Power battery internal structure damage detection method, system, equipment and medium

Abstract

The invention relates to a power battery internal structure damage detection method, system and device and a medium. The method comprises the following steps: firstly, acquiring a battery internal medium interface reflection signal, and extracting first characteristic data containing a reflection wave amplitude value and frequency distribution; charge and discharge vibration signals are synchronously collected, and second characteristic data containing characteristic frequency and amplitude modulation are extracted through wavelet transform; and generating comprehensive feature data containing ultrasonic and vibration features by using a pre-trained neural network fusion algorithm, inputting the comprehensive feature data into a damage evaluation model to position a damage position, and generating region description. And finally, in combination with the region description, the quantized numerical value and the matching degree with a preset feature library, identifying a damage type and generating a structured detection report. According to the method, accurate positioning of the damage position, quantitative evaluation of the degree and accurate identification of the type are realized, the problems of insufficient positioning precision and low type identification accuracy in a traditional detection method are solved, the safety of the power battery is improved, and the service life of the power battery is prolonged.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault diagnosis, and particularly relates to a power battery internal structure damage detection method, system, device and medium. BACKGROUND

[0002] With the rapid development of the new energy automobile industry, the safety and reliability of power batteries as core energy storage components directly affect the running safety of the whole vehicle. Under the working conditions of long-term charge-discharge cycle, high-low temperature environment alternation and mechanical vibration, the internal structure of the power battery is prone to damage such as separator breakage, pole piece shedding and electrolyte leakage. If such damage is not detected in time, it may cause serious problems such as thermal runaway and capacity attenuation. At present, power battery damage detection mainly relies on a single physical field signal (such as ultrasonic detection or vibration monitoring), which has limitations such as one-sided characteristic information, insufficient damage positioning accuracy, low type recognition accuracy, and is difficult to meet the early warning and accurate diagnosis needs of internal structure damage under complex working conditions. SUMMARY

[0003] Therefore, it is necessary to provide a power battery internal structure damage detection method, system, device and medium which can realize accurate positioning of damage position, quantitative evaluation of damage degree and accurate identification of damage type, and help to improve the safety and service life of power batteries.

[0004] In a first aspect, the application provides a power battery internal structure damage detection method, comprising:

[0005] Obtaining the reflection signals of different medium interfaces in the power battery, and extracting first characteristic data including reflection amplitude and frequency distribution; collecting vibration signals in the charge-discharge process, and extracting second characteristic data including characteristic frequency and amplitude modulation after wavelet transform decomposition and reconstruction.

[0006] Using a pre-trained neural network fusion algorithm to fuse and calculate the first characteristic data and the second characteristic data, to generate comprehensive characteristic data containing multi-dimensional damage features; the multi-dimensional damage features include ultrasonic features and vibration features reflecting the abnormality of the internal structure of the battery.

[0007] Inputting the comprehensive characteristic data into a pre-trained damage evaluation algorithm model to position and analyze the abnormal values in the characteristic data and determine the damage position, and extracting spatial geometric features to generate a damage area description.

[0008] Combining the damage area description with the matching degree between the quantitative values related to the damage degree of the comprehensive characteristic data and each template in the preset damage type feature library to identify the specific type of damage, and generating a detection report containing a structured field.

[0009] In one of the embodiments, a pre-trained neural network fusion algorithm is used to fuse the first feature data and the second feature data to generate comprehensive feature data containing multi-dimensional damage features, including:

[0010] The first feature data and the second feature data are integrated to construct an original feature set containing ultrasonic features and vibration features; the ultrasonic features correspond to the physical interface characteristics of the internal structure of the battery; the vibration features correspond to the dynamic response characteristics under the working state of the battery.

[0011] A pre-trained dual-channel neural network fusion algorithm is used to perform spatial dimension matching on the ultrasonic and vibration features in the original feature set, calculate the key damage feature weight through the attention mechanism, and output preliminary fusion feature data.

[0012] The preliminary fusion feature data is analyzed through time-frequency domain joint analysis to mine transient damage response features, and a damage feature set containing multi-dimensional damage features is constructed by combining spatial distribution features; the damage feature set includes damage response intensity, distribution range, and dynamic change trend.

[0013] Based on the damage feature set, redundant features are removed and key features are enhanced to generate dimensionally optimized comprehensive feature data.

[0014] In one of the embodiments, the key damage feature weight is calculated by the following formula:

[0015]

[0016] where ω i represents the weight of the i-th key damage feature, s i represents the importance score generated by the self-attention mechanism for the i-th feature, τ represents the temperature coefficient, the value range is (0, 1], n represents the total number of features, I(f i ,f k ) represents the mutual information of features f i and f k , ||f i -f k ||2 represents the Euclidean distance of the feature vector, and σ represents the Gaussian kernel bandwidth parameter.

[0017] In one of the embodiments, the comprehensive feature data is input into a pre-trained damage evaluation algorithm model to perform positioning analysis on abnormal values in the feature data to determine the damage location, extract spatial geometric features to generate damage area description, including:

[0018] The feature vector is extracted from the dimensionally optimized comprehensive feature data; the feature vector contains time series parameters and spatial distribution parameters related to damage; the time series parameters include the dynamic change value of the damage feature with the charge and discharge cycle; the spatial distribution parameters include the coordinate distribution value of the feature abnormal region.

[0019] The feature vector is input into a pre-trained damage assessment algorithm model to detect the values deviating from the normal feature distribution range in the vector, and an outlier distribution set containing abnormal feature values and their corresponding indexes is output; the damage assessment algorithm model uses an isolation forest algorithm to calculate the isolation path length of the feature vector in the hyperplane to determine the abnormality.

[0020] Based on the outlier distribution set, a density clustering algorithm is used for spatial clustering analysis of outliers, and the damage location corresponding to the outliers is located according to the clustering center coordinates to generate a damage coordinate set composed of three-dimensional coordinate points.

[0021] The spatial geometric features of the damage location are extracted from the damage coordinate set to generate a damage area description containing damage area boundary coordinates, area, and spatial form; the spatial geometric features include the aggregation range and center distance of the coordinate points and the area contour parameters.

[0022] In one embodiment, the specific type of damage is identified by combining the damage area description with the quantitative values related to the damage degree of the comprehensive feature data and the matching degree of each template in the preset damage type feature library, and a detection report containing structured fields is generated, including:

[0023] Based on the damage area description, a principal component analysis algorithm is used for dimension reduction processing to extract key feature dimensions and generate a standardized feature vector; the damage area description contains damage area boundary coordinates, area, spatial form, and geometric feature parameters.

[0024] By calculating the distribution density and gradient change rate of each dimension parameter in the standardized feature vector, distribution feature data representing the damage diffusion trend and aggregation form are obtained; the standardized feature vector contains area contour complexity, spatial distribution entropy, and geometric center offset.

[0025] Extract the damage feature mode from the distribution feature data and the comprehensive feature data; the damage feature mode includes diffusion rate, aggregation density peak, ultrasonic reflection wave distortion rate, and vibration signal harmonic distortion degree.

[0026] Calculate the cosine similarity of the damage feature mode and each template in the preset damage type feature library, and determine the type corresponding to the template with the highest similarity and exceeding the matching threshold as the specific type of damage; the damage type feature library contains feature templates of typical damages such as diaphragm damage, pole shedding, and electrolyte leakage.

[0027] If the cosine similarity is higher than the matching threshold, the detection report containing the structured fields is generated by integrating the three-dimensional coordinate range of the damage location, the severity level, and the matching degree value of the type; if it is lower than the matching threshold, the secondary feature mode extraction and matching process is triggered until the matching threshold is met; the matching threshold is set based on the accuracy verification of historical detection data; the structured fields include the location parameter, the level identifier, the matching degree score, and the type determination result.

[0028] In one of the embodiments, the distribution density and the gradient change rate of each dimension parameter in the standardized feature vector are calculated by the following formula:

[0029]

[0030] wherein, p j represents the distribution density of the jth dimension parameter in the standardized feature vector, represents the average value of the jth dimension parameter in the damage area, and V represents the three-dimensional volume of the damage area, which is calculated by spatial integration based on the boundary coordinates, represents the gradient change rate of the jth dimension parameter at a certain point (x, y, z) in space, and f j (x, y, z) represents the feature value of the dimension parameter at the point (x, y, z) in space, and the partial derivative is calculated by the three-dimensional difference method, taking the difference value of the feature value of the point (x, y, z) and the adjacent sampling point divided by the spatial distance.

[0031] In one of the embodiments, the damage feature mode in the distribution feature data and the comprehensive feature data is extracted, including:

[0032] Based on the distribution feature data, the volume change amount of the damage area in the continuous charge and discharge cycle is calculated to obtain the diffusion rate representing the damage expansion speed; the volume change amount is obtained by calculating the volume difference of the boundary coordinates of the adjacent cycles.

[0033] The distribution density of each dimension parameter in the distribution feature data is subjected to spatial interpolation processing to generate a three-dimensional density field, and the maximum value point in the density field is located by the extreme value detection algorithm, and the density value corresponding to the maximum value point is determined as the aggregation density peak value.

[0034] Based on the comprehensive feature data, the frequency distribution data in the ultrasonic feature is subjected to Fourier transform, the ratio of the energy sum of the abnormal frequency components to the total signal energy is calculated, and the ultrasonic reflection wave distortion rate is obtained; the abnormal frequency components are the frequencies deviating from the normal frequency band.

[0035] According to the comprehensive feature data, the time domain signal of the vibration feature is subjected to harmonic analysis, the fundamental wave and the high-order harmonic components are separated, the proportion of the amplitude sum of all high-order harmonics to the fundamental wave amplitude is calculated to obtain the vibration signal harmonic distortion degree.

[0036] The damage characteristic mode is obtained by combining the diffusion rate, the aggregation density peak value, the ultrasonic reflection wave distortion rate and the vibration signal harmonic distortion degree.

[0037] In a second aspect, the application further provides a power battery internal structure damage detection system, which comprises:

[0038] The signal acquisition module is configured to acquire reflection signals of different medium interfaces in the power battery, and extract first characteristic data including reflection amplitude and frequency distribution; and acquire vibration signals in the charging and discharging process, and extract second characteristic data including characteristic frequency and amplitude modulation after wavelet transform decomposition and reconstruction.

[0039] The characteristic fusion module is configured to perform fusion calculation on the first characteristic data and the second characteristic data by using a pre-trained neural network fusion algorithm, and generate comprehensive characteristic data containing multi-dimensional damage characteristics; the multi-dimensional damage characteristics include ultrasonic characteristics and vibration characteristics reflecting the abnormality of the internal structure of the battery.

[0040] The damage positioning module is configured to input the comprehensive characteristic data into a pre-trained damage evaluation algorithm model to perform positioning analysis on abnormal values in the characteristic data and determine the damage position, extract spatial geometric characteristics and generate a damage area description.

[0041] The evaluation and reporting module is configured to identify the specific type of damage by combining the damage area description, the quantitative value related to the damage degree of the comprehensive characteristic data and the matching degree of each template in the pre-set damage type characteristic library, and generate a detection report containing a structured field.

[0042] In a third aspect, the application further provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the foregoing method when executing the computer program.

[0043] In a fourth aspect, the application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the foregoing method.

[0044] The power battery internal structure damage detection method, system, computer device and storage medium, first, the reflection signals of different medium interfaces in the power battery are acquired, the first characteristic data containing the reflection amplitude and frequency distribution are extracted, and the vibration signals in the charging and discharging process are collected, and the second characteristic data containing the characteristic frequency and amplitude modulation are extracted after wavelet transform decomposition and reconstruction; then, the first characteristic data and the second characteristic data are fused and calculated by using the pre-trained neural network fusion algorithm, to generate comprehensive characteristic data containing multi-dimensional damage characteristics of ultrasonic characteristics and vibration characteristics reflecting the internal structure abnormalities of the battery; then, the comprehensive characteristic data is input into the pre-trained damage evaluation algorithm model, to position and analyze the abnormal values in the characteristic data to determine the damage position, and to extract the spatial geometric characteristics to generate the damage area description; finally, the specific type of damage is identified by combining the damage area description, the quantitative values related to the damage degree in the comprehensive characteristic data, and the matching degrees of each template in the preset damage type characteristic library, to generate a detection report containing a structured field. The method fuses the ultrasonic and vibration multi-source characteristic data, makes up for the limitation of the feature information in the single signal detection, improves the integrity and effectiveness of the damage characteristics, realizes the accurate positioning of the damage position, the quantitative evaluation of the degree and the accurate identification of the type by means of the neural network fusion and the damage evaluation algorithm model, solves the problems of insufficient positioning accuracy and low type identification accuracy in the traditional detection method, provides reliable technical support for the early warning and accurate diagnosis of the internal structure damage of the power battery, and helps to improve the safety and service life of the power battery. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating any creative labor.

[0046] Figure 1 The flowchart of the power battery internal structure damage detection method provided by the embodiment of the present application;

[0047] Figure 2 The structural block diagram of the power battery internal structure damage detection system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0049] In one of the embodiments, asFigure 1 As shown, the application provides a method for detecting internal structural damage of a power battery, which can include the following steps:

[0050] In step S101, the reflection signals of different medium interfaces inside the power battery are acquired, and first characteristic data including reflection amplitude and frequency distribution are extracted; vibration signals during the charging and discharging process are collected, and second characteristic data including characteristic frequency and amplitude modulation are extracted after wavelet transform decomposition and reconstruction.

[0051] Specifically, the reflection signals of different medium interfaces (such as the interface between the pole piece and the diaphragm, the interface between the electrolyte and the shell, etc.) inside the power battery are acquired by an ultrasonic detection device, and the first characteristic data reflecting the physical characteristics of the interface are extracted from the reflection signals, including the amplitude parameter (characterizing the interface reflection intensity) and the frequency distribution information (reflecting the interface structural integrity) of the reflection wave; at the same time, the vibration signals of the battery during the charging and discharging cycle process are collected by a vibration sensor, and the vibration signals are decomposed by wavelet transform to separate different frequency components, and the second characteristic data are extracted after signal reconstruction, including the characteristic frequency (corresponding to a specific structural vibration mode) and the amplitude modulation parameter (reflecting the amplitude variation law of the vibration signal) related to the dynamic response of the battery.

[0052] In step S102, the first characteristic data and the second characteristic data are fused and calculated by using a pre-trained neural network fusion algorithm to generate comprehensive characteristic data containing multi-dimensional damage features; the multi-dimensional damage features include ultrasonic features and vibration features reflecting the abnormality of the internal structure of the battery.

[0053] The extracted first characteristic data (ultrasonic features) and second characteristic data (vibration features) are input into a pre-trained neural network fusion algorithm, which uses a double-channel architecture to respectively map and align the dimensions of the two types of features, and realizes the associated fusion of ultrasonic features and vibration features through spatial dimension matching; during the fusion process, the attention mechanism is introduced to strengthen the weight of key damage features, and the feature information sensitive to damage is highlighted, and finally the comprehensive characteristic data containing multi-dimensional damage features are output. Among them, the multi-dimensional damage features cover ultrasonic features (such as interface reflection abnormalities) and vibration features (such as dynamic response distortion) reflecting the abnormality of the internal structure of the battery.

[0054] In step S103, the comprehensive characteristic data are input into a pre-trained damage evaluation algorithm model to locate and analyze the abnormal values in the characteristic data to determine the damage position, and the spatial geometric features are extracted to generate a damage area description.

[0055] Specifically, the generated comprehensive feature data is input into a pre-trained damage assessment algorithm model. The model identifies abnormal values deviating from the normal distribution in the feature data through an outlier detection algorithm (such as the Isolation Forest algorithm), and preliminarily locates the potential damage area in combination with the spatial index information corresponding to the abnormal values. Then, a density clustering algorithm is used to perform spatial clustering analysis on the abnormal values, and the specific position of the damage is determined by calculating the cluster center coordinates to generate a damage coordinate set composed of three-dimensional coordinate points. Finally, the spatial geometric features are extracted from the damage coordinate set, including the aggregation range, center distance, and area contour parameters of the coordinate points, and based on these features, a damage area description containing the boundary coordinates, area, and spatial form of the damage area is generated.

[0056] In step S104, the specific type of damage is identified by combining the damage area description with the matching degree of the quantitative values related to the damage degree in the comprehensive feature data and each template in the preset damage type feature library, and a detection report containing structured fields is generated.

[0057] Based on the damage area description, the quantitative values related to the damage degree in the comprehensive feature data (such as the damage area, abnormal amplitude of feature parameters, etc.) are combined to judge the severity level of the damage by comparing with the preset threshold. At the same time, the damage feature pattern (such as the diffusion rate, ultrasonic distortion rate, etc.) in the distribution feature data and the comprehensive feature data is extracted, the cosine similarity of this pattern with each template in the preset damage type feature library is calculated, and the type corresponding to the template with the highest similarity and exceeding the matching threshold is determined as the specific type of damage. Finally, the three-dimensional coordinate range of the damage location, the severity level, the type matching degree, and other information are integrated to generate a detection report containing position parameters, level identification, matching degree score, and type determination results, etc. structured fields.

[0058] The power battery internal structure damage detection method first acquires reflection signals of different medium interfaces in the power battery, extracts first characteristic data containing reflection amplitude and frequency distribution, simultaneously collects vibration signals in the charging and discharging process, extracts second characteristic data containing characteristic frequency and amplitude modulation after wavelet transform decomposition and reconstruction; then a pre-trained neural network fusion algorithm is used to fuse and calculate the first characteristic data and the second characteristic data, to generate comprehensive characteristic data containing multi-dimensional damage characteristics reflecting the ultrasonic characteristics and vibration characteristics of the battery internal structure abnormalities; then the comprehensive characteristic data is input into a pre-trained damage evaluation algorithm model to position and analyze the abnormal values in the characteristic data to determine the damage position, and to extract spatial geometric characteristics to generate a damage area description; finally, the damage area description, the quantitative values related to the damage degree in the comprehensive characteristic data, and the matching degrees with each template in the preset damage type characteristic library are combined to identify the specific type of damage and generate a detection report containing a structured field. This method fuses ultrasonic and vibration multi-source characteristic data, makes up for the limitations of feature information in single signal detection, improves the integrity and effectiveness of damage characteristics, and realizes accurate positioning of damage position, quantitative evaluation of damage degree and accurate identification of damage type by means of neural network fusion and damage evaluation algorithm model, solves the problems of insufficient positioning accuracy and low type recognition accuracy in traditional detection methods, provides reliable technical support for early warning and accurate diagnosis of power battery internal structure damage, and helps to improve the safety and service life of power batteries.

[0059] In one embodiment, the pre-trained neural network fusion algorithm is used to fuse and calculate the first characteristic data and the second characteristic data to generate comprehensive characteristic data containing multi-dimensional damage characteristics, which can include the following steps:

[0060] Step S201, integrate the first characteristic data and the second characteristic data to construct an original feature set containing ultrasonic characteristics and vibration characteristics; the ultrasonic characteristics correspond to the physical interface characteristics of the battery internal structure; the vibration characteristics correspond to the dynamic response characteristics of the battery under working state.

[0061] Preferably, the extracted first characteristic data (ultrasonic characteristics) and second characteristic data (vibration characteristics) are integrated to form an original feature set containing two types of characteristics. The ultrasonic characteristics are derived from the reflection signals of different medium interfaces in the battery, which directly correspond to the physical interface characteristics (such as interface integrity, medium density change, etc.) of the battery internal structure; the vibration characteristics are derived from the processing results of the vibration signals in the charging and discharging process, which correspond to the dynamic response characteristics (such as structural vibration mode, mechanical property change, etc.) of the battery under working state.

[0062] Step S202, the pre-trained dual-channel neural network fusion algorithm is adopted to perform spatial dimension matching on the ultrasonic and vibration features in the original feature set, the attention mechanism is calculated to obtain the weight of the key damage feature, and the preliminary fusion feature data is output.

[0063] Further, the pre-trained dual-channel neural network fusion algorithm is adopted to process the ultrasonic features and vibration features in the original feature set. The algorithm performs feature mapping on the two types of features through two parallel channels, and realizes the alignment of the features in the spatial scale through the spatial dimension matching mechanism. The attention mechanism is introduced in the fusion process, the weight of the key damage feature is calculated according to the sensitivity of the feature to the damage, the feature with a larger contribution to the damage identification is given a higher weight, and finally the preliminary fusion feature data integrating the key information of the ultrasonic and vibration is output.

[0064] Step S203, the preliminary fusion feature data is analyzed through time-frequency domain joint analysis to mine the transient damage response features, and the spatial distribution features are combined to construct a damage feature set containing multi-dimensional damage features. The damage feature set includes damage response intensity, distribution range and dynamic change trend.

[0065] The output preliminary fusion feature data is analyzed through time-frequency domain joint analysis, the transient response change (such as signal mutation, amplitude fluctuation) when the damage occurs is captured through time domain analysis, the frequency characteristics (such as abnormal frequency component, frequency spectrum shift) related to the damage are extracted through frequency domain analysis, so as to mine the transient damage response features; meanwhile, the spatial distribution information (such as the position and range of the feature abnormal area) of the features is combined to construct a damage feature set containing multi-dimensional damage features, which specifically includes damage response intensity (feature abnormal amplitude), distribution range (abnormal area spatial span) and dynamic change trend (evolution law of the feature with time).

[0066] Step S204, based on the damage feature set, the redundant features are removed and the key features are enhanced to generate dimension-optimized comprehensive feature data.

[0067] Based on the constructed damage feature set, the feature selection algorithm (such as variance analysis, mutual information filtering) is adopted to remove the redundant features (such as highly correlated or non-contributing features to damage identification), reduce the data dimension and noise interference; at the same time, the feature enhancement technology (such as feature scaling, principal component analysis) is adopted to improve the representation ability of the key features, highlight the discrimination of the effective features, and finally generate dimension-optimized comprehensive feature data.

[0068] Specifically, first, by integrating the first feature data (ultrasonic features) and the second feature data (vibration features), an original feature set containing two types of features is constructed, wherein the ultrasonic features correspond to the physical interface characteristics of the internal structure of the battery, and the vibration features correspond to the dynamic response characteristics under the working state of the battery, providing a complete original data basis for subsequent fusion; then, a pre-trained dual-channel neural network fusion algorithm is used to match the spatial dimensions of the ultrasonic and vibration features in the original feature set, the weight of the key damage feature is calculated through the attention mechanism, the feature information sensitive to damage is strengthened, and the preliminary fusion feature data is output; the preliminary fusion feature data is analyzed in time domain and frequency domain, the transient damage response features are mined, and a multi-dimensional damage feature set containing damage response intensity, distribution range and dynamic change trend is constructed in combination with the spatial distribution features; finally, based on the damage feature set, redundant features are removed and key features are enhanced, irrelevant or repetitive information is removed, the representation ability of effective features is improved, and dimension-optimized comprehensive feature data is generated.

[0069] In this embodiment, through systematic integration and optimization of multi-source features, the advantages of ultrasonic features and vibration features are complementary, the problem of one-sidedness of single feature information is solved, and the integrity and distinguishability of damage features are improved; with the help of dual-channel neural network fusion and attention mechanism, the weight of key damage features is effectively strengthened, and noise interference is reduced; through time domain-frequency domain joint analysis and dimension optimization, the dynamic and spatial characteristics of damage are further mined, which helps to improve the accuracy and reliability of internal structure damage detection of power batteries.

[0070] In one of the embodiments, the weight of the key damage feature can be calculated by the following formula:

[0071]

[0072] wherein ω i represents the weight of the i-th key damage feature, s i represents the importance score generated by the self-attention mechanism for the i-th feature, τ represents the temperature coefficient, the value range is (0, 1], n represents the total number of features, I(f i ,f k ) represents the mutual information of features f i and f k , ||f i -f k ||2 represents the Euclidean distance of the feature vector, and σ represents the Gaussian kernel bandwidth parameter.

[0073] The embodiment introduces importance scores generated by self-attention mechanisms, inter-feature mutual information, and Euclidean distance, and dynamically adjusts temperature coefficients and Gaussian kernel bandwidth parameters, so as to accurately quantify the contribution of different features to damage identification. The exponential function is used to strengthen the weight proportion of high importance features, and the relevance and distance information between features are used to filter redundant interference, so that the key damage features obtain more reasonable weight distribution in the fusion process, effectively improving the pertinence and accuracy of feature fusion, providing more discriminative input features for subsequent damage evaluation models, and helping to improve the accuracy and reliability of the internal structure damage detection of the power battery.

[0074] In one of the embodiments, the comprehensive feature data is input into a pre-trained damage evaluation algorithm model to analyze and determine the damage position of the abnormal value in the feature data, and to extract the spatial geometric features to generate the damage area description, which can include the following steps:

[0075] Step S301, extracting a feature vector from the dimensionally optimized comprehensive feature data; the feature vector contains time series parameters and spatial distribution parameters related to damage; the time series parameters include dynamic change values of damage features with charging and discharging cycles; the spatial distribution parameters include coordinate distribution values of feature abnormal regions.

[0076] Preferably, from the dimensionally optimized comprehensive feature data, a feature vector is screened and extracted, which contains two types of core parameters: one is the time series parameter, which records the dynamic change value of the damage feature with the charging and discharging cycle (such as the feature amplitude fluctuation and frequency offset under different cycles), which reflects the evolution trend of the damage; the second is the spatial distribution parameter, which covers the coordinate distribution value of the feature abnormal region (such as the coordinate set of the feature parameter abnormal point in the three-dimensional space), which represents the spatial position information of the damage. Through this step, the high-dimensional comprehensive feature data is converted into a structured feature vector.

[0077] Step S302, inputting the feature vector into a pre-trained damage evaluation algorithm model to detect the values deviating from the normal feature distribution range in the vector, and outputting an abnormal value distribution set containing abnormal feature values and their corresponding indexes; the damage evaluation algorithm model uses the isolation forest algorithm to calculate the isolation path length of the feature vector in the hyperplane to determine the abnormality.

[0078] Illustratively, the damage evaluation algorithm model constructs a hyperplane in a high-dimensional space, calculates the isolation path length (i.e. the number of hyperplane cuts required for the sample to be isolated) of each feature vector sample, and the vectors with path length significantly deviating from the normal sample distribution range are determined as abnormal. The model outputs an abnormal value distribution set containing abnormal feature values (specific parameter values deviating from the normal range) and their corresponding indexes (marking the position and associated coordinates of the abnormal value in the feature vector), realizing the preliminary screening and positioning marking of the potential damage features.

[0079] Step S303, based on the abnormal value distribution set, a density clustering algorithm is used for spatial clustering analysis of abnormal values, the abnormal value corresponding damage position is located according to the clustering center coordinates, and a damage coordinate set composed of three-dimensional coordinate points is generated.

[0080] By setting the neighborhood radius and the minimum number of contained points, abnormal values with similar spatial positions are aggregated into clustering clusters, each clustering cluster represents a potential damage area; the center coordinates (clustering center) of each clustering cluster are calculated to locate the damage position corresponding to the abnormal value, and the three-dimensional coordinates of all abnormal points in the clustering cluster are summarized to generate a damage coordinate set composed of three-dimensional coordinate points. This step eliminates the interference of isolated noise points through clustering, and realizes the accurate spatial positioning of the damage position.

[0081] Step S304, the spatial geometric features of the damage position are extracted from the damage coordinate set, and a damage area description containing the damage area boundary coordinates, area and spatial form is generated; the spatial geometric features include the aggregation range and center distance of the coordinate points and the area contour parameters.

[0082] Spatial geometric features: including the aggregation range of coordinate points (spatial boundary determined by coordinate extreme value), center distance (distance between clustering center and battery geometric center) and area contour parameters (such as boundary point curvature, contour complexity); based on these features, the boundary coordinates (vertex coordinates of spatial boundary), area (surface area of boundary enclosed region) and spatial form (such as point, line or surface distribution) of the damage area are further calculated, and finally a structured damage area description is generated.

[0083] Specifically, first, a feature vector is extracted from the dimensionally optimized comprehensive feature data, which contains time series parameters related to damage (such as dynamic change value of damage feature with charge and discharge cycle) and spatial distribution parameters (such as coordinate distribution value of feature abnormal area); then the feature vector is input into the pre-trained damage evaluation algorithm model, the model uses the isolation forest algorithm to calculate the isolation path length of the feature vector in the hyperplane, detects the values deviating from the normal feature distribution range in the vector, and outputs an abnormal value distribution set containing abnormal feature values and their corresponding indexes; based on the abnormal value distribution set, a density clustering algorithm is used for spatial clustering analysis of abnormal values, the abnormal value corresponding damage position is located according to the clustering center coordinates, and a damage coordinate set composed of three-dimensional coordinate points is generated; finally, the spatial geometric features (including the aggregation range of coordinate points, the center distance and the area contour parameters) are extracted from the damage coordinate set, and a damage area description containing the damage area boundary coordinates, the area and the spatial form is generated.

[0084] The embodiment solves the problem of poor adaptability to complex feature distribution in traditional anomaly detection by extracting multi-dimensional feature vectors and combining the isolated forest algorithm to accurately detect abnormal values; the spatial clustering of abnormal values is realized by means of the density clustering algorithm, the accurate positioning of the damage position is realized, and the positioning deviation caused by the misjudgment of a single abnormal point is avoided; the spatial distribution characteristics of the damage are quantified completely by extracting the spatial geometric features to generate a structured damage area description. Through multi-step data processing and algorithm cooperation, the objectivity and reliability of the internal damage detection of the power battery are improved.

[0085] In one embodiment, the specific type of damage is identified by combining the damage area description with the quantitative values of the comprehensive feature data related to the damage degree and the matching degree of each template in the preset damage type feature library, and a detection report containing a structured field is generated, which can include the following steps:

[0086] Step S401, based on the damage area description, the principal component analysis algorithm is used for dimension reduction processing, the key feature dimensions are extracted, and a standardized feature vector is generated; the damage area description includes damage area boundary coordinates, area, spatial form and geometric feature parameters.

[0087] The dimension reduction processing extracts the principal components with the largest variance contribution in the data set, eliminates redundant dimensions, and retains the key feature dimensions that have the most significant impact on damage features; then the key features are standardized (such as mean normalization and standard deviation scaling), and a standardized feature vector containing area contour complexity (representing the irregularity of the boundary), spatial distribution entropy (reflecting the uniformity of the coordinate point distribution), and geometric center offset (describing the offset distance of the damage center from the battery reference center) is generated.

[0088] Step S402, by calculating the distribution density and gradient change rate of each dimension parameter in the standardized feature vector, distribution feature data representing the damage diffusion trend and aggregation form are obtained; the standardized feature vector includes area contour complexity, spatial distribution entropy and geometric center offset.

[0089] Preferably, the distribution density is calculated by the ratio of the average value of the feature parameter in the damage area to the three-dimensional volume of the area, representing the aggregation intensity of the feature parameter; the gradient change rate is calculated by the three-dimensional difference method to calculate the change rate of the feature value of each point in space (i.e. the ratio of the difference value of the feature value of adjacent sampling points to the spatial distance), reflecting the degree of change of the feature parameter with position. The final distribution feature data comprehensively represent the diffusion trend (gradient change direction and rate) and aggregation form (density distribution rule) of the damage.

[0090] Step S403, extract the damage feature mode in the distribution feature data and the comprehensive feature data; the damage feature mode includes diffusion rate, aggregation density peak, ultrasonic reflection wave distortion rate and vibration signal harmonic distortion degree.

[0091] The diffusion rate (reflecting the damage expansion speed based on the ratio of the damage volume change amount in a continuous period to the time interval) and the aggregation density peak value (the maximum density value determined by the three-dimensional density field extreme value detection, representing the intensity of the most concentrated area of the damage) are extracted from the distribution feature data; at the same time, the ultrasonic reflection wave distortion rate (the ratio of the abnormal frequency energy to the total energy, reflecting the interface structure abnormality) and the vibration signal harmonic distortion degree (the proportion of the sum of the amplitudes of the high-order harmonics to the amplitude of the fundamental wave, representing the dynamic response abnormality) are extracted from the comprehensive feature data.

[0092] In step S404, the cosine similarity of the damage feature mode and each template in the preset damage type feature library is calculated, and the type corresponding to the template with the highest similarity and exceeding the matching threshold is determined as the specific type of the damage; the damage type feature library includes the feature templates of typical damages such as diaphragm breakage, pole shedding and electrolyte leakage.

[0093] Further, the feature library includes the standard feature vector templates of typical damages such as diaphragm breakage, pole shedding and electrolyte leakage, the similarity is calculated by the dot product of the feature mode vector and the template vector divided by the product of the lengths, and the value closer to 1 indicates a higher matching degree. The damage type corresponding to the template with the highest similarity and exceeding the preset matching threshold is determined as the specific type of the damage currently detected.

[0094] In step S405, if the cosine similarity is higher than the matching threshold, the three-dimensional coordinate range of the damage position, the severity level and the type matching degree value are integrated to generate a detection report including a structured field; if the cosine similarity is lower than the matching threshold, the secondary feature mode extraction and matching process is triggered until the matching threshold is met; the matching threshold is set based on the accuracy verification of the historical detection data; the structured field includes the position parameters, the level identification, the matching degree score and the type determination result.

[0095] Preferably, if the calculated cosine similarity is higher than the matching threshold (the threshold is set based on the accuracy verification of the historical detection data to ensure the recognition reliability), the three-dimensional coordinate range of the damage position, the severity level (based on the comparison result of the quantitative value and the threshold) and the type matching degree value are integrated to generate a structured detection report including the position parameters, the level identification, the matching degree score and the type determination result; if the similarity is lower than the threshold, the secondary feature mode extraction and matching process is triggered, and the process returns to step S403 (such as supplementing the feature dimension or adjusting the feature weight), until the threshold requirement is met.

[0096] Specifically, first, based on the damage area description (including boundary coordinates, area, spatial form and geometric feature parameters), principal component analysis algorithm is used for dimension reduction processing, key feature dimensions are extracted, and a standardized feature vector containing area contour complexity, spatial distribution entropy and geometric center offset is generated; by calculating the distribution density and gradient change rate of each dimension parameter in the standardized feature vector, the distribution characteristic data representing the damage diffusion trend and aggregation form are obtained; then, from the distribution characteristic data and the comprehensive characteristic data, the damage feature mode containing diffusion rate, aggregation density peak value, ultrasonic reflection wave distortion rate and vibration signal harmonic distortion degree is extracted; the cosine similarity of the feature mode and each template in the preset damage type feature library (containing typical templates such as diaphragm damage, pole piece shedding and electrolyte leakage) is calculated, and the template type with the highest similarity and exceeding the matching threshold is determined as the specific type of damage. Finally, if the similarity meets the standard, the damage position three-dimensional coordinate range, severity level and matching degree value are integrated to generate a detection report containing position parameters, level identification and other structured fields; if it does not meet the standard, the secondary feature extraction and matching process is triggered until the threshold is met, and the matching threshold is set based on the historical detection data accuracy verification.

[0097] In this embodiment, principal component analysis is used to realize dimension reduction and standardization of damage features, reduce data redundancy and focus on key features; the distribution density and gradient change rate are calculated to accurately capture the damage diffusion and aggregation characteristics, providing a quantitative basis for type identification; combined with cosine similarity matching and dynamic threshold mechanism, accurate identification and report generation of damage type are realized. Through multi-dimensional feature fusion and algorithm cooperation, the limitations of single feature evaluation are solved.

[0098] In one of the embodiments, the distribution density and gradient change rate of each dimension parameter in the standardized feature vector can be calculated by the following formula:

[0099]

[0100] wherein, ρ j represents the distribution density of the jth dimension parameter in the standardized feature vector, represents the average value of the jth dimension parameter in the damage area, and V represents the three-dimensional volume of the damage area, which is calculated by spatial integration based on the boundary coordinates, represents the gradient change rate of the jth dimension parameter at a certain point (x, y, z) in space, and f j (x, y, z) represents the feature value of the dimension parameter at the point (x, y, z) in space, and the partial derivative is calculated by three-dimensional difference method, taking the difference value of the feature values of the point (x, y, z) and the adjacent sampling point divided by the spatial distance.

[0101] The embodiment accurately quantifies the aggregation strength of the characteristic parameters in the damage area by correlating the average value of the dimensional parameters with the three-dimensional volume of the damage area, and effectively captures the dynamic change trend of the parameters with the spatial position based on the three-dimensional difference method to calculate the change rate of the spatial point characteristic value. The volume of the damage area is obtained by spatial integration to ensure the accuracy of the density calculation, and the gradient change rate is quantified by the ratio of the difference value and the distance between adjacent sampling points, so that the distribution characteristic data can objectively reflect the diffusion trend and aggregation form of the damage, and provide accurate spatial characteristic quantification basis for subsequent damage characteristic pattern extraction and type recognition, which helps to improve the scientificity and reliability of the internal structure damage detection of the power battery.

[0102] In one of the embodiments, extracting the damage characteristic pattern in the distribution characteristic data and the comprehensive characteristic data can include the following steps:

[0103] Step S501, based on the distribution characteristic data, calculating the volume change amount of the damage area in the continuous charge and discharge cycle to obtain the diffusion rate representing the damage expansion speed; the volume change amount is obtained by calculating the volume difference of the boundary coordinates of the adjacent cycles.

[0104] Preferably, based on the damage area information recorded in the distribution characteristic data, the change of the damage area in the continuous charge and discharge cycle is tracked: the three-dimensional volumes of the damage areas in the corresponding cycles are calculated respectively through the boundary coordinates of the damage areas in the adjacent two cycles (such as through spatial integration or polygon volume formula calculation), and the difference between the two is the volume change amount; the volume change amount is divided by the time interval (i.e. the charge and discharge cycle length) of the two cycles, and the obtained value is the diffusion rate, which directly represents the expansion speed of the damage area with time and reflects the dynamic evolution trend of the damage.

[0105] Step S502, performing spatial interpolation processing on the distribution density of each dimensional parameter in the distribution characteristic data to generate a three-dimensional density field, and locating the maximum value point in the density field by an extreme value detection algorithm to determine that the density value corresponding to the maximum value point is the aggregation density peak value.

[0106] The spatial interpolation algorithm (such as Kriging interpolation, inverse distance weighted interpolation) is used to process the discrete density data to generate a continuous three-dimensional density field, and realize the smooth representation of the density in the spatial range; then the extreme value detection algorithm (such as gradient descent method, local maximum value search) is used to scan the three-dimensional density field to locate the point with the maximum value (maximum value point), and the density value corresponding to the point is the aggregation density peak value, which is used to quantify the strength of the most concentrated area of the damage characteristic parameters in the space and reflect the aggregation form characteristics of the damage.

[0107] Step S503, based on the comprehensive feature data, Fourier transform is performed on the frequency distribution data in the ultrasonic feature to calculate the ratio of the energy sum of abnormal frequency components to the total signal energy, and the ultrasonic reflection wave distortion rate is obtained. The abnormal frequency component is the frequency deviating from the normal frequency band.

[0108] Based on the ultrasonic feature in the comprehensive feature data, the frequency distribution data is extracted and Fourier transformed to convert the time domain signal into a frequency domain signal to obtain the complete frequency component distribution. By comparing with the preset normal frequency band (ultrasonic frequency range in the undamaged state), abnormal frequency components deviating from the normal frequency band are screened out. The energy sum of all abnormal frequency components (obtained by integrating the square of the amplitude of the frequency component) is calculated, and the ratio with the total energy of the ultrasonic signal (the total energy of all frequency components) is calculated to obtain the ultrasonic reflection wave distortion rate, which is used to represent the degree of ultrasonic signal distortion caused by the abnormal interface structure inside the battery.

[0109] Step S504, according to the comprehensive feature data, the time domain signal of the vibration feature is subjected to harmonic analysis to separate the fundamental wave and high-order harmonic components, and the vibration signal harmonic distortion degree is obtained by calculating the proportion of the amplitude sum of all high-order harmonics to the fundamental wave amplitude.

[0110] According to the vibration feature in the comprehensive feature data, the time domain vibration signal is subjected to harmonic analysis (such as Fourier series decomposition) to separate the fundamental wave component (main frequency component of vibration) and high-order harmonic component (integer multiple frequency component of the fundamental wave frequency) in the signal. The amplitude sum of all high-order harmonic components and the amplitude of the fundamental wave component are calculated respectively. The ratio of the amplitude sum of the high-order harmonics to the fundamental wave amplitude is calculated to obtain the vibration signal harmonic distortion degree, which is used to represent the degree of vibration signal waveform distortion caused by internal structure damage under the working state of the battery.

[0111] Step S505, combining the diffusion rate, the aggregation density peak value, the ultrasonic reflection wave distortion rate and the vibration signal harmonic distortion degree to obtain the damage feature mode.

[0112] The damage feature mode comprehensively quantifies the core characteristics of the internal structure damage of the power battery by fusing the dynamic evolution, spatial aggregation and multi-physical field signal abnormal characteristics of the damage.

[0113] Specifically, first, based on the distribution feature data, the diffusion rate is obtained by calculating the volume change of the damage area in the continuous charge and discharge cycle, wherein the volume change is determined by the volume difference of the boundary coordinates of adjacent cycles, representing the expansion speed of the damage; after the spatial interpolation processing of the distribution density of each dimension parameter in the distribution feature data is performed to generate a three-dimensional density field, the maximum value point of the density field is located by using the extreme value detection algorithm, and the corresponding density value is the aggregation density peak value, reflecting the aggregation intensity of the damage. At the same time, based on the comprehensive feature data, the ratio of the energy sum of the abnormal frequency component (the frequency deviating from the normal frequency band) to the total signal energy is calculated by performing Fourier transform on the frequency distribution data of the ultrasonic feature, to obtain the ultrasonic reflection wave distortion rate; after the time domain signal of the vibration feature is analyzed by harmonic analysis, the sum of the amplitudes of all high-order harmonic components and the proportion of the fundamental wave amplitude are calculated to obtain the harmonic distortion degree of the vibration signal. Finally, the diffusion rate, the aggregation density peak value, the ultrasonic reflection wave distortion rate and the vibration signal harmonic distortion degree are combined to form a complete damage feature mode.

[0114] The embodiment realizes comprehensive quantification of damage characteristics through multi-dimensional feature extraction: the diffusion rate and the aggregation density peak value represent the physical morphological changes of the damage from the angles of spatial evolution and aggregation intensity, and the ultrasonic reflection wave distortion rate and the vibration signal harmonic distortion degree reflect the internal structural damage characteristics from the angle of physical signal abnormality. The extraction methods of each feature are based on accurate quantitative calculation, ensuring the objectivity and reliability of the feature values, and avoiding the limitations of single feature representation. Through the integration of the damage feature mode formed by the four types of features, the dynamic expansion, spatial aggregation and signal abnormality characteristics of the damage are comprehensively covered, providing a comprehensive and accurate feature basis for subsequent damage type recognition and severity level determination, effectively improving the accuracy and integrity of the internal structural damage detection of the power battery.

[0115] In one of the embodiments, as shown in Figure 2 The application further provides a power battery internal structural damage detection system, which can include:

[0116] The signal acquisition module 601 is configured to acquire the reflection signals of different medium interfaces in the power battery, extract the first feature data including the reflection wave amplitude and frequency distribution, and acquire the vibration signal in the charge and discharge process, and extract the second feature data including the characteristic frequency and amplitude modulation after wavelet transform decomposition and reconstruction.

[0117] The feature fusion module 602 is configured to perform fusion calculation on the first feature data and the second feature data by using a pre-trained neural network fusion algorithm, and generate comprehensive feature data containing multi-dimensional damage features; the multi-dimensional damage features include ultrasonic features and vibration features reflecting the abnormality of the internal structure of the battery.

[0118] The damage positioning module 603 is configured to input the comprehensive feature data into a pre-trained damage evaluation algorithm model to perform positioning analysis on abnormal values in the feature data to determine a damage position, and extract spatial geometric features to generate a damage region description.

[0119] The evaluation and reporting module 604 is configured to identify a specific type of damage by combining the damage region description, quantitative values related to a damage degree in the comprehensive feature data, and matching degrees of each template in a preset damage type feature library, and generate a detection report containing a structured field.

[0120] The above power battery internal structure damage detection system, the signal acquisition module acquires the reflection signals of different medium interfaces in the power battery, extracts the first feature data containing the reflection amplitude and frequency distribution, and simultaneously acquires the vibration signals in the charging and discharging process, and extracts the second feature data containing the characteristic frequency and amplitude modulation after wavelet transform decomposition and reconstruction; the feature fusion module adopts a pre-trained neural network fusion algorithm to fuse and calculate the first feature data and the second feature data, and generates comprehensive feature data containing multi-dimensional damage features of ultrasonic features and vibration features reflecting the abnormalities of the battery internal structure; the damage positioning module inputs the comprehensive feature data into a pre-trained damage evaluation algorithm model to perform positioning analysis on abnormal values in the feature data to determine a damage position, and extracts spatial geometric features to generate a damage region description; the evaluation and reporting module combines the damage region description, quantitative values related to a damage degree in the comprehensive feature data, and matching degrees of each template in a preset damage type feature library to identify a specific type of damage, and generates a detection report containing a structured field.

[0121] The embodiment realizes accurate acquisition of multi-source feature data through the signal acquisition module, provides comprehensive input for damage detection; realizes complementary advantages of ultrasonic and vibration features through the feature fusion module, and improves the integrity and effectiveness of damage features; realizes accurate positioning of a damage position and regional quantitative description through the damage positioning module; and utilizes the evaluation and reporting module to complete damage type identification and structured report generation. The modules work cooperatively to form a complete detection process from signal acquisition to result output, solve the limitations of single signal detection, improve the accuracy, comprehensiveness and reliability of power battery internal structure damage detection, and provide strong technical support for battery safety management.

[0122] In one of the embodiments, the application also provides an internal damage detection scenario of a certain ternary lithium battery after 500 cycles of cyclic charging and discharging, which can include:

[0123] Firstly, the signal acquisition module obtains the reflection signals of the interfaces between the electrode-tab-membrane and electrolyte-housing through the ultrasonic probe, extracts the reflection amplitude (range 0-5V) and frequency distribution (1MHz-5MHz) as the first characteristic data; meanwhile, the vibration signals (sampling frequency 10kHz) during the 1C charge-discharge process are collected by the vibration sensor, and after wavelet transform decomposition and reconstruction, the characteristic frequency and amplitude modulation coefficient in the 100Hz-500Hz frequency band are extracted as the second characteristic data.

[0124] The feature fusion module calls the pre-trained dual-channel neural network fusion algorithm to match the spatial dimensions of the two types of features, calculates the ultrasonic reflection amplitude weight 0.62 and vibration characteristic frequency weight 0.38 through the attention mechanism, and outputs the preliminary fusion features containing damage response intensity and spatial distribution range; after joint time-frequency domain analysis and redundant feature elimination, the dimension-optimized comprehensive feature data is generated.

[0125] The damage positioning module inputs the comprehensive feature data into the Isolation Forest algorithm model, detects 3 characteristic abnormal values (deviating from the normal distribution threshold 1.8 times), locates the damage coordinate set as three-dimensional coordinate points (X: 52mm, Y: 38mm, Z: 2mm) through density clustering analysis, extracts the spatial geometric features to generate the damage area description: boundary coordinate range 50-55mm (X-axis), 35-42mm (Y-axis), area about 28mm 2 , irregular sheet distribution.

[0126] The evaluation and reporting module generates a standardized feature vector based on the damage area description, calculates the diffusion rate 0.03mm 3 / cycle, peak aggregation density 0.85, constructs the damage feature mode combined with the ultrasonic reflection wave distortion rate 12% and the vibration signal harmonic distortion 8%, matches with the pre-set feature library, and the cosine similarity of the membrane damage template reaches 0.91 (threshold 0.75), finally generates the detection report, the structured field shows the three-dimensional coordinate range of the damage position, the severity level is moderate, the type matching degree is 91%, and the damage type is determined as local membrane damage.

[0127] It should be understood that, although the steps in the flowcharts related to the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least some of the steps in the flowcharts related to the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least some of the other steps or the steps or stages in the other steps.

[0128] In an embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the power battery internal structure damage detection method, system, device and medium as described above when executing the computer program.

[0129] In an embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the method embodiments described above when executed by a processor.

[0130] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts are described in the part of the method embodiments. The device embodiments described above are merely illustrative, and the components described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e. they can be located in one place or distributed on multiple network units. Some or all of the modules can be selected to achieve the purpose of the present disclosure according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0131] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be pointed out that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for detecting damage to the internal structure of a power cell, characterized in that The method comprises: acquiring reflection signals of different medium interfaces inside the power battery, and extracting first characteristic data including reflection amplitude and frequency distribution; collecting vibration signals in the charging and discharging process, and extracting second characteristic data including characteristic frequency and amplitude modulation after wavelet transform decomposition and reconstruction; using a pre-trained neural network fusion algorithm to perform fusion calculation on the first characteristic data and the second characteristic data, to generate comprehensive characteristic data containing multi-dimensional damage characteristics; the multi-dimensional damage characteristics include ultrasonic characteristics and vibration characteristics reflecting abnormal internal structure of the battery; inputting the comprehensive characteristic data into a pre-trained damage evaluation algorithm model to perform positioning analysis on abnormal values in the characteristic data to determine damage positions, extract spatial geometric characteristics to generate damage area descriptions; combining the damage area descriptions, the quantitative values related to the damage degree of the comprehensive characteristic data, and the matching degrees of each template in the pre-set damage type characteristic library to identify the specific types of damage, and generate a detection report containing a structured field.

2. The method of claim 1, wherein, The method comprises: integrating the first characteristic data and the second characteristic data to construct an original characteristic set containing ultrasonic characteristics and vibration characteristics; the ultrasonic characteristics correspond to physical interface characteristics of the internal structure of the battery; the vibration characteristics correspond to dynamic response characteristics under the working state of the battery; using a pre-trained dual-channel neural network fusion algorithm to perform spatial dimension matching on the ultrasonic and vibration characteristics in the original characteristic set, calculating key damage characteristic weights through an attention mechanism, and outputting preliminary fusion characteristic data; mining transient damage response characteristics through time-domain and frequency-domain joint analysis on the preliminary fusion characteristic data, and constructing a damage characteristic set containing multi-dimensional damage characteristics in combination with spatial distribution characteristics; the damage characteristic set includes damage response intensity, distribution range, and dynamic change trend; based on the damage characteristic set, performing redundant feature elimination and key feature enhancement to generate dimension-optimized comprehensive characteristic data.

3. The method of claim 2, wherein, The key damage characteristic weight is calculated by the following formula: where ω i represents the weight of the i-th key damage feature, s i represents the importance score generated by the self-attention mechanism of the i-th feature, τ represents the temperature coefficient, the value range is (0, 1], n represents the total number of features, I(f i ,f k ) represents the mutual information of the feature f i and f k , ||f i -f k ||2 represents the Euclidean distance of the feature vector, and σ represents the Gaussian kernel bandwidth parameter.

4. The method of claim 1, wherein, The method comprises: extracting a feature vector from the dimension-optimized comprehensive characteristic data; the feature vector contains time series parameters and spatial distribution parameters related to damage; the time series parameters include dynamic change values of damage characteristics with the charging and discharging cycle; the spatial distribution parameters include coordinate distribution values of characteristic abnormal regions; inputting the feature vector into a pre-trained damage evaluation algorithm model to detect values deviating from the normal characteristic distribution range in the vector, and outputting an abnormal value distribution set containing abnormal characteristic values and their corresponding indexes; the damage evaluation algorithm model uses an isolation forest algorithm to calculate the isolation path length of the feature vector in a hyperplane to determine abnormalities. Based on the set of abnormal value distribution, a density clustering algorithm is used for spatial clustering analysis of the abnormal values, and a damage coordinate set composed of three-dimensional coordinate points is generated according to the clustering center coordinates to locate the damage positions of the abnormal values; The spatial geometric features of the damage positions are extracted from the damage coordinate set to generate a damage area description containing damage area boundary coordinates, area and spatial form; the spatial geometric features include the aggregation range and center distance of the coordinate points and the area contour parameters.

5. The method of claim 1, wherein, The specific type of the damage is identified by matching the damage area description with the quantitative values related to the damage degree of the comprehensive feature data and the matching degrees of each template in the preset damage type feature library, and a detection report containing structured fields is generated, including: Based on the damage area description, a principal component analysis algorithm is used for dimension reduction processing to extract key feature dimensions and generate a standardized feature vector; the damage area description contains damage area boundary coordinates, area, spatial form and geometric feature parameters; Distribution feature data representing the damage diffusion trend and aggregation form are obtained by calculating the distribution density and gradient change rate of each dimension parameter in the standardized feature vector; the standardized feature vector contains area contour complexity, spatial distribution entropy and geometric center offset; Damage feature patterns in the distribution feature data and the comprehensive feature data are extracted; the damage feature patterns include diffusion rate, aggregation density peak, ultrasonic reflection wave distortion rate and vibration signal harmonic distortion degree; The cosine similarity of the damage feature patterns and each template in the preset damage type feature library is calculated, and the type corresponding to the template with the highest similarity and exceeding the matching threshold is determined as the specific type of the damage; the damage type feature library contains feature templates of typical damages such as diaphragm damage, pole piece shedding and electrolyte leakage; If the cosine similarity is higher than the matching threshold, a detection report containing structured fields is generated by integrating the three-dimensional coordinate range of the damage position, the severity level and the type matching degree value; if it is lower than the matching threshold, a secondary feature pattern extraction and matching process is triggered until the matching threshold is met; the matching threshold is set based on the accuracy verification of historical detection data; the structured fields include position parameters, level identification, matching degree score and type determination result.

6. The method of claim 5, wherein, The distribution density and gradient change rate of each dimension parameter in the standardized feature vector are calculated by the following formula: where, ρ j represents the distribution density of the jth dimension parameter in the normalized feature vector, represents the average value of the jth dimension parameter in the damage area, V represents the three-dimensional volume of the damage area, which is calculated by spatial integration through the boundary coordinates, represents the gradient change rate of the jth dimension parameter at a certain point (x, y, z) in space, f j (x, y, z) represents the eigenvalue of the dimension parameter at the spatial point (x, y, z), and the partial derivative is calculated by the three-dimensional difference method, taking the difference value of the eigenvalue of the spatial point (x, y, z) and the adjacent sampling point divided by the spatial distance.

7. The method of claim 5, wherein, The extraction of the damage feature patterns in the distribution feature data and the comprehensive feature data includes: Based on the distribution feature data, the volume change of the damage area in consecutive charge and discharge cycles is calculated to obtain a diffusion rate representing the damage expansion speed; the volume change is obtained by calculating the volume difference of the boundary coordinates of adjacent cycles; The distribution density of each dimension parameter in the distribution feature data is processed by spatial interpolation to generate a three-dimensional density field, and the maximum point in the density field is located by an extreme value detection algorithm to determine the density value corresponding to the maximum point as the aggregation density peak; performing Fourier transform on frequency distribution data in the ultrasonic feature based on the comprehensive feature data, calculating a ratio of an energy sum of abnormal frequency components to a total signal energy to obtain an ultrasonic reflection wave distortion rate; the abnormal frequency components are frequencies deviating from normal frequency bands; performing harmonic analysis on the time domain signal of the vibration feature according to the comprehensive feature data, separating fundamental wave and high-order harmonic components, and calculating a proportion of an amplitude sum of all high-order harmonics to a fundamental wave amplitude to obtain a vibration signal harmonic distortion degree; combining the diffusion rate, the aggregation density peak value, the ultrasonic reflection wave distortion rate, and the vibration signal harmonic distortion degree to obtain the damage feature mode.

8. A system for detecting damage to the internal structure of a power cell, characterized by The system comprises: a signal acquisition module configured to acquire reflection signals of different medium interfaces inside the power battery, extract first feature data including reflection wave amplitudes and frequency distributions, and acquire vibration signals in a charging and discharging process, and extract second feature data including characteristic frequencies and amplitude modulations after wavelet transform decomposition and reconstruction; a feature fusion module configured to perform fusion calculation on the first feature data and the second feature data by using a pre-trained neural network fusion algorithm, and generate comprehensive feature data containing multi-dimensional damage features; the multi-dimensional damage features include ultrasonic features and vibration features reflecting abnormal structures inside the battery; a damage positioning module configured to input the comprehensive feature data into a pre-trained damage evaluation algorithm model to perform positioning analysis on abnormal values in the feature data and determine a damage position, extract spatial geometric features, and generate a damage area description; an evaluation and reporting module configured to identify a specific type of damage by combining the damage area description, a quantitative value related to a damage degree of the comprehensive feature data, and a matching degree of each template in a pre-set damage type feature library, and generate a detection report containing structured fields. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 7.

Images