Battery pack defect identification method, device and equipment

Abstract

The invention discloses a battery pack defect identification method, device and equipment, and belongs to the technical field of battery pack quality detection.The method comprises the steps that a dual-channel defect identification model is constructed, and dual channels comprise a 2D image channel and a 3D point cloud channel; determining a 2D recognition result and a 3D recognition result according to the 2D image data and the 3D point cloud data of the battery pack based on the dual-channel defect recognition model; when the 2D recognition result and the 3D recognition result meet a first preset condition, extracting damage features of the 2D recognition result and structural features of the 3D recognition result; and carrying out weighted fusion on the damage features and the structural features, and identifying the defects of the battery pack according to the features after weighted fusion. According to the invention, through combination of 2D image deep learning and a 3D point cloud template matching technology, accurate identification of battery pack defects is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of battery pack quality detection, and particularly to a method, device and equipment for identifying battery pack defects. Background Art

[0002] As one of the core functional components of new energy vehicles, the battery pack is responsible for storing electrical energy and providing power for the vehicle. It consists of multiple battery cells and is managed and controlled by a specific battery management system (BMS). Due to the influence of various factors during the production and use of battery cells, the performance of each cell inside the battery pack gradually becomes inconsistent. If this inconsistency is not discovered and processed in time, it may lead to safety accidents. Therefore, it is particularly important to conduct a comprehensive and accurate detection of the battery pack.

[0003] Traditional defect identification technologies for retired battery packs mainly rely on manual visual inspection, which is inefficient and affected by subjective factors. Using a single 2D vision can only identify surface damage and cannot detect three-dimensional structural deformations such as bulges and pits. Pure 3D point cloud detection has a high complexity in data processing and poor real-time performance, making it difficult to accurately identify tiny defects.

[0004] Therefore, there is an urgent need for a method, device and equipment for identifying battery pack defects to solve the technical problem of being unable to accurately identify battery pack defects in the existing technology. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, device and equipment for identifying battery pack defects, which can combine 2D image deep learning and 3D point cloud template matching technology to achieve accurate identification of battery pack defects.

[0006] To solve the above technical problems, on the one hand, the present invention provides a method for identifying battery pack defects, including: Construct a dual-channel defect identification model, where the dual-channel includes a 2D image channel and a 3D point cloud channel; Based on the dual-channel defect identification model, determine 2D identification results and 3D identification results according to the 2D image data and 3D point cloud data of the battery pack; When the 2D identification results and 3D identification results meet the first preset condition, extract the damage features of the 2D identification results and the structural features of the 3D identification results; Perform weighted fusion on the damage features and structural features, and identify the defects of the battery pack according to the features after weighted fusion.

[0007] In a possible implementation manner, constructing the dual-channel defect identification model includes: Taking 2D images as input and damaged marked images as output, construct a damage identification model for the 2D image channel; Construct a deformation recognition model for the 3D point cloud channel, with 3D point cloud data as the input and deformed marked point cloud data as the output; The damage recognition model of the 2D image channel and the deformation recognition model of the 3D point cloud channel form a dual-channel defect recognition model.

[0008] In a possible implementation, constructing the damage recognition model for the 2D image channel includes: Based on the YOLOv7-tiny model, with 2D images as the input and damaged marked images as the output, construct an initial damage recognition model; Optimize the initial damage recognition model based on the Retinex algorithm and the attention mechanism to obtain an optimized damage recognition model; Based on the generative adversarial network, expand the 2D images of the training battery pack to obtain a training dataset, and iterate the optimized damage recognition model according to the training dataset to obtain the damage recognition model for the 2D image channel.

[0009] In a possible implementation, the deformation recognition model includes a preprocessing layer, a coarse matching layer, a fine registration layer, a deformation index calculation layer, and an output layer; The preprocessing layer is used to filter and denoise the input data of the model; The coarse matching layer is used to coarsely register the preprocessed 3D point cloud data with a preset template based on the SAC-IA algorithm; The fine registration layer is used to precisely align the 3D point cloud data with the preset template based on the iterative closest point algorithm on the basis of coarse registration; The deformation index calculation layer is used to calculate the deformation amount between the 3D point cloud data and the preset template based on the Hausdorff distance algorithm according to the precisely aligned result; The output layer is used to determine the deformation position and deformation degree according to the deformation amount, mark the deformation of the 3D point cloud data according to the deformation position and deformation degree, and output the deformed marked point cloud data.

[0010] In a possible implementation, the weighted fusion of the damage features and the structural features includes: Determine the first weight combination set according to the confidence of the historical 2D recognition result and the matching score of the historical 3D recognition result; Based on the cross-validation method, perform a network search on the first weight combination set, select the optimal weight combination to obtain the second weight combination; Dynamically optimize the second weight combination according to the real-time environmental data to obtain the target weight combination; Perform weighted fusion on the damage features and the structural features according to the target weight combination.

[0011] In a possible implementation manner, dynamically optimizing the second weight combination according to the real-time environment data to obtain a target weight combination includes: Based on a sliding window mechanism, dynamically adjusting and optimizing the second weight combination according to the acquired real-time environment data to obtain a target weight combination, wherein the weight coefficient adjustment formula for the 2D channel in the target weight combination is: , wherein, is the weight coefficient of the 2D channel in the target weight combination, is the weight coefficient of the 2D channel in the second weight combination, is the light change parameter; The weight coefficient adjustment formula for the 3D channel in the target weight combination is: , wherein, is the weight coefficient of the 3D channel in the target weight combination, is the weight coefficient of the 3D channel in the second weight combination, is the point cloud noise level.

[0012] In a possible implementation manner, after defect detection of the battery pack, it further includes: When the defect of the battery pack meets the second preset condition, calculating the remaining capacity of the battery pack according to the acquired battery pack voltage data and temperature data; When the remaining capacity of the battery pack meets the third preset condition, evaluating the consistency of the battery pack according to the voltage data and the electrochemical impedance.

[0013] In a possible implementation manner, when the remaining capacity of the battery pack meets the third preset condition, evaluating the consistency of the battery pack according to the voltage data and the electrochemical impedance includes: Based on the voltage curve dispersion, obtaining the first consistency result of the battery pack according to the voltage data in the preset state of charge interval; Based on the equivalent circuit model, extracting the electrochemical impedance of the battery pack, analyzing the mapping relationship between the state of charge and the electrochemical impedance, and evaluating the second consistency result of the battery pack according to the mapping relationship; Determining the consistency of the battery pack according to the first consistency result and the second consistency result.

[0014] In a second aspect, the present invention further provides a battery pack defect identification device, including: A model construction module, configured to construct a dual-channel defect identification model, where the dual-channel includes a 2D image channel and a 3D point cloud channel; An initial defect detection module, configured to determine 2D recognition results and 3D recognition results based on the dual-channel defect recognition model according to the 2D image data and 3D point cloud data of the battery pack; A feature extraction module, configured to extract the damage features of the 2D recognition results and the structural features of the 3D recognition results when the 2D recognition results and the 3D recognition results meet a first preset condition; A defect recognition module, configured to perform weighted fusion on the damage features and the structural features, and recognize the defects of the battery pack according to the features after weighted fusion.

[0015] In a third aspect, the present invention further provides a device, including a memory and a processor, wherein the memory is configured to store programs and data; the processor is coupled to the memory and configured to execute the programs stored in the memory to implement the battery pack defect recognition method as described above.

[0016] The beneficial effects of the present invention are as follows: First, a dual-channel defect recognition model including a 2D image channel and a 3D point cloud channel is constructed. This model can simultaneously process the 2D image and 3D point cloud data of the battery pack, recognize the damage defects and deformation defects of the battery pack, and double verification improves the detection reliability; Then, the dual-channel model is used to recognize the 2D image data and 3D point cloud data of the battery pack to obtain 2D recognition results and 3D recognition results; When the 2D recognition results and the 3D recognition results meet the preset conditions, the damage features of the 2D recognition results and the structural features of the 3D recognition results are extracted. This process reduces the computational complexity of the recognition process by eliminating the battery packs with severe defects; Finally, dynamic weighted fusion is performed on the damage features and the structural features, so that the features after weighted fusion can better reflect the defect characteristics of the battery pack, thereby accurately recognizing the state of the battery pack; In this embodiment, the 2D image deep learning and the 3D point cloud template matching technology are used to respectively recognize the damage features and deformation features of the battery pack, and the damage features and deformation features are fused and analyzed to recognize the state of the battery pack, improving the accuracy and efficiency of battery state detection. Description of the Drawings

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a schematic flowchart of an embodiment of the battery pack defect recognition method provided by the present invention; Figure 2 Provided by the present invention Figure 1Flow diagram of an embodiment of step S104; Figure 3 Flow diagram of an embodiment of the battery pack consistency evaluation provided by the present invention; Figure 4 Provided by the present invention Figure 3 Flow diagram of the first embodiment of step S302 in; Figure 5 Flow diagram of the battery pack defect detection and consistency evaluation provided by the present invention; Figure 6 Structural diagram of an embodiment of the battery pack defect identification device provided by the present invention; Figure 7 Structural diagram of an embodiment of the battery pack defect identification equipment provided by the present invention. Detailed implementation manners

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise specified, "a plurality of" means two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships, for example: A and / or B, which can mean: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The descriptions such as "first" and "second" involved in the embodiments of the present invention are only for descriptive purposes, and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" can explicitly or implicitly include at least one of the features.

[0022] Referring to "embodiment" in this article means that the specific features, structures or characteristics described in connection with the embodiment can be included in at least one embodiment of the present invention. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0023] The present invention provides a battery pack defect identification method, device and equipment, which will be described separately below.

[0024] Figure 1The figure is a schematic flowchart of an embodiment of the battery pack defect recognition method provided by the present invention. As shown in Figure 1 the figure, the battery pack defect recognition method includes: S101. Construct a dual-channel defect recognition model, where the dual-channel includes a 2D image channel and a 3D point cloud channel; It should be noted that through the model management software on the electronic terminal, a dual-channel model is constructed to analyze and recognize the 2D image and 3D point cloud data of the battery pack respectively. In this embodiment, a binocular structured light 3D camera is used to obtain point cloud data with an accuracy of 0.05 mm, and a 2D industrial camera is used to collect 20 million pixel RGB images as 2D image data. These 2D image data and 3D point cloud data are uploaded to the electronic terminal, and defect recognition is performed through the model. Among them, the electronic terminal can implement image data processing and model management functions, and can be a PC, a mobile terminal, a portable terminal, etc.

[0025] S102. Based on the dual-channel defect recognition model, determine the 2D recognition result and the 3D recognition result according to the 2D image data and 3D point cloud data of the battery pack; It should be noted that a 2D industrial camera is used to collect 20 million pixel RGB images as 2D image data, and a binocular structured light 3D camera is used to obtain point cloud data with an accuracy of 0.05 mm. These 2D image data and 3D point cloud data of the battery pack are input into the dual-channel defect recognition model to identify and mark the damaged details in the 2D image data of the battery pack, and details such as the degree and position of the damage are marked in detail. The details with deformed structures in the 3D point cloud data are marked, and details such as the position and degree of the deformation are marked in detail.

[0026] S103. When the 2D recognition result and the 3D recognition result meet the first preset condition, extract the damage feature of the 2D recognition result and the structure feature of the 3D recognition result; It should be noted that the damage detection condition and the deformation detection condition are set respectively. When any one of the 2D recognition result and the 3D recognition result of the battery pack exceeds the corresponding detection condition, for example, the damage or deformation at the key position of the battery pack, serious damage, and serious deformation will all lead to the danger during the use of the battery pack. These battery packs are directly eliminated and no further processing of the following steps is performed. When the 2D recognition result and the 3D recognition result both meet the first preset condition, that is, it is impossible to directly determine whether the battery pack can be reused through the 2D recognition result and the 3D recognition result. At this time, the feature information of the 2D recognition result and the 3D recognition result with defect marking information needs to be extracted. This process is a rough screening of the 2D recognition result and the 3D recognition result. Battery packs with serious damage and deformation are directly eliminated to avoid unnecessary subsequent processing.

[0027] Further, it should be noted that representative features are extracted from the 2D recognition result image with damage marks and the 3D recognition result point cloud with deformation marks respectively. For the 2D image, algorithms such as SIFT and SURF are used to extract feature points, and the specific method is not limited. For the 3D point cloud data, algorithms such as FPFH and PFH are used to extract feature descriptors, and the specific method is not limited. These features can reflect the essential information of the data and provide an accurate data basis for subsequent matching and fusion.

[0028] S104. Perform weighted fusion on the damage features and structural features, and determine the defects of the battery pack according to the features after weighted fusion.

[0029] In this embodiment, a dual-channel defect recognition model including a 2D image channel and a 3D point cloud channel is constructed. This model can simultaneously process the 2D image and 3D point cloud data of the battery pack to identify the damage defects and deformation defects of the battery pack, and double verification improves the detection reliability. The dual-channel model is used to identify the 2D image data and 3D point cloud data of the battery pack to obtain 2D recognition results and 3D recognition results. When the 2D recognition results and 3D recognition results meet the preset conditions, the damage features of the 2D recognition results and the structural features of the 3D recognition results are extracted. This process can eliminate the battery packs with serious defects and reduce the computational amount in the recognition process. Perform weighted fusion on the damage features and structural features, and determine the defects of the battery pack according to the features after weighted fusion. Perform dynamic weighted fusion on the damage features and structural features, so that the features after weighted fusion can better reflect the defect characteristics of the battery pack, thereby accurately identifying the state of the battery pack. In this embodiment, the damage features and deformation features of the battery pack are respectively identified by 2D image deep learning and 3D point cloud template matching technology, and the damage features and deformation features are fused and analyzed to identify the state of the battery pack, improving the accuracy and efficiency of battery state detection.

[0030] In some embodiments of the present invention, constructing a dual-channel defect recognition model includes: Taking a 2D image as the input and a damage marked image as the output, constructing a damage recognition model for the 2D image channel; It should be noted that in this embodiment, an improved YOLOv7-tiny model is used to construct a damage recognition model for the 2D image channel by adding light normalization processing at the input end and introducing an attention mechanism at the output end to strengthen the features of the damaged area.

[0031] In some embodiments of the present invention, constructing a damage recognition model for the 2D image channel includes: Based on the YOLOv7-tiny model, taking a 2D image as the input and a damage marked image as the output, constructing an initial damage recognition model; It should be noted that based on the YOLOv7-tiny model, the YOLOv7-tiny model is lightweight and efficient, suitable for real-time monitoring tasks.

[0032] The initial damaged recognition model is optimized based on the Retinex algorithm and the attention mechanism to obtain an optimized damaged recognition model; It should be noted that the Retinex algorithm is introduced into the input layer of the model to preprocess the 2D image data. The brightness and contrast of the image are improved through the Retinex algorithm, especially in the case of uneven illumination. Applying the Retinex algorithm to the input image can improve the visibility of the damaged area, thereby improving the detection accuracy; the attention mechanism is introduced into the output layer of the model to guide the model to focus on the important areas in the image, thereby improving the detection performance, enabling the model to adaptively adjust the weights of the feature maps and highlighting the damaged area.

[0033] Based on the generative adversarial network, the 2D images of the training battery packs are augmented to obtain a training dataset, and the optimized damaged recognition model is iterated according to the training dataset to obtain a damaged recognition model for the 2D image channel.

[0034] It should be noted that the generative adversarial network GAN is used to generate synthetic images similar to the 2D images of real battery packs to augment the training dataset. The generative adversarial network GAN consists of a generator and a discriminator. Through continuous adversarial training, the generator can generate high-quality synthetic images. The generated synthetic images are mixed with the real images to form a richer training dataset. The optimized damaged recognition model is iteratively trained using the augmented training dataset to further improve the generalization ability and detection accuracy of the model. During the iteration process, monitor the performance metrics of the model (such as accuracy, recall, F1 score, etc.) and adjust the hyperparameters as needed.

[0035] In this embodiment, the YOLOv7-tiny model is optimized through the Retinex algorithm and the attention mechanism, and the training model is augmented through the generative adversarial network, enabling the damaged recognition model for the 2D image channel to accurately detect the damaged areas in the 2D images of the battery packs.

[0036] Taking 3D point cloud data as the input and deformed marked point cloud data as the output, a deformation recognition model for the 3D point cloud channel is constructed; In some embodiments of the present invention, the deformation recognition model includes a preprocessing layer, a rough matching layer, a fine registration layer, a deformation index calculation layer, and an output layer; The preprocessing layer is used to filter and denoise the input data of the model; It should be noted that in this embodiment, the methods for filtering and denoising 3D point cloud data include, but are not limited to, using a voxel grid filter and a statistical outlier removal (SOR) filter. The voxel grid filter divides the point cloud data into a three-dimensional voxel grid, where each voxel represents a small cubic space. Then, for each voxel, a representative point (usually the center point or centroid) of all the points within it is selected to replace all the points in that voxel. The voxel size is determined by the density of the point cloud and the required level of detail. In this embodiment, the voxel size is set to 0.5 mm. The statistical outlier removal (SOR) filter calculates the average distance from each point to its multiple nearest neighbor points and evaluates whether the point is an outlier based on this distance. If the average distance of a point is greater than a certain threshold (which is usually calculated based on the average distance and standard deviation of all points), then the point is considered an outlier and is removed.

[0037] The coarse matching layer is used to perform coarse registration of the preprocessed 3D point cloud data with a preset template based on the SAC-IA algorithm. Specifically, a group of points in the 3D point cloud data is randomly selected and matched with the corresponding points in the template. By calculating the transformation matrix (rotation and translation) of the matching point pairs, an initial registration result is obtained. The optimal transformation matrix is selected according to the score of the registration result (such as the number of inliers) to obtain the coarse registration result.

[0038] The fine registration layer is used to perform precise alignment of the 3D point cloud data with the preset template based on the iterative closest point (ICP) algorithm on the basis of the coarse registration. Specifically, based on the iterative closest point (ICP) algorithm, the closest point pairs between the point cloud data and the template are found, the rigid body transformation matrix between these point pairs is calculated, and the position of the point cloud data is updated by applying this matrix. It is checked whether the convergence condition is met (such as the error being less than a certain threshold or reaching the maximum number of iterations). If it is met, the iteration stops, and finally, the precise alignment result is obtained.

[0039] The deformation index calculation layer is used to calculate the deformation amount between the 3D point cloud data and the preset template based on the Hausdorff distance algorithm according to the precise alignment result. Specifically, the distance from each point in the point cloud data to the nearest point on the template, and the distance from each point on the template to the nearest point in the point cloud data are calculated. The maximum value of these distances is taken as the Hausdorff distance, which reflects the maximum degree of mismatch between the point cloud data and the template. The deformation degree is quantified by calculating the average Hausdorff distance.

[0040] The output layer is used to determine the deformation position and deformation degree according to the deformation amount, perform deformation marking on the 3D point cloud data according to the deformation position and deformation degree, and output the deformed marked point cloud data.

[0041] Through the above structural layer, a complete 3D point cloud channel deformation detection model is constructed, which can process 3D point cloud data. Through steps such as preprocessing, rough matching, fine registration, deformation index calculation, and output layer, the deformed marked point cloud data is finally output to accurately identify the deformation defects of the battery pack in the 3D point cloud data.

[0042] The damage recognition model of the 2D image channel and the deformation recognition model of the 3D point cloud channel form a dual-channel defect recognition model.

[0043] In this embodiment, by fusing the damage recognition model of the 2D image channel and the deformation recognition model of the 3D point cloud channel, dual-channel defect detection of the battery pack is realized, and the defect state of the battery pack can be comprehensively and accurately identified.

[0044] In some embodiments of the present invention, as Figure 2 shown, Figure 2 is a schematic flowchart of an embodiment of step S104 provided by the present invention, including: Figure 1 S201. Determine the first weight combination set according to the confidence of the historical 2D recognition result and the matching score of the historical 3D recognition result; It should be noted that in the model training and model verification stages, historical recognition results are collected, and these recognition results include 2D recognition results and the confidence of 2D recognition results as well as 3D recognition results and the matching scores of 3D recognition results. The confidence of 2D recognition results represents the credibility or accuracy of 2D recognition results, and the matching score of 3D recognition results represents the matching degree of 3D recognition results with the true value or other sensor data, etc. Based on these historical data, different weight combinations are tried to find the weight combination that can balance the influence of 2D and 3D recognition results, and these weight combinations constitute the first weight combination set.

[0045] Further, it should be noted that the 2D image recognition result is sensitive to damages such as cracks and scratches on the surface of the battery pack, and surface damages directly affect the battery safety, so higher weights should be given. At the same time, 2D detection is vulnerable to the interference of light and occlusion; the 3D point cloud matching is robust to structural deformations such as bulges and depressions, but has a high computational complexity and is sensitive to minor defects. As an auxiliary verification, the weight should be slightly lower than that of 2D detection.

[0046] S202. Based on the cross-validation method, perform a network search on the first weight combination set, select the optimal weight combination, and obtain the second weight combination; It should be noted that in the first weight combination, the cross-validation method is used to evaluate each weight combination to find the weight combination with the optimal performance. In this embodiment, it is obtained through experimental data that when the weight combination is [0.6, 0.4], the F1 score is 98.7%, and when the weight combination is [0.5, 0.5], the F1 score is 97.2%. This indicates that the performance of the weight combination [0.6, 0.4] is better than that of the weight combination [0.5, 0.5].

[0047] S203. Dynamically optimize the second weight combination according to the real-time environmental data to obtain the target weight combination; It should be noted that different real-time environments of the battery pack will affect the recognition results of the 2D channel and the 3D channel. For example, lighting conditions, occlusion situations, and sensor noise, etc. According to the real-time environmental data, the second weight combination is dynamically adjusted to adapt to the changes in the current environment.

[0048] In some embodiments of the present invention, dynamically optimizing the second weight combination according to the real-time environmental data to obtain the target weight combination includes: based on the sliding window mechanism, dynamically adjusting and optimizing the second weight combination according to the acquired real-time environmental data to obtain the target weight combination. Among them, the adjustment formula for the weight coefficient of the 2D channel in the target weight combination is: , where, is the weight coefficient of the 2D channel in the target weight combination, is the weight coefficient of the 2D channel in the second weight combination, is the lighting change parameter; The adjustment formula for the weight coefficient of the 3D channel in the target weight combination is: , where, is the weight coefficient of the 3D channel in the target weight combination, is the weight coefficient of the 3D channel in the second weight combination, is the point cloud noise level.

[0049] It should be noted that the current lighting brightness and intensity are analyzed through the light sensor, and at the same time, the noise level of the 3D point cloud sensor is analyzed. The sliding window mechanism is used to smooth the real-time data to reduce the influence of data fluctuations on weight adjustment. The size of the sliding window can be adjusted according to specific application requirements.

[0050] S204. Perform weighted fusion on the damage features and the structure features according to the target weight combination.

[0051] In this embodiment, the initial weight combination is first determined through the confidence of the 2D channel and the matching score of the 3D channel, and then this initial weight combination is optimized and dynamically adjusted according to real-time environmental data, enabling the recognition result to be more adaptable to the changing environment and ensuring the accuracy of the recognition result.

[0052] In some embodiments of the present invention, as Figure 3 shown, Figure 3 is a schematic flowchart of an embodiment of the battery pack consistency evaluation provided by the present invention, including: S301. When the defects of the battery pack meet the second preset condition, calculate the remaining capacity of the battery pack according to the obtained battery pack voltage data and temperature data; It should be noted that when the weighted fusion features further identify serious damage and deformation, that is, when they do not meet the second preset condition, the battery pack is continued to be eliminated. For the battery pack that meets the second preset condition, it indicates that the battery pack has almost no damage and deformation that affect the use safety of the battery pack. Then, the next step of detecting the battery pack is required to evaluate the remaining capacity and consistency of the battery pack. The voltage data and temperature data of the battery pack are obtained through a constant current charge and discharge experiment on the battery, and the remaining capacity of the battery pack is calculated based on these voltage data and temperature data.

[0053] S302. When the remaining capacity of the battery pack meets the third preset condition, evaluate the consistency of the battery pack according to the voltage data and electrochemical impedance.

[0054] It should be noted that a preset condition is set for the remaining capacity of the battery pack. When the battery pack capacity exceeds this preset condition, it indicates that the battery pack can be reused, and then the consistency of the battery pack is evaluated, and the whole battery pack is used for cascade utilization according to the battery pack that meets the consistency requirements.

[0055] In some embodiments of the present invention, as Figure 4 shown, Figure 4 is the schematic flowchart of an embodiment of step S302 provided by the present invention, including: Figure 4 S401. Based on the voltage curve dispersion, obtain the first consistency result of the battery pack according to the voltage data in the preset state of charge interval; Specifically, a charge and discharge experiment is carried out on the battery pack, and the voltage change data of each single battery in the battery pack during the charge and discharge process is collected at specific charge and discharge stages, and a voltage change curve is drawn. The more discrete the voltage change curve is, the greater the performance difference of each single battery in the battery pack is. On the contrary, it indicates that the consistency of each single battery in the battery pack is better.

[0056] ​S402. Based on the equivalent circuit model, extract the electrochemical impedance of the battery pack, analyze the mapping relationship between the state of charge and the electrochemical impedance, and evaluate the second consistency result of the battery pack according to the mapping relationship; Specifically, construct an equivalent circuit model of the battery pack, extract the electrochemical impedance of each single battery in the battery pack through this model, analyze the electrochemical impedance of the battery under different states of charge, and analyze the mapping relationship between the state of charge and the electrochemical impedance. If the difference in the electrochemical impedance of each single battery is small, it indicates that the battery pack has good consistency in electrochemical characteristics. Otherwise, it indicates that the battery pack has poor consistency in electrochemical characteristics.

[0057] S403. Determine the consistency of the battery pack according to the first consistency result and the second consistency result.

[0058] Specifically, determine the overall consistency of the battery pack through the two results of the voltage curve dispersion and the electrochemical impedance. If both of these results show good consistency, it is considered that the overall consistency of the battery pack is good, and the battery pack can be reused in a cascade manner. If the overall consistency of the battery pack is poor, corresponding measures need to be further taken for the battery pack to improve the consistency of the battery pack.

[0059] In this embodiment, the consistency of the battery pack is determined by the voltage curve dispersion and the electrochemical impedance, and the performance and safety of the battery pack are ensured through multi-dimensional data analysis.

[0060] To illustrate the method flow of the technical solution in detail, as Figure 5 shown, Figure 5 is a schematic flow chart of battery pack defect detection and consistency evaluation provided by the present invention.

[0061] Specifically, first, detect the battery pack through a dual-channel defect recognition model, initially screen out the battery packs that meet the first preset condition, extract the characteristics of the dual-channel recognition results of the screened battery packs, and after weighted fusion of the characteristics, perform a second defect recognition according to the fused characteristics, and perform a second screening according to the defect recognition results to screen out the battery packs that meet the second preset condition. Calculate the remaining capacity of the battery pack according to the voltage data and temperature data of these battery packs, perform a third screening according to the remaining capacity of the battery pack to screen out the battery packs whose remaining capacity meets the third preset condition, and then evaluate the consistency of these screened battery packs through the voltage curve dispersion and the electrochemical impedance, and perform overall cascade utilization on the battery packs that meet the consistency evaluation conditions.

[0062] To better implement the battery pack defect recognition method in the embodiments of the present invention, correspondingly, on the basis of the battery pack defect recognition method, as Figure 6 shown, the embodiments of the present invention also provide a battery pack defect recognition device 600, including: A model construction module 601 for constructing a dual-channel defect recognition model, where the dual-channel includes a 2D image channel and a 3D point cloud channel; An initial defect detection module 602 for determining a 2D recognition result and a 3D recognition result based on the dual-channel defect recognition model according to the 2D image data and 3D point cloud data of the battery pack; A feature extraction module 603 for extracting the damage features of the 2D recognition result and the structural features of the 3D recognition result when the 2D recognition result and the 3D recognition result meet the first preset condition; A defect recognition module 604 for performing weighted fusion on the damage features and structural features, and recognizing the defects of the battery pack according to the features after weighted fusion.

[0063] The battery pack defect recognition device 600 provided in the above embodiments can implement the technical solutions described in the embodiments of the battery pack defect recognition method. The specific implementation principles of the above modules or units can be referred to the corresponding content in the embodiments of the battery pack defect recognition method, which will not be elaborated here.

[0064] As Figure 7 shown, the present invention also correspondingly provides a battery pack defect recognition device 700. The battery pack defect recognition device 700 includes a processor 701, a memory 702, and a display 703. Figure 7 Only some components of the battery pack defect recognition device 700 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0065] In some embodiments, the processor 701 may be a central processing unit (CPU), a microprocessor, or other data processing chips, for running the program code stored in the memory 702 or processing data, such as the battery pack defect recognition method in the present invention.

[0066] In some embodiments of the present invention, the processor 701 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 701 may be local or remote. In some embodiments, the processor 701 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination of the above.

[0067] The memory 702 can be an internal storage unit of the battery pack defect identification device 700 in some embodiments, such as the hard disk or memory of the battery pack defect identification device 700. The memory 702 can also be an external storage device of the battery pack defect identification device 700 in other embodiments, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the battery pack defect identification device 700.

[0068] Furthermore, the memory 702 can include both the internal storage unit and the external storage device of the battery pack defect identification device 700. The memory 702 is used to store the application software and various types of data for installing the battery pack defect identification device 700.

[0069] The display 703 can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. in some embodiments. The display 703 is used to display the information of the battery pack defect identification device 700 and to display a visual user interface. The components 701 - 703 of the battery pack defect identification device 700 communicate with each other through the system bus.

[0070] In some embodiments of the present invention, when the processor 701 executes the battery pack defect identification program in the memory 702, the following steps can be achieved: Construct a dual-channel defect identification model, where the dual-channel includes a 2D image channel and a 3D point cloud channel; Based on the dual-channel defect identification model, determine the 2D identification result and the 3D identification result according to the 2D image data and the 3D point cloud data of the battery pack; When the 2D identification result and the 3D identification result meet the first preset condition, extract the damage features of the 2D identification result and the structural features of the 3D identification result; Dynamically adjust the preset weight coefficient according to the real-time environment data, and perform weighted fusion on the damage features and the structural features according to the adjusted weight coefficient, and identify the defects of the battery pack according to the features after weighted fusion.

[0071] It should be understood that when the processor 701 executes the battery pack defect identification program in the memory 702, in addition to the above functions, other functions can also be achieved. For details, please refer to the description of the corresponding method embodiments above.

[0072] Furthermore, the embodiments of the present invention do not specifically limit the type of the battery pack defect identification device 700 mentioned. The battery pack defect identification device 700 can be a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, or other portable battery pack defect identification devices. Exemplary embodiments of the portable battery pack defect identification device include, but are not limited to, portable battery pack defect identification devices running IOS, Android, Microsoft, or other operating systems. The above-mentioned portable battery pack defect identification device can also be other portable battery pack defect identification devices. It should also be understood that in some other embodiments of the present invention, the battery pack defect identification device 700 may not be a portable battery pack defect identification device, but a desktop computer with a touch-sensitive surface (such as a touch panel).

[0073] Correspondingly, the embodiments of the present invention also provide a computer-readable storage medium. The computer-readable storage medium is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the battery pack defect identification method provided by the above-mentioned various method embodiments can be implemented.

[0074] Those skilled in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disk, a read-only memory, or a random access memory, etc.

[0075] The above has introduced in detail the battery pack defect identification method, device, and equipment provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A battery pack defect identification method, characterized in that: include: Constructing a dual-channel defect recognition model, wherein the dual channels include a 2D image channel and a 3D point cloud channel; Based on the dual-channel defect recognition model, determine a 2D recognition result and a 3D recognition result according to the 2D image data and the 3D point cloud data of the battery pack; When the 2D recognition result and the 3D recognition result meet a first preset condition, extracting a damage feature of the 2D recognition result and a structural feature of the 3D recognition result; The damage features and the structural features are weightedly fused, and defects of the battery pack are identified based on the weighted fused features.

2. The battery pack defect identification method according to claim 1, characterized in that: Build a dual-channel defect recognition model, including: Taking 2D image as input and damage mark image as output, a damage recognition model of 2D image channel is constructed; Taking 3D point cloud data as input and deformation marked point cloud data as output, a deformation recognition model of 3D point cloud channel is constructed; The damage recognition model of the 2D image channel and the deformation recognition model of the 3D point cloud channel constitute a dual-channel defect recognition model.

3. The battery pack defect identification method according to claim 2, characterized in that: Build a damage recognition model for 2D image channels, including: Based on the YOLOv7-tiny model, the initial damage recognition model is constructed with 2D images as input and damage mark images as output; The initial damage recognition model is optimized based on the Retinex algorithm and the attention mechanism to obtain an optimized damage recognition model; Based on the generative adversarial network, the 2D image of the training battery pack is expanded to obtain a training data set, and the optimized damage recognition model is iterated according to the training data set to obtain a damage recognition model of the 2D image channel.

4. The battery pack defect identification method according to claim 2, characterized in that: The deformation recognition model includes a preprocessing layer, a coarse matching layer, a fine registration layer, a deformation index calculation layer and an output layer; The preprocessing layer is used to filter and denoise the input data of the model; The coarse matching layer is used to coarsely align the preprocessed 3D point cloud data with the preset template based on the SAC-IA algorithm; The fine registration layer is used to accurately align the 3D point cloud data with the preset template based on the rough registration based on the iterative closest point algorithm; The deformation index calculation layer is used to calculate the deformation between the 3D point cloud data and the preset template based on the Hausdorff distance algorithm according to the result of precise alignment; The output layer is used to determine the deformation position and deformation degree according to the deformation amount, mark the 3D point cloud data according to the deformation position and deformation degree, and output the deformation marked point cloud data.

5. The battery pack defect identification method according to claim 3, characterized in that: The damage features and the structural features are weightedly fused, including: Determining a first weight combination set according to the confidence of the historical 2D recognition results and the matching score of the historical 3D recognition results; Based on the cross-validation method, a network search is performed on the first weight combination set, and the optimal weight combination is selected to obtain the second weight combination; Dynamically optimize the second weight combination according to the real-time environmental data to obtain a target weight combination; The damage features and the structural features are weightedly fused according to the target weight combination.

6. The battery pack defect identification method according to claim 5, characterized in that: Dynamically optimizing the second weight combination according to the real-time environmental data to obtain a target weight combination includes: Based on the sliding window mechanism, the second weight combination is dynamically adjusted and optimized according to the acquired real-time environmental data to obtain the target weight combination, where the weight coefficient adjustment formula of the 2D channel in the target weight combination is: ， in, is the weight coefficient of the 2D channel in the target weight combination, is the weight coefficient of the 2D channel in the second weight combination, is the illumination variation parameter; The weight coefficient adjustment formula of the 3D channel in the target weight combination is: ， in, is the weight coefficient of the 3D channel in the target weight combination, is the weight coefficient of the 3D channel in the second weight combination, is the point cloud noise level.

7. The battery pack defect identification method according to claim 1, characterized in that: After the defect detection of the battery pack, it also includes: When the defect of the battery pack meets a second preset condition, calculating the remaining capacity of the battery pack according to the acquired voltage data and temperature data of the battery pack; When the remaining capacity of the battery pack meets the third preset condition, the consistency of the battery pack is evaluated based on the voltage data and the electrochemical impedance.

8. The battery pack defect identification method according to claim 7, characterized in that: When the remaining capacity of the battery pack meets the third preset condition, the consistency of the battery pack is evaluated according to the voltage data and the electrochemical impedance, including: Based on the voltage curve dispersion, a first consistency result of the battery pack is obtained according to the voltage data of a preset state of charge interval; Extracting the electrochemical impedance of the battery pack based on the equivalent circuit model, analyzing the mapping relationship between the state of charge and the electrochemical impedance, and evaluating the second consistency result of the battery pack according to the mapping relationship; The consistency of the battery pack is determined according to the first consistency result and the second consistency result.

9. A battery pack defect identification device, characterized in that: include: A model building module, used to build a dual-channel defect recognition model, wherein the dual channels include a 2D image channel and a 3D point cloud channel; An initial defect detection module, used to determine a 2D recognition result and a 3D recognition result according to the 2D image data and the 3D point cloud data of the battery pack based on the dual-channel defect recognition model; A feature extraction module, configured to extract a damage feature of the 2D recognition result and a structural feature of the 3D recognition result when the 2D recognition result and the 3D recognition result meet a first preset condition; The defect recognition module is used to perform weighted fusion on the damage features and structural features, and identify defects of the battery pack according to the weighted fusion features.

10. A battery pack defect identification device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the battery pack defect identification method described in any one of claims 1 to 8.

Images