Battery pack appearance defect detection method, device, equipment, medium and program product

Abstract

The embodiments of the present application provide a method, apparatus, device, medium, and program product for detecting defects in the appearance of a battery pack, wherein the method includes: determining an initial transformation matrix between two adjacent frames of point cloud data acquired based on the acquisition sequence of multiple frames of point cloud data corresponding to the battery pack; wherein the acquisition perspectives of each frame of point cloud data are different; adjusting the initial transformation matrix based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features to obtain a transformation matrix corresponding to each frame of point cloud data; converting the multiple frames of point cloud data based on the transformation matrix to obtain a fused point cloud structure of the battery pack; and performing defect detection on the battery pack based on the fused point cloud structure to obtain a detection result. By constraining the point cloud conversion through the global structural features, point cloud quality, and local geometric features of the multiple frames of point cloud data, a complete and accurate battery pack structure is reconstructed, thereby improving the accuracy of surface detection.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of battery technology, and are related to, but not limited to, a method, apparatus, device, medium, and program product for detecting appearance defects of a battery pack. Background Art

[0002] Point cloud data is widely used in current applications such as 3D modeling and appearance inspection. For example, when inspecting the appearance of a battery pack, the surface may have a metal casing with uneven structures, curved surfaces, or slight deformations. Image recognition is difficult to quantify depth information when detecting defects, and reflections and shadows from the metal on the 2D image can lead to inaccurate inspection results. Therefore, 3D modeling using point cloud data is necessary for accurate appearance inspection.

[0003] However, the solutions provided by related technologies mostly focus on single-view or small-scale point cloud registration optimization, which often suffers from occlusion, noise, and low point cloud quality in some areas, making it impossible to achieve complete, accurate, and high-fidelity 3D reconstruction. Battery packs are large, and the sizes and types of defects vary widely. When using the methods described in related technologies for registration, a single point cloud data set cannot cover the entire battery pack, or when multiple point cloud data sets are stitched together, it cannot resolve the low point cloud quality and local optimal point cloud issues in some areas. As a result, the reconstructed battery pack point cloud structure cannot be used for accurate defect detection. Summary of the Invention

[0004] In order to solve the problems existing in the related technologies, the embodiments of the present application provide a battery pack appearance defect detection method, device, equipment, medium and program product, which constrains the point cloud conversion through the global structural characteristics, point cloud quality and local geometric characteristics of multi-frame point cloud data, reconstructs a complete and accurate battery pack structure, and improves the accuracy of surface detection.

[0005] In a first aspect, the present application provides a method for detecting defects in the appearance of a battery pack, which method includes: determining an initial transformation matrix between two frames of point cloud data in adjacent acquisition sequences based on the acquisition sequence of multiple frames of point cloud data corresponding to the battery pack; wherein the acquisition perspectives of each frame of point cloud data are different; based on the global structural characteristics of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and local geometric characteristics, adjusting the initial transformation matrix to obtain a transformation matrix corresponding to each frame of point cloud data; converting multiple frames of point cloud data based on the transformation matrix to obtain a fused point cloud structure of the battery pack; and performing defect detection on the battery pack based on the fused point cloud structure to obtain a detection result.

[0006] In the above embodiment, multi-frame point cloud data collected from different perspectives can supplement local data from different angles, reduce blind spots, and reconstruct a complete battery pack structure; the embodiment of the present application adjusts the initial transformation matrix through the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features. It can adjust the transformation matrix based on point cloud data with high point cloud quality, reduce the influence of noise, and make the conversion of point cloud data more reliable. The transformation matrix is ​​constrained by local features and global features during the registration process, so that the transformation matrix can improve the registration accuracy while reducing the registration deviation caused by local optimality, and obtain a more accurate fused point cloud structure of the battery pack. Defect detection is performed based on the fused point cloud structure, which can improve the accuracy and efficiency of battery pack surface defect detection, reduce the false detection rate of manual and two-dimensional image detection, and provide reliable quality assurance for new energy battery manufacturing.

[0007] In some embodiments, the battery pack appearance defect detection method also includes: extracting multi-scale features of each point in each frame of point cloud data to obtain multi-scale features of each point in each frame of point cloud data; fusing the multi-scale features of each point in each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, and the global structural features of each frame of point cloud data to obtain fused features of each point in each frame of point cloud data; determining the initial transformation matrix between two frames of point cloud data in adjacent acquisition sequences based on the acquisition sequence of the multi-frame point cloud data corresponding to the battery pack, including: obtaining a preset transformation matrix between two frames of point cloud data in adjacent acquisition sequences; determining the initial transformation matrix between two frames of point cloud data in adjacent acquisition sequences based on the preset transformation matrix, the fused features of each point in each frame of point cloud data, and the initial position of each point in each frame of point cloud data in the initial coordinate system of each frame of point cloud data.

[0008] In the above embodiment, by adjusting the change matrix through the fusion features corresponding to the two point clouds, the feature information of the point cloud surface can be better utilized to achieve point cloud matching and improve the accuracy of point cloud registration.

[0009] In some embodiments, multi-scale feature extraction is performed on each point in each frame of point cloud data to obtain multi-scale features of each point in each frame of point cloud data, including: feature extraction is performed on multiple neighborhoods of different radii corresponding to each point in each frame of point cloud data to obtain multiple neighborhood features of each point; multiple neighborhood features of each point are spliced ​​to obtain multi-scale features of each point in each frame of point cloud data.

[0010] In some embodiments, based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features, the initial transformation matrix is ​​adjusted to obtain the transformation matrix corresponding to each frame of point cloud data, including: constructing a global optimization function based on two frames of point cloud data in adjacent acquisition orders and the corresponding initial transformation matrix; determining the initial transformation matrix when the global optimization function meets the first target condition as the optimization matrix; based on the point cloud quality of each point in each frame of point cloud data, performing weighted calculation on the optimization matrix to obtain a weighted transformation matrix between two frames of point cloud data in adjacent acquisition orders; based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data, adjusting the weighted transformation matrix to obtain the transformation matrix corresponding to each frame of point cloud data.

[0011] In the above embodiment, two frames of point cloud data in adjacent acquisition order and the corresponding initial transformation matrix are globally optimized so that multiple local point clouds can be spliced ​​into a complete point cloud, providing high-precision input data for battery pack defect detection; the optimization matrix is ​​adjusted according to the point cloud quality of each point in each frame of point cloud data, which can suppress the interference of noise areas and avoid the registration from falling into local optimality, making the registration more dependent on reliable areas and improving the registration accuracy; finally, by combining local geometric features and global structural features to comprehensively constrain the point cloud transformation, the registration result can reflect local details while maintaining overall consistency.

[0012] In some embodiments, the weighted transformation matrix is ​​adjusted based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data to obtain the transformation matrix corresponding to each frame of point cloud data, including: constructing a target optimization function based on the global structural features of each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix between two frames of point cloud data in adjacent acquisition order, the local weight coefficient and the global weight coefficient; and determining the weighted transformation matrix when the target optimization function meets the second target condition as the transformation matrix.

[0013] In some embodiments, multiple frames of point cloud data are converted based on a transformation matrix to obtain a fused point cloud structure of the battery pack, including: based on the transformation matrix corresponding to each frame of point cloud data, each frame of point cloud data is converted to a global coordinate system to obtain multiple frames of converted point cloud data; and multiple frames of converted point cloud data are spliced ​​to obtain a fused point cloud structure.

[0014] In the above embodiment, by converting each frame of point cloud data into a global coordinate system and splicing them, these local point cloud data can be integrated into a complete three-dimensional model, thereby being able to more comprehensively and accurately reflect the geometric structure of the battery pack, provide accurate input data for defect detection of the battery pack, and improve the accuracy of defect detection.

[0015] In some embodiments, the battery pack appearance defect detection method also includes: based on the neighborhood of each point in each frame of point cloud data, performing local feature extraction on each point in each frame of point cloud data to obtain local geometric features of each point in each frame of point cloud data; performing feature extraction at different levels on each frame of point cloud data to obtain global structural features of each frame of point cloud data; based on the average distance between each point in the corresponding neighborhood of each point in each frame of point cloud data, determining the point density of each point in each frame of point cloud data; based on the local geometric features and point density of each point in each frame of point cloud data, determining the point cloud quality of each point in each frame of point cloud data.

[0016] In the above embodiment, detailed information on the battery pack surface (such as dents and cracks) can be captured through local feature extraction, while the overall shape of the battery pack can be captured through global feature extraction, and the point cloud quality of each point can be obtained, ensuring that stable and reliable features can still be extracted in complex scenarios.

[0017] In some embodiments, feature extraction at different levels is performed on each frame of point cloud data to obtain global structural features of each frame of point cloud data, including: performing local feature extraction at different levels multiple times on each frame of point cloud data to obtain multiple local features corresponding to each frame of point cloud data; performing global convolution on multiple local features corresponding to each frame of point cloud data to obtain global structural features corresponding to each frame of point cloud data.

[0018] In some embodiments, the point cloud quality of each point in each frame of point cloud data is determined based on the local geometric features and point density of each point in each frame of point cloud data, including: determining the covariance matrix of the neighborhood corresponding to each point and the covariance eigenvalue of each point based on the local geometric features of each point in each frame of point cloud data; determining the noise of each point in each frame of point cloud data based on the covariance eigenvalue; performing weighted calculation on the noise and point density of each point in each frame of point cloud data to obtain the point cloud quality of each point in each frame of point cloud data.

[0019] In a second aspect, an embodiment of the present application provides a battery pack appearance defect detection device, which includes: a determination module for determining an initial transformation matrix between two frames of point cloud data in adjacent acquisition sequences based on the acquisition sequence of multiple frames of point cloud data corresponding to the battery pack; an adjustment module for adjusting the initial transformation matrix based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features to obtain a transformation matrix corresponding to each frame of point cloud data; a conversion module for converting multiple frames of point cloud data based on the transformation matrix to obtain a fused point cloud structure of the battery pack; and a defect detection module for performing defect detection on the battery pack based on the fused point cloud structure to obtain a detection result.

[0020] In a third aspect, an embodiment of the present application provides a battery pack appearance defect detection device, which includes: a memory for storing executable instructions; and a processor for implementing the above-mentioned battery pack appearance defect detection method when executing the executable instructions stored in the memory.

[0021] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned battery pack appearance defect detection method.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product or a computer program, which includes executable instructions, and the executable instructions are stored in a computer-readable storage medium; when the processor of the battery pack appearance defect detection device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the above-mentioned battery pack appearance defect detection method is implemented.

[0023] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the structure of a battery pack appearance defect detection device provided in an embodiment of the present application;

[0025] Figure 2 This is an optional process diagram of the battery pack appearance defect detection method provided in the embodiment of the present application. Figure 1 ;

[0026] Figure 3 This is an optional process diagram of the battery pack appearance defect detection method provided in the embodiment of the present application. Figure 2 ;

[0027] Figure 4 This is an optional process diagram of the battery pack appearance defect detection method provided in the embodiment of the present application. Figure 3 ;

[0028] Figure 5 This is an optional flowchart of the battery pack detection process provided in an embodiment of the present application;

[0029] Figure 6 This is an optional flowchart of point cloud stitching provided in an embodiment of the present application;

[0030] Figure 7 This is an optional structural diagram of point cloud stitching provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0032] In the following description, reference is made to "some embodiments," which describe a subset of all possible embodiments. However, it will be understood that "some embodiments" may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by those skilled in the art to which the embodiments of this application pertain. The terms used in the embodiments of this application are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0033] Currently, due to the limited acquisition range of sensors, when a large-scale three-dimensional environmental model needs to be constructed, a single sensor cannot cover the entire target area. Therefore, point cloud data from multiple sensors or multiple perspectives must be registered and stitched together to form a complete large-scale point cloud. In order to obtain a complete large-scale point cloud, point clouds from different perspectives or different positions must be effectively registered and stitched together. However, due to possible displacement, rotation, and scale differences between point clouds, the registration process is complex and prone to errors. In addition, point cloud data often contains noise, and due to occlusion, reflection, and other reasons, the collected point cloud may be incomplete, affecting the subsequent registration and stitching accuracy.

[0034] Point cloud stitching techniques in related technologies can be categorized into feature-based registration methods, iterative closest point (ICP) algorithms, and global optimization methods. The ICP algorithm is commonly used for point cloud registration, but it relies on initial alignment and is susceptible to error accumulation when registering multiple point clouds, resulting in poor stitching results. Feature-based registration extracts geometric features from point clouds for matching, but when stitching over large areas, feature matching is unstable and susceptible to noise and occlusion. Global optimization methods, such as the Globally Optimal Iterative Closest Point (GO-ICP), avoid local optimality through global search, but are computationally intensive, unable to meet real-time requirements, and exhibit poor performance in large-scale stitching. These methods are effective in simple scenarios, but when stitching over large areas, they still face challenges such as low accuracy, high computational complexity, and poor noise immunity.

[0035] For inspection targets like battery packs, which are large and often have uneven surfaces (such as heat sinks and sealing edges), curved surfaces, or subtle deformations (such as assembly errors and collision dents), two-dimensional images make it difficult to accurately quantify depth information. Battery pack surfaces may be made of reflective materials (such as aluminum alloy casings) or dark coatings, making traditional optical imaging susceptible to interference from reflections and shadows. Furthermore, battery packs require inspection of hidden areas such as sides, bottoms, and seams. Existing registration and stitching methods are mostly optimized for single-viewpoint or small-scale point clouds. When stitching multiple large-scale point clouds, they cannot effectively address inter-sensor displacement, rotation, occlusion, and noise. They also cannot resolve low point cloud quality and local optimality in certain areas, making accurate defect detection impossible and computationally inefficient. Therefore, a new inspection method is needed that can efficiently stitch multiple point clouds while ensuring detection accuracy to form a complete, multi-viewpoint, large-scale point cloud for precise battery pack surface inspection.

[0036] In order to alleviate the problems existing in the related art, the applicant believes that the multi-perspective point cloud data can be transformed. The multi-perspective point cloud data can cover the entire surface of the battery pack, reduce occlusion, and eliminate blind spots that cannot be captured by a single perspective. It is suitable for the complex structure of the battery pack surface; and the present application adjusts the transformation matrix through local features, global features and point cloud quality to avoid local optimality and interference from low-quality areas, which can improve the alignment accuracy and obtain the accurate point cloud structure of the battery pack, thereby improving the accuracy of battery pack defect detection.

[0037] Based on the above considerations, the inventors have conducted in-depth research and provided a method for detecting defects in the appearance of a battery pack. The method can obtain multi-frame point cloud data by adopting different acquisition perspectives. Based on the acquisition order of the multi-frame point cloud data corresponding to the battery pack, the initial conversion matrix between two frames of point cloud data in adjacent acquisition orders is determined. Based on the global structural characteristics of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data and the local geometric characteristics, the initial conversion matrix is ​​adjusted to obtain the conversion matrix corresponding to each frame of point cloud data. Based on the conversion matrix, the multi-frame point cloud data is converted to obtain a fused point cloud structure of the battery pack. Based on the fused point cloud structure, the battery pack is subjected to defect detection to obtain a detection result.

[0038] In this way, the multi-frame point cloud data collected by the embodiment of the present application from different perspectives can supplement the local data from different angles, reduce blind spots, and reconstruct the complete battery pack structure; the embodiment of the present application adjusts the initial transformation matrix through the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features. It can adjust the transformation matrix based on point cloud data with high point cloud quality, reduce the influence of noise, and make the conversion of point cloud data more reliable. The transformation matrix is ​​constrained by local features and global features in the alignment process, so that the transformation matrix can improve the alignment accuracy while reducing the alignment deviation caused by local optimality, and obtain a more accurate fused point cloud structure of the battery pack. Defect detection is performed based on the fused point cloud structure, which can improve the accuracy and efficiency of battery pack surface defect detection, reduce the false detection rate of manual and two-dimensional image detection, and provide reliable quality assurance for new energy battery manufacturing.

[0039] The application of new energy batteries in life and industry is becoming more and more extensive. New energy batteries are not only used in energy storage power systems such as hydropower, thermal power, wind power and solar power stations, but are also widely used in electric vehicles such as electric bicycles, electric motorcycles, electric cars, as well as in aerospace and other fields. With the continuous expansion of the application field of power batteries, the market demand is also constantly expanding. In the embodiments of the present application, the battery involved may be a battery cell, also known as a battery cell. A battery cell refers to a basic unit that can realize the mutual conversion of chemical energy and electrical energy, which can be used to make a battery module or battery pack, and thus used to supply power to electrical devices. A battery cell may be a secondary battery, which refers to a battery cell that can continue to be used by activating the active material by charging after the battery cell is discharged. The battery cell may be a lithium-ion battery, a sodium-ion battery, a sodium-lithium-ion battery, a lithium metal battery, a sodium metal battery, a lithium-sulfur battery, a magnesium-ion battery, a nickel-hydrogen battery, a nickel-cadmium battery, a lead-acid battery, etc., and the embodiments of the present application are not limited to this.

[0040] In the embodiments of this application, a cell can refer to any shape, such as a square cell or a round cell. A cell generally refers to a battery cell, which is one of the basic units that make up a battery. A cell is the core component of a battery, responsible for storing and releasing electrical energy. Cells can include lithium-ion cells, lithium-polymer cells, nickel-metal hydride cells, and others. The embodiments of this application do not limit the type of cell; the type can be selected based on the actual application scenario. In the embodiments of this application, a cell is the core component of a battery pack. A battery pack typically includes multiple cells, which are combined to provide the required energy capacity and voltage. A battery pack comprises at least battery cells, a battery management system (BMS), a housing, a wiring harness, connectors, and interfaces. These components work together to combine the battery cells into a fully functional battery pack for use in various applications. For example, the battery pack can be used in electric vehicles, energy storage systems, portable electrical devices, solar energy systems, wind energy systems, emergency backup power supplies, power tools, or electric bicycles, etc. The present application embodiment does not impose any restrictions on this, and the specific selection can be made according to the actual application scenario.

[0041] It should be noted that the battery pack can use different types of battery cells, such as lithium-ion batteries, nickel-metal hydride batteries, lithium polymer batteries, etc., depending on the actual application needs and performance requirements.

[0042] In the embodiment of the present application, the battery may also be a single physical module including one or more battery cells to provide higher voltage and capacity. When there are multiple battery cells, the multiple battery cells are connected in series, in parallel or in hybrid via a busbar.

[0043] The following describes an exemplary application of the battery pack appearance defect detection device according to an embodiment of the present application. The battery pack appearance defect detection device provided in the embodiment of the present application can be executed by a processor of a computer device. During implementation, the computer device can be any suitable device with data processing capabilities. It is understood that in battery industrial production, the computer device can refer to any one of a programmable logic controller (PLC), a single-chip microcomputer, an intermediate computer, and a host computer. It can also be a server, a laptop, a tablet computer, a desktop computer, a smartphone, etc. The computer device can also refer to a lower computer, such as an industrial computer, a programmable logic controller (PLC), etc. In some embodiments, the computer device may include a memory and a processor, wherein the memory stores a computer program that can be executed on the processor, and when the processor executes the program, the battery pack appearance defect detection method is implemented.

[0044] Figure 1 is a structural diagram of a battery pack appearance defect detection device provided in an embodiment of the present application, Figure 1 The battery pack appearance defect detection device 10 shown may include at least one processor 110, a memory 150, at least one network interface 120, and a user interface 130. The various components in the battery pack appearance defect detection device 10 are coupled together via a bus system 140. It is understood that the bus system 140 is used to achieve connection and communication between these components. In addition to including a data bus, the bus system 140 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clarity, Figure 1 Various buses are labeled as bus system 140 .

[0045] The processor 110 may be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor may be a microprocessor or any conventional processor, etc.

[0046] The user interface 130 includes one or more output devices 131 that enable presentation of media content, and one or more input devices 132 .

[0047] The memory 150 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, and the like. The memory 150 may optionally include one or more storage devices physically remote from the processor 110. The memory 150 may include volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory may be read-only memory (ROM), and volatile memory may be random access memory (RAM). The memory 150 described in the embodiments of the present application is intended to include any suitable type of memory. In some embodiments, the memory 150 is capable of storing data to support various operations. Examples of such data include programs, modules, and data structures, or subsets or supersets thereof, as exemplified below.

[0048] Operating system 151, including system programs for processing various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, and driver layer, which are used to implement various basic services and process hardware-based tasks;

[0049] A network communication module 152 for reaching other computing devices via one or more (wired or wireless) network interfaces 120 , exemplary network interfaces 120 including Bluetooth, WiFi, and USB;

[0050] The input processing module 153 is configured to detect one or more inputs or interactions from one of the one or more input devices 132 .

[0051] In some embodiments, the apparatus provided in the embodiments of the present application may be implemented in software. Figure 1 A battery pack appearance defect detection device 154 stored in memory 150 is shown. This battery pack appearance defect detection device 154 can be a battery pack appearance defect detection device in a battery pack appearance defect detection device. It can be software in the form of a program or plug-in, and includes the following software modules: a determination module 1541, an adjustment module 1542, a conversion module 1543, and a defect detection module 1544. These modules are logical and can be arbitrarily combined or further separated according to the functions they implement. The functions of each module will be described below.

[0052] In other embodiments, the apparatus provided in the embodiments of the present application may be implemented in hardware. As an example, the apparatus provided in the embodiments of the present application may be a processor in the form of a hardware decoding processor, which is programmed to execute the battery pack appearance defect detection method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor may adopt one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0053] In the embodiments of the present application, the battery pack appearance defect detection method can be used to detect surface defects on structures such as battery packs, battery modules, single cells, or electrode sheets. Therefore, the battery pack appearance defect detection equipment can be any of a programmable logic controller (PLC), a single-chip microcomputer, an intermediate computer, or a host computer at any appearance inspection station on a battery production line. The technical solution of this application will be described in detail below with reference to the accompanying drawings.

[0054] Figure 2 This is an optional process diagram of the battery pack appearance defect detection method provided in the embodiment of the present application. Figure 1 ,like Figure 2 As shown, the battery pack appearance defect detection method provided in the embodiment of the present application can be implemented through steps S201 to S204:

[0055] Step S201: determining an initial transformation matrix between two adjacent frames of point cloud data acquired in a sequence based on an acquisition sequence of multiple frames of point cloud data corresponding to the battery pack; wherein each frame of point cloud data has a different acquisition perspective.

[0056] In an embodiment of the present application, when the battery pack arrives at the designated position after entering the station, the host computer can obtain the geometric structure information of the battery pack, plan the scanning path of the battery pack based on the scanning range of the scanner, and control the scanner to scan the battery pack based on the scanning path to obtain multiple frames of point cloud data. There may be overlapping areas between adjacent point cloud data.

[0057] Here, each frame of point cloud data can be a spatial data set of multiple points, and each point contains three-dimensional coordinates and other additional information (such as color, intensity, etc.).

[0058] The scanner can be a Light Detection and Ranging (LiDAR) system, a depth camera, a laser radar, or a millimeter-wave radar. The scanning path can have multiple acquisition points. During scanning based on the scanning path, data is collected once at each acquisition point, resulting in multiple raw signals. The raw signals are converted into point cloud data. For example, millimeter-wave radar signals require analog-to-digital conversion and filtering, and depth images are converted into 3D point clouds using an intrinsic parameter matrix. LiDAR data may require triangulation algorithms to generate 3D point clouds.

[0059] In the embodiment of the present application, the different acquisition perspectives of each frame of point cloud data can supplement local data through different angles, reduce blind spots, include more geometric features (such as edges, contours, planes, etc.), improve the stability and accuracy of alignment, and reconstruct a more comprehensive battery pack structure.

[0060] In an embodiment of the present application, the initial transformation matrix may be the geometric transformation parameters required to align the spatial position of the point cloud data Pj (current frame) to the coordinate system of the point cloud data Pi (previous frame) through rotation and translation transformation.

[0061] When stitching and registering multiple frames of point cloud data, the transformation and stitching is performed based on the relative position and posture between adjacent frames, that is, the initial transformation matrix. In the embodiments of the present application, a nearest neighbor search can be used to match feature point pairs of adjacent frame point cloud data. The rotation matrix and translation vector between two adjacent point cloud data can be calculated through methods such as singular value decomposition (SVD), and the rotation matrix and translation vector are combined to obtain the initial transformation matrix.

[0062] Here, the initial transformation matrix can optimize the relative position between each pair of adjacent point clouds, and the initial transformation matrix can be used to perform registration between adjacent point clouds.

[0063] Step S202: Based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features, the initial transformation matrix is ​​adjusted to obtain the transformation matrix corresponding to each frame of point cloud data.

[0064] In an embodiment of the present application, the global structural features of each frame of point cloud data are used to describe the overall shape, size and spatial distribution characteristics of the point cloud frame. They can be obtained by performing global feature extraction on each point cloud data. For example, they can be extracted through a global pooling method, or they can be obtained by aggregating multi-level global information and expanding the receptive field layer by layer, expanding the neighborhood range of each layer, capturing a wider range of contextual relationships, and finally outputting global features.

[0065] Here, the point cloud quality of each point in each frame of point cloud data is used to evaluate the acquisition reliability of the point cloud data, including data integrity, noise level and density distribution, which can be determined by the point density and noise level of the point cloud.

[0066] The local geometric features of each point in each frame of point cloud data are used to describe the geometric characteristics of a specific area or point in the point cloud, which is used for detail matching and defect location. They can be obtained by extracting features from the neighborhood of each point.

[0067] In some embodiments, the initial transformation matrix is ​​adjusted based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features. This may refer to global optimization of multiple initial change matrices to obtain multiple optimization matrices, so that multiple local point clouds can be spliced ​​into a complete point cloud based on the optimization matrix. The initial transformation matrix is ​​then weighted calculated based on the point cloud quality of each point in each frame of point cloud data, which can give high-quality point clouds (such as points with low noise and significant structural features) a greater weight, and obtain a weighted feature matrix, so that key structural points can be more accurately matched in the transformation matrix alignment. In each alignment, local features and global features can be used to jointly constrain the transformation between point clouds to obtain a transformation matrix, thereby improving the alignment accuracy.

[0068] Here, global optimization can adopt a global optimization strategy based on particle swarm optimization (PSO) and genetic algorithm (GA) to improve the relative transformation between point clouds and avoid local optimality.

[0069] In an embodiment of the present application, the transformation matrix can be a set of multiple matrices, with each frame of point cloud data corresponding to one matrix. The transformation matrix can realize the rotation and translation of each frame of point cloud data to align all point cloud data into a coordinate system.

[0070] Step S203: convert the multi-frame point cloud data based on the conversion matrix to obtain a fused point cloud structure of the battery pack.

[0071] In some embodiments, after obtaining the transformation matrix, the coordinates of each frame of point cloud data can be multiplied by the corresponding transformation matrix, and finally the multiple frames of point cloud data are aligned into a coordinate system to obtain a fused point cloud structure of the battery pack. The fused point cloud structure presents the three-dimensional geometric features of the battery pack, including details such as surface concave-convex structure, electrode distribution, welds and surface defects.

[0072] Step S204: Based on the fused point cloud structure, perform defect detection on the battery pack to obtain a detection result.

[0073] In the embodiments of the present application, defect detection can include detecting surface dents, unevenness, cracks, assembly errors, etc. The fused point cloud structure can be used for battery pack status analysis. For example, by calculating the point cloud normal vector and curvature features, dent detection, shell deformation, or weld defects can be achieved. The fused point cloud structure can be compared with the battery pack design model to determine whether there are assembly errors. A neural network can be used to identify whether there are cracks on the battery pack surface, thereby obtaining the battery pack inspection results.

[0074] In an embodiment of the present application, multi-frame point cloud data collected from different perspectives can supplement local data from different angles, reduce blind spots, and reconstruct a complete battery pack structure; the embodiment of the present application adjusts the initial transformation matrix through the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features. The transformation matrix can be adjusted based on point cloud data with high point cloud quality, which can reduce the influence of noise and make the conversion of point cloud data more reliable. The transformation matrix is ​​constrained in the alignment process through local features and global features, so that the transformation matrix can improve the alignment accuracy while reducing the alignment deviation caused by local optimality, thereby obtaining a more accurate fused point cloud structure for the battery pack. Defect detection is performed based on the fused point cloud structure, which can improve the accuracy and efficiency of battery pack surface defect detection, reduce the false detection rate of manual and two-dimensional image detection, and provide reliable quality assurance for new energy battery manufacturing.

[0075] To address the scale differences of battery pack surface defects (such as small dents and large-area deformations) and capture geometric information at different scales, the embodiment of the present application can extract multi-scale features of point cloud data using neighborhood radii of different sizes. Therefore, the battery pack appearance defect detection method provided by the embodiment of the present application can also include steps S1 and S2:

[0076] Step S1: extract multi-scale features from each point in each frame of point cloud data to obtain multi-scale features of each point in each frame of point cloud data.

[0077] In some embodiments, step S1 may be implemented by steps S11 and S12:

[0078] Step S11 : performing feature extraction on multiple neighborhoods with different radii corresponding to each point in each frame of point cloud data to obtain multiple neighborhood features of each point.

[0079] In the embodiment of the present application, for each point in the point cloud data , you can build multiple neighborhoods with different radii , , each neighborhood corresponds to a different scale, and feature extraction is performed on the points in each neighborhood. The features of the points in the field are extracted to obtain the neighborhood features corresponding to the neighborhood, and finally multiple neighborhood features of each point are obtained.

[0080] In some embodiments, a small radius neighborhood can detect micron-level scratches or bumps, and a large radius neighborhood can identify defects such as overall flatness and overall battery deformation.

[0081] Step S12: splicing multiple neighborhood features of each point to obtain multi-scale features of each point in each frame of point cloud data.

[0082] In some embodiments, multiple neighborhood features of each point are spliced ​​together to obtain a multi-scale feature of each point. , as shown in formula (1):

[0083] (1);

[0084] in, Indicates a point Local features at the kth scale.

[0085] The embodiment of the present application uses multi-scale feature extraction to achieve small-radius neighborhood capture of local details and large-radius neighborhood coverage of wider areas, and can flexibly handle feature extraction of sparse and dense areas. The combination of the two can simultaneously detect local looseness and overall deformation of the battery pack surface, avoid missed detection caused by a single scale, and improve the accuracy of battery pack appearance inspection.

[0086] Step S2: Fusing the multi-scale features of each point in each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, and the global structural features of each frame of point cloud data to obtain the fused features of each point in each frame of point cloud data.

[0087] In the embodiment of the present application, the local geometric features of each point in each frame of point cloud data are , the global structural characteristics of each frame point cloud data are , the multi-scale features of each point in each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, and the global structural features of each frame of point cloud data can be fused through the Fusion function, as shown in formula (2):

[0088] (2);

[0089] in, Represents the fusion feature corresponding to the i-th point in the point cloud.

[0090] Correspondingly, step S201 can be implemented through steps S2011 and S2022:

[0091] Step S2011: Obtain a preset transformation matrix between two frames of point cloud data in an adjacent acquisition sequence.

[0092] In an embodiment of the present application, since the shape and parameters of the battery pack are known, the approximate relative rotation and translation between two frames of point cloud data in adjacent acquisition sequences are known, and these values ​​can be used as a preset transformation matrix; or a voxel grid-based downsampling method can be used to quickly align two frames of point cloud data in adjacent acquisition sequences to obtain a preset transformation matrix.

[0093] Step S2012: Determine the initial transformation matrix between two adjacent frames of point cloud data based on the preset transformation matrix, the fusion features of each point in each frame of point cloud data, and the initial position of each point in the initial coordinate system of each frame of point cloud data.

[0094] In some embodiments, by comparing two adjacent point cloud data and The relative transformation between them is matched to obtain the initial transformation matrix , and use the fusion features corresponding to the two point clouds to adjust the change matrix, so that the feature information of the point cloud surface can be better used to achieve point cloud matching, as shown in formulas (3) and (4):

[0095] (3);

[0096] exp (4);

[0097] in, and Point cloud and The fusion feature of the k-th point in Point cloud and The initial positions of the kth point in the initial coordinate systems of point clouds Pi and Pj are is a preset transformation matrix, and σ is a parameter that controls the speed of weight decay, which can be set based on prior experience.

[0098] In the embodiment of the present application, by adjusting the change matrix through the fusion features corresponding to the two point clouds, the feature information of the point cloud surface can be better utilized to achieve point cloud matching and improve the accuracy of point cloud registration.

[0099] Figure 3 This is an optional process diagram of the battery pack appearance defect detection method provided in the embodiment of the present application. Figure 2 ,like Figure 3As shown, step S202 in the battery pack appearance defect detection method provided in the embodiment of the present application can be implemented through steps S301 to S304:

[0100] Step S301: construct a global optimization function based on two frames of point cloud data in adjacent acquisition order and the corresponding initial transformation matrix.

[0101] In the embodiment of the present application, on the basis of step-by-step registration, any feasible global optimization strategy (such as global least squares and matrix decomposition, deep learning or exchange average algorithm, etc.) can be adopted to further improve the relative transformation between point clouds and avoid local optimality. The constructed global optimization function S is shown in formula (5):

[0102] (5);

[0103] Step S302: Determine the initial transformation matrix when the global optimization function satisfies the first target condition as the optimization matrix.

[0104] In some embodiments, the first target condition may be an initial transformation matrix that minimizes the global optimization function S. The optimization matrix may be obtained by formula (6):

[0105] (6);

[0106] Formula (6) represents the global transformation optimization between multiple point clouds, ensuring that the registration result of each pair of point clouds can meet the global optimal solution.

[0107] In some embodiments, the optimization matrix may be a set of multiple matrices, with each point cloud data corresponding to one matrix.

[0108] Step S303: Based on the point cloud quality of each point in each frame of point cloud data, a weighted calculation is performed on the optimization matrix to obtain a weighted transformation matrix between two frames of point cloud data in adjacent acquisition sequences.

[0109] In the embodiments of this application, The point cloud quality of the kth point in the point cloud data Pi can be determined based on its point density and noise level. Point density refers to the number of points per unit volume, measured by the average distance between points in the neighborhood of the kth point. The noise level refers to the degree to which a point deviates from the true surface, measured by calculating the change in the local plane normal using local features. Areas of low density or high noise are considered to be low quality.

[0110] During the process of stitching multiple point clouds, different point clouds may have different qualities, such as density and noise level. Therefore, embodiments of the present application introduce a dynamic weighting mechanism to adjust the optimization matrix based on point cloud quality. Based on the quality of different points in the point cloud data, the contribution weight of each point in the matrix calculation can be determined to adjust the optimization matrix. For example, high-quality areas (high density, low noise, and complete features) are given higher weights during the registration process, while low-quality areas are given lower weights to reduce their negative impact on the overall registration results. Weight adjustment can be achieved through key point optimization or sparse Mixed-ICP algorithms, etc.

[0111] The weighted transformation matrix after weighting can be calculated by formula (7):

[0112] (7);

[0113] Step S304 : Based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix is ​​adjusted to obtain a transformation matrix corresponding to each frame of point cloud data.

[0114] In some embodiments, during the step-by-step registration process, in order to more accurately stitch point clouds, the present application can improve the robustness and accuracy of the registration process by combining local geometric features with global structural features. For example, during each registration, local geometric features and global structural features can be used to jointly constrain the transformation between point clouds, and the weighted transformation matrix can be adjusted to obtain the transformation matrix corresponding to each frame of point cloud data.

[0115] Therefore, step S304 can be implemented through steps S3041 and S3042:

[0116] Step S3031: Construct a target optimization function based on the global structural features of each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix between two frames of point cloud data in adjacent acquisition order, the local weight coefficient and the global weight coefficient.

[0117] In some embodiments, the target optimization function X may be as shown in formula (8):

[0118] (8);

[0119] in, and is the weight coefficient of local features and global features, which can be set based on experience. Point Cloud A collection of local geometric features, It is a point cloud global structural features.

[0120] Step S3032: Determine the weighted transformation matrix when the target optimization function satisfies the second target condition as the conversion matrix.

[0121] In some embodiments, the target optimization function satisfying the second target condition may mean that, as shown in formula (9), the weighted transformation matrix that minimizes the target optimization function X is determined as the transformation matrix of the point cloud data Pi.

[0122] (9);

[0123] Here, the transformation matrix is ​​a set of matrices, one for each point cloud. After transforming each point cloud using Tcombined, the registered point clouds are directly added together to obtain the complete point cloud.

[0124] In an embodiment of the present application, two frames of point cloud data in an adjacent acquisition sequence and the corresponding initial transformation matrix are globally optimized so that multiple local point clouds can be spliced ​​into a complete point cloud, providing high-precision input data for battery pack defect detection; the optimization matrix is ​​adjusted according to the point cloud quality of each point in each frame of point cloud data, which can suppress the interference of noise areas and avoid the registration from falling into local optimality, making the registration more dependent on reliable areas and improving the registration accuracy; finally, the point cloud transformation is comprehensively constrained by combining local geometric features and global structural features, so that the registration result can reflect local details while maintaining overall consistency.

[0125] In some embodiments, step S203 may be implemented by steps S2031 and S2032:

[0126] Step S2031 : Based on the transformation matrix corresponding to each frame of point cloud data, transform each frame of point cloud data into a global coordinate system to obtain multiple frames of transformed point cloud data.

[0127] In an embodiment of the present application, the transformation matrix contains rotation and translation information. For each frame of point cloud data, its corresponding transformation matrix is ​​used to transform it from the original coordinate system to the global coordinate system. This can be achieved through matrix multiplication, multiplying the coordinates of each point in the point cloud with the transformation matrix.

[0128] Step S2032: splice multiple frames of converted point cloud data to obtain a fused point cloud structure.

[0129] In an embodiment of the present application, after multiple frames of point cloud data are converted into a global coordinate system, the point sets of all point cloud data can be merged into a large point set through point cloud stitching or merging operations, thereby obtaining a fused point cloud structure of the battery pack to achieve visualization of the battery pack.

[0130] Here, duplicate points and noise points can be removed during stitching to improve the quality of the fused point cloud structure.

[0131] In an embodiment of the present application, by converting each frame of point cloud data into a global coordinate system and splicing them, these local point cloud data can be integrated into a complete three-dimensional model, thereby being able to more comprehensively and accurately reflect the geometric structure of the battery pack, provide accurate input data for defect detection of the battery pack, and improve the accuracy of defect detection.

[0132] Figure 4 This is an optional process diagram of the battery pack appearance defect detection method provided in the embodiment of the present application. Figure 3 ,like Figure 4 As shown, the battery pack appearance defect detection method provided in the embodiment of the present application may further include steps S401 to S404:

[0133] Step S401 : Based on the neighborhood of each point in each frame of point cloud data, local features are extracted for each point in each frame of point cloud data to obtain local geometric features of each point in each frame of point cloud data.

[0134] In the embodiment of the present application, local feature extraction may refer to the process of extracting features of each point by using a local perception mechanism (such as k-nearest neighbor search) to extract features of points in the neighborhood of each point through a convolution operation, thereby forming local geometric features of each point in each frame of point cloud data. The local feature vector It can be expressed by formula (10):

[0135] (10);

[0136] in, The first Points, is the neighborhood of the point, Represents a local convolution operation.

[0137] Step S402: extract features at different levels from each frame of point cloud data to obtain global structural features of each frame of point cloud data.

[0138] In an embodiment of the present application, performing feature extraction at different levels on each frame of point cloud data may refer to performing feature extraction at different levels through multi-layer convolution operations to obtain global structural features of each frame of point cloud data.

[0139] In some embodiments, step S402 may be implemented by steps S4021 and S4022:

[0140] Step S4021: Perform multiple local feature extractions at different levels on each frame of point cloud data to obtain multiple local features corresponding to each frame of point cloud data.

[0141] Here, we can use convolution kernels of different sizes to extract local features of each frame of point cloud data multiple times at different levels. The first layer uses a small convolution kernel (such as 3×3) to extract fine features, and the second layer uses a slightly larger convolution kernel (such as 5×5) to extract more extensive features, and obtain multiple local features corresponding to each frame of point cloud data. .

[0142] Step S4022: Perform global convolution on multiple local features corresponding to each frame of point cloud data to obtain global structural features corresponding to each frame of point cloud data.

[0143] In an embodiment of the present application, local features can be aggregated through feature aggregation, fully connected layers or global convolution to obtain global structural features, so as to aggregate context information layer by layer. For example, local features at different levels can be aggregated using skip connections to obtain global structural features; global convolution can also be used to convert multiple local features into global structural features.

[0144] In some embodiments, the global structural features As shown in formula (11):

[0145] (11);

[0146] in, is the local feature of the k-th point, Represents the global convolution operation, which outputs the global structural features corresponding to each point cloud data.

[0147] Step S403: Determine the point density of each point in each frame of point cloud data based on the average distance between each point in the neighborhood corresponding to each point in each frame of point cloud data.

[0148] In the embodiment of the present application, the point density can be calculated by calculating the kth point in the point cloud data Pi It is measured by the average distance between each point in the neighborhood of and the kth point, as shown in formula (12):

[0149] (12);

[0150] Among them, Daverage is the average distance between each point and the kth point in the neighborhood corresponding to the kth point, and e is a small constant used to avoid division by zero errors.

[0151] Step S404: determining the point cloud quality of each point in each frame of point cloud data based on the local geometric features and point density of each point in each frame of point cloud data.

[0152] In the embodiment of the present application, the noise level of each point can be obtained through local geometric features.

[0153] In some embodiments, step S404 may be implemented by steps S4041 and S4043:

[0154] Step S4041: Based on the local geometric features of each point in each frame of point cloud data, determine the covariance matrix of the neighborhood corresponding to each point and the covariance eigenvalue of each point.

[0155] In an embodiment of the present application, the covariance matrix can reflect the overall distribution of the data, and the degree of concentration or dispersion of the data can be determined by the covariance eigenvalues. After the local geometric features of each point in each frame of point cloud data have been determined, the covariance matrix of the neighborhood of each point can be determined based on the local geometric features. For example, the centroid of all points in the neighborhood of each point is calculated based on the local geometric features, the coordinates of the points in the neighborhood are centered relative to the centroid, and the covariance matrix is ​​calculated using the centered coordinates.

[0156] Perform eigendecomposition on the covariance matrix to obtain the covariance eigenvalues .

[0157] Step S4042: Determine the noise of each point in each frame of point cloud data based on the covariance eigenvalue.

[0158] In the embodiment of the present application, the noise of each point in each frame of point cloud data can be determined by using the covariance eigenvalue, which can be achieved by formula (13):

[0159] (13);

[0160] in, are the eigenvalues ​​of the covariance matrix, is the minimum eigenvalue, reflecting the noise level. Here, when the point cloud data is locally a perfect plane, Approaching 0, the noise level approaches 0, which is in line with expectations.

[0161] Step S4043: Perform weighted calculation on the noise and point density of each point in each frame of point cloud data to obtain the point cloud quality of each point in each frame of point cloud data.

[0162] In the embodiment of the present application, the point cloud quality of each point can be achieved by formula (14):

[0163] (14);

[0164] in, Point Cloud The point density of the kth point in , Point Cloud The noise of the kth point, , It can be an empirical parameter.

[0165] The embodiment of the present application can capture detailed information (such as dents and cracks) on the surface of the battery pack through local feature extraction, and at the same time capture the overall shape of the battery pack through global feature extraction, and obtain the point cloud quality of each point, ensuring that stable and reliable features can still be extracted in complex scenarios.

[0166] The following describes an exemplary application of the embodiments of the present application in a practical application scenario.

[0167] In the embodiment of the present application, in view of the large amount of point cloud data, large pack area, and large variations in defect size in the battery pack surface inspection scenario, there is a need to stitch multiple large-scale point clouds. A multi-scale point cloud stitching method (GROMS) based on global registration optimization is proposed. Through deep learning-enhanced multi-scale feature extraction and global convergence optimization methods, the problem of effective registration and stitching of multiple limited-range point clouds is solved.

[0168] In the pack inspection scenario, the battery pack surface may have defects such as dents and cracks, and traditional inspection methods have difficulty dealing with noise, occlusion, and local optimality in point cloud data. To meet the needs of high-precision and high-efficiency point cloud registration and defect detection in pack inspection tasks, this application achieves efficient and robust point cloud registration and defect detection for packs through multi-level feature extraction, step-by-step registration and global optimization, dynamic weighting and adaptive adjustment, and the combination of local and global information.

[0169] Among them, multi-level feature extraction can refer to the noise and occlusion problems in the point cloud scanning of the Pack surface. Local feature extraction (which can be through k-nearest neighbor search and local convolution) is performed through a new multi-scale hierarchical feature extraction network (MS-HFEN) to capture the detailed information of the battery pack surface (such as dents and cracks). At the same time, global feature extraction (aggregating context information layer by layer) is used to capture the overall shape of the battery pack, ensuring that robust features can still be extracted in complex scenarios.

[0170] Gradual registration and global optimization can refer to the situation in pack inspection tasks where a single scan can only capture a partial point cloud due to the scanner's limited field of view. This application generates an initial transformation matrix through gradual registration and combines it with particle swarm optimization (PSO) and genetic algorithms (GA) for global optimization, ensuring that multiple partial point clouds can be stitched together into a complete point cloud, providing high-precision input data for subsequent defect detection.

[0171] Dynamic weighting and adaptive adjustment can refer to the problem of uneven point cloud quality on the Pack surface (such as low point cloud density or high noise in some areas), introducing a point cloud quality scoring mechanism, and dynamically adjusting the registration weight, thereby improving the stability and accuracy of the detection results.

[0172] Combining local and global information can mean that in pack inspection, both local defects (such as dents) and overall shape (such as battery pack surface flatness) need to be inspected. This application combines local geometric information with global structural information to comprehensively constrain point cloud transformations, ensuring that the registration results reflect both local details and maintain overall consistency.

[0173] Figure 5 This is an optional flow chart of the battery pack detection process provided in the embodiment of the present application, such as Figure 5 As shown, the battery pack detection process provided in the embodiment of the present application can be implemented through steps S501 to S504:

[0174] Step S501: The battery pack enters a designated position of a testing station.

[0175] In an embodiment of the present application, a laser or infrared sensor is provided at the entrance of the inspection station to detect whether the battery pack has entered the station. The pack to be inspected enters the inspection area of ​​the inspection station. The host computer can control the ranging sensor or metal detection sensor to detect the approximate position of the battery pack to detect whether the battery pack has reached the designated inspection position.

[0176] Step S502: collect data from the battery pack to obtain multiple frames of point cloud data.

[0177] In an embodiment of the present application, the host computer controls a movable mechanism (such as a robotic arm) to clamp or manually hold a scanner (such as a lidar, depth camera, etc.) to perform an all-round scan of the battery pack to obtain multiple local point clouds. Each point in the local point cloud contains three-dimensional coordinates and other additional information (such as color, intensity, etc.).

[0178] Step S503: perform registration and splicing processing on the multi-frame point cloud data to obtain a complete point cloud.

[0179] The multi-scale point cloud stitching method (GROMS) of the embodiment of the present application is used to stitch multiple local point clouds into a complete point cloud.

[0180] Step S504: perform defect detection on the battery pack based on the complete point cloud.

[0181] The embodiments of the present application can use a point cloud-level defect detection algorithm based on complete point cloud data (for example, by calculating the local curvature and normal direction changes of the point cloud to identify surface anomalies (such as dents, cracks, etc.)) to identify defects on the surface of the battery pack (such as dents, cracks, etc.).

[0182] The embodiments of the present application solve common problems such as noise, occlusion and local optimality in point cloud registration through feature extraction, step-by-step optimization, dynamic weighting and information fusion, and also provide an efficient and reliable solution for Pack detection, which can improve the registration accuracy.

[0183] In the embodiments of this application, Figure 6 This is an optional flow chart of point cloud stitching provided in the embodiment of the present application, such as Figure 6 As shown, the battery pack detection process provided in the embodiment of the present application can be implemented through steps S601 to S607:

[0184] Step S601: Acquire multiple frames of point cloud data.

[0185] In an embodiment of the present application, the scanner moves one position to collect one frame of point cloud data, and multiple frames of point cloud data are obtained after the collection of the entire battery pack is completed.

[0186] Step S602: Perform multi-level feature extraction on the multi-frame point cloud data to obtain local features, multi-scale features, global features and fusion features.

[0187] In point cloud registration problems, dealing with noise, occlusion, scale differences, and other issues usually requires extracting features from different scales (or levels). In the pack inspection task, there may be defects such as dents and cracks on the surface of the battery pack, and traditional methods have difficulty extracting effective features from point clouds with severe noise and occlusion. In order to effectively capture the global and local structural information of the point cloud, this application provides a multi-scale hierarchical feature extraction network (MS-HFEN, Multi-Scale Hierarchical Feature Extraction Network), which includes the extraction of local features (i.e., local geometric features), multi-scale features, global features (i.e., global structural features), and fused features.

[0188] Among them, local feature extraction can refer to the use of local perception mechanism (such as k-nearest neighbor search) in the feature extraction process of each layer, extracting the local geometric features of each point through convolution operation, and forming the local feature vector (i.e. local geometric features) of each point in each frame of point cloud data. , local eigenvector It can be expressed by formula (15):

[0189] (15);

[0190] in, The first Points, is the neighborhood of the point, Represents a local convolution operation.

[0191] Multi-scale feature extraction can refer to designing a multi-scale feature extraction module to deal with the scale differences of the Pack surface point cloud (such as small dents and large area deformation) and capture geometric information of different scales. Multi-scale features are extracted through neighborhood radii of different sizes: for each point , construct multiple neighborhoods with different radii , , corresponding to different scales. For each scale neighborhood, use the local feature extraction module to extract features, and splice the multi-scale features to obtain the multi-scale features of each point , as shown in formula (16):

[0192] (16);

[0193] in, Indicates a point Local features at the kth scale.

[0194] Global feature extraction can refer to the process of increasing the receptive field layer by layer through multi-level global information aggregation, so that the features of each layer can be combined with more point cloud context information. Finally, the global feature is output. , to represent the overall shape information of the point cloud, global features (i.e. global structural features) , as shown in formula (17):

[0195] (17);

[0196] in, is the local feature of the kth layer, Represents the global convolution operation, which outputs the global features corresponding to each local point cloud.

[0197] Feature fusion can be achieved by fusing local and global features to obtain a comprehensive feature vector , this vector contains both local detail information and global morphological information, as shown in formula (18):

[0198] (18);

[0199] in, Represents the fusion feature corresponding to the i-th point in the point cloud.

[0200] Step S603: gradually register multiple frames of point cloud data based on the fusion features to obtain an initial transformation matrix between two adjacent point cloud data acquisition sequences.

[0201] In some embodiments, step-by-step registration refers to aligning multiple frames of point cloud data to the same coordinate system. First, the point cloud and The relative transformation between two adjacent point cloud data is preliminarily matched to obtain the initial transformation matrix (i.e., the initial conversion matrix) , and use the corresponding features of the two point clouds to adjust the change matrix, so that the feature information of the surface pack point cloud can be better used to achieve point cloud matching. Then, gradually optimize the relative position between each pair of point clouds and use the initial transformation matrix for registration. As shown in formulas (19) and (20):

[0202] (19);

[0203] exp (20);

[0204] in, and Point cloud and The fusion feature of the position of the k-th point in Point cloud and The position of the kth point in is the initial matrix between manually entered point clouds, and σ is a parameter that controls the speed of weight decay.

[0205] Step S604: Globally optimize the initial transformation matrix to obtain a global optimized matrix.

[0206] On the basis of step-by-step registration, a global optimization strategy based on particle swarm optimization (PSO) and genetic algorithm (GA) can be used to further improve the relative transformation between point clouds and avoid local optimality. The optimization objective is shown in formula (21):

[0207] (twenty one);

[0208] This formula represents the global transformation optimization between multiple point clouds, ensuring that the registration result of each pair of point clouds can meet the global optimal solution.

[0209] Step S605: Dynamically weight the global optimization matrix to obtain a weighted transformation matrix.

[0210] In some embodiments, during the process of stitching multiple point clouds, different point clouds may have different qualities, such as point cloud density, noise level, etc. Therefore, the present invention introduces a dynamic weighting mechanism to adjust the transformation matrix according to the point cloud quality.

[0211] Set the quality function of the point cloud to , then the weighted transformation matrix It can be expressed as formula (22):

[0212] (twenty two);

[0213] in, is the quality scoring function of the k-th point cloud point, which is usually determined according to the point density and noise level of the point cloud. It can be expressed as formula (23):

[0214] (twenty three);

[0215] in, Point Cloud The point density of the kth point, Point Cloud The noise of the kth point, , It can be an empirical parameter.

[0216] Among them, the point density can be calculated by calculating the kth point in the point cloud Pi It is measured by the average distance between each point and the kth point in the local neighborhood of , as shown in formula (24):

[0217] (twenty four);

[0218] Where Daverage is the average distance of the k-th point, and e is a small constant used to avoid division by zero errors.

[0219] The noise level can be analyzed by local geometric features, and the change of the normal vector of the local plane at the kth point is calculated using principal component analysis, as shown in formula (25):

[0220] (25);

[0221] in, are the eigenvalues ​​of the covariance matrix, is the minimum eigenvalue, reflecting the noise level. When the point cloud is locally a perfect plane, Approaching 0, the noise level approaches 0, which is in line with expectations.

[0222] Step S606: Adjust the weighted transformation matrix based on the local features and the global features to obtain the transformation matrix of each frame of point cloud data.

[0223] Step S607: Based on the transformation matrix of each frame of point cloud data, the point cloud data of each frame is spliced ​​to obtain a complete point cloud.

[0224] In order to more accurately stitch point clouds during the step-by-step registration process, the present embodiment combines local geometric information with global structural information to improve the robustness and accuracy of the registration process. During each registration, local features and global features can be used to jointly constrain the transformation between point clouds, thereby improving registration accuracy. This process can be expressed as formula (26) by optimizing the objective function:

[0225] (26);

[0226] in, and is the weight coefficient of local features and global features (can be set manually), Point Cloud A collection of local features, It is a point cloud global characteristics.

[0227] Here, Tcombined can be a set of matrices, with each point cloud data corresponding to a transformation matrix. After transforming each point cloud data through Tcombined, the registered point clouds are directly added together to obtain the complete point cloud.

[0228] Figure 7 This is an optional structural diagram of point cloud stitching provided by the embodiment of the present application, such as Figure 7 As shown, 701 is the point cloud Point cloud data, 702 is point cloud Point cloud data, 703 is point cloud With point cloud Point cloud structure after registration.

[0229] In order to ensure the effectiveness of the multi-scale feature extraction network (MS-HFEN) and the point cloud registration algorithm, the multi-scale feature extraction network and the point cloud registration algorithm can be trained. The training process is as follows:

[0230] First, you can use a public point cloud dataset (such as ModelNet40, KITTI dataset) or a custom dataset as training data. The dataset contains multiple point cloud samples and their corresponding ground truth transformation matrices.

[0231] In order to train the multi-scale hierarchical feature extraction network (MS-HFEN) and the registration algorithm, the loss function can be designed as shown in formula (27):

[0232] =λ1 +λ2 (27);

[0233] in, is the feature extraction loss; is the registration loss.

[0234] Contrastive Loss can be used to train the feature extraction network. The goal of contrastive loss is to make the feature vectors of similar point clouds closer in the feature space, while the feature vectors of dissimilar point clouds are farther apart. The feature extraction loss can be expressed as formula (28):

[0235] (28);

[0236] in, Point cloud The eigenvector of . is the Euclidean distance between eigenvectors. is a label, when If it is a similar point cloud, otherwise . is the interval parameter that controls the minimum distance between dissimilar point clouds.

[0237] In order to optimize the accuracy of point cloud registration, the mean square error (MSE) of point cloud registration can be used as the loss function to measure the error between the predicted transformation matrix and the true transformation matrix. The registration loss can be shown as formula (29):

[0238] (29);

[0239] in, is the predicted transformation matrix. , Point cloud The position of the kth point in . N is the number of points in the point cloud.

[0240] During training, the convolutional layer parameters of the multi-scale feature extraction network (MS-HFEN) are initialized, and the Adam optimizer is used for training. The initial learning rate and learning rate decay strategy are set. The sample point cloud data is input into the multi-scale hierarchical feature extraction network (MS-HFEN) to extract local features and global features. Based on the extracted features, the transformation matrix is ​​calculated. The loss function is used. Calculate the error and update the network parameters through back propagation. Repeat the above process until the loss function converges or the predetermined number of training rounds is reached.

[0241] The embodiments of the present application significantly improve the accuracy of stitching multiple point clouds, especially in complex environments, by combining multi-level feature extraction with global optimization; the multi-scale features extracted by deep learning enable the present application to enhance its noise and occlusion resistance and have greater robustness when facing noise and occlusion; 3) through dynamic weighting and step-by-step alignment strategies, it reduces computational redundancy, improves the efficiency of the stitching process, and can better cope with large-scale point cloud data; it can handle the stitching of large-scale and multiple point clouds to meet the needs of large-scale three-dimensional modeling and environmental reconstruction.

[0242] based on Figure 1 A battery pack appearance defect detection device 154 is provided. The battery pack appearance defect detection device 154 may include a determination module 1541, an adjustment module 1542, a conversion module 1543 and a defect detection module 1544, wherein the determination module 1541 is used to determine the initial conversion matrix between two frames of point cloud data in adjacent acquisition sequences based on the acquisition sequence of multiple frames of point cloud data corresponding to the battery pack; wherein the acquisition perspectives of each frame of point cloud data are different; the adjustment module 1542 is used to adjust the initial conversion matrix based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data and the local geometric features, to obtain the conversion matrix corresponding to each frame of point cloud data; the conversion module 1543 is used to convert multiple frames of point cloud data based on the conversion matrix to obtain a fused point cloud structure of the battery pack; the defect detection module 1544 is used to perform defect detection on the battery pack based on the fused point cloud structure to obtain a detection result.

[0243] In some embodiments, the battery pack appearance defect detection device also includes: a multi-scale feature extraction module, which is used to extract multi-scale features of each point in each frame of point cloud data to obtain multi-scale features of each point in each frame of point cloud data; a fusion module, which is used to fuse the multi-scale features of each point in each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, and the global structural features of each frame of point cloud data to obtain the fusion features of each point in each frame of point cloud data; the determination module 1541 is also used to obtain a preset transformation matrix between two frames of point cloud data in adjacent acquisition sequences; based on the preset transformation matrix, the fusion features of each point in each frame of point cloud data, and the initial position of each point in each frame of point cloud data in the initial coordinate system of each frame of point cloud data, the initial transformation matrix between two frames of point cloud data in adjacent acquisition sequences is determined.

[0244] In some embodiments, the multi-scale feature extraction module is further used to extract features from multiple neighborhoods of different radii corresponding to each point in each frame of point cloud data to obtain multiple neighborhood features of each point; and to splice the multiple neighborhood features of each point to obtain multi-scale features of each point in each frame of point cloud data.

[0245] In some embodiments, the adjustment module 1542 is also used to construct a global optimization function based on two frames of point cloud data in adjacent acquisition sequences and the corresponding initial transformation matrix; the initial transformation matrix when the global optimization function meets the first target condition is determined as the optimization matrix; based on the point cloud quality of each point in each frame of point cloud data, the optimization matrix is ​​weightedly calculated to obtain a weighted transformation matrix between two frames of point cloud data in adjacent acquisition sequences; based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix is ​​adjusted to obtain the transformation matrix corresponding to each frame of point cloud data.

[0246] In some embodiments, the adjustment module 1542 is also used to construct a target optimization function based on the global structural features of each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix between two frames of point cloud data in adjacent acquisition order, the local weight coefficient and the global weight coefficient; and the weighted transformation matrix when the target optimization function meets the second target condition is determined as the conversion matrix.

[0247] In some embodiments, the conversion module 1543 is also used to convert each frame of point cloud data to a global coordinate system based on the conversion matrix corresponding to each frame of point cloud data to obtain multiple frames of converted point cloud data; and splice the multiple frames of converted point cloud data to obtain a fused point cloud structure.

[0248] In some embodiments, the battery pack appearance defect detection device also includes: a local feature extraction module, which is used to perform local feature extraction on each point in each frame of point cloud data based on the neighborhood of each point in each frame of point cloud data, and obtain the local geometric features of each point in each frame of point cloud data; a feature extraction module, which is used to perform feature extraction at different levels on each frame of point cloud data, and obtain the global structural features of each frame of point cloud data; a first determination module, which is used to determine the point density of each point in each frame of point cloud data based on the average distance between each point in the corresponding neighborhood of each point in each frame of point cloud data; a second determination module, which is used to determine the point cloud quality of each point in each frame of point cloud data based on the local geometric features and point density of each point in each frame of point cloud data.

[0249] In some embodiments, the feature extraction module is also used to perform multiple local feature extractions at different levels on each frame of point cloud data to obtain multiple local features corresponding to each frame of point cloud data; and perform global convolution on the multiple local features corresponding to each frame of point cloud data to obtain global structural features corresponding to each frame of point cloud data.

[0250] In some embodiments, the second determination module is also used to determine the covariance matrix of the neighborhood corresponding to each point and the covariance eigenvalue of each point based on the local geometric features of each point in each frame of point cloud data; determine the noise of each point in each frame of point cloud data based on the covariance eigenvalue; perform weighted calculation on the noise and point density of each point in each frame of point cloud data to obtain the point cloud quality of each point in each frame of point cloud data.

[0251] It should be noted that the description of the device embodiment of the present application is similar to the description of the method embodiment described above, and has similar beneficial effects as the method embodiment, so it will not be repeated. For technical details not disclosed in the device embodiment, please refer to the description of the method embodiment of the present application for understanding.

[0252] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements some or all of the steps in the above method. The computer-readable storage medium may be transient or non-transient.

[0253] An embodiment of the present application provides a computer program, including computer-readable code. When the computer-readable code runs in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0254] Embodiments of the present application provide a computer program product comprising a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps of the above-described method. The computer program product may be implemented in hardware, software, or a combination thereof. In some embodiments, the computer program product is embodied as a computer storage medium. In other embodiments, the computer program product is embodied as a software product, such as a software development kit (SDK).

[0255] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned steps / processes does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments.

[0256] This application uses descriptions of directions or positional relationships such as "up", "down", "top", "bottom", "front", "back", "inside" and "outside" to facilitate the description of this application, and does not indicate or imply that the device referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, it should not be understood as limiting the scope of protection of this application.

[0257] It should also be noted that, in the description of this application, unless otherwise specified or limited, the terms "installed," "connected," and "connected" should be understood broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to direct connections or indirect connections through an intermediary. Those skilled in the art will understand the specific meanings of these terms in this application based on the specific circumstances.

[0258] It should be noted that, in this application, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0259] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0260] The units described above as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, the functional units in the various embodiments of the present application may all be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0261] The above are merely examples of the present application and are not intended to limit the scope of protection of the present application. Any modifications, equivalent replacements, and improvements made within the spirit and scope of the present application are included in the scope of protection of the present application.

Claims

1. A method for detecting appearance defects of a battery pack, characterized in that: The battery pack appearance defect detection method includes: Determining, based on the order in which multiple frames of point cloud data corresponding to the battery pack are collected, an initial transformation matrix between two adjacent frames of point cloud data collected from different perspectives; wherein the initial transformation matrix is ​​determined based on a preset transformation matrix between the two adjacent frames of point cloud data collected from different perspectives, and the characteristics and positions of each point in each frame of point cloud data; Adjusting the initial transformation matrix based on the global structural characteristics of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric characteristics to obtain a transformation matrix corresponding to each frame of point cloud data; the point cloud quality includes point density and noise level, and the noise level is the degree to which the point deviates from the actual surface of the battery pack; Converting the multi-frame point cloud data based on the conversion matrix to obtain a fused point cloud structure of the battery pack; Based on the fused point cloud structure, performing defect detection on the battery pack to obtain a detection result; The initial transformation matrix is ​​adjusted based on the global structural features of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric features to obtain the transformation matrix corresponding to each frame of point cloud data, including: Based on two frames of point cloud data acquired in adjacent order and the corresponding initial transformation matrix, a global optimization function is constructed; Determine the initial transformation matrix when the global optimization function satisfies the first objective condition as the optimization matrix; Based on the point cloud quality of each point in each frame of point cloud data, the optimization matrix is ​​weightedly calculated to obtain a weighted transformation matrix between two frames of point cloud data in adjacent acquisition sequences; Based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix is ​​adjusted to obtain a conversion matrix corresponding to each frame of point cloud data.

2. The battery pack appearance defect detection method according to claim 1, characterized in that: The battery pack appearance defect detection method further includes: Perform multi-scale feature extraction on each point in each frame of point cloud data to obtain multi-scale features of each point in each frame of point cloud data; The multi-scale features of each point in each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, and the global structural features of each frame of point cloud data are fused to obtain the fusion features of each point in each frame of point cloud data; The determining of the initial conversion matrix between two frames of point cloud data in adjacent acquisition sequences based on the acquisition sequence of the multiple frames of point cloud data corresponding to the battery pack includes: Obtain the preset transformation matrix between two frames of point cloud data acquired in adjacent order; Based on the preset transformation matrix, the fusion features of each point in each frame of point cloud data and the initial position of each point in each frame of point cloud data in the initial coordinate system of each frame of point cloud data, the initial transformation matrix between two frames of point cloud data in adjacent acquisition order is determined.

3. The method for detecting appearance defects of a battery pack according to claim 2, wherein: The multi-scale feature extraction of each point in each frame of point cloud data to obtain the multi-scale features of each point in each frame of point cloud data includes: Feature extraction is performed on multiple neighborhoods with different radii corresponding to each point in each frame of point cloud data to obtain multiple neighborhood features of each point; Multiple neighborhood features of each point are spliced ​​together to obtain multi-scale features of each point in each frame of point cloud data.

4. The method for detecting appearance defects of a battery pack according to claim 1, wherein: The weighted transformation matrix is ​​adjusted based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data to obtain a transformation matrix corresponding to each frame of point cloud data, including: Constructing a target optimization function based on the global structural features of each frame of point cloud data, the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix between two adjacent frames of point cloud data, the local weight coefficient and the global weight coefficient; The weighted transformation matrix when the objective optimization function satisfies the second objective condition is determined as the conversion matrix.

5. The method for detecting appearance defects of a battery pack according to any one of claims 1 to 4, characterized in that: The converting the multi-frame point cloud data based on the conversion matrix to obtain a fused point cloud structure of the battery pack includes: Based on the transformation matrix corresponding to each frame of point cloud data, each frame of point cloud data is transformed into a global coordinate system to obtain multiple frames of transformed point cloud data; The multiple frames of converted point cloud data are spliced ​​to obtain the fused point cloud structure.

6. The method for detecting appearance defects of a battery pack according to any one of claims 1 to 4, characterized in that: The battery pack appearance defect detection method further includes: Extracting local features of each point in each frame of point cloud data based on the neighborhood of each point in each frame of point cloud data to obtain local geometric features of each point in each frame of point cloud data; Performing feature extraction at different levels on each frame of point cloud data to obtain global structural features of each frame of point cloud data; Determine the point density of each point in each frame of point cloud data based on the average distance between each point in the neighborhood corresponding to each point in each frame of point cloud data; Based on the local geometric features and point density of each point in each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data is determined.

7. The method for detecting appearance defects of a battery pack according to claim 6, wherein: The extracting features at different levels from each frame of point cloud data to obtain global structural features of each frame of point cloud data includes: Performing multiple local feature extractions at different levels on each frame of point cloud data to obtain multiple local features corresponding to each frame of point cloud data; Global convolution is performed on multiple local features corresponding to each frame of point cloud data to obtain global structural features corresponding to each frame of point cloud data.

8. The method for detecting appearance defects of a battery pack according to claim 7, wherein: Determining the point cloud quality of each point in each frame of point cloud data based on the local geometric features and point density of each point in each frame of point cloud data includes: Determining the covariance matrix of the neighborhood corresponding to each point and the covariance eigenvalue of each point based on the local geometric features of each point in each frame of point cloud data; Determining the noise of each point in each frame of point cloud data based on the covariance eigenvalue; A weighted calculation is performed on the noise and point density of each point in each frame of point cloud data to obtain the point cloud quality of each point in each frame of point cloud data.

9. A battery pack appearance defect detection device, characterized in that: The battery pack appearance defect detection device includes: a determination module for determining, based on the order in which multiple frames of point cloud data corresponding to the battery pack are collected, an initial transformation matrix between two frames of point cloud data collected in adjacent sequences; the initial transformation matrix is ​​determined based on a preset transformation matrix between the two frames of point cloud data collected in adjacent sequences, and features and positions of each point in each frame of point cloud data; an adjustment module, configured to adjust the initial transformation matrix based on the global structural characteristics of each frame of point cloud data, the point cloud quality of each point in each frame of point cloud data, and the local geometric characteristics, to obtain a transformation matrix corresponding to each frame of point cloud data; the point cloud quality includes point density and noise level, where the noise level is the degree to which the points deviate from the actual surface of the battery pack; a conversion module, configured to convert the multi-frame point cloud data based on the conversion matrix to obtain a fused point cloud structure of the battery pack; a defect detection module, configured to perform defect detection on the battery pack based on the fused point cloud structure and obtain a detection result; Among them, the adjustment module is also used to construct a global optimization function based on two frames of point cloud data in adjacent acquisition sequences and the corresponding initial transformation matrix; the initial transformation matrix when the global optimization function meets the first target condition is determined as the optimization matrix; based on the point cloud quality of each point in each frame of point cloud data, the optimization matrix is ​​weighted calculated to obtain a weighted transformation matrix between two frames of point cloud data in adjacent acquisition sequences; based on the global structural features of each frame of point cloud data and the local geometric features of each point in each frame of point cloud data, the weighted transformation matrix is ​​adjusted to obtain the transformation matrix corresponding to each frame of point cloud data.

10. A battery pack appearance defect detection device, characterized in that: The battery pack appearance defect detection equipment includes: A memory for storing executable instructions; a processor for implementing the battery pack appearance defect detection method according to any one of claims 1 to 8 when executing the executable instructions stored in the memory.

11. A computer-readable storage medium, characterized in that Executable instructions are stored, which are used to cause the processor to execute the executable instructions to implement the battery pack appearance defect detection method according to any one of claims 1 to 8.

12. A computer program product, characterized in that The computer program product includes executable instructions stored in a computer-readable storage medium; When the processor of the battery pack appearance defect detection device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the battery pack appearance defect detection method according to any one of claims 1 to 8 is implemented.

Images