A lithium battery lamination quality space distribution feature extraction method based on X-ray layer-by-layer detection and cooperation with an MES system

Abstract

The present application relates to a kind of lithium battery lamination quality space distribution feature extraction methods based on X-ray layer-by-layer detection and MES system cooperation, comprising the following steps: S1: MES system receives the layer-by-layer alignment detection data of target lithium ion battery from the data interface of X-ray detection equipment, constructs alignment deviation space distribution matrix;S2: extract overall statistical characteristics;S3: extract interlayer gradient feature;S4: extract head-tail asymmetry feature;S5: extract four-corner consistency feature;S6: extract local anomaly feature;S7: feature vector construction and dimension reduction.Compared with the traditional method using only range scalar, the present application can more fully utilize the layer-by-layer space distribution information, and the performance prediction accuracy is expected to be significantly improved.MES system executes automatic quality classification according to the prediction result and drives production line actuator to shunt, realizes the end-to-end full-automatic closed-loop management and control of lamination station to production line shunting action.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing and quality inspection technology of lithium-ion batteries, specifically involving a method for extracting spatial distribution features of lithium battery stack quality and predicting electrochemical performance driven by a manufacturing execution system (MES) and working in conjunction with an X-ray layer-by-layer alignment inspection device. Background Technology

[0002] The stacking process of lithium-ion batteries is one of the core steps in cell manufacturing. During the stacking process, the positive electrode, negative electrode, and separator need to be stacked alternately, ensuring that the alignment accuracy between each layer meets the process requirements. The alignment accuracy directly affects the electrochemical performance and safety performance of the cell: excessive alignment deviation will lead to uneven coverage of the anode relative to the cathode, which in turn will cause inconsistent lithium-ion transport paths in local areas, uneven current density distribution during charging and discharging, and ultimately lead to problems such as local lithium deposition and uneven SEI film growth, affecting the cell's capacity, internal resistance, and self-discharge performance.

[0003] Currently, battery production lines commonly use X-ray inspection equipment to check the alignment of stacked cells. Advanced X-ray inspection systems can scan multiple corners of the cell layer by layer, acquiring the anode / cathode edge position information of each electrode layer, thereby calculating the alignment deviation value (unit: mm) for each corner and each layer. Taking a typical square stacked cell as an example, a cell typically has about 50 layers inspected at each of the four corners, obtaining approximately 200 alignment deviation data points, which are then uploaded in real-time to the MES system for archiving via the equipment communication interface.

[0004] However, existing technologies for utilizing X-ray alignment data have significant shortcomings, specifically:

[0005] (1) The judgment method is crude. The current production line MES system mainly uses a simple threshold judgment of "whether the range value exceeds the standard" to process X-ray data. That is, it only focuses on whether the difference between the maximum and minimum deviations in all layers exceeds the upper and lower limits of the process. This method compresses the rich layer-by-layer spatial distribution information into a single range scalar, and a large amount of valuable spatial distribution information is discarded, which cannot support the refined evaluation of the stack quality.

[0006] (2) Single feature dimension. Even if some literature or practice uses statistical measures such as mean and maximum value, it only stays at the level of first-order statistical features and fails to capture the change pattern of alignment deviation in the interlayer direction and angular direction, such as the interlayer gradient trend of deviation, the correlation between different angular positions and other higher-order spatial features.

[0007] (3) Insufficient correlation with electrochemical performance. Existing methods cannot effectively distinguish between the two distinct alignment modes of "overall offset but uniform" and "local abrupt change and non-uniformity". Electrochemical studies have shown that the latter has a significantly greater impact on cell performance than the former, especially on the self-discharge rate (K value).

[0008] Therefore, there is an urgent need for a method that is deeply integrated with X-ray inspection equipment and MES system, can systematically extract spatial distribution features from layer-by-layer aligned data and achieve electrochemical performance prediction, so as to comprehensively characterize the multidimensional characteristics of stack quality and provide a data foundation for quality grading, process optimization and anomaly diagnosis. Summary of the Invention

[0009] To address the aforementioned technical problems, the first objective of this invention is to provide a method for extracting spatial distribution features of lithium-ion battery stack quality and predicting performance. This method deeply integrates the output data of X-ray inspection equipment with the MES system, transforming the high-dimensional, layer-by-layer aligned raw data into stack quality feature vectors with clear physical meaning and controllable dimensions. The extracted multi-dimensional spatial distribution feature vectors are then stored in the MES database as a unified data foundation for subsequent prediction of cell electrochemical performance (capacity, internal resistance, K-value), process diagnosis, and quality grading. The second objective of this invention is to provide a system for extracting spatial distribution features of lithium-ion battery stack quality.

[0010] To achieve the first objective mentioned above, the present invention adopts the following technical solution:

[0011] A method for extracting the spatial distribution features of lithium battery stack quality based on X-ray layer-by-layer inspection and MES system collaboration, employing an MES system and X-ray inspection equipment, and automatically triggered by the MES system after receiving layer-by-layer alignment inspection data uploaded by the X-ray inspection equipment, includes the following steps:

[0012] S1: The MES system receives layer-by-layer alignment detection data of the target lithium-ion cell from the data interface of the X-ray inspection equipment. The data includes the anode-to-cathode alignment deviation values ​​obtained by the X-ray inspection equipment scanning the cell's M corner positions and N layers of electrodes at each corner position. The MES system constructs the received data into an M×N spatial distribution matrix D, where D(i,j) represents the alignment deviation value of the j-th layer at the i-th corner position.

[0013] S2: The MES system performs global statistical calculations on the spatial distribution matrix D to extract overall statistical features, including global mean, global standard deviation, global range, range at each corner, global skewness, and global kurtosis.

[0014] S3: The MES system performs differential analysis and linear trend analysis on the interlayer orientation of the spatial distribution matrix D to extract interlayer gradient features, including interlayer gradient intensity G_rms, trend slope β_i at each angle, and maximum continuous drift features.

[0015] S4: The MES system divides the N layers of data at each corner into the head region corresponding to the initial stage of the stacking and the tail region corresponding to the final stage of the stacking. It calculates the standardized difference between the head mean and the tail mean and extracts the head-tail asymmetry features that reflect the degree of thermal drift of the stacking machine.

[0016] S5: The MES system calculates the Pearson correlation coefficient matrix between the M corner deviation sequences and extracts the four-corner consistency features, including the mean corner correlation coefficient r_mean, the mean coefficient of variation CV_mean, and the diagonal consistency index ΔC_diag.

[0017] S6: The MES system uses the IQR method to perform local anomaly detection on the spatial distribution matrix D, and extracts the proportion of anomaly points p_outlier and the length of the longest continuous anomaly segment L_cluster;

[0018] S7: The MES system summarizes the features extracted in steps S2 to S6 into an original feature vector F_raw. After standardization and feature selection, it outputs a stacking quality feature vector F. The stacking quality feature vector F is then bound to the unique code of the lithium-ion cell and stored in the database of the MES system.

[0019] As a preferred embodiment, in step S3, the interlayer gradient intensity G_rms is calculated by the MES system according to the following steps: First, for each corner position i, the change in the positioning deviation of the adjacent layer electrode is calculated as ΔD(i,j)=D(i,j+1)-D(i,j), and then the root mean square value is calculated for all elements of the difference matrix ΔD, G_rms = sqrt((1 / (M×(N-1))) × ΣiΣj (ΔD(i,j))²); G_rms reflects the layer-by-layer fluctuation intensity of the electrode positioning accuracy during the continuous stacking process of the stacking machine.

[0020] As a preferred option, in step S3, the trend slope β_i is calculated using the least squares method: for each corner position i, a linear regression is performed with layer number j as the independent variable and D(i,j) as the dependent variable to obtain β_i; β_i quantifies the systematic positioning drift rate of the stacker during the continuous stacking process, with the unit being mm / layer.

[0021] As a preferred embodiment, in step S4, the head region is the first k layers, and the tail region is the last k layers, k=floor(N / 5); the formula for calculating the head-tail asymmetry coefficient is A_i=(μ_tail_i - μ_head_i) / σ_i, where μ_tail_i is the tail mean of the i-th corner position, μ_head_i is the head mean of the i-th corner position, and σ_i is the standard deviation of the deviation sequence of the i-th corner position; the MES system calculates the global head-tail asymmetry index A_mean=(1 / M)×Σ|A_i|, and automatically triggers the zero-point verification command of the stacker when A_mean exceeds the preset threshold.

[0022] As a preferred option, in step S5, the diagonal consistency index ΔC_diag=|C(1,3)-C(2,4)|, where C(p,q) is the Pearson correlation coefficient of the deviation sequence between the p-th and q-th corners; when ΔC_diag is large, the MES system generates an electrode warping warning, which triggers the incoming material inspection process for review.

[0023] As a preferred option, in step S6, when the MES system detects that the length of the longest continuous abnormal segment L_cluster is not less than the preset layer threshold, it automatically generates a manual review instruction and marks the cell as isolated, waiting for physical sampling inspection confirmation before it can be released.

[0024] As a preferred option, in step S7, redundant features are first removed based on the Spearman correlation coefficient threshold, and then the top-ranked features are retained based on the random forest feature importance ranking. Finally, the dimension d of the feature vector F satisfies d≤20 and the retained features cover no less than 95% of the variance information of the original 29-dimensional features.

[0025] As a preferred embodiment, the method further includes step S8: the MES system calls a pre-trained and deployed machine learning regression model, taking the feature vector F output in step S7 as input, and outputs the prediction result of at least one of the following: capacity, AC internal resistance, and K value; the machine learning regression model can be constructed using XGBoost, LightGBM, or random forest algorithms; the MES system performs quality grading on the battery cell based on the prediction result, and the grading result triggers the production line execution mechanism to perform release or isolation actions on the battery cell; the quality grading can be divided into four levels based on the K value prediction result: release, key tracking, pending re-inspection, and interception.

[0026] As a preferred option, M=4 and N=50 correspond to the X-ray inspection configuration of 50 layers at each of the four corners of the square laminated cell; after receiving 200 data points uploaded by the X-ray inspection equipment, the MES system completes all steps S1 to S8 and outputs the quality grading results before the cell leaves the inspection station.

[0027] To achieve the second objective mentioned above, the present invention employs the following technical solution:

[0028] A lithium battery stacking quality spatial distribution feature extraction system includes an X-ray inspection device for scanning the stacked cells layer by layer and outputting alignment deviation data of M corner positions × N layers;

[0029] The MES system is communicatively connected to the X-ray inspection equipment, receives layer-by-layer alignment inspection data in real time, and performs the spatial distribution feature extraction calculation of the stacked sheet quality according to the method described in any of the above, and stores the extracted feature vector F in the MES database; and the MES system further performs electrochemical performance prediction and quality grading calculation on the feature vector F.

[0030] The production line actuator communicates with the MES system and performs subsequent actions on the battery cell based on the feature extraction results or quality grading instructions output by the MES system.

[0031] The X-ray inspection equipment, MES system, and production line actuators together constitute a wafer stacking quality control loop.

[0032] As a preferred embodiment, the MES system also includes a process diagnostic module, which is used to periodically generate equipment health status reports based on the extracted spatial distribution feature vectors: when G_rms continuously exceeds a preset threshold, an early warning of the stacking machine positioning accuracy is output; when A_mean is continuously high, the thermal drift compensation calibration of the stacking machine is triggered; when ΔC_diag is abnormal, the incoming electrode warpage re-inspection process is triggered in association; thus realizing preventive maintenance of the stacking equipment status and adaptive optimization of process parameters.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] This invention constructs a spatial distribution matrix from X-ray layer-by-layer aligned data, systematically extracting 29 original features from multiple dimensions such as inter-layer direction, angular direction, and local anomalies, fully utilizing the spatial distribution information contained in the layer-by-layer data. Compared with traditional methods that only utilize first-order statistics such as range, the multi-dimensional spatial distribution feature vector extracted by this invention provides a more discriminative input for the prediction model, and is expected to significantly improve the prediction accuracy of K value, capacity, and internal resistance.

[0035] Each feature extracted in this invention has a clear physical meaning and corresponds to the specific manufacturing equipment status: G_rms reflects the layer-by-layer fluctuation of the real-time positioning accuracy of the stacking machine, which can be used to identify the vibration or wear status of the stacking machine; A_mean reflects the degree of thermal drift of the equipment and can trigger preventive calibration; r_mean and ΔC_diag can diagnose electrode warping problems and guide the adjustment of incoming material inspection standards. These features provide process engineers with directly operable quantitative diagnostic basis.

[0036] The method of this invention is executed throughout the MES system. It can output performance prediction results and quality grades within seconds after the cell completes X-ray inspection. Compared with the traditional post-evaluation method that relies on capacity testing (which takes several hours to tens of hours), it advances the discovery time of quality problems from the capacity testing station to the stacking station, which greatly shortens the flow time of defective products in the production line and reduces manufacturing losses.

[0037] The method framework of this invention does not depend on a specific number of corners M or layers N, and can be applied to different types of stacked cells (square, pouch) and different configurations of X-ray inspection equipment. It has good versatility and scalability, and is easy to deploy uniformly in multi-production line and multi-model environments. Attached Figure Description

[0038] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute a limitation thereof.

[0039] Figure 1 This is a schematic diagram of the integrated architecture of the method of the present invention with the MES system and X-ray inspection equipment, showing the data flow (X-ray equipment → MES system → feature extraction module → prediction model → quality grading output) and the triggering relationship between each module.

[0040] Figure 2 This diagram illustrates the construction of the spatial distribution matrix of X-ray layer-by-layer alignment deviation. Taking a 4-corner × 50-layer matrix as an example, it shows the physical correspondence of matrix elements D(i,j) and the definition of the directions of positive / negative deviations.

[0041] Figure 3 is a schematic diagram illustrating the principle of inter-layer gradient feature extraction, in which... Figure 3a This is a schematic diagram of the interlayer difference sequence (ΔD) at cell corner position 1 with a low G_rms. Figure 3b This is a schematic diagram of the layer-by-layer deviation sequence of a cell with a low G_rms angle position 1; Figure 3c This is a schematic diagram of the interlayer difference sequence (ΔD) at cell corner position 1 with a certain high G_rms. Figure 3d This is a schematic diagram of the layer-by-layer deviation sequence of the cell corner position 1 with a certain high G_rms.

[0042] Figure 4 is a schematic diagram of the head-to-tail asymmetry feature, in which... Figure 4a and Figure 4b Examples of the mean difference and asymmetry coefficient A_i for the head region (first 10 layers) and tail region (last 10 layers) are shown for small asymmetry and large asymmetry, respectively, and typical waveform characteristics of device thermal drift are marked.

[0043] Figure 5 is a schematic diagram of the four-corner consistency analysis, in which... Figure 5a and Figure 5b Heatmaps of the 4×4 correlation coefficient matrix are shown for two typical modes: r_mean close to 1 (overall translational bias) and r_mean close to 0 (random warping bias).

[0044] Figure 6 is a schematic diagram of local anomaly detection. Figure 6a and Figure 6b Examples of IQR anomaly detection boundaries and calculation of the clustering degree (L_cluster) of anomalies in the interlayer direction are shown respectively.

[0045] Figure 7 shows the method of the present invention ( Figure 7b ) and traditional range feature method ( Figure 7a The graph shows the comparison of the prediction performance of K-values, illustrating the comparison results of R² and RMSE.

[0046] Figure 8 shows the importance analysis of SHAP features. Figure 8a , Figure 8b and Figure 8c The contribution of each spatial distribution feature to the prediction of K value, capacity, and internal resistance is displayed in order. Detailed Implementation

[0047] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0048] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0049] Furthermore, in the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0050] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more, unless explicitly defined otherwise.

[0051] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0052] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0054] like Figure 1 As shown in Figure 8, the implementation environment of this embodiment is the MES (Manufacturing Execution System) of a lithium-ion battery production line. This production line uses a stacking process to manufacture square aluminum-cased lithium-ion cells (LFP system, nominal capacity 280Ah). After the cells are stacked, they are automatically sent to the X-ray inspection station by the stacking machine.

[0055] After completing layer-by-layer scanning, the X-ray inspection equipment (any model) uploads the inspection data to the MES system in real time via a standard industrial communication interface (OPC-UA or a custom TCP / IP protocol). Upon receiving the data, the MES system immediately triggers the feature extraction and prediction process of this invention, without requiring manual intervention.

[0056] The X-ray inspection configuration is as follows: the number of inspection corners M=4 (corresponding to corners 1, 2, 3, and 4 of the cell respectively), the number of inspection layers N=50 for each corner (covering all stacked layers of the cell from the top to the bottom), and the output of anode / cathode alignment deviation value (unit: mm) for each corner and each layer. A total of 200 data points are output for a single cell.

[0057] Meanwhile, the MES system collects test data of the same batch of cells at subsequent capacity testing stations, including capacity (Ah), AC internal resistance (mΩ), and K value (mV / day) obtained from OCV static testing, as performance labels for model training and validation.

[0058] A method for extracting the spatial distribution features of lithium battery stack quality based on X-ray layer-by-layer inspection and MES system collaboration includes the following steps:

[0059] Step S1: Spatial distribution matrix construction. Taking a single battery cell as an example, the MES system organizes the 200 deviation data points (4 corners × 50 layers, totaling 200 layers) received from the X-ray device into a 4×50 matrix D.

[0060] Corner / Layer Level 1 2nd floor ... 50th floor unit Angle 1 D(1,1) D(1,2) ... D(1,50) mm Angle 2 D(2,1) D(2,2) ... D(2,50) mm Angle 3 D(3,1) D(3,2) ... D(3,50) mm Angle 4 D(4,1) D(4,2) ... D(4,50) mm

[0061] Each element D(i,j) directly corresponds to the physical measurement output value (unit: mm) of the X-ray inspection equipment at the i-th angular position and j-th layer, and has a clear spatial physical coordinate meaning. A positive value of the alignment deviation indicates that the anode is offset outward (overhang increases), and a negative value indicates that the anode is offset inward (overhang decreases).

[0062] Step S2: Overall statistical feature extraction. The MES system calculates the following overall statistical features for matrix D:

[0063] The global mean μ_global = (1 / (M×N))×Σ_iΣ_jD(i,j);

[0064] Global standard deviation σ_global=sqrt((1 / (M×N-1))×Σ_iΣ_j(D(i,j)-μ_global)²);

[0065] Global range R_global = max(D) - min(D); Angular ranges R_1, R_2, R_3, R_4;

[0066] Global skewness S_global=(1 / (M×N))×Σ_iΣ_j((D(i,j)-μ_global) / σ_global)³;

[0067] Global kurtosis K_global=(1 / (M×N))×Σ_iΣ_j((D(i,j)-μ_global) / σ_global) 4 -3;

[0068] The global mean μ_global and global standard deviation σ_global characterize the overall level and dispersion of cell alignment deviation; the global skewness S_global and kurtosis K_global characterize the asymmetry of the alignment deviation distribution and the degree of concentration of extreme values. Skewness reflects the existence of a systematic unidirectional offset tendency, while kurtosis reflects the existence of a concentration of extreme deviations. In one example, the typical range of μ_global is 0.05-0.15 mm, and the typical range of σ_global is 0.02-0.08 mm.

[0069] Step S3: Interlayer gradient feature extraction. The MES system performs differential analysis and linear trend analysis on the column directions (interlayer directions) of the spatial distribution matrix D to extract the following interlayer gradient features, which are used to quantify the positioning accuracy fluctuations and systematic drift of the stacking machine during continuous stacking:

[0070] The MES system performs layer-by-layer difference analysis on the deviation sequence of each corner position: ΔD(i,j)=D(i,j+1)-D(i,j), to obtain a 4×49 difference matrix.

[0071] The formula for calculating the interlayer gradient intensity is: G_rms=sqrt((1 / (4×49))×Σ_iΣ_j(ΔD(i,j))²);

[0072] In one example, when G_rms is greater than approximately 0.03 mm / layer, it indicates that the stacking machine is experiencing layer-by-layer positioning jitter, which may be related to a decrease in the mechanical precision of the equipment drive system. The MES system can automatically generate equipment inspection reminders based on this. When β_i is greater than approximately 0.001 mm / layer, it can be determined that there is a significant systematic drift trend.

[0073] The trend slope was calculated using the least squares method: for each corner i, a linear regression was performed with layer j as the independent variable and D(i,j) as the dependent variable to obtain β_i. β_i > 0.001 mm / layer indicates a significant systematic drift trend.

[0074] If there are interlayer fluctuations (larger G_rms) or systematic drift (significantly non-zero β_i) in the electrode positioning accuracy during the stacking process of lithium-ion battery cells, the overhang distribution of each electrode layer will be inconsistent, leading to differences in the anode active area and lithium-ion transport path in different regions of the cell. During charge-discharge cycles, the uneven distribution of local current density accelerates the irregular growth of the SEI film and local electrolyte consumption in local areas. The inhomogeneity of the SEI film is one of the main physical mechanisms leading to an increase in the self-discharge rate (K value). Therefore, G_rms and β_i can quantify this physical process from its source and provide a physical basis for predicting the K value.

[0075] Step S4: Asymmetric feature extraction at the beginning and end. Assume the number of segmented layers k=10. The MES system calculates:

[0076] Head mean: μ_head_i=(1 / 10)×Σ_{j=1}^{10}D(i,j);

[0077] Tail mean: μ_tail_i=(1 / 10)×Σ_{j=41}^{50}D(i,j);

[0078] Head-tail asymmetry coefficient: A_i=(μ_tail_i-μ_head_i) / σ_i;

[0079] Global head-to-tail asymmetry index: A_mean = (1 / 4) × Σ_i|A_i|;

[0080] During long-term continuous operation, the mechanical system of the stacking machine experiences thermal expansion due to temperature changes, leading to a systematic decrease in the electrode positioning accuracy at the end of the stack (tail end), manifested as an increase in the A_mean value. The systematic under-sizing of the overhang in the tail region causes localized excess or deficiency of anode active material in this area, resulting in a non-uniform initial SEI film during the formation process. This leads to uneven lithiation of the cell in the thickness direction, ultimately resulting in a higher AC internal resistance.

[0081] In one example, A_mean can take values ​​in the range of 0.1 to 0.8; when A_mean is greater than a preset threshold (such as 0.5), the MES system marks the cell as "suspected impact of equipment thermal drift" and triggers the automatic zero-point verification process of the stacking machine.

[0082] Step S5: Corner Consistency Feature Extraction. The MES system performs correlation analysis on the row directions (corner directions) of the spatial distribution matrix D to extract the following corner consistency features, which are used to determine the formation mechanism of lamination deviation (overall systematic displacement vs. random positioning error vs. electrode warping):

[0083] The MES system calculates the Pearson correlation coefficient matrix C (a 4×4 symmetric matrix) of the four angular deviation sequences:

[0084] Mean angular correlation coefficient: r_mean=(1 / 6)×(C(1,2)+C(1,3)+C(1,4)+C(2,3)+C(2,4)+C(3,4))

[0085] Diagonal consistency index: ΔC_diag=|C(1,3)-C(2,4)|;

[0086] Mean coefficient of variation at each level: CV_mean = (1 / 50)×Σ_j(std(D(:,j)) / |mean(D(:,j))|);

[0087] r_mean close to 1 corresponds to the overall positioning offset of the entire electrode sheet (synchronous movement of each corner); a lower r_mean and a larger ΔC_diag correspond to random rotation or warping of the electrode sheet. The two modes have different mechanisms of influence on cell performance. Classification diagnosis helps process engineers to accurately locate the root cause of the problem.

[0088] When r_mean approaches 0 (independent variation at each corner), it indicates random rotation or local warping of the electrode, rather than overall translation. In this case, the effective overlap area of ​​the electrode (the region corresponding to the positive and negative electrodes) decreases locally, directly leading to a reduction in the area of ​​active material that can participate in the electrochemical reaction, manifested as capacity loss. An increase in CV_mean quantifies the degree of this area loss; therefore, the four-corner consistency characteristic has direct physical significance for capacity prediction.

[0089] Step S6: Local anomaly feature extraction. The MES system uses the IQR (interquartile range) method to perform local anomaly detection on the spatial distribution matrix D, identifying and quantifying occasional alignment misalignment events during the stacking process.

[0090] (a) Calculate the first quartile Q1 and the third quartile Q3 of all elements in matrix D. Let IQR = Q3 - Q1. Use Q1-1.5×IQR and Q3+1.5×IQR as the anomaly detection boundary. Count the number of outliers n_outlier and the percentage of outliers p_outlier=n_outlier / (4×50).

[0091] (b) Calculate the length of the longest continuous abnormal segment L_cluster in the layer number direction of the abnormal point. The larger L_cluster is, the more likely it is that a continuous misalignment has occurred in a certain stage of the stacking process, which requires triggering equipment alarm verification. When L_cluster≥5, the MES system triggers a manual review instruction to perform manual sampling inspection on the cell.

[0092] (c) Record the deviation value D_max with the largest absolute value in matrix D and its location (i_max, j_max) to provide process engineers with precise positioning information of the most serious deviation.

[0093] Step S7: Feature Vector Construction and Dimensionality Reduction. The MES system summarizes all the features extracted in the above steps and constructs the original feature vector F_raw, with the following dimensional composition:

[0094] Feature categories Includes features Dimension Overall statistical characteristics μ, σ, R_global, R_1~R_4, S, K 9 Interlayer gradient features G_rms, β_1~β_4, β_mean, β_std, L_max 8 asymmetry between head and tail A_1~A_4, A_mean 5 Four-corner consistency feature r_mean, ΔC_diag, CV_mean 3 Local abnormal features n_outlier, p_outlier, L_cluster, D_max 4 total 29

[0095] After Z-score standardization of F_raw, redundant features are removed using a Spearman correlation coefficient threshold (|ρ|>0.9), and then the top d-dimensional features are retained based on random forest feature importance ranking. In this embodiment, after feature selection, d=15-20 dimensions can retain more than 95% of the information. Finally, the feature vector F is bound to the unique cell code and written into the MES database.

[0096] Step S8 (Optional Implementation): In one optional implementation, real-time prediction and quality grading of electrochemical performance are performed as follows: the MES system calls a pre-deployed XGBoost regression model, using the feature vector F as input, to predict the capacity, AC internal resistance, and K value. The prediction results are output approximately 2-5 seconds after the cell leaves the X-ray inspection station. The MES system then performs quality grading according to the following grading rules:

[0097] (a) If the predicted K value is <0.5mV / day and the predicted internal resistance is within the nominal range: Quality grade [Excellent], direct release;

[0098] (b) Predicted K value is 0.5-1.0mV / day: Quality level [Good], released but marked as a key tracking cell;

[0099] (c) Predicted K value > 1.0 mV / day: Quality level [Pending re-inspection], isolated to the re-inspection area awaiting confirmation of the measured K value;

[0100] (d) If the predicted K value is >2.0mV / day or the predicted capacity is less than 95% of the nominal value: Quality level [Interception], triggering automatic isolation and issuing a process alarm.

[0101] The application and expected effects of the method of this invention are as follows: In implementation, a certain batch of square battery cell samples from a production line can be collected. The feature vectors obtained by processing the X-ray layer-by-layer aligned data through steps S1 to S7 are used as model inputs. The measured capacity, AC internal resistance, and K value of the same batch of battery cells at the capacity testing station are used as labels. The training set and test set are divided according to a certain ratio. The XGBoost regression model is trained using the method of this invention (multidimensional spatial distribution feature vector) and the traditional method (only using first-order features such as the range of each corner) respectively, and the performance is compared. Since there is a clear physical causal chain between the features extracted by this invention and the electrochemical performance indicators—interlayer gradient features correspond to the positioning accuracy fluctuation of the stacking machine, head and tail asymmetry features correspond to equipment thermal drift, and four-corner consistency features correspond to electrode translation and warping modes—compared with the traditional method that only uses the range scalar, it can make fuller use of the layer-by-layer spatial distribution information. Therefore, it is expected to obtain a higher prediction determination coefficient and a lower prediction error. In addition, interpretability analysis methods such as SHAP (SHapley Additive exPlanations) can be combined to rank the contributions of each spatial distribution feature to the prediction of K value, capacity, and internal resistance, so as to verify the physical rationality and interpretability of the proposed features and assist in process diagnosis.

[0102] This invention utilizes structured feature extraction and multi-model prediction of layer-by-layer alignment deviation data output in real time by an X-ray inspection system. This enables immediate prediction of cell performance and quality grading upon completion of the lamination process, guiding production line process adjustments and early interception of abnormal cells. Furthermore, the spatial distribution features extracted by this invention are not purely mathematical transformations of the original data, but rather quantitative representations of various physical deviation modes (mechanical drift, thermal deformation, random errors, and local misalignments) in the lamination process. Each feature has a clear physical causal chain with electrochemical performance indicators, thus effectively supporting the establishment of performance prediction models.

[0103] A lithium battery stacking quality spatial distribution feature extraction system includes an X-ray inspection device for scanning the stacked cells layer by layer and outputting alignment deviation data of M corner positions × N layers;

[0104] The MES system is communicatively connected to the X-ray inspection equipment, receives layer-by-layer alignment inspection data in real time, and performs feature extraction, performance prediction, and quality grading calculation according to the method described in any of the above.

[0105] The production line actuator communicates with the MES system and performs physical actions to release or isolate the battery cells according to the quality grading instructions output by the MES system.

[0106] The X-ray inspection equipment, MES system, and production line actuators together form a fully automated closed-loop control loop for wafer stacking quality, achieving end-to-end automation from layer-by-layer alignment physical measurement to production line diversion action execution.

[0107] The MES system also includes a process diagnostic module, which periodically generates equipment health status reports based on the extracted spatial distribution feature vectors: when G_rms continuously exceeds a preset threshold, it outputs a positioning accuracy warning for the stacking machine; when A_mean is continuously high, it triggers thermal drift compensation calibration for the stacking machine; when ΔC_diag is abnormal, it triggers a re-inspection process for the warpage of the incoming electrode sheets; thus realizing preventive maintenance of the stacking equipment status and adaptive optimization of process parameters.

[0108] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0109] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for extracting the spatial distribution features of lithium battery stack quality based on X-ray layer-by-layer inspection and MES system collaboration, characterized in that, The process employs an MES system and X-ray inspection equipment, with the MES system automatically triggering execution after receiving the layer-by-layer alignment inspection data uploaded by the X-ray inspection equipment. The process includes the following steps: S1: The MES system receives layer-by-layer alignment detection data of the target lithium-ion cell from the data interface of the X-ray inspection equipment. The data includes the anode-to-cathode alignment deviation values ​​obtained by the X-ray inspection equipment scanning the cell's M corner positions and N layers of electrodes at each corner position. The MES system constructs the received data into an M×N spatial distribution matrix D, where D(i,j) represents the alignment deviation value of the j-th layer at the i-th corner position. S2: The MES system performs global statistical calculations on the spatial distribution matrix D to extract overall statistical features, including global mean, global standard deviation, global range, range at each corner, global skewness, and global kurtosis. S3: The MES system performs differential analysis and linear trend analysis on the interlayer orientation of the spatial distribution matrix D to extract interlayer gradient features, including interlayer gradient intensity G_rms, trend slope β_i at each angle, and maximum continuous drift features. S4: The MES system divides the N layers of data at each corner into the head region corresponding to the initial stage of the stacking and the tail region corresponding to the final stage of the stacking. It calculates the standardized difference between the head mean and the tail mean and extracts the head-tail asymmetry features that reflect the degree of thermal drift of the stacking machine. S5: The MES system calculates the Pearson correlation coefficient matrix between the M corner deviation sequences and extracts the four-corner consistency features, including the mean corner correlation coefficient r_mean, the mean coefficient of variation CV_mean, and the diagonal consistency index ΔC_diag. S6: The MES system uses the IQR method to perform local anomaly detection on the spatial distribution matrix D, and extracts the proportion of anomaly points p_outlier and the length of the longest continuous anomaly segment L_cluster; S7: The MES system summarizes the features extracted in steps S2 to S6 into an original feature vector F_raw. After standardization and feature selection, it outputs a stacking quality feature vector F. The stacking quality feature vector F is then bound to the unique code of the lithium-ion cell and stored in the database of the MES system.

2. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer detection and MES system collaboration as described in claim 1, characterized in that, In step S3, the interlayer gradient intensity G_rms is calculated by the MES system according to the following steps: First, for each corner position i, the change in the positioning deviation of adjacent layers of electrode sheets is calculated as ΔD(i,j)=D(i,j+1)-D(i,j). Then, the root mean square value is calculated for all elements of the difference matrix ΔD, G_rms = sqrt((1 / (M×(N-1))) × ΣiΣj(ΔD(i,j))²). G_rms reflects the layer-by-layer fluctuation intensity of the electrode sheet positioning accuracy during the continuous stacking process of the stacking machine.

3. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in claim 1, characterized in that: In step S3, the trend slope β_i is calculated using the least squares method: for each corner position i, a linear regression is performed with layer number j as the independent variable and D(i,j) as the dependent variable to obtain β_i; β_i quantifies the systematic positioning drift rate of the stacker during the continuous stacking process, with the unit being mm / layer.

4. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in claim 1, characterized in that: In step S4, the head region is the first k layers, and the tail region is the last k layers, k=floor(N / 5); the formula for calculating the head-tail asymmetry coefficient is A_i=(μ_tail_i - μ_head_i) / σ_i, where μ_tail_i is the tail mean of the i-th corner position, μ_head_i is the head mean of the i-th corner position, and σ_i is the standard deviation of the deviation sequence of the i-th corner position; the MES system calculates the global head-tail asymmetry index A_mean=(1 / M)×Σ|A_i|, and automatically triggers the zero-point verification command of the stacker when A_mean exceeds the preset threshold.

5. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in claim 1, characterized in that: In step S5, the diagonal consistency index ΔC_diag=|C(1,3)-C(2,4)|, and C(p,q) is the Pearson correlation coefficient of the deviation sequence between the p-th and q-th angle positions; When ΔC_diag is large, the MES system generates an electrode warpage warning, which triggers the incoming material inspection process for review.

6. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in claim 1, characterized in that: In step S6, when the MES system detects that the length of the longest continuous abnormal segment L_cluster is not less than the preset layer threshold, it automatically generates a manual review instruction and marks the cell as isolated, waiting for physical sampling inspection confirmation before it can be released.

7. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in claim 1, characterized in that: In step S7, feature selection first removes redundant features based on the Spearman correlation coefficient threshold, and then retains the top-ranked features based on the random forest feature importance ranking; finally, the dimension d of the feature vector F satisfies d≤20 and the retained features cover no less than 95% of the variance information of the original 29-dimensional features.

8. The method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in claim 1, characterized in that: The method further includes step S8: the MES system calls a pre-trained and deployed machine learning regression model, takes the feature vector F output in step S7 as input, and outputs the prediction result of at least one of the following: capacity, AC internal resistance, and K value; the machine learning regression model can be constructed using XGBoost, LightGBM, or random forest algorithms; the MES system performs quality grading on the battery cell based on the prediction result, and the grading result triggers the production line execution mechanism to perform release or isolation actions on the battery cell; the quality grading can be divided into four levels based on the K value prediction result: release, key tracking, pending re-inspection, and interception.

9. A method for extracting the spatial distribution features of lithium battery stacking quality based on X-ray layer-by-layer inspection and MES system collaboration as described in any one of claims 1 to 8, characterized in that: M=4, N=50, corresponding to the X-ray inspection configuration of 50 layers at each of the four corners of the square laminated cell; after receiving 200 data points uploaded by the X-ray inspection equipment, the MES system completes all steps S1 to S8 and outputs the quality grading results before the cell leaves the inspection station.

10. A system for extracting the spatial distribution characteristics of lithium battery stacked mass, characterized in that: Includes X-ray inspection equipment, used to scan the stacked cells layer by layer and output alignment deviation data of M corner positions × N layers; The MES system is communicatively connected to the X-ray inspection equipment, receives layer-by-layer alignment inspection data in real time, and performs the extraction and calculation of the spatial distribution features of the stacked sheet quality according to any one of claims 1 to 9, storing the extracted feature vector F in the MES database; and the MES system further performs electrochemical performance prediction and quality grading calculation on the feature vector F. The production line actuator communicates with the MES system and performs subsequent actions on the battery cell based on the feature extraction results or quality grading instructions output by the MES system. The X-ray inspection equipment, MES system, and production line actuators together constitute a wafer stacking quality control loop.

11. The lithium battery stack mass spatial distribution feature extraction system according to claim 10, characterized in that: The MES system also includes a process diagnostic module, which periodically generates equipment health status reports based on the extracted spatial distribution feature vectors: when G_rms continuously exceeds a preset threshold, it outputs a positioning accuracy warning for the stacking machine; when A_mean is continuously high, it triggers thermal drift compensation calibration for the stacking machine; when ΔC_diag is abnormal, it triggers a re-inspection process for the warpage of the incoming electrode sheets; thus realizing preventive maintenance of the stacking equipment status and adaptive optimization of process parameters.

Images