Artificial intelligence visual inspection method for battery manufacturing defects and adaptive correction system

Abstract

This invention relates to the field of battery testing technology, specifically to an AI visual inspection method and adaptive correction system for battery manufacturing defects. The method includes acquiring multimodal data of the battery to be inspected; outputting a defect correlation map based on the multimodal data, wherein the defect correlation map correlates the electrolyte and coating of the battery; extracting material gene features from the defect correlation map using an AI visual inspection model; performing transfer calculations on the material gene features to obtain root cause classification and defect location coordinates; performing equipment parameter adjustment operations based on the root cause classification and defect location coordinates to obtain battery manufacturing defect detection results after equipment parameter adjustment; and inputting the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization to obtain an optimized AI visual inspection model, thereby improving the accuracy and response speed of visual inspection.

Description

Technical Field

[0001] This invention relates to the field of battery inspection technology, specifically to an AI visual inspection method and adaptive correction system for battery manufacturing defects. Background Technology

[0002] With the rapid development of electric vehicles, portable electronics and other fields, the safety and reliability of key components such as lithium-ion batteries are becoming increasingly important. Manufacturing defects such as electrode misalignment and uneven coating may lead to serious consequences such as short circuits and thermal runaway, threatening personal safety and causing economic losses.

[0003] Currently, the industry mainly relies on manual visual inspection or visual inspection solutions based on conventional machine vision technology. However, these solutions have drawbacks such as low inspection efficiency, high false detection rate, and lack of real-time feedback capability on high-speed production lines, making it difficult to identify and handle defects in a timely manner.

[0004] Therefore, how to improve the accuracy and response speed of battery visual inspection is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To improve the accuracy and response speed of battery visual inspection, this application provides an AI visual inspection method and adaptive correction system for battery manufacturing defects.

[0006] The AI ​​visual detection method for battery manufacturing defects provided in this application adopts the following technical solution:

[0007] An AI-based visual detection method for battery manufacturing defects includes:

[0008] Acquire multimodal data of the battery under test, and output a defect correlation map based on the multimodal data. The defect correlation map is associated with the electrolyte and coating of the battery under test.

[0009] By using an AI visual inspection model, material gene features are extracted from the defect association map. The transfer calculation of material gene features is performed to obtain the root cause classification and defect location coordinates.

[0010] Based on the root cause classification and defect location coordinates, perform equipment parameter adjustment operations to obtain the battery manufacturing defect detection results after equipment parameter adjustment;

[0011] The battery manufacturing defect detection results are input into the AI ​​visual inspection model for optimization, resulting in an optimized AI visual inspection model.

[0012] Further steps for acquiring multimodal data of the battery under test include:

[0013] A tunable pulse is emitted to the battery under test. The tunable pulse penetrates the electrolyte layer of the battery under test through the battery surface. The molecular vibration decay signal in the electrolyte layer is captured by a quantum dot array to obtain a microbubble distribution map. The diffraction pattern on the surface of the battery under test is acquired to generate an optical diffraction image. The surface temperature of the battery under test is sampled, and a thermal field distribution video stream is generated based on the surface temperature distribution.

[0014] The multimodal data includes microbubble distribution maps, optical diffraction images, and thermal field distribution video streams.

[0015] Furthermore, the steps for outputting a defect correlation map based on multimodal data include:

[0016] Spatial registration and temporal synchronization are performed on the multimodal data to obtain a multimodal dataset;

[0017] Feature extraction and reconstruction were performed on the multimodal dataset to obtain a defect association map.

[0018] Furthermore, the steps for spatial registration and temporal synchronization of multimodal data to obtain a multimodal dataset include:

[0019] A unified spatial coordinate system was established based on the chessboard calibration method;

[0020] The microbubble distribution map, optical diffraction image, and thermal field distribution video stream are aligned to a unified spatial coordinate system using an affine transformation matrix to obtain the registration matrix.

[0021] After triggering the temporal synchronization of the microbubble distribution map, optical diffraction image, and thermal field distribution video stream, a multimodal dataset is generated by combining the registration matrix. The multimodal dataset includes microbubble density distribution data, coating thickness deviation data, and electrolyte wetting state binary mask.

[0022] Furthermore, the steps for feature extraction and reconstruction of the multimodal dataset to obtain the defect association map include:

[0023] Input the microbubble density distribution data and coating thickness deviation data into the fusion analysis module to identify the coupling relationship;

[0024] By combining the low-temperature unwetted areas marked by the electrolyte wetting state binary mask, a defect correlation map is obtained.

[0025] Furthermore, the steps of extracting material gene features from the defect association map, performing transfer calculations on these material gene features, and obtaining root cause classification and defect location coordinates include:

[0026] After extracting the bond energy decay rate, pyrolysis rate and thermal stress coefficient based on the defect correlation map, the probability of local bonding failure is obtained based on the bond energy decay rate, the electrolyte decomposition risk value is obtained based on the pyrolysis rate, and the current collector fatigue accumulation index is obtained based on the thermal stress coefficient.

[0027] The relationship between the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector and the preset root cause judgment conditions is determined to obtain the root cause classification; and the coordinates of the defect location are determined based on the spatial location mapped by the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector.

[0028] Furthermore, based on root cause classification and defect location coordinates, the steps for performing equipment parameter adjustment operations include:

[0029] Based on the root cause classification and defect location coordinates, determine the equipment to be adjusted and the adjustment amount;

[0030] Based on the equipment to be adjusted and the adjustment amount, a simulation adjustment operation is performed to obtain the defect prediction recurrence rate, and it is determined whether the defect prediction recurrence rate is less than the preset defect prediction recurrence rate.

[0031] If the defect prediction recurrence rate is less than the preset defect prediction recurrence rate, then the parameters of the equipment to be adjusted are adjusted based on the adjustment amount.

[0032] If the predicted defect recurrence rate is greater than the preset predicted defect recurrence rate, then the step of determining the equipment to be adjusted and the adjustment amount based on the root cause classification and defect location coordinates is executed.

[0033] Furthermore, the steps for obtaining battery manufacturing defect detection results after equipment parameter adjustment include:

[0034] Obtain the battery defect rate after adjusting the equipment parameters, and determine whether the battery defect rate is less than the preset battery defect rate;

[0035] If the battery defect rate is less than the preset battery defect rate, the battery manufacturing defect detection result is generated based on the battery defect rate.

[0036] Furthermore, the steps for inputting the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization, resulting in the optimized AI visual inspection model, include:

[0037] The battery manufacturing defect detection results are used to generate a cryptographic gradient tensor through local training. The cryptographic gradient tensor is then stored on the blockchain and federated to obtain the aggregated global gradient.

[0038] The AI ​​visual detection model is optimized by aggregating the global gradient, resulting in an optimized AI visual detection model.

[0039] This application also provides an adaptive correction system applied to the AI ​​visual detection method for battery manufacturing defects as described above. The adaptive correction system includes:

[0040] The data acquisition module is used to acquire multimodal data of the battery under test and output a defect correlation map based on the multimodal data. The defect correlation map is associated with the electrolyte and coating of the battery under test.

[0041] The data calculation module is used to extract material gene features from the defect association map through the AI ​​visual detection model, perform transfer calculations on the material gene features, and obtain the root cause classification and defect location coordinates.

[0042] The data detection module is used to perform equipment parameter adjustment operations based on root cause classification and defect location coordinates, and obtain the battery manufacturing defect detection results after the equipment parameter adjustment.

[0043] The model optimization module is used to input the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization, resulting in an optimized AI visual inspection model.

[0044] Beneficial effects achieved:

[0045] This application proposes an AI (Artificial Intelligence) visual inspection method for battery manufacturing defects. Specifically, it involves acquiring multimodal data of the battery to be inspected, outputting a defect correlation map based on the multimodal data, wherein the defect correlation map correlates the electrolyte and coating of the battery to be inspected; extracting material gene features from the defect correlation map using an AI visual inspection model, performing transfer calculations on the material gene features to obtain root cause classification and defect location coordinates; performing equipment parameter adjustment operations based on the root cause classification and defect location coordinates to obtain the battery manufacturing defect detection results after equipment parameter adjustment; and inputting the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization to obtain an optimized AI visual inspection model.

[0046] In this application, by integrating multimodal data acquisition, defect correlation map analysis, material gene feature root cause tracing, and equipment closed-loop control processes, the accuracy and response speed of visual inspection are synergistically improved: At the data acquisition level, multimodal data is collected simultaneously, and a defect correlation map of electrolyte and coating is constructed through an AI visual inspection model, breaking through the limitations of conventional single sensors, fully covering defect morphology and material interaction relationships, and significantly improving the completeness of abnormal feature recognition and classification accuracy; At the defect diagnosis level, through graph network transfer calculation of material gene features, the source of defects is accurately traced, avoiding only identifying superficial errors, thus improving the root cause classification accuracy; At the response control level, based on the real-time output of defect location and root cause classification, the equipment execution parameters are adjusted in conjunction, compressing conventional manual intervention into an instantaneous closed loop on the production line, while using each inspection result to dynamically optimize the adaptive capability of the AI ​​visual inspection model, continuously reducing the false detection rate under new material processes and improving computational efficiency, ultimately achieving an effective improvement in visual inspection accuracy and speed. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating an AI visual detection method for battery manufacturing defects according to this application.

[0048] Figure 2 This is a flowchart illustrating the process of generating a defect correlation diagram based on multimodal data acquisition in this application.

[0049] Figure 3 This is a flowchart illustrating the defect correlation graph analysis in this application;

[0050] Figure 4 This is a flowchart illustrating the material genome characteristic root cause classification and tracing and equipment closed-loop control process in this application;

[0051] Figure 5 This is a schematic diagram of an adaptive correction system according to this application. Detailed Implementation

[0052] The following is in conjunction with the appendix Figure 1-5 This application will be described in further detail.

[0053] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present 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, and therefore should not be construed as a limitation of the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0054] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" 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 between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0055] This application discloses an AI visual detection method for battery manufacturing defects.

[0056] Please refer to Figure 1 In one embodiment of this application, an AI visual detection method for battery manufacturing defects includes steps S10-S40:

[0057] Step S10: Obtain multimodal data of the battery to be tested, and output a defect correlation map based on the multimodal data. The defect correlation map is associated with the electrolyte and coating of the battery to be tested.

[0058] In the battery manufacturing process, rigorous defect detection is essential to ensure battery safety and consistency. Currently, the industry commonly employs manual visual inspection or conventional machine vision solutions. Manual visual inspection relies on operator experience to sample and inspect each battery individually, resulting in extremely low efficiency and a significant risk of missing invisible defects, such as electrolyte microbubbles and microcracks within the coating. While conventional machine vision improves inspection speed, it can only acquire visible light images of the surface and cannot penetrate the internal battery structure to perceive the physical relationship between electrolyte flow and coating microstructure, leading to a persistently high false positive rate.

[0059] To address the aforementioned problems, this embodiment acquires multimodal data of the battery under test and integrates spectral imaging and thermal field sensing technologies that can penetrate the electrolyte layer. It reveals the intrinsic relationship between electrolyte state and coating deformation from different physical dimensions. By constructing a spatiotemporal mapping of electrolyte microbubble distribution, coating stress concentration area and thermal conduction anomaly, it provides a data foundation for subsequent in-depth analysis based on physical field coupling.

[0060] The output defect correlation map, by integrating the spatial distribution characteristics of both electrolyte and coating parameters—for example, mapping the bubble density gradient to the corresponding coating thickness reduction rate—visually presents the causal chain of "electrolyte abnormality → coating damage." For instance, a high bubble density in a certain area shows a strong correlation with a sudden reduction in coating thickness directly above it. This map directly drives the correction of coating pressure or vacuum injection parameters, elevating the fragmented characterization data in conventional methods into a diagnostic basis that can guide equipment adjustments.

[0061] Step S20: Using an AI visual detection model, extract the material gene features from the defect association map, perform transfer calculations on the material gene features, and obtain the root cause classification and defect location coordinates.

[0062] It should be noted that the AI ​​visual inspection model in this embodiment is a visual inspection system for battery manufacturing defects built on a deep neural network. It learns the inherent patterns of a large amount of historical battery defect data through pre-training, and achieves the ability to analyze the multi-physics defect correlation map.

[0063] Extracting material gene features from defect correlation maps is a key step in mapping observed macroscopic physical anomalies, such as coating thinning and bubble aggregation, to microscopic material failure mechanisms. Material gene features, through physical field data reconstruction, reveal the essential laws of field material molecular bond energy decay, solvent pyrolysis reaction rate, etc., which can overcome the limitation of conventional detection that can only perceive surface phenomena.

[0064] The purpose of performing transfer calculations on the extracted material gene features is to simulate the dynamic propagation path of defects in the material microstructure. Using graph neural networks (GNNs) as a carrier, cross-scale causal inference is achieved, such as the transfer chain of bubble pulse pressure → coating molecular chain breakage probability. The root cause classification obtained at the end accurately determines the nature of the failure, such as binder dissociation or electrolyte decomposition. Meanwhile, the defect location coordinates pinpoint the core target of process intervention. Together, they provide accurate positioning and essential diagnosis basis for real-time correction of equipment parameters, enabling battery manufacturing to shift from passively intercepting defects to actively eliminating failure mechanisms, effectively improving the intelligence and convenience of battery testing.

[0065] Step S30: Based on the root cause classification and defect location coordinates, perform equipment parameter adjustment operation to obtain the battery manufacturing defect detection results after equipment parameter adjustment.

[0066] In the process of battery testing, the essence of adjusting equipment parameters based on root cause classification and defect location coordinates lies in breaking through the limitations of conventional testing, which only focuses on the identification of defect appearance. Although conventional battery testing solutions can detect phenomena such as coating cracks and abnormal coating bubbles, they cannot trace the essential mechanism of process failure, such as the breakage of binder molecular chains and the decomposition of electrolyte solvents. This leads to blind and delayed adjustment of equipment parameters. For example, manual parameter adjustment relies on experience and trial and error, which takes too long and still cannot prevent the recurrence of the same type of defect.

[0067] This embodiment, by locking the root cause classification and the precise location coordinates of the defect, can target and correct specific parameters of a specific device, thereby eliminating the defect generation mechanism from the source.

[0068] Furthermore, by acquiring the battery manufacturing defect detection results after the equipment parameters were adjusted, a closed-loop verification link was constructed: when the new battery to be tested after the parameters were corrected passes through online testing, the real-time feedback of the defect distribution changes can not only verify the effectiveness of the equipment parameter adjustment, but also provide incremental data for the dynamic optimization of the AI ​​visual inspection model, enabling the system to have continuous proactive optimization capabilities.

[0069] Compared to conventional battery testing technologies, this embodiment transforms battery manufacturing from passive defect interception to active process immunity. By diagnosing root cause classification and locating defect coordinates in real time, and adjusting equipment parameters in a targeted manner, it effectively improves the accuracy and convenience of equipment parameter adjustment to a certain extent, avoiding the inaccuracy and lag of manual experience-based adjustments.

[0070] Step S40: Input the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization to obtain the optimized AI visual inspection model.

[0071] The purpose of inputting the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization is to solve the inherent defects of conventional inspection solutions that cannot adapt to dynamic interferences such as material changes and equipment aging in battery manufacturing. Conventional inspection solutions rely on models built with fixed training sets, which have increased false detection rates when faced with new defects or process fluctuations, require frequent shutdowns for readjustment, and have excessively long response delays.

[0072] This embodiment optimizes the AI ​​visual inspection model online by inputting battery manufacturing defect detection results, enabling the AI ​​visual inspection model to have autonomous adaptive capabilities. It upgrades the conventional passive interception to active learning. The optimized AI visual inspection model can not only identify historical defects with higher accuracy, but also predictively capture negative risks induced by process fluctuations. Thus, the inspection function is transformed from post-event screening of defective products to pre-event defect prevention. Therefore, compared with conventional inspection solutions, it can significantly reduce the defective product production rate and avoid the increased costs caused by too many defective products.

[0073] This embodiment injects the AI ​​visual inspection model with the ability to update and optimize in real time, enabling the battery production line to maintain relatively stable inspection accuracy under variable environments such as material iteration and equipment wear.

[0074] Specifically, step S40 includes steps S41 to S42:

[0075] Step S41: Locally train the battery manufacturing defect detection results to generate an encrypted gradient tensor, perform blockchain notarization and federated aggregation on the encrypted gradient tensor to obtain the aggregated global gradient.

[0076] Step S42: Optimize the AI ​​visual detection model using the aggregated global gradient to obtain the optimized AI visual detection model.

[0077] In this embodiment, when the battery manufacturing defect detection results are trained locally through edge computing nodes, an adversarial training mechanism is adopted to enhance the model's ability to identify low-confidence samples and novel defects: the generator creates defect morphology perturbation samples that simulate process fluctuations, such as topological variants of coating cracks, while the discriminator learns subtle feature differences through a visual converter and a hybrid architecture of U-Net++. During the training process, the gradient matrix of the model parameters is calculated, and an encrypted gradient tensor with node digital signatures and timestamps is generated by the AES-256 encryption algorithm.

[0078] The encrypted gradient tensor is tamper-proofed through the blockchain notarization system—the SHA-256 hash operation is performed to generate a unique data fingerprint and write it into the Hyperledger. At the same time, the federated aggregation center is triggered to logarithmically weight the multi-source gradients according to the historical accuracy of the nodes, generating an aggregated gradient that incorporates global knowledge (i.e., the aggregated global gradient).

[0079] When optimizing the AI ​​visual inspection model based on this aggregated gradient, the gradient data is first decrypted and the convolution kernel weight parameters are updated. Then, a dedicated recognition channel for newly added defect types is expanded. Finally, the confidence threshold adaptive algorithm is reconstructed to output an AI visual inspection model with continuous optimization capabilities.

[0080] The optimized AI visual inspection model not only improves the accuracy of historical defect identification, but also incorporates new anomalies caused by process fluctuations into the inspection scope through an incremental learning mechanism, forming an intelligent closed loop of dynamic collaboration between inspection capabilities and the manufacturing environment.

[0081] Reference Figure 2 As shown, a feasible implementation method for generating a defect correlation diagram based on multimodal data acquisition is as follows, including steps S11 to S13:

[0082] Step S11: A tunable pulse is emitted to the battery under test. The tunable pulse penetrates the electrolyte layer of the battery under test through the battery surface and captures the molecular vibration decay signal in the electrolyte layer through the quantum dot array to obtain a microbubble distribution map; and the diffraction pattern on the surface of the battery under test is acquired to generate an optical diffraction image; and the surface temperature of the battery under test is sampled, and a thermal field distribution video stream is generated based on the surface temperature distribution; wherein, the multimodal data includes the microbubble distribution map, the optical diffraction image and the thermal field distribution video stream.

[0083] In this embodiment, a tunable pulse is directionally emitted to the battery under test that enters the detection range via a terahertz tunable transmitter. The tunable pulse penetrates the battery shell and electrode layer and acts on the electrolyte layer. Its specific frequency band excites the resonance of electrolyte solvent molecules. The attenuation intensity and phase shift of the molecular vibration signal are captured in real time by the quantum dot array in the quantum dot detector to obtain the molecular vibration attenuation signal and reconstruct the three-dimensional spatial distribution map of microbubbles in the electrolyte layer (i.e., microbubble distribution map).

[0084] Synchronously triggered coherent light, modulated by a Fresnel zone plate, illuminates the surface of the battery coating. The microstructure of the coating surface modulates the incident light wavefront, forming a diffraction wavefield containing morphological information. A high-speed industrial camera captures the interference fringe image of this diffraction wavefield on the imaging plane with a microsecond-level exposure time. During this process, the optical path difference caused by changes in coating thickness is directly reflected in changes in fringe spacing, orientation, and intensity.

[0085] After obtaining the original diffraction pattern based on the diffraction wave field and interference fringe image, the optical diffraction image is reconstructed through an iterative constrained projection algorithm. Specifically, the intensity distribution of the captured interference fringes is first used to simulate the propagation process of light waves from the coating surface to the imaging plane based on the angular spectrum propagation theory. In the iterative calculation, the initial phase estimate of the coating surface is continuously corrected until the simulated diffraction pattern matches the actual captured fringes with a matching degree of more than 99%. After outputting the corresponding phase distribution, the phase distribution is directly converted into the physical thickness of the coating to generate the optical diffraction image.

[0086] Meanwhile, the infrared thermal imager scans the battery surface temperature field at a sampling frequency of 30Hz, combines an ambient temperature compensation algorithm to eliminate interference, and dynamically marks areas with temperatures below the regional average, such as marking persistently low-temperature areas below 0.5℃, generating a time- and space-continuous thermal field distribution video stream.

[0087] The multimodal data obtained through the above steps can provide a cross-physical field coupled data foundation for subsequent defect correlation analysis.

[0088] Step S12: Spatial registration and temporal synchronization are performed on the multimodal data to obtain a multimodal dataset.

[0089] The core significance of spatial registration and temporal synchronization of multimodal data lies in overcoming the fragmented nature of multi-source sensor data in battery manufacturing and testing. Specifically, due to differences in physical location and sampling time drift, the microbubble distribution, coating thickness variation, and thermal conduction state of the electrolyte field layer are spatially misaligned and temporally disjointed. This means that when the thermal imager shows an abnormal temperature for the same physical event, the optical system may not have captured the deformation response of the associated coating, or even incorrectly associate bubble aggregation with coating cracking in non-adjacent areas.

[0090] Spatial registration eliminates sensor bias through coordinate system mapping, ensuring that the three-dimensional positioning points of microbubble clusters within the electrolyte are precisely aligned with the spatial coordinates of the coating directly above them. This exposes the true coupling relationship between bubble hydraulic impact and coating stress distortion. Temporal synchronization eliminates the causal inversion caused by sampling delays from different sensing units, reconstructing a strict sequential logical chain of multi-physics field changes. The spatiotemporally calibrated multimodal dataset constructs a chain of evidence for the electrolyte dynamics corresponding to the microbubble distribution map, the coating mechanical response corresponding to the optical diffraction image, and the thermodynamic state corresponding to the thermal field distribution video stream. For example, when the data reveals that the peak microbubble density consistently occurs 0.5 seconds earlier than the abrupt change in coating thickness directly above it, and the thermal field gradient change conforms to the fluid-solid energy transfer model, the system can pinpoint the complete failure path: "fluid injection turbulence induces bubble aggregation → hydraulic impact leads to coating fatigue → increased thermal resistance causes localized low temperatures."

[0091] The generation of defect correlation maps must rely on such strictly aligned multimodal datasets to elevate discrete abnormal signals into diagnosable process root causes, avoid the isolated analysis of phenomena by conventional inspections, and transform battery manufacturing from passive sampling to active closed-loop control, realizing the transformation of defect attribution from probabilistic guessing to inevitable laws.

[0092] In one feasible implementation, step S12 includes steps S121 to S123:

[0093] Step S121: Establish a unified spatial coordinate system based on the checkerboard calibration method.

[0094] Step S122: Align the microbubble distribution map, optical diffraction image, and thermal field distribution video stream to a unified spatial coordinate system using an affine transformation matrix to obtain a registration matrix.

[0095] Step S123: After triggering the time synchronization of the microbubble distribution map, optical diffraction image and thermal field distribution video stream, a multimodal dataset is generated by combining the registration matrix. The multimodal dataset includes microbubble density distribution data, coating thickness deviation data and electrolyte wetting state binary mask.

[0096] A high-precision checkerboard calibration plate is installed at a fixed position on the battery conveyor belt. When the battery passes the detection position, the terahertz detection system, optical diffraction camera, and infrared thermal imager simultaneously capture images of the battery with the checkerboard pattern. The computer vision algorithm extracts the coordinates of the checkerboard corner points in the images of each system. Using the bubble coordinate system of the terahertz detection system as a reference, the spatial mapping relationship between the thickness image of the optical diffraction camera and the checkerboard feature points in the infrared thermal imager is calculated to establish a unified spatial coordinate system.

[0097] When aligning the microbubble distribution map, the unified spatial coordinate system mentioned above is directly used as the reference space. The pixel coordinates of the thickness image corresponding to the optical diffraction image are mapped to the bubble coordinate system to calculate the first affine matrix. The thickness data is resampled and registered through bilinear interpolation. The pixels of the thermal field image corresponding to the thermal field distribution video stream are mapped to the bubble coordinate system to generate the second affine matrix to achieve frame-by-frame coordinate transformation. Finally, the registration matrix of the three sets of data under the unified spatial grid is obtained.

[0098] Timing synchronization is achieved through hardware triggering: the PLC controller sends an edge trigger signal the instant the battery enters the detection zone, the terahertz tunable transmitter starts a tunable pulse sequence, and the quantum dot detector records the initial timestamp value. Simultaneously, the trigger signal activates the laser to emit coherent light and initiates exposure by the high-speed industrial camera. The infrared thermal imager starts scanning 0.3ms after receiving the trigger signal. All data stream headers embed the same trigger sequence ID and timestamp.

[0099] Finally, the registration matrix is ​​applied to the original data stream. For example, each three-dimensional coordinate point in the microbubble distribution map is projected onto the two-dimensional coating thickness plane through an affine transformation and superimposed with the corresponding timestamp of the thermal field frame data to generate a multimodal dataset. The microbubble density distribution data is generated from the mapping relationship between bubble volume and coordinates, the coating thickness deviation data is derived from the difference between the registered diffraction image and the reference thickness model, and the electrolyte wetting state binary mask is converted into a 0-1 matrix based on the spatial distribution of the sustained low temperature zone below the regional average temperature, such as below 0.5℃, according to the thermal field data. The three constitute a spatiotemporally strictly aligned multimodal dataset for subsequent defect correlation map analysis.

[0100] Step S13: Perform feature extraction and reconstruction on the multimodal dataset to obtain a defect association map.

[0101] In this embodiment, the core value of feature extraction and reconstruction of multimodal datasets lies in solving the fundamental defect of isolated physical fields in conventional battery testing. That is, when microbubble density, coating thickness deviation and electrolyte wetting state are used as independent characterization quantities, they can only reflect local abnormal phenomena, such as simple coating thinning or poor local wetting. They cannot reveal the stress distortion of the coating caused by bubble aggregation, thereby blocking the coupling failure chain of electrolyte wetting.

[0102] Feature extraction involves mapping the microbubble density gradient to hydraulic impact force field characteristics, reconstructing abrupt changes in coating thickness as adhesive molecular chain stress characteristics, and converting the wetting state binary mask into thermodynamic gradient diffusion characteristics. This transforms fragmented data into quantitative evidence exposing cross-physical causality, such as demonstrating that a sudden decrease in coating thickness above a high-bubble-density area is inevitably accompanied by surrounding wetting degradation. The generated electrolyte-coating defect correlation map is a clear expression of this coupling mechanism. It locates overlapping areas of the triple anomalies of bubbles, coating, and wetting through spatially layered thermograms. For example, if the bubble density peak coincides with the location of maximum coating thinning and is surrounded by a low-temperature wetting band, it provides irrefutable physical logic input for subsequent diagnosis. For instance, when the map shows a region simultaneously exhibiting high hydraulic impact characteristics, high molecular chain stress characteristics, and thermal diffusion hindrance characteristics, the system directly determines it as "inaccurate injection parameters inducing a chain failure," thus precisely triggering correction of the target equipment rather than blindly adjusting equipment parameters.

[0103] In one feasible implementation, step S13 includes steps S131 to S132:

[0104] Step S131: Input the microbubble density distribution data and coating thickness deviation data into the fusion analysis module to identify the coupling relationship.

[0105] Step S132: Combine the low-temperature unwetted areas marked by the electrolyte wetting state binary mask to obtain the defect correlation map.

[0106] In this embodiment, when microbubble density distribution data and coating thickness deviation data are input into the fusion analysis module, the system automatically scans the gradient similarity of the two types of data through a spatial convolutional neural network (SCNN). When it is detected that the spatial coordinate overlap between the contour boundary of the high-density microbubble area reflected by the microbubble density distribution data and the coating thickness deviation area reflected by the coating thickness deviation data exceeds a set threshold, such as a surge in coating thinning within a 2mm radius of the bubble aggregation center, it is determined that there is a hydraulic impact coupling relationship between the two.

[0107] At this point, combining the continuous low-temperature unwetted areas marked by the binary mask of the electrolyte wetting state, the system performs three-dimensional probabilistic fusion on the triple data. If the spatial distribution of the low-temperature unwetted area simultaneously covers the peak area of ​​bubble density and the area of ​​sudden drop in coating thickness, a red warning layer is superimposed on the multi-layer probability map; if the low-temperature unwetted area only overlaps with a single factor, a yellow risk layer is superimposed; isolated anomalies are marked with blue monitoring marks, and finally a defect correlation map presented as a pseudo-color heat map is generated. Its essence is to transform the spatiotemporal coupling relationship of bubble dynamic pressure, coating mechanical stress and thermal retardation effect into a quantitative visual model that can drive root cause classification diagnosis.

[0108] Reference Figure 3 As shown, a feasible implementation method for defect correlation graph analysis is as follows, including steps S21-S22:

[0109] Step S21: After extracting the bond energy decay rate, pyrolysis rate and thermal stress coefficient based on the defect correlation map, the probability of local bonding failure is obtained based on the bond energy decay rate, the electrolyte decomposition risk value is obtained based on the pyrolysis rate, and the current collector fatigue accumulation index is obtained based on the thermal stress coefficient.

[0110] First, the regions representing abrupt changes in coating thickness in the defect correlation map are mapped to the binder molecular dynamics model. The bond energy attenuation amplitude is calculated based on the functional relationship between the abnormal deformation of the coating and the yield strength of the material. For the extraction of the pyrolysis rate, the bubble volume gradient is converted into the pyrolysis rate by using the electrolyte microbubble density distribution and solvent molecule pyrolysis activation energy model in the defect correlation map. The thermal stress coefficient is derived from the coupled calculation of the electrolyte unwetted area and the current collector expansion coefficient in the defect correlation map, and the cumulative amount of lattice distortion is derived by using the thermal conduction hysteresis characteristics.

[0111] The subsequent process of obtaining the local bond failure probability based on the bond energy decay rate is as follows: call the molecular chain breakage probability function model, input the bond energy decay rate into the Monte Carlo simulation system, and output the local bond failure probability based on random sampling of the bond energy potential well depth.

[0112] The risk value of electrolyte decomposition is calculated by using the integral reaction equation of the decomposition rate, combined with the solvent concentration half-life model to predict the irreversible decomposition threshold.

[0113] The fatigue accumulation index of the current collector is generated using the lattice slip displacement algorithm. When the thermal stress coefficient exceeds the material recrystallization critical condition, the dislocation density accumulation function is triggered to output the fatigue index.

[0114] The calculated values ​​quantify the microscopic mechanisms of material gene characteristics into decision factors that can guide process modification, providing physical and logical support for subsequent root cause classification and parameter adjustment.

[0115] Step S22: Determine the relationship between the probability of local bonding failure, the risk value of electrolyte decomposition, and the fatigue accumulation index of the current collector and the preset root cause judgment conditions to obtain the root cause classification; and determine the coordinates of the defect location based on the spatial location mapped by the probability of local bonding failure, the risk value of electrolyte decomposition, and the fatigue accumulation index of the current collector.

[0116] When classifying root causes, the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector are simultaneously input into the root cause classification decision tree model. This model incorporates the physical logic rules of material failure mechanisms. If the probability of adhesion failure exceeds the molecular chain breakage threshold, it is directly classified as adhesive failure; if the risk value of electrolyte decomposition continuously exceeds the solvent chemical bond cleavage critical line, it is classified as electrolyte decomposition; if the cumulative fatigue index of the current collector reaches the lattice slip criterion, it points to current collector fatigue; if all three exceed the standard simultaneously, the dominant root cause is output according to the failure priority. This hierarchical judgment mechanism ensures that the microscopic mechanisms of the material genome are transformed into process-operable diagnostic conclusions without damage. The preset root cause judgment conditions are the molecular chain breakage threshold, the solvent chemical bond cleavage critical line, and the lattice slip criterion.

[0117] To determine the coordinates of the defect location, the system maps the bond energy decay rate, fracture rate, and fatigue index calculated for each spatial node back to the original detection coordinate system. It then uses a probability density clustering algorithm to identify outlier clusters: extracting continuous spatial regions with a failure probability greater than 95% and calculating their geometric centers as the core coordinates. When multiple failure mechanisms are superimposed in the same region, a three-dimensional thermodynamic field diffusion simulation is automatically triggered, defining the scope of influence based on the probability decay boundary.

[0118] It provides spatial-level precise positioning for equipment parameter adjustment, solving derivative problems such as excessive parameter adjustment range and overcompensation caused by fuzzy spatial positioning in conventional solutions.

[0119] Reference Figure 4 As shown, a feasible implementation method for the material gene characteristic root cause classification tracing and equipment closed-loop control process is as follows, including steps S31~S35:

[0120] Step S31: Determine the equipment to be adjusted and the adjustment amount based on the root cause classification and defect location coordinates.

[0121] Step S32: Perform simulation adjustment operation based on the equipment to be adjusted and the adjustment amount to obtain the defect prediction recurrence rate, and determine whether the defect prediction recurrence rate is less than the preset defect prediction recurrence rate.

[0122] Step S33: If the defect prediction recurrence rate is less than the preset defect prediction recurrence rate, then the parameters of the equipment to be adjusted are adjusted based on the adjustment amount.

[0123] In this embodiment, when determining the equipment to be adjusted and the adjustment amount based on the root cause classification and defect location coordinates, the system calls a preset equipment mapping rule library. If the root cause classification is adhesive failure, the coating machine is locked as the equipment to be adjusted. Based on the spatial weight distribution of the defect location coordinates, such as the distance between the coating thinning zone and the center of the scraper, the scraper pressure compensation amount is automatically generated. When the root cause classification is electrolyte decomposition, the injection machine is selected as the equipment to be adjusted, and the adjustment amount is dynamically calculated based on the integral value of the bubble density gradient at the defect coordinates. The root cause classification of current collector fatigue triggers the roller press parameter correction, and its adjustment amount is derived in reverse from the spatial decay characteristics of the fatigue index.

[0124] Subsequently, when performing simulation adjustments based on the equipment to be adjusted and the adjustment amount, the digital twin system loads the new equipment parameters and the battery production line dynamics model, simulates 200 consecutive production cycles, and counts the defect recurrence frequency. The bonding failure recurrence rate is calculated iteratively by the molecular bond breakage probability model, while the electrolyte decomposition risk is predicted by the solvent concentration diffusion equation. Finally, the weighted polymerization outputs the defect prediction recurrence rate.

[0125] If the defect prediction recurrence rate is lower than the preset defect prediction recurrence rate, an adjustment instruction is issued to the corresponding equipment to be adjusted; if it exceeds the preset defect prediction recurrence rate, the equipment to be adjusted and the adjustment amount combination are rematched. Through repeated iterations until the defect prediction recurrence rate converges to the target, a self-verification and self-optimization error prevention mechanism for process parameters is formed.

[0126] It should be noted that the preset defect prediction recurrence rate is the upper limit of the acceptable defect prediction recurrence rate.

[0127] Step S34: Obtain the battery defect rate after adjusting the equipment parameters, and determine whether the battery defect rate is less than the preset battery defect rate.

[0128] Step S35: If the battery defect rate is less than the preset battery defect rate, generate battery manufacturing defect detection results based on the battery defect rate.

[0129] It should be noted that the preset battery defect rate is the upper limit of the acceptable battery defect rate.

[0130] When obtaining the battery defect rate after equipment parameter adjustment, the system initiates targeted monitoring on 50 batches of continuously produced batteries. Based on the defect location coordinates, it locks the same spatial area for multimodal re-inspection, such as rescanning the electrolyte layer at the coordinate point with a bubble detector and re-measuring the coating thickness with an optical system. Based on the detection results, it counts the number of recurrences of the same type of defect and calculates the battery defect rate by combining it with the total batch sample. This battery defect rate is compared with the preset battery defect rate in real time. If the battery defect rate is less than the preset battery defect rate, it is determined that the equipment parameter adjustment is effective, and the system automatically generates battery manufacturing defect detection results including the defect distribution elimination rate. If the battery defect rate is greater than the preset battery defect rate, it triggers the iterative optimization of root cause classification diagnosis and equipment parameter adjustment in reverse. Based on the new defect characteristics captured in the re-inspection, such as the appearance of new crack morphology in the coating, the root cause classification weights are updated, and the equipment to be adjusted and the adjustment amount are corrected based on the coordinate offset of the recurrence area until the battery defect rate after adjustment continuously converges to a safe range. This fundamentally avoids the blindness of conventional experience-based parameter adjustment and effectively improves the accuracy and response speed of battery detection.

[0131] This application also provides an adaptive correction system, referring to... Figure 5 As shown, the adaptive correction system includes:

[0132] The data acquisition module 10 is used to acquire multimodal data of the battery under test and output a defect correlation map based on the multimodal data. The defect correlation map is associated with the electrolyte and coating of the battery under test.

[0133] Data calculation module 20 is used to extract material gene features from defect association maps through AI visual detection models, perform transfer calculations on material gene features, and obtain root cause classification and defect location coordinates.

[0134] The data detection module 30 is used to perform equipment parameter adjustment operations based on root cause classification and defect location coordinates, and obtain battery manufacturing defect detection results after equipment parameter adjustment;

[0135] The model optimization module 40 is used to input the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization, so as to obtain the optimized AI visual inspection model.

[0136] Optionally, the data acquisition module 10 is also used for:

[0137] A tunable pulse is emitted to the battery under test. The tunable pulse penetrates the electrolyte layer of the battery under test through the battery surface. The molecular vibration decay signal in the electrolyte layer is captured by a quantum dot array to obtain a microbubble distribution map. The diffraction pattern on the surface of the battery under test is acquired to generate an optical diffraction image. The surface temperature of the battery under test is sampled, and a thermal field distribution video stream is generated based on the surface temperature distribution.

[0138] The multimodal data includes microbubble distribution maps, optical diffraction images, and thermal field distribution video streams.

[0139] Optionally, the data acquisition module 10 is also used for:

[0140] Spatial registration and temporal synchronization are performed on the multimodal data to obtain a multimodal dataset;

[0141] Feature extraction and reconstruction were performed on the multimodal dataset to obtain a defect association map.

[0142] Optionally, the data acquisition module 10 is also used for:

[0143] A unified spatial coordinate system was established based on the chessboard calibration method;

[0144] The microbubble distribution map, optical diffraction image, and thermal field distribution video stream are aligned to a unified spatial coordinate system using an affine transformation matrix to obtain the registration matrix.

[0145] After triggering the temporal synchronization of the microbubble distribution map, optical diffraction image, and thermal field distribution video stream, a multimodal dataset is generated by combining the registration matrix. The multimodal dataset includes microbubble density distribution data, coating thickness deviation data, and electrolyte wetting state binary mask.

[0146] Optionally, the data acquisition module 10 is also used for:

[0147] Input the microbubble density distribution data and coating thickness deviation data into the fusion analysis module to identify the coupling relationship;

[0148] By combining the low-temperature unwetted areas marked by the electrolyte wetting state binary mask, a defect correlation map is obtained.

[0149] Optionally, the data calculation module 20 is also used for:

[0150] After extracting the bond energy decay rate, pyrolysis rate and thermal stress coefficient based on the defect correlation map, the probability of local bonding failure is obtained based on the bond energy decay rate, the electrolyte decomposition risk value is obtained based on the pyrolysis rate, and the current collector fatigue accumulation index is obtained based on the thermal stress coefficient.

[0151] The relationship between the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector and the preset root cause judgment conditions is determined to obtain the root cause classification; and the coordinates of the defect location are determined based on the spatial location mapped by the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector.

[0152] Optionally, the data detection module 30 is also used for:

[0153] Based on the root cause classification and defect location coordinates, determine the equipment to be adjusted and the adjustment amount;

[0154] Based on the equipment to be adjusted and the adjustment amount, a simulation adjustment operation is performed to obtain the defect prediction recurrence rate, and it is determined whether the defect prediction recurrence rate is less than the preset defect prediction recurrence rate.

[0155] If the defect prediction recurrence rate is less than the preset defect prediction recurrence rate, then the parameters of the equipment to be adjusted are adjusted based on the adjustment amount.

[0156] If the predicted defect recurrence rate is greater than the preset predicted defect recurrence rate, then the step of determining the equipment to be adjusted and the adjustment amount based on the root cause classification and defect location coordinates is executed.

[0157] Optionally, the data detection module 30 is also used for:

[0158] Obtain the battery defect rate after adjusting the equipment parameters, and determine whether the battery defect rate is less than the preset battery defect rate;

[0159] If the battery defect rate is less than the preset battery defect rate, the battery test results are generated based on the battery defect rate.

[0160] Optionally, the model optimization module 40 is also used for:

[0161] The battery manufacturing defect detection results are used to generate a cryptographic gradient tensor through local training. The cryptographic gradient tensor is then stored on the blockchain and federated to obtain the aggregated global gradient.

[0162] The AI ​​visual detection model is optimized by aggregating the global gradient, resulting in an optimized AI visual detection model.

[0163] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An AI visual inspection method for battery manufacturing defects, characterized in that, include: Acquire multimodal data of the battery under test, and output a defect correlation map based on the multimodal data, wherein the defect correlation map is associated with the electrolyte and coating of the battery under test; The AI ​​visual detection model extracts material gene features from the defect association map, performs transfer calculations on the material gene features, and obtains root cause classification and defect location coordinates. The AI ​​visual detection model is a visual detection system for battery manufacturing defects built on a deep neural network. It learns the inherent laws of a large amount of historical battery defect data through pre-training, and realizes the ability to analyze multi-physics field defect association maps. Based on the root cause classification and the defect location coordinates, perform equipment parameter adjustment operations to obtain the battery manufacturing defect detection results after equipment parameter adjustment; The battery manufacturing defect detection results are input into the AI ​​visual detection model for optimization, resulting in an optimized AI visual detection model. The steps for obtaining the multimodal data of the battery under test include: A tunable pulse is emitted to the battery under test. The tunable pulse penetrates the electrolyte layer of the battery under test through the battery surface. The molecular vibrational attenuation signal in the electrolyte layer is captured by a quantum dot array to obtain a microbubble distribution map. The diffraction pattern on the battery surface of the battery under test is acquired to generate an optical diffraction image. The surface temperature of the battery under test is sampled, and a thermal field distribution video stream is generated based on the surface temperature distribution. The multimodal data includes the microbubble distribution map, the optical diffraction image, and the thermal field distribution video stream; The steps of extracting material gene features from the defect association map and performing transfer calculations on the material gene features to obtain root cause classification and defect location coordinates include: After extracting the bond energy decay rate, pyrolysis rate and thermal stress coefficient based on the defect correlation map, the probability of local bonding failure is obtained based on the bond energy decay rate, the electrolyte decomposition risk value is obtained based on the pyrolysis rate, and the current collector fatigue accumulation index is obtained based on the thermal stress coefficient. The relationship between the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the fatigue accumulation index of the current collector and the preset root cause judgment conditions is determined to obtain the root cause classification; and the coordinates of the defect location are determined according to the spatial location mapped by the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the fatigue accumulation index of the current collector. The step of performing equipment parameter adjustment based on the root cause classification and the defect location coordinates includes: Based on the root cause classification and the coordinates of the defect location, determine the equipment to be adjusted and the adjustment amount; Based on the device to be adjusted and the adjustment amount, a simulation adjustment operation is performed to obtain the defect prediction recurrence rate, and it is determined whether the defect prediction recurrence rate is less than the preset defect prediction recurrence rate. If the predicted defect recurrence rate is less than the preset predicted defect recurrence rate, then the parameters of the device to be adjusted are adjusted based on the adjustment amount. If the predicted defect recurrence rate is greater than the preset predicted defect recurrence rate, then the step of determining the equipment to be adjusted and the adjustment amount based on the root cause classification and the defect location coordinates is executed.

2. The AI ​​visual detection method for battery manufacturing defects according to claim 1, characterized in that, The step of outputting a defect correlation map based on the multimodal data includes: Spatial registration and temporal synchronization are performed on the multimodal data to obtain a multimodal dataset; Feature extraction and reconstruction are performed on the multimodal dataset to obtain a defect association map.

3. The AI ​​visual detection method for battery manufacturing defects according to claim 2, characterized in that, The steps of spatial registration and temporal synchronization of the multimodal data to obtain a multimodal dataset include: A unified spatial coordinate system was established based on the chessboard calibration method; The microbubble distribution map, the optical diffraction image, and the thermal field distribution video stream are aligned to the unified spatial coordinate system using an affine transformation matrix to obtain a registration matrix. After triggering the time synchronization of the microbubble distribution map, the optical diffraction image, and the thermal field distribution video stream, the multimodal dataset is generated by combining the registration matrix. The multimodal dataset includes microbubble density distribution data, coating thickness deviation data, and electrolyte wetting state binary mask.

4. The AI ​​visual detection method for battery manufacturing defects according to claim 3, characterized in that, The steps of extracting and reconstructing features from the multimodal dataset to obtain the defect association map include: The microbubble density distribution data and the coating thickness deviation data are input into the fusion analysis module to identify the coupling relationship; The defect correlation map is obtained by combining the low-temperature unwetted areas marked by the electrolyte wetting state binary mask.

5. The AI ​​visual detection method for battery manufacturing defects according to claim 4, characterized in that, The steps for obtaining the battery test results after adjusting the device parameters include: Obtain the battery defect rate after adjusting the equipment parameters, and determine whether the battery defect rate is less than the preset battery defect rate; If the battery defect rate is less than the preset battery defect rate, the battery detection result is generated based on the battery defect rate.

6. The AI ​​visual detection method for battery manufacturing defects according to claim 5, characterized in that, The step of inputting the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization to obtain the optimized AI visual inspection model includes: The battery manufacturing defect detection results are used to generate an encrypted gradient tensor through local training. The encrypted gradient tensor is then stored on the blockchain and federated to obtain the aggregated global gradient. The AI ​​visual detection model is optimized by aggregating the global gradients to obtain the optimized AI visual detection model.

7. An adaptive correction system, characterized in that, The adaptive correction system is applied to the AI ​​visual detection method for battery manufacturing defects as described in any one of claims 1 to 6, comprising: The data acquisition module is used to acquire multimodal data of the battery under test and output a defect correlation map based on the multimodal data, wherein the defect correlation map is associated with the electrolyte and coating of the battery under test. The data calculation module is used to extract material gene features from the defect association map through the AI ​​visual detection model, perform transfer calculation on the material gene features, and obtain root cause classification and defect location coordinates. The AI ​​visual detection model is a visual detection system for battery manufacturing defects built on a deep neural network. It learns the inherent laws of a large amount of historical battery defect data through pre-training to achieve the ability to analyze multi-physics field defect association maps. The data detection module is used to perform equipment parameter adjustment operations based on the root cause classification and the defect location coordinates, and obtain the battery manufacturing defect detection results after the equipment parameter adjustment. The model optimization module is used to input the battery manufacturing defect detection results into the AI ​​visual inspection model for optimization, so as to obtain the optimized AI visual inspection model. The data acquisition module is also used to emit tunable pulses to the battery under test. The tunable pulses penetrate the electrolyte layer of the battery under test through the battery surface and capture the molecular vibration decay signal in the electrolyte layer through the quantum dot array to obtain a microbubble distribution map; and to acquire the diffraction pattern on the battery surface of the battery under test to generate an optical diffraction image; and to sample the surface temperature of the battery under test and generate a thermal field distribution video stream based on the surface temperature distribution. The multimodal data includes microbubble distribution maps, optical diffraction images, and thermal field distribution video streams. The data calculation module is also used to extract bond energy decay rate, pyrolysis rate and thermal stress coefficient based on defect correlation map, and then obtain the probability of local bonding failure based on bond energy decay rate, the risk value of electrolyte decomposition based on pyrolysis rate, and the fatigue accumulation index of current collector based on thermal stress coefficient. The relationship between the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector and the preset root cause judgment conditions is determined to obtain the root cause classification; and the coordinates of the defect location are determined based on the spatial location mapped by the probability of localized adhesion failure, the risk value of electrolyte decomposition, and the cumulative fatigue index of the current collector. The data detection module is also used to determine the equipment to be adjusted and the adjustment amount based on the root cause classification and defect location coordinates; Based on the equipment to be adjusted and the adjustment amount, a simulation adjustment operation is performed to obtain the defect prediction recurrence rate, and it is determined whether the defect prediction recurrence rate is less than the preset defect prediction recurrence rate. If the defect prediction recurrence rate is less than the preset defect prediction recurrence rate, then the parameters of the equipment to be adjusted are adjusted based on the adjustment amount. If the predicted defect recurrence rate is greater than the preset predicted defect recurrence rate, then the step of determining the equipment to be adjusted and the adjustment amount based on the root cause classification and defect location coordinates is executed.

Images