Intelligent detection method and system for energy storage battery production line processes

Abstract

This invention provides an intelligent detection method and system for energy storage battery production line processes, relating to the technical field of energy storage battery production detection. The method involves obtaining the electrical performance parameters of battery cells; collecting AC impedance and temperature data to generate a performance distribution map; identifying abnormal areas during charging and determining the defect type; performing pulse discharge tests on abnormal areas to calculate the capacity loss rate; and marking defective products based on the capacity loss rate and adjusting process parameters. This method can accurately identify performance defects in battery cells, improve detection accuracy and production efficiency, and effectively reduce the defective product rate.

Description

Technical Field

[0001] The present invention relates to energy storage battery production detection technology, and in particular to an energy storage battery production line process intelligent detection method and system. Background Art

[0002] Energy storage batteries are key components in the new energy sector, and their production quality directly impacts the performance and lifespan of the entire energy storage system. With the rapid development of energy storage technology, the demand for intelligent and automated battery production lines continues to increase. Traditional battery production lines rely primarily on manual inspection and simple electrical performance tests to determine battery quality, an approach that is inefficient and prone to missed inspections.

[0003] In recent years, some advanced battery production lines have begun to adopt automated equipment and online testing systems to evaluate battery performance by measuring parameters such as battery voltage and internal resistance. However, these methods still have some limitations. First, a single electrical performance parameter test cannot fully reflect the internal state and potential defects of the battery. Second, existing testing methods often only provide an overall performance assessment and have difficulty accurately locating abnormal areas within the battery. Finally, the lack of evaluation of the battery's dynamic performance makes it impossible to accurately predict the battery's performance during actual use.

[0004] To improve the production quality and consistency of energy storage batteries, a more comprehensive and accurate intelligent detection method is urgently needed. This method should be able to comprehensively analyze the multi-dimensional parameters of the battery, accurately locate and classify internal anomalies, and automatically adjust the production process based on the detection results, thereby continuously optimizing the production process and improving the overall performance and yield rate of the battery. Summary of the Invention

[0005] The embodiments of the present invention provide a method and a system that can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is an intelligent detection method for energy storage battery production line processes, comprising:

[0008] Obtain the electrical performance parameters of battery cells on the energy storage battery production line;

[0009] Multiple testing points are arranged along the length and circumference of the battery cell, and the AC impedance data and temperature data of the battery cell at each testing point are collected to generate a performance distribution map of the battery cell;

[0010] Charge the battery cells with constant current and constant voltage until they are fully charged, record the temperature change data of the battery cells during the charging process, identify abnormal areas of the battery cells in the performance distribution diagram, and determine the performance defect type corresponding to the abnormal area based on the distribution characteristics and temperature change data of the abnormal area;

[0011] Select corresponding pulse discharge test parameters according to the performance defect type, perform a pulse discharge test on the abnormal area to obtain the dynamic response characteristics of the abnormal area, and calculate the capacity loss rate of the abnormal area based on the dynamic response characteristics and the performance defect type;

[0012] When the capacity loss rate exceeds the preset loss rate threshold, the corresponding battery cell will be marked as defective, and the robot will be controlled to transfer the defective products to the defective product collection area. At the same time, the process parameters will be adjusted according to the distribution characteristics of the capacity loss rate, and the adjusted process parameters will be sent to the production line control system.

[0013] In an optional embodiment,

[0014] Multiple test points are arranged along the length and circumference of the battery cell. The AC impedance data and temperature data of the battery cell at each test point are collected to generate a performance distribution diagram of the battery cell, including:

[0015] Arrange multiple detection rings along the length of the battery cell, evenly arrange multiple detection points around each detection ring, construct a detection point array on the surface of the battery cell, apply an AC signal to the detection point array, collect voltage and current signals of the detection point array, and calculate AC impedance data of the detection point array;

[0016] collecting temperature data of the detection point array, establishing a corresponding relationship between AC impedance data and temperature data at each detection point of the detection point array, and digitally filtering the AC impedance data and temperature data to obtain filtered AC impedance data and filtered temperature data of the detection point array;

[0017] Based on the distribution coordinates of the detection point array, the filtered AC impedance data and filtered temperature data are reconstructed using an interpolation algorithm to generate the AC impedance distribution surface and temperature distribution surface of the battery cell surface;

[0018] Normalizing the AC impedance distribution surface and the temperature distribution surface to obtain normalized AC impedance distribution data and normalized temperature distribution data of the battery cell surface, and calculating the AC impedance weight coefficient and temperature weight coefficient of the battery cell;

[0019] The normalized AC impedance distribution data and the normalized temperature distribution data are weighted and superimposed according to the AC impedance weight coefficient and the temperature weight coefficient to generate a performance distribution diagram of the battery cell.

[0020] In an optional embodiment,

[0021] Based on the distribution coordinates of the detection point array, the filtered AC impedance data and filtered temperature data are reconstructed using an interpolation algorithm to generate the AC impedance distribution surface and temperature distribution surface of the battery cell surface, including:

[0022] Calculating the Euclidean distances between adjacent detection points in the detection point array to generate a distance matrix, calculating the detection point density distribution value based on the distance matrix, comparing the detection point density distribution value with a density threshold to perform region division, and generating detection area identification data;

[0023] Performing wavelet transform filtering on the AC impedance data and the temperature data, and partitioning and storing the filtered AC impedance data and filtered temperature data according to the detection area identification data;

[0024] Calculate the coefficient of variation of each partition, and assign an interpolation algorithm based on the comparison result of the coefficient of variation with a preset variation threshold. When the coefficient of variation is less than the variation threshold, assign the cubic spline interpolation algorithm; when the coefficient of variation is greater than or equal to the variation threshold, assign the Kriging interpolation algorithm;

[0025] Execute the assigned interpolation algorithm to perform reconstruction operations to obtain partitioned reconstructed data, and construct a physical constraint optimization function based on the internal resistance continuity and temperature conduction characteristics of the battery cell surface. Perform conjugate gradient optimization on the partitioned reconstructed data, calculate the spatial gradient value of the optimized data, determine the grid point fusion window parameters for adaptive fusion, and obtain fused reconstructed data.

[0026] The error between the data at the verification point and the fused reconstructed data is calculated. When the error value exceeds the preset error threshold, the interpolation algorithm parameters are updated and the interpolation reconstruction is returned until the error value does not exceed the preset error threshold, generating the AC impedance distribution surface and temperature distribution surface of the battery cell surface.

[0027] In an optional embodiment,

[0028] Identify abnormal areas of battery cells in the performance distribution diagram. Based on the distribution characteristics and temperature change data of the abnormal areas, determine the performance defect types corresponding to the abnormal areas, including:

[0029] Calculate the Gaussian curvature value of each detection point in the performance distribution map. When the Gaussian curvature value exceeds the background field average value, mark the corresponding point as a performance abnormality feature point. Based on the position information of the performance abnormality feature point, expand the abnormal area with the continuity of the performance value as the criterion. When the performance value difference between adjacent detection points exceeds the expansion threshold, mark it as a region boundary point. Connect the region boundary points to obtain the performance abnormality region.

[0030] Extracting area, roundness, and irregularity information of the performance abnormality region to generate a morphological feature vector, calculating a shape tensor of the performance abnormality region based on the morphological feature vector, and obtaining extended features of the performance abnormality region;

[0031] collecting temperature time series data based on the boundary range of the performance abnormality area, calculating the temperature rise rate and temperature fluctuation frequency of the temperature time series data, and obtaining the temperature change characteristics of the performance abnormality area;

[0032] Combining the morphological feature vector, the extended feature, and the temperature change feature of the performance abnormality region to construct a defect feature matrix of the performance abnormality region, and dividing the performance abnormality region into defect regions to be classified according to the defect feature matrix;

[0033] The defective area to be classified is matched with a pre-established battery cell defect feature library through differential feature coding and a double-layer local sensitive hash index structure, and the battery cell defect type corresponding to the abnormal performance area is determined based on the matching result.

[0034] In an optional embodiment,

[0035] The defective area to be classified is matched with the pre-established battery cell defect feature library through differential feature coding and a double-layer local sensitive hash index structure. The battery cell defect type corresponding to the abnormal performance area is determined based on the matching results, including:

[0036] Obtaining the morphological features, extended features, and temperature features of the defect area to be classified and performing feature encoding to generate a feature fingerprint of the defect area to be classified;

[0037] Constructing a two-layer locality-sensitive hash index structure, using a random projection hash function to perform a first-level match on the feature fingerprint with a pre-established battery cell defect feature library to determine candidate defect types, and using an angle-sensitive hash function to perform a second-level match on the candidate defect types to obtain preliminary matching results;

[0038] Calculating the discriminant information entropy of the morphological features, the extended features, and the temperature features at different time scales, determining a feature importance index based on the discriminant information entropy, and dynamically adjusting the feature weights of the preliminary matching results based on the feature importance index to obtain an optimized matching result;

[0039] Constructing a feature association graph of the optimized matching result, the feature association graph includes a feature node set, a feature association edge set, and an association weight matrix, and using a spectral clustering method to decompose the feature association graph to obtain feature subgraphs of different defect modes;

[0040] Repeated feature matching sampling is performed on the defect area to be classified, and a feature matching probability distribution is calculated. When the highest matching probability is greater than a preset probability threshold, the battery cell defect type corresponding to the defect area to be classified is determined.

[0041] In an optional embodiment,

[0042] Performing a pulse discharge test on the abnormal area to obtain the dynamic response characteristics of the abnormal area, and calculating the capacity loss rate of the abnormal area based on the dynamic response characteristics and performance defect type includes:

[0043] Performing a pulse discharge test on the abnormal area to obtain a voltage response curve, performing a wavelet transform on the voltage response curve to obtain a time-frequency domain feature spectrum, extracting frequency features, energy features, and phase features from the time-frequency domain feature spectrum, and constructing a dynamic response feature of the abnormal area;

[0044] Performing parameter identification based on electrochemical impedance spectroscopy data to obtain polarization resistance and capacitance parameters of the abnormal region, and combining the polarization resistance and capacitance parameters with the dynamic response characteristics to form a characteristic parameter set of the abnormal region;

[0045] Projecting the characteristic parameter set of the abnormal area into a high-dimensional feature space, using nonlinear feature decomposition to obtain coupling features between parameters, setting a coupling strength threshold according to the performance defect type, extracting coupling features exceeding the coupling strength threshold, and constructing a parameter coupling matrix;

[0046] A capacity decay characteristic curve is determined as a reference benchmark based on the performance defect type, and the parameter coupling matrix is ​​fitted to the capacity decay characteristic curve to obtain an actual capacity decay trend curve. The deviation of the actual capacity decay trend curve relative to the reference benchmark is calculated, and different penalty weights are set for positive deviations and negative deviations to construct an asymmetric loss function. The capacity loss rate of the abnormal area is calculated based on the asymmetric loss function.

[0047] In an optional embodiment,

[0048] Adjusting process parameters based on the distribution characteristics of capacity loss rate includes:

[0049] Statistically analyzing the capacity loss rate data of defective products, obtaining a probability density function of the capacity loss rate using a kernel density estimation method, extracting peak distribution features of the probability density function, and determining abnormal process parameters based on the number of peak distribution features;

[0050] Extracting historical data of abnormal process parameters from process parameter records corresponding to defective products, calculating a change in the abnormal process parameters based on the historical data, calculating a sensitivity coefficient based on a correspondence between the change and a change in the capacity loss rate, and calculating an adjustment amount for the process parameters based on the sensitivity coefficient;

[0051] A PID controller is used to dynamically optimize the adjustment amount of the process parameters, wherein the real-time change of the capacity loss rate is used as the input signal of the PID controller, and the output signal of the PID controller is used as the correction amount of the process parameters. The process parameters are updated based on the correction amount to achieve adjustment of the process parameters according to the distribution characteristics of the capacity loss rate.

[0052] According to a second aspect of the embodiments of the present invention,

[0053] Provided is an intelligent detection system for energy storage battery production line processes, including:

[0054] The first unit is used to obtain the electrical performance parameters of battery cells on the energy storage battery production line;

[0055] The second unit is used to arrange multiple detection points along the length and circumference of the battery cell, collect AC impedance data and temperature data of the battery cell at each detection point, and generate a performance distribution map of the battery cell;

[0056] The third unit is used to charge the battery cells with constant current and constant voltage to a fully charged state, record the temperature change data of the battery cells during the charging process, identify abnormal areas of the battery cells in the performance distribution diagram, and determine the performance defect type corresponding to the abnormal area based on the distribution characteristics and temperature change data of the abnormal area;

[0057] A fourth unit is configured to select corresponding pulse discharge test parameters according to the performance defect type, perform a pulse discharge test on the abnormal area to obtain a dynamic response characteristic of the abnormal area, and calculate a capacity loss rate of the abnormal area based on the dynamic response characteristic and the performance defect type;

[0058] The fifth unit is used to mark the corresponding battery cell as defective when the capacity loss rate exceeds the preset loss rate threshold, and control the robot to transfer the defective products to the defective product collection area. At the same time, it adjusts the process parameters according to the distribution characteristics of the capacity loss rate and sends the adjusted process parameters to the production line control system.

[0059] According to a third aspect of the embodiments of the present invention,

[0060] An electronic device is provided, comprising:

[0061] processor;

[0062] a memory for storing processor-executable instructions;

[0063] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0064] According to a fourth aspect of the embodiments of the present invention,

[0065] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0066] In this embodiment, by acquiring the electrical performance parameters and multi-point AC impedance data of battery cells, combined with temperature change data, it is possible to accurately identify abnormal areas and performance defect types in battery cells, thereby improving the accuracy and reliability of detection. Pulse discharge testing is used to obtain the dynamic response characteristics of abnormal areas, and based on this, the capacity loss rate is calculated. This allows for a quantitative assessment of the degree of performance loss in battery cells, providing an objective basis for determining defective products. Automatically adjusting process parameters based on the distribution characteristics of the capacity loss rate and providing feedback to the production line control system achieves closed-loop control of the production process, helping to continuously optimize the production process and improve product yield. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of an intelligent detection method for energy storage battery production line processes according to an embodiment of the present invention;

[0068] Figure 2 This is a simulation diagram comparing the performance of different interpolation algorithms in this embodiment;

[0069] Figure 3 This is a performance comparison chart of a double-layer locality-sensitive hash index according to an embodiment of the present invention;

[0070] Figure 4 This is a kernel density estimation analysis diagram of the capacity loss rate in an embodiment of the present invention. DETAILED DESCRIPTION

[0071] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0072] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0073] Figure 1FIG. 1 is a flow chart of an intelligent detection method for energy storage battery production line process according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0074] Obtain the electrical performance parameters of battery cells on the energy storage battery production line;

[0075] Multiple testing points are arranged along the length and circumference of the battery cell, and the AC impedance data and temperature data of the battery cell at each testing point are collected to generate a performance distribution map of the battery cell;

[0076] Charge the battery cells with constant current and constant voltage until they are fully charged, record the temperature change data of the battery cells during the charging process, identify abnormal areas of the battery cells in the performance distribution diagram, and determine the performance defect type corresponding to the abnormal area based on the distribution characteristics and temperature change data of the abnormal area;

[0077] Select corresponding pulse discharge test parameters according to the performance defect type, perform a pulse discharge test on the abnormal area to obtain the dynamic response characteristics of the abnormal area, and calculate the capacity loss rate of the abnormal area based on the dynamic response characteristics and the performance defect type;

[0078] When the capacity loss rate exceeds the preset loss rate threshold, the corresponding battery cell will be marked as defective, and the robot will be controlled to transfer the defective products to the defective product collection area. At the same time, the process parameters will be adjusted according to the distribution characteristics of the capacity loss rate, and the adjusted process parameters will be sent to the production line control system.

[0079] Energy storage battery production lines refer to automated or semi-automated production systems used for mass production of energy storage battery products. These systems typically include multiple process stages, including cell assembly, liquid filling and packaging, chemical composition and capacity analysis, and testing and screening. A battery cell is the most basic functional unit in an energy storage battery and is typically an independent, unpackaged cell. During testing, each cell is independently inspected and evaluated. Its performance directly impacts the stability and safety of the resulting battery pack.

[0080] Electrical performance parameters, including but not limited to voltage, current, resistance, internal resistance, and conductivity, are used to quantify the basic electrical characteristics of battery cells under static or dynamic operating conditions. Obtaining these parameters is the basis for subsequently evaluating battery performance distribution and identifying defective areas.

[0081] Constant-current and constant-voltage charging, a process that begins by charging the battery at a constant current until the voltage reaches a set value, then switches to a constant voltage and maintains it until the current drops to a cutoff value, is a commonly used standardized test method for battery activation and performance evaluation. Pulse discharge test parameters, including pulse current amplitude, pulse duration, and intermittent period, are adaptively adjusted based on different defect types to induce and quantify the non-steady-state discharge response behavior in abnormal areas.

[0082] In an optional embodiment, multiple detection points are arranged along the length and circumference of the battery cell, and AC impedance data and temperature data of the battery cell at each detection point are collected to generate a performance distribution diagram of the battery cell, including:

[0083] Arrange multiple detection rings along the length of the battery cell, evenly arrange multiple detection points around each detection ring, construct a detection point array on the surface of the battery cell, apply an AC signal to the detection point array, collect voltage and current signals of the detection point array, and calculate AC impedance data of the detection point array;

[0084] collecting temperature data of the detection point array, establishing a corresponding relationship between AC impedance data and temperature data at each detection point of the detection point array, and digitally filtering the AC impedance data and temperature data to obtain filtered AC impedance data and filtered temperature data of the detection point array;

[0085] Based on the distribution coordinates of the detection point array, the filtered AC impedance data and filtered temperature data are reconstructed using an interpolation algorithm to generate the AC impedance distribution surface and temperature distribution surface of the battery cell surface;

[0086] Normalizing the AC impedance distribution surface and the temperature distribution surface to obtain normalized AC impedance distribution data and normalized temperature distribution data of the battery cell surface, and calculating the AC impedance weight coefficient and temperature weight coefficient of the battery cell;

[0087] The normalized AC impedance distribution data and the normalized temperature distribution data are weighted and superimposed according to the AC impedance weight coefficient and the temperature weight coefficient to generate a performance distribution diagram of the battery cell.

[0088] In this embodiment, an array of test points is constructed on the surface of the battery cell to collect data and generate a performance distribution map. Specifically, a test ring is placed every 50 mm along the length of the battery cell, and each test ring has a test point at 45-degree angles around the circumference. For a battery cell length of 500 mm, 10 test rings are placed along the length, each with 8 test points, for a total of 80 test points in the test point array. The test points are made of gold-plated copper with a diameter of 1 mm. The probes are placed at a 30-degree angle to the battery surface to ensure good electrical contact.

[0089] When applying an AC signal to the detection point array, an impedance analyzer is used to generate the AC signal. The signal frequency range is 0.1Hz to 1000Hz, the frequency scan interval is 10 points / decade, and the AC signal amplitude is 1 / 20 of the rated capacity of the battery. The voltage signal and current signal of each detection point are collected separately by a digital multimeter with a high-precision voltage acquisition module. The sampling frequency is set to 20 times the signal frequency, and the sampling time is 5 times the signal period. When collecting data, the data of the 8 detection points on the first detection ring are collected first, and then the data of all detection rings are collected. The collected voltage and current signals are calculated using the built-in algorithm of the impedance analyzer to obtain the AC impedance data of the detection point.

[0090] A PT100 temperature sensor was installed at each test point to collect temperature data. The temperature sensor has a measurement accuracy of 0.1°C, a range of -50°C to 150°C, and a sampling frequency of 1Hz. The temperature sensor acquires data through a temperature acquisition module, using a four-wire wiring system to eliminate the effects of lead resistance. When establishing a correspondence between AC impedance data and temperature data, the test point number, location coordinates, AC impedance value, and temperature value are stored in an SQL database to facilitate subsequent data processing and analysis.

[0091] To eliminate noise interference in the collected data, a bandpass filter was used to filter the AC impedance and temperature data. The filter was fourth-order and had a passband of 0.01 Hz to 100 Hz. The real and imaginary parts of the AC impedance data were filtered separately before the filtered AC impedance data was synthesized. The temperature data was directly bandpass filtered. This filtered data effectively removed power frequency interference and high-frequency noise, preserving the valid measurement signal.

[0092] When reconstructing data based on the distributed coordinates of the test point array, a three-dimensional coordinate system is first established for the battery cell surface, with the center of one end of the battery cell as the origin, the axial direction as the Z axis, the radial direction as the R axis, and the circumferential angle as θ. The coordinates of each test point can be expressed as (R, θ, Z). A cubic spline interpolation algorithm is used to spatially reconstruct the filtered AC impedance data and temperature data, with the interpolation nodes being the position coordinates of the test points. In the axial direction, 50 interpolation points are used between adjacent test rings; in the circumferential direction, 36 interpolation points are used between adjacent test points to generate continuous AC impedance distribution surfaces and temperature distribution surfaces.

[0093] When normalizing the distribution surface, the maximum values ​​of the AC impedance and temperature data are selected as the normalization benchmarks. For lithium-ion batteries, for example, at an ambient temperature of 25°C, the AC impedance data on the battery surface range from 150mΩ to 180mΩ, and the temperature data range from 26°C to 29°C. Using 180mΩ and 29°C as the normalization benchmarks for the AC impedance and temperature, respectively, the AC impedance and temperature data are mapped to the range of 0 to 1 through normalization. Normalization eliminates dimensional differences between different physical quantities, facilitating subsequent data fusion.

[0094] The weight coefficients are calculated based on the degree of data dispersion, using the data standard deviation as the basis for weight assignment. For the example data above, the standard deviation of the normalized AC impedance data was calculated to be 0.15, and the standard deviation of the temperature-normalized data was calculated to be 0.1. Based on the ratio of the standard deviations, the AC impedance weight coefficient was determined to be 0.6, and the temperature weight coefficient was determined to be 0.4. The weight coefficient setting reflects the degree of influence of different parameters on battery performance, with parameters with larger standard deviations receiving higher weights.

[0095] The performance distribution map is generated using a weighted overlay method. The normalized AC impedance distribution data is multiplied by a weighting factor of 0.6, and the temperature distribution data is multiplied by a weighting factor of 0.4. The two are added together to obtain the performance index for each point on the battery cell surface. The performance index ranges from 0 to 1, and a jet color spectrum is used for visualization. The performance index from 0 to 1 is mapped to a gradient from blue to red. Blue areas indicate good performance (performance index close to 0), while red areas indicate poor performance (performance index close to 1).

[0096] In this embodiment, by constructing an array of detection points on the surface of the battery cell and collecting multi-dimensional data, accurate characterization and visual analysis of battery performance are achieved. The use of high-precision data acquisition equipment and reasonable signal processing methods effectively reduces the impact of measurement noise and improves data reliability. Through three-dimensional spatial interpolation reconstruction and normalization processing, a continuously distributed performance index surface is obtained, breaking through the limitations of traditional discrete point detection methods. The introduction of weight coefficients takes into account the degree of influence of different parameters on performance, making performance evaluation more scientific and reasonable. The visualization method based on chromatographic mapping intuitively displays the distribution characteristics of battery surface performance, making it easier for quality control personnel to quickly identify abnormal areas. This solution can be used for quality inspection and screening in the battery production process. By analyzing the performance distribution characteristics, process anomalies can be discovered in a timely manner, guiding process optimization, and improving product yield. It has important practical application value.

[0097] In an optional embodiment, based on the distribution coordinates of the detection point array, an interpolation algorithm is used to reconstruct the filtered AC impedance data and the filtered temperature data to generate an AC impedance distribution surface and a temperature distribution surface on the surface of the battery cell, including:

[0098] Calculating the Euclidean distances between adjacent detection points in the detection point array to generate a distance matrix, calculating the detection point density distribution value based on the distance matrix, comparing the detection point density distribution value with a density threshold to perform region division, and generating detection area identification data;

[0099] Performing wavelet transform filtering on the AC impedance data and the temperature data, and partitioning and storing the filtered AC impedance data and filtered temperature data according to the detection area identification data;

[0100] Calculate the coefficient of variation of each partition, and assign an interpolation algorithm based on the comparison result of the coefficient of variation with a preset variation threshold. When the coefficient of variation is less than the variation threshold, assign the cubic spline interpolation algorithm; when the coefficient of variation is greater than or equal to the variation threshold, assign the Kriging interpolation algorithm;

[0101] Execute the assigned interpolation algorithm to perform reconstruction operations to obtain partitioned reconstructed data, and construct a physical constraint optimization function based on the internal resistance continuity and temperature conduction characteristics of the battery cell surface. Perform conjugate gradient optimization on the partitioned reconstructed data, calculate the spatial gradient value of the optimized data, determine the grid point fusion window parameters for adaptive fusion, and obtain fused reconstructed data.

[0102] The error between the data at the verification point and the fused reconstructed data is calculated. When the error value exceeds the preset error threshold, the interpolation algorithm parameters are updated and the interpolation reconstruction is returned until the error value does not exceed the preset error threshold, generating the AC impedance distribution surface and temperature distribution surface of the battery cell surface.

[0103] For example, we first calculate the Euclidean distances between adjacent detection points in the detection point array to generate a distance matrix. Specifically, we select any two adjacent detection points, calculate the difference in their coordinates in 3D space, and then take the square root of the sum to obtain the Euclidean distance. This process is repeated for all pairs of adjacent points, ultimately resulting in an NxN distance matrix, where N is the total number of detection points.

[0104] The density distribution value of each detection point is calculated based on the distance matrix. A kernel density estimation method can be used. With each detection point as the center, the distances from other points to that point are calculated, and the Gaussian kernel function is applied to obtain the density value. For example, the density value of a detection point may be 0.85. The density distribution value of the detection point is then compared with a preset density threshold (such as 0.7) to divide the detection area. Points with density values ​​above the threshold are classified as high-density areas, while points below the threshold are classified as low-density areas. This generates detection area identification data.

[0105] The original collected AC impedance and temperature data are processed through wavelet transform filtering. The db4 wavelet can be used to perform a five-layer decomposition of the data, removing high-frequency noise and reconstructing the filtered data. Based on the detection area identifiers generated previously, the filtered data are stored in the corresponding high-density and low-density area datasets. The coefficient of variation (CV) for each partition is calculated. The CV is equal to the standard deviation divided by the mean and reflects the degree of data dispersion. For example, the CV coefficient of the AC impedance data for a high-density area is 0.15, while that for a low-density area is 0.35.

[0106] The calculated coefficient of variation is compared with a preset variation threshold (e.g., 0.25) to assign appropriate interpolation algorithms to different regions. Regions with a coefficient of variation less than the threshold use the cubic spline interpolation algorithm, while regions with a coefficient of variation greater than or equal to the threshold use the kriging interpolation algorithm. The assigned interpolation algorithm is then executed to perform the reconstruction operation. For cubic spline interpolation, cubic spline functions are constructed in the x and y directions, and then interpolated in the z direction to obtain the reconstructed data. For kriging interpolation, the experimental variogram is first calculated, the theoretical variogram model is fitted, and then the kriging equation is used for interpolation.

[0107] A physically constrained optimization function is constructed based on the internal resistance continuity and temperature conductivity characteristics of the battery cell surface. Constraints can be set for the smoothness of the internal resistance and temperature gradients, as well as the correlation between internal resistance and temperature. The conjugate gradient method is applied to the partitioned reconstructed data, and the optimization is iterated until the objective function converges or the maximum number of iterations is reached.

[0108] Calculate the spatial gradient of the optimized data and determine the grid point fusion window parameters. Use the central difference method to calculate the gradient, and dynamically adjust the fusion window size based on the gradient value. The larger the gradient value, the smaller the window. Use the determined window parameters to perform a weighted average fusion of the partition boundary data to obtain the fused and reconstructed data.

[0109] Select a few validation points, compare the measured data with the fused reconstructed data, and calculate the root mean square error (RMS). If the error exceeds a preset threshold (e.g., 5%), update the interpolation algorithm parameters and re-run the interpolation and reconstruction process. This can be done by adjusting the variogram model parameters of the kriging interpolation or modifying the knot positions of the cubic spline. Repeat this process until the error falls below the preset threshold.

[0110] Based on the fused reconstructed data that meets the accuracy requirements, a 3D surface fitting method is used to generate the AC impedance distribution surface and temperature distribution surface of the battery cell surface. A bicubic spline surface fitting algorithm can be used to divide the grid on the xy plane, construct a bicubic polynomial surface patch for each grid cell, and finally splice it to obtain the complete distribution surface.

[0111] Figure 2 This is a simulation diagram comparing the performance of different interpolation algorithms in this embodiment. Figure 2 The figure compares the reconstruction performance of cubic spline interpolation and kriging interpolation in regions with different data characteristics. The left side shows the reconstruction performance in regions with smooth data, and the right side shows the reconstruction performance in regions with abrupt data changes. The curve shows how the reconstruction error changes with the density of interpolation points. Simulation results confirm that cubic spline interpolation has lower computational complexity and good reconstruction accuracy in smooth regions, while kriging interpolation better preserves local features in regions with abrupt data changes. Compared to a single interpolation method, this adaptive interpolation strategy significantly reduces the computational burden while maintaining reconstruction accuracy.

[0112] In the prior art, a single interpolation algorithm is usually used to reconstruct battery surface data, without considering the distribution characteristics of the detection points and the variability of the data, resulting in insufficient reconstruction accuracy. The present application divides the area by calculating the detection point density and the data variation coefficient, adaptively assigns different interpolation algorithms according to the data characteristics, and introduces a physical constraint optimization function to optimize the reconstruction results, thereby achieving high-precision reconstruction of the battery surface performance distribution. Compared with the prior art, the solution proposed in this application can adapt to the data characteristics of different detection areas, and uses Kriging interpolation to ensure reconstruction accuracy in sparse point areas and areas with large data fluctuations, and uses cubic spline interpolation to improve computational efficiency in dense point areas and areas with stable data. Through physical constraint optimization and adaptive fusion of grid points, it is ensured that the reconstruction results conform to the continuity of the battery internal resistance distribution and the conduction characteristics of the temperature field, thereby improving the physical rationality of the reconstruction results. The error feedback iterative mechanism ensures that the reconstruction accuracy meets the requirements, and ultimately achieves accurate characterization of the battery surface performance distribution, providing a reliable basis for battery performance evaluation and quality control.

[0113] In an optional embodiment, identifying an abnormal area of ​​a battery cell in a performance distribution diagram, and determining a performance defect type corresponding to the abnormal area based on distribution characteristics and temperature change data of the abnormal area includes:

[0114] Calculate the Gaussian curvature value of each detection point in the performance distribution map. When the Gaussian curvature value exceeds the background field average value, mark the corresponding point as a performance abnormality feature point. Based on the position information of the performance abnormality feature point, expand the abnormal area with the continuity of the performance value as the criterion. When the performance value difference between adjacent detection points exceeds the expansion threshold, mark it as a region boundary point. Connect the region boundary points to obtain the performance abnormality region.

[0115] Extracting area, roundness, and irregularity information of the performance abnormality region to generate a morphological feature vector, calculating a shape tensor of the performance abnormality region based on the morphological feature vector, and obtaining extended features of the performance abnormality region;

[0116] collecting temperature time series data based on the boundary range of the performance abnormality area, calculating the temperature rise rate and temperature fluctuation frequency of the temperature time series data, and obtaining the temperature change characteristics of the performance abnormality area;

[0117] Combining the morphological feature vector, the extended feature, and the temperature change feature of the performance abnormality region to construct a defect feature matrix of the performance abnormality region, and dividing the performance abnormality region into defect regions to be classified according to the defect feature matrix;

[0118] The defective area to be classified is matched with a pre-established battery cell defect feature library through differential feature coding and a double-layer local sensitive hash index structure, and the battery cell defect type corresponding to the abnormal performance area is determined based on the matching result.

[0119] In this embodiment, the Gaussian curvature value of each detection point in the performance distribution diagram is first calculated. The Gaussian curvature calculation is achieved by analyzing the performance value change trend of each detection point and its surrounding points. Specifically, for each point (i, j) in the performance distribution diagram, the points in the 3×3 area around it are selected to construct a local surface, and then the Gaussian curvature value K(i, j) of the point is calculated. In practical applications, when the Gaussian curvature value K(i, j) of a certain point exceeds the background field mean Kmean, the point is marked as a performance abnormality feature point. For example, in the above-mentioned 18650 battery case, the background field Gaussian curvature mean Kmean is 0.0023. When a point K(i, j) value is detected to be 0.0089, the point is marked as an abnormal feature point.

[0120] Based on the marked performance anomaly feature points, the abnormal area is expanded. Starting from each abnormal feature point, the performance values ​​of adjacent points are expanded in all directions. When the performance value difference between adjacent points exceeds the preset expansion threshold, the point is marked as a region boundary point. In actual cases, the expansion threshold is set to 2 times the standard deviation of the performance value, which is approximately 0.15. By connecting all boundary points, a closed performance anomaly area outline is formed. In the above case, three obvious abnormal areas are identified, located in the upper, middle and lower areas of the battery.

[0121] The morphological features of the performance anomaly areas are extracted. The area S, circularity C, and irregularity I of each anomaly area are calculated to form a morphological feature vector [S, C, I]. Area S is calculated by counting the number of pixels within the area; circularity C is calculated by comparing the ratio of the area's perimeter to the perimeter of a circle of equal area; and irregularity I is expressed as the standard deviation of the curvature of the area boundary. In this case, the morphological feature vector of the upper anomaly area is [324, 0.78, 0.35], indicating an area of ​​324 pixels, a circularity of 0.78, and an irregularity of 0.35.

[0122] Based on the morphological eigenvectors, the shape tensor of the performance anomaly region is calculated to obtain expansion features. The shape tensor describes the primary expansion direction and extent of the anomaly region. By analyzing the distribution of points within the region and calculating the eigenvalues ​​and eigenvectors of the shape tensor, the region's principal axis direction and major-minor axis ratio are determined. In this case, the principal axis direction of the central anomaly region is 67 degrees, and the major-minor axis ratio is 2.3, indicating that the region exhibits an elliptical expansion along the 67-degree direction.

[0123] Temperature time series data is collected based on the boundary of the performance anomaly region. An infrared thermal imager is used to continuously monitor the anomaly region, recording temperature data every 10 seconds for 30 minutes. The temperature rise rate dT / dt and temperature fluctuation frequency f are calculated as temperature variation features. In this case, the temperature rise rate of the lower anomaly region is 0.15°C / minute, and the temperature fluctuation frequency is 0.033 Hz, indicating slow but continuous heating in this region. The morphological feature vectors, expansion features, and temperature variation features are combined to construct a defect feature matrix M for the performance anomaly region. Matrix M encompasses the geometry, expansion trend, and thermodynamic properties of the anomaly region, providing multidimensional feature information for defect type identification. In this case, the defect feature matrix for the upper anomaly region is [324, 0.78, 0.35, 43, 1.8, 0.08, 0.012], where the first three values ​​represent morphological features, the middle two represent expansion features, and the last two represent temperature variation features. Finally, using differential feature encoding and a two-layer locality-sensitive hashing index structure, the defect region to be classified is matched against a pre-established battery cell defect feature library. The defect feature library contains common battery defect types, such as lithium dendrites, SEI film anomalies, and active material shedding. Each defect type has a corresponding feature matrix range. By calculating the similarity between the feature matrix of the region to be classified and the feature matrix of each type in the library, the most matching defect type is determined. In this case, the upper abnormal region was identified as the lithium dendrite growth region, the middle abnormal region was identified as the SEI film anomaly region, and the lower abnormal region was identified as the active material shedding region, with similarities of 92.3%, 88.7%, and 95.1%, respectively.

[0124] In this embodiment, the abnormal feature points are identified by calculating the Gaussian curvature value of the performance distribution map, and the region is expanded based on the continuity of the performance value, thereby achieving accurate positioning of the abnormal region. By extracting the morphological features, expansion features, and temperature change features of the abnormal region, a multi-dimensional defect feature matrix is ​​constructed, and an objective defect characterization system is established. Differential feature coding and a two-layer locally sensitive hash index structure are used for feature matching to improve the accuracy and efficiency of defect type identification. This solution breaks through the limitations of traditional manual judgment, realizes the automatic identification and classification of battery performance defects, and provides reliable technical support for battery quality control and defective product analysis. At the same time, the established defect feature library can be continuously accumulated and updated to continuously improve the defect identification capability.

[0125] In an optional embodiment, the defective area to be classified is matched with a pre-established battery cell defect feature library through differential feature coding and a double-layer local sensitive hash index structure, and the battery cell defect type corresponding to the abnormal performance area is determined based on the matching result, including:

[0126] Obtaining the morphological features, extended features, and temperature features of the defect area to be classified and performing feature encoding to generate a feature fingerprint of the defect area to be classified;

[0127] Constructing a two-layer locality-sensitive hash index structure, using a random projection hash function to perform a first-level match on the feature fingerprint with a pre-established battery cell defect feature library to determine candidate defect types, and using an angle-sensitive hash function to perform a second-level match on the candidate defect types to obtain preliminary matching results;

[0128] Calculating the discriminant information entropy of the morphological features, the extended features, and the temperature features at different time scales, determining a feature importance index based on the discriminant information entropy, and dynamically adjusting the feature weights of the preliminary matching results based on the feature importance index to obtain an optimized matching result;

[0129] Constructing a feature association graph of the optimized matching result, the feature association graph includes a feature node set, a feature association edge set, and an association weight matrix, and using a spectral clustering method to decompose the feature association graph to obtain feature subgraphs of different defect modes;

[0130] Repeated feature matching sampling is performed on the defect area to be classified, and a feature matching probability distribution is calculated. When the highest matching probability is greater than a preset probability threshold, the battery cell defect type corresponding to the defect area to be classified is determined.

[0131] This embodiment first obtains the morphological features, extended features, and temperature features of the defect area to be classified, and performs feature encoding to generate a feature fingerprint of the defect area to be classified. Specifically, the morphological features include geometric parameters such as the area, perimeter, and roundness of the defect area; the extended features include the grayscale value distribution and texture features of the defect area; and the temperature features include the maximum temperature, average temperature, and temperature gradient of the defect area. When encoding these features, a 256-bit binary encoding method is used, in which the morphological features occupy 64 bits, the extended features occupy 128 bits, and the temperature features occupy 64 bits. For example, the feature fingerprint of a defect area to be classified may be: 1010...1101 (morphological features) 1100...0011 (extended features) 0101...1010 (temperature features).

[0132] A two-layer locality-sensitive hash index structure is constructed. The first layer uses a random projection hash function to match the feature fingerprint with a pre-established library of battery cell defect signatures to determine candidate defect types. In specific implementation, 20 8-bit random projection hash functions are used to map the 256-bit feature fingerprint into 20 8-bit hash values. Samples with the same hash value are then searched in the signature library as candidate defect types. The second layer uses an angle-sensitive hash function to further match the candidate defect types and obtain preliminary matching results. The angle-sensitive hash function uses 30 hyperplanes to divide the feature space into different regions and calculates the probability that the sample to be classified and the candidate sample fall in the same region. A higher probability indicates a higher degree of match.

[0133] The discriminant information entropy of morphological, extended, and temperature features at different time scales is calculated, and the feature importance index is determined based on the discriminant information entropy. Specifically, the discriminant information entropy of each feature is calculated at time scales of 1 minute, 5 minutes, 30 minutes, and 1 hour. A higher discriminant information entropy indicates a stronger discriminative ability at that time scale. The feature importance index is then calculated by taking the weighted average of the discriminant information entropies at different time scales. For example, the importance index of the morphological feature is 0.4, that of the extended feature is 0.35, and that of the temperature feature is 0.25. Based on the feature importance index, the feature weights of the preliminary matching results are dynamically adjusted to obtain an optimized matching result.

[0134] A feature association graph of the optimized matching results is constructed. The feature association graph consists of a feature node set, a feature association edge set, and an association weight matrix. The feature node set contains all matched features, the feature association edge set represents the association relationship between features, and the association weight matrix describes the strength of the association between features. For example, the feature association graph of a matching result may contain 10 feature nodes and 30 association edges, with association weights between 0 and 1. Spectral clustering is used to decompose the feature association graph to obtain feature subgraphs for different defect patterns. Spectral clustering first calculates the Laplacian matrix of the graph, then performs eigenvalue decomposition on the matrix. Finally, the K-means algorithm is used to cluster the feature vectors to obtain feature subgraphs.

[0135] Repeated feature matching sampling is performed on the defective area to be classified, and the feature matching probability distribution is calculated. Specifically, repeated sampling is performed 100 times, each time randomly selecting 80% of the features for matching. The number of matches for each defect type is counted, and the matching probability is calculated. When the highest matching probability exceeds a preset probability threshold (e.g., 0.8), the battery cell defect type corresponding to the defective area to be classified is determined.

[0136] Suppose a suspected defective area is detected on a battery production line, and its defect type needs to be determined. First, the characteristics of the area are obtained: in terms of morphological characteristics, the area is 2.5 square centimeters, the perimeter is 6.8 centimeters, and the roundness is 0.85; in terms of extended characteristics, the average grayscale value is 128 and the texture complexity is 0.6; in terms of temperature characteristics, the maximum temperature is 45°C, the average temperature is 42°C, and the temperature gradient is 3°C / cm. These features are encoded into a 256-bit binary feature fingerprint. Then, a two-layer local sensitive hash index structure is used for matching. After the first layer of random projection hash matching, five candidate defect types are obtained: cracks, bubbles, foreign matter, wrinkles, and deformation. After the second layer of angle sensitive hash matching, preliminary matching results are obtained: cracks (matching degree 0.8), bubbles (matching degree 0.6), and foreign matter (matching degree 0.4).

[0137] Then, the feature importance index was calculated, resulting in a morphological feature of 0.45, an extended feature of 0.3, and a temperature feature of 0.25. Based on these weights, the matching results were adjusted to obtain the optimized matching results: cracks (matching degree 0.75) and bubbles (matching degree 0.65). A feature association graph was constructed, containing 8 feature nodes and 20 associated edges. Spectral clustering was used to decompose the graphs to obtain two feature subgraphs, corresponding to the two defect modes of cracks and bubbles, respectively. Finally, 100 repeated feature matching samples were performed to obtain the matching probability distribution: cracks 85%, bubbles 12%, and others 3%. Since the matching probability of 85% for cracks is greater than the preset threshold of 80%, the type of the defect area is determined to be a crack.

[0138] Figure 3 This is a performance comparison chart of the double-layer locality sensitive hash index according to the embodiment of the present invention. Figure 3The figure shows four performance curves comparing the query time of different feature matching schemes for battery cell defect identification. The horizontal axis represents the feature library size, ranging from 1,000 to 100,000 entries; the vertical axis represents the query time, ranging from 0 to 120 milliseconds. The four different schemes are distinguished by square, circle, diamond, and triangle markers, respectively. The proposed dual-layer locality-sensitive hashing index structure, which uses a combination of random projection hashing and angle-sensitive hashing (denoted by triangles in the figure), shows the most gradual increase in query time, from 8.2 milliseconds with a feature library of 1,000 entries to 23.8 milliseconds with 100,000 entries. The traditional single-layer LSH scheme, shown by squares, saw its query time increase from 12.5 milliseconds to 67.3 milliseconds. The cosine similarity-based scheme, shown by circles, saw its query time increase from 18.7 milliseconds to 98.6 milliseconds. The Euclidean distance-based scheme, shown by diamonds, saw its query time increase from 21.4 milliseconds to 105.2 milliseconds. The performance curve clearly demonstrates that the dual-layer locality-sensitive hash index structure significantly improves query efficiency while maintaining matching accuracy through differentiated feature encoding and a dual-layer matching mechanism. This solution demonstrates significant performance advantages, particularly in large-scale feature library scenarios, with slow query time growth and good scalability, providing efficient technical support for the rapid identification of battery cell defects.

[0139] In the prior art, the matching of battery defect features usually adopts single feature comparison or simple similarity calculation, without considering the temporal changes and correlation of features, resulting in low defect recognition accuracy and susceptibility to noise interference. The present application realizes fast coarse matching and precise matching of features by constructing a two-layer local sensitive hash index structure, which greatly improves the matching efficiency. The feature weights are dynamically adjusted by calculating the discriminant information entropy of features at different time scales, so that the matching process can adapt to changes in the importance of features. Constructing a feature association graph and performing spectral clustering decomposition reveals the deep correlation between defect features and improves the accuracy of defect pattern recognition. The repeated feature matching sampling and probability distribution calculation method are used to effectively reduce the impact of random errors and enhance the reliability of defect type determination. This scheme breaks through the limitations of traditional single feature matching and establishes a complete multi-feature fusion recognition mechanism, which significantly improves the accuracy and robustness of battery defect type identification and provides more reliable technical support for battery quality control.

[0140] In an optional embodiment, performing a pulse discharge test on the abnormal region to obtain a dynamic response characteristic of the abnormal region, and calculating the capacity loss rate of the abnormal region based on the dynamic response characteristic and the performance defect type includes:

[0141] Performing a pulse discharge test on the abnormal area to obtain a voltage response curve, performing a wavelet transform on the voltage response curve to obtain a time-frequency domain feature spectrum, extracting frequency features, energy features, and phase features from the time-frequency domain feature spectrum, and constructing a dynamic response feature of the abnormal area;

[0142] Performing parameter identification based on electrochemical impedance spectroscopy data to obtain polarization resistance and capacitance parameters of the abnormal region, and combining the polarization resistance and capacitance parameters with the dynamic response characteristics to form a characteristic parameter set of the abnormal region;

[0143] Projecting the characteristic parameter set of the abnormal area into a high-dimensional feature space, using nonlinear feature decomposition to obtain coupling features between parameters, setting a coupling strength threshold according to the performance defect type, extracting coupling features exceeding the coupling strength threshold, and constructing a parameter coupling matrix;

[0144] A capacity decay characteristic curve is determined as a reference benchmark based on the performance defect type, and the parameter coupling matrix is ​​fitted to the capacity decay characteristic curve to obtain an actual capacity decay trend curve. The deviation of the actual capacity decay trend curve relative to the reference benchmark is calculated, and different penalty weights are set for positive deviations and negative deviations to construct an asymmetric loss function. The capacity loss rate of the abnormal area is calculated based on the asymmetric loss function.

[0145] For example, a pulse discharge test is first performed on the abnormal area to obtain a voltage response curve. During the test, a constant current pulse discharge method can be used, the pulse current amplitude is set to 1C, the pulse duration is 10ms, the pulse interval is 100ms, and 100 pulses are applied continuously. The voltage response of the entire process is recorded by a high-precision voltage acquisition device with a sampling frequency of not less than 1kHz to obtain a curve of voltage change over time. The obtained voltage response curve is subjected to wavelet transform to obtain a time-frequency domain characteristic spectrum. The db4 wavelet basis function can be selected to perform a 5-layer wavelet decomposition. The time-frequency domain characteristic spectrum obtained after decomposition contains the energy distribution information of the voltage response at different frequencies and time scales.

[0146] Frequency, energy, and phase features are extracted from the time-frequency domain feature map. The frequency feature selects the center frequency of each frequency band; the energy feature calculates the energy percentage of each frequency band; and the phase feature extracts the phase angle of the signal in each frequency band. For example, the center frequencies of the five frequency bands can be 10 Hz, 50 Hz, 100 Hz, 500 Hz, and 1 kHz, respectively, with corresponding energy percentages of 15%, 25%, 30%, 20%, and 10%, and phase angles of 30°, 45°, 60°, 90°, and 120°, respectively. These features together constitute the dynamic response characteristics of the abnormal region. Parameter identification is then performed based on the electrochemical impedance spectroscopy data to obtain the polarization resistance and capacitance parameters of the abnormal region. Equivalent circuit fitting can be used, using the Randles equivalent circuit model and the least squares method to fit the impedance spectroscopy data to obtain the polarization resistance Rp and double-layer capacitance Cdl values. For example, the fitting results in Rp = 0.5Ω and Cdl = 10mF.

[0147] Polarization resistance and capacitance parameters are combined with dynamic response characteristics to form a characteristic parameter set for abnormal regions. This parameter set includes the frequency characteristics, energy characteristics, phase characteristics, as well as polarization resistance and capacitance parameters extracted above, for a total of 13 parameters.

[0148] Project the characteristic parameter set of the abnormal region into a high-dimensional feature space. Kernel principal component analysis (KPCA) can be used with a Gaussian kernel function to map the 13-dimensional parameters into a 100-dimensional feature space. In this high-dimensional feature space, nonlinear eigendecomposition is used to extract the coupling characteristics between the parameters. Kernelized independent component analysis (ICA) can be used to extract nonlinear correlations between the parameters.

[0149] Set a coupling strength threshold based on the type of performance defect. For example, for lithium-ion battery capacity decay, the coupling strength threshold can be set to 0.7. Coupling features exceeding this threshold are extracted and a parameter coupling matrix is ​​constructed. This matrix reflects the strong correlations between parameters and may have fewer dimensions than the original number of parameters.

[0150] Determine a capacity decay characteristic curve based on the type of performance defect as a reference benchmark. For lithium-ion batteries, the capacity decay curve under standard cycle life testing can be used as a benchmark, typically exhibiting an exponential decay pattern. Fit the parameter coupling matrix to the capacity decay characteristic curve using support vector regression with a radial basis kernel function to obtain the actual capacity decay trend curve.

[0151] Calculate the deviation of the actual capacity attenuation trend curve relative to the reference benchmark. The capacity difference between the two curves can be calculated at multiple cycle points to obtain a series of deviation values. Set different penalty weights for positive deviations and negative deviations to construct an asymmetric loss function. For example, the weight for the positive deviation (actual capacity is higher than the benchmark) can be set to 0.8, and the weight for the negative deviation (actual capacity is lower than the benchmark) can be set to 1.2, reflecting a conservative estimate of capacity loss. The capacity loss rate of the abnormal area is calculated based on the asymmetric loss function. Specifically, the weighted average of each deviation point can be used as the final capacity loss rate. For example, if the calculated weighted average deviation is 5%, it means that the capacity loss rate of the abnormal area is 5%.

[0152] In this embodiment, the dynamic response characteristics are obtained by pulse discharge testing, and the parameters are identified by combining the electrochemical impedance spectroscopy data to fully characterize the dynamic performance characteristics of the abnormal area. The nonlinear characteristic decomposition method is used to reveal the coupling relationship between the parameters, overcoming the limitation of the traditional linear analysis method that it is difficult to express the coupling effect of complex parameters. According to the type of performance defect, the appropriate capacity decay characteristic curve is selected as the reference benchmark, and the capacity loss is evaluated by the asymmetric loss function, which effectively distinguishes the degree of influence of different types of defects on capacity decay. The accurate quantitative evaluation of the capacity loss in the abnormal area of ​​the battery is achieved, which provides a reliable basis for battery performance prediction and life evaluation. At the same time, the established evaluation method has strong versatility and can be applied to different types of battery defect analysis.

[0153] In practical applications, the parameters in each step can be adjusted appropriately based on the specific battery type and usage scenario. For example, the parameters of the pulse discharge test can be set based on the battery's rated capacity and maximum allowable discharge rate; the number of wavelet transform layers can be selected based on the required frequency resolution; and the parameter coupling strength threshold can be determined based on different types of performance defects. Through these flexible adjustments, the method can be adapted to various battery types and diverse application scenarios.

[0154] The last eight hours of historical data for these two parameters were extracted from the process parameter records in the production management system, with a recording interval of one minute. For the roller pressure parameter, the set value was 15 MPa, but the actual fluctuation range was 14.5 MPa to 15.5 MPa. The set value for the coating thickness parameter was 80 μm, but the actual fluctuation range was 78 μm to 82 μm. By analyzing this historical data, the changing trends of the process parameters were calculated.

[0155] The change data of process parameters are paired with the battery capacity loss rate data produced in the corresponding period in time, and the corresponding relationship between the parameter change and the capacity loss rate change is established. The difference between the parameter value at each time point and the previous time point is taken as the parameter change, and the difference between the capacity loss rate of the corresponding batch of batteries and the baseline value is taken as the capacity loss rate change. The sensitivity coefficient is calculated using these paired data. If a change of 0.1MPa in roller pressure causes a change of approximately 0.5% in capacity loss rate, the sensitivity coefficient is 5; a change of 1μm in coating thickness causes a change of approximately 0.3% in capacity loss rate, and the sensitivity coefficient is 0.3. Based on these sensitivity coefficients, combined with the deviation between the current capacity loss rate and the target value, the adjustment amount of the process parameters is calculated. For example, when the capacity loss rate is 8% higher than the target value, it is recommended to reduce the roller pressure by 0.2MPa and increase the coating thickness by 2μm.

[0156] To achieve smooth adjustment of process parameters, a PID controller was used for dynamic optimization. The PID controller's proportional coefficient was set to 0.8, the integral time constant to 120 seconds, and the differential time constant to 30 seconds. The deviation between the real-time acquired capacity loss rate and the target value was used as the PID controller's input signal. The controller's output adjustment signal, after amplitude limiting, served as the process parameter correction. The roll pressure correction was limited to ±0.5 MPa, and the coating thickness correction was limited to ±3 μm.

[0157] In actual application, the control system collects capacity loss rate data every minute. After the PID controller calculates the correction value, the industrial computer sends the corrected parameters to the corresponding equipment controller. After receiving the new parameters, the equipment controller adjusts the equipment operation through the inverter or servo system to achieve real-time update of process parameters. The system also records the changes in capacity loss rate before and after the parameter adjustment to evaluate the adjustment effect.

[0158] Figure 4 This is a kernel density estimation analysis diagram of the capacity loss rate according to an embodiment of the present invention, as shown in FIG. Figure 4The figure shows the results of analyzing the distribution characteristics of the capacity loss rate of defective battery products using the kernel density estimation method. The figure clearly shows that by processing the data of 100 defective product samples with a Gaussian kernel function (bandwidth 0.8), two significant peaks were identified in the capacity loss rate distribution, located at 7% and 12%, with peak probability densities reaching 0.19 and 0.18, respectively. This bimodal distribution indicates the existence of two major types of defective patterns, which are difficult to accurately capture using traditional histogram statistics. As can be seen from the figure, compared with the step-like distribution presented by traditional histogram statistics (using discrete interval statistics), the kernel density estimation method of this technical solution can more accurately describe the continuous distribution characteristics of the data, providing a reliable mathematical basis for the subsequent identification of abnormal process parameters. The figure verifies that the kernel density estimation method used in this technical solution can accurately identify the distribution pattern of the capacity loss rate, providing a data foundation for locating abnormal process parameters.

[0159] In this example, kernel density estimation was used to analyze the distribution characteristics of the capacity loss rate, enabling accurate identification of unqualified patterns and precise location of abnormal process parameters. By establishing a correspondence between process parameter changes and capacity loss rate changes, sensitivity coefficients were calculated, and a quantitative basis for parameter adjustment was established. Dynamic optimization using a PID controller enabled real-time automatic adjustment of process parameters, avoiding the lag and subjectivity of manual adjustments. This method achieved adaptive optimization of process parameters based on product quality characteristics, significantly improving product quality stability and reducing the failure rate. The method is highly versatile and can be applied to the optimization and adjustment of other process parameters.

[0160] According to a second aspect of the embodiments of the present invention,

[0161] Provided is an intelligent detection system for energy storage battery production line processes, the system comprising:

[0162] The first unit is used to obtain the electrical performance parameters of battery cells on the energy storage battery production line;

[0163] The second unit is used to arrange multiple detection points along the length and circumference of the battery cell, collect AC impedance data and temperature data of the battery cell at each detection point, and generate a performance distribution map of the battery cell;

[0164] The third unit is used to charge the battery cells with constant current and constant voltage to a fully charged state, record the temperature change data of the battery cells during the charging process, identify abnormal areas of the battery cells in the performance distribution diagram, and determine the performance defect type corresponding to the abnormal area based on the distribution characteristics and temperature change data of the abnormal area;

[0165] A fourth unit is configured to select corresponding pulse discharge test parameters according to the performance defect type, perform a pulse discharge test on the abnormal area to obtain a dynamic response characteristic of the abnormal area, and calculate a capacity loss rate of the abnormal area based on the dynamic response characteristic and the performance defect type;

[0166] The fifth unit is used to mark the corresponding battery cell as defective when the capacity loss rate exceeds the preset loss rate threshold, and control the robot to transfer the defective products to the defective product collection area. At the same time, it adjusts the process parameters according to the distribution characteristics of the capacity loss rate and sends the adjusted process parameters to the production line control system.

[0167] According to a third aspect of the embodiments of the present invention,

[0168] An electronic device is provided, comprising:

[0169] processor;

[0170] a memory for storing processor-executable instructions;

[0171] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0172] According to a fourth aspect of the embodiments of the present invention,

[0173] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0174] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent detection method for energy storage battery production line process, characterized in that: include: Obtain the electrical performance parameters of battery cells on the energy storage battery production line; Multiple testing points are arranged along the length and circumference of the battery cell, and the AC impedance data and temperature data of the battery cell at each testing point are collected to generate a performance distribution map of the battery cell; Charge the battery cells with constant current and constant voltage until they are fully charged, record the temperature change data of the battery cells during the charging process, identify abnormal areas of the battery cells in the performance distribution diagram, and determine the performance defect type corresponding to the abnormal area based on the distribution characteristics and temperature change data of the abnormal area; Select corresponding pulse discharge test parameters according to the performance defect type, perform a pulse discharge test on the abnormal area to obtain the dynamic response characteristics of the abnormal area, and calculate the capacity loss rate of the abnormal area based on the dynamic response characteristics and the performance defect type; When the capacity loss rate exceeds the preset loss rate threshold, the corresponding battery cell is marked as defective and the robot is controlled to transfer the defective product to the defective product collection area. At the same time, the process parameters are adjusted according to the distribution characteristics of the capacity loss rate and the adjusted process parameters are sent to the production line control system. Identify abnormal areas of battery cells in the performance distribution diagram. Based on the distribution characteristics and temperature change data of the abnormal areas, determine the performance defect types corresponding to the abnormal areas, including: Calculate the Gaussian curvature value of each detection point in the performance distribution map. When the Gaussian curvature value exceeds the background field average value, mark the corresponding point as a performance abnormality feature point. Based on the position information of the performance abnormality feature point, expand the abnormal area with the continuity of the performance value as the criterion. When the performance value difference between adjacent detection points exceeds the expansion threshold, mark it as a region boundary point. Connect the region boundary points to obtain the performance abnormality region. Extracting area, roundness, and irregularity information of the performance abnormality region to generate a morphological feature vector, calculating a shape tensor of the performance abnormality region based on the morphological feature vector, and obtaining extended features of the performance abnormality region; collecting temperature time series data based on the boundary range of the performance abnormality area, calculating the temperature rise rate and temperature fluctuation frequency of the temperature time series data, and obtaining the temperature change characteristics of the performance abnormality area; Combining the morphological feature vector, the extended feature, and the temperature change feature of the performance abnormality region to construct a defect feature matrix of the performance abnormality region, and dividing the performance abnormality region into defect regions to be classified according to the defect feature matrix; The defective area to be classified is matched with a pre-established battery cell defect feature library through differential feature coding and a double-layer local sensitive hash index structure, and the battery cell defect type corresponding to the abnormal performance area is determined based on the matching result.

2. The method according to claim 1, characterized in that Multiple test points are arranged along the length and circumference of the battery cell. The AC impedance data and temperature data of the battery cell at each test point are collected to generate a performance distribution diagram of the battery cell, including: Arrange multiple detection rings along the length of the battery cell, evenly arrange multiple detection points around each detection ring, construct a detection point array on the surface of the battery cell, apply an AC signal to the detection point array, collect voltage and current signals of the detection point array, and calculate AC impedance data of the detection point array; collecting temperature data of the detection point array, establishing a corresponding relationship between AC impedance data and temperature data at each detection point of the detection point array, and digitally filtering the AC impedance data and temperature data to obtain filtered AC impedance data and filtered temperature data of the detection point array; Based on the distribution coordinates of the detection point array, the filtered AC impedance data and filtered temperature data are reconstructed using an interpolation algorithm to generate the AC impedance distribution surface and temperature distribution surface of the battery cell surface; Normalizing the AC impedance distribution surface and the temperature distribution surface to obtain normalized AC impedance distribution data and normalized temperature distribution data of the battery cell surface, and calculating the AC impedance weight coefficient and temperature weight coefficient of the battery cell; The normalized AC impedance distribution data and the normalized temperature distribution data are weighted and superimposed according to the AC impedance weight coefficient and the temperature weight coefficient to generate a performance distribution diagram of the battery cell.

3. The method according to claim 2, characterized in that Based on the distribution coordinates of the detection point array, the filtered AC impedance data and filtered temperature data are reconstructed using an interpolation algorithm to generate the AC impedance distribution surface and temperature distribution surface of the battery cell surface, including: Calculating the Euclidean distances between adjacent detection points in the detection point array to generate a distance matrix, calculating the detection point density distribution value based on the distance matrix, comparing the detection point density distribution value with a density threshold to perform region division, and generating detection area identification data; Performing wavelet transform filtering on the AC impedance data and the temperature data, and partitioning and storing the filtered AC impedance data and filtered temperature data according to the detection area identification data; Calculate the coefficient of variation of each partition, and assign an interpolation algorithm based on the comparison result of the coefficient of variation with a preset variation threshold. When the coefficient of variation is less than the variation threshold, assign the cubic spline interpolation algorithm; when the coefficient of variation is greater than or equal to the variation threshold, assign the Kriging interpolation algorithm; Execute the assigned interpolation algorithm to perform reconstruction operations to obtain partitioned reconstructed data, and construct a physical constraint optimization function based on the internal resistance continuity and temperature conduction characteristics of the battery cell surface. Perform conjugate gradient optimization on the partitioned reconstructed data, calculate the spatial gradient value of the optimized data, determine the grid point fusion window parameters for adaptive fusion, and obtain fused reconstructed data. The error between the data at the verification point and the fused reconstructed data is calculated. When the error value exceeds the preset error threshold, the interpolation algorithm parameters are updated and the interpolation reconstruction is returned until the error value does not exceed the preset error threshold, generating the AC impedance distribution surface and temperature distribution surface of the battery cell surface.

4. The method according to claim 1, wherein The defective area to be classified is matched with the pre-established battery cell defect feature library through differential feature coding and a double-layer local sensitive hash index structure. The battery cell defect type corresponding to the abnormal performance area is determined based on the matching results, including: Obtaining the morphological features, extended features, and temperature features of the defect area to be classified and performing feature encoding to generate a feature fingerprint of the defect area to be classified; Constructing a two-layer locality-sensitive hash index structure, using a random projection hash function to perform a first-level match on the feature fingerprint with a pre-established battery cell defect feature library to determine candidate defect types, and using an angle-sensitive hash function to perform a second-level match on the candidate defect types to obtain preliminary matching results; Calculating the discriminant information entropy of the morphological features, the extended features, and the temperature features at different time scales, determining a feature importance index based on the discriminant information entropy, and dynamically adjusting the feature weights of the preliminary matching results based on the feature importance index to obtain an optimized matching result; Constructing a feature association graph of the optimized matching result, the feature association graph includes a feature node set, a feature association edge set, and an association weight matrix, and using a spectral clustering method to decompose the feature association graph to obtain feature subgraphs of different defect modes; Repeated feature matching sampling is performed on the defect area to be classified, and a feature matching probability distribution is calculated. When the highest matching probability is greater than a preset probability threshold, the battery cell defect type corresponding to the defect area to be classified is determined.

5. The method according to claim 1, wherein Performing a pulse discharge test on the abnormal area to obtain the dynamic response characteristics of the abnormal area, and calculating the capacity loss rate of the abnormal area based on the dynamic response characteristics and performance defect type includes: Performing a pulse discharge test on the abnormal area to obtain a voltage response curve, performing a wavelet transform on the voltage response curve to obtain a time-frequency domain feature spectrum, extracting frequency features, energy features, and phase features from the time-frequency domain feature spectrum, and constructing a dynamic response feature of the abnormal area; Performing parameter identification based on electrochemical impedance spectroscopy data to obtain polarization resistance and capacitance parameters of the abnormal region, and combining the polarization resistance and capacitance parameters with the dynamic response characteristics to form a characteristic parameter set of the abnormal region; Projecting the characteristic parameter set of the abnormal area into a high-dimensional feature space, using nonlinear feature decomposition to obtain coupling features between parameters, setting a coupling strength threshold according to the performance defect type, extracting coupling features exceeding the coupling strength threshold, and constructing a parameter coupling matrix; A capacity decay characteristic curve is determined as a reference benchmark based on the performance defect type, and the parameter coupling matrix is ​​fitted to the capacity decay characteristic curve to obtain an actual capacity decay trend curve. The deviation of the actual capacity decay trend curve relative to the reference benchmark is calculated, and different penalty weights are set for positive deviations and negative deviations to construct an asymmetric loss function. The capacity loss rate of the abnormal area is calculated based on the asymmetric loss function.

6. The method according to claim 1, characterized in that Adjusting process parameters based on the distribution characteristics of capacity loss rate includes: Statistically analyzing the capacity loss rate data of defective products, obtaining a probability density function of the capacity loss rate using a kernel density estimation method, extracting peak distribution features of the probability density function, and determining abnormal process parameters based on the number of peak distribution features; Extracting historical data of abnormal process parameters from process parameter records corresponding to defective products, calculating a change in the abnormal process parameters based on the historical data, calculating a sensitivity coefficient based on a correspondence between the change and a change in the capacity loss rate, and calculating an adjustment amount for the process parameters based on the sensitivity coefficient; A PID controller is used to dynamically optimize the adjustment amount of the process parameters, wherein the real-time change of the capacity loss rate is used as the input signal of the PID controller, and the output signal of the PID controller is used as the correction amount of the process parameters. The process parameters are updated based on the correction amount to achieve adjustment of the process parameters according to the distribution characteristics of the capacity loss rate.

7. An intelligent detection system for energy storage battery production line processes, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain the electrical performance parameters of battery cells on the energy storage battery production line; The second unit is used to arrange multiple detection points along the length and circumference of the battery cell, collect AC impedance data and temperature data of the battery cell at each detection point, and generate a performance distribution map of the battery cell; The third unit is used to charge the battery cells with constant current and constant voltage to a fully charged state, record the temperature change data of the battery cells during the charging process, identify abnormal areas of the battery cells in the performance distribution diagram, and determine the performance defect type corresponding to the abnormal area based on the distribution characteristics and temperature change data of the abnormal area; A fourth unit is configured to select corresponding pulse discharge test parameters according to the performance defect type, perform a pulse discharge test on the abnormal area to obtain a dynamic response characteristic of the abnormal area, and calculate a capacity loss rate of the abnormal area based on the dynamic response characteristic and the performance defect type; The fifth unit is used to mark the corresponding battery cell as defective when the capacity loss rate exceeds the preset loss rate threshold, and control the robot to transfer the defective products to the defective product collection area. At the same time, it adjusts the process parameters according to the distribution characteristics of the capacity loss rate and sends the adjusted process parameters to the production line control system.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images