Battery defect detection method and system based on multi-mode sensor

Abstract

The invention discloses a battery defect detection method and system based on a multi-modal sensor, and the method comprises the steps: carrying out the image collection of a to-be-detected battery, and obtaining a to-be-detected image; analyzing the to-be-detected image according to a preset image processing algorithm, and determining a target area suspected to have defects and visual features in the image; scanning the target area, and extracting tactile features; obtaining a visual preliminary defect probability corresponding to the visual feature by using a random forest algorithm, and obtaining a tactile preliminary defect probability corresponding to the tactile feature; and according to the visual preliminary defect probability and the tactile preliminary defect probability, obtaining a target defect probability under the target probability distribution based on a D-S theory, so as to determine whether a target defect exists in the target area according to the target defect probability. Through a mode of combining a touch sensor and image acquisition, the 2D visual data and the 3D touch data are fused in a multi-modal manner, so that the reliability of final judgment is improved, and the cost of detection equipment is effectively controlled.

Description

Technical Field

[0001] The present disclosure relates to the technical field of battery detection, and in particular, to a battery defect detection method and system based on multi-modal sensors. Background Art

[0002] In the fields of battery manufacturing and quality inspection, the integrity of the battery appearance and the welding quality are related to the service life, reliability of the battery, and the overall safety of the end product. In the actual production process, various problems may occur on the battery appearance, such as scratches, pits, oxidation, dimensional deviation, etc. These defects may directly lead to serious safety accidents such as battery leakage, fire, or even explosion.

[0003] Currently, in the detection of battery appearance and welding quality, the existing detection technologies have many limitations. Although the traditional 2D vision detection technology can detect the battery appearance and welding defects to a certain extent, since the battery case is usually made of metal material and has high reflectivity, this makes the traditional 2D vision system easily interfered by the reflective effect of the metal surface during the detection process, resulting in adverse phenomena such as light spots and shadows in the image, thus affecting the accuracy and reliability of the detection, and it is difficult to effectively identify the subtle defects on the highly reflective surface such as the aluminum case. To overcome the above deficiencies, some enterprises consider using 3D cameras for detection, which can improve the detection accuracy and reliability to a certain extent, but the equipment cost of 3D cameras is high. For most battery manufacturing enterprises, this undoubtedly increases a huge production cost burden and limits their large-scale application.

[0004] Therefore, those skilled in the art urgently need to develop a new technical solution to solve the above problems. Summary of the Invention

[0005] To overcome the problems existing in the related technologies, the present disclosure provides a battery defect detection method and system based on multi-modal sensors.

[0006] According to the first aspect of the embodiments of the present disclosure, a battery defect detection method based on multi-modal sensors is provided. The method is applied to a battery defect detection system based on multi-modal sensors. The system includes: an image acquisition module, a tactile sensor module, a processor, and a motion control platform. The method includes:

[0007] Acquire an image of the battery to be detected on the motion control platform through the image acquisition module to obtain a to-be-detected image;

[0008] Analyze the to-be-detected image by the processor according to a preset image processing algorithm, determine the target area suspected of having defects in the image, and extract the visual features corresponding to the target area;

[0009] The target area is scanned by the tactile sensor, and the processor extracts tactile features according to the scanning result of the target area;

[0010] The processor uses the random forest algorithm to obtain the preliminary visual defect probability corresponding to the visual features and the preliminary tactile defect probability corresponding to the tactile features;

[0011] The processor obtains the target defect probability under the target probability assignment according to the preliminary visual defect probability and the preliminary tactile defect probability based on the D-S theory, so as to determine whether there is a target defect in the target area according to the target defect probability.

[0012] Optionally, the step that the processor obtains the target defect probability under the target probability assignment according to the preliminary visual defect probability and the preliminary tactile defect probability based on the D-S theory, so as to determine whether there is a target defect in the target area, includes:

[0013] The preliminary visual defect probability and the preliminary tactile defect probability are converted into the basic probability assignment of visual defects and the basic probability assignment of tactile defects of the D-S theory, and the conflict coefficient of the image acquisition module and the tactile sensor is calculated;

[0014] If the conflict coefficient is greater than or equal to the preset coefficient threshold, the basic probability assignment of visual defects and the basic probability assignment of tactile defects are corrected by the weight coefficient of the image acquisition module and the weight coefficient of the tactile sensor, and the target defect probability under the target probability assignment is obtained according to the corrected basic probability assignment of visual defects and the corrected basic probability assignment of tactile defects;

[0015] If the conflict coefficient is less than the preset coefficient threshold, the target defect probability under the target probability assignment is obtained according to the basic probability assignment of visual defects and the basic probability assignment of tactile defects;

[0016] If the target defect probability is greater than or equal to the preset defect threshold corresponding to the target defect, it is determined that there is a target defect in the target area;

[0017] If the target defect probability is less than the preset defect threshold corresponding to the target defect, it is determined that there is no target defect in the target area.

[0018] Optionally, the step of obtaining the target defect probability under the target probability assignment according to the corrected basic probability assignment of visual defects and the corrected basic probability assignment of tactile defects, or the step of obtaining the target defect probability under the target probability assignment according to the basic probability assignment of visual defects and the basic probability assignment of tactile defects, includes:

[0019] Determine the target defect probability according to the corrected basic probability assignment of visual defects and the corrected basic probability assignment of tactile defects, or according to the basic probability assignment of visual defects and the basic probability assignment of tactile defects, through a preset target defect probability formula.

[0020] Among them, the target defect probability formula is:

[0021] , is the target defect probability in the target probability assignment. is the probability of existing defects in the basic probability assignment of visual defects. is the probability of existing defects in the basic probability of tactile defects, and K is the conflict coefficient.

[0022] Optionally, the correction of the basic probability assignment of visual defects and the basic probability assignment of tactile defects by the weight coefficient of the image acquisition module and the weight coefficient of the tactile sensor includes:

[0023] Correct the basic probability assignment of visual defects through the weight coefficient of the image acquisition module to obtain the corrected basic probability assignment of visual defects. , where is the probability of existing defects in the basic probability assignment of visual defects. is the weight coefficient of the image acquisition module.

[0024] Correct the basic probability assignment of tactile defects through the weight coefficient of the tactile sensor to obtain the corrected basic probability assignment of tactile defects. , where is the probability of existing defects in the basic probability of tactile defects. is the weight coefficient of the tactile sensor.

[0025] Optionally, the motion control platform includes a robotic arm and a placement table. The placement table is used to place the battery to be detected, and the robotic arm is used to move the battery to be detected; the image acquisition of the battery to be detected on the motion control platform by the image acquisition module includes:

[0026] Acquire the image to be detected of one detection surface of the battery to be detected through the image acquisition module on the placement table.

[0027] Move the battery to be detected by the robotic arm during the image acquisition process to ensure that the images to be detected of all detection surfaces of the battery to be detected are acquired.

[0028] Optionally, a planar coordinate system is set on the placement table. The scanning of the target area by the tactile sensor includes:

[0029] Determine the coordinate position of the target area in the plane coordinate system;

[0030] Determine the contact path when the tactile sensor scans the target area according to the coordinate position;

[0031] Scan the target area through the tactile sensor according to the contact path.

[0032] According to the second aspect of the disclosed embodiments of the present invention, there is provided a battery defect detection system based on a multi-modal sensor, which is used to implement the battery defect detection method based on a multi-modal sensor described in the first aspect of the disclosed embodiments of the present invention. The system includes: an image acquisition module, a tactile sensor module, a processor, and a motion control platform;

[0033] The image acquisition module and the tactile sensor module are arranged above the motion control platform and are electrically connected to the processor respectively. The motion control platform is used to place the battery to be detected.

[0034] Optionally, the motion control platform includes a robotic arm and a placement table. The placement table is used to place the battery to be detected, and the robotic arm is used to move the battery to be detected.

[0035] Optionally, the robotic arm is used to move the battery to be detected during the process of acquiring images to ensure that images of all detection surfaces of the battery to be detected are acquired. And after the processor analyzes the image of the battery to be detected through an image processing algorithm and determines the target area, the robotic arm moves the battery to be detected to align the detection surface corresponding to the target area with the tactile sensor.

[0036] Optionally, a plane coordinate system is provided on the placement table.

[0037] In summary, through the technical solutions in the disclosed embodiments of the present invention, the following beneficial effects can be brought:

[0038] (1) First, use random forest to classify visual features and tactile features, output the preliminary defect probability, and then use the D-S theory to fuse the confidence degrees of multiple sensors, and adopt a multi-modal fusion algorithm to improve the reliability of the final determination;

[0039] (2) Use relatively low-cost devices such as a high-resolution industrial camera and an array tactile sensor as the image acquisition module and the tactile sensor. Compared with expensive 3D cameras, the cost in the process of battery defect detection is effectively controlled;

[0040] (3) Further verify the "suspected defects" in the target area through the tactile sensor. Concentrate the area that the tactile sensor focuses on detecting in the target area, which can solve the problem of low working efficiency of the tactile sensor and effectively improve the battery defect detection efficiency.

[0041] Other features and advantages of the present invention will be described in detail in the following specific implementation section. Brief Description of the Drawings

[0042] The drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific implementation, they are used to explain the present disclosure, but do not constitute a limitation to the present disclosure. In the drawings:

[0043] Figure 1 is a schematic flowchart of a battery defect detection method based on a multi-modal sensor shown according to an exemplary embodiment;

[0044] Figure 2 is according to Figure 1 shown is a schematic flowchart of an image acquisition method;

[0045] Figure 3 is according to Figure 1 shown is a flowchart schematic of a method for a tactile sensor to scan a target area;

[0046] Figure 4 is according to Figure 1 shown is a schematic flowchart of a method for determining a target defect;

[0047] Figure 5 is a schematic structural diagram of a battery defect detection system based on a multi-modal sensor shown according to an exemplary embodiment. Specific Implementation

[0048] The following details the specific implementation of the present invention disclosed in conjunction with the drawings. It should be understood that the specific implementation described herein is only for explaining and understanding the present disclosure and does not limit the present disclosure.

[0049] The battery defect detection method in the disclosed embodiments of the present invention combines visual detection and tactile detection to collect battery housing data in multiple dimensions, and fuses the data in multiple dimensions (2D visual data + 3D tactile data) to obtain a high-precision detection result and identify defects on the surface of the battery housing. This battery defect detection method is applied to a battery defect detection system based on a multi-modal sensor. The system includes: an image acquisition module, a tactile sensor module, a processor, and a motion control platform. The image acquisition module and the tactile sensor module are respectively used to collect data in the visual dimension and the tactile dimension. The processor is used to analyze the above data, perform fusion processing on the data, and then obtain the result of defect detection. The motion control platform can be used to place the battery housing to be detected and move the battery housing so that each surface to be detected is aligned with the image acquisition module or the tactile sensor.

[0050] Figure 1 is a schematic flowchart of a battery defect detection method based on a multi-modal sensor shown according to an exemplary embodiment, as Figure 1 shown, the method includes:

[0051] In step 101, the image acquisition module is used to acquire an image of the battery to be detected on the motion control platform to obtain an image to be detected.

[0052] Exemplarily, the motion control platform includes a robotic arm and a placement table. The placement table is used to place the battery to be detected, and the robotic arm is used to move the battery to be detected. It can be understood that the housing of each battery to be detected includes several detection surfaces. When acquiring the image to be detected through the image acquisition module, it is necessary to completely acquire the images to be detected of each detection surface.

[0053] Specifically, Figure 2 is according to Figure 1 shown, a schematic flowchart of an image acquisition method, as Figure 2 shown, this step 101 includes:

[0054] In step 1011, the image acquisition module is used to acquire the battery to be detected on the placement table to obtain an image to be detected of one detection surface of the battery to be detected.

[0055] In step 1012, the robotic arm is used to move the battery to be detected during the image acquisition process to ensure that the images to be detected of all detection surfaces of the battery to be detected are acquired.

[0056] Exemplarily, the robotic arm is used to move the battery to be detected on the placement table so that each detection surface is aligned with the image acquisition module to complete the image acquisition of this detection surface.

[0057] In step 102, the processor analyzes the image to be detected according to a preset image processing algorithm, determines the target area suspected of having defects in the image, and extracts the visual features corresponding to the target area.

[0058] Exemplarily, when the processor analyzes the image to be detected, each detection surface is analyzed separately. The preset image processing algorithms include existing image processing methods such as multi-light source imaging algorithm, edge enhancement algorithm, and contour extraction algorithm. The specific steps for analyzing the image to be detected according to the preset image processing algorithm are as follows: for each detection surface, the images to be detected of the detection surface collected by the image acquisition modules at different angles are fused to highlight the defect features. Use Sobel or Scharr operators to calculate the image gradient of the image to be detected on the detection surface, strengthen the linear features (for example, scratches usually appear as long strip-shaped high-gradient areas), determine the contour of the weld area through contour extraction technology, and adopt methods such as local contrast enhancement to highlight the concave area. In addition, the oxidation area, size deviation area, weld porosity area, crack area, welding slag area, false welding area, etc. on the surface of the battery case can also be determined by existing image processing algorithms, which will not be elaborated in the disclosed embodiments of the present invention. It can be understood that the above weld area, concave area, weld porosity area, crack area, etc. are all target areas suspected of having defects. The reason why the above target areas are called suspected of having defects is that the 2D visual data collected by the image acquisition module is easily affected by the reflection of the metal outer surface and other environmental light, resulting in certain deviations in the detection results. To ensure the accuracy of the defect detection results of each detection surface of the battery case, the defect detection results obtained from the 2D visual data are called target areas suspected of having defects, and then the "suspected defects" are further verified through the following steps 103-105.

[0059] The visual features corresponding to the target area include: texture features (LBP, Haralick), shape descriptors (Hu moments), and deep learning features (output of the middle layer of CNN).

[0060] In step 103, the target area is scanned by the tactile sensor, and the processor extracts tactile features according to the scanning results of the target area.

[0061] Exemplarily, the "suspected defects" in the target area are further verified by the tactile sensor to identify the contact force / pressure distribution, false welding, structural deformation, scratches, pits, convex point deformation, etc. in the target area. The tactile features corresponding to the target area include: mean / variance of pressure distribution, force sense time series features (zero crossing rate), and surface roughness.

[0062] A plane coordinate system is set on the placement table. Specifically, Figure 3 It is based onFigure 1 The flowchart schematic diagram of a method for a tactile sensor to scan a target area is shown, as Figure 3 shown. Step 103 includes:

[0063] In step 1031, determine the coordinate position of the target area in the plane coordinate system.

[0064] In step 1032, determine the contact path when the tactile sensor scans the target area according to the coordinate position.

[0065] In step 1033, scan the target area by the tactile sensor according to the contact path.

[0066] Exemplarily, the plane coordinate system takes the lower left corner vertex of the placement table as the center, the lower side edge of the placement table as the straight line where the x-axis is located, and the left side edge of the placement table as the straight line where the y-axis is located. Each point in the placement table can find a corresponding coordinate position in the plane coordinate system. After determining the coordinate position corresponding to the target area, a reasonable contact path for the tactile sensor to scan can be set according to the position of the target area and the size of the target area, so that the tactile sensor scans according to the specified contact path.

[0067] In addition, it should be noted that after determining the target area, the robotic arm can be used to move the battery to be detected so that the detection surface corresponding to the target area is aligned with the tactile sensor, which is convenient for the tactile sensor to perform the next scanning action.

[0068] In step 104, the processor uses the random forest algorithm to obtain the visual preliminary defect probability corresponding to the visual feature and the tactile preliminary defect probability corresponding to the tactile feature.

[0069] Exemplarily, a random forest classifier is established, and the visual feature and the tactile feature are respectively used as the inputs of the random forest classifier, and the visual preliminary defect probability and the tactile preliminary defect probability are obtained according to the output of the random forest classifier.

[0070] In step 105, the processor obtains the target defect probability under the target probability assignment according to the visual preliminary defect probability and the tactile preliminary defect probability based on the D-S theory, so as to determine whether there is a target defect in the target area according to the target defect probability.

[0071] Exemplarily, according to the visual preliminary defect probability and the tactile preliminary defect probability, they are converted into the output confidence under the D-S theory, and the probability assignment of the target defect is obtained, so as to determine whether there is a target defect in the target area. Among them, the D-S theory is short for Dempster-Shafer theory, which is a method for dealing with uncertainty problems and describes uncertain information through "interval estimation" rather than "point estimation". In the disclosed embodiments of the present invention, the random forest is first used to classify the visual features and tactile features to output the preliminary defect probability, and then the D-S theory is used to fuse the confidences of multiple sensors, and a multi-modal fusion algorithm is adopted to improve the reliability of the final determination.

[0072] Specifically, Figure 4 is based on Figure 1 FIG. shows a schematic flowchart of a method for determining a target defect. As Figure 4 shown, step 105 includes:

[0073] In step 1051, the visual preliminary defect probability and the tactile preliminary defect probability are converted into the visual defect basic probability assignment and the tactile defect basic probability assignment of the D-S theory, and the conflict coefficient of the image acquisition module and the tactile sensor is calculated.

[0074] Exemplarily, the calculation formula of the conflict coefficient K is: , is the probability of the existence of a defect in the visual defect basic probability assignment, is the probability of the existence of a defect in the tactile defect basic probability.

[0075] In step 1052, if the conflict coefficient is greater than or equal to the preset coefficient threshold, the visual defect basic probability assignment and the tactile defect basic probability assignment are corrected by the weight coefficient of the image acquisition module and the weight coefficient of the tactile sensor, and the target defect probability under the target probability assignment is obtained according to the corrected visual defect basic probability assignment and the corrected tactile defect basic probability assignment.

[0076] Exemplarily, if the conflict coefficient is greater than or equal to the preset coefficient threshold, it indicates that there is a high contradiction between the 2D visual data collected by the image acquisition module and the 3D tactile data collected by the tactile sensor. At this time, it is necessary to correct the visual defect basic probability assignment and the tactile defect basic probability assignment.

[0077] Specifically, correcting the visual defect basic probability assignment and the tactile defect basic probability assignment by the weight coefficient of the image acquisition module and the weight coefficient of the tactile sensor includes:

[0078] Correcting the visual defect basic probability assignment by the weight coefficient of the image acquisition module to obtain the corrected visual defect basic probability assignment , where is the probability of a defect in the basic probability assignment for this visual defect, is the weight coefficient of the image acquisition module.

[0079] The basic probability assignment of this tactile defect is corrected by the weight coefficient of the tactile sensor to obtain the corrected basic probability assignment of the tactile defect , where, is the probability of a defect in the basic probability of this tactile defect, is the weight coefficient of the tactile sensor. The weight coefficient of the image acquisition module and the weight coefficient of the tactile sensor are obtained through experiments.

[0080] In step 1053, if the conflict coefficient is less than the preset coefficient threshold, the target defect probability under the target probability assignment is obtained according to the basic probability assignment of this visual defect and the basic probability assignment of the tactile defect.

[0081] Exemplarily, if the conflict coefficient is less than the preset coefficient threshold, it indicates that there is no high contradiction between the 2D visual data collected by the image acquisition module and the 3D tactile data collected by the tactile sensor, and it is not necessary to correct the basic probability assignment of the visual defect and the basic probability assignment of the tactile defect. The target defect probability can be directly determined through the basic probability assignment of the visual defect and the basic probability assignment of the tactile defect.

[0082] Specifically, the target defect probability under the target probability assignment is obtained according to the corrected basic probability assignment of the visual defect and the corrected basic probability assignment of the tactile defect, or, the target defect probability under the target probability assignment is obtained according to the basic probability assignment of this visual defect and the basic probability assignment of the tactile defect, including:

[0083] According to the corrected basic probability assignment of the visual defect and the corrected basic probability assignment of the tactile defect, or, according to the basic probability assignment of this visual defect and the basic probability assignment of the tactile defect, the target defect probability is determined through a preset target defect probability formula,

[0084] where, the target defect probability formula is:

[0085] , is the target defect probability in this target probability assignment, is the probability of a defect in the basic probability assignment of this visual defect, is the probability of a defect in the basic probability of this tactile defect, and K is the conflict coefficient.

[0086] In step 1054, if the target defect probability is greater than or equal to the preset defect threshold corresponding to this target defect, it is determined that there is a target defect in this target area.

[0087] In step 1055, if the target defect probability is less than the preset defect threshold corresponding to the target defect, it is determined that there is no target defect in the target area.

[0088] Exemplarily, it can be understood that the types of target defects include: welds, dents, oxidation, welding slag, false soldering, etc. The annotation library contains the preset defect thresholds corresponding to each target type. By analyzing the 2D visual data + 3D tactile data, the target defect type can be initially determined. If the target defect probability is greater than or equal to the preset defect threshold corresponding to the target defect type, it is determined that there is the target defect in the target area; otherwise, it is determined that there is no such target defect in the target area. For the preset defect thresholds in the standard library, a high-quality annotation library (covering the main defect types) can be established in the initial stage, active learning can be introduced in the middle stage to reduce the incremental cost, and data combination and transfer learning can be combined in the long term to achieve efficient and sustainable model optimization.

[0089] Figure 5 It is a schematic structural diagram of a battery defect detection system based on a multi-modal sensor shown according to an exemplary embodiment. As Figure 5 shown, the system 500 includes: an image acquisition module 510, a tactile sensor module 520, a processor (not shown in the figure), and a motion control platform (including a robotic arm 531 and a placement table 532); the image acquisition module 510 and the tactile sensor module 520 are arranged above the motion control platform (an additional image acquisition module 510 can also be added to the side of the motion control platform for acquiring images of other detection surfaces of the battery to be detected), and are respectively electrically connected to the processor, and the motion control platform is used for placing the battery to be detected.

[0090] Specifically, the image acquisition module 510 can be a high-resolution industrial camera, which contains a circularly polarized light source to suppress reflection and structured light, and can emit light sources at different angles from different colors to suppress reflection. The tactile sensor 520 can be an array tactile sensor. The prices of high-resolution industrial cameras and array tactile sensors are relatively low, which can effectively control the cost in the process of battery defect detection.

[0091] The image acquisition module 510 is used to acquire images of the battery to be detected on the motion control platform, so as to obtain the images to be detected; the tactile sensor 520 is used to scan the target area; the processor is used to analyze the images to be detected according to the preset image processing algorithm, determine the target area suspected of having defects in the images and extract the visual features corresponding to the target area; extract the tactile features according to the scanning results of the target area; use the random forest algorithm to obtain the preliminary visual defect probability corresponding to the visual features, and obtain the preliminary tactile defect probability corresponding to the tactile features; according to the preliminary visual defect probability and the preliminary tactile defect probability, obtain the target defect probability under the target probability assignment according to the D-S theory, so as to determine whether there are target defects in the target area according to the target defect probability.

[0092] The motion control platform includes a robotic arm 531 and a placement table 532. The placement table 532 is used to place the battery to be detected, and the robotic arm 531 is used to move the battery to be detected. During the image acquisition process, the battery to be detected is moved to ensure that the images of all detection surfaces of the battery to be detected are acquired. Moreover, after the processor analyzes the image of the battery to be detected through the image processing algorithm and determines the target area, the battery to be detected is moved to align the detection surface corresponding to the target area with the tactile sensor.

[0093] A planar coordinate system is set on the placement table. Determine the coordinate position of the target area in the planar coordinate system; determine the contact path when the tactile sensor scans the target area according to the coordinate position; scan the target area by the tactile sensor according to the contact path.

[0094] In summary, the present invention discloses a battery defect detection method and system based on multi-modal sensors. The method includes: acquiring images of the battery to be detected to obtain the images to be detected; analyzing the images to be detected according to the preset image processing algorithm to determine the target area suspected of having defects in the images and the visual features; scanning the target area to extract the tactile features; using the random forest algorithm to obtain the preliminary visual defect probability corresponding to the visual features, and obtaining the preliminary tactile defect probability corresponding to the tactile features; according to the preliminary visual defect probability and the preliminary tactile defect probability, obtaining the target defect probability under the target probability assignment according to the D-S theory, so as to determine whether there are target defects in the target area according to the target defect probability. It can combine the tactile sensor with image acquisition, fuse 2D visual data and 3D tactile data in a multi-modal manner, improve the reliability of the final determination, and effectively control the cost of the detection equipment.

[0095] The preferred embodiments of the present disclosure have been described in detail above in conjunction with the accompanying drawings. However, the present disclosure is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present disclosure, various simple modifications can be made to the technical solutions of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0096] In addition, it should be noted that, in the various specific technical features described in the above specific embodiments, they can be combined in any suitable manner without conflict. To avoid unnecessary repetition, the present disclosure will not separately describe various possible combination manners.

[0097] Furthermore, any combination can be made between various different embodiments of the present disclosure, as long as it does not violate the idea of the present disclosure, and it should also be regarded as the content disclosed by the present disclosure.

Claims

1. A battery defect detection method based on multimodal sensors, characterized in that, The method is applied to a battery defect detection system based on multi-modal sensors. The system includes an image acquisition module, a tactile sensor module, a processor, and a motion control platform. The method includes: Using the image acquisition module to acquire an image of the battery to be detected on the motion control platform to obtain a to-be-detected image; Using the processor to analyze the to-be-detected image according to a preset image processing algorithm, determine a target area suspected of having a defect in the image, and extract visual features corresponding to the target area; Scanning the target area with the tactile sensor, and using the processor to extract tactile features according to the scanning result of the target area; Using the processor to obtain a visual preliminary defect probability corresponding to the visual features and a tactile preliminary defect probability corresponding to the tactile features by using a random forest algorithm; Using the processor to obtain a target defect probability under a target probability assignment according to the visual preliminary defect probability and the tactile preliminary defect probability based on the D-S theory, so as to determine whether there is a target defect in the target area according to the target defect probability.

2. The battery defect detection method based on a multi-modal sensor according to claim 1, wherein The step of using the processor to obtain a target defect probability under a target probability assignment according to the visual preliminary defect probability and the tactile preliminary defect probability based on the D-S theory, so as to determine whether there is a target defect in the target area includes: Converting the visual preliminary defect probability and the tactile preliminary defect probability into a visual defect basic probability assignment and a tactile defect basic probability assignment of the D-S theory, and calculating a conflict coefficient between the image acquisition module and the tactile sensor; If the conflict coefficient is greater than or equal to a preset coefficient threshold, correcting the visual defect basic probability assignment and the tactile defect basic probability assignment by using an image acquisition module weight coefficient and a tactile sensor weight coefficient, and obtaining a target defect probability under a target probability assignment according to the corrected visual defect basic probability assignment and the corrected tactile defect basic probability assignment; If the conflict coefficient is less than the preset coefficient threshold, obtaining a target defect probability under a target probability assignment according to the visual defect basic probability assignment and the tactile defect basic probability assignment; If the target defect probability is greater than or equal to a preset defect threshold corresponding to the target defect, determining that there is a target defect in the target area; If the target defect probability is less than the preset defect threshold corresponding to the target defect, determining that there is no target defect in the target area.

3. The battery defect detection method based on a multi-modal sensor according to claim 2, characterized in that, The step of obtaining a target defect probability under a target probability assignment according to the corrected visual defect basic probability assignment and the corrected tactile defect basic probability assignment, or the step of obtaining a target defect probability under a target probability assignment according to the visual defect basic probability assignment and the tactile defect basic probability assignment includes: Determining a target defect probability according to the corrected visual defect basic probability assignment and the corrected tactile defect basic probability assignment, or according to the visual defect basic probability assignment and the tactile defect basic probability assignment, by using a preset target defect probability formula. Wherein, the target defect probability formula is: , is the target defect probability in the target probability assignment, is the probability of a defect existing in the basic probability assignment of the visual defect, is the probability of a defect existing in the basic probability of the tactile defect, and K is the conflict coefficient.

4. The battery defect detection method based on a multi-modal sensor according to claim 2, characterized in that, The correction of the basic probability assignment of visual defects and the basic probability assignment of tactile defects by the weight coefficient of the image acquisition module and the weight coefficient of the tactile sensor includes: Modify the basic probability assignment of the visual defect by the weight coefficient of the image acquisition module to obtain the modified basic probability assignment of the visual defect , where is the probability of a defect existing in the basic probability assignment of the visual defect, is the weight coefficient of the image acquisition module; Modify the basic probability assignment of the tactile defect through the weight coefficient of the tactile sensor to obtain the modified basic probability assignment of the tactile defect , where is the probability of a defect existing in the basic probability of the tactile defect, is the weight coefficient of the tactile sensor.

5. The battery defect detection method based on multimodal sensors according to claim 1, characterized in that, The motion control platform includes a robotic arm and a placement table. The placement table is used to place the battery to be detected, and the robotic arm is used to move the battery to be detected; The image acquisition of the battery to be detected on the motion control platform by the image acquisition module to obtain a to-be-detected image includes: The image acquisition module acquires the battery to be detected on the placement table to obtain a to-be-detected image of one detection surface of the battery to be detected; The robotic arm moves the battery to be detected during the image acquisition process to ensure that the to-be-detected images of all detection surfaces of the battery to be detected are acquired.

6. The battery defect detection method based on a multi-modal sensor according to claim 5, wherein, A planar coordinate system is set on the placement table. The scanning of the target area by the tactile sensor includes: Determine the coordinate position of the target area in the planar coordinate system; Determine the contact path of the tactile sensor when scanning the target area according to the coordinate position; The tactile sensor scans the target area according to the contact path.

7. A battery defect detection system based on multi-modal sensors, characterized in that, The system is used to implement the battery defect detection method based on multi-modal sensors described in any one of claims 1-6. The system includes: an image acquisition module, a tactile sensor module, a processor, and a motion control platform; The image acquisition module and the tactile sensor module are arranged above the motion control platform and are electrically connected to the processor respectively. The motion control platform is used to place the battery to be detected.

8. The battery defect detection system based on a multi-modal sensor according to claim 7, characterized in that The motion control platform includes a robotic arm and a placement table. The placement table is used to place the battery to be detected, and the robotic arm is used to move the battery to be detected.

9. The battery defect detection system based on multi-modal sensors according to claim 8, wherein, The robotic arm is used to move the battery to be detected during the image acquisition process to ensure that the images of all detection surfaces of the battery to be detected are acquired. Also, after the processor analyzes the image of the battery to be detected through an image processing algorithm and determines the target area, the robotic arm moves the battery to be detected to align the detection surface corresponding to the target area with the tactile sensor.

10. The battery defect detection system based on a multi-modal sensor according to claim 7, wherein A planar coordinate system is set on the placement table.

Images