A method for evaluating the wetting uniformity of liquid electrolyte batteries

Abstract

This invention relates to the field of battery quality assessment technology, specifically disclosing a method for assessing the uniformity of wetting in liquid electrolyte batteries. The method includes: acquiring and processing a distribution map of the wetting degree of the target battery; extracting global statistics, local spatial heterogeneity, defect quantification, and topological features from the processed image and performing feature processing; dividing the processed features into optimizable factor groups and structural constraint factor groups, determining the basic total weight of each group and the basic weight of each feature, obtaining the final weight of each feature through sample adaptive fine-tuning and intra-group weight normalization, and calculating a basic uniformity score; establishing a set of risk features for normal wetting morphology and a set of risk features for poor wetting morphology, and calculating a comprehensive morphological health score; finally, generating a comprehensive grade based on the basic uniformity score and the comprehensive morphological health score. This invention improves assessment accuracy, can distinguish between normal and poor wetting morphology, and guides process optimization or design improvement.

Description

Technical Field

[0001] This invention relates to the field of liquid electrolyte battery quality assessment technology, and in particular to a method for assessing the uniformity of wetting in liquid electrolyte batteries. Background Technology

[0002] Liquid electrolyte batteries are chemical power sources that use a liquid substance as an ion conductor (i.e., electrolyte) and rely on the directional migration of ions in the liquid to achieve charge transfer between the positive and negative electrodes during charging and discharging. Liquid lithium-ion batteries are currently the mainstream lithium battery application, using electrolyte as the medium for lithium-ion transfer between the positive and negative electrodes. As a core energy storage device in the new energy field, the performance of liquid lithium-ion batteries is closely related to the effectiveness of electrolyte wetting. Whether the electrolyte can uniformly and fully wet the electrode materials and separator inside the battery directly determines the battery's internal resistance consistency, cycle life, and safety performance.

[0003] In the current production process of liquid electrolyte batteries, the assessment of electrolyte wetting uniformity mainly relies on two traditional methods: The first is destructive sampling inspection, which involves randomly selecting finished batteries for disassembly and judging the wetting condition by observing wetting marks on the electrodes and separators and detecting local electrolyte content. Although this method can provide intuitive results, it can only cover the sampled products and cannot achieve full batch inspection. Moreover, the disassembly process directly damages the batteries, resulting in resource waste. At the same time, disassembly inspection is a post-production inspection, and even if wetting defects are found, it is impossible to trace and adjust the completed production process, making it difficult to achieve real-time quality control. The second method is non-destructive X-ray inspection, which involves taking batteries that have been filled with electrolyte and left to stand for a certain period of time, performing X-ray scanning to obtain a two-dimensional grayscale image of the battery's interior, extracting the area ratio of defective areas, grayscale mean, and standard deviation from the two-dimensional grayscale image, comparing the feature values ​​with preset feature thresholds, and determining whether the battery wetting is uniform based on the comparison results. This method does not damage the battery, allowing it to continue to be used after testing. It visualizes the battery's internal structure using a two-dimensional grayscale image, offering some quantitative capability. Feature extraction is relatively simple, computationally efficient, and easy to implement in engineering. However, this method extracts only a single feature, lacks sufficient information dimensionality, and cannot describe the spatial heterogeneity and topological morphology of the wetting process. Simple threshold comparison leads to coarse judgment results; it cannot distinguish the specific types of wetting uniformity problems and cannot guide the direction of process improvement.

[0004] In summary, existing assessment methods suffer from insufficient information dimensions, coarse judgment of wetting uniformity, inability to distinguish between wetting morphology risk types, and inability to differentiate between optimizable and structural limiting factors. These shortcomings make it difficult to meet the demand for precise control of wetting quality in the large-scale production of liquid electrolyte batteries. There is an urgent need for an efficient, accurate, and full-process coverage method for assessing wetting uniformity. Summary of the Invention

[0005] This invention addresses the problems in the prior art, such as insufficient information dimensions, coarse judgment of wetting uniformity, inability to distinguish wetting morphology risk types, and inability to distinguish between optimizable and structural limiting factors, by proposing a method for evaluating the wetting uniformity of liquid electrolyte batteries.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for evaluating the wetting uniformity of a liquid electrolyte battery, characterized by comprising the following steps:

[0007] Obtain a wetting degree distribution map of the target battery, and process the wetting degree distribution map to obtain a processed image;

[0008] Features are extracted from the processed image, and the extracted features are processed to obtain the processed features. The features are selected from at least one of global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features.

[0009] The processed features are divided into optimizable factor groups and structural constraint factor groups. The basic total weight of each group and the basic weight of each feature are determined. The basic weight of each feature is subjected to intra-group sample adaptive fine-tuning and intra-group weight normalization to obtain the final weight of each feature within the group. Based on the processed feature values ​​and their final weights, the uniformity basic score is calculated.

[0010] Based on expert experience, morphological risk features are selected from the processed features and divided into the normal infiltration morphological risk feature set and the adverse infiltration morphological risk feature set. The basic weights of each morphological risk feature are normalized within the set to obtain the normalized weights of each feature within the set. Based on the processed feature values ​​and their normalized weights of each feature within the set, the comprehensive morphological health score is calculated.

[0011] A comprehensive rating is generated based on the uniformity baseline score and the overall morphological health score.

[0012] Optionally, the infiltration degree distribution map is processed, including the following steps:

[0013] Adaptive ROI extraction is performed on the wetting degree distribution state map of the target battery to obtain the effective region of the wetting degree distribution state map;

[0014] Denoising is performed on the effective area of ​​the infiltration degree distribution state map to obtain the processed infiltration degree distribution state map.

[0015] Optionally, the infiltration degree distribution map is processed, including the following steps:

[0016] Obtain a pre-constructed baseline map of the distribution of infiltration levels;

[0017] After extracting the ROI from the wettability distribution state map of the target battery according to preset parameters, the difference calculation is performed between the ROI and the baseline wettability distribution state map to obtain the wettability distribution state difference map of the target battery.

[0018] Adaptive ROI extraction is performed on the infiltration degree distribution state difference map to obtain the effective region of the infiltration degree distribution state difference map;

[0019] Denoising is performed on the effective area of ​​the infiltration degree distribution difference map to obtain the processed infiltration degree distribution difference map.

[0020] Optionally, the construction of the infiltration degree distribution baseline map includes the following steps:

[0021] For the infiltration degree distribution state map of several candidate reference batteries, ROI extraction is performed according to uniform preset parameters to obtain several candidate reference battery ROI images, wherein the ROI extraction parameters of the candidate reference batteries and the target battery are consistent.

[0022] Denoising is performed on the ROI images of several candidate reference batteries using uniform parameters to suppress acquisition noise, resulting in ROI images of several candidate reference batteries after processing.

[0023] The same features are extracted from the ROI images of each candidate reference battery after processing, and the features are normalized to obtain normalized feature values. The features extracted from the candidate reference battery and the target battery are consistent.

[0024] Based on the normalized feature values, the comprehensive similarity between every two candidate reference cells is calculated, and the average similarity of each candidate reference cell is calculated based on the comprehensive similarity.

[0025] The top K candidate reference batteries are selected as baseline batteries based on the average similarity from high to low, where K is a preset number;

[0026] The average pixel value at the same pixel position is calculated for the ROI image after processing the K reference batteries to generate the infiltration degree distribution state reference map.

[0027] Optionally, the global infiltration statistical features include at least one of the following: discrete uniformity features, extreme value span difference features, symmetry offset features, steepness and cusp features, interval proportion features, and global distribution pattern features.

[0028] The local spatial heterogeneous infiltration features include at least one of the following: sub-block local mean discrete features, region infiltration gradient difference features, and window texture spatial heterogeneity features;

[0029] The infiltration defect quantification features include at least one of the following: anomaly threshold statistical features, defect connected domain morphological scale features, defect spatial distribution clustering features, and defect boundary transition gradient features;

[0030] The topological morphological features include at least one of the following: topological connected domain morphological features, skeleton extension structure features, and topological clustering orientation features.

[0031] Optionally, the step of performing feature processing on each extracted feature to obtain the processed feature values ​​includes the following steps:

[0032] Based on the correlation between the extracted feature values ​​and the electrolyte wetting uniformity, the features are classified into one of the following: negative features, positive features, and bidirectional features. Among them, the larger the feature value of the negative feature, the worse the wetting uniformity; the larger the feature value of the positive feature, the better the wetting uniformity; and the larger the feature value of the bidirectional feature deviates from the target value, the worse the wetting uniformity.

[0033] For negative features, retain the original values; for positive features, convert them to negative indices through linear inversion; for bidirectional features, convert them to negative indices based on the degree of deviation.

[0034] Hyperbolic tangent normalization is performed on the negative features and the transformed negative indices, mapping them to the [0,1] interval to obtain the hyperbolic tangent normalized eigenvalues ​​of each feature.

[0035] Optionally, the step of dividing the processed features into a group of optimizable factors and a group of structural limiting factors includes:

[0036] Features that show a response relationship between characteristic changes and process parameter adjustments are grouped into optimizable factor groups;

[0037] Features whose characteristics are difficult to fundamentally change within the scope of process optimization and require design improvement are classified into the structural constraint factor group;

[0038] The determination of the basic total weight of each group and the basic weight of each feature includes:

[0039] The basic total weight of each group and the basic weight of each feature are determined by a weight determination method, wherein the weight determination method includes at least one of the analytic hierarchy process, the Delphi method or the entropy weight method;

[0040] And it satisfies the following: the sum of the basic weights of each feature in each group is equal to the total basic weight of that group, the sum of the total basic weights of all groups is 1, and the sum of the basic weights of all features is 1.

[0041] Optionally, the process of performing intra-group sample adaptive fine-tuning and intra-group weight normalization on the basic weights of each feature includes:

[0042] Within the same factor group, feature adjustment coefficients are calculated based on the preset fine-tuning intensity coefficients of each feature and the correlation coefficients with the long-term performance indicators of the battery; the basic weights of each feature within the group are fine-tuned using the feature adjustment coefficients to obtain the fine-tuning weights of each feature within the group.

[0043] For each factor group, calculate the sum of the fine-tuned weights of each feature within the group as the total fine-tuned weights of the group; calculate the scaling factor for the group, which is equal to the total base weights of the group divided by the total fine-tuned weights of the group; multiply the fine-tuned weights of each feature within the group by the scaling factor to obtain the final weights of each feature within the group.

[0044] The calculation of the uniformity baseline score based on the feature values ​​and their final weights after feature processing includes:

[0045] Calculate the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each feature and their corresponding final weights, and then subtract the sum of the products from 1 to obtain the uniformity base score.

[0046] Optionally, the set-based weight normalization process for the basic weights of each risk characteristic includes:

[0047] Calculate the normalized weight of each feature in the risk feature set of normal infiltration morphology and the risk feature set of adverse infiltration morphology respectively. The normalized weight of each morphological risk feature is equal to the basic weight of that feature divided by the sum of the basic weights of all morphological risk features in the set.

[0048] The calculation of the comprehensive physical health score includes:

[0049] Calculate the risk score for normal morphology and the risk score for adverse morphology. The risk score for normal morphology is the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each morphological risk feature in the set of normal infiltration morphological risk features and their corresponding normalized weights. The risk score for adverse morphology is the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each morphological risk feature in the set of adverse infiltration morphological risk features and their corresponding normalized weights.

[0050] Calculate the overall morphological health score, where the overall morphological health score is: ,in, , These are the preset weights for the risk scores of normal and adverse states, respectively. 1.

[0051] Optionally, the generation of the comprehensive grade based on the uniformity baseline score and the comprehensive morphological health score includes:

[0052] The uniformity base score and the comprehensive morphological health score are respectively cropped to the [0,1] interval to obtain the uniformity score and the morphological health score;

[0053] The overall rating will be determined in the following order:

[0054] If the uniformity score is less than the third threshold for uniformity or the morphological health score is less than the third threshold for morphological health, it is deemed unqualified.

[0055] Otherwise, if the uniformity score is greater than or equal to the first threshold of uniformity and the morphological health score is greater than or equal to the first threshold of morphological health, it is judged as excellent.

[0056] Otherwise, if the uniformity score is greater than or equal to the second threshold for uniformity and the morphological health score is greater than or equal to the second threshold for morphological health, then it is considered qualified.

[0057] Otherwise, it is determined that optimization is needed;

[0058] Among them, the first threshold for uniformity > the second threshold for uniformity > the third threshold for uniformity, and the first threshold for morphological health > the second threshold for morphological health > the third threshold for morphological health.

[0059] Compared with the prior art, the present invention includes at least one of the following beneficial technical effects:

[0060] First, to address the problem of insufficient information dimensions in existing technologies, this invention extracts features including at least one of global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features. Each category contains multiple sub-features, forming a multi-dimensional feature system that covers multiple perspectives such as discrete uniformity, extreme value span, symmetry offset, steepness, gradient difference, texture heterogeneity, connected domain morphology, skeleton extension, and aggregation orientation. This provides a comprehensive analysis of whether the battery electrolyte wetting is uniform, which helps improve the evaluation accuracy.

[0061] Second, to address the problem of coarse determination of immersion uniformity in existing technologies, this invention divides the features into an optimizable group and a structurally constrained group. Based on the correlation between the features and the long-term performance of the battery, the basic weights are adaptively fine-tuned and normalized within the group, so that the weights can better reflect the actual immersion state and improve the accuracy of the determination.

[0062] Third, to address the problem that existing technologies cannot distinguish between different types of infiltration morphology risks, this invention constructs a set of risk characteristics for normal infiltration morphology and a set of risk characteristics for adverse infiltration morphology, calculates risk scores for normal infiltration morphology and adverse infiltration morphology respectively, and then generates a comprehensive morphological health score. This score is combined with the uniformity baseline score to determine the comprehensive level, which can distinguish between normal and adverse infiltration morphology, facilitating targeted improvements.

[0063] Fourth, to address the problem that existing technologies cannot distinguish between optimizable and structural limiting factors, this invention categorizes features whose characteristic changes are responsive to process parameter adjustments into the optimizable factor group, and categorizes features whose characteristic changes are difficult to fundamentally change within the process optimization range and require design improvements into the structural limiting factor group. After grouping, the weight of each feature is calculated separately, which can prioritize optimizable factors when guiding subsequent process improvements, avoid making ineffective adjustments to structural limiting factors, and improve improvement efficiency. Attached Figure Description

[0064] Figure 1 A flowchart illustrating the steps of a method for evaluating the wetting uniformity of a liquid electrolyte battery;

[0065] Figure 2 The wettability distribution diagram of a liquid lithium-ion battery provided in Example 1;

[0066] Figure 3 This is a wettability distribution diagram of a liquid lithium-ion battery after treatment, as provided in Example 1.

[0067] Figure 4 This is a wettability distribution diagram of a liquid lithium-ion battery provided in Example 2;

[0068] Figure 5 This is a reference diagram showing the wetting degree distribution of a liquid lithium-ion battery provided in Example 2;

[0069] Figure 6 This is a difference diagram of the wetting degree distribution after treatment of a liquid lithium-ion battery, provided in Example 2.

[0070] Figure 7 This is a module connection diagram for a liquid electrolyte battery wetting uniformity evaluation system. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] like Figure 1 As shown, this embodiment provides a method for evaluating the wetting uniformity of a liquid electrolyte battery, including the following steps:

[0073] Obtain a wetting degree distribution map of the target battery, and process the wetting degree distribution map to obtain a processed image;

[0074] Features are extracted from the processed image, and the extracted features are processed to obtain the processed features. The features are selected from at least one of global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features.

[0075] The processed features are divided into optimizable factor groups and structural constraint factor groups. The basic total weight of each group and the basic weight of each feature are determined. The basic weight of each feature is subjected to intra-group sample adaptive fine-tuning and intra-group weight normalization to obtain the final weight of each feature within the group. Based on the processed feature values ​​and their final weights, the uniformity basic score is calculated.

[0076] Based on expert experience, morphological risk features are selected from the processed features and divided into the normal infiltration morphological risk feature set and the adverse infiltration morphological risk feature set. The basic weights of each morphological risk feature are normalized within the set to obtain the normalized weights of each feature within the set. Based on the processed feature values ​​and their normalized weights of each feature within the set, the comprehensive morphological health score is calculated.

[0077] A comprehensive rating is generated based on the uniformity baseline score and the overall morphological health score.

[0078] The target battery of this application, i.e., the battery to be tested, refers to a liquid electrolyte battery used for electrolyte distribution detection. Specifically, it can be a lithium-ion battery using an organic liquid electrolyte, or other types of liquid electrolyte batteries, which will not be listed here.

[0079] In this embodiment, it is specifically necessary to explain that the generation of the wetting degree distribution map includes the following steps:

[0080] The target battery is subjected to ultrasonic scanning to obtain a set of ultrasonic transmission signals to be tested corresponding to the ROI region. The ROI region is the desired electrolyte wetting area of ​​the battery under test. The set of ultrasonic transmission signals to be tested includes several ultrasonic transmission signals to be tested, and each ultrasonic transmission signal to be tested corresponds to a scanning detection point in the ROI region.

[0081] Extract the corresponding transmission signal feature value for each ultrasound transmission signal to be tested, and use the pre-constructed feature-electrolyte content model to generate the electrolyte content solution value corresponding to the transmission signal feature value;

[0082] Extract the maximum value of all electrolyte solutions, and normalize all electrolyte solutions based on the maximum value to generate the corresponding wetting normalization value for each electrolyte solution.

[0083] Based on all the normalized infiltration values, a state map of the infiltration degree distribution is generated.

[0084] It should be understood that when performing ultrasonic testing on the electrolyte distribution within a target battery, the target battery should first be ultrasonically scanned. Specifically, this involves a transmission-type ultrasonic scan of the target battery. The method and process of performing this transmission-type ultrasonic scan can be consistent with existing technologies and will not be elaborated upon here. Furthermore, after the electrolyte is injected into the target battery, it will distribute within its predetermined desired wetting area. To improve detection accuracy, the ultrasonic scan information can be imaged first, and then the ROI can be determined based on the desired wetting area. The ROI is typically consistent with or slightly larger than the desired wetting area of ​​the electrolyte. The desired wetting area is generally not lower than the battery body area.

[0085] To determine the electrolyte distribution within the battery under test, the transmission signal characteristics of each ultrasonic transmission signal need to be extracted. Since different wetting states at the scanning detection point lead to varying ultrasonic wave transmission intensities, the transmission signal characteristics should at least reflect intensity information, thus indicating the electrolyte content at that point. Alternatively, features reflecting the time dimension can also be used. Specifically, intensity characteristics include, but are not limited to, at least one of PPV value, amplitude, attenuation coefficient, and envelope area; time dimension characteristics include, but are not limited to, TOF value. The transmission signal characteristics can be selected according to the testing requirements and will not be listed individually here. The specific calculation methods for each transmission signal characteristic are consistent with existing technologies and will not be elaborated further.

[0086] After extracting the transmission signal features of each ultrasound transmission signal to be inspected, a pre-constructed feature-electrolyte content model is used to calculate the electrolyte content value corresponding to each feature value. The feature-electrolyte content model needs to be pre-constructed before detection. Since the transmission signal features are related to the electrolyte content at the corresponding scanning detection point, the electrolyte content value at each point can be obtained through the feature-electrolyte content model.

[0087] Based on the calculated electrolyte content at all scanning detection points, the electrolyte distribution within the ROI region of the target battery can be determined, thus characterizing the electrolyte wetting state of that region. The calculated electrolyte content reflects the electrolyte content at each point, allowing analysis of the electrolyte distribution at different locations within the ROI region, achieving effective detection of the electrolyte distribution within the battery under test. This method acquires transmission signal sets through ultrasonic scanning and combines them with a model to calculate the electrolyte content, offering advantages such as good real-time performance, high efficiency, and high accuracy.

[0088] In this embodiment, it should be specifically noted that the feature-electrolyte content model is constructed through data fitting, which specifically includes creating a fitting dataset, solving the battery dataset for each reference battery, and fitting the model.

[0089] Creating the fitted dataset includes:

[0090] Provide several reference cells with known electrolyte injection amounts but not entirely identical electrolyte wetting distribution states;

[0091] Each reference cell was subjected to ultrasonic scanning, and the reference ultrasonic transmission signal group corresponding to the ROI region within the reference cell was obtained.

[0092] Based on the reference ultrasonic transmission signal sets of all reference cells and the electrolyte injection amount of each reference cell, the battery dataset for each reference cell is calculated, and a fitting dataset is formed based on the battery datasets of all reference cells.

[0093] Each reference battery's battery dataset includes several fitted data samples. Each fitted data sample includes a reference electrolyte content per unit area and a transmission fitted feature value corresponding to the reference electrolyte content per unit area. The type of the transmission fitted feature value is consistent with the type of the transmission signal feature value.

[0094] For each reference battery's battery dataset, the number of fitted data samples within the battery dataset is consistent, and the fitted data samples within all reference datasets have a consistent distribution.

[0095] As explained above, multiple reference batteries are provided when creating the fitting dataset. The total mass of electrolyte injected into each reference battery is known, and the electrolyte wetting distribution of each reference battery is not exactly the same. For each reference battery, the same ultrasonic scanning method as the battery under test is used to obtain the reference ultrasonic transmission signal set corresponding to its ROI region (which is consistent with the ROI of the battery under test). Based on the reference ultrasonic transmission signal set and the electrolyte injection volume of each reference battery, a corresponding battery dataset is generated. The datasets for all reference batteries are created in the same way, and the number of fitting data samples in each dataset is consistent.

[0096] Solving the battery dataset for each reference battery includes:

[0097] Based on the corresponding transmission reference feature value of each reference ultrasound transmission signal in the reference ultrasound transmission signal group, a reference feature distribution histogram of each reference cell is constructed to characterize the distribution state of the transmission reference feature value of the corresponding reference cell. The type of transmission reference feature value is consistent with the type of transmission signal feature value.

[0098] For each reference cell, the histogram of reference feature distribution is divided into reference feature intervals, and the area of ​​each reference feature sub-interval is calculated.

[0099] When dividing the reference feature intervals, the number of reference feature sub-intervals obtained by dividing each reference feature distribution histogram is the same, the reference feature interval ranges of the corresponding reference feature sub-intervals are consistent, and the number of reference feature sub-intervals obtained is no more than the number of reference batteries.

[0100] For each reference cell, based on the area of ​​each reference feature sub-interval and the electrolyte injection amount of the reference cell, an electrolyte content distribution state equation is constructed to characterize the equivalent relationship between sub-interval area - electrolyte content per unit area - electrolyte injection amount.

[0101] Based on the state equation of electrolyte content distribution of all reference cells, the electrolyte content per unit area of ​​each reference characteristic sub-interval is calculated.

[0102] For each reference cell, the transmission fitting feature value of each reference feature sub-interval is determined, and a fitting data sample of the reference cell is formed based on the determined transmission fitting feature value and the corresponding electrolyte content per unit area.

[0103] It should be understood that when there are multiple reference cells, a battery dataset for each reference cell should be calculated and generated separately. When generating the battery dataset for each reference cell, a reference feature distribution histogram of the reference cell should be constructed. The horizontal axis of the reference feature distribution histogram corresponds to the transmission reference feature value, and the vertical axis of the reference feature distribution histogram represents the number of each transmission reference feature value.

[0104] It should be noted that for each reference cell, based on the transmission reference characteristic values ​​of all its reference ultrasonic transmission signals, the frequency of each characteristic value is counted, and a reference characteristic distribution histogram of the reference cell is constructed with the characteristic value on the x-axis and the frequency on the y-axis.

[0105] After constructing the histogram of the reference feature distribution for each reference cell, the reference feature interval should be divided. Specifically, this includes: determining the distribution range of the transmission reference feature values ​​based on the transmission reference feature values ​​of all reference cells, and configuring the reference feature interval division parameters based on the determined distribution range. The reference feature interval division parameters include at least the number of reference feature sub-intervals and the corresponding interval range of each reference feature sub-interval.

[0106] As explained above, the total distribution range of transmission reference characteristic values ​​for all reference cells is determined as a unified abscissa interval. This interval is then divided into multiple reference characteristic sub-intervals, with all reference cells having the same number of sub-intervals, consistent interval ranges, and the number of sub-intervals not exceeding the number of reference cells. The division can be uniform or other methods. For each reference cell, the electrode area corresponding to the scanning detection point within each sub-interval is calculated to obtain the interval area. (Unit: mm) 2 Based on the known total electrolyte injection volume of this reference battery. (Unit: g), establish the equation: Where n is the number of subintervals. Electrolyte content per unit area corresponding to the i-th sub-interval (unit: g / mm²) 2 By solving the equations of all reference cells simultaneously, the value of each subinterval can be calculated. The specific solution method can be found using existing techniques (such as the least squares method), which will not be elaborated here.

[0107] The fitting model, specifically, uses methods such as least squares to fit the dataset, obtaining the parameters of the feature-electrolyte content model. To improve the detection accuracy of electrolyte distribution, the feature-electrolyte content model can employ a double-weighted logistic composite model. Specifically, when using a double-weighted logistic composite model, for the feature-electrolyte content model constructed through data fitting, we have: ,in, These are characteristic values ​​of the transmitted signal. Electrolyte content per unit area , These are the weighting coefficients. , , , , , These are the parameters determined through data fitting. It should be noted that the fitted dataset is used to fit the double-weighted Logistic composite model to determine the values ​​of parameters A, B, C, D, E, and F. Once the parameters are determined, the feature-electrolyte content model is constructed. The weighting coefficients α and β can be set empirically.

[0108] It should be noted that since ultrasonic scanning cannot completely cover all points within the ROI area, to improve detection accuracy, interpolation (such as linear interpolation) can be performed on the ultrasonic transmission signals of two adjacent scanning points to generate interpolated transmission signals. The transmission signal feature values ​​of the interpolated signals are extracted, and the electrolyte content is calculated using a feature-electrolyte content model. Finally, the electrolyte content of the original scanning points and the interpolated points is combined to jointly determine the electrolyte distribution in the ROI area.

[0109] To eliminate absolute differences in electrolyte content between different batteries and scanning batches, and to visually demonstrate the relative uniformity of electrolyte distribution, the calculated electrolyte content values ​​at all scanning points within the ROI area were normalized. Specifically, this included the following steps:

[0110] Determine the maximum value of all electrolyte solutions and normalize all electrolyte solutions using the maximum value to generate the corresponding wetting normalization value for each electrolyte solution.

[0111] Based on all the normalized infiltration values, a state map of the infiltration degree distribution is generated.

[0112] Specifically, the maximum value of the calculated electrolyte content at all scanning detection points is extracted. It should be noted that the maximum calculated electrolyte content can also be calculated based on the porosity of the electrode material and the electrolyte capacity per unit volume of void space. A normalized wetting value is then calculated for each detection point based on the maximum calculated electrolyte content. These normalized wetting values ​​are then arranged according to the spatial coordinates of the corresponding scanning detection points to generate a two-dimensional distribution map, called the wetting degree distribution map. Figure 2 , 4 These represent the wetting degree distribution diagrams provided in Examples 1 and 2, respectively.

[0113] Example 1 provides a method for evaluating the wetting uniformity of a liquid electrolyte battery, comprising the following steps:

[0114] S1. Obtain using the method described above. Figure 2 The diagram showing the distribution of the degree of infiltration is presented, and the diagram is processed, including the following steps:

[0115] Adaptive ROI extraction is performed on the wetting degree distribution state map of the target battery to obtain the effective region of the wetting degree distribution state map;

[0116] Denoising is performed on the effective area of ​​the infiltration degree distribution state map to obtain the processed infiltration degree distribution state map.

[0117] In one implementation, the adaptive ROI extraction includes the following steps:

[0118] The infiltration degree distribution state map is binarized based on an adaptive edge threshold to obtain an initial binary mask.

[0119] A morphological closing operation is performed on the initial binary mask, and the connected component is labeled on the closing operation result. The connected region with the largest area is selected as the main battery region.

[0120] Morphological etching is performed on the main area of ​​the battery to obtain an analysis ROI mask. The minimum bounding rectangle of the analysis ROI mask is used as the clipping boundary. After clipping, the effective area of ​​the wetting degree distribution state map is obtained.

[0121] If the area of ​​the erosion result is less than a preset percentage of the total area, or if multiple connected regions with similar areas (the ratio of adjacent connected region areas is less than a preset percentage) are detected, the adaptive ROI extraction is deemed to have failed, and the process reverts to full-image analysis with a corresponding warning.

[0122] For example, the adaptive ROI extraction specifically includes the following steps:

[0123] Calculate the adaptive edge threshold (Otsu's method or maximum entropy thresholding method can be used), compare the pixel values ​​in the immersion distribution state map with the adaptive edge threshold, if the pixel value is greater than the adaptive edge threshold, assign the value 1 to the pixel, otherwise assign the value 0. After traversing the pixels and assigning values, the initial binary mask is obtained.

[0124] A morphological closing operation is performed on the initial binary mask. This morphological closing operation includes dilation followed by erosion. The dilation operation involves using a circular structuring element with a radius of 3 to 5 pixels (the size of the structuring element can be adjusted appropriately according to the image resolution) to dilate the initial binary mask. The center of the structuring element is used to scan every pixel position of the initial binary mask. If at least one pixel within the area covered by the structuring element has a value of 1, the value of the output pixel position corresponding to the center of the structuring element is set to 1; otherwise, it is set to 0. The erosion operation involves using the same structuring element to erode the dilation result. The center of the structuring element is used to scan every pixel position of the dilation result. If all pixels within the area covered by the structuring element have a value of 1, the value of the output pixel position corresponding to the center of the structuring element is set to 1; otherwise, it is set to 0.

[0125] The corrosion results are labeled with connected components to identify all independent connected regions. The area of ​​each connected region is calculated, and the connected region with the largest area is retained as the main body region of the battery.

[0126] Morphological erosion is performed on the main battery area using circular structural elements with a radius of 2 to 4 pixels to obtain an analysis ROI mask. The minimum bounding rectangle of the analysis ROI mask is used as the clipping boundary. After clipping, the effective area of ​​the wetting degree distribution state map is obtained.

[0127] If the area of ​​corrosion in the main battery region is less than 10% of the total image area (this percentage can be adjusted according to the actual battery size), the ROI extraction is deemed to have failed, and the process reverts to full image analysis with a warning. Reverting to full image analysis means skipping adaptive ROI extraction and setting an analysis ROI mask (i.e., a full 1 mask with the same size as the wetting distribution map) for the wetting degree distribution map. At this point, no effective cropped region is generated. If multiple connected components with similar areas are detected, it indicates a possible battery segmentation error, and manual review is recommended. "Similar areas" means that the area ratio of two adjacent connected components arranged in descending order of area is less than 1.2 (this threshold is only an example and can be adjusted according to actual image features).

[0128] To eliminate isolated noise, minute artifacts, or edge burrs introduced by ultrasonic scanning, image acquisition, or threshold segmentation, while preserving the true boundaries and local variation characteristics of the electrolyte distribution as much as possible, a denoising operation can be further performed after obtaining the effective area of ​​the wetting degree distribution map. The denoising employs one or more combinations of median filtering, Gaussian filtering, bilateral filtering, morphological opening operations, or small-area connected component removal. In a preferred embodiment, median filtering denoising is used, specifically including the following steps:

[0129] A median filter with a window size of 3×3 can be used to calculate the neighborhood median for each pixel position in the effective area of ​​the immersion degree distribution state map to obtain the initial filtering result;

[0130] Calculate the absolute deviation between the center pixel value and the median of its neighborhood, and count the number of pixels in the 8-neighborhood that are close to the center pixel value (i.e., absolute deviation ≤ absolute deviation allowable value). If the absolute deviation > absolute deviation warning value and the number of pixels in the 8-neighborhood that are close to the center pixel value ≤ 2, then the center pixel is determined to be an isolated noise point, and the filtering result is accepted (i.e., replaced by the median). Otherwise, the center pixel is determined to belong to the true boundary or uniform region, and the original value is retained.

[0131] For pixels in the boundary region where a complete 3×3 neighborhood cannot be obtained, the original value is directly retained.

[0132] It should be noted that the allowable and warning values ​​for absolute deviation in median filtering denoising can be preset based on the numerical range of the infiltration distribution map and the image quality. For example, when the infiltration normalization value ranges from 0 to 1, these values ​​can be set to 0.05 and 0.15 respectively; alternatively, the global standard deviation σ can be used, set to 0.5σ and 1.5σ respectively. Specific values ​​can be adjusted based on the actual noise level, aiming to effectively distinguish isolated noise points from the true boundary, and are not limited to a single value.

[0133] like Figure 3 As shown, Figure 3 for Figure 2 The distribution map of the degree of infiltration obtained after the above step S1.

[0134] S2. Extract features from the infiltration degree distribution state map after processing in step S1, and perform feature processing on the extracted features to obtain the feature values ​​of each feature after processing.

[0135] Specifically, features are extracted from the processed infiltration degree distribution map. These features include at least one of global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features. Preferably, the features include global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features.

[0136] Global immersion statistical features take the entire test area of ​​the battery as a whole, ignoring local details. By statistically analyzing the immersion degree values ​​of all pixels, it describes the central tendency, dispersion, and distribution pattern of the overall immersion level. Global immersion statistical features are not concerned with spatial location, but only with the overall statistical regularity of the values. In one embodiment, the global immersion statistical features include at least one of the following: discrete uniformity features, extreme value span difference features, symmetry offset features, steepness and sharpness features, interval proportion features, and global distribution pattern features. Among them, the discrete uniformity feature quantifies the overall fluctuation amplitude and dispersion of the battery's global immersion values ​​through statistical measures, reflecting the immersion uniformity. The discrete uniformity feature focuses on the overall dispersion level of the values, without involving extreme values, distribution shape, or segment proportion. Specifically, the discrete uniformity feature includes at least one of the following: standard deviation, interquartile range, mean absolute deviation, and root mean square deviation. Extreme value span difference features characterize the span of infiltration values ​​at extreme values ​​or within a wide quantile range. They are used to identify local extreme regions or measure overall dispersion, ignoring intermediate distribution patterns. Specifically, the extreme value span difference features include at least one of range and decimal point (the difference between the 90th and 10th percentiles). Symmetry offset features quantify the asymmetry and offset characteristics of the infiltration value distribution, used to identify morphological defects such as left and right skewness. Symmetry offset features specifically characterize the direction and magnitude of distribution skewness. Specifically, the symmetry offset features include at least one of mean-median offset and left-right quantile sample proportion difference. The mean-median offset refers to the difference between the mean and the median, and the left-right quantile sample proportion difference refers to the difference between the percentage of pixels below a preset left quantile (e.g., the 10th percentile) and the percentage of pixels above a preset right quantile (e.g., the 90th percentile). Steepness and spiking characteristics quantify the sharpness or gentleness of the main peak and the relative thickness of the tail, used to evaluate the concentration of infiltration values ​​in the main peak region and the dispersion in the tail. Specifically, the steepness and spiking characteristics include at least one of interquartile steepness coefficient and MAD normalized steepness, where the interquartile steepness coefficient refers to the ratio of the interquartile range to the decimal range, and the MAD normalized steepness refers to the ratio of the mean absolute deviation to the standard deviation. Interval proportion characteristics divide the infiltration value range into several continuous intervals and statistically analyze the proportion of pixels in each interval to the total number of pixels, reflecting the area proportion of different infiltration levels. Interval proportion characteristics are based on threshold segmentation. Specifically, the interval proportion characteristics include at least one of central interval sample proportion and first and last tail sample proportion, where the central interval sample proportion refers to the proportion of pixels whose infiltration values ​​fall within a preset central interval (e.g., within a certain range near the median), and the first and last tail sample proportion refers to the proportion of pixels whose infiltration values ​​fall within a preset first and last tail interval (e.g., below a certain low threshold or above a certain high threshold).The global distribution trend characteristics describe the degree of disorder, uncertainty, or order in the overall distribution of infiltrating values. They are used to assess the complexity and uniformity of the distribution. The global distribution trend characteristics do not involve spatial location and are calculated only based on numerical statistical characteristics. Specifically, the global distribution trend characteristics include information entropy, where the information entropy H is calculated as follows: ,in, The probability that the infiltration degree value falls into the i-th preset interval is given by n, where n is the preset number of interval divisions, which can be set according to the analysis precision, such as n=5 or n=10. It should be noted that the above-mentioned standard deviation, interquartile range, mean absolute deviation, root mean square deviation, range, etc. are all commonly used terms in the field of statistics, and their definitions and standard calculation formulas are well known to those skilled in the art, and will not be elaborated here.

[0137] Local spatial heterogeneity features, based on block partitioning, sliding windows, or neighborhood difference, calculate the infiltration differences in local regions, quantitatively characterizing the spatial non-uniformity and structural changes in infiltration distribution. These features focus on where the non-uniformity lies and how drastic the changes are, rather than overall numerical values ​​or specific defects. In one embodiment, the local spatial heterogeneous infiltration features include at least one of the following: sub-block local mean discrete features, regional infiltration gradient difference features, and window texture spatial heterogeneity features. The sub-block local mean discrete feature divides the area to be tested into several non-overlapping sub-blocks, calculates the mean of the infiltration degree within each sub-block, and performs discrete statistical analysis on the mean of all sub-blocks to evaluate the differences in the average infiltration level of different local areas. Specifically, the sub-block local mean discrete feature includes at least one of the following: sub-block mean range, sub-block mean standard deviation, sub-block mean interquartile range, sub-block mean mean absolute deviation, sub-block mean mean absolute deviation, and maximum sub-block mean deviation. The maximum sub-block mean deviation is the maximum value of the absolute deviation, and its calculation steps are as follows: calculate the absolute deviation between the mean of the infiltration degree within each sub-block and the mean of the global infiltration degree, and take the maximum value among all sub-blocks. The regional infiltration gradient difference feature calculates the spatial gradient of the infiltration degree and performs statistical analysis on the gradient field to characterize the intensity, direction, and overall trend of spatial changes in the infiltration degree. Specifically, the regional infiltration gradient difference feature includes at least one of the following: gradient standard deviation, gradient mean absolute deviation, gradient mean absolute deviation, gradient range, gradient mean absolute value, gradient root mean square, gradient maximum absolute value, gradient mean local variance, average gradient magnitude of adjacent sub-blocks, gradient difference of average gradient of adjacent sub-blocks, slope of trend component, and spatial gradient magnitude. Among them, the gradient mean local variance is used to characterize the intensity of local fluctuations in the gradient field. Its calculation steps are as follows: first, calculate the gradient magnitude of each pixel; then, take a 3×3 (or 5×5) neighborhood window centered on the pixel and calculate the local variance of the gradient magnitude within the window; finally, take the mean of the local variances of all pixels. The average gradient magnitude of adjacent sub-blocks is used to characterize the drastic change in the degree of wetting between sub-blocks. Its calculation steps are as follows: divide the area to be measured into several non-overlapping sub-blocks, calculate the average gradient magnitude within each sub-block, and then average the absolute values ​​of the differences in average gradient magnitudes between all pairs of adjacent sub-blocks (horizontal and / or vertical directions). For local heterogeneous component roughness, based on a sliding window, calculate the ratio of the standard deviation to the mean of the pixel values ​​within each window (adding a very small positive number to prevent division by zero), and then average the results across all windows. For the mean of window spatial dispersion, based on a sliding window, calculate the variance of the wetting degree values ​​within each window, and then average the variances across all windows.The window texture space heterogeneity class features use a sliding window to traverse the test area, extract texture features within each window, and perform statistical analysis on the window features to evaluate the complexity and structural heterogeneity of pixel value distribution within a local window. Specifically, the window texture space heterogeneity class features include at least one of the following: local heterogeneous component roughness, window gray-level mean standard deviation, window gray-level variance mean, window gray-level range mean, window gray-level entropy mean, window gray-level entropy standard deviation, window texture roughness mean, window spatial dispersion mean, and window local information entropy mean. It should be noted that the terms sub-block mean range, sub-block mean standard deviation, sub-block mean interquartile range, sub-block mean average absolute deviation, sub-block mean average absolute deviation, gradient standard deviation, gradient average absolute deviation, gradient average absolute deviation, gradient range, gradient absolute mean (gradient magnitude mean), gradient root mean square, gradient absolute maximum, trend component slope, spatial gradient magnitude, window grayscale mean standard deviation, window grayscale variance mean, window grayscale range mean, window grayscale entropy mean, window grayscale entropy standard deviation, window texture roughness mean, and window local information entropy mean are all commonly used terms in statistics or image processing. Those skilled in the art are familiar with their definitions and standard calculation methods, and their specific formulas will not be elaborated here. For the window grayscale entropy mean and window grayscale entropy standard deviation, the grayscale entropy (local information entropy) involved can be calculated using the aforementioned information entropy formula, and will not be repeated here.

[0138] The quantification features of wetting defects first extract abnormal weak wetting regions (such as dry areas, bubbles, impurities, etc.) through threshold segmentation, and then quantify these regions through connected component analysis, morphological measurement, area proportion, and aggregation degree to specifically characterize the presence, severity, and spatial distribution of defects. The quantification features of wetting defects only target regions judged as abnormal, are completely isolated from normal region statistics, and are dedicated to defect identification and classification. In one embodiment, the quantification features of wetting defects include at least one of the following: abnormal threshold statistical features, defect connected component morphological scale features, defect spatial distribution aggregation features, and defect boundary transition gradient features. Among them, the anomaly threshold statistical features only target the set of abnormal pixels determined by threshold segmentation, and statistically analyze the distribution characteristics of their infiltration values ​​(such as central tendency, dispersion, information content, etc.). They focus on the abnormal values ​​themselves and do not involve the spatial morphology or positional relationship of the defects. Specifically, the anomaly threshold statistical features include at least one of the following: abnormal grid proportion, average abnormal amplitude, root mean square of abnormal amplitude, maximum anomaly deviation, interquartile range outside the threshold, and anomaly information entropy. The abnormal grid proportion refers to the proportion of grids containing abnormal pixels (infiltration levels below a preset threshold) to the total number of grids when the test area is divided into several equally sized grids. The maximum anomaly deviation refers to the maximum absolute value of the difference between the infiltration level values ​​of all abnormal pixels and the preset threshold. The interquartile range outside the threshold only calculates the difference between the upper and lower quartiles of the infiltration level values ​​of abnormal pixels. The defect connected component morphological scale feature extracts the geometric shape and size (such as defect number, area, shape factor, etc.) of each connected component after marking the abnormal regions as connected components. It focuses on the defect geometry and does not involve spatial relationships or boundary transitions between defects. Specifically, the defect connected component morphological scale feature includes at least one of the following: number of abnormal region blocks, average area of ​​abnormal regions, total area of ​​abnormal regions, and shape factor of abnormal regions. The shape factor is calculated for each abnormal connected component, with roundness calculated as: roundness = 4π × area / perimeter. 2 Roundness can be used as a shape factor; a roundness closer to 1 indicates a defect is closer to a circle, while a smaller roundness indicates a more elongated defect. Defect spatial distribution clustering features describe the degree of spatial clustering of multiple defect connected domains (such as average distance, local density, clustering index, etc.). These features focus on the spatial relationships between defects, evaluating whether they are isolated and scattered or clustered. Specifically, the defect spatial distribution clustering features include at least one of defect clustering degree and local defect density. The defect clustering degree uses the nearest neighbor index, calculated as the ratio of the average nearest neighbor distance of the centroids of all defect connected domains to the expected distance of a random distribution. The expected distance of the random distribution is... ,in, Defect density (number of defects per unit area). Local defect density is calculated by taking each defect connected region as the center, within a circular area of ​​a preset radius (e.g., 10 pixels), calculating the ratio of the total area of ​​other defects within that circular area to the area of ​​the circular area, and then averaging the ratios for all defect connected regions. This average is used as the local defect density. Defect boundary transition gradient features are calculated along the boundary of the defect connected region, including the gradient magnitude, gradient range, and mean difference between the inner and outer sides of the boundary. These features focus on the boundary changes between the defect and normal areas, characterizing the degree of transition (clear or gradual). Specifically, the defect boundary transition gradient features include at least one of the following: average gradient magnitude of the defect boundary, gradient range of the defect boundary, and mean difference of infiltration on both sides of the defect boundary. It should be noted that the mean abnormal amplitude, root mean square of abnormal amplitude, number of abnormal region blocks, average area of ​​abnormal region, total area of ​​abnormal region, average gradient magnitude of defect boundary, gradient range of defect boundary, and mean difference of infiltration on both sides of defect boundary are all common knowledge in the fields of image processing or morphological analysis, and their standard definitions are known to those skilled in the art; therefore, they will not be elaborated upon here.

[0139] Topological morphological features ignore the specific numerical value of the wetting degree and only focus on the overall morphological configuration features of the highly wetting area (or normally wetting connected area) in the battery test area, such as geometric connectivity, number of pores, skeleton structure, and Euler number. Topological morphological features do not reflect the strength or difference of wetting; they only describe pure topological geometric attributes such as whether normally wetting areas are connected and whether their shapes are regular. In one embodiment, the topological morphological features include at least one of topological connected domain morphological features, skeleton extension structure features, and topological aggregation orientation features. Specifically, the topological connected domain morphological features, based on the binarized normally wetting area, extract the overall number and scale information of connected domains, focusing on macroscopic connectivity statistics such as the number of blocks and the size of the largest block. Specifically, the topological connected domain morphological features include at least one of the maximum connected domain area ratio and connected domain number density, where the maximum connected domain area ratio reflects the concentration of the main region, and the connected domain number density reflects the degree of regional fragmentation. The skeleton extension structure features involve extracting the skeleton of the largest connected region, analyzing its topological structure, and focusing on the skeleton extension characteristics such as the shape of the veins within the largest block (e.g., endpoints, branches, total length). Specifically, the skeleton extension structure features include at least one of the following: the number of skeleton endpoints, the number of skeleton branches, and the total skeleton length. The number of skeleton endpoints reflects the number of branches at the end of the region, the number of skeleton branches reflects the bifurcation complexity, and the total skeleton length reflects the elongation of the region. The topological clustering orientation features are calculated based on all binarized normal infiltrated regions, focusing on the overall directional distribution of all normal infiltrated regions. Specifically, the topological clustering orientation features include at least one of the following: principal direction intensity and direction entropy. The principal direction intensity reflects whether there is a dominant extension direction. The calculation steps for the principal direction intensity are: for the binarized normal infiltrated regions, calculate the principal direction angle of each connected region (e.g., obtained through image moments or the minimum bounding rectangle), and set 0... o ~180 o The azimuth range is divided into K equally spaced intervals (e.g., K=18, every 10). o The proportion of connected components within each interval is statistically analyzed, and the interval with the largest proportion is taken as the principal direction intensity. Directional entropy reflects the uniformity of direction distribution (the smaller the entropy value, the more concentrated the direction). Directional entropy is calculated based on the principal direction angle distribution histogram, calculating the entropy value of the probability of each interval. It should be noted that the proportion of the largest connected component area, the density of connected components, the number of endpoints of the largest connected component skeleton, the number of skeleton branch points, and the total skeleton length are all well-known terms in the field of image processing, and their standard definitions and calculation methods are familiar to those skilled in the art; therefore, they will not be elaborated upon here.

[0140] It should be noted that all the above features are extracted based on pixel values ​​in the processed infiltration distribution map. Extraction methods include, but are not limited to: pixel-level traversal of the entire test area to directly statistically analyze the global numerical distribution and obtain global statistical features; dividing the test area into sub-blocks or using a sliding window to perform discrete analysis, gradient calculation, or texture analysis on each local region to extract local spatial heterogeneity features; separating abnormal regions through threshold segmentation, and then performing connected component labeling, morphological measurement, spatial clustering analysis, and boundary gradient calculation on the abnormal pixel set to extract defect quantification features; and obtaining topological morphology features through binarization, connected component analysis, skeleton extraction, and orientation statistics. Those skilled in the art can use corresponding image processing algorithms (such as block statistics, gradient operators, window traversal, threshold segmentation, connected component labeling, skeletonization, etc.) to achieve feature extraction based on the above feature types. These image processing algorithms are all common knowledge in the field, and their specific implementation steps will not be elaborated here.

[0141] In the actual extraction process, at least one of the above four types of features, or at least one subordinate feature of each type of feature, can be selected and used in combination, depending on the type of battery to be tested, the detection accuracy requirements, and the computing resources. There is no unique limitation here.

[0142] After extracting the above features, since the dimensions, distribution characteristics and correlation with electrolyte wetting quality of different features may be different, a unified preprocessing is required to eliminate the influence of dimensions, unify the direction of indicators, and finally normalize them to the same scale for subsequent comprehensive evaluation.

[0143] In one implementation, the step of performing feature processing on the extracted features to obtain the processed feature values ​​includes the following steps:

[0144] Based on the correlation between the extracted feature values ​​and the electrolyte wetting uniformity, the features are classified into one of the following: negative features, positive features, and bidirectional features. Among them, the larger the feature value of the negative feature, the worse the wetting uniformity; the larger the feature value of the positive feature, the better the wetting uniformity; and the larger the feature value of the bidirectional feature deviates from the target value, the worse the wetting uniformity.

[0145] For negative features, retain the original values; for positive features, convert them to negative indices through linear inversion; for bidirectional features, convert them to negative indices based on the degree of deviation.

[0146] Hyperbolic tangent normalization is performed on the negative features and the transformed negative indices, mapping them to the [0,1] interval to obtain the hyperbolic tangent normalized eigenvalues ​​of each feature.

[0147] Specifically, based on the correlation between each feature value and the electrolyte wetting uniformity, the extracted features are divided into negative features, positive features, and bidirectional features. Among these, the larger the feature value of a negative feature, the worse the wetting uniformity. The extracted features include: standard deviation, interquartile range, mean absolute deviation, root mean square deviation, range, decimal point distance, interquartile kurtosis coefficient, MAD normalized kurtosis, proportion of first and last samples, information entropy, sub-block mean range, sub-block mean standard deviation, sub-block mean interquartile range, sub-block mean mean absolute deviation, sub-block mean mean absolute deviation, maximum sub-block mean deviation, gradient standard deviation, gradient mean absolute deviation, gradient mean absolute deviation, gradient range, gradient absolute value mean, gradient root mean square, gradient absolute value maximum, gradient local variance mean, average gradient magnitude of adjacent sub-blocks, gradient difference between adjacent sub-blocks, spatial gradient magnitude, local heterogeneous component roughness, and window grayscale mean standard deviation. The following features are considered: mean window grayscale variance, mean window grayscale range, mean window grayscale entropy, mean window grayscale entropy standard deviation, mean window texture roughness, mean window spatial dispersion, mean window local information entropy, percentage of abnormal meshes, mean abnormal amplitude, root mean square of abnormal amplitude, maximum abnormal deviation, interquartile range outside the threshold, abnormal information entropy, number of abnormal region blocks, average area of ​​abnormal regions, total area of ​​abnormal regions, shape factor of abnormal regions, defect clustering degree, local defect density, average magnitude of defect boundary gradient, range of defect boundary gradient, mean difference between wetting values ​​on both sides of the defect boundary, number density of connected components, number of endpoints of the largest connected component skeleton, number of branch points of the skeleton, total length of the skeleton, and intensity of the principal direction. All negative features do not require changing their direction, i.e., they retain their original values.

[0148] The larger the feature value of a positive feature, the better the infiltration uniformity. Among the features extracted above, the proportion of samples in the central interval, the proportion of the area of ​​the largest connected region, and the directional entropy are all positive features. Positive features are converted to negative indices through linear inversion, that is, the original values ​​of the positive features are converted into linearly inverted values. Specifically, let X be the reference maximum value of a certain positive feature. max If the original value of this feature is X, then the transformed negative index value X' = X max -X, in specific implementation, X max The maximum value of this feature from historical high-quality samples, the maximum value of this feature from all samples to be tested in the current batch, or the theoretical maximum value preset according to process requirements can be used. Those skilled in the art can choose an appropriate method based on the actual data distribution, and no single limitation is made here. The linear inverse transformation ensures that the larger the original value of the positive feature (the better the wetting uniformity), the smaller the transformed value.

[0149] The greater the deviation of the feature value from the target value in a two-way feature, the worse the infiltration uniformity. Among the features extracted above, the mean-median offset, the difference in sample proportion between the left and right quantiles, and the slope of the trend component are all two-way features. For two-way features, the deviation is converted into a negative indicator. Specifically, for a given two-way feature, the original value of the feature is X, and the optimal value O and the maximum acceptable deviation T are preset. The deviation is then calculated. The deviation degree D is used as the negative index value after conversion. The larger the deviation degree D, the worse the wetting uniformity. It should be noted that for bidirectional features, the optimal value O is usually set directly according to the physical meaning of the feature. For example, for mean-median offset, the ideal state is no offset or no trend, so O=0 is set. The maximum acceptable deviation T can be determined based on historical high-quality samples, set based on process requirements or experience, or set based on theoretical boundaries. Among them, based on historical high-quality samples, the high quantile (e.g., 95%) of the absolute deviation of qualified samples can be taken as T.

[0150] After unifying the feature directions, since the dimensions and numerical ranges of different features may vary significantly, normalization is required to map them to a unified [0,1] interval. This implementation uses the hyperbolic tangent function for normalization, which has the advantage of robustness to extreme values, and the transformed value range naturally falls within (0,1). Specifically, for any negative index (including negative features that retain their original values, positive features after linear inverse transformation, and bidirectional features after deviation transformation), let the feature value before hyperbolic tangent normalization be y, then the feature value after hyperbolic tangent normalization is... The calculation formula is: ,in, Let be the hyperbolic tangent function, defined as M is the reference median, IQR is the reference interquartile range, and c is a scaling parameter (c is generally set to 1). It should be noted that the reference median M and reference interquartile range IQR are used to standardize the feature value y, ensuring that most of the data falls within a suitable interval. They can be determined using any of the following methods: based on historical high-quality samples, based on empirical or theoretical settings, or based on adaptive methods for the current batch of samples. Based on historical high-quality samples, this involves: collecting a batch of battery samples with known good electrolyte wetting uniformity (historical high-quality sample set); calculating the feature value in each sample to obtain a set of data; taking the median of this set of data as M, and taking the interquartile range of this set of data as IQR.

[0151] In battery production line inspection, sensor malfunctions, image anomalies, or algorithm boundary failures may lead to missing feature values; simultaneously, extreme outliers (such as rare defects) may exist in the data. The aforementioned denoising only handles pixel-level noise, and feature calculation may still result in missing or outlier values. In one implementation, the feature processing of the extracted features further includes missing value imputation and outlier truncation. Specifically, missing value imputation includes: negative and positive features are imputed using the historical median, and bidirectional features are imputed using the historical median of deviation. Outlier truncation includes: defining normal threshold intervals for negative, positive, and bidirectional features, where the normal threshold interval is... Where M is the reference median and IQR is the reference interquartile range. This is a preset coefficient. Values ​​exceeding the corresponding normal threshold range are considered abnormal. Feature values ​​exceeding the upper limit of the normal threshold range are truncated to the upper limit, and feature values ​​below the lower limit are truncated to the lower limit. It should be noted that the preset coefficient... The settings can be configured based on the distribution characteristics of historical samples and the required detection accuracy. For example, A value of 2.5 is acceptable, and those skilled in the art can determine this value through experiments or statistical experience; it is not a single, definitive value here. The values ​​for the reference median M and the reference interquartile range (IQR) are as described above and will not be repeated here.

[0152] S3. Divide the processed features into optimizable factor groups and structural constraint factor groups, determine the basic total weight of each group and the basic weight of each feature, perform intra-group sample adaptive fine-tuning and intra-group weight normalization on the basic weight of each feature to obtain the final weight of each feature in the group, and calculate the uniformity basic score based on the processed feature value and its final weight.

[0153] In one implementation, dividing the processed features into a group of optimizable factors and a group of structural limiting factors includes:

[0154] Features that show a response relationship between characteristic changes and process parameter adjustments are grouped into optimizable factor groups;

[0155] Features whose characteristics are difficult to fundamentally change within the scope of process optimization and require design improvement are classified into the structural constraint factor group.

[0156] It should be noted that the purpose of grouping is to distinguish between features significantly affected by process parameters (optimizable) and features limited by battery structural design (requiring design improvement). The eigenvalues ​​of the optimizable factor group show a responsive relationship with process parameters such as electrolyte injection volume, formation current, and aging temperature; adjusting the process can significantly change the eigenvalues, reflecting the impact of process level on wetting uniformity. The eigenvalues ​​of the structural limiting factor group are difficult to fundamentally change within the scope of process optimization, mainly constrained by design factors such as electrode thickness, porosity, separator material, and tab position, requiring improvement from the design source. In practical implementation, initial grouping based on expert experience can be followed by experimental verification. For example, the experimental verification steps are: keeping the structure unchanged, changing process parameters; if the coefficient of variation of the eigenvalue exceeds a preset threshold, or the absolute value of the correlation coefficient with at least one process parameter is >0.5 (p<0.05), then it is classified as an optimizable factor group; otherwise, it is classified as a structural limiting factor group. Those skilled in the art can choose specific methods based on actual data; no single limitation is made here.

[0157] The features processed in step S2 include: standard deviation, interquartile range, mean absolute deviation, root mean square deviation, range, decimal point distance, interquartile steepness coefficient, MAD normalized steepness, percentage of samples at the beginning and end, information entropy, percentage of samples in the central interval, mean-median offset, difference in percentage of samples between left and right quantiles, range of sub-block mean, standard deviation of sub-block mean, interquartile range of sub-block mean, mean absolute deviation of sub-block mean, mean absolute deviation of sub-block mean, maximum deviation of sub-block mean, standard deviation of gradient, mean absolute deviation of gradient, mean absolute deviation of gradient, gradient range, mean absolute value of gradient, root mean square of gradient, maximum absolute value of gradient, mean local variance of gradient, average magnitude of gradient between adjacent sub-blocks, gradient difference between the mean of adjacent sub-blocks, and slope of trend components. Rate, spatial gradient magnitude, local heterogeneous component roughness, window gray level mean standard deviation, window gray level variance mean, window gray level range mean, window gray level entropy mean, window gray level entropy standard deviation, window texture roughness mean, window spatial dispersion mean, window local information entropy mean, abnormal mesh proportion, abnormal magnitude mean, abnormal magnitude root mean square, maximum abnormal deviation, outside threshold interquartile range, abnormal information entropy, number of abnormal region blocks, average area of ​​abnormal region, total area of ​​abnormal region, abnormal region shape factor, defect clustering degree, local defect density, defect boundary gradient mean magnitude, defect boundary gradient range, difference in mean infiltration values ​​on both sides of the defect boundary, connected component number density, and maximum connected component area proportion can all be classified into optimizable factor groups. The number of endpoints of the maximum connected domain skeleton, the number of skeleton branch points, the total length of the skeleton, the principal direction strength, and the directional entropy can all be classified into the structural constraint factor group. These mainly reflect the skeleton morphology and directional distribution of the connected domain. These characteristics are strongly correlated with design factors such as electrode rolling texture, diaphragm structure, and cell winding method. They are difficult to fundamentally change within the scope of conventional process optimization and usually require improvement from the design source.

[0158] In one implementation, determining the basic total weight of each group and the basic weight of each feature includes:

[0159] The basic total weight of each group and the basic weight of each feature are determined by a weight determination method, wherein the weight determination method includes at least one of the analytic hierarchy process, the Delphi method or the entropy weight method;

[0160] And it satisfies the following: the sum of the basic weights of each feature in each group is equal to the total basic weight of that group, the sum of the total basic weights of all groups is 1, and the sum of the basic weights of all features is 1.

[0161] In this embodiment, the basic total weight of each group and the basic weight of each feature within each group are calculated using a weight determination method. The weight determination method includes at least one of the analytic hierarchy process (AHP), the Delphi method, or the entropy weight method. Regardless of the method used, the final determined weights must satisfy the following constraints: the sum of the basic weights of each feature within each group equals the basic total weight of that group; the sum of the basic total weights of all groups equals 1; and the sum of the basic weights of all features equals 1.

[0162] In one implementation, the analytic hierarchy process (AHP) is used to determine the total basic weights of each group. and the basic weights of each feature The Analytic Hierarchy Process (AHP) quantifies expert experience into weights by constructing a judgment matrix. The specific steps include: (1) establishing a hierarchical structure model, taking the comprehensive evaluation of the electrolyte wetting uniformity of the evaluation target as the target layer; taking the optimizable factor group and the structural constraint factor group as the criterion layer; and taking the specific features within each factor group as the indicator layer; (2) constructing a judgment matrix, in which domain experts compare the importance of the two factor groups in the criterion layer with respect to the target layer, using the 1-9 scale method (1 represents equal importance, 9 represents extreme importance), and constructing a 2×2 judgment matrix. For example, if the optimizable factor group is considered to be slightly more important than the structural constraint factor group, then take 3. Similarly, for each feature within each factor group, construct the intra-group judgment matrix. (3) calculating the weight vector, using the sum-product method to calculate the eigenvector corresponding to the largest eigenvalue of the judgment matrix and normalizing it. Taking the sum-product method as an example, the specific steps include: normalizing each column of the judgment matrix; summing the normalized matrix by row; and normalizing the row sum vector to obtain the weight vector. The weight vector of the criterion layer is the basic total weight of each group. (e.g., the total weight of the optimizable factor group) The basic total weight of the structural constraint factor group ,and For each factor group, multiply the feature weight vector within the group by the total base weight of that group. That is, to obtain the basic weights of each feature. (4) Consistency test: Calculate the consistency ratio CR of the judgment matrix. The formula is: CR = CI / RI, where CI is the consistency index, and the formula for calculating CI is: ,in, To determine the largest eigenvalue of the judgment matrix, n is the order of the judgment matrix. Specifically, at the criterion level, the order n of the judgment matrix equals the number of factor groups involved in the comparison (i.e., 2). Within each factor group, the order n of the judgment matrix equals the number of features contained in that group (let's say m). RI is the random consistency index. When CR < 0.1, the consistency of the judgment matrix is ​​considered acceptable; otherwise, the judgment matrix needs to be adjusted. It should be noted that the details of the judgment matrix construction, the specific numerical methods for eigenvector calculation, and the complete value table of the random consistency index RI involved in the steps are all common knowledge in the field and will not be elaborated here. Those skilled in the art can use existing technologies to implement this according to actual application needs, as long as the consistency verification requirements are met.

[0163] Of course, the Delphi method or entropy weight method can also be used. The Delphi method reaches a consensus through multiple rounds of anonymous expert surveys, specifically including: (1) Expert scoring: Select several experts in battery technology and design to independently score the importance of each group and each feature. The scoring scale can be 0-10 points or 1-9 points, and the higher the score, the more important it is. (2) Statistics and feedback: Collect the first round of scoring results and calculate the mean, median and standard deviation of each weight item. Anonymously feed back the statistical results to all experts and invite experts to conduct the next round of scoring based on the feedback. (3) Convergence judgment: Repeat the above process until the standard deviation of each weight item is less than the preset threshold (such as 0.05), or the change in the mean between the two rounds is less than 5%. At this time, the expert opinions are considered to have converged. (4) Normalization: Take the mean of each weight item in the last round, and normalize the basic total weight (between group weight) and the relative weight of the feature within each group respectively, so that the sum of the weights between groups is 1 and the sum of the relative weights within groups is 1, and then multiply by the basic total weight of the corresponding group. The basic weights of each feature are obtained. Thus, the constraints are satisfied. It should be noted that in the Delphi method, the number of experts is generally recommended to be 5 to 15, and the specific number can be determined according to the actual number of experts in the field; the preset threshold (such as standard deviation <0.05, mean change <5%) is only an example, and can be adjusted according to actual needs during implementation to ensure convergence. The entropy weight method objectively calculates the weight based on the dispersion of the data itself, and is suitable for situations with sufficient historical samples. Specifically, it includes: (1) constructing a data matrix, collecting the feature values ​​of multiple battery samples, forming an m×n original data matrix, where m is the number of samples and n is the number of features. (2) using the normalized value of each feature to calculate the information entropy, and calculating the proportion of the i-th sample under the j-th feature. The calculation formula is as follows: Then the entropy value of the j-th feature The calculation formula is as follows: ,in, Satisfy 0≤ ≤1 and =1. (3) Calculate the objective weight, the objective weight of the j-th feature. The calculation formula is as follows: And objective weight The global weight is the sum of all feature weights, which is 1. (4) Grouping: Based on the factor group to which the feature belongs, the objective weights of the features within the group are normalized so that their sum equals the basic total weight of the group. The basic total weight of each group can be preset (e.g., equal weight 0.5), given by experts, or redistributed by calculating the sum of the information entropy of all features within each group using the entropy weight method. It should be noted that in the entropy weight method, each feature value should first undergo the aforementioned preprocessing (filling missing values, truncation of outliers, unification of direction, and hyperbolic tangent normalization) to ensure that the dimensions are consistent and the values ​​are non-negative. If the feature value has not been normalized to the [0,1] interval, it needs to be processed by minimum-maximum standardization or mean-standard deviation standardization to make the data suitable for entropy calculation. The basic total weight of each feature in the grouping process (e.g., equal weight 0.5, expert scoring, or entropy weight redistribution) can also be obtained by the analytic hierarchy process (AHP) or other subjective weighting methods. Those skilled in the art can choose according to the actual situation, and there is no unique limitation here.

[0164] Regardless of the weight determination method used, if the initial calculation results do not meet the following constraints: the sum of the basic weights of each feature within each group equals the total basic weight of that group; the sum of the total basic weights of all groups is 1; and the sum of the basic weights of all features is 1, then a secondary normalization process is required. Specifically, the secondary normalization process includes inter-group weight normalization and intra-group feature weight normalization. For inter-group weight normalization, the initial value of the total basic weight of each group is set to... The total basic weight of each group after normalization The calculation formula is as follows: This ensures that the sum of the basic total weights of each group is 1. Within-group feature weights are normalized. Let the initial sum of the feature weights within a group be S, and the initial value of the basic weight of a feature within that group be... The normalized base weights of a certain feature within this group The calculation formula is as follows: It should be noted that the base weights are normalized based on a certain feature within the group. As the basic weight of this feature .

[0165] In one implementation, the process of performing intra-group sample adaptive fine-tuning and intra-group weight normalization on the basic weights of each feature includes:

[0166] Within the same factor group, feature adjustment coefficients are calculated based on the preset fine-tuning intensity coefficients of each feature and the correlation coefficients with the long-term performance indicators of the battery; the basic weights of each feature within the group are fine-tuned using the feature adjustment coefficients to obtain the fine-tuning weights of each feature within the group.

[0167] For each factor group, calculate the sum of the fine-tuned weights of each feature within the group as the total fine-tuned weights of the group; calculate the scaling factor for the group, which is equal to the total base weights of the group divided by the total fine-tuned weights of the group; multiply the fine-tuned weights of each feature within the group by the scaling factor to obtain the final weights of each feature within the group.

[0168] After determining the basic weights for each group and feature, to further ensure that the weights reflect the individual characteristics of specific samples (i.e., the correlation between features and long-term battery performance), this invention introduces a sample adaptive fine-tuning step. This step is performed within the same factor group and includes setting a preset fine-tuning intensity coefficient, calculating the correlation coefficient, calculating the feature adjustment coefficient, and calculating the fine-tuning weights. The preset fine-tuning intensity coefficient... This represents the sensitivity of the j-th feature to changes in long-term battery performance, i.e., the strength of the impact of feature value fluctuations on long-term battery performance, with a preset fine-tuning strength coefficient. The values ​​can be pre-assigned by experts based on the physical meaning of the features, and the value range is usually [0,2]. The default value can be 1. For example, for key features such as discrete uniformity and defect area ratio, A value of 1.2 is acceptable; if a certain feature is not desired to participate in fine-tuning, then the value of that feature... The value can be 0. Preset fine-tuning intensity coefficient. It can also be determined through sensitivity analysis (changing process parameters and observing the ratio of characteristic changes to performance changes). Preset fine-tuning strength coefficient. Once determined, maintain consistency for the same batch of samples. Long-term battery performance metrics include, but are not limited to, cycle life, capacity retention, and internal resistance growth rate. Correlation coefficients. The linear correlation between the j-th feature and the long-term performance index of the battery is represented by the Pearson correlation coefficient, calculated using the following formula: Where N is the number of historical samples used to calculate the correlation coefficient. Let the value of the k-th sample be the value of the j-th feature. Let the j-th feature be the arithmetic mean of all N samples. Let be the long-term performance index value of the battery for the k-th sample. To ensure that the sign of the correlation coefficient has a clear physical meaning (i.e., the larger the eigenvalue, the worse the wetting uniformity, and the worse the long-term performance index reflects), it should be ensured that... Consistent with the feature direction, i.e. The larger the value, the worse the long-term performance of the battery. If the original performance index is larger and the performance is worse (such as capacity decay rate, internal resistance growth rate), then the original value is retained; if the original performance index is larger and the performance is better (such as cycle life), then a monotonically decreasing conversion is required (such as taking the reciprocal or taking the negative value). This is the arithmetic mean of the long-term performance metric over all N samples. Feature adjustment coefficient. The feature adjustment coefficient in this embodiment combines the fine-tuning intensity coefficient and the correlation coefficient. The calculation formula is: If calculated ≤0, can be set =0.1 or other extremely small positive numbers. Of course, other methods can also be used to calculate the characteristic adjustment coefficient. ,like As long as the principle of adjusting the weights based on the stronger the correlation with long-term performance is reflected, those skilled in the art can choose according to actual circumstances; no single limitation is made here. The basic weights of each feature within the same factor group should be adjusted accordingly. Multiply by the corresponding feature adjustment coefficient To obtain fine-tuned weights Fine-tuning the weights This reflects the correlation between the current sample's features and its long-term performance.

[0169] After completing the sample adaptive fine-tuning, the fine-tuning weights of each feature are obtained. Subsequently, because the fine-tuning process may have altered the sum of the feature weights within the same factor group, causing this sum to no longer equal the group's base total weight, it is necessary to perform in-group weight normalization to ensure that the sum of the fine-tuned weights of each feature within the group equals the group's base total weight. In this embodiment, the weight normalization process within each group is performed independently on a per-factor-group basis, specifically including: (1) calculating the sum of the fine-tuned weights. For the current factor-group, let m be the total number of features in the group, and the fine-tuning weight of each feature be... The formula for calculating the sum S of the fine-tuned weights of all features within the group is: (2) Calculate the scaling factor. The scaling factor K is used to adjust the sum of the fine-tuned weights to be consistent with the base total weights. The calculation formula is: K = / S. (3) Calculate the final weights by multiplying the fine-tuned weights of each feature in the group by the scaling factor to obtain the final weights of that feature. The calculation formula is as follows: After in-group weight normalization, the final sum of the weights of all features within the factor group is... Strictly equal to the group's total base weight .

[0170] In one implementation, calculating the uniformity baseline score based on the feature values ​​after feature processing and their final weights includes:

[0171] Calculate the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each feature and their corresponding final weights, and then subtract the sum of the products from 1 to obtain the uniformity base score.

[0172] In one implementation, the hyperbolic tangent normalized eigenvalues ​​and their final weights after feature processing are described. Calculate the basic score for uniformity Among them, the basic score of uniformity Subtract 1 the hyperbolic tangent normalized eigenvalues ​​after feature processing and their final weights within each group. The sum of the products, that is, ,in, Let be the processed hyperbolic tangent normalized eigenvalue of the j-th feature, n be the total number of features involved in the evaluation, and be the homogeneity baseline score. It is used to quantify the uniformity of electrolyte wetting as a whole.

[0173] It should be noted that in actual numerical calculations, due to floating-point operation errors (such as rounding errors and cumulative errors), a theoretical value of 1.0 may be calculated as 1.0000000001, slightly greater than 1. To avoid this situation causing the uniformity base score to have a negative value (i.e., 1 − slightly greater than 1 = a negative minimum), it can be set that if the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of the above features and the corresponding final weights is greater than 1, then the uniformity base score is set to 0 as a defensive measure to ensure that the score is always in the [0,1] interval.

[0174] S4. Based on expert experience, select morphological risk features from the processed features and classify them into the normal infiltration morphological risk feature set and the adverse infiltration morphological risk feature set. Normalize the weights of each morphological risk feature within the set to obtain the normalized weights of each feature within the set. Calculate the comprehensive morphological health score based on the processed feature values ​​and normalized weights of each feature within the set.

[0175] In one implementation, the in-set weight normalization processing of the basic weights of each risk characteristic includes:

[0176] Calculate the normalized weights of each feature in the risk feature set of normal infiltration morphology and the risk feature set of adverse infiltration morphology, respectively. The normalized weight of each morphological risk feature is equal to the basic weight of that feature divided by the sum of the basic weights of all morphological risk features in the set.

[0177] Based on expert experience, morphological risk features are selected from the processed features and divided into a normal infiltration morphological risk feature set and an adverse infiltration morphological risk feature set. It should be noted that the normal infiltration morphological risk feature set and the adverse infiltration morphological risk feature set are mutually exclusive, and the features in the two sets are completely different. That is, no feature belongs to both the normal set and the adverse set at the same time. This ensures that the same feature will not make duplicate contributions to the two risk scores, avoiding assessment redundancy and logical conflicts.

[0178] Experts judged based on the physical meaning of the features and their correlation with morphological risk. From the features processed in step S3, they extracted features that could assess the overall morphological health, namely, features that assessed the risk of normal infiltrative morphology and the risk of adverse infiltrative morphology. The screened morphological risk features constituted the sets of features for normal infiltrative morphological risk. and set of risk characteristics of adverse infiltration patterns Specifically, Including features related to normal infiltration morphology, the proportion of the largest connected region area and the intensity of the principal direction among the features processed in step S3 can be classified into the risk feature set of normal infiltration morphology. . Including features related to undesirable infiltration morphology, the features processed in step S3, such as connected component density, number of endpoints of the largest connected component skeleton, number of skeleton branch points, total skeleton length, directional entropy, percentage of abnormal meshes, mean abnormal amplitude, root mean square of abnormal amplitude, maximum abnormal deviation, interquartile range outside the threshold, abnormal information entropy, number of abnormal region blocks, average area of ​​abnormal regions, total area of ​​abnormal regions, shape factor of abnormal regions, defect clustering, local defect density, average magnitude of gradient at defect boundary, range of gradient at defect boundary, and difference in mean infiltration values ​​on both sides of the defect boundary, can be classified into the set of risk features for undesirable infiltration morphology. .

[0179] Due to the basic weights of each feature The weights are predetermined globally using the analytic hierarchy process, the Delphi method, or the entropy weight method. To assess the relative importance within risk sets of similar morphologies, the basic weights of each risk characteristic within each set need to be normalized. Specifically, the total weight of the risk characteristic set of normal infiltration morphologies... The calculation formula is as follows: The total weight of the set of risk characteristics of adverse infiltration patterns The calculation formula is as follows: .for The normalized weight of each feature j in the set is: The calculation formula is as follows: ;for The normalized weight of each feature j in the set is: The calculation formula is as follows: .

[0180] In one implementation, calculating the comprehensive morphological health score includes:

[0181] Calculate the risk score for normal morphology and the risk score for adverse morphology. The risk score for normal morphology is the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each morphological risk feature in the set of normal infiltration morphological risk features and their corresponding normalized weights. The risk score for adverse morphology is the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each morphological risk feature in the set of adverse infiltration morphological risk features and their corresponding normalized weights.

[0182] Calculate the overall morphological health score, where the overall morphological health score is: ,in, , These are the preset weights for the risk scores of normal and adverse states, respectively. 1.

[0183] Calculate the risk score for normal morphology and the risk score for adverse morphology, where the risk score for normal morphology is... The formula used to quantify the risk level reflected by a set of normal morphological features is as follows: Unfavorable risk score The formula for quantifying the risk level reflected by a set of abnormal morphological features is as follows: (4) Calculate the overall morphological health score. Comprehensive physical health score The formula used to comprehensively evaluate the health level of battery immersion morphology is as follows: ,in, , These are the preset weights for the risk scores of normal and adverse states, respectively. 1. It should be noted that the preset weights... , It can be determined based on expert experience or historical data. For example, if it is believed that adverse morphology has a more serious impact on health, settings can be made accordingly. > If both are equally important, they can be set and 0.5. In specific implementation, it can also be determined by the AHP (Analytic Hierarchy Process) or experimental optimization. This embodiment does not limit the specific value, and those skilled in the art can set it according to actual needs.

[0184] S5. A comprehensive rating is generated based on the uniformity baseline score and the overall morphological health score, including:

[0185] The uniformity base score and the comprehensive morphological health score are respectively cropped to the [0,1] interval to obtain the uniformity score and the morphological health score;

[0186] The overall rating will be determined in the following order:

[0187] If the uniformity score is less than the third threshold for uniformity or the morphological health score is less than the third threshold for morphological health, it is deemed unqualified.

[0188] Otherwise, if the uniformity score is greater than or equal to the first threshold of uniformity and the morphological health score is greater than or equal to the first threshold of morphological health, it is judged as excellent.

[0189] Otherwise, if the uniformity score is greater than or equal to the second threshold for uniformity and the morphological health score is greater than or equal to the second threshold for morphological health, then it is considered qualified.

[0190] Otherwise, it is determined that optimization is needed;

[0191] Among them, the first threshold for uniformity > the second threshold for uniformity > the third threshold for uniformity, and the first threshold for morphological health > the second threshold for morphological health > the third threshold for morphological health.

[0192] To obtain a uniform base score After calculating the overall morphological health score, this embodiment further generates a comprehensive grade for battery wetting uniformity. The comprehensive grade is divided into four levels: Excellent, Acceptable, Requires Optimization, and Unacceptable, each corresponding to different process optimization or design improvement suggestions. Specifically, due to the uniformity base score... and overall physical health score During the calculation process, numerical errors may slightly exceed the theoretical range [0,1]. Boundary clipping is required to ensure that the score falls within the valid interval. Among these, the uniformity base score is... Cut to the [0,1] interval to obtain the uniformity score. Its formula is: The overall physical health score will be calculated. Cropping to the [0,1] interval yields the overall morphological health score. Its formula is: Three sets of uniformity thresholds and morphological health thresholds were set, among which, the first uniformity threshold... > Uniformity second threshold > Third threshold for uniformity The first threshold for physical health >Second threshold for morphological health >Third threshold for morphological health It should be noted that these thresholds can be preset based on the statistical distribution of historical high-quality samples or process requirements. For example, =0.9、 =0.75、 =0.6; =0.85、 =0.7、 =0.55. The above value is only an example. In actual implementation, it can be adjusted according to the battery type, application scenario and quality control standards. It will not be elaborated here.

[0193] The system uses a priority order to evaluate each battery sample, ensuring that each sample matches only one level. The specific rules are as follows:

[0194] (1) If the uniformity fraction < Third threshold for uniformity Or physical health score <Third threshold for morphological health> If so, it is deemed unqualified;

[0195] (2) If the unqualified condition is not met, i.e., condition (1) is not met, and the uniformity score is... ≥ Uniformity first threshold And the physical health score ≥ First threshold of morphological health If so, it is judged as excellent;

[0196] (3) If the above conditions are not met, i.e., conditions (1) or (2) are not met, and the uniformity fraction is... ≥ Uniformity first threshold And the physical health score ≥ First threshold of morphological health If so, it is deemed qualified;

[0197] (4) If none of the above conditions are met, that is, if none of the conditions (1), (2) and (3) are met, it is determined that optimization is required.

[0198] The overall grade of a battery is determined based on the above criteria. If a battery is deemed unqualified, it indicates significant defects or severe unevenness, requiring immediate production halt and investigation of the cause. If a battery is deemed excellent, it indicates excellent wetting uniformity and can serve as a benchmark sample. If a battery is deemed qualified, it indicates that the battery meets basic quality requirements and is permitted for normal shipment. If a battery is deemed to require optimization, it indicates that the battery has not yet met the qualified standard but has not yet constituted an unqualified battery, and it is recommended to make minor adjustments to process parameters or local design improvements.

[0199] Example 2 provides a method for evaluating the wetting uniformity of a liquid electrolyte battery, comprising the following steps:

[0200] S1. Obtain using the method described above. Figure 4The aforementioned infiltration degree distribution map is processed by the following steps:

[0201] Obtain a pre-constructed baseline map of the distribution of infiltration levels;

[0202] After extracting the ROI from the wettability distribution state map of the target battery according to preset parameters, the difference calculation is performed between the ROI and the baseline wettability distribution state map to obtain the wettability distribution state difference map of the target battery.

[0203] Adaptive ROI extraction is performed on the infiltration degree distribution state difference map to obtain the effective region of the infiltration degree distribution state difference map;

[0204] Denoising is performed on the effective area of ​​the infiltration degree distribution difference map to obtain the processed infiltration degree distribution difference map.

[0205] This embodiment provides a method for evaluating the uniformity of wetting in a liquid electrolyte battery, by constructing a baseline diagram of the wetting degree distribution (such as...). Figure 5 As shown), and compared with the wetting distribution state diagram of the target battery (as shown). Figure 4 The difference operation is performed on the target battery to obtain the difference map of the wetting degree distribution state, and the wetting uniformity of the target battery is further evaluated based on the difference map of the wetting degree distribution state.

[0206] In one embodiment, the construction of the infiltration degree distribution baseline map includes the following steps:

[0207] For the infiltration degree distribution state map of several candidate reference batteries, ROI extraction is performed according to uniform preset parameters to obtain several candidate reference battery ROI images, wherein the ROI extraction parameters of the candidate reference batteries and the target battery are consistent.

[0208] Denoising is performed on the ROI images of several candidate reference batteries using uniform parameters to suppress acquisition noise, resulting in ROI images of several candidate reference batteries after processing.

[0209] The same features are extracted from the ROI images of each candidate reference battery after processing, and the features are normalized to obtain normalized feature values. The features extracted from the candidate reference battery and the target battery are consistent.

[0210] Based on the normalized feature values, the comprehensive similarity between every two candidate reference cells is calculated, and the average similarity of each candidate reference cell is calculated based on the comprehensive similarity.

[0211] The top K candidate reference batteries are selected as baseline batteries based on average similarity from high to low, where K is a preset number.

[0212] The average pixel value at the same pixel position is calculated for the ROI image after processing the K reference batteries to generate the infiltration degree distribution state reference map.

[0213] Specifically, a batch of batteries with the same model as the target battery, stable production process, and qualified inspection results are selected as candidate reference batteries. The number of candidate reference batteries should be sufficient (e.g., no less than 30) to ensure statistical representativeness. For each candidate reference battery, ROI extraction is performed using the same preset parameters as the target battery (i.e., the same cropping boundary, ensuring consistent length and width dimensions of the battery cropping), resulting in the ROI image of each candidate reference battery. The ROI extraction parameters for all candidate reference batteries are kept consistent to ensure that the ROI size and image spatial range of all candidate reference batteries are consistent. The same feature set as the target battery is extracted from the processed ROI image of each candidate reference battery, that is, all features extracted from the candidate reference battery and the target battery are completely identical. For specific feature types and extraction methods, please refer to the relevant description in Example 1, which will not be repeated here. For each extracted feature, min-max normalization or Z-score normalization is used to scale the feature values ​​of all candidate reference batteries to the same order of magnitude to eliminate the influence of dimensions. For any two candidate reference batteries c1 and c2, based on their normalized feature vectors... , Calculate the overall similarity score. The overall similarity score can be calculated using any of the following: Euclidean distance similarity, cosine similarity, or Pearson correlation coefficient. Cosine similarity is preferred because it is not affected by the absolute numerical value of the features and focuses more on directional consistency. The calculation formula is as follows: For each candidate reference cell i, calculate the average similarity between it and all other candidate reference cells. A higher average similarity indicates better consistency between the cell and the overall sample. Sort the candidate reference cells from highest to lowest average similarity and select the top K cells as baseline cells, where K is a preset number that can be set according to actual needs (e.g., K can be 5 or 10). Then, process the ROI images of the selected K baseline cells and place them at the same pixel location. The arithmetic mean of the pixel values ​​is calculated using the following formula: ,in, For the k-th baseline battery ROI image at pixel location Pixel value at that location, The baseline map of the infiltration degree distribution at the pixel location The pixel value at each location is calculated. This calculation is repeated for all pixel locations to obtain a complete baseline map of the infiltration degree distribution.

[0214] The target battery immersion distribution map is subjected to ROI extraction using the same preset parameters for the aforementioned candidate reference batteries to obtain a target battery ROI image with the same spatial range as the immersion distribution baseline map. Then, the difference between the target battery ROI image and the immersion distribution baseline map is calculated at the same pixel position using the following formula: ,in, For the target battery ROI image at pixel location Pixel value at that location, The difference map of the infiltration degree distribution at the pixel location The pixel value at each location is calculated. This calculation is repeated for all pixel locations to obtain the complete infiltration degree distribution state difference map.

[0215] Adaptive ROI extraction is performed on the infiltration degree distribution difference map using the same adaptive ROI extraction method as described above (see the description of adaptive ROI extraction in Example 1), which will not be repeated here. This yields the effective region of the infiltration degree distribution difference map. The same denoising process is then performed on the effective region of the infiltration degree distribution difference map (see the description of denoising the effective region in Example 1), which will not be repeated here. This results in the processed infiltration degree distribution difference map (e.g., ...). Figure 6 (As shown), used for subsequent feature extraction and evaluation.

[0216] The method for extracting features from the processed infiltration degree distribution state difference map and performing feature processing on the features in Example 2 (i.e., step S2) and the subsequent steps S3-S5 can refer to the relevant description in Example 1, and will not be repeated here.

[0217] like Figure 7 As shown, this embodiment provides a liquid electrolyte battery wetting uniformity evaluation system, including an image processing module, an image feature extraction module, an image feature processing module, a feature weight calculation module, a target index calculation module, a comprehensive level determination module, and a database. The image processing module is connected to the image feature extraction module, the image feature extraction module is connected to the image feature processing module and the feature weight calculation module, the image feature processing module and the feature weight calculation module are connected to the target index calculation module, the target index calculation module is connected to the comprehensive level determination module, and all modules in the system are connected to the database.

[0218] After acquiring the saturation distribution map of the target battery, the image processing module processes the saturation distribution map to obtain a processed image. The specific processing flow includes: adaptive ROI extraction, noise reduction, and (optionally) difference calculation with the baseline image (see Example 2). The processed output image is a standardized image that matches the features to be evaluated.

[0219] The image feature extraction module is used to extract features from the processed image. The extracted features include, but are not limited to, at least one of the following: global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features. For specific feature items and calculation methods, please refer to the feature list and definitions in Example 1. The image feature extraction module outputs the original values ​​of each feature.

[0220] The image feature processing module is used to process the extracted features to obtain the feature values ​​of each feature after processing. The feature processing includes: direction unification (positive, negative, bidirectional conversion) and hyperbolic tangent normalization, and (optional) missing value filling and outlier truncation. For details of the specific steps, please refer to the feature processing section in Embodiment 1. The output of the image feature processing module is the processed value of each feature in the [0,1] interval.

[0221] The feature weight calculation module performs optimizable / structural constraint group weight calculation and morphological risk group weight calculation. The optimizable / structural constraint group weight calculation includes: dividing the processed features into optimizable factor groups and structural constraint factor groups (grouping rules are shown in Example 1); determining the basic total weight of each group and the basic weight of each feature using the analytic hierarchy process, Delphi method, or entropy weight method; then performing sample adaptive fine-tuning and intra-group normalization (fine-tuning formula and normalization steps are shown in Example 1) to obtain the final weight of each feature. The morphological risk group weight calculation includes: selecting morphological risk features from the processed features and dividing them into mutually exclusive sets of normal infiltration morphological risk features and adverse infiltration morphological risk features (division basis is shown in the explanation of morphological risk grouping in Example 1); performing intra-set normalization on the basic weights of features within each set (i.e., dividing each feature weight by the sum of the basic weights within the set) to obtain the normalized weight of each morphological risk feature within its respective set. All intermediate data involved in the above calculations (such as judgment matrices, expert scoring results, historical sample statistics, etc.) are stored in the database.

[0222] The target index calculation module calculates the uniformity base score based on the processed hyperbolic tangent normalized eigenvalues ​​of each feature and the corresponding final weights within the group. It calculates the normal infiltration morphology risk score based on the hyperbolic tangent normalized eigenvalues ​​of each feature within the normal infiltration morphology risk feature set and the corresponding normalized weights within the set. It calculates the adverse infiltration morphology risk score based on the hyperbolic tangent normalized eigenvalues ​​of each feature within the adverse infiltration morphology risk feature set and the corresponding normalized weights within the set. It calculates the comprehensive morphological health score based on the normal infiltration morphology risk score and the adverse infiltration morphology risk score. The specific calculation formulas are shown in Example 1 for the relevant uniformity base score, normal infiltration morphology risk score, adverse infiltration morphology risk score, and comprehensive morphological health score. At the same time, the uniformity base score and the comprehensive morphological health score are truncated to the [0,1] interval.

[0223] The comprehensive level determination module determines the comprehensive level based on the uniformity base score and the comprehensive morphological health score. Specifically, the comprehensive level is determined according to the determination rules described in Example 1. The threshold required for determination can be pre-stored in the database or set by the user.

[0224] The database stores data information for all modules in the system, including but not limited to: the original infiltration distribution map, processed images, feature extraction results, feature processing parameters, weight calculation results, intermediate calculated values, final evaluation scores, overall grades, and historical sample data. The database supports data reading, writing, querying, and backup.

[0225] It should be noted that the specific algorithm implementations of each module in this embodiment (such as the morphological parameters extracted from the ROI, the type and window size of the denoising filter, the specific statistics of feature extraction, the construction method of the AHP judgment matrix, the method for calculating the correlation coefficient of adaptive fine-tuning, the reference mean and standard deviation of hyperbolic tangent normalization, the threshold for comprehensive level determination, etc.) can all adopt the technical solutions disclosed in the foregoing method embodiments. Those skilled in the art, after reading the foregoing embodiments, can modularize these methods and integrate them into this system without creative effort. All steps not elaborated in detail in this embodiment are performed with reference to the corresponding descriptions in Embodiments 1 and 2, and will not be repeated here.

[0226] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for evaluating the wetting uniformity of a liquid electrolyte battery, characterized in that, Includes the following steps: Obtain a wetting degree distribution map of the target battery, and process the wetting degree distribution map to obtain a processed image; Features are extracted from the processed image, and the extracted features are processed to obtain the processed features. The features are selected from at least one of global statistical features, local spatial heterogeneity features, defect quantification features, and topological morphology features. The processed features are divided into optimizable factor groups and structural constraint factor groups. The basic total weight of each group and the basic weight of each feature are determined. The basic weight of each feature is subjected to intra-group sample adaptive fine-tuning and intra-group weight normalization to obtain the final weight of each feature within the group. Based on the processed feature values ​​and their final weights, the uniformity basic score is calculated. Based on expert experience, morphological risk features are selected from the processed features and divided into the normal infiltration morphological risk feature set and the adverse infiltration morphological risk feature set. The basic weights of each morphological risk feature are normalized within the set to obtain the normalized weights of each feature within the set. Based on the processed feature values ​​and their normalized weights of each feature within the set, the comprehensive morphological health score is calculated. A comprehensive rating is generated based on the uniformity baseline score and the overall morphological health score.

2. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 1, characterized in that, The process of processing the infiltration degree distribution map includes the following steps: Adaptive ROI extraction is performed on the wetting degree distribution state map of the target battery to obtain the effective region of the wetting degree distribution state map; Denoising is performed on the effective area of ​​the infiltration degree distribution state map to obtain the processed infiltration degree distribution state map.

3. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 1, characterized in that, The process of processing the infiltration degree distribution map includes the following steps: Obtain a pre-constructed baseline map of the distribution of infiltration levels; After extracting the ROI from the wettability distribution state map of the target battery according to preset parameters, the difference calculation is performed between the ROI and the baseline wettability distribution state map to obtain the wettability distribution state difference map of the target battery. Adaptive ROI extraction is performed on the infiltration degree distribution state difference map to obtain the effective region of the infiltration degree distribution state difference map; Denoising is performed on the effective area of ​​the infiltration degree distribution difference map to obtain the processed infiltration degree distribution difference map.

4. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 3, characterized in that, The construction of the baseline map of the infiltration degree distribution includes the following steps: For the infiltration degree distribution state map of several candidate reference batteries, ROI extraction is performed according to uniform preset parameters to obtain several candidate reference battery ROI images, wherein the ROI extraction parameters of the candidate reference batteries and the target battery are consistent. Denoising is performed on the ROI images of several candidate reference batteries using uniform parameters to suppress acquisition noise, resulting in ROI images of several candidate reference batteries after processing. The same features are extracted from the ROI images of each candidate reference battery after processing, and the features are normalized to obtain normalized feature values. The features extracted from the candidate reference battery and the target battery are consistent. Based on the normalized feature values, the comprehensive similarity between every two candidate reference cells is calculated, and the average similarity of each candidate reference cell is calculated based on the comprehensive similarity. The top K candidate reference batteries are selected as baseline batteries based on the average similarity from high to low, where K is a preset number; The average pixel value at the same pixel position is calculated for the ROI image after processing the K reference batteries to generate the infiltration degree distribution state reference map.

5. A method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 1 or 4, characterized in that, The global infiltration statistical features include at least one of the following: discrete uniformity features, extreme value span difference features, symmetry offset features, steepness and spiking features, interval proportion features, and global distribution pattern features; The local spatial heterogeneous infiltration features include at least one of the following: sub-block local mean discrete features, region infiltration gradient difference features, and window texture spatial heterogeneity features; The infiltration defect quantification features include at least one of the following: anomaly threshold statistical features, defect connected domain morphological scale features, defect spatial distribution clustering features, and defect boundary transition gradient features; The topological morphological features include at least one of the following: topological connected domain morphological features, skeleton extension structure features, and topological clustering orientation features.

6. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 1, characterized in that, The step of performing feature processing on each extracted feature to obtain the processed feature value includes the following steps: Based on the correlation between the extracted feature values ​​and the electrolyte wetting uniformity, the features are classified into one of the following: negative features, positive features, and bidirectional features. Among them, the larger the feature value of the negative feature, the worse the wetting uniformity; the larger the feature value of the positive feature, the better the wetting uniformity; and the larger the feature value of the bidirectional feature deviates from the target value, the worse the wetting uniformity. For negative features, retain the original values; for positive features, convert them to negative indices through linear inversion; for bidirectional features, convert them to negative indices based on the degree of deviation. Hyperbolic tangent normalization is performed on the negative features and the transformed negative indices, mapping them to the [0,1] interval to obtain the hyperbolic tangent normalized eigenvalues ​​of each feature.

7. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 6, characterized in that, The process of dividing the processed features into an optimizable factor group and a structural constraint factor group includes: Features that show a response relationship between characteristic changes and process parameter adjustments are grouped into optimizable factor groups; Features whose characteristics are difficult to fundamentally change within the scope of process optimization and require design improvement are classified into the structural constraint factor group; The determination of the basic total weight of each group and the basic weight of each feature includes: The basic total weight of each group and the basic weight of each feature are determined by a weight determination method, wherein the weight determination method includes at least one of the analytic hierarchy process, the Delphi method or the entropy weight method; And it satisfies the following: the sum of the basic weights of each feature in each group is equal to the total basic weight of that group, the sum of the total basic weights of all groups is 1, and the sum of the basic weights of all features is 1.

8. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 7, characterized in that, The process of adaptive fine-tuning of the basic weights of each feature within the group and normalizing the weights within the group includes: Within the same factor group, feature adjustment coefficients are calculated based on the preset fine-tuning intensity coefficients of each feature and the correlation coefficients with the long-term performance indicators of the battery; the basic weights of each feature within the group are fine-tuned using the feature adjustment coefficients to obtain the fine-tuning weights of each feature within the group. For each factor group, calculate the sum of the fine-tuned weights of each feature within the group as the total fine-tuned weights of the group; calculate the scaling factor for the group, which is equal to the total base weights of the group divided by the total fine-tuned weights of the group; multiply the fine-tuned weights of each feature within the group by the scaling factor to obtain the final weights of each feature within the group. The calculation of the uniformity baseline score based on the feature values ​​and their final weights after feature processing includes: Calculate the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each feature and their corresponding final weights, and then subtract the sum of the products from 1 to obtain the uniformity base score.

9. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 8, characterized in that, The process of normalizing the basic weights of each risk characteristic within the set includes: Calculate the normalized weight of each feature in the risk feature set of normal infiltration morphology and the risk feature set of adverse infiltration morphology respectively. The normalized weight of each morphological risk feature is equal to the basic weight of that feature divided by the sum of the basic weights of all morphological risk features in the set. The calculation of the comprehensive physical health score includes: Calculate the risk score for normal morphology and the risk score for adverse morphology. The risk score for normal morphology is the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each morphological risk feature in the set of normal infiltration morphological risk features and their corresponding normalized weights. The risk score for adverse morphology is the sum of the products of the hyperbolic tangent normalized eigenvalues ​​of each morphological risk feature in the set of adverse infiltration morphological risk features and their corresponding normalized weights. Calculate the overall morphological health score, where the overall morphological health score is: ,in, , These are the preset weights for the risk scores of normal and adverse states, respectively.

1.

10. The method for evaluating the wetting uniformity of a liquid electrolyte battery according to claim 9, characterized in that, The generation of a comprehensive grade based on a uniform baseline score and a comprehensive morphological health score includes: The uniformity base score and the comprehensive morphological health score are respectively cropped to the [0,1] interval to obtain the uniformity score and the morphological health score; The overall rating will be determined in the following order: If the uniformity score is less than the third threshold for uniformity or the morphological health score is less than the third threshold for morphological health, it is deemed unqualified. Otherwise, if the uniformity score is greater than or equal to the first threshold of uniformity and the morphological health score is greater than or equal to the first threshold of morphological health, it is judged as excellent. Otherwise, if the uniformity score is greater than or equal to the second threshold for uniformity and the morphological health score is greater than or equal to the second threshold for morphological health, then it is considered qualified. Otherwise, it is determined that optimization is needed; Among them, the first threshold for uniformity > the second threshold for uniformity > the third threshold for uniformity, and the first threshold for morphological health > the second threshold for morphological health > the third threshold for morphological health.

Images