A Smart X-ray Image Detection Method for Defects in Solid-State Battery Electrodes

Abstract

This application relates to an intelligent X-ray image detection method for defects in solid-state battery electrodes, comprising: S1, acquiring an X-ray transmission image of the solid-state battery electrode to be detected; establishing an X-ray attenuation physical model; using the X-ray attenuation physical model to perform scattering correction calculation and thickness non-uniformity compensation calculation on the X-ray transmission image to obtain a structural correction image; S2, performing layered structural segmentation calculation on the structural correction image to obtain structural feature maps corresponding to the electrode material layer, solid electrolyte layer, and interface transition region, respectively; S3, calculating interface continuity parameters, grayscale statistical parameters, and texture feature parameters from the structural feature maps to form a multi-dimensional numerical feature vector; converting the multi-dimensional numerical feature vector into numerical defect characterization parameters; S4, inputting the numerical defect characterization parameters into a pre-trained defect recognition model for classification and localization calculation, and outputting defect type and defect spatial location information; S5, outputting the electrode defect detection result.

Description

Technical Field

[0001] This application relates to the field of detection technology, and in particular to an intelligent X-ray image detection method for defects in solid-state battery electrodes. Background Technology

[0002] With the rapid development of the new energy industry, solid-state batteries, due to their high energy density, high safety, and good cycle stability, are gradually becoming an important development direction for next-generation energy storage devices. Compared with traditional liquid electrolyte lithium-ion batteries, solid-state batteries use solid electrolytes instead of organic liquid electrolytes. Their internal structure typically consists of an electrode active material layer, a solid electrolyte layer, and an interface transition region between the two. The contact state between the multi-layer structure directly affects the ion transport efficiency and the overall battery performance.

[0003] In the manufacturing process of solid-state batteries, processes such as electrode coating, cold pressing, and heat treatment can easily lead to defects such as poor interlayer contact, interface delamination, abnormal porosity, uneven material distribution, and foreign object inclusions. These defects are often located inside the electrode and are difficult to detect with the naked eye, but they can cause localized current density anomalies and stress concentrations during battery charge-discharge cycles, leading to accelerated capacity decay or even premature failure. Therefore, high-precision, non-destructive internal defect detection of solid-state battery electrodes is of great significance.

[0004] Currently, the main inspection methods for battery electrodes include visual inspection, electrical performance sampling, and non-destructive testing methods based on X-ray imaging. Among these, X-ray transmission imaging or industrial CT technology can penetrate the multi-layered structure of the electrode to achieve visual observation of internal defects, and therefore is gradually being applied in the battery manufacturing field.

[0005] Existing X-ray image-based detection methods mostly employ traditional image processing or deep learning algorithms to identify gray-level anomalies. For example, threshold segmentation, edge detection, or convolutional neural network models are used to classify defects in gray-level difference regions of the image. However, solid-state battery electrodes are multilayer composite material structures with significant differences in the effective atomic number, electron density, and thickness distribution of each layer. Furthermore, factors such as coating thickness fluctuations and compaction density variations during the manufacturing process lead to thickness artifacts, scattering noise, and interlayer gray-level overlap in X-ray images. If the imaging physical processes are not compensated for, relying solely on gray-level information for detection can easily result in misjudgments or missed detections.

[0006] However, the inventors have discovered at least the following technical problems in the related technologies: how to accurately analyze the multilayer structure of solid-state battery electrodes based on the X-ray physical imaging mechanism, construct physical defect characterization parameters that can reflect the integrity of interface contact and the spatial uniformity of materials, and achieve stable and reliable intelligent identification under small sample conditions. Summary of the Invention

[0007] One objective of this application is to provide an intelligent X-ray image detection method for defects in solid-state battery electrodes, at least to address the aforementioned problems.

[0008] To achieve the above objectives, some embodiments of this application provide an intelligent X-ray image detection method for defects in solid-state battery electrodes, including:

[0009] S1. Obtain an X-ray transmission image of the solid-state battery electrode to be tested; based on the design material composition parameters and layer thickness parameters of the electrode, establish an X-ray attenuation physical model including the linear attenuation coefficient of the material and the layer thickness parameters; use the X-ray attenuation physical model to perform scattering correction calculation and thickness non-uniformity compensation calculation on the X-ray transmission image to obtain a structural correction image.

[0010] S2. Based on the spatial arrangement design parameters of the electrode material layer, solid electrolyte layer and interface transition region, perform layered structural segmentation calculation on the structure correction image to obtain structural feature maps corresponding to the electrode material layer, solid electrolyte layer and interface transition region respectively.

[0011] S3. Calculate the interface continuity parameter, grayscale statistical parameter and texture feature parameter from the structural feature map to form a multidimensional numerical feature vector; according to the solid-state battery failure mechanism model, convert the multidimensional numerical feature vector into numerical defect characterization parameters for quantifying interface contact integrity and material spatial uniformity.

[0012] S4. Input the numerical defect characterization parameters into the pre-trained defect identification model for classification and localization calculation. The loss function of the defect identification model includes a stacking order consistency constraint term, which is used to restrict the spatial adjacency relationship of each layer structure in the prediction result to conform to the physical stacking order of the electrode. Output defect type and defect spatial location information.

[0013] S5, Output electrode defect detection results.

[0014] Compared with related technologies, the solution provided in this application uses a preset physical model to preprocess the image, transforming the original grayscale image, which is affected by scattering interference and thickness fluctuations, into a structure-corrected image that reflects the essential characteristics of the material. The algorithm automatically identifies three core regions: electrodes, electrolytes, and interfaces, and extracts multi-dimensional features that can quantify defects from them. Finally, a deep learning model with physical constraints is used to accurately identify and locate defects such as cracks, pores, and foreign objects. This process solves the pain points of traditional visual inspection, which is unable to penetrate the complex multi-layer structure of solid-state batteries and cannot quantify the contact quality of internal interfaces, thus improving the quality control accuracy of the all-solid-state battery production line. Attached Figure Description

[0015] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0016] Figure 1 This is a flowchart of the method provided in the embodiments of this disclosure;

[0017] Figure 2 This is a flowchart from another perspective of the method provided in the embodiments of this disclosure;

[0018] Figure 3 This is a flowchart from another perspective of the method provided in the embodiments of this disclosure. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0021] In this disclosure, the terms "upper," "lower," "inner," "middle," "outer," "front," and "rear," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for better description of the embodiments of this disclosure and their implementations, and are not intended to limit the indicated devices, elements, or components to having a specific orientation, or to require them to be constructed and operated in a specific orientation. Furthermore, some of the aforementioned terms may be used to indicate other meanings besides orientation or positional relationship; for example, the term "upper" may in some cases indicate a dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in the embodiments of this disclosure according to the specific circumstances.

[0022] Furthermore, the terms "set up," "connect," and "fix" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this disclosure according to the specific circumstances.

[0023] Unless otherwise stated, the term "multiple" means two or more.

[0024] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0025] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in the present disclosure can be combined with each other.

[0027] Combination Figures 1 to 3 As shown in the embodiments of this disclosure, a method for intelligent detection of defects in solid-state battery electrodes using X-ray images is provided, comprising:

[0028] S1. Obtain X-ray transmission images of the solid-state battery electrode to be tested; based on the design material composition parameters and layer thickness parameters of the electrode, establish an X-ray attenuation physical model that includes the linear attenuation coefficient of the material and the layer thickness parameters; use the X-ray attenuation physical model to perform scattering correction calculations and thickness non-uniformity compensation calculations on the X-ray transmission images to obtain a structural correction image.

[0029] S2. Based on the spatial arrangement design parameters of the electrode material layer, solid electrolyte layer and interface transition region, perform layered structural segmentation calculation on the structural correction image to obtain structural feature maps corresponding to the electrode material layer, solid electrolyte layer and interface transition region respectively.

[0030] S3. Calculate the interface continuity parameters, grayscale statistical parameters, and texture feature parameters from the structural feature map to form a multidimensional numerical feature vector; based on the solid-state battery failure mechanism model, convert the multidimensional numerical feature vector into numerical defect characterization parameters for quantifying interface contact integrity and material spatial uniformity.

[0031] S4. Input the numerical defect characterization parameters into the pre-trained defect identification model for classification and localization calculation. The loss function of the defect identification model includes a stacking order consistency constraint term, which is used to restrict the spatial adjacency relationship of each layer structure in the prediction result to conform to the physical stacking order of the electrode. Output the defect type and defect spatial location information.

[0032] S5, Output electrode defect detection results.

[0033] This embodiment provides an intelligent X-ray image detection method for defects in solid-state battery electrodes. The overall process includes: X-ray physical imaging modeling and image correction, layered structure analysis, construction of defect physical feature parameters, intelligent defect identification, and process feedback and risk assessment. This method can be deployed on online electrode inspection production lines and can also be used for offline quality analysis.

[0034] In actual production online inspection, this embodiment uses a high-resolution X-ray detector to capture real-time transmission images of solid-state battery electrodes. First, a pre-defined physical model is used to preprocess the images, transforming the original grayscale images, which are affected by scattering interference and thickness fluctuations, into structurally corrected images reflecting the essential characteristics of the material. Next, the algorithm automatically identifies three core regions: electrodes, electrolyte, and interfaces, and extracts multi-dimensional features that can quantify defects. Finally, a deep learning model with physical constraints is used to accurately identify and locate defects such as cracks, pores, and foreign objects. This process solves the pain points of traditional visual inspection, which struggles to penetrate the complex multi-layered structure of solid-state batteries and cannot quantify the quality of internal interface contacts, thus improving the quality control accuracy of the all-solid-state battery production line.

[0035] In addition, the imaging system includes: a microfocus X-ray source, a flat panel detector, a conveyor belt mechanism, a standard stepped test block, and an edge computing unit. The X-ray source operates in the energy spectrum range of 20 keV to 120 keV.

[0036] For the corrected image Calculate the gradient: ;

[0037] Constructing the structure tensor: .

[0038] In multiscale Calculate its eigenvalues ;

[0039] Define directional consistency: Used to distinguish layered structural regions.

[0040] Multidimensional defect feature vectors, forming a vector:

[0041]

[0042] Numerical defect characterization parameters are obtained through normalization.

[0043] Optionally, establishing an X-ray attenuation physical model includes: calculating the linear attenuation coefficient under a preset X-ray energy spectrum based on the effective atomic number and electron density parameters of each material component of the electrode; and constructing an equivalent attenuation coefficient matrix of the multilayer composite medium by combining the design thickness parameters of each layer.

[0044] The thickness non-uniformity compensation calculation includes: converting the transmission intensity of each pixel in the transmission image into path integral attenuation values; solving the thickness deviation field based on the equivalent attenuation coefficient matrix; and using the thickness deviation field to perform pixel-level gain correction processing on the transmission image to reduce artifacts caused by thickness fluctuations.

[0045] The system pre-calculates the attenuation characteristics of lithium metal, sulfide, or oxide electrolytes under a specific energy spectrum based on their elemental composition. During the inspection process, when slight thickness fluctuations occur in the electrode due to uneven coating, the compensation algorithm automatically calculates the thickness deviation field and corrects the pixel grayscale gain. Its technical advantage lies in eliminating false defect interference in the image, ensuring the consistency of features in subsequent inspections, and accurately distinguishing true density anomalies caused by internal material porosity even when electrode thickness fluctuates within acceptable limits.

[0046] For any pixel, its transmission intensity satisfies:

[0047]

[0048] in: For incident intensity, It is the linear attenuation coefficient.

[0049] For multilayer composite electrode structures, we have:

[0050]

[0051] in: For the first The linear attenuation coefficient of the layer material, This corresponds to the layer thickness.

[0052] The material attenuation coefficient is obtained through the effective atomic number. and electron density Obtained through calculation.

[0053] Perform a logarithmic transformation on the transmission image:

[0054] The thickness deviation is calculated by inversely using the equivalent attenuation coefficient matrix:

[0055]

[0056] in, For design thickness.

[0057] After constructing the thickness deviation field, gain compensation is performed on the image:

[0058]

[0059] in, It is a piecewise linear gain function.

[0060] Optionally, the hierarchical segmentation calculation includes: calculating the local gradient direction consistency parameter of the image using a multi-scale structural tensor algorithm; constructing a variational segmentation model containing data terms and regularization terms; and iteratively optimizing the variational segmentation model using the local gradient direction consistency parameter to obtain the pixel occupancy probability map of each layer.

[0061] This embodiment does not rely on a simple grayscale threshold during the segmentation stage. Instead, it utilizes a multi-scale structural tensor to capture the directional derivative features of the electrode interlayer interface. Through iteration of the variational segmentation model, the electrode is spatially and precisely divided into different physical layer probability maps. This method performs exceptionally well when dealing with solid-state batteries where materials have similar refractive indices and blurred interfaces. It can effectively separate the electrolyte layer from the electrode material layer, providing a precise geometric basis for analyzing interfacial contact failures.

[0062] Constructing the energy functional:

[0063]

[0064] in: For the segmentation results, These are the weighting coefficients.

[0065] Iterative optimization using gradient descent:

[0066]

[0067] The final results are: electrode layer probability map, electrolyte layer probability map, and interface transition region probability map.

[0068] Optionally, the interface continuity parameter is obtained by calculating the gray-level gradient continuity in the interface normal direction; the texture feature parameter includes gray-level co-occurrence matrix feature or local binary mode feature; the defect characterization parameter also includes the pore tortuosity parameter composed of the ratio of pore connectivity path length to Euclidean distance, and the interface sharpness parameter calculated from the interface curvature distribution.

[0069] In practical implementation, the system not only extracts conventional texture information but also focuses on calculating the gradient continuity of the interface normal and the tortuosity of the pores. For example, by calculating the ratio of the actual path of the pores to the straight-line distance, the degree of obstruction to ion transport can be assessed. The interface sharpness parameter can reflect the potential for stress concentration between the electrolyte and the electrode. These parameters transform qualitative image information into quantitative indicators that can be used for electrical performance prediction, enabling the detection results to be directly fed back into the battery's electrochemical performance evaluation system.

[0070] Calculate the gray-level gradient along the interface normal:

[0071]

[0072] Calculate its standard deviation ; The larger the value, the less continuous the interface.

[0073] Extract the pore connectivity; calculate the pore tortuosity parameter:

[0074]

[0075] in, The shortest path length. The distance is Euclidean.

[0076] Calculate the interface sharpness parameter based on curvature:

[0077]

[0078] Optionally, the loss function of the defect identification model includes a stacked topology consistency constraint term, which is constructed based on the physical adjacency relationship between each layer of the electrode. During the training phase, the defect identification model generates simulated defect samples through a generative adversarial network and is jointly trained with real electrode samples through transfer learning.

[0079] This embodiment introduces prior topological constraints during the model training phase, such as stipulating that the electrolyte layer must be located between the positive and negative electrode material layers. Simultaneously, it utilizes a Generative Adversarial Network (GAN) to simulate various extreme or rare defect samples, such as micro-delamination or micro-cracks, and combines this with transfer learning to improve the model's generalization ability. The technical effect is to avoid misjudgments by the AI ​​model that violate physical common sense and to solve the problem of insufficient training caused by the scarcity of real defect samples in the early stages of solid-state battery production.

[0080] It should be noted that the defect identification model in this embodiment is not a general image classification network, but rather deeply integrates the physical topological prior of solid-state battery electrodes into its loss function. During the training phase, the model receives a multi-dimensional vector after structural correction and feature extraction. Its total loss function consists of the basic classification loss and a layer order consistency constraint term. Specifically, this constraint term establishes a topological adjacency matrix to impose a high penalty on the spatial distribution of pixels in the prediction results that violate the physical logic of "positive electrode-transition layer-electrolyte-transition layer-negative electrode". For example, if the model directly identifies the electrolyte layer on the surface of the current collector in the prediction result while omitting the active material layer, the constraint term will generate significant gradient feedback, forcing the model to correct its spatial logic judgment.

[0081] Furthermore, addressing the scarcity of genuine defect samples (such as internal microcracks and localized chemical inhomogeneities) in the early stages of solid-state battery production, this embodiment utilizes a generator to simulate pseudo-defect images conforming to a decay physics model within the feature space, and then fuses them with genuinely acquired normal electrode images. This approach not only expands the training set but also enables the model to learn the essential evolutionary characteristics of defects under X-rays through transfer learning. Through this training mechanism, the model's accuracy in identifying subtle layered defects within the electrode is significantly improved, effectively avoiding false alarms caused by the lack of physical common sense in traditional AI models.

[0082] Add the following to the loss function:

[0083]

[0084] in: ; To predict the adjacency matrix; This is the physical layer adjacency matrix.

[0085] Constructing Generative Adversarial Networks:

[0086]

[0087] in, These are physical constraint terms.

[0088] The network structure adopts a convolutional neural network or a Transformer structure, and the input is a defect feature vector and a structural probability map.

[0089] Optionally, after outputting the electrode defect detection results, the local current density distribution parameters and stress concentration parameters are calculated using a preset performance evolution prediction model based on the defect type and defect spatial location information, and an electrode failure risk score is generated.

[0090] After detecting defects, this embodiment further inputs the spatial information of the defects into an evolution model to simulate the distribution of current density around the defects. If a pore causes excessively high local current density, the system will give a high-risk score and warn of possible dendrite growth. This upgrades the detection from simply "finding defects" to "predicting lifetime," providing a scientific basis for the graded screening of solid-state batteries.

[0091] This embodiment demonstrates the leap from two-dimensional transmission detection to three-dimensional spatial evaluation. The system collects structurally corrected images from multiple projection angles and uses filtered backprojection or iterative reconstruction algorithms to reconstruct the microscopic three-dimensional morphology inside the electrode. Based on this, the algorithm focuses on extracting the three-dimensional curvature parameters at the interface. Since the local stress at the electrolyte-electrode interface is highly correlated with the interface shape during solid-state battery cycling, high curvature regions often mean greater stress concentration during volume expansion.

[0092] After completing the 3D feature extraction, the system further invokes a pre-defined performance evolution prediction model. This model transforms the detected geometric parameters, such as interface contact integrity, porosity, and interface curvature, into boundary conditions in electrochemical simulation, thereby calculating the distribution of local current density. If a significant surge in current density is detected in a certain area, it indicates that the electrolyte-electrode contact is too tight or that conductive foreign matter is present, which can easily induce lithium dendrite growth. Finally, the system integrates current density non-uniformity and stress concentration parameters to generate a quantified electrode failure risk score. This scoring mechanism allows the quality inspection results to directly guide downstream cell sorting, ensuring that only high-quality, low-risk electrodes enter subsequent assembly processes.

[0093] Construct a local current density model based on the defect location:

[0094]

[0095] Calculate stress concentration:

[0096]

[0097] Establish a failure risk score:

[0098]

[0099] Will Feedback is sent to the coating pressure control module and the cold pressing line pressure control module to achieve closed-loop regulation.

[0100] Optionally, acquiring X-ray transmission images includes acquiring high-energy spectral transmission images and low-energy spectral transmission images; calculating pixel-level effective atomic number and electron density parameters based on the dual-energy spectral attenuation model, and obtaining low-atomic-number array sub-images and high-atomic-number array sub-images based on a material decomposition algorithm.

[0101] The implementation uses high- and low-energy X-rays for simultaneous imaging. By utilizing the differences in the response of different materials to different energy spectra, the effective atomic number of each pixel is calculated. This can effectively distinguish between impurities and active materials with similar atomic numbers, and achieve precise stripping of internal components of solid-state batteries composed of light elements (such as lithium) and heavy elements (such as transition metals), greatly improving the sensitivity of foreign matter detection.

[0102] For example, two sets of energy spectra are used: high-energy E1 and low-energy E2.

[0103] Analytical system of equations:

[0104]

[0105]

[0106] Solve for pixel-level effective atomic number and electron density.

[0107] This generates low atomic number array sub-maps and high atomic number array sub-maps, which are used to distinguish between solid electrolytes and metal inclusions.

[0108] Optionally, the defect identification model also receives the production process parameters corresponding to the electrode, including cold pressing pressure, baking temperature and coating gap; the defect identification model calculates the correlation weight between defect features and production process parameters through an attention mechanism and outputs the potential process anomaly category.

[0109] This embodiment inputs real-time process data such as cold pressing pressure and baking temperature from the production line along with image features into the model. Through an attention mechanism, the model can automatically detect the strong correlation between "abnormal coating gaps" and "electrode edge cracking." In this way, not only can defects be detected, but the root cause of the defects can also be traced, helping to optimize production parameters and shorten the process debugging cycle.

[0110] Optionally, the three-dimensional topography of the structural correction image is reconstructed based on multi-angle projection data, and the three-dimensional curvature parameters of the interface are calculated; the defect characterization parameters include the local stress concentration factor calculated from the three-dimensional curvature parameters.

[0111] By collecting X-ray projections from multiple angles, the embodiment utilizes a tomographic reconstruction algorithm to reconstruct the three-dimensional geometry inside the electrode. Based on this, the three-dimensional curvature of the interface is calculated, enabling precise identification of local wrinkles or stress concentration points that are difficult to detect in two-dimensional images.

[0112] Collect multi-angle projection data; employ a filtered back-projection algorithm:

[0113]

[0114] Calculate the 3D curvature of the interface:

[0115]

[0116] in, Principal curvature.

[0117] Optionally, the training of the defect identification model includes a pre-training phase based on a mask-based self-supervised task, and a data augmentation training phase based on generative adversarial networks.

[0118] This embodiment utilizes a large number of unlabeled normal poleme images for self-supervised training during the model development phase, enabling the model to deeply understand the standard structure of polemes. Subsequently, targeted defect enhancement training is performed using a GAN. This training strategy significantly reduces the reliance on manually labeled data.

[0119] Optionally, after establishing the X-ray attenuation physical model, a reference attenuation value is obtained by setting a standard stepped test block, and the parameters of the attenuation physical model are calibrated in real time based on the reference attenuation value.

[0120] A set of standard stepped test blocks of known density is configured in the testing equipment. The system automatically scans the test blocks during operation intervals to obtain the reference attenuation value. The technical advantage of this embodiment is that it effectively offsets the measurement deviation caused by X-ray source aging or detector temperature drift, ensuring that the defect detection standard for different batches and time periods remains at the same baseline, thus guaranteeing the stability of large-scale mass production quality testing.

[0121] For example, standard stepped test blocks are arranged at the edge of the conveyor belt.

[0122] Real-time acquisition of its attenuation value ,like: Then correct the attenuation model parameters. .

[0123] The calibration cycle can be set to 1 to 10 seconds.

[0124] In some alternative embodiments, the acquisition of the structure-corrected image also includes establishing a scattering kernel function library based on Monte Carlo simulation; the scattering correction calculation improves the contrast of the interface transition region by iteratively convolving the scattering kernel function library with the initially reconstructed density map to subtract the incoherent scattering component from the original transmission intensity.

[0125] In some alternative embodiments, the layer order consistency constraint is implemented by constructing a graph convolutional neural network (GNN), which maps the identified layer structure to graph nodes and defines the physical adjacency relationship between layers as edge constraints. By calculating the topological loss of node features, cross-layer identification results that violate physical logic are forcibly filtered out.

[0126] In some optional embodiments, the defect identification model also integrates a process deviation early warning module; this module analyzes the evolution trend of defect characterization parameters of consecutive batches of electrodes through a long short-term memory network (LSTM), calculates the spatiotemporal correlation entropy between defect distribution characteristics and fluctuations in production process parameters, and automatically outputs process parameter adjustment instructions before the defect rate reaches the early warning threshold.

[0127] In some alternative embodiments, when performing layered structure segmentation calculations, adaptive anisotropic diffusion filtering is applied to the electrode edge region to eliminate grayscale abrupt interference caused by edge diffraction; and the edge shrinkage vector field is extracted to identify overflow or shrinkage defects of the solid electrolyte layer in the edge-sealing process.

[0128] In some alternative embodiments, the performance evolution prediction model also includes parameter mapping logic based on an electrochemical impedance spectroscopy (EIS) equivalent circuit, which converts the calculated interface contact integrity and porosity tortuosity parameters into a local charge transfer impedance distribution map, and quantifies the rate performance consistency of the electrode by calculating the non-uniformity factor of the overall impedance.

[0129] In some alternative embodiments, real-time calibration also includes using a multi-energy spectrum detector to capture the energy spectrum hardening effect caused by X-ray tube voltage fluctuations; and dynamically compensating the energy spectrum response parameters in the X-ray attenuation physical model by analyzing the changes in the low-energy to high-energy ratio of the standard stepped test block in real time, thereby ensuring the absolute stability of the test results in complex production environments.

[0130] Regarding the accuracy of the physical model, by introducing Monte Carlo scattering kernel functions and energy spectrum hardening compensation, the system can handle the detection scenarios of solid-state batteries with higher energy and more complex thicknesses, solving the problem of blurred imaging of large-thickness electrodes. This allows X-rays to not only visualize images, but also provide density and thickness data with absolute physical meaning, much like a high-precision sensor.

[0131] Regarding the logical completeness of the algorithm, a graph convolutional neural network is introduced to handle the stacked topological relationships, elevating the detection task from simple image classification to the level of physical logical reasoning. This method effectively utilizes the inherent physical arrangement rules of solid-state batteries (such as the inevitable order of positive electrode-electrolyte-negative electrode) to constrain AI predictions, greatly reducing the probability of false alarms at complex interfaces. Simultaneously, targeted algorithms for edge diffraction and process offsets enable the detection system to possess stronger industrial anti-interference capabilities and forward-looking predictive abilities.

[0132] By mapping defect parameters to electrochemical impedance distribution and process adjustment instructions, this approach achieves a leap from "static detection" to "dynamic feedback." It no longer simply provides a pass / fail conclusion, but deeply participates in battery performance evaluation and production control. For example, by identifying the spatiotemporal correlation entropy between defects and process parameters, the system can clearly inform the operator whether the current interface cracking is caused by fluctuations in cold pressing pressure or by periodic oscillations in the coating gap, thus truly realizing the core pivotal role of intelligent detection in intelligent manufacturing.

[0133] In the laboratory or pilot-scale stage, for solid-state batteries using novel electrolytes such as sulfides or oxides, this embodiment focuses on the in-depth deconstruction of the material's microstructure. Using a dual-energy-density attenuation model, researchers can clearly separate the low-atomic-number active material from the high-atomic-number solid electrolyte component in images, even if they highly overlap in the thickness direction, obtaining their independent distribution maps through material decomposition algorithms. At this point, the porosity tortuosity parameter and the three-dimensional curvature parameter of the interface become key indicators for evaluating battery performance. By analyzing the continuity of the interface normal grayscale gradient, researchers can quantify the impact of different sintering processes on the interface transition region, thereby accurately predicting the charge transfer impedance. This application not only identifies defects but also provides a scientific quantitative evaluation basis for material formulation and interface modification schemes through a data-driven approach.

[0134] In large-scale industrial production lines, this embodiment can be used for real-time quality control after electrode coating and cold pressing. Since the electrodes may experience slight vibrations or coating thickness fluctuations during high-speed transport, the system performs pixel-level gain correction on the X-ray transmission image in real time by solving for the thickness deviation field. This process effectively filters artifacts caused by mechanical vibration, ensuring that detection accuracy does not decrease with increasing production speed. Simultaneously, the system performs real-time dynamic calibration using standard stepped test blocks to compensate for energy drift caused by long-term operation of the X-ray source. In this scenario, process parameter correlation is particularly important. When the detection model detects a decrease in interface contact integrity, it can automatically correlate with the current cold pressing pressure data and use an attention mechanism to determine whether insufficient pressure has led to interlayer porosity, thereby achieving closed-loop optimization of the production process and early detection and removal of defects.

[0135] Furthermore, this embodiment can achieve a leap from "defect detection" to "risk prediction" by constructing a complex failure mechanism model. During the detection process, the multi-scale structural tensor algorithm can define the physical boundary between the electrode and the electrolyte, even if there are extremely fine stress cracks inside the electrode. The system not only outputs the spatial location of the defect, but also calculates the risk of a surge in local current density and stress concentration parameters caused by the defect based on the performance evolution prediction model. If the connection path of a certain internal pore may induce lithium dendrite penetration, the system will give a failure risk score and trigger an early warning. In addition, to address the problem of insufficient new defect samples, generative adversarial networks and masked self-supervised tasks play a core role, ensuring that when faced with complex defect morphologies that have never been seen before, the model can still make classification and grading judgments that conform to physical logic based on its understanding of the physical stacking order of the electrode.

[0136] The foregoing description and accompanying drawings fully illustrate embodiments of the present disclosure to enable those skilled in the art to practice them. Other embodiments may include structural and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included or substituted for parts and features of other embodiments. Embodiments of the present disclosure are not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from its scope. The scope of the present disclosure is limited only by the appended claims, and the foregoing embodiments should be considered exemplary and non-limiting.

Claims

1. A method for intelligent X-ray image detection of defects in solid-state battery electrodes, characterized in that, include: S1. Obtain an X-ray transmission image of the solid-state battery electrode to be tested; based on the design material composition parameters and layer thickness parameters of the electrode, establish an X-ray attenuation physical model including the linear attenuation coefficient of the material and the layer thickness parameters; use the X-ray attenuation physical model to perform scattering correction calculation and thickness non-uniformity compensation calculation on the X-ray transmission image to obtain a structural correction image. S2. Based on the spatial arrangement design parameters of the electrode material layer, solid electrolyte layer and interface transition region, perform layered structural segmentation calculation on the structure correction image to obtain structural feature maps corresponding to the electrode material layer, solid electrolyte layer and interface transition region respectively. S3. Calculate the interface continuity parameter, grayscale statistical parameter and texture feature parameter from the structural feature map to form a multidimensional numerical feature vector; Based on the solid-state battery failure mechanism model, the multidimensional numerical feature vector is converted into numerical defect characterization parameters for quantifying interface contact integrity and material spatial uniformity. S4. Input the numerical defect characterization parameters into the pre-trained defect identification model for classification and localization calculation. The loss function of the defect identification model includes a stacking order consistency constraint term, which is used to restrict the spatial adjacency relationship of each layer structure in the prediction result to conform to the physical stacking order of the electrode. Output defect type and defect spatial location information. S5, Output electrode defect detection results.

2. The method according to claim 1, characterized in that, The establishment of the physical model for X-ray attenuation includes: Based on the effective atomic number and electron density parameters of each material component of the electrode, the linear attenuation coefficient under the preset X-ray energy spectrum is calculated. Based on the design thickness parameters of each layer, an equivalent attenuation coefficient matrix of the multilayer composite medium is constructed. The thickness non-uniformity compensation calculation includes: The transmission intensity of each pixel in the transmission image is converted into path integral attenuation value; The thickness deviation field is solved based on the equivalent attenuation coefficient matrix. The thickness deviation field is used to perform pixel-level gain correction on the transmission image to reduce artifacts caused by thickness fluctuations.

3. The method according to claim 1, characterized in that, The hierarchical structure segmentation calculation includes: The multi-scale structure tensor algorithm is used to calculate the local gradient direction consistency parameter of the image; Construct a variational partitioning model that includes data items and regularization terms; The variational segmentation model is iteratively optimized using the local gradient direction consistency parameter to obtain the pixel occupancy probability map of each layer structure.

4. The method according to claim 1, characterized in that: The interface continuity parameter is obtained by calculating the gray-level gradient continuity in the interface normal direction; The texture feature parameters include gray-level co-occurrence matrix features or local binary pattern features; The defect characterization parameters also include the pore tortuosity parameter, which is the ratio of the pore connectivity path length to the Euclidean distance, and the interface sharpness parameter, which is calculated from the interface curvature distribution.

5. The method according to claim 1, characterized in that: The loss function of the defect identification model includes a stacked topology consistency constraint term, which is constructed based on the physical adjacency relationship between each layer of the electrode. The defect identification model generates simulated defect samples through a generative adversarial network during the training phase, and is jointly trained with real electrode samples through transfer learning.

6. The method according to claim 1, characterized in that: After outputting the electrode defect detection results, the local current density distribution parameters and stress concentration parameters are calculated using a preset performance evolution prediction model based on the defect type and defect spatial location information, and an electrode failure risk score is generated.

7. The method according to claim 1, characterized in that: Acquiring X-ray transmission images includes acquiring high-energy spectral transmission images and low-energy spectral transmission images; The effective atomic number and electron density parameters at the pixel level are calculated based on the dual-energy spectral attenuation model, and the low atomic number array sub-image and the high atomic number array sub-image are obtained based on the material decomposition algorithm.

8. The method according to claim 1, characterized in that: The defect identification model also receives production process parameters corresponding to the electrode, including cold pressing pressure, baking temperature and coating gap. The defect identification model calculates the correlation weight between defect features and production process parameters through an attention mechanism and outputs the category of potential process anomalies.

9. The method according to claim 1, characterized in that: Three-dimensional topography reconstruction is performed on the structure correction image based on multi-angle projection data, and the three-dimensional curvature parameters of the interface are calculated. The defect characterization parameters include the local stress concentration factor calculated from the three-dimensional curvature parameters.

10. The method according to claim 1, characterized in that: The training of the defect identification model includes a pre-training phase based on a mask-based self-supervised task, and a data augmentation training phase based on generative adversarial networks.

Images