A method and system for visually detecting surface defects of a power battery integrated busbar

Abstract

This application provides a visual inspection method and system for surface defects of integrated busbars in power batteries. The method includes: acquiring multimodal images and a view transformation matrix of the integrated busbar; wherein the multimodal images include structured light depth images, visible light images, and infrared thermal images from multiple viewpoints; generating geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps using a cross-modal fusion neural network based on the structured light depth images, visible light images, infrared thermal images, and view transformation matrix; identifying defect types based on the geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps, obtaining corresponding diagnostic results and confidence levels for each; and determining the final detection result based on the diagnostic results and confidence levels through comprehensive reasoning using a fuzzy logic system. This approach can improve the accuracy of defect detection in integrated busbars of power batteries.

Description

Technical Field

[0001] This application relates to the field of power battery technology, specifically to a visual inspection method and system for surface defects of integrated busbars in power batteries. Background Technology

[0002] With the rapid development of new energy vehicles and the energy storage industry, extremely high requirements have been placed on the energy density, safety, reliability, and service life of power battery packs. The integrated busbar (CCS), as the core electrical connection and signal acquisition component of the battery pack, directly determines the overall performance and safety of the battery system. The CCS integrates various heterogeneous components such as FFC / FPC (flexible printed circuit board), copper-aluminum busbars, and plastic structural parts. Its manufacturing process is complex, and any minor surface or internal defects, such as connector soldering defects, insulation damage, conductor deformation, or foreign matter contamination, can potentially lead to serious safety accidents such as localized overheating, short circuits, or even thermal runaway during use.

[0003] Currently, in the production and manufacturing process of CCS, quality inspection mainly relies on traditional manual visual inspection, but this method suffers from problems such as strong subjectivity, low efficiency, and high rate of missed detection. Summary of the Invention

[0004] This application aims to provide a visual inspection method and system for surface defects of integrated busbars in power batteries, which can improve the accuracy of defect detection in integrated busbars of power batteries.

[0005] The technical solution of this application is implemented as follows: In a first aspect, embodiments of this application provide a visual inspection method for surface defects of a power battery integrated busbar, the method comprising: Acquire multimodal images and view transformation matrices of the integrated busbar of the power battery; wherein the multimodal images include structured light depth images from multiple viewpoints, visible light images from multiple viewpoints, and infrared thermal images from multiple viewpoints; Based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix, a geometric point cloud map, a visible light point cloud map, and an infrared point cloud map are generated through a cross-modal fusion neural network. Based on the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, defect type identification is performed to obtain the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, respectively. Based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, a fuzzy logic system is used for comprehensive reasoning to determine the final detection result.

[0006] In the above scheme, obtaining the multimodal image and view transformation matrix of the integrated power battery busbar includes: Determine a preset acquisition angle; wherein, the preset acquisition angle includes multiple acquisition angles; Based on the multiple acquisition angles, the multimodal images of the power battery integrated busbar are acquired by the multimodal acquisition module; The first coordinate system of the multimodal acquisition module and the second coordinate system of the power battery integrated busbar are determined; and the view transformation matrix is ​​determined based on the first coordinate system and the second coordinate system.

[0007] In the above scheme, the cross-modal fusion neural network includes a multi-branch feature extraction encoder, a cost volume construction and view fusion module, and a point cloud generation decoder; The process of generating geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps through a cross-modal fusion neural network based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix includes: The multi-branch feature extraction encoder extracts and fuses features from the structured light depth image, the visible light image, and the infrared thermal image to obtain multi-view fused cross-modal features. The cost volume construction and view fusion module performs 3D projection on the fused cross-modal features of the multi-viewpoint based on the view transformation matrix to obtain a three-dimensional feature map; and performs mesh division and normalization processing on the three-dimensional feature map through a 3D cost volume mesh and a 3D regularization network to obtain a standardized mesh feature image. The point cloud generation decoder generates point clouds from the standardized grid feature image to obtain the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map.

[0008] In the above scheme, the point cloud generation decoder includes a 3D to 2D projection layer, a coordinate regression branch, an attribute regression branch, and a point cloud attribute assignment layer. The step of generating point clouds from the standardized grid feature image using the point cloud generation decoder to obtain the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map includes: The standardized grid feature image is filtered through the 3D to 2D projection layer to obtain a filtered standardized grid feature image. The geometric point cloud map is obtained by extracting the surface value from the filtered standardized grid feature image through the coordinate regression branch. Through the attribute regression branch, color regression and temperature regression are performed on the geometric point cloud map to obtain color and temperature values. Based on the color value and the temperature value, the geometric point cloud map is interpolated through the point cloud attribute assignment layer to obtain the visible light point cloud map and the infrared point cloud map.

[0009] In the above scheme, the step of identifying defect types based on the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map to obtain the diagnostic results and confidence levels corresponding to each of the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map includes: By using a geometric deformation sensing network, the geometric point cloud map is used to identify the defect type, and the geometric defect diagnosis result and first confidence level are obtained. By using a surface appearance perception network, feature extraction, color and spatial weighting processing, and defect classification and identification are performed on the visible light point cloud image to obtain appearance defect diagnosis results and a second confidence level. By using a thermophysical property sensing network, feature extraction, temporal feature fusion, and defect classification and identification are performed on the infrared point cloud image to obtain potential defect diagnosis results and a third confidence level.

[0010] In the above scheme, the geometric deformation sensing network includes a point cloud voxelization layer, a 3D convolutional coding layer, a geometric feature extraction layer, and a decoding layer; The step of identifying the defect type of the geometric point cloud image through a geometric deformation sensing network to obtain a geometric defect diagnosis result and a first confidence level includes: The point cloud voxelization layer transforms the geometric point cloud map into a three-dimensional voxel mesh map. The 3D convolutional coding layer performs three convolutions on the three-dimensional voxel mesh to obtain convolutional features; The geometric feature extraction layer calculates channel attention weights and performs weighted processing on the convolutional features based on the channel attention weights to obtain enhanced features; multi-scale geometric feature extraction is then performed on the enhanced features to obtain geometric features. The geometric features are identified through the decoding layer to obtain the geometric defect diagnosis result and the first confidence level corresponding to the geometric defect diagnosis result; wherein, the geometric defect diagnosis result includes the type and location of the geometric defect.

[0011] In the above scheme, the step of determining the final detection result by performing comprehensive reasoning through a fuzzy logic system based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, respectively, includes: Determine the material information of the power battery integrated busbar, as well as the environmental information of the power battery integrated busbar; Based on the material information and the environmental information, the confidence levels corresponding to the geometric point cloud map, the visible light point cloud map and the infrared point cloud map are dynamically corrected to obtain the final confidence level. The fuzzy logic system is used to comprehensively reason about the diagnostic results corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, as well as the final confidence level, to obtain the final detection result. The final detection result includes one of the following: qualified, scrapped, reworked, or requesting manual re-inspection.

[0012] Secondly, embodiments of this application provide a visual inspection system for surface defects of integrated busbars in power batteries. The system includes: an acquisition module, a generation module, an identification module, and a determination module. The acquisition module is used to acquire multimodal images and view transformation matrices of the integrated busbar of the power battery; wherein, the multimodal images include structured light depth images from multiple viewpoints, visible light images from multiple viewpoints, and infrared thermal images from multiple viewpoints. The generation module is used to generate geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix through a cross-modal fusion neural network. The identification module is used to identify the defect type based on the geometric point cloud map, the visible light point cloud map and the infrared point cloud map, and obtain the diagnosis results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map and the infrared point cloud map respectively; The determining module is used to determine the final detection result by performing comprehensive reasoning through a fuzzy logic system based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, respectively.

[0013] Thirdly, embodiments of this application provide a visual inspection device for surface defects of integrated busbars in power batteries, comprising: a processor and a memory; wherein, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in the first aspect.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing executable instructions for causing a processor to perform the method described in the first aspect.

[0015] This application provides a visual inspection method and system for surface defects of a power battery integrated busbar. The method includes: acquiring multimodal images and a view transformation matrix of the power battery integrated busbar; wherein the multimodal images include structured light depth images, visible light images, and infrared thermal images from multiple viewpoints; generating geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps through a cross-modal fusion neural network based on the structured light depth images, visible light images, infrared thermal images, and the view transformation matrix; performing defect type identification based on the geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps to obtain the diagnostic results and confidence levels corresponding to each of the geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps; and determining the final detection result by performing comprehensive reasoning through a fuzzy logic system based on the diagnostic results and confidence levels corresponding to each of the geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps. The aforementioned solution, by integrating visible light, three-dimensional geometry, and infrared thermophysical information, can not only identify surface defects such as scratches and stains, but also accurately detect geometric anomalies such as deformation and coplanarity. Furthermore, it can diagnose potential electrical faults that traditional visual methods cannot detect, such as poor soldering and internal cracks. Simultaneously, by comprehensively reasoning about the power battery integrated busbar using the diagnostic results and confidence levels of each of the visible light, three-dimensional geometry, and infrared thermophysical information, the final detection result is determined, thus improving the accuracy of defect detection for the power battery integrated busbar. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0017] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0018] Figure 1 This is an optional flowchart illustrating a visual inspection method for surface defects of an integrated busbar in a power battery, provided in an embodiment of this application. Figure 2 A schematic diagram of a visual inspection system for surface defects of an integrated busbar in a power battery, provided in an embodiment of this application; Figure 3This is a schematic diagram of the structure of a visual inspection device for surface defects of an integrated busbar in a power battery, provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.

[0020] Unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this application is for the purpose of describing embodiments of this application only and is not intended to be limiting of this application.

[0021] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0022] This application provides a visual inspection method for surface defects on the integrated busbar of a power battery. Figure 1 This is an optional flowchart illustrating a visual inspection method for surface defects of an integrated busbar in a power battery, provided in an embodiment of this application. Figure 1 The steps shown are explained.

[0023] S101. Obtain multimodal images and view transformation matrices of the integrated busbar of the power battery; wherein, the multimodal images include structured light depth images from multiple viewpoints, visible light images from multiple viewpoints, and infrared thermal images from multiple viewpoints.

[0024] In some embodiments of this application, the multimodal image includes structured light depth images from multiple perspectives, visible light images from multiple perspectives, and infrared thermal images from multiple perspectives.

[0025] In some embodiments of this application, a visual inspection method for surface defects of integrated busbars in power batteries is adapted to power battery defect inspection scenarios.

[0026] In some embodiments of this application, a visual inspection method for surface defects of integrated busbars of power batteries is adapted to a visual inspection system for surface defects of integrated busbars of power batteries.

[0027] In some embodiments of this application, a preset acquisition angle is determined; wherein, the preset acquisition angle includes multiple acquisition angles; based on the multiple acquisition angles, a multimodal image of the power battery integrated busbar is acquired by a multimodal acquisition module; a first coordinate system of the multimodal acquisition module and a second coordinate system of the power battery integrated busbar are determined; and a viewpoint transformation matrix is ​​determined based on the first coordinate system and the second coordinate system.

[0028] S102. Based on structured light depth images, visible light images, infrared thermal images, and viewpoint transformation matrices, geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps are generated through a cross-modal fusion neural network.

[0029] In some embodiments of this application, the cross-modal fusion neural network includes a multi-branch feature extraction encoder, a cost volume construction and view fusion module, and a point cloud generation decoder.

[0030] In some embodiments of this application, a multi-branch feature extraction encoder is used to extract and fuse features from structured light depth images, visible light images, and infrared thermal images to obtain multi-view fused cross-modal features. Through the cost volume construction and view fusion module, the multi-view fusion cross-modal features are 3D projected based on the view transformation matrix to obtain a three-dimensional feature map; and the three-dimensional feature map is meshed and normalized through a 3D cost volume mesh and a 3D regularization network to obtain a standardized mesh feature image; and point cloud generation decoder is used to generate point clouds from the standardized mesh feature image to obtain geometric point cloud map, visible light point cloud map and infrared point cloud map.

[0031] S103. Based on geometric point cloud map, visible light point cloud map and infrared point cloud map, perform defect type identification and obtain the corresponding diagnostic results and confidence levels of geometric point cloud map, visible light point cloud map and infrared point cloud map respectively.

[0032] In some embodiments of this application, a geometric deformation sensing network is used to identify the defect type of a geometric point cloud image to obtain a geometric defect diagnosis result and a first confidence level; a surface appearance sensing network is used to extract features, perform color and spatial weighting processing, and classify defects in a visible light point cloud image to obtain an appearance defect diagnosis result and a second confidence level; a thermophysical property sensing network is used to extract features, fuse temporal features, and classify defects in an infrared point cloud image to obtain a potential defect diagnosis result and a third confidence level.

[0033] S104. Based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, visible light point cloud map, and infrared point cloud map, the final detection result is determined by comprehensive reasoning through a fuzzy logic system.

[0034] In some embodiments of this application, the material information of the integrated busbar of the power battery and the environmental information of the integrated busbar of the power battery are determined; based on the material information and environmental information, the confidence levels corresponding to the geometric point cloud map, visible light point cloud map and infrared point cloud map are dynamically corrected to obtain the final confidence level; through a fuzzy logic system, the diagnostic results and the final confidence levels corresponding to the geometric point cloud map, visible light point cloud map and infrared point cloud map are comprehensively reasoned to obtain the final detection result; wherein, the final detection result includes one of qualified, scrapped, reworked and requested for manual re-inspection.

[0035] For example, confidence-weighted approach: Each perception network assigns an initial confidence score to its output defect diagnosis result. The system then dynamically adjusts this score based on contextual information such as environmental conditions and component materials to obtain the final confidence score. Fuzzy logic decision-making: The diagnosis results and their final confidence scores from the three networks are input into a fuzzy logic system. This system performs comprehensive reasoning based on a pre-defined expert rule base (e.g., threshold-based judgment) and outputs a clear final decision, such as "qualified," "scrapped," "reworked," or "request manual re-inspection."

[0036] 1. Scrap-level defect - high-risk cold solder joint: Scenario: When the thermophysical property sensing network reports “cold solder joint” or “imperfect via” with high confidence (e.g., >0.85), and the geometric deformation sensing network reports “coplanarity anomaly” or “deformation” at the same location with medium to high confidence (e.g., >0.7).

[0037] Verdict: Scrapped.

[0038] Reason: The simultaneous occurrence of electrical connection defects and mechanical structural abnormalities indicates a serious and irreversible manufacturing defect, making it impossible to guarantee product reliability.

[0039] 2. Scrap-grade defect - severe physical damage: Scenario: When the surface appearance perception network reports serious defects such as "insulation layer damage" or "copper leakage" and the confidence level is high (e.g., >0.8).

[0040] Verdict: Scrapped.

[0041] Reason: This defect directly leads to electrical insulation failure, posing a short circuit risk and a significant safety hazard.

[0042] 3. Refurbishment-grade defects - potential thermal risks: Scenario: When the thermophysical property sensing network reports potential defects such as "internal cracks" with a medium to high confidence level (e.g., >0.7), but the surface appearance sensing network does not detect any obvious anomalies at the corresponding locations (low confidence level).

[0043] Verdict: Repair required.

[0044] Reason: There is a potential risk of failure, but the defect is not visible on the surface and can be repaired through rework (such as re-crimping).

[0045] Understandably, this application, by integrating visible light, three-dimensional geometry, and infrared thermophysical information, can not only identify surface scratches and stains, but also accurately detect geometric anomalies such as deformation and coplanarity. Furthermore, it can diagnose potential electrical faults that traditional visual methods cannot detect, such as poor soldering and internal cracks. Simultaneously, by comprehensively reasoning about the power battery integrated busbar using the diagnostic results and confidence levels of each of the visible light, three-dimensional geometry, and infrared thermophysical information, the final detection result is determined, thus improving the accuracy of defect detection for the power battery integrated busbar.

[0046] In some embodiments of this application, S101 can be implemented by S201-S203, as follows: S201. Determine the preset acquisition angle; wherein, the preset acquisition angle includes multiple acquisition angles.

[0047] S202. Based on multiple acquisition angles, multimodal images of the power battery integrated busbar are acquired through a multimodal acquisition module.

[0048] S203. Determine the first coordinate system of the multimodal acquisition module and the second coordinate system of the power battery integrated busbar; and determine the view transformation matrix based on the first and second coordinate systems.

[0049] For example, a structured light projector, a high-resolution visible light camera, and an infrared thermal imager are rigidly fixed on a platform to form a multimodal acquisition module. Before testing, the multimodal acquisition module is first calibrated with high precision. This process not only obtains the intrinsic parameters of each sensor, but more importantly, it also obtains their relative positions and orientations (i.e., extrinsic parameters), thereby establishing a unified module coordinate system. The multimodal acquisition module is mounted on a high-precision rotary displacement stage. The transformation matrix of the module coordinate system relative to the world coordinate system (based on the CCS component) is accurately measured and recorded at each preset acquisition angle (e.g., 0°, 60°, 120°). These transformation matrices are the "view transformation matrix" for the acquired data at each viewpoint. The displacement stage is controlled to rotate to each preset angle, and acquisition is synchronously triggered at each angle. a) A set of structured light sequence images, and calculate the structured light depth image from that viewpoint.

[0050] b. A high-resolution visible light image.

[0051] c. An infrared thermal image.

[0052] In some embodiments of this application, S102 can be implemented by S301-S303, as follows: S301. By using a multi-branch feature extraction encoder, feature extraction and feature fusion are performed on structured light depth images, visible light images, and infrared thermal images to obtain multi-view fused cross-modal features.

[0053] For example, a multi-branch feature extraction encoder sequentially includes an input layer, a shared feature extraction layer, and a feature fusion layer; wherein, the input layer includes a visible light image branch, an infrared image branch, and a depth map branch, the three branches being used to receive different images, as detailed below: Visible light image branch: Receive RGB images (224×224×3) to obtain visible light images; Infrared image branch: Receives thermal image (224×224×1), i.e., infrared thermal image; Depth map branch: Receives a depth map (224×224×1), i.e., a structured light depth image.

[0054] The shared feature extraction layer consists of multiple independent branches, as detailed below: Conv2D(32, 3×3) → ReLU → BatchNorm; Conv2D(32, 3×3) → ReLU → BatchNorm → MaxPool(2×2); Conv2D(64, 3×3) → ReLU → BatchNorm; Conv2D(64, 3×3) → ReLU → BatchNorm → MaxPool(2×2); Conv2D(128, 3×3) → ReLU → BatchNorm; Conv2D(128, 3×3) → ReLU → BatchNorm → MaxPool(2×2), Feature extraction is performed using the shared feature extraction layers corresponding to the structured light depth image, visible light image, and infrared thermal image, to obtain the multi-view features corresponding to each of the structured light depth image, visible light image, and infrared thermal image.

[0055] The feature fusion layer concatenates the 128-dimensional feature maps of the three branches along the channel dimension to obtain a 384-channel feature map; Conv2D(256, 1×1) → ReLU # Dimensionality reduction and fusion of cross-modal features, that is, fusing the multi-view features corresponding to the structured light depth image, visible light image and infrared thermal image to obtain multi-view fused cross-modal features.

[0056] S302. Through the cost volume construction and view fusion module, the multi-view fusion cross-modal features are 3D projected based on the view transformation matrix to obtain a three-dimensional feature map; and the three-dimensional feature map is meshed and normalized through a 3D cost volume mesh and a 3D regularization network to obtain a standardized mesh feature image.

[0057] For example, the cost volume construction and view fusion module sequentially includes a cost volume construction layer and a 3D regularization network. The cost volume construction layer uses a view transformation matrix to project the fused cross-modal features from multiple views onto a unified 3D space to obtain a 3D feature map; it constructs a 3D cost volume mesh (64×64×64×256), and uses the 3D cost volume mesh to divide the 3D feature map to obtain multiple mesh feature images.

[0058] The 3D regularization network consists of four different branches, as follows: 3D Conv(128, 3×3×3) → ReLU → BatchNorm; 3D Conv(64, 3×3×3) → ReLU → BatchNorm; 3D Conv(32, 3×3×3) → ReLU → BatchNorm; 3D Conv(16, 3×3×3) → ReLU → BatchNorm; The 3D regularization network is used to normalize and standardize multiple grid feature images.

[0059] S303. Using a point cloud generation decoder, point cloud is generated from the standardized grid feature image to obtain geometric point cloud map, visible light point cloud map and infrared point cloud map.

[0060] In some embodiments of this application, the point cloud generation decoder includes a 3D to 2D projection layer, a coordinate regression branch, an attribute regression branch, and a point cloud attribute assignment layer.

[0061] In some embodiments of this application, a 3D-to-2D projection layer is used to filter the standardized grid feature image to obtain a filtered standardized grid feature image; a coordinate regression branch is used to extract isosurface values ​​from the filtered standardized grid feature image to obtain a geometric point cloud map; an attribute regression branch is used to perform color regression and temperature regression processing on the geometric point cloud map to obtain color and temperature values; based on the color and temperature values, an interpolation processing layer is used to interpolate the geometric point cloud map to obtain a visible light point cloud map and an infrared point cloud map.

[0062] For example, the point cloud generation decoder sequentially includes a 3D to 2D projection layer, a coordinate regression branch, an attribute regression branch, and a point cloud attribute assignment layer, as detailed below: The 3D to 2D projection layer consists of two branches, as follows: 3D deconvolution (16, 3×3×3) → ReLU → Output size (128×128×128×16); 3D deconvolution (8, 3×3×3) → ReLU → Output size (256×256×256×8); The standardized grid feature image is filtered by a 3D-to-2D projection layer to obtain a filtered standardized grid feature image.

[0063] The coordinate regression branch includes: Conv3D(4, 1×1×1) → Tanh activation # Output XYZ coordinates + confidence; Extract isosurfaces through Marching Cubes algorithm to generate basic point cloud P_geom (i.e., geometric point cloud map).

[0064] The attribute regression branches include: RGB attribute header: Conv3D(3, 1×1×1) → Sigmoid #Regress RGB color; Thermal attribute header: Conv3D(1, 1×1×1) → ReLU #Regress temperature value. Point cloud attribute assignment layer: Interpolates RGB values ​​and temperature values ​​to the corresponding positions in the base point cloud to generate the final point cloud; the final point cloud includes P_rgb: containing XYZ coordinates and RGB color; P_ir: containing XYZ coordinates and temperature value. The visible light point cloud image P_rgb is a dense point cloud, where each point contains not only three-dimensional coordinates (X, Y, Z) but also its RGB color value. The infrared point cloud image P_ir is a dense point cloud that corresponds strictly point-to-point to P_rgb, where each point contains three-dimensional coordinates (X, Y, Z) and its thermal intensity value (Temperature).

[0065] The result is two dense, coordinate-aligned point clouds: P_rgb and P_ir. Furthermore, the original multi-view structured light depth maps are also fused into a more complete, single 3D point cloud, P_geom.

[0066] In some embodiments of this application, S103 can be implemented by S401-S403, as follows: S401. Using a geometric deformation sensing network, defect type identification is performed on the geometric point cloud map to obtain the geometric defect diagnosis result and the first confidence level.

[0067] In some embodiments of this application, the geometric deformation sensing network includes a point cloud voxelization layer, a 3D convolutional coding layer, a geometric feature extraction layer, and a decoding layer.

[0068] In some embodiments of this application, a point cloud voxelization layer is used to convert a geometric point cloud map into a three-dimensional voxel mesh map; a 3D convolutional coding layer is used to perform three convolutions on the three-dimensional voxel mesh map to obtain convolutional features; a geometric feature extraction layer is used to calculate channel attention weights and to perform weighted processing on the convolutional features based on the channel attention weights to obtain enhanced features; multi-scale geometric feature extraction is performed on the enhanced features to obtain geometric features; a decoding layer is used to identify defects in the geometric features to obtain a geometric defect diagnosis result and a first confidence level corresponding to the geometric defect diagnosis result; wherein, the geometric defect diagnosis result includes the type and location of the geometric defect.

[0069] For example, the input to the geometric deformation sensing network is a pure geometric point cloud P_geom; the output of the geometric deformation sensing network is the type, location, and deformation parameters of the geometric defect.

[0070] The task of the geometric deformation sensing network is to employ a voxelized high-resolution network to focus on both macroscopic and microscopic geometric features. Accurate detection is achieved by analyzing the spatial distribution and normal vector field of the point cloud. 1. Warping and deformation of plastic structural components; 2. The copper and aluminum busbars are uneven; 3. Insufficient coplanarity of the connecting terminals.

[0071] The geometric deformation perception network consists of a point cloud voxelization layer, a 3D convolutional coding layer, a geometric feature extraction layer, and a decoding layer, specifically: 1. Point cloud voxelization layer: Converts point clouds into regular 3D voxel meshes, output size: 64×64×64×1; 2. The 3D convolutional coding layer includes: 3D Conv (kernel=3, stride=1, filters=32) + ReLU + BatchNorm; 3D MaxPool (kernel=2, stride=2); 3D Conv (kernel=3, stride=1, filters=64) + ReLU + BatchNorm; 3D MaxPool (kernel=2, stride=2); 3D Conv (kernel=3, stride=1, filters=128) + ReLU + BatchNorm.

[0072] 3. The geometric feature extraction layer includes a 3D attention module and 3D spatial pyramid pooling; The 3D attention module is used to calculate channel attention weights to enhance important features; 3D spatial pyramid pooling extracts geometric features through multi-scale receptive fields.

[0073] 4. The decoding layer includes global average pooling, fully connected layer, ReLU, fully connected layer, and Softmax.

[0074] The decoding layer outputs the type, location, and deformation parameters of the geometric defect. The type and location of the geometric defect are the geometric defect diagnosis results; the deformation parameters are the first confidence level.

[0075] S402. Through the surface appearance perception network, feature extraction, color and spatial weighting processing, and defect classification and identification are performed on the visible light point cloud image to obtain the appearance defect diagnosis result and the second confidence level.

[0076] For example, the input of the surface appearance perception network is a visible light point cloud P_rgb with RGB information; the output of the surface appearance perception network is the type, boundary and confidence level of the appearance defect.

[0077] The task of the surface appearance perception network is to use a neural network based on a point cluster attention mechanism to keenly capture local anomalies in color and texture. It is specifically designed for recognition of: 1. Surface scratches (linear color abnormality); 2. Stains (clumps of abnormal color); 3. Damaged insulation layer / exposed copper (material discoloration).

[0078] The surface appearance perception network consists of a feature extraction backbone network, a color-space attention layer, and a defect classification layer, specifically: 1. Feature Extraction Backbone Network: Multi-scale grouping (MSG) is used to extract local features, specifically: Level 1: Sample 512 center points, radius 0.1, MLP[16,16,32]; Level 2: Sample 128 center points, radius 0.2, MLP[32,32,64]; Level 3: Sample 32 center points, radius 0.4, MLP[64,64,128].

[0079] 2. Color-Spatial Attention Layer: This layer consists of a dual-branch attention module, a channel attention module, and a spatial attention module.

[0080] Bi-branch attention: focusing on color anomalies and spatial distribution anomalies respectively; Channel attention: Weighted importance of color channel features; Spatial attention: Focusing on areas suspected of having defects.

[0081] 3. The defect classification layer includes a max pooling layer, a fully connected layer, ReLU, a fully connected layer (M), and Sigmoid activation.

[0082] The defect classification layer outputs the type, boundary, and confidence level of the appearance defect. The type and boundary of the appearance defect are the appearance defect diagnosis results; the confidence level is the second confidence level.

[0083] S403. Through a thermophysical property sensing network, feature extraction, temporal feature fusion, and defect classification and identification are performed on the infrared point cloud image to obtain potential defect diagnosis results and third confidence level.

[0084] For example, the input to the thermophysical property sensing network is an infrared point cloud P_ir with temperature information; the output of the thermophysical property sensing network is the type, location, and risk level of the potential defect.

[0085] The task of the thermophysical property sensing network is to treat the point cloud as a graph structure through a spatiotemporal graph convolutional network, and simultaneously analyze the spatial distribution and temporal evolution of temperature (from multi-angle thermal sequences). Based on this, it diagnoses potential faults caused by poor connectivity. 1. Cold solder joint (stable localized hot spot); 2. Incomplete through-hole sealing (abnormal thermal diffusion pattern); 3. Internal cracks (unique thermal response phase characteristics).

[0086] The thermophysical property sensing network includes a feature extraction layer, a temporal feature fusion layer, and a classification layer.

[0087] 1. The feature extraction layer consists of three consecutive sets of concatenated modules. These three sets of modules each include a convolutional layer, an activation layer, another convolutional layer, an activation layer, and a pooling layer, as follows: Conv2D(64, 3×3) → ReLU → Conv2D(64, 3×3) → ReLU → MaxPool(2×2); Conv2D(128, 3×3) → ReLU → Conv2D(128, 3×3) → ReLU → MaxPool(2×2); Conv2D(256, 3×3) → ReLU → Conv2D(256, 3×3) → ReLU → MaxPool(2×2).

[0088] 2. The temporal feature fusion module includes global average pooling, temporal concatenation, 1D temporal convolution, and global max pooling, specifically: Global average pooling: For example, pooling the 256×28×28 feature map at each time step into a 256-dimensional vector; Time series concatenation: For example, concatenating 256-dimensional vectors of T time steps into a T×256 matrix; 1D temporal convolution: For example, Conv1D(128, kernel=3) captures temperature change patterns over time; Global max pooling (temporal dimension): for example, obtaining a 128-dimensional temporal feature vector; 3. Classification layer: This includes, in order, a fully connected layer, ReLU, another fully connected layer, and a Sigmoid layer.

[0089] The classification layer outputs the type, location, and risk level of potential defects. The type and location of potential defects are the diagnostic results of potential defects; the risk level is the third confidence level.

[0090] It should be noted that defect classification and identification are performed on geometric point cloud images, visible light point cloud images, and infrared point cloud images respectively, yielding corresponding diagnostic results and confidence levels for each. Geometric point cloud images, visible light point cloud images, and infrared point cloud images correspond to different sensing networks; specifically, geometric point cloud images correspond to a geometric deformation sensing network; visible light point cloud images correspond to a surface appearance sensing network; and infrared point cloud images correspond to a thermophysical property sensing network.

[0091] Understandably, this application represents a leap from "isolated detection" to "holographic perception," significantly enhancing detection capabilities. By innovatively fusing visible light, three-dimensional geometry, and infrared thermophysical information, the system can not only identify surface defects such as scratches and stains, but also accurately detect geometric anomalies such as deformation and coplanarity. Furthermore, it can diagnose potential electrical faults that traditional vision cannot detect, such as poor soldering and internal cracks. This multimodal fusion method overcomes the limitations of single-sensor detection, achieving comprehensive and thorough diagnosis of CCS component quality.

[0092] Based on the above embodiments of the method for visual inspection of surface defects on integrated busbars of power batteries, this application also provides a visual inspection system for surface defects on integrated busbars of power batteries, such as... Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of a visual inspection system for surface defects of an integrated busbar in a power battery, provided in an embodiment of this application. The system includes: an acquisition module 201, a generation module 202, an identification module 203, and a determination module 204. The acquisition module 201 is used to acquire multimodal images and view transformation matrices of the integrated busbar of the power battery; wherein, the multimodal images include structured light depth images from multiple viewpoints, visible light images from multiple viewpoints, and infrared thermal images from multiple viewpoints. The generation module 202 is used to generate geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix through a cross-modal fusion neural network. The identification module 203 is used to identify the defect type based on the geometric point cloud map, the visible light point cloud map and the infrared point cloud map, and obtain the diagnosis results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map and the infrared point cloud map respectively; The determining module 204 is used to determine the final detection result by performing comprehensive reasoning through a fuzzy logic system based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, respectively.

[0093] In some embodiments of this application, the determining module 204 is further configured to determine a preset acquisition angle; wherein the preset acquisition angle includes multiple acquisition angles; based on the multiple acquisition angles, the multimodal image of the power battery integrated busbar is acquired by the multimodal acquisition module; a first coordinate system of the multimodal acquisition module and a second coordinate system of the power battery integrated busbar are determined; and the viewpoint transformation matrix is ​​determined based on the first coordinate system and the second coordinate system.

[0094] In some embodiments of this application, the cross-modal fusion neural network includes a multi-branch feature extraction encoder, a cost volume construction and view fusion module, and a point cloud generation decoder; The generation module 202 is further configured to perform feature extraction and feature fusion on the structured light depth image, the visible light image, and the infrared thermal image through the multi-branch feature extraction encoder to obtain multi-view fused cross-modal features; through the cost volume construction and view fusion module, perform 3D projection on the multi-view fused cross-modal features based on the view transformation matrix to obtain a three-dimensional feature map; and perform mesh division and normalization processing on the three-dimensional feature map through a 3D cost volume mesh and a 3D regularization network to obtain a standardized mesh feature image; and perform point cloud generation on the standardized mesh feature image through the point cloud generation decoder to obtain the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map.

[0095] In some embodiments of this application, the point cloud generation decoder includes a 3D to 2D projection layer, a coordinate regression branch, an attribute regression branch, and a point cloud attribute assignment layer. The generation module 202 is further configured to: filter the standardized grid feature image through the 3D-to-2D projection layer to obtain a filtered standardized grid feature image; extract isosurface values ​​from the filtered standardized grid feature image through the coordinate regression branch to obtain the geometric point cloud map; perform color regression and temperature regression processing on the geometric point cloud map through the attribute regression branch to obtain color and temperature values; and perform interpolation processing on the geometric point cloud map based on the color and temperature values ​​through the point cloud attribute assignment layer to obtain the visible light point cloud map and the infrared point cloud map.

[0096] In some embodiments of this application, the identification module 203 is further configured to: identify the defect type of the geometric point cloud image through a geometric deformation perception network to obtain a geometric defect diagnosis result and a first confidence level; extract features, perform color and spatial weighting processing, and classify defects in the visible light point cloud image through a surface appearance perception network to obtain an appearance defect diagnosis result and a second confidence level; and extract features, fuse temporal features, and classify defects in the infrared point cloud image through a thermophysical property perception network to obtain a potential defect diagnosis result and a third confidence level.

[0097] In some embodiments of this application, the geometric deformation sensing network includes a point cloud voxelization layer, a 3D convolutional coding layer, a geometric feature extraction layer, and a decoding layer; The recognition module 203 is further configured to: convert the geometric point cloud map into a three-dimensional voxel mesh map through the point cloud voxelization layer; perform three convolutions on the three-dimensional voxel mesh map through the 3D convolutional coding layer to obtain convolutional features; calculate channel attention weights through the geometric feature extraction layer and perform weighted processing on the convolutional features based on the channel attention weights to obtain enhanced features; perform multi-scale geometric feature extraction on the enhanced features to obtain geometric features; and perform defect identification on the geometric features through the decoding layer to obtain the geometric defect diagnosis result and the first confidence level corresponding to the geometric defect diagnosis result; wherein, the geometric defect diagnosis result includes the type and location of the geometric defect.

[0098] In some embodiments of this application, the determining module 204 is further configured to determine the material information of the power battery integrated busbar and the environmental information of the power battery integrated busbar; based on the material information and the environmental information, dynamically correct the confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map respectively to obtain the final confidence level; through the fuzzy logic system, comprehensively reason about the diagnostic results corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map and the final confidence level to obtain the final detection result; wherein, the final detection result includes one of qualified, scrapped, reworked, and requesting manual re-inspection.

[0099] Based on the above embodiments of the method for visual inspection of surface defects on integrated busbars of power batteries, this application also provides a visual inspection device for surface defects on integrated busbars of power batteries, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of a visual inspection device for surface defects of an integrated busbar in a power battery, provided in an embodiment of this application. The device 3 includes a processor 301 and a memory 302. The memory 302 stores a computer program; the processor 301 retrieves and runs the computer program from the memory to execute a visual inspection method for surface defects of an integrated busbar in a power battery as described in the above embodiment.

[0100] In the embodiments of this application, the processor 301 described above can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above processor function can also be other types, and the embodiments of this application do not specifically limit it.

[0101] This application provides a computer-readable storage medium storing a computer program for implementing, when executed by a processor, a visual inspection method for surface defects of a power battery integrated busbar as described in any of the above embodiments.

[0102] For example, the program instructions corresponding to the visual inspection method for surface defects of integrated power battery busbars in this embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the program instructions corresponding to the visual inspection method for surface defects of integrated power battery busbars in the storage media are read or executed by an electronic device, the visual inspection method for surface defects of integrated power battery busbars as described in any of the above embodiments can be realized.

[0103] Furthermore, in the embodiments of this application, the functional modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.

[0104] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the embodiments in this application are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, these will not be repeated here.

[0106] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0107] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.

[0108] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0109] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0110] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0111] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0112] The above description is merely an embodiment of this application, but the protection scope of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A visual inspection method for surface defects of integrated busbars in power batteries, characterized in that, The method includes: Acquire multimodal images and view transformation matrices of the integrated busbar of the power battery; wherein the multimodal images include structured light depth images from multiple viewpoints, visible light images from multiple viewpoints, and infrared thermal images from multiple viewpoints; Based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix, a geometric point cloud map, a visible light point cloud map, and an infrared point cloud map are generated through a cross-modal fusion neural network. Based on the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, defect type identification is performed to obtain the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, respectively. Based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, a fuzzy logic system is used for comprehensive reasoning to determine the final detection result.

2. The method according to claim 1, characterized in that, The acquisition of the multimodal image and view transformation matrix of the integrated busbar of the power battery includes: Determine a preset acquisition angle; wherein, the preset acquisition angle includes multiple acquisition angles; Based on the multiple acquisition angles, the multimodal images of the power battery integrated busbar are acquired by the multimodal acquisition module; The first coordinate system of the multimodal acquisition module and the second coordinate system of the power battery integrated busbar are determined; and the view transformation matrix is ​​determined based on the first coordinate system and the second coordinate system.

3. The method according to claim 1, characterized in that, The cross-modal fusion neural network includes a multi-branch feature extraction encoder, a cost volume construction and view fusion module, and a point cloud generation decoder; The process of generating geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps through a cross-modal fusion neural network based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix includes: The multi-branch feature extraction encoder extracts and fuses features from the structured light depth image, the visible light image, and the infrared thermal image to obtain multi-view fused cross-modal features. The cost volume construction and view fusion module performs 3D projection on the fused cross-modal features of the multi-viewpoint based on the view transformation matrix to obtain a three-dimensional feature map; and performs mesh division and normalization processing on the three-dimensional feature map through a 3D cost volume mesh and a 3D regularization network to obtain a standardized mesh feature image. The point cloud generation decoder generates point clouds from the standardized grid feature image to obtain the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map.

4. The method according to claim 3, characterized in that, The point cloud generation decoder includes a 3D to 2D projection layer, a coordinate regression branch, an attribute regression branch, and a point cloud attribute assignment layer. The step of generating point clouds from the standardized grid feature image using the point cloud generation decoder to obtain the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map includes: The standardized grid feature image is filtered through the 3D to 2D projection layer to obtain a filtered standardized grid feature image. The geometric point cloud map is obtained by extracting the surface value from the filtered standardized grid feature image through the coordinate regression branch. Through the attribute regression branch, color regression and temperature regression are performed on the geometric point cloud map to obtain color and temperature values. Based on the color value and the temperature value, the geometric point cloud map is interpolated through the point cloud attribute assignment layer to obtain the visible light point cloud map and the infrared point cloud map.

5. The method according to claim 1, characterized in that, The defect type identification is performed based on the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map to obtain the diagnostic results and confidence levels corresponding to each of the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, including: By using a geometric deformation sensing network, the geometric point cloud map is used to identify the defect type, and the geometric defect diagnosis result and first confidence level are obtained. By using a surface appearance perception network, feature extraction, color and spatial weighting processing, and defect classification and identification are performed on the visible light point cloud image to obtain appearance defect diagnosis results and a second confidence level. By using a thermophysical property sensing network, feature extraction, temporal feature fusion, and defect classification and identification are performed on the infrared point cloud image to obtain potential defect diagnosis results and a third confidence level.

6. The method according to claim 5, characterized in that, The geometric deformation sensing network includes a point cloud voxelization layer, a 3D convolutional coding layer, a geometric feature extraction layer, and a decoding layer. The step of identifying the defect type of the geometric point cloud image through a geometric deformation sensing network to obtain a geometric defect diagnosis result and a first confidence level includes: The point cloud voxelization layer transforms the geometric point cloud map into a three-dimensional voxel mesh map. The 3D convolutional coding layer performs three convolutions on the three-dimensional voxel mesh to obtain convolutional features; The geometric feature extraction layer calculates channel attention weights and performs weighted processing on the convolutional features based on the channel attention weights to obtain enhanced features; multi-scale geometric feature extraction is then performed on the enhanced features to obtain geometric features. The geometric features are identified through the decoding layer to obtain the geometric defect diagnosis result and the first confidence level corresponding to the geometric defect diagnosis result; wherein, the geometric defect diagnosis result includes the type and location of the geometric defect.

7. The method according to claim 1, characterized in that, The diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map are used to perform comprehensive reasoning through a fuzzy logic system to determine the final detection result, including: Determine the material information of the power battery integrated busbar, as well as the environmental information of the power battery integrated busbar; Based on the material information and the environmental information, the confidence levels corresponding to the geometric point cloud map, the visible light point cloud map and the infrared point cloud map are dynamically corrected to obtain the final confidence level. The fuzzy logic system is used to comprehensively reason about the diagnostic results corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, as well as the final confidence level, to obtain the final detection result; wherein, the final detection result includes one of the following: qualified, scrapped, reworked, and requesting manual re-inspection.

8. A visual inspection system for surface defects of integrated busbars in power batteries, characterized in that, The visual inspection system for surface defects of the integrated busbar of the power battery includes: an acquisition module, a generation module, an identification module, and a determination module, wherein... The acquisition module is used to acquire multimodal images and view transformation matrices of the integrated busbar of the power battery; wherein, the multimodal images include structured light depth images from multiple viewpoints, visible light images from multiple viewpoints, and infrared thermal images from multiple viewpoints. The generation module is used to generate geometric point cloud maps, visible light point cloud maps, and infrared point cloud maps based on the structured light depth image, the visible light image, the infrared thermal image, and the viewpoint transformation matrix through a cross-modal fusion neural network. The identification module is used to identify the defect type based on the geometric point cloud map, the visible light point cloud map and the infrared point cloud map, and obtain the diagnosis results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map and the infrared point cloud map respectively; The determining module is used to determine the final detection result by performing comprehensive reasoning through a fuzzy logic system based on the diagnostic results and confidence levels corresponding to the geometric point cloud map, the visible light point cloud map, and the infrared point cloud map, respectively.

9. A visual inspection device for surface defects of integrated busbars in power batteries, characterized in that, include: Processor and memory, of which, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The device stores executable instructions for causing a processor to execute the method according to any one of claims 1 to 7.

Images