Lithium battery valuable metal detection system and method

Abstract

The application relates to the technical field of battery detection, and provides a lithium battery valuable metal detection system and method. A sensing module is used to accurately acquire a surface topography image of a lithium battery and perform region segmentation, the surface roughness of each region is calculated to adjust laser pulse energy, and the detection requirement is adapted; an excitation module determines the spatial distribution coordinates of each sampling point according to the corresponding sampling points of different regions, and takes into account the detection accuracy and detection efficiency; a detection module performs matrix correction processing on spectral data, the chemical component content of all sampling points is arranged into a component sequence in spatial order, the component sequence is processed by using a long short-term memory network, and a detection result is output; a control module generates a valuable metal content report and links to match recovery process parameters, when the number of abnormal sampling points is greater than a preset number threshold, only the resampling and detection of the corresponding abnormal region are triggered, the abnormal processing efficiency is greatly improved, and the detection cost and time cost are reduced.

Description

Technical Field

[0001] This application relates to the field of battery testing technology, specifically to a system and method for detecting valuable metals in lithium batteries. Background Technology

[0002] In the process of recycling and reusing lithium batteries, accurate detection of valuable metals is a prerequisite for optimizing recycling processes and making efficient use of resources. At present, technologies such as laser-induced breakdown spectroscopy and X-ray fluorescence spectroscopy are mainly used to extract the content of valuable metal components by exciting the lithium battery electrodes to generate spectral signals.

[0003] The core logic of existing detection systems is to acquire information about the electrode surface and perform laser excitation on the electrode, process the spectral data and output the component content, ultimately providing a reference for the recycling process. In recent years, the industry has gradually pursued improvements in detection accuracy and efficiency, and has attempted to improve detection results by optimizing spectral processing algorithms and adjusting laser excitation parameters, so as to adapt to the detection needs of different types of lithium batteries and reduce secondary damage to the electrode during the detection process.

[0004] Despite considering practical application scenarios and testing requirements, several shortcomings remain to be addressed. Existing testing systems can only simply distinguish between the positive and negative electrode regions of lithium batteries, failing to accurately identify damaged areas. This results in poor laser excitation effects, affecting the authenticity and validity of spectral data. Furthermore, the use of uniformly distributed sampling points without considering the testing priority and structural characteristics of the positive and negative electrodes and damaged areas impacts testing accuracy and efficiency. Moreover, the interference of the lithium battery electrode substrate on the characteristic spectral lines of valuable metals is not fully considered, and the spatial correlation characteristics of sampling points are not explored, leading to significant deviations in the calculation of valuable metal content and inaccurate identification of abnormal sampling points.

[0005] Based on the shortcomings of the existing technology, the technical problem to be solved by this application is how to achieve accurate detection of valuable metals by comprehensively identifying lithium batteries and matching lasers to acquire data from different sampling points. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application provides a system and method for detecting valuable metals in lithium batteries.

[0007] In the first aspect, this application provides a lithium battery valuable metal detection system, which includes: a sensing module, an excitation module, a detection module and a control module; The sensing module is used to acquire surface morphology images of the lithium battery and perform region segmentation to identify the positive electrode region, negative electrode region and damaged region, and calculate the surface roughness of the lithium battery to adjust the laser pulse energy; The excitation module is used to set basic sampling points in the positive electrode region, sparse sampling points in the damaged region, and reference sampling points in the negative electrode region to determine the spatial distribution coordinates of different sampling points, and to excite the lithium battery point by point according to the laser pulse energy to obtain the spectral data of different sampling points of the lithium battery. The detection module is used to perform matrix correction processing on spectral data, extract the chemical composition content of different sampling points, construct a composition sequence by spatially ordering the chemical composition content of all sampling points on the same lithium battery, process the composition sequence through a long short-term memory network, and output the valuable metal content and abnormal sampling points of the lithium battery. The control module is used to generate a report on the content of valuable metals and link the recycling process parameters. When the number of abnormal sampling points exceeds the preset threshold, it triggers resampling and detection of the corresponding area.

[0008] As an optional implementation, the execution of region segmentation includes: Wavelet decomposition is performed on the surface morphology image of the lithium battery to obtain low-frequency and high-frequency components. The undulation gradient features and texture distribution features of the surface morphology image are extracted respectively and fused to form a region discrimination vector. Image pixels are clustered and divided into candidate regions based on region discrimination vectors. Based on the prior structural relationship of lithium battery electrodes, spatial topological constraints are applied to all candidate regions to obtain pre-partitions. Optimize all pre-partitions based on the mutual exclusion constraints between regions, and perform fluctuation state verification on the region discrimination vector of the optimized pre-partitions. If the verification passes, output the positive region, negative region and damaged region; otherwise, re-execute feature extraction and re-execute region segmentation.

[0009] As an optional implementation, the spatial topology constraints include the aspect ratio of the lithium battery electrode, the area ratio of the region, and the relative positional relationship between the positive and negative electrodes, eliminating isolated candidate regions and correcting the corresponding region boundaries; the mutual exclusion constraints represent that the positive and negative electrode regions are mutually exclusive and adjacent, and the damaged region is superimposed on the positive or negative electrode region; the fluctuation state of the region discrimination vector is the variance fluctuation value of the undulating gradient feature and texture distribution feature within each pre-partition, and the verification is passed when the variance fluctuation value is not greater than the fluctuation threshold.

[0010] As an optional implementation, adjusting the laser pulse energy includes: Calculate the mean and gradient fluctuation values ​​of the surface roughness within each region, and assign a first weight, a second weight, and a third weight to the positive region, the damaged region, and the negative region, respectively, with the first weight being greater than the third weight, which is greater than the second weight. Based on the weighted fusion of global roughness and its gradient fluctuation value, the initial pulse energy of the corresponding region is generated, and the energy upper and lower limit constraints and the matching of the electrode excitation depth of the initial pulse energy are verified. If the verification passes, the initial pulse energy is output as the laser pulse energy. If the verification fails, the first weight, the second weight, and the third weight are iteratively adjusted and the initial pulse energy is regenerated until the verification passes.

[0011] As an optional implementation, calculating the mean and gradient fluctuation values ​​of the surface roughness within each region includes: The high-frequency components in the surface topography image are reconstructed to obtain the topography height values ​​of different pixels. Taking each pixel in different regions as the center, a preset neighborhood window is selected, and the arithmetic mean deviation of the topography height values ​​of all pixels in the neighborhood window is calculated to obtain the surface roughness of a single pixel. The surface roughness of all single pixels in each region is weighted and averaged to obtain the mean of the surface roughness of the corresponding region. The continuous rate of change of the surface roughness of single pixels in the region is calculated, and the root mean square error of the continuous rate of change is used as the gradient fluctuation value of the surface roughness.

[0012] As an optional implementation, exciting the lithium battery point-by-point according to the laser pulse energy includes: Based on the spatial distribution coordinates of the basic sampling points corresponding to the positive electrode region, the sparse sampling points corresponding to the damaged region, and the reference sampling points corresponding to the negative electrode region, the laser pulse energy of the corresponding regions is matched respectively. Excitation of the reference sampling point is performed according to the matching laser pulse energy to obtain the reference excitation reference, so as to calibrate the laser pulse energy of the corresponding region. Then, according to the adjacency relationship of the same region in the spatial distribution coordinates, the basic sampling point and the sparse sampling point are excited one by one in sequence. The system acquires the original spectral intensity of different sampling points in real time, determines whether it meets the preset signal threshold, and saves the spectral data of the corresponding sampling point if it does not meet the threshold. The system then re-excites the corresponding sampling point until the preset signal threshold is met.

[0013] As an optional implementation, the extraction of chemical composition content at different sampling points includes: Baseline correction and noise filtering are performed sequentially on the spectral data from different sampling points, and the matrix correction coefficients of the lithium battery are constructed based on the processed spectral data corresponding to the reference sampling points. Intensity compensation is performed point-by-point for the corresponding bands of the spectral data after processing the base sampling points and sparse sampling points, based on the matrix correction coefficient. The peak positions of the characteristic spectral lines corresponding to the valuable metals are located in the intensity-compensated spectral data. Peak area integration is performed on the characteristic spectral lines to obtain the peak intensity. The chemical composition content of the corresponding sampling points is extracted by combining the pre-stored quantitative standard curve of the composition.

[0014] As an optional implementation, the valuable metal content and abnormal sampling points of the output lithium battery include: According to the spatial order of the spatial distribution coordinates corresponding to different sampling points, the chemical composition content of all sampling points on the same lithium battery is sorted in sequence to form a composition sequence; Feature extraction and data fitting of component sequences are performed using a long short-term memory network. The fitting result is the time-series fusion value of the chemical component content corresponding to the sampling point. Combined with the pre-stored component quantitative standard curve, the time-series fusion value is mapped to the valuable metal content of lithium battery. By comparing the chemical component content of each sampling point in the component sequence with the average chemical component content of sampling points in the same region, abnormal sampling points are identified.

[0015] As an optional implementation, the triggering of resampling and detection of the corresponding area includes: if the number of abnormal sampling points is greater than a preset number threshold, locating the target area to which the abnormal sampling points belong, and supplementing the target area with corresponding sampling points, and re-exciting and detecting the valuable metal content of the lithium battery point by point according to the laser pulse energy, until the number of abnormal sampling points is not greater than the preset number threshold.

[0016] Secondly, this application provides a method for detecting valuable metals in lithium batteries. The method includes: acquiring a surface morphology image of the lithium battery and performing region segmentation to identify the positive electrode region, the negative electrode region and the damaged region, and calculating the surface roughness of the lithium battery to adjust the laser pulse energy. Basic sampling points were set in the positive electrode region, sparse sampling points were set in the damaged region, and reference sampling points were set in the negative electrode region to determine the spatial distribution coordinates of different sampling points. The lithium battery was excited point by point according to the laser pulse energy to obtain the spectral data of different sampling points of the lithium battery. Matrix correction is performed on the spectral data, the chemical composition content of different sampling points is extracted, the chemical composition content of all sampling points on the same lithium battery is arranged in spatial order to form a composition sequence, the composition sequence is processed by a long short-term memory network, and the valuable metal content and abnormal sampling points of the lithium battery are output. Generate a report on the content of valuable metals and link it to the recycling process parameters. When the number of abnormal sampling points exceeds the preset threshold, trigger resampling and detection of the corresponding area.

[0017] Compared with the prior art, the beneficial effects of this application are: by accurately acquiring the surface morphology image of the lithium battery through the sensing module and performing region segmentation, the positive electrode region, negative electrode region and damaged region are clearly identified, and the surface roughness of each region is calculated simultaneously to adjust the laser pulse energy, so that the laser pulse energy is adapted to the structural characteristics and detection requirements of different regions, and the spectral data truly reflects the characteristic information of valuable metals.

[0018] The excitation module sets corresponding sampling points according to different regions to determine the spatial distribution coordinates of each sampling point, ensuring sufficient sampling in the core area while balancing detection accuracy and efficiency. The detection module performs matrix correction processing on the spectral data to counteract the interference of the lithium battery electrode matrix on the characteristic spectral lines of valuable metals. It constructs a component sequence by spatially ordering the chemical composition content of all sampling points and processes the component sequence through a long short-term memory network to mine the spatial correlation features of the sampling points, thereby improving the calculation accuracy of the overall valuable metal content of the lithium battery. At the same time, it outputs abnormal sampling points to provide a basis for subsequent anomaly handling.

[0019] The control module integrates and generates a valuable metal content report and links it to match the corresponding recycling process parameters, so that the test results can directly guide the recycling process. It connects the data flow between testing and recycling, avoids the deviation caused by manual judgment of process parameters, and triggers resampling and testing only in the corresponding abnormal area when the number of abnormal sampling points exceeds the preset threshold, rather than global resampling. This greatly improves the efficiency of abnormal handling, reduces testing and time costs, and is suitable for the actual application scenarios of lithium battery recycling.

[0020] This application forms a closed loop for the detection of valuable metals in lithium batteries, from surface morphology perception, region segmentation, and laser energy adjustment, to sampling excitation, spectral processing, and component extraction, and then to report generation, recycling linkage, and anomaly resampling, ensuring the stability of system operation; at the same time, each module can flexibly adjust parameters according to the detection requirements of different types of lithium batteries, with strong scalability and adaptability to the detection needs of different scenarios. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a system flowchart of the lithium battery valuable metal detection system provided in the embodiments of this application; Figure 2 The flowchart for calculating the mean and gradient fluctuation values ​​of the surface roughness within each region is provided in the embodiments of this application. Figure 3 This is a flowchart of a method for detecting valuable metals in lithium batteries provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application more apparent and understandable, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0023] Example 1: like Figure 1 The diagram shown is a system flowchart of a lithium battery valuable metal detection system according to an embodiment of this application. The system includes a sensing module, an excitation module, a detection module, and a control module.

[0024] The sensing module is used to acquire surface topography images of the lithium battery and perform region segmentation to identify the positive electrode region, negative electrode region and damaged region, and calculate the surface roughness of the lithium battery to adjust the laser pulse energy.

[0025] Furthermore, performing region segmentation includes: Wavelet decomposition is performed on the surface morphology image of the lithium battery to obtain low-frequency and high-frequency components. The undulation gradient features and texture distribution features of the surface morphology image are extracted respectively and fused to form a region discrimination vector. Image pixels are clustered and divided into candidate regions based on region discrimination vectors. Based on the prior structural relationship of lithium battery electrodes, spatial topological constraints are applied to all candidate regions to obtain pre-partitions. Optimize all pre-partitions based on the mutual exclusion constraints between regions, and perform fluctuation state verification on the region discrimination vector of the optimized pre-partitions. If the verification passes, output the positive region, negative region and damaged region; otherwise, re-execute feature extraction and re-execute region segmentation.

[0026] Among them, the spatial topology constraints include the aspect ratio of lithium battery electrode sheets, the area ratio of the region, and the relative positional relationship between the positive and negative electrodes, so as to eliminate isolated candidate regions and correct the corresponding region boundaries; the mutual exclusion constraints represent that the positive and negative electrode regions are mutually exclusive and adjacent, and the damaged region is superimposed on the positive or negative electrode region, and the region overlap and boundary blurring are eliminated through masking operation; the fluctuation state of the region discrimination vector is the variance fluctuation value of the undulating gradient features and texture distribution features within each pre-partition, and the verification is passed when the variance fluctuation value is not greater than the fluctuation threshold.

[0027] The prerequisite for performing region segmentation is to obtain feature parameters that can effectively distinguish between positive electrode region, negative electrode region and damaged region. The surface morphology of lithium battery electrode sheet carries two types of information: overall structural undulation and local texture details. These two types of information are superimposed. If features are extracted directly, it will lead to feature confusion and make it impossible to accurately distinguish between the three types of regions. Positive electrode region, negative electrode region and damaged region have inherent differences in overall contour undulation and local texture distribution. Overall contour undulation features reflect the macro structure of the region, while local texture features reflect the micro state of the region. Relying on a single type of feature cannot fully characterize the morphological differences of different regions.

[0028] The acquired lithium battery surface morphology image is processed by wavelet decomposition. Wavelet decomposition divides the image into low-frequency and high-frequency components. The low-frequency component corresponds to the low-frequency information of the image, representing the overall contour and macroscopic undulation of the lithium battery electrode surface, and is not affected by local minor defects. The high-frequency component corresponds to the high-frequency information of the image, representing the local texture, minor defects and roughness distribution of the lithium battery electrode surface, accurately capturing the microscopic differences between the damaged area and the positive and negative electrode areas.

[0029] The wavelet basis is selected to adapt to the features of the lithium battery surface morphology image. The db4 wavelet basis is used for two-level decomposition. The db4 wavelet basis has good smoothness and tight support, effectively preserving the contour features and texture details of the image and avoiding feature distortion during the decomposition process. After decomposition, the undulation gradient features are extracted from the low-frequency components. The Sobel operator is used to calculate the gradient value of each pixel in the window with a 3×3 pixel calculation window. The average gradient value is taken as the undulation gradient feature of the window. The dimension of the undulation gradient feature is determined according to the image resolution. For a 512×512 pixel surface morphology image, a 3D undulation gradient feature is extracted.

[0030] Texture distribution features are extracted from high-frequency components. The gray-level co-occurrence matrix method is used to calculate the contrast, entropy, and correlation of the gray-level co-occurrence matrix as texture distribution features, forming a 3D texture distribution feature. Feature fusion is performed by dimensional concatenation. The extracted undulation gradient features and texture distribution features are concatenated in sequence to form a region discrimination vector. For example, the 3D undulation gradient features and 3D texture distribution features are concatenated to form a 6D region discrimination vector. Each pixel corresponds to a 6D region discrimination vector. Each dimension in the vector corresponds to a different morphological feature parameter, which comprehensively represents the morphological attributes of the region to which the pixel belongs.

[0031] This enables the effective separation of the overall structural features and local detail features of the lithium battery surface morphology image, avoiding the interference between the two types of features and the resulting feature distortion. The region discrimination vector comprehensively and accurately represents the morphological features of each pixel, providing feature support for subsequent pixel clustering and ensuring the accuracy of clustering.

[0032] The region discrimination vector clusters the image pixels, which can initially obtain the pixel sets corresponding to the positive electrode region, negative electrode region, and damaged region. However, pixel division based solely on the clustering algorithm does not consider the actual physical structure of the lithium battery electrode sheet, resulting in a large number of isolated and distorted candidate regions that do not conform to the actual structure of the electrode sheet. Such candidate regions have no practical detection significance. If they are directly used for subsequent processing, it will increase the workload of subsequent region optimization and cannot guarantee the accuracy of region segmentation. At the same time, lithium battery electrode sheets have inherent prior structural relationships. The aspect ratio, area ratio, and relative position of the positive and negative electrode regions are all subject to clear physical constraints, and the distribution of damaged regions also follows specific rules.

[0033] Based on region discriminant vectors, the K-means clustering algorithm is used to classify image pixels. Three clusters are preset, corresponding to pixel sets in positive, negative, and damaged regions, respectively. During cluster initialization, the region discriminant vectors of three different pixels are randomly selected as initial cluster centers. The number of iterations is set to 50, and the iteration terminates when the change in cluster centers is less than 10. -3 This ensures the stability of the clustering results; after clustering, candidate regions are obtained, each candidate region corresponds to a set of pixels, initially corresponding to positive, negative or damaged regions.

[0034] Based on the prior structural relationships of lithium battery electrodes, spatial topological constraints are applied for screening. These constraints include three types: aspect ratio, area ratio, and relative positional relationship between the positive and negative electrodes. The aspect ratio constraint is set according to the conventional physical dimensions of lithium battery electrodes, which typically range from 3:1 to 6:1. For candidate regions obtained through clustering, the aspect ratio of their circumscribed rectangle is calculated. If the aspect ratio exceeds this range, it is considered a deformed candidate region and is eliminated. The area ratio constraint is set according to the conventional distribution ratio of the positive and negative electrode regions on the electrode. The positive electrode region typically accounts for 40% to 60% of the total electrode area, the negative electrode region typically accounts for 30% to 50%, and the damaged region typically does not exceed 10% of the total electrode area. The area ratio of each candidate region to the total electrode area is calculated, and candidate regions whose area ratio does not conform to the above ranges are eliminated.

[0035] The relative positional constraints of the positive and negative electrodes are set according to the structural design of the lithium battery electrode sheet. The positive and negative electrodes are distributed adjacently along the length of the electrode sheet, without intersection or separation. For candidate areas that are initially determined to be positive and negative electrodes, their relative positional relationship is checked, and isolated candidate areas that deviate from the positional relationship are eliminated. After screening, the boundaries of the remaining candidate areas are corrected. Morphological expansion and etching operations are used to eliminate burrs and depressions at the boundaries of the candidate areas, making the boundaries of the candidate areas smoother and more regular, and finally obtaining pre-partitions that conform to the actual physical structure of the electrode sheet.

[0036] This enables the initial classification of pixels. Then, invalid candidate regions that do not conform to the actual structure of the electrode are eliminated through spatial topological constraints. This makes the pre-partitioning fit the actual physical structure of the lithium battery electrode, improves the rationality of the region segmentation, reduces the workload of subsequent region optimization, and ensures the final accuracy of the region segmentation.

[0037] Although the pre-partitions obtained by spatial topological constraint screening already conform to the basic physical structure of lithium battery electrodes, there may still be problems such as regional overlap and blurred boundaries. This is mainly because during the clustering process, the regional discrimination vector features of some pixels are similar, resulting in unclear cluster boundaries and thus regional overlap. At the same time, if the pixel features within the pre-partitions fluctuate greatly, it will lead to uneven morphological features within the region, making it impossible to accurately represent the attributes of a single region and affecting the accuracy of subsequent steps.

[0038] The pre-partitioning is optimized based on the mutual exclusion constraints between regions. The mutual exclusion constraints clearly define the spatial relationship between the positive electrode region, the negative electrode region, and the damaged region. That is, the positive electrode region and the negative electrode region are mutually exclusive and adjacent, with no overlapping areas, and are distributed adjacently along the length of the electrode sheet. The damaged region does not exist independently, but is superimposed on the positive electrode region or the negative electrode region, and partially overlaps with the positive electrode region or the negative electrode region. However, it cannot cross both the positive electrode region and the negative electrode region at the same time. That is, the spatial range of the damaged region only covers the single electrode region to which it is attached. To achieve the above constraints, masking operations are used for optimization. A corresponding binary mask is generated for each pre-partition. In the binary mask, the pixel value corresponding to the pre-partition is set to 1, and the pixel value corresponding to the non-pre-partition is set to 0.

[0039] For the positive and negative electrode pre-partitions, a bitwise AND operation is performed on their binary masks to obtain the mask of the overlapping region. Pixels in the overlapping region are then redistributed to pre-partitions with better feature matching to eliminate the overlap between the positive and negative electrode regions. For the damaged pre-partition, a bitwise AND operation is performed with the binary masks of the positive and negative electrode pre-partitions respectively to determine the overlapping region between the damaged pre-partition and the positive and negative electrode pre-partitions. The overlapping region is retained, and isolated damaged pixels that do not overlap with any electrode region are removed to ensure that the damaged region is superimposed on the positive and negative electrode regions. For regions with blurred boundaries, the similarity between the boundary pixels and the region discrimination vectors of adjacent pre-partitions is calculated. The boundary pixels are then redistributed to pre-partitions with higher similarity to correct the boundary blurring problem.

[0040] After optimization, the fluctuation state of the region discrimination vector is checked for each pre-partition. The fluctuation state of the region discrimination vector is the variance fluctuation value of the region discrimination vector of all pixels within each pre-partition. The variance fluctuation value is calculated as follows: calculate the variance of each dimension of the region discrimination vector of all pixels within the pre-partition, and take the mean of the variances of each dimension as the variance fluctuation value of the pre-partition. The variance fluctuation value represents the uniformity of the features within the pre-partition. The smaller the variance fluctuation value, the more uniform the features within the pre-partition and the more accurate the region division. The preset fluctuation threshold is 10% of the mean of the region discrimination vector, which is determined according to the feature distribution of the region discrimination vector to ensure that pre-partitions with uniform features can be effectively selected.

[0041] If the variance fluctuation value of the pre-partition is not greater than the fluctuation threshold, the verification is deemed successful and the pre-partition is a valid region with uniform features. If the variance fluctuation value of the pre-partition is greater than the fluctuation threshold, the features within the pre-partition are deemed to be non-uniform and the region division is biased. The feature extraction and region segmentation process is then re-executed until a verified pre-partition is obtained. After verification, the optimized pre-partition is identified as a positive region, a negative region, and a damaged region, and the coordinate range and boundary contour of the region are output, thus completing the entire region segmentation process.

[0042] This eliminates the problems of overlapping and blurred boundaries in the pre-partitioned areas, ensuring that the spatial relationship of the areas meets the actual needs; it ensures the uniformity and consistency of the morphological features within each area, achieving accurate and clear division of the positive, negative, and damaged areas, meeting the requirements of subsequent inspection processes for area accuracy; the area result is the prerequisite for spatial positioning in the subsequent processes of the entire inspection system. If there is a deviation in the area segmentation, it will lead to distortion in subsequent surface roughness calculations, unreasonable adjustment of laser pulse energy, and deviation of sampling points from the target area, thereby affecting the inspection accuracy of the entire inspection system.

[0043] Specifically, adjusting the laser pulse energy includes: Calculate the mean and gradient fluctuation values ​​of the surface roughness within each region, and assign a first weight, a second weight, and a third weight to the positive region, the damaged region, and the negative region, respectively, with the first weight being greater than the third weight, which is greater than the second weight. Based on the weighted fusion of global roughness and its gradient fluctuation value, the initial pulse energy of the corresponding region is generated, and the energy upper and lower limit constraints and the matching of the electrode excitation depth of the initial pulse energy are verified. If the verification passes, the initial pulse energy is output as the laser pulse energy. If the verification fails, the first weight, the second weight, and the third weight are iteratively adjusted and the initial pulse energy is regenerated until the verification passes.

[0044] Among them, such as Figure 2 As shown, the calculation of the mean and gradient fluctuation values ​​of the surface roughness within each region includes: The high-frequency components in the surface topography image are reconstructed to obtain the topography height values ​​of different pixels. Taking each pixel in different regions as the center, a preset neighborhood window is selected, and the arithmetic mean deviation of the topography height values ​​of all pixels in the neighborhood window is calculated to obtain the surface roughness of a single pixel. The surface roughness of all single pixels in each region is weighted and averaged to obtain the mean of the surface roughness of the corresponding region. The continuous rate of change of the surface roughness of single pixels in the region is calculated, and the root mean square error of the continuous rate of change is used as the gradient fluctuation value of the surface roughness.

[0045] The calculation of surface roughness is based on the microscopic height variation data of the lithium battery electrode surface. However, the high-frequency components of the surface morphology image obtained through the region segmentation process are only frequency domain feature data and cannot be used for the quantitative characterization of microscopic morphology. A two-dimensional discrete wavelet inverse reconstruction method matching the previous wavelet decomposition is adopted to process the high-frequency components of the surface morphology image retained after region segmentation. The reconstruction process uses the db4 wavelet basis and the reconstruction parameters of the 2-level decomposition. For example, taking a lithium battery electrode surface morphology image with a resolution of 512×512 pixels as the processing object, after two-dimensional discrete wavelet inverse reconstruction, each pixel in the image corresponds to a morphology height value that characterizes the degree of microscopic unevenness. This height value is a relative quantized value, and the value distribution range is set to 0 to 255. For example, the morphology height value of a certain pixel in the positive electrode region after reconstruction is 186, the height value of a certain pixel in the negative electrode region is 122, and the height value of a certain pixel in the damaged region is 213. The difference in pixel height values ​​in different regions can directly reflect the basic state of local microscopic undulations.

[0046] This enables the conversion of frequency domain features to spatial domain height data, retains detailed information on the microstructure of the electrode surface, eliminates the interference of macroscopic contours on microscopic features, and provides data for surface roughness calculation.

[0047] The morphological height value of a single pixel can only reflect the microscopic height of a single point and cannot characterize the surface undulation of the surrounding local area. The core definition of surface roughness is the microscopic undulation deviation of the local surface. The positive, negative and damaged areas determined after region segmentation are used as the boundaries to limit the range. Within each region, a neighborhood window of a preset size is selected with each pixel as the center to carry out statistical calculations. The preset neighborhood window is a 3×3 pixel rectangular window. This window size can cover the local microscopic undulation range around a single pixel, which meets the basic requirements for local characterization of surface roughness.

[0048] The arithmetic mean of the height values ​​of all pixels within the neighborhood window is calculated. Then, the absolute deviation between the height value of the center pixel and this arithmetic mean is calculated, which is the arithmetic mean deviation. This absolute deviation is used as the surface roughness of the corresponding center pixel. The larger the value, the more severe the local micro-ripples at that location, and the higher the roughness. For example, taking a center pixel in the positive polarity region as an example, the height values ​​of the nine pixels in its 3×3 neighborhood window are 186, 179, 183, 181, 185, 178, 180, 182, and 177, respectively. First, the arithmetic mean of this set of values ​​is calculated as 181. Then, the absolute deviation between the height value of the center pixel (186) and the average value (181) is calculated, resulting in a surface roughness of 5 for the single pixel. Taking a center pixel in the damaged region as another example, the height values ​​in its neighborhood window fluctuate greatly. The calculated surface roughness of the single pixel is 12, which can intuitively reflect the more severe local micro-ripples in the damaged region.

[0049] This transforms single-point height data into local micro-undulation characterization data, enabling precise quantification of single-pixel surface roughness, eliminating the random error of single-point pixel height, and allowing the roughness parameter to truly reflect the local micro-state of the electrode surface.

[0050] The surface roughness of a single pixel is discrete microscopic data and cannot be directly used for regional adjustment of laser pulse energy. The degree of continuous change in roughness within a region will affect the stability of laser excitation. The surface roughness of all single pixels within a single region is calculated by weighting the pixels according to the uniformity of their distribution within the region. The weight of the central pixel is slightly higher than that of the edge pixels. The value obtained after weighting the average is the mean of the surface roughness of the region. This mean reflects the average roughness of the electrode surface within the region.

[0051] For example, the positive electrode region contains 1200 effective single-pixel roughness data points, and after weighted averaging, the average surface roughness of the positive electrode region is 6; the negative electrode region contains 950 effective single-pixel roughness data points, and the average roughness after weighted averaging is 4; the damaged region contains 320 effective single-pixel roughness data points, and the average roughness after weighted averaging is 10.

[0052] The continuous rate of change of surface roughness of a single pixel is calculated. The continuous rate of change is the amplitude of the change in surface roughness between adjacent pixels, which is used to characterize the spatial rate of change of roughness. After obtaining the continuous rate of change of all adjacent pixels in the region, the mean square error of all continuous rate of change is calculated. This mean square error is used as the gradient fluctuation value of the surface roughness of the corresponding region. The larger the gradient fluctuation value, the more drastic the spatial change of roughness in the region. For example, after the mean square error calculation, the gradient fluctuation value of the set of continuous rate of change of roughness of adjacent pixels in the positive region is 1.2; the gradient fluctuation value corresponding to the negative region is 0.8; and the gradient fluctuation value corresponding to the damaged region is 2.5. The difference in gradient fluctuation values ​​among the three types of regions reflects the spatial change characteristics of roughness within each region.

[0053] This enables the conversion of micro-roughness characteristics into macro-regional parameters. The mean roughness value reflects the overall roughness of the region, while the gradient fluctuation value reflects the spatial variation characteristics of roughness. This comprehensively characterizes the influence of regional surface morphology on laser excitation and provides data basis for energy difference adjustment.

[0054] The detection priorities and laser excitation requirements for different regions of lithium battery electrodes differ fundamentally. The positive electrode region is the core area for detecting valuable metals and requires stable and sufficient laser excitation energy. The negative electrode region serves as a reference area with secondary excitation requirements. Damaged areas have structural defects and require reduced energy to avoid secondary damage. First, second, and third weights are assigned to the positive electrode region, damaged area, and negative electrode region, respectively, with the first weight being greater than the third weight, which is greater than the second weight. This weight relationship perfectly matches the detection priority of each region.

[0055] For example, based on the hardware parameters of the detection system and the characteristics of the electrode material, the first weight corresponding to the positive electrode region is set to 0.5, the third weight corresponding to the negative electrode region is set to 0.3, and the second weight corresponding to the damaged region is set to 0.2. This ensures that the weight of the core detection positive electrode region is the highest, while keeping the weight of the damaged region at the lowest level, which meets the laser excitation safety and detection requirements of different regions. The weight configuration process only assigns values ​​and does not change the original values ​​of the roughness parameters of each region, so as to calculate the global roughness.

[0056] This transforms the detection priorities of different regions into quantifiable computational parameters, enabling the laser energy adjustment process to prioritize the needs of the core detection area while also taking into account the protection requirements of the damaged areas.

[0057] The roughness parameter of a single region cannot reflect the laser energy requirements of the overall morphology of the electrode. The weighted fusion process multiplies the average surface roughness and gradient fluctuation value of each region by the corresponding region weight, and then sums the calculation results of all regions to obtain the weighted fused global roughness parameter. This parameter provides a unified energy calibration benchmark for the entire region, which is used to correct the initial energy corresponding to the roughness parameter of each region. This ensures that the initial pulse energy of each region is both adapted to its own morphological characteristics and takes into account the consistency of the overall morphology of the electrode, avoiding detection deviation caused by excessive energy differences between regions.

[0058] For example, the global roughness mean is obtained by weighted fusion of the roughness mean values ​​of each region, i.e., 0.5×6+0.3×4+0.2×10=3+1.2+2=6.2; the global roughness gradient fluctuation value is obtained by weighted fusion of the gradient fluctuation values ​​of each region, i.e., 0.5×1.2+0.3×0.8+0.2×2.5=0.6+0.24+0.5=1.34; the initial pulse energy is generated based on the global roughness parameter as the reference and the roughness parameter of each region as the core, generating the initial pulse energy of the corresponding region. The mapping coefficient between the global roughness mean and the pulse energy is set to 10, and the basic energy reference is obtained as 62mJ. Then, the basic energy reference is corrected according to the roughness mean and gradient fluctuation value of each region. The correction coefficient is determined by the ratio of the gradient fluctuation value of each region to the global roughness gradient fluctuation value. That is, the larger the gradient fluctuation value of each region, the larger the correction amplitude, and the initial pulse energy of each region is obtained.

[0059] To avoid excessive correction amplitude exceeding the safe energy range of the laser equipment, a scaling factor of 5 is set to control the correction amplitude within ±20%, meaning the maximum impact amplitude of the correction coefficient ÷ 5 does not exceed 20%. The positive electrode region is the core detection area, with a high content of valuable metals, requiring sufficient laser energy to effectively excite the characteristic spectrum. Therefore, the energy is increased from the baseline energy to ensure excitation effect. The negative electrode region serves as a reference calibration area, providing a matrix correction benchmark. Its excitation energy requirement is lower than that of the positive electrode region, and the negative electrode region has a low average roughness and small gradient fluctuations, thus requiring a lower energy. The damaged area has a fragile structure, high average roughness, and large gradient fluctuations; high energy can easily exacerbate electrode damage, therefore, a lower energy is necessary.

[0060] For example, in the positive electrode region, the average roughness is 6, the gradient fluctuation is 1.2, the correction factor is 1.2 ÷ 1.34 ≈ 0.9, and the initial pulse energy is 62 mJ × (1 + 0.9 ÷ 5) ≈ 62 × 1.18 ≈ 73 mJ. Because the positive electrode region is the core detection area and the gradient fluctuation is small, the energy is increased to ensure the excitation effect. In the negative electrode region, the average roughness is 4, the gradient fluctuation is 0.8, the correction factor is 0.8 ÷ 1.34 ≈ 0.6, and the initial pulse energy is 62 mJ. J×(1-0.6÷5)≈62×0.88≈55mJ, the negative electrode region is the reference region, the energy is slightly lower than the basic benchmark, which is suitable for its low roughness characteristics; the average roughness of the damaged region is 10, the gradient fluctuation value is 2.5, the correction coefficient is 2.5÷1.34≈1.87, the initial pulse energy is 62mJ×(1-1.87÷5)≈62×0.626≈39mJ, the damaged region has large gradient fluctuation and fragile structure, so the energy is reduced to avoid secondary damage.

[0061] This enables the integration of roughness parameters across multiple regions, allowing the initial pulse energy to be precisely matched with the overall and local morphological features of the electrode, thus providing target parameters to be tested for compliance verification.

[0062] The initial pulse energy is generated based solely on the roughness parameters, without considering the hardware constraints of the laser equipment and the process requirements for electrode excitation. The energy value may exceed the rated range of the equipment or fail to match the effective excitation depth of the electrode. The initial pulse energy is compared with the preset rated energy range of the laser equipment. The safe energy range of the laser equipment is set to 20mJ to 80mJ. The aforementioned generated initial pulse energies are all within this range, satisfying the upper and lower limit constraints of energy.

[0063] The electrode excitation depth matching verification determines whether the excitation depth corresponding to the initial pulse energy matches the effective detection layer thickness of the electrode. The valuable metal distribution layer thickness of the electrode is 15 μm. The excitation depth corresponding to 73 mJ laser energy in the positive electrode region is 17 μm, 55 mJ in the negative electrode region corresponds to 14 μm, and 39 mJ in the damaged region corresponds to 10 μm. All of these conditions must cover the effective detection layer without penetrating the substrate, thus meeting the excitation depth matching requirement. If both conditions are met, the verification is considered passed; otherwise, it is considered failed. If the initial pulse energy in the positive electrode region, after correction, is 82 mJ, exceeding the equipment's upper limit of 80 mJ, the energy upper and lower limit constraint verification for that region is considered failed. If the initial pulse energy in the damaged region is 18 mJ, below the equipment's lower limit of 20 mJ, and the corresponding excitation depth is only 7 μm, failing to cover the valuable metal distribution layer, the excitation depth matching verification is considered failed.

[0064] This process eliminates initial pulse energies that do not meet hardware and process requirements, preventing equipment malfunctions or detection failures due to energy anomalies. The initial pulse energy that passes the verification is used as the final laser pulse energy output, while the energy that fails the verification needs to enter the weighted iteration process to provide a basis for energy closed-loop optimization.

[0065] When the initial pulse energy verification fails, the energy parameters cannot be directly output. The iterative adjustment process strictly maintains the core relationship that the first weight is greater than the third weight, which is greater than the second weight, and only makes minor adjustments to the weights. The adjustment range is determined based on the reason for the verification failure. For example, if the initial pulse energy of 82mJ in the positive electrode region exceeds the equipment's upper limit, the weights of the core region need to be reduced. The first weight is adjusted from 0.5 to 0.45, the third weight from 0.3 to 0.32, and the second weight remains unchanged at 0.2. The adjusted weights are then divided by the sum to complete normalization. The process involves adjusting the weights: first weight 0.464, second weight 0.206, and third weight 0.33. After adjustment, the weighted fusion yields a global roughness mean of 5.86. Based on this, the initial pulse energy is adjusted to 67mJ for the positive electrode region, 51mJ for the negative electrode region, and 36mJ for the damaged region. The values ​​for each region are within the safe range of 20mJ to 80mJ, and the excitation depth matches the detection requirements, thus passing the verification. The iteration process has no fixed limit on the number of iterations and terminates with the verification passing. Each adjustment maintains the core ranking relationship of the weights.

[0066] This enables closed-loop optimization of laser pulse energy, allowing energy parameters to gradually adapt to hardware constraints and process requirements without changing the regional detection priority, thus ensuring the rationality and stability of the energy. The laser pulse energy after verification serves as the execution parameter for the excitation module to excite each sampling point, providing the energy basis for the subsequent sampling point excitation process. It is the core link connecting surface morphology detection and laser excitation operation.

[0067] The excitation module is used to set basic sampling points in the positive electrode region, sparse sampling points in the damaged region, and reference sampling points in the negative electrode region to determine the spatial distribution coordinates of different sampling points. It also excites the lithium battery point by point according to the laser pulse energy to obtain the spectral data of different sampling points of the lithium battery.

[0068] Specifically, the excitation of the lithium battery point by point according to the laser pulse energy includes: Based on the spatial distribution coordinates of the basic sampling points corresponding to the positive electrode region, the sparse sampling points corresponding to the damaged region, and the reference sampling points corresponding to the negative electrode region, the laser pulse energy of the corresponding regions is matched respectively. Excitation of the reference sampling point is performed according to the matching laser pulse energy to obtain the reference excitation reference, so as to calibrate the laser pulse energy of the corresponding region. Then, according to the adjacency relationship of the same region in the spatial distribution coordinates, the basic sampling point and the sparse sampling point are excited one by one in sequence. The system acquires the original spectral intensity of different sampling points in real time, determines whether it meets the preset signal threshold, and saves the spectral data of the corresponding sampling point if it does not meet the threshold. The system then re-excites the corresponding sampling point until the preset signal threshold is met.

[0069] The sampling point types in different regions are fundamentally different, and the laser pulse energy obtained from the previous adjustment in each region has been adapted to its surface roughness characteristics. If the sampling point coordinates are not accurately matched with the laser pulse energy of the corresponding region, the excitation energy of the sampling point will be mismatched with the regional morphology and sampling type. In particular, if the basic sampling points in the core detection area have insufficient energy, they will not be able to effectively excite the spectrum of valuable metals; if the sparse sampling points in the damaged area have too high energy, it will aggravate the damage of the electrode; if the reference sampling point has energy deviation, it will affect the accuracy of the subsequent calibration benchmark.

[0070] Based on the spatial distribution coordinates of the basic sampling points corresponding to the positive electrode region, the sparse sampling points corresponding to the damaged region, and the reference sampling points corresponding to the negative electrode region, the excitation module deploys the two-dimensional spatial coordinates (X, Y) of the sampling points in each region according to the region segmentation results. These coordinates have been determined in the previous process to accurately locate the specific position of each sampling point on the electrode. After adjustment and verification, the laser pulse energy of each region is output. That is, the positive electrode region, the negative electrode region, and the damaged region correspond to different energies, which are matched with the average surface roughness, gradient fluctuation value, and weight configuration of each region. The core of the matching is to establish the correspondence between the sampling point coordinates, the region affiliation, and the laser energy. Specifically, the spatial distribution coordinates of all sampling points and the corresponding region affiliation information are extracted first, and then the laser pulse energy of each region after verification is called. The coordinates of all sampling points in the same region are bound to the laser pulse energy of that region to form a matching relationship, ensuring that each sampling point can call the excitation pulse energy of its region.

[0071] For example, assuming that after prior adjustments, the laser pulse energy in the positive electrode region is 73 mJ, in the negative electrode region it is 55 mJ, and in the damaged region it is 39 mJ; the coordinates of the basic sampling points deployed in the positive electrode region are (100, 200), (100, 210), (110, 200)..., and the coordinates of all basic sampling points in this region are bound to the laser pulse energy of 73 mJ; the coordinates of the reference sampling points deployed in the negative electrode region are (300, 200), (300, 210), (310, 200)..., and all are bound to the laser pulse energy of 55 mJ; the coordinates of the sparse sampling points deployed in the damaged region are (200, 250), (210, 260)..., and all are bound to the laser pulse energy of 39 mJ, thus completing the matching of all sampling point coordinates with the corresponding region's laser pulse energy.

[0072] This allows for the matching of excitation pulse energy at sampling points with regional morphology and sampling type, ensuring that the excitation pulse energy at each sampling point meets its detection requirements, effectively exciting the spectrum of valuable metals, and guaranteeing the reliability of subsequent calibration. Sparse sampling points obtain appropriate energy to avoid exacerbating electrode damage.

[0073] Although the laser pulse energy adjusted in the early stage has been verified by hardware and process, there may be slight deviations in the actual excitation process due to minor factors such as the uniformity of electrode material and ambient temperature. If all sampling points are directly excited, the consistency of spectral data will be poor. In addition, the adjacency relationship of sampling points in the same area determines the excitation order. Excitation according to the adjacency relationship avoids interference to adjacent sampling points during the excitation process, ensures that the excitation environment of each sampling point is consistent, and further improves the reliability of spectral data.

[0074] The reference excitation benchmark refers to the original spectral intensity and corresponding excitation energy parameters of the reference sampling point obtained after excitation. This reference excitation benchmark is used to determine whether there is a deviation in the energy of the current laser pulse and to fine-tune and calibrate the energy of the corresponding region. Within the same region, the coordinates of the sampling points are associated with each other according to their adjacent positions (such as left and right adjacent, or top and bottom adjacent), which is the spatial distribution coordinates of the same region. Excitation according to this relationship can ensure the continuity of the excitation process and avoid disordered excitation across regions and positions. After the excitation and calibration of the reference sampling points are completed, the basic sampling points are excited first, and then the sparse sampling points are excited.

[0075] First, the excitation module calls the laser pulse energy that matches the reference sampling point and performs point-by-point excitation on all reference sampling points in the negative electrode area. The original spectral intensity of each reference sampling point is acquired in real time. The average spectral intensity of all reference sampling points is used as the reference excitation benchmark. The benchmark is compared with the preset reference spectral intensity range. If the benchmark is within the range, it means that the current laser pulse energy is within the range and no calibration is required. If the benchmark is below the range, it means that the energy is insufficient, and the laser pulse energy of the corresponding area is slightly increased. If the benchmark is above the range, it means that the energy is too high, and it is slightly decreased, thus completing the calibration of the laser pulse energy of the corresponding area.

[0076] After calibration, the basic sampling points and sparse sampling points are excited sequentially according to their adjacency in the same region. First, for the basic sampling points in the positive electrode region, the laser pulse energy of the positive electrode region is called point by point in the adjacency order from left to right and from top to bottom according to the coordinates. After the positive electrode region is excited, the sparse sampling points in the damaged area are excited point by point in the adjacency order in the same region according to the calibration order, to ensure that each sampling point is excited with the calibrated energy.

[0077] For example, continuing with the aforementioned energy values, after excitation of the reference sampling point in the negative electrode region, the average spectral intensity obtained is 1200 counts. Here, counts is the original counting unit output by the spectrometer detector, representing the number of photons received by the detector within a specific integration time. It is a relative measurement value, characterizing the strength of the spectral signal under the same equipment and the same integration conditions. That is, the reference excitation standard is 1200 counts, and the preset reference spectral intensity range is 1100~1300 counts. The reference excitation standard is within the range, so no calibration is required, and the energy of 55mJ is maintained unchanged. The basic sampling points in the positive electrode region are excited point by point with 73mJ energy in the adjacent order of (100,200)→(100,210)→(110,200)→…; the sparse sampling points in the damaged region are excited point by point with 39mJ energy in the adjacent order of (200,250)→(210,260)→…; if the average spectral intensity of the reference sampling point is 1050 counts, which is lower than the preset reference spectral intensity range, the laser energy in the negative electrode region is increased to 56mJ, and the positive electrode region is increased to 74mJ and the damaged region is increased to 40mJ, and then the subsequent excitation is performed.

[0078] This eliminates laser energy deviations caused by minor factors, ensuring that the excitation energy of all sampling points is within a reasonable range; excitation is performed sequentially according to the adjacency relationship and detection priority within the same region, ensuring that the excitation process is orderly and efficient, taking into account the needs of both core and auxiliary detection points, and guaranteeing the comprehensiveness and reliability of spectral data.

[0079] During the excitation process at each sampling point, accidental factors such as surface contamination, severe local damage to the electrode, and excitation angle deviation may result in insufficient original spectral intensity at some sampling points. Directly saving such spectral data would lead to insufficient extraction of chemical components due to weak signals, affecting the accuracy of the detection results. After the excitation module excites each sampling point, the detection module synchronously acquires the spectral signal of that sampling point and converts it into a quantifiable original spectral intensity to reflect the strength of the spectral signal. The preset signal threshold refers to the minimum effective value of the preset spectral intensity based on the detection accuracy requirements of the detection system. Spectral signals below this threshold cannot be used for subsequent component extraction. The setting of this preset signal threshold is matched with the performance of the laser equipment, the electrode material, and the detection requirements for valuable metal content.

[0080] For sampling points with spectral intensities lower than a preset signal threshold, the excitation operation is repeated until the spectral intensity reaches or exceeds the preset signal threshold, ensuring that each sampling point can provide valid spectral data. In practice, after the excitation module excites each sampling point, the detection module collects the original spectral intensity of the sampling point in real time and immediately compares it with the preset signal threshold. If the spectral intensity is greater than or equal to the preset signal threshold, the excitation of the sampling point is deemed valid, and the spectral data of the corresponding sampling point, including the original spectral intensity, sampling point coordinates, and excitation energy, is saved. If the spectral intensity is lower than the preset signal threshold, the excitation is deemed invalid, and the excitation module is controlled to re-excite the sampling point. During re-excitation, the calibrated laser pulse energy matched to the sampling point is used, and the excitation angle is finely adjusted to ensure that the excitation angle is perpendicular to the electrode surface. The excitation and verification process is repeated until the original spectral intensity of the sampling point meets the preset signal threshold, and then its spectral data is saved. If the threshold is still not met after multiple re-excitations (e.g., 3 times), the sampling point is deemed invalid, its coordinates are recorded and processed subsequently, and the number of valid sampling points is ensured to meet the detection requirements.

[0081] For example, if the preset signal threshold is set to 1000 counts, after the initial excitation of a basic sampling point in the positive region, the original spectral intensity is 920 counts, which is lower than the signal threshold and is therefore deemed invalid. The sampling point is then re-excited, and after fine-tuning the excitation angle, the obtained spectral intensity is 1080 counts, which meets the signal threshold, and the spectral data of the sampling point is saved. After the initial excitation of a sparse sampling point in the damaged region, the spectral intensity is 850 counts. After two re-excitations, the spectral intensity reaches 1020 counts, which meets the signal threshold and is saved. If a reference sampling point still has a spectral intensity of 950 counts after three re-excitations, it is deemed an invalid sampling point, and its coordinates (300, 220) are recorded. Subsequent sampling points can be added at this location for re-excitation.

[0082] This process eliminates invalid spectral data, ensuring that each sampling point provides a valid spectral signal that meets the detection requirements, thus guaranteeing the integrity and reliability of the spectral data. It also improves the success rate of re-excitation. The output valid spectral data serves as input data for subsequent processing, and the validity and accuracy of the spectral data determine the precision of subsequent component extraction, ensuring the overall reliability of the detection system.

[0083] The detection module performs matrix correction on the spectral data, extracts the chemical composition content of different sampling points, constructs a composition sequence by spatially ordering the chemical composition content of all sampling points on the same lithium battery, processes the composition sequence through a long short-term memory network, and outputs the valuable metal content and abnormal sampling points of the lithium battery.

[0084] Furthermore, the chemical composition content at different sampling points was extracted, including: Baseline correction and noise filtering are performed sequentially on the spectral data from different sampling points, and the matrix correction coefficients of the lithium battery are constructed based on the processed spectral data corresponding to the reference sampling points. Intensity compensation is performed point-by-point for the corresponding bands of the spectral data after processing the base sampling points and sparse sampling points, based on the matrix correction coefficient. The peak positions of the characteristic spectral lines corresponding to the valuable metals are located in the intensity-compensated spectral data. Peak area integration is performed on the characteristic spectral lines to obtain the peak intensity. The chemical composition content of the corresponding sampling points is extracted by combining the pre-stored quantitative standard curve of the composition.

[0085] The raw spectral data obtained through sampling point excitation in the early stage suffers from baseline drift and noise interference due to factors such as laser excitation noise, electrode surface scattering, and electronic noise of the detection equipment. If used directly for subsequent component extraction, it will lead to deviations in the identification of characteristic spectral lines. At the same time, the lithium battery electrode substrate (such as copper foil, aluminum foil, binder, etc.) will absorb and interfere with the characteristic spectral lines of valuable metals, resulting in distortion of the characteristic spectral line intensity and failing to directly reflect the true content of valuable metals. Baseline correction refers to the process of eliminating baseline drift in spectral data. Baseline drift refers to the slow shift of spectral intensity with wavelength, which does not reflect the characteristics of valuable metals and needs to be corrected back to the reference level.

[0086] Noise filtering refers to filtering out random noise in spectral data while retaining the effective signals corresponding to the characteristic spectral lines of valuable metals. The matrix correction coefficient is a quantization coefficient used to counteract the interference of the lithium battery matrix on the characteristic spectral lines. It is calculated based on the spectral data of the reference sampling points and reflects the intensity and pattern of matrix interference. Specifically, baseline correction and noise filtering are performed sequentially on the spectral data of all sampling points. Baseline correction uses a polynomial fitting method, selecting the bands without characteristic spectral lines (i.e., background bands) in the spectrum as the fitting benchmark. The baseline curve is obtained by fitting using the least squares method. The baseline curve is then subtracted from the original spectral data to complete the baseline correction. Noise filtering uses a Gaussian filtering algorithm, setting the filtering window size to 5 wavelength points. The baseline-corrected spectral data is smoothed to filter out random noise and retain the effective characteristic signals.

[0087] For example, spectral data with a wavelength range of 200-800 nm are selected, where 200-250 nm and 750-800 nm are background bands without characteristic spectral lines. Polynomial fitting is performed on the original spectral data of a certain reference sampling point to obtain a baseline curve. The intensity at 300 nm in the original spectrum is 1250 counts, and the corresponding intensity of the baseline curve is 250 counts. After correction, the spectral intensity at that wavelength is 1000 counts. Then, Gaussian filtering is applied to filter out random fluctuations in the spectral data of the sampling point, making the spectral curve smoother and ensuring that the characteristic spectral lines are clearly distinguishable.

[0088] After preprocessing, the matrix correction coefficient of the lithium battery is constructed. The preprocessed spectral data of all reference sampling points are extracted, and the bands corresponding to the characteristic spectral lines of valuable metals are selected. For example, the characteristic spectral line wavelength of cobalt is 350nm and that of nickel is 380nm. The ratio of the spectral intensity of the reference sampling point in this band to the preset pure matrix spectral intensity is calculated. This ratio is the matrix correction coefficient, which is used to characterize the degree of interference of the matrix on the characteristic spectral line of this band. For example, if the 350nm characteristic spectral line band of cobalt is selected, the average spectral intensity of the preprocessed reference sampling point is 1000 counts, and the preset pure matrix spectral intensity in this band is 200 counts, the matrix correction coefficient is calculated to be 1000 / 200=5. This matrix correction coefficient can be used to subsequently offset the matrix interference in the spectral data of the basic sampling point and sparse sampling point.

[0089] This eliminates the influence of irrelevant interference factors on spectral data, making the characteristic spectral lines of valuable metals clearer and ensuring that the spectral data truly reflects the characteristic information of valuable metals; it also accurately quantifies the interference patterns of the lithium battery substrate, providing a basis for subsequent intensity compensation and avoiding distortion of characteristic spectral line intensity caused by substrate interference.

[0090] Even after preprocessing, the spectral data from the basic and sparse sampling points still suffer from interference from the lithium battery matrix, causing the intensity of the characteristic spectral lines of valuable metals to be lower than the true value, making them unsuitable for direct component quantification. The corresponding band refers to the wavelength band where the characteristic spectral lines of valuable metals are located. Different valuable metals correspond to different characteristic spectral line bands, and intensity compensation is only performed for this band to avoid ineffective compensation for non-characteristic bands. For each basic sampling point and each sparse sampling point, compensation is performed separately based on the matrix correction coefficient to ensure the specificity and accuracy of the compensation. Since there are slight differences in the degree of matrix interference at different sampling points, point-by-point compensation can avoid the deviation caused by uniform compensation.

[0091] Specifically, the constructed matrix correction coefficients are first called to identify the bands corresponding to the characteristic spectral lines of different valuable metals. Then, for each basic sampling point and sparse sampling point, the intensity value of the corresponding characteristic band in the preprocessed spectral data is extracted. This intensity value is multiplied by the matrix correction coefficient to obtain the compensated spectral intensity value, thus completing the intensity compensation. During the compensation process, only the intensity of the characteristic spectral line bands is adjusted, while the spectral data of non-characteristic bands remain unchanged to avoid affecting the accuracy of subsequent peak position positioning.

[0092] For example, continuing with the matrix correction factor of 5, for a basic sampling point in the positive electrode region, the spectral intensity of cobalt in the 350nm characteristic band after preprocessing is 800 counts. After multiplying by the matrix correction factor of 5, the compensated spectral intensity is 4000 counts. This value restores the true intensity of the cobalt characteristic spectral lines and cancels out matrix interference. For a sparse sampling point in the damaged area, the spectral intensity of nickel in the 380nm characteristic band after preprocessing is 600 counts. After multiplying by the matrix correction factor of 5, the compensated spectral intensity is 3000 counts, completing the intensity compensation for this sampling point. For multiple valuable metals (such as cobalt, nickel, and manganese), the same matrix correction factor is used for point-by-point compensation for their corresponding characteristic bands to ensure that the true intensity of all valuable metal characteristic spectral lines is restored.

[0093] This precisely counteracts the interference of the lithium battery substrate on the characteristic spectral lines of valuable metals, restores the true intensity of the characteristic spectral lines, solves the intensity distortion caused by substrate interference, and ensures that the spectral data accurately reflects the actual content characteristics of valuable metals; it also avoids ineffective adjustments to non-characteristic bands and ensures the integrity of the spectral data.

[0094] In the intensity-compensated spectral data, the characteristic spectral lines of valuable metals are clearly presented, but the specific positions of each characteristic spectral line of valuable metals need to be determined by peak position positioning to avoid confusing the characteristic signals of different valuable metals; peak area integration can convert the intensity of the characteristic spectral lines into quantifiable peak intensity parameters, which have a clear quantitative relationship with the content of valuable metals; the pre-stored component quantitative standard curve is constructed based on standard samples with known contents, realizing the accurate mapping between peak intensity and valuable metal content.

[0095] Peak location for characteristic spectral lines of valuable metals refers to determining the wavelength position corresponding to the peak value of each valuable metal's characteristic spectral line. Each valuable metal has its own characteristic spectral line wavelength, such as 350nm for cobalt, 380nm for nickel, and 400nm for manganese. Peak location can accurately identify the characteristic signals of different valuable metals. Peak area integration to obtain peak intensity refers to integrating the spectral intensity near the peak value of the characteristic spectral line. The integration result is the peak intensity, which quantifies the total intensity of the characteristic spectral line and is positively correlated with the content of the valuable metal. The pre-stored quantitative standard curve refers to the linear fitting curve between peak intensity and content constructed by measuring the peak intensity of characteristic spectral lines of standard samples with known valuable metal contents in advance, so as to realize the mapping from peak intensity to content.

[0096] Specifically, the peak positions of the characteristic spectral lines corresponding to valuable metals are located in the intensity-compensated spectral data. A peak detection algorithm is used, and a peak threshold is set (1.5 times higher than the surrounding spectral intensity). The spectral data is traversed to identify the peak wavelengths corresponding to the characteristic spectral lines of each valuable metal, confirming the accuracy of the peak position and avoiding misidentification of noise peaks as characteristic peaks. For example, for the spectral data of a certain basic sampling point after compensation, the peak detection algorithm identifies three peak wavelengths of 350nm, 380nm, and 400nm, which correspond to the characteristic spectral lines of cobalt, nickel, and manganese, respectively. It is confirmed that the peak position is consistent with the preset characteristic wavelength, thus completing the peak position location.

[0097] After peak location is completed, peak area integration is performed on each characteristic spectral line. The integration interval is defined with the peak wavelength as the center. For example, 345-355nm corresponds to the characteristic spectral line of cobalt. The trapezoidal integration method is used to integrate the compensated spectral intensity within this interval. The integration result is the peak intensity of the characteristic spectral line of the valuable metal. For example, the integration interval of the characteristic spectral line of cobalt is 345-355nm, and the integrated value of the compensated spectral intensity within this interval is 40000 counts·nm, which is the peak intensity of cobalt. The integration interval of nickel is 375-385nm, and the integrated value is 30000 counts·nm, which is the peak intensity of nickel.

[0098] The chemical component content is extracted by combining pre-stored quantitative standard curves. These pre-stored quantitative standard curves are linear fitting curves, with peak intensity on the x-axis and valuable metal content on the y-axis. Each valuable metal corresponds to a standard curve. The peak intensity obtained by integrating the peak area is substituted into the corresponding standard curve, and the chemical component content of the valuable metal corresponding to that sampling point is calculated through linear interpolation. For example, the fitting equation of the pre-stored quantitative standard curve for cobalt is y=0.0001x, where y is the content (%) and x is the peak intensity (counts·nm). Substituting the cobalt peak intensity of 40000 counts·nm into the standard curve, the cobalt content at that sampling point is 4.0%. Substituting the nickel peak intensity of 30000 counts·nm into the corresponding standard curve, where the fitting equation is y=0.00012x, the nickel content is calculated to be 3.6%. Similarly, the chemical component content of valuable metals at all basic sampling points and sparse sampling points is extracted.

[0099] This allows for the accurate identification of characteristic signals of different valuable metals, avoiding signal confusion; it transforms the intensity of characteristic spectral lines into quantifiable peak intensity parameters, establishing a correlation between characteristic signals and the content of valuable metals; and it achieves a precise mapping from peak intensity to chemical component content, ensuring that the extraction results of chemical component content at each sampling point are accurate and reliable, thus fulfilling the core detection objective of the detection system.

[0100] Specifically, the valuable metal content and abnormal sampling points of the output lithium battery include: According to the spatial order of the spatial distribution coordinates corresponding to different sampling points, the chemical composition content of all sampling points on the same lithium battery is sorted in sequence to form a composition sequence; Feature extraction and data fitting of component sequences are performed using a long short-term memory network. The fitting result is the time-series fusion value of the chemical component content corresponding to the sampling point. Combined with the pre-stored component quantitative standard curve, the time-series fusion value is mapped to the valuable metal content of lithium battery. By comparing the chemical component content of each sampling point in the component sequence with the average chemical component content of sampling points in the same region, abnormal sampling points are identified.

[0101] The chemical composition content of each sampling point obtained in the early stage is discrete local detection data, which does not reflect the spatial distribution relationship of the sampling points on the lithium battery electrode. The Long Short-Term Memory Network (LSTM) highly depends on the ordered input structure for feature extraction of sequence data. Disordered composition data cannot enable the network to discover the spatial distribution pattern of valuable metals on the electrode. The ordered composition sequence is the basis for subsequent global content fitting and abnormal sampling point judgment.

[0102] Spatial distribution coordinates refer to the two-dimensional planar coordinates of each sampling point on the lithium battery electrode determined by the excitation module when setting up the sampling points, including the positioning values ​​of the horizontal and vertical axes; spatial order refers to the unified sorting rules based on the fixed orientation of the lithium battery electrode, usually using a combination of ascending horizontal coordinate and ascending vertical coordinate under the same horizontal coordinate; composition sequence refers to the ordered data set formed after arranging according to the above spatial order, with the chemical composition content of the sampling points as the element.

[0103] Specifically, the spatial distribution coordinates and corresponding chemical composition content of all valid sampling points of the same lithium battery are extracted. Taking the left side of the electrode as the starting position of the horizontal axis and the top side as the starting position of the vertical axis, the sampling points are first sorted according to the horizontal axis coordinate values ​​from smallest to largest. Sampling points with the same horizontal axis coordinates are then sorted according to the vertical axis coordinate values ​​from smallest to largest. The sorted chemical composition contents are then combined sequentially to form a one-dimensional composition sequence.

[0104] For example, the effective sampling points of a certain lithium battery include the positive electrode basic sampling point, the damaged sparse sampling point, and the negative electrode reference sampling point. The corresponding spatial coordinates and cobalt chemical composition contents are (100,200) corresponding to 4.0%, (100,210) corresponding to 4.1%, (200,250) corresponding to 3.8%, and (300,200) corresponding to 1.2%. After being sorted in spatial order, the final constructed composition sequence is [4.0%, 4.1%, 3.8%, 1.2%].

[0105] This integrates discrete local component data into an ordered sequence with spatial correlation, preserves the spatial distribution characteristics between sampling points, eliminates the interference of data disorder on subsequent network processing, and enables the component sequence to truly reflect the spatial distribution of valuable metals on the electrode.

[0106] The composition sequence is only a collection of local chemical composition contents at each sampling point and cannot directly characterize the overall valuable metal content of the lithium battery. Among them, the Long Short-Term Memory Network is a recurrent neural network suitable for sequence data feature processing, which can avoid the information loss problem that occurs when conventional neural networks process sequences. Feature extraction refers to the network extracting spatial distribution, local fluctuations and other related features of valuable metal content from the composition sequence through internal neuron operations. Data fitting refers to the network completing the numerical fitting operation of global composition data based on the extracted spatial features. The temporal fusion value is the global quantitative value output by the network after fitting, which integrates the spatial features of the composition content of all sampling points.

[0107] Specifically, the constructed component sequence is input into a pre-trained Long Short-Term Memory (LSTM) network. The network receives the component sequence data through the input layer, extracts the spatial features of valuable metal content through the hidden layer, and then performs data fitting and outputs a temporal fusion value through the output layer. This temporal fusion value is an overall content estimate in the content dimension. The pre-stored component quantification standard curve provides a calibration benchmark for the content dimension during the network training phase, ensuring that the output temporal fusion value is consistent with the content definition of the component quantification standard curve. Finally, the temporal fusion value is mapped to the valuable metal content of the lithium battery. For example, if the aforementioned component sequence [4.0%, 4.1%, 3.8%, 1.2%] is input into the LSM network, after feature extraction and data fitting, the output temporal fusion value is 3.2%, which is the overall valuable metal content of the lithium battery.

[0108] This enables in-depth mining and global fusion of the spatial features of the component sequence, avoids the random bias of local data from a single sampling point, and ensures the quantitative standardization of the overall valuable metal content through standard curve mapping, so that the output results objectively reflect the overall valuable metal content of the lithium battery.

[0109] The positive, negative, and damaged areas of a lithium battery electrode each possess stable compositional distribution characteristics. The chemical composition content of normal sampling points within the same area should be within a relatively uniform range. A significant deviation between the content of a single sampling point and the regional average usually indicates an abnormality in the excitation, spectral acquisition, or component extraction process at that sampling point, or a local abnormality at the corresponding location of the electrode. Here, "same area" refers to the positive, negative, and damaged areas determined through prior regional segmentation, and the regional assignment of each sampling point has been calibrated during the sampling point deployment phase. The average chemical composition content refers to the arithmetic mean of the chemical composition content of all valid sampling points within a single area, serving as a benchmark reference value for the chemical composition content of that area.

[0110] Specifically, based on the regional affiliation of the sampling points, all sampling points are divided into the positive electrode basic sampling point group, the damaged and sparse sampling point group, and the negative electrode reference sampling point group. The arithmetic mean of the chemical composition content of each sampling point in each group is calculated to obtain the content mean of each region. The chemical composition content of each sampling point in the composition sequence is numerically compared with the content mean of the region to which the sampling point belongs to complete the identification of abnormal sampling points.

[0111] For example, the positive electrode basic sampling point group includes two sampling points: (100, 200) with a chemical composition content of 4.0% and (100, 210) with a chemical composition content of 4.1%, with a regional average chemical composition content of 4.05%. The damaged and sparse sampling point (200, 250) with a chemical composition content of 3.8% is the only sampling point in the corresponding region, with a regional average of 3.8%. The negative electrode reference sampling point (300, 200) with a chemical composition content of 1.2% is the only sampling point in the corresponding region, with a regional average of 1.2%. If the chemical composition content of a sampling point in the positive electrode region is 2.0%, and the deviation from the regional average of 4.05% exceeds the reasonable range, then the sampling point is determined to be an abnormal sampling point. If the deviation of the sampling point content from the regional average is within the reasonable range, then it is determined to be a normal sampling point.

[0112] This avoids misjudgments caused by differences in component content in different areas, and only identifies local deviation points within the area, improving the accuracy and specificity of abnormal sampling point determination, and effectively locating the sampling point location with problems in the detection process; it provides a basis for the subsequent control module to determine whether to trigger resampling detection, and the accuracy of abnormal sampling point determination determines the rationality of starting the resampling process and the accuracy of its execution.

[0113] The control module is used to generate a report on the content of valuable metals and link the recycling process parameters. When the number of abnormal sampling points exceeds the preset threshold, it triggers resampling and detection of the corresponding area.

[0114] Specifically, triggering resampling and detection of the corresponding region includes: The valuable metal content of lithium batteries, the regional affiliation and spatial distribution coordinates of abnormal sampling points are integrated to generate a valuable metal content report, and the corresponding recycling process parameters are matched accordingly. If the number of abnormal sampling points exceeds the preset threshold, the target area to which the abnormal sampling points belong is located, and corresponding sampling points are added within the target area. The sampling points before and after the addition are re-excited according to the laser pulse energy, and the valuable metal content of the lithium battery is detected until the number of abnormal sampling points does not exceed the preset threshold.

[0115] The data on the valuable metal content and abnormal sampling points of lithium batteries output from the initial process are scattered test results and cannot be directly applied to subsequent testing, archiving, and recycling processes. Furthermore, the valuable metal content level of the lithium battery and the distribution of abnormal sampling points directly determine the choice of recycling process. Different contents and abnormal states correspond to different recycling process parameters. The valuable metal content report is a standardized test document that integrates the overall valuable metal content of the lithium battery, the regional attribution information of the abnormal sampling points, and the spatial distribution coordinates of the abnormal sampling points. Regional attribution clarifies whether the abnormal sampling point is located in the positive electrode region, negative electrode region, or damaged region, while the spatial distribution coordinates are the two-dimensional positioning data of the abnormal sampling point on the electrode. Recycling process parameters refer to the recycling execution parameters, such as the crushing strength, sorting power, and melting temperature, adapted to different valuable metal contents and different electrode abnormal states. Linkage matching refers to the process of retrieving corresponding matching items from a pre-stored parameter library based on the test results.

[0116] Specifically, the valuable metal content of lithium batteries, the regional affiliation and spatial distribution coordinates of all abnormal sampling points are extracted and integrated according to the standard format of the test report. The overall content value, the number of abnormal sampling points in each region and the corresponding coordinate information are clearly marked to form a complete valuable metal content report. After the report is generated, based on the valuable metal content level and the regional distribution of abnormal sampling points in the report, the data is matched with the pre-stored recycling process parameter library to retrieve the recycling process parameters corresponding to the current test results.

[0117] For example, a lithium battery was tested and found to have an overall valuable metal content of 3.2%. There were 4 abnormal sampling points, all belonging to the positive electrode region, with spatial coordinates of (110,200), (110,210), (120,200), and (120,210). After integrating the above data, a standardized content report was generated. This content was matched with the intermediate recycling process parameters in the pre-stored parameter library corresponding to the abnormal state. The parameters retrieved after matching were crushing strength level 2, sorting power 60%, and smelting temperature 850℃, thus completing the linkage matching of recycling process parameters.

[0118] This enables standardized output of test results, generating archiveable and reusable content reports. Simultaneously, precise linkage and matching directly link test data with the recycling process, opening up data pathways between the testing and recycling stages and avoiding deviations caused by manual judgment of process parameters.

[0119] When the number of abnormal sampling points exceeds the preset threshold, it means that the validity and reliability of the current detection data cannot meet the detection requirements. A single abnormal sampling point can be regarded as a random error, while a batch of abnormal points indicates that there is a systematic deviation in the sampling, excitation, or spectral detection process. If the data is used directly, it will lead to the distortion of the final detection result. At the same time, it is not necessary to resample the entire area of ​​the lithium battery. Only the target area to which the abnormal sampling point belongs needs to be supplemented and retested, which can reduce redundant operations. It is also necessary to use an iterative detection method until the number of abnormal sampling points meets the preset requirements to ensure the validity of the final detection data.

[0120] The preset quantity threshold refers to the maximum allowed number of abnormal sampling points, which is the critical value for the system to determine whether to start resampling. It is set according to the detection importance and sampling density differences of different regions. The positive pole region is the core detection region with the highest sampling point density and the most stringent detection accuracy requirements, so the quantity threshold is the lowest. The damaged region has sparse sampling points and the lowest detection priority, so the quantity threshold is slightly higher. The negative pole region is the reference region, and the quantity threshold is between the two. The target region refers to the single or multiple regions to which the abnormal sampling points are concentrated, that is, the region containing abnormal sampling points in the positive pole region, the negative pole region, and the damaged region. In the target region, according to the layout rules of the original sampling point type in the region, the corresponding type of sampling points are added. Using the laser pulse energy that has been calibrated in the early stage, the original sampling points and the newly added sampling points in the target region are uniformly subjected to excitation, spectral acquisition, component extraction, and abnormality determination retesting operations.

[0121] First, the total number of abnormal sampling points is compared with a preset threshold. If the number of abnormal sampling points is greater than the preset threshold, a resampling process is initiated. The target area is located based on the regional affiliation information of the abnormal sampling points. Within the target area, sampling points of the same type as the original sampling points are added, following the original layout density and spacing rules. After the sampling point addition is completed, the laser pulse energy of the corresponding area, which has been previously verified and calibrated, is called. The original sampling points before addition and the newly added sampling points after addition are uniformly excited point by point. Then, the entire process of spectral data preprocessing, matrix correction, chemical component content extraction, component sequence construction, and abnormal sampling point determination is repeated to obtain a new round of abnormal sampling point count. The new abnormal sampling point count is compared with the preset threshold again. If it is still greater than the threshold, the sampling point layout position is fine-tuned within the target area and the re-excitation and detection process is repeated until the number of abnormal sampling points is no greater than the preset threshold, at which point the iteration process is terminated.

[0122] For example, the threshold number of abnormal sampling points is 2 for the positive electrode region, 3 for the negative electrode region, and 4 for the damaged region. The abnormal sampling points detected so far are all concentrated in the positive electrode region, and the number is 4, which is greater than the threshold number. In the positive electrode region, according to the original layout rules of the basic sampling points, 2 basic sampling points are added, with coordinates (115, 205) and (125, 205) respectively. Using the laser pulse energy of 73mJ corresponding to the positive electrode region, the original 4 abnormal sampling points and the 2 newly added supplementary sampling points in the positive electrode region are uniformly re-excited point by point. After completing the whole process detection, the number of abnormal sampling points in the new round is 2, which is equal to the preset threshold number of the positive electrode region, and the resampling and detection process is terminated.

[0123] This enables precise initiation of the resampling process, allowing for targeted local resampling, improving the efficiency of the testing process, gradually eliminating abnormal data caused by systematic biases, and ensuring that the final test results meet the reliability requirements set by the system. The qualified test data obtained after the iteration is terminated will replace the original abnormal data, be used to update the valuable metal content report, and re-link and match the recycling process parameters, forming a complete closed loop of testing, retesting, and output, providing data support for the final testing archiving and recycling execution.

[0124] Example 2: like Figure 3 The diagram shows a flowchart of a method for detecting valuable metals in lithium batteries, as provided in this application embodiment. The method includes: The surface morphology image of the lithium battery is acquired and region segmentation is performed to identify the positive electrode region, negative electrode region and damaged region, and the surface roughness of the lithium battery is calculated to adjust the laser pulse energy; Basic sampling points were set in the positive electrode region, sparse sampling points were set in the damaged region, and reference sampling points were set in the negative electrode region to determine the spatial distribution coordinates of different sampling points. The lithium battery was excited point by point according to the laser pulse energy to obtain the spectral data of different sampling points of the lithium battery. Matrix correction is performed on the spectral data, the chemical composition content of different sampling points is extracted, the chemical composition content of all sampling points on the same lithium battery is arranged in spatial order to form a composition sequence, the composition sequence is processed by a long short-term memory network, and the valuable metal content and abnormal sampling points of the lithium battery are output. Generate a report on the content of valuable metals and link it to the recycling process parameters. When the number of abnormal sampling points exceeds the preset threshold, trigger resampling and detection of the corresponding area.

[0125] Since the principle of the method in this application embodiment is similar to that of the system described in this application embodiment, the implementation of the method is the same as that of the system, and the repeated parts will not be described again.

Claims

1. A lithium battery valuable metal detection system, characterized in that, include: The module consists of a sensing module, an excitation module, a detection module, and a control module. The sensing module is used to acquire surface topography images of the lithium battery and perform region segmentation to identify the positive electrode region, negative electrode region and damaged region, and calculate the surface roughness of the lithium battery to adjust the laser pulse energy; The adjustment of laser pulse energy includes: Calculate the mean and gradient fluctuation values ​​of the surface roughness within each region, and assign a first weight, a second weight, and a third weight to the positive region, the damaged region, and the negative region, respectively, with the first weight being greater than the third weight, which is greater than the second weight. Based on the weighted fusion of global roughness and its gradient fluctuation value, the initial pulse energy of the corresponding region is generated, and the energy upper and lower limit constraints and the matching of the electrode excitation depth of the initial pulse energy are verified. If the verification passes, the initial pulse energy is output as the laser pulse energy. If the verification fails, the first weight, the second weight, and the third weight are iteratively adjusted and the initial pulse energy is regenerated until the verification passes. The calculation of the mean and gradient fluctuation values ​​of the surface roughness within each region includes: The high-frequency components in the surface topography image are reconstructed to obtain the topography height values ​​of different pixels. Taking each pixel in different regions as the center, a preset neighborhood window is selected, and the arithmetic mean deviation of the topography height values ​​of all pixels in the neighborhood window is calculated to obtain the surface roughness of a single pixel. The surface roughness of all single pixels in each region is weighted and averaged to obtain the mean of the surface roughness of the corresponding region. The continuous rate of change of the surface roughness of single pixels in the region is calculated, and the root mean square error of the continuous rate of change is used as the gradient fluctuation value of the surface roughness. The excitation module is used to set basic sampling points in the positive electrode region, sparse sampling points in the damaged region, and reference sampling points in the negative electrode region to determine the spatial distribution coordinates of different sampling points, and to excite the lithium battery point by point according to the laser pulse energy to obtain the spectral data of different sampling points of the lithium battery. The detection module is used to perform matrix correction processing on spectral data, extract the chemical composition content of different sampling points, construct a composition sequence by spatially ordering the chemical composition content of all sampling points on the same lithium battery, process the composition sequence through a long short-term memory network, and output the valuable metal content and abnormal sampling points of the lithium battery. The control module is used to generate a report on the content of valuable metals and link the recycling process parameters. When the number of abnormal sampling points exceeds the preset threshold, it triggers resampling and detection of the corresponding area.

2. The lithium battery valuable metal detection system as described in claim 1, characterized in that, The execution region segmentation includes: Wavelet decomposition is performed on the surface morphology image of the lithium battery to obtain low-frequency and high-frequency components, so as to extract the undulation gradient features and texture distribution features of the surface morphology image respectively, and fuse them to form a region discrimination vector. Image pixels are clustered and divided into candidate regions based on region discrimination vectors. Based on the prior structural relationship of lithium battery electrodes, spatial topological constraints are applied to all candidate regions to obtain pre-partitions. Optimize all pre-partitions based on the mutual exclusion constraints between regions, and perform fluctuation state verification on the region discrimination vector of the optimized pre-partitions. If the verification passes, output the positive region, negative region and damaged region; otherwise, re-execute feature extraction and re-execute region segmentation.

3. The lithium battery valuable metal detection system as described in claim 2, characterized in that, The spatial topology constraints include the aspect ratio of the lithium battery electrode, the area ratio of the region, and the relative positional relationship between the positive and negative electrodes, in order to eliminate isolated candidate regions and correct the corresponding region boundaries; the mutual exclusion constraints represent that the positive and negative electrode regions are mutually exclusive and adjacent, and the damaged region is superimposed on the positive or negative electrode region; the fluctuation state of the region discrimination vector is the variance fluctuation value of the undulating gradient feature and texture distribution feature within each pre-partition, and the verification is passed when the variance fluctuation value is not greater than the fluctuation threshold.

4. The lithium battery valuable metal detection system as described in claim 3, characterized in that, The step-by-step excitation of the lithium battery based on laser pulse energy includes: Based on the spatial distribution coordinates of the basic sampling points corresponding to the positive electrode region, the sparse sampling points corresponding to the damaged region, and the reference sampling points corresponding to the negative electrode region, the laser pulse energy of the corresponding regions is matched respectively. Excitation of the reference sampling point is performed according to the matching laser pulse energy to obtain the reference excitation reference, so as to calibrate the laser pulse energy of the corresponding region. Then, according to the adjacency relationship of the same region in the spatial distribution coordinates, the basic sampling point and the sparse sampling point are excited one by one in sequence. The system acquires the original spectral intensity of different sampling points in real time, determines whether it meets the preset signal threshold, and saves the spectral data of the corresponding sampling point if it does not meet the threshold. The system then re-excites the corresponding sampling point until the preset signal threshold is met.

5. The lithium battery valuable metal detection system as described in claim 4, characterized in that, The chemical composition content extracted from different sampling points includes: Baseline correction and noise filtering are performed sequentially on the spectral data from different sampling points, and the matrix correction coefficients of the lithium battery are constructed based on the processed spectral data corresponding to the reference sampling points. Intensity compensation is performed point-by-point for the corresponding bands of the spectral data after processing the base sampling points and sparse sampling points, based on the matrix correction coefficient. The peak positions of the characteristic spectral lines corresponding to the valuable metals are located in the intensity-compensated spectral data. Peak area integration is performed on the characteristic spectral lines to obtain the peak intensity. The chemical composition content of the corresponding sampling points is extracted by combining the pre-stored quantitative standard curve of the composition.

6. The lithium battery valuable metal detection system as described in claim 5, characterized in that, The valuable metal content and abnormal sampling points of the output lithium battery include: According to the spatial order of the spatial distribution coordinates corresponding to different sampling points, the chemical composition content of all sampling points on the same lithium battery is sorted in sequence to form a composition sequence; Feature extraction and data fitting of component sequences are performed using a long short-term memory network. The fitting result is the time-series fusion value of the chemical component content corresponding to the sampling point. Combined with the pre-stored component quantitative standard curve, the time-series fusion value is mapped to the valuable metal content of lithium battery. By comparing the chemical component content of each sampling point in the component sequence with the average chemical component content of sampling points in the same region, abnormal sampling points are identified.

7. The lithium battery valuable metal detection system as described in claim 6, characterized in that, The triggering of resampling and detection of the corresponding area includes: if the number of abnormal sampling points is greater than a preset number threshold, locating the target area to which the abnormal sampling points belong, and supplementing the target area with corresponding sampling points, and re-exciting and detecting the valuable metal content of the lithium battery point by point according to the laser pulse energy, until the number of abnormal sampling points is not greater than the preset number threshold.

8. A method for detecting valuable metals in lithium batteries, implemented based on the lithium battery valuable metal detection system according to any one of claims 1-7, characterized in that, include: The surface morphology image of the lithium battery is acquired and region segmentation is performed to identify the positive electrode region, negative electrode region and damaged region, and the surface roughness of the lithium battery is calculated to adjust the laser pulse energy; Basic sampling points were set in the positive electrode region, sparse sampling points were set in the damaged region, and reference sampling points were set in the negative electrode region to determine the spatial distribution coordinates of different sampling points. The lithium battery was excited point by point according to the laser pulse energy to obtain the spectral data of different sampling points of the lithium battery. Matrix correction is performed on the spectral data, the chemical composition content of different sampling points is extracted, the chemical composition content of all sampling points on the same lithium battery is arranged in spatial order to form a composition sequence, the composition sequence is processed by a long short-term memory network, and the valuable metal content and abnormal sampling points of the lithium battery are output. Generate a report on the content of valuable metals and link it to the recycling process parameters. When the number of abnormal sampling points exceeds the preset threshold, trigger resampling and detection of the corresponding area.

Images