of the defect by calculating the size of the projection area.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the computer program is executed by a processor, the method for detecting defects in power battery laser welding based on three-dimensional vision as described in any one of claims 1 to 9 is implemented.

Citation Information

Cited By

  • Solid-state laser radar module packaging method and system

    CN119634959A

  • A solid-state laser radar module packaging method and system

    CN119634959B

  • Welding seam point cloud region segmentation method and system based on multi-feature dynamic threshold analysis

    CN119850643A

  • Image processing method and system based on fusion growing point optimization algorithm and Unet network

    CN120125583A

  • Intelligent planning method for welding path of welding seam

    CN120972763A