Battery top cover airtightness detection device and detection method

Abstract

The invention discloses a battery top cover airtightness detection device and detection method, and relates to the technical field of sealing detection. The method comprises the following steps: applying a preset dynamic pressure sequence to a pressurizing area covering a detection cavity outside a battery top cover, and synchronously acquiring the battery top cover to obtain a vibration response signal; performing time segmentation on the vibration response signal to obtain a sub-vibration signal set; inputting the sub-vibration signal set into a preset feature extraction module to obtain a vibration feature set; fourier transform is carried out on each vibration characteristic to obtain a frequency image set; and performing airtightness analysis on the battery top cover according to the frequency image set to obtain a detection result. According to the method, vibration response is obtained through dynamic pressure sequence excitation, air tightness judgment is achieved in combination with sliding window time domain segmentation and Fourier transform frequency domain analysis, tiny defects can be effectively recognized, sealing dynamic changes can be adapted, the risk that a battery structure is damaged due to improper loads is eliminated, and the reliability of the battery structure is improved. Therefore, precise detection of the air tightness of the battery top cover is realized.

Description

Technical Field

[0001] This invention belongs to the field of sealing detection technology, specifically relating to a battery top cover airtightness detection device and detection method. Background Technology

[0002] With the rapid development of new energy vehicles, energy storage power stations, and other fields, the safety performance and reliability of power batteries such as lithium-ion batteries have become a core focus of the industry. As a key component of the battery pack sealing system, the airtightness of the battery top cover directly determines the retention of electrolyte inside the battery and the isolation effect between the cell and the external environment. Once the seal fails, it can easily lead to safety accidents such as electrolyte leakage, cell short circuit, and thermal runaway. Therefore, efficient and accurate airtightness testing of the battery top cover is a necessary procedure in battery production and maintenance.

[0003] Patent application CN120538746A discloses a method, apparatus, and battery pack for verifying the sealing performance of a battery pack. The method includes: constructing a battery pack model, including a top cover and a housing; the top cover has a first edge, and the housing has a second edge that fits against the first edge; connecting the first and second edges with bolts; and sandwiching a sealing gasket between the first and second edges; applying a preload to the bolts, pointing towards the contact planes where the first and second edges respectively contact the sealing gasket, to compress the thickness of the sealing gasket and generate compressive deformation; applying a pressure load from inside the housing to outside the housing to resist the preload and compressive deformation; identifying the deformation of the sealing gasket under compression, and defining the area where the deformation is greater than or equal to a deformation threshold as a compression area; and determining that the target battery pack's sealing performance is qualified under the preload and pressure loads when a continuous compression area is formed on the contact plane.

[0004] However, this method only determines the sealing performance based on the compression deformation of the sealing gasket and the continuity of the pressing area. It relies on the structural mechanical parameters of the connection between the shell and the top cover. It cannot capture the dynamic response of small airtight defects, is difficult to adapt to the dynamic changes in sealing performance throughout the entire life cycle, and is also prone to damage to the battery pack structure and affect the safety of the cells due to improper load or model deviation. Therefore, it cannot achieve reliable sealing performance testing. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that relying solely on the compression deformation of the sealing gasket and the continuity of the compression area as the judgment criteria can lead to safety hazards in the battery pack structure and cells, and also has the limitation of limited application range, making it impossible to reliably detect the battery sealing performance. Therefore, this invention proposes a battery top cover airtightness detection device and detection method.

[0006] In a first aspect of this invention, a method for detecting the airtightness of a battery top cover is first proposed, the method comprising:

[0007] A preset dynamic pressure sequence is applied to the pressurized area of ​​the detection cavity covering the outside of the battery top cover, and the vibration response signal is synchronously acquired by the battery top cover;

[0008] The vibration response signal is time-divided by a preset sliding window to obtain a sub-vibration signal set;

[0009] The sub-vibration signal set is input into a preset feature extraction module to obtain a vibration feature set;

[0010] Perform a Fourier transform on each vibration feature in the vibration feature set to obtain a frequency map set;

[0011] The airtightness of the battery top cover was analyzed based on the frequency atlas to obtain the test results.

[0012] Optionally, the working principle of the preset feature extraction module includes:

[0013] A multi-scale temporal feature set is obtained by performing a preset convolution operation on each sub-vibration signal in the sub-vibration signal set;

[0014] The multi-scale temporal feature set is split according to the scale dimension to obtain a single-scale feature set;

[0015] Variational mode decomposition is performed on the target single-scale feature to extract a preset number of intrinsic mode components. The preset number of intrinsic mode components are then screened and their waveforms are preserved to obtain joint time-domain features characterizing a specific vibration mode at that scale. The target single-scale feature is any single-scale feature in the set of single-scale features.

[0016] The mean, variance, kurtosis, and energy percentage of all joint temporal features are calculated to obtain the scale-aware vector set;

[0017] Using the scale-aware vectors in the scale-aware vector set as nodes, the feature similarity between any two nodes is calculated as the edge weight to construct a fully connected scale relationship graph.

[0018] Iterative graph convolutional propagation is performed on the scale relationship graph. By aggregating the features of each node and its first-order neighboring nodes and redistributing the node state according to the edge weight, a set of adaptive fusion weight coefficients containing global topological dependencies is finally output.

[0019] After weighted summation of the adaptive fusion weight coefficients and the multi-scale temporal feature set, temporal alignment, standardization, and length normalization are performed to obtain the vibration feature set.

[0020] Optionally, the pre-defined convolution operation process includes:

[0021] A multi-channel temporal feature set is obtained by performing multi-channel filtering on the sub-vibration signal set through a one-dimensional convolutional layer.

[0022] A deep feature set is obtained by performing step-by-step feature mapping on the target multi-channel temporal features through multiple cascaded convolutional layers; the target multi-channel temporal feature is any one of the multi-channel temporal features in the multi-channel temporal feature set.

[0023] Multi-scale temporal features are obtained by cross-channel integration of deep features in the deep feature set;

[0024] The multi-scale time-domain feature set is obtained by statistically analyzing all multi-scale time-domain features.

[0025] Optionally, the airtightness analysis of the battery based on the frequency atlas to obtain the test results includes:

[0026] By using a preset weight vector corresponding to each defect category, each frequency map in the frequency map set is weighted and summed to generate a frequency activation map set representing the intensity of the characteristic frequency response of each candidate fault category.

[0027] For each candidate defect category, the frequency activation map set corresponding to the frequency activation map set is selected along the frequency dimension to obtain the candidate key activation map set under that defect category by selecting the frequency activation map set with an activation intensity greater than the preset activation threshold.

[0028] A similarity matrix is ​​obtained by performing similarity calculations on every two candidate key activation graphs in the candidate key activation graph set;

[0029] Based on the similarity matrix, the mechanical energy screening of paired candidate key activation maps that exceed the preset similarity threshold is performed, and candidate key activation maps with higher average activation intensity are retained to finally obtain the key activation map set;

[0030] The core frequency band corresponding to the key activation map is obtained by extracting significant continuous frequency bands with activation intensity exceeding an adaptive threshold along the frequency dimension of the target key activation map; the target key activation map is any one of the key activation maps in the set of key activation maps.

[0031] The test results are obtained by judging the defects of the battery top cover based on all core frequency bands.

[0032] Optionally, the test results obtained by judging defects in the battery top cover based on all core frequency bands include:

[0033] The intensity composite value characterizing the anomaly scale is obtained by statistically analyzing the number of frequency points and their average activation intensity across all core frequency bands.

[0034] The bandwidth of all core discrimination frequency bands on the frequency axis is measured to obtain the bandwidth value that characterizes the breadth of the anomaly distribution;

[0035] An anomaly score is obtained by weighting the comprehensive intensity value and the frequency band span value.

[0036] The abnormal score is compared with a preset judgment threshold. If the abnormal score exceeds the preset judgment threshold, the airtightness of the battery top cover is determined to be unqualified.

[0037] In a second aspect of the invention, a battery top cover airtightness detection device is provided, comprising:

[0038] The acquisition module is used to apply a preset dynamic pressure sequence to the pressurized area of ​​the detection cavity covering the outside of the battery top cover, and synchronously acquire the vibration response signal of the battery top cover;

[0039] The time segmentation module is used to divide the vibration response signal into sub-vibration signal sets by a preset sliding window;

[0040] A vibration feature generation module is used to input the sub-vibration signal set into a preset feature extraction module to obtain a vibration feature set;

[0041] The frequency map generation module is used to perform Fourier transform on each vibration feature in the vibration feature set to obtain a frequency map set;

[0042] The airtightness analysis module is used to perform airtightness analysis on the battery top cover based on the frequency atlas to obtain the test results.

[0043] Optionally, the working principle of the preset feature extraction module includes:

[0044] A multi-scale temporal feature set is obtained by performing a preset convolution operation on each sub-vibration signal in the sub-vibration signal set;

[0045] The multi-scale temporal feature set is split according to the scale dimension to obtain a single-scale feature set;

[0046] Variational mode decomposition is performed on the target single-scale feature to extract a preset number of intrinsic mode components. The preset number of intrinsic mode components are then screened and their waveforms are preserved to obtain joint time-domain features characterizing a specific vibration mode at that scale. The target single-scale feature is any single-scale feature in the set of single-scale features.

[0047] The mean, variance, kurtosis, and energy percentage of all joint temporal features are calculated to obtain the scale-aware vector set;

[0048] Using the scale-aware vectors in the scale-aware vector set as nodes, the feature similarity between any two nodes is calculated as the edge weight to construct a fully connected scale relationship graph.

[0049] Iterative graph convolutional propagation is performed on the scale relationship graph. By aggregating the features of each node and its first-order neighboring nodes and redistributing the node state according to the edge weight, a set of adaptive fusion weight coefficients containing global topological dependencies is finally output.

[0050] After weighted summation of the adaptive fusion weight coefficients and the multi-scale temporal feature set, temporal alignment, standardization, and length normalization are performed to obtain the vibration feature set.

[0051] Optionally, the pre-defined convolution operation process includes:

[0052] A multi-channel temporal feature set is obtained by performing multi-channel filtering on the sub-vibration signal set through a one-dimensional convolutional layer.

[0053] A deep feature set is obtained by performing stepwise feature mapping on the target multi-channel temporal features through multiple cascaded convolutional layers; the target multi-channel temporal feature is any one of the multi-channel temporal features in the multi-channel temporal feature set.

[0054] Multi-scale temporal features are obtained by cross-channel integration of deep features in the deep feature set;

[0055] The multi-scale time-domain feature set is obtained by statistically analyzing all multi-scale time-domain features.

[0056] Optionally, the airtightness analysis module includes:

[0057] The frequency activation module is used to perform a weighted summation of each frequency map in the frequency map set using a preset weight vector corresponding to each defect category, and generate a frequency activation map set representing the intensity of the characteristic frequency response of each candidate fault category.

[0058] The candidate key activation map generation module is used to filter out the frequency activation maps with activation intensity greater than a preset activation threshold along the frequency dimension for each candidate defect category to obtain the candidate key activation map set for that defect category.

[0059] The similarity calculation module is used to perform similarity calculation on every two candidate key activation graphs in the candidate key activation graph set to obtain a similarity matrix;

[0060] The key activation map generation module is used to filter paired candidate key activation maps that exceed a preset similarity threshold according to the similarity matrix, retain the candidate key activation maps with higher average activation intensity, and finally obtain the key activation map set.

[0061] The core frequency band generation module is used to extract significant continuous frequency bands with activation intensities exceeding an adaptive threshold along the frequency dimension of the target key activation map to obtain the core frequency band corresponding to the key activation map; the target key activation map is any one of the key activation maps in the set of key activation maps.

[0062] The test result generation module is used to determine the defects of the battery top cover based on all core frequency bands and obtain the test results.

[0063] Optionally, the detection result generation module includes:

[0064] The intensity comprehensive value calculation module is used to count the number of frequency points and their average activation intensity in all core frequency bands to obtain an intensity comprehensive value that characterizes the scale of the anomaly.

[0065] The bandwidth span value calculation module is used to measure the coverage width of all core discrimination bandwidths on the frequency axis to obtain the bandwidth span value that characterizes the breadth of the abnormal distribution;

[0066] The weighting module is used to weight the comprehensive intensity value and the frequency band span value to obtain an anomaly score;

[0067] The judgment module is used to compare the abnormal score with a preset judgment threshold. If the abnormal score exceeds the preset judgment threshold, the airtightness of the battery top cover is determined to be unqualified.

[0068] The beneficial effects of this invention are as follows: This invention proposes a method for detecting the airtightness of a battery top cover. A preset dynamic pressure sequence is applied to the pressurized area of ​​the external detection cavity of the battery top cover, and vibration response signals are simultaneously acquired. The signals are then time-divided using a preset sliding window to obtain a sub-vibration signal set. This sub-vibration signal set is then input into a preset feature extraction module to obtain a vibration feature set. Subsequently, Fourier transforms are performed on each vibration feature to generate a frequency spectrum. Finally, the airtightness of the battery top cover is analyzed based on the frequency spectrum, and the detection results are output. This method uses dynamic pressure sequence excitation to obtain the vibration response, and combines sliding window time-domain segmentation with Fourier transform frequency-domain analysis to determine airtightness. It not only effectively identifies minute defects and adapts to dynamic changes in sealing performance, but also eliminates the risk of battery structural damage due to improper load, thereby achieving precise detection of the airtightness of the battery top cover. Attached Figure Description

[0069] The present invention will now be further described with reference to the accompanying drawings.

[0070] Figure 1 A flowchart of a battery top cover airtightness detection method provided in an embodiment of the present invention;

[0071] Figure 2 This is a schematic diagram of a battery top cover airtightness detection device provided in an embodiment of the present invention. Detailed Implementation

[0072] The technical methods of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0073] 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.

[0074] This invention provides a method for detecting the airtightness of a battery top cover. See also... Figure 1 , Figure 1 A flowchart illustrating a battery top cover airtightness testing method provided in an embodiment of the present invention. The method includes the following steps:

[0075] S101, apply a preset dynamic pressure sequence to the pressurization area of ​​the detection cavity covering the outside of the battery top cover, and synchronously collect the vibration response signal of the battery top cover;

[0076] S102, the vibration response signal is time-divided by a preset sliding window to obtain a sub-vibration signal set;

[0077] S103, input the sub-vibration signal set into the preset feature extraction module to obtain the vibration feature set;

[0078] S104, Perform Fourier transform on each vibration feature in the vibration feature set to obtain the frequency map set;

[0079] S105, the airtightness analysis of the battery top cover was performed based on the frequency atlas to obtain the test results.

[0080] This invention provides a method for detecting the airtightness of a battery top cover. The method involves first applying a preset dynamic pressure sequence to the pressurized area of ​​the external detection cavity of the battery top cover, while simultaneously acquiring its vibration response signal. After time-domain segmentation of the vibration response signal through a preset sliding window, a sub-vibration signal set is obtained. This sub-vibration signal set is then input into a preset feature extraction module to acquire a vibration feature set. Fourier transform processing is performed on each vibration feature to generate a frequency atlas. Finally, the airtightness of the battery top cover is analyzed based on the frequency atlas, and the detection results are output. This method relies on the acquisition of vibration response under dynamic pressure excitation, time-domain signal segmentation via a sliding window, and frequency-domain feature mining using Fourier transform to determine airtightness. It can effectively identify minute airtightness defects, adapt to dynamic changes in sealing performance, and eliminate the risk of battery structural damage caused by improper loads, thereby achieving precise detection of the battery top cover's airtightness.

[0081] In one implementation, the length of the preset sliding window is set by technicians based on the actual working conditions of the vibration signal sampling frequency.

[0082] In one implementation, the applied preset dynamic pressure sequence is precisely calculated and calibrated. Its pressure peak, fluctuation frequency, and duration are all controlled within the load-bearing threshold of the battery top cover structure and the safe tolerance range of the battery cell. It can be used as an excitation source to induce mechanical vibration in the top cover structure without damaging the battery casing, internal battery cell, and various sealing components. If there are airtight defects such as microcracks or weld leaks in the top cover, the stiffness and damping characteristics of its local structure will differ from those of the intact area. At the same time, the defect will also generate fluid-structure interaction effects such as turbulence and throttling due to gas leakage, thus generating vibration coupling signals different from those of the intact area. The synchronous acquisition of vibration response signals is essentially to capture the correlation physical information between dynamic pressure excitation and the airtight state and structural characteristics of the top cover, transforming the implicit characteristics of airtight defects into quantifiable vibration signals. This provides core raw data for subsequent accurate defect identification through feature extraction and frequency domain analysis. The dynamic design of its pressure sequence can also cover the vibration response characteristics of airtight defects of different types and scales, ensuring that the raw signal can fully carry the physical characterization of the defects.

[0083] In one embodiment, the working principle of the preset feature extraction module includes:

[0084] A multi-scale temporal feature set is obtained by performing a pre-defined convolution operation on each sub-vibration signal in the sub-vibration signal set;

[0085] The multi-scale temporal feature set is split according to the scale dimension to obtain a single-scale feature set;

[0086] Variational mode decomposition is performed on the single-scale feature of the target to extract a preset number of intrinsic mode components. The preset number of intrinsic mode components are then screened and their waveforms are preserved to obtain the joint time-domain features that characterize the specific vibration mode at that scale. The single-scale feature of the target is any single-scale feature in the single-scale feature set.

[0087] The mean, variance, kurtosis, and energy percentage of all joint temporal features are calculated to obtain the scale-aware vector set;

[0088] Using the scale-aware vectors in the scale-aware vector set as nodes, calculate the feature similarity between any two nodes as the edge weight, and construct a fully connected scale relationship graph.

[0089] Iterative graph convolution propagation is performed on the scale relationship graph. By aggregating the features of each node and its first-order neighboring nodes and redistributing the node state according to the edge weight, a set of adaptive fusion weight coefficients containing global topological dependencies is finally output.

[0090] After weighted summation of the adaptive fusion weight coefficients and the multi-scale time-domain feature set, temporal alignment, standardization, and length normalization are performed to obtain the vibration feature set.

[0091] In one implementation, the preset quantity is set by technical personnel.

[0092] In one implementation, the input signal is decomposed into a preset number of intrinsic mode components, each of which has a tightly supported characteristic in the frequency domain and its center frequency is adaptively determined. This process achieves quasi-orthogonal decomposition of the signal in the time-frequency domain, which can separate different vibration modes aliased within the same scale. The subsequent screening is based on the physical meaning and statistical significance of each component: first, it is evaluated whether the center frequency of each component falls within the known fault characteristic frequency band; second, it is evaluated whether its energy is significantly higher than the noise floor; and finally, it is checked whether its waveform has physical rationality. By selecting and retaining key components, their complete time-domain waveforms are directly retained and aligned and superimposed. The resulting joint time-domain feature is essentially a set of time-domain waveforms of all significant vibration modes at this scale, retaining the complete amplitude, phase, and modulation information of the original oscillation, providing the most direct waveform data for subsequent physical feature quantization.

[0093] In one implementation, this step quantifies the joint temporal features at each scale using four types of statistical indicators; the joint temporal feature sequence at each scale is x = [x1, x2, ..., x...]. N Perform four-dimensional statistical quantification, where N is the sequence length; first, calculate the mean. This value reflects the DC offset of the signal and can indicate the vibration deviation caused by structural asymmetry; then the variance is calculated. The total energy of the wave after removing the DC component from the characterization signal directly reflects the excitation intensity of the vibration mode at this scale; further calculation of kurtosis... in The fourth central moment is used to measure the sharpness of the signal distribution. When there are periodic impact pulses caused by minute leaks, the K value will be significantly higher than the baseline value of 3 for the normal distribution. Finally, the energy proportion R = E is calculated. k / E t ,in This represents the characteristic energy at this scale. The total energy across all M scales is represented by a proportion that reflects the relative importance of the vibration mode in the global energy distribution. These four statistics constitute a four-dimensional sensing vector characterizing the scale. The sensing vectors of all scales form a scale sensing vector set, which numerically encodes the statistical distribution characteristics of the signal in multi-scale space.

[0094] In one implementation, the four-dimensional statistical feature vector of each scale is treated as an independent node in the graph. The weight of the connection edge between any two nodes is determined by calculating the directional similarity of their feature vectors in multidimensional space. The higher the similarity value, the greater the connection weight, indicating that the two scales are statistically similar and may originate from the same physical excitation or have a coupled vibration transmission relationship. By establishing connections for all node pairs to form a fully connected graph, this structure fully expresses the potential network of connections between features at each scale. Without pre-setting any connection constraints, it provides input graph data containing complete topological relationships for subsequent graph convolutional learning in the form of an adjacency matrix.

[0095] In one implementation, the process can accurately separate multi-dimensional fault features related to airtightness from complex vibration responses, effectively suppress the interference of environmental noise and irrelevant vibration modes, and explicitly model the physical correlation between features at different scales through graph convolution mechanism, ultimately generating a standardized feature representation with strong discriminativeness, high robustness and physical interpretability, providing a reliable data foundation for subsequent frequency domain analysis and accurate airtightness determination.

[0096] In one embodiment, the pre-defined convolution operation process includes:

[0097] A multi-channel temporal feature set is obtained by performing multi-channel filtering on the sub-vibration signal set through a one-dimensional convolutional layer;

[0098] A deep feature set is obtained by performing step-by-step feature mapping on the target multi-channel temporal features through multiple cascaded convolutional layers; the target multi-channel temporal feature is any one of the multi-channel temporal features in the multi-channel temporal feature set.

[0099] Multi-scale temporal features are obtained by cross-channel integration of deep features in the deep feature set;

[0100] The multi-scale time-domain feature set is obtained by statistically analyzing all multi-scale time-domain features.

[0101] In one implementation, the 1D convolutional layer uses a pre-sized, unbiased convolutional kernel, which essentially performs frequency domain filtering on the vibration signal. The multi-channel convolutional kernel can separate the features of different frequency bands in the signal, thereby outputting a multi-channel time-series feature set. Multiple cascaded convolutional layers perform deep coupling of high-frequency and low-frequency information on the initial multi-channel time-series features through layer-by-layer convolutional operations. Each convolutional kernel maintains a consistent frequency filtering intensity under norm constraints, preventing the deep network from losing key frequency information, thus transforming the shallow time-domain features into a deep feature set containing multi-frequency band coupling information. Cross-channel integration of deep features is essentially the fusion of power and phase of the time-series features of different frequency bands, breaking the frequency band limitations of a single channel and generating multi-scale time-domain features with multi-frequency band information. Finally, all multi-scale time-domain features are statistically analyzed to form a feature set, which not only preserves the time-domain dynamic characteristics of the vibration signal but also integrates the frequency features of multiple frequency bands.

[0102] In one embodiment, the test results obtained by performing airtightness analysis on the battery based on a frequency atlas include:

[0103] By using a pre-defined weight vector corresponding to each defect category, a weighted summation is performed on each frequency map in the frequency map set to generate a frequency activation map set representing the intensity of the characteristic frequency response of each candidate fault category.

[0104] For each candidate defect category, the frequency activation map set corresponding to the frequency activation map set is selected along the frequency dimension to obtain the candidate key activation map set under that defect category by selecting the frequency activation map set with an activation intensity greater than the preset activation threshold.

[0105] A similarity matrix is ​​obtained by calculating the similarity between every two candidate key activation maps in the candidate key activation map set;

[0106] Based on the similarity matrix, paired candidate key activation maps that exceed a preset similarity threshold are filtered out, and candidate key activation maps with higher average activation intensity are retained to finally obtain the key activation map set;

[0107] The core frequency band corresponding to the key activation map is obtained by extracting significant continuous frequency bands with activation intensity exceeding an adaptive threshold along the frequency dimension of the target key activation map; the target key activation map is any key activation map in the key activation map set.

[0108] The test results are obtained by judging the defects of the battery top cover based on all core frequency bands.

[0109] In one implementation, the preset weight vector corresponding to each defect category refers to a vector composed of numerical weights configured for different defect types. It can assign differentiated importance weights to various defects in the defect identification or evaluation process, thereby improving the accuracy of judgment. It is set by technicians based on historical experience. The preset similarity threshold and preset activation threshold are also set by technicians.

[0110] In one implementation, thresholding segmentation and continuous region extraction along the frequency dimension are performed on the key activation map. First, an adaptive threshold is calculated based on the activation intensity distribution of all frequency points in the key activation map. This threshold is usually dynamically determined based on statistical characteristics to ensure robustness against noise and background fluctuations. Then, the entire frequency axis is scanned to identify all continuous frequency intervals where the activation intensity of each frequency point exceeds the adaptive threshold. These continuous intervals are the significant frequency bands. Finally, these significant frequency bands are verified and merged. If the interval between adjacent frequency bands is less than a preset minimum gap, they are merged into the same frequency band, and isolated frequency bands that are too short are removed. The final set of continuous significant frequency bands is the core frequency band corresponding to the key activation map. These frequency bands collectively represent the characteristic frequency response regions most relevant to the fault hypothesis.

[0111] In one implementation, the process uses category-oriented weighted aggregation, adaptive threshold filtering, and graph similarity redundancy removal to accurately extract key frequency bands highly correlated with various faults from multi-channel frequency features, effectively eliminating noise interference and redundant responses. Through an intensity-based competitive filtering mechanism, the most discriminative activation modes are retained. Finally, based on the comprehensive features of the core frequency band, defect determination is performed, significantly improving the pertinence, robustness, and interpretability of fault diagnosis, and achieving an efficient and reliable mapping from complex spectra to clear fault modes.

[0112] In one embodiment, the defect determination of the battery top cover based on all core frequency bands to obtain the detection result includes:

[0113] The intensity composite value characterizing the anomaly scale is obtained by statistically analyzing the number of frequency points and their average activation intensity across all core frequency bands.

[0114] The bandwidth of all core discrimination frequency bands on the frequency axis is measured to obtain the bandwidth value that characterizes the breadth of the anomaly distribution;

[0115] An anomaly score is obtained by weighting the overall intensity value and the frequency band span value.

[0116] The abnormal score is compared with the preset judgment threshold. If the abnormal score exceeds the preset judgment threshold, the airtightness of the battery top cover is determined to be unqualified.

[0117] In one implementation, the preset judgment threshold is set by a technician.

[0118] In one implementation, the frequency domain characteristics of all core frequency bands are quantitatively statistically analyzed. The total number of frequency points is obtained by traversing each frequency band and accumulating the total number of discrete frequency points contained therein. At the same time, the arithmetic mean of the activation intensity of all frequency points is calculated. Then, the total number of frequency points is multiplied by the average activation intensity to generate a comprehensive intensity value that reflects the distribution density of anomalous frequency components and the average energy level. By identifying the minimum start frequency and the maximum end frequency of all core frequency bands on the frequency axis, the difference between the two is calculated to obtain the frequency band span value that characterizes the extreme boundary range of the anomalous component distribution. This value reflects the distribution breadth of anomalous vibration in the frequency domain. The larger the span, the more complex or severe the leakage mode usually is.

[0119] Based on the same inventive concept, this invention also provides a battery top cover airtightness detection device. See also Figure 2 , Figure 2 A schematic diagram of a battery top cover airtightness detection device provided in an embodiment of the present invention includes:

[0120] The acquisition module is used to apply a preset dynamic pressure sequence to the pressurized area of ​​the detection cavity covering the outside of the battery top cover, and synchronously acquire the vibration response signal of the battery top cover;

[0121] The time division module is used to divide the vibration response signal into sub-vibration signal sets by a preset sliding window;

[0122] The vibration feature generation module is used to input the sub-vibration signal set into the preset feature extraction module to obtain the vibration feature set;

[0123] The frequency map generation module is used to perform Fourier transform on each vibration feature in the vibration feature set to obtain a frequency map set;

[0124] The airtightness analysis module is used to perform airtightness analysis on the battery top cover based on the frequency atlas to obtain the test results.

[0125] This invention provides a battery top cover airtightness testing device. It first applies a preset dynamic pressure sequence to the pressurized area of ​​the external detection cavity of the battery top cover, simultaneously acquiring its vibration response signal. After time-domain segmentation of the vibration response signal through a preset sliding window, a sub-vibration signal set is obtained. This sub-vibration signal set is then input into a preset feature extraction module to obtain a vibration feature set. Fourier transform processing is performed on each vibration feature to generate a frequency atlas. Finally, the airtightness of the battery top cover is analyzed based on the frequency atlas, and the test results are output. This method relies on the acquisition of vibration response under dynamic pressure excitation, time-domain signal segmentation through a sliding window, and frequency-domain feature mining through Fourier transform to achieve airtightness determination. It can not only effectively identify minute airtightness defects and adapt to dynamic changes in sealing performance, but also eliminate the risk of battery structural damage caused by improper loads, thereby achieving accurate detection of battery top cover airtightness.

[0126] In one embodiment, the working principle of the preset feature extraction module includes:

[0127] A multi-scale temporal feature set is obtained by performing a pre-defined convolution operation on each sub-vibration signal in the sub-vibration signal set;

[0128] The multi-scale temporal feature set is split according to the scale dimension to obtain a single-scale feature set;

[0129] Variational mode decomposition is performed on the single-scale feature of the target to extract a preset number of intrinsic mode components. The preset number of intrinsic mode components are then screened and their waveforms are preserved to obtain the joint time-domain features that characterize the specific vibration mode at that scale. The single-scale feature of the target is any single-scale feature in the single-scale feature set.

[0130] The mean, variance, kurtosis, and energy percentage of all joint temporal features are calculated to obtain the scale-aware vector set;

[0131] Using the scale-aware vectors in the scale-aware vector set as nodes, calculate the feature similarity between any two nodes as the edge weight, and construct a fully connected scale relationship graph.

[0132] Iterative graph convolution propagation is performed on the scale relationship graph. By aggregating the features of each node and its first-order neighboring nodes and redistributing the node state according to the edge weight, a set of adaptive fusion weight coefficients containing global topological dependencies is finally output.

[0133] After weighted summation of the adaptive fusion weight coefficients and the multi-scale time-domain feature set, temporal alignment, standardization, and length normalization are performed to obtain the vibration feature set.

[0134] In one embodiment, the pre-defined convolution operation process includes:

[0135] A multi-channel temporal feature set is obtained by performing multi-channel filtering on the sub-vibration signal set through a one-dimensional convolutional layer;

[0136] A deep feature set is obtained by performing stepwise feature mapping on the target multi-channel temporal features through multiple cascaded convolutional layers; the target multi-channel temporal feature is any one of the multi-channel temporal features in the multi-channel temporal feature set.

[0137] Multi-scale temporal features are obtained by cross-channel integration of deep features in the deep feature set;

[0138] The multi-scale time-domain feature set is obtained by statistically analyzing all multi-scale time-domain features.

[0139] In one embodiment, the airtightness analysis module includes:

[0140] The frequency activation module is used to perform weighted summation on each frequency map in the frequency map set by using a preset weight vector corresponding to each defect category, so as to generate a frequency activation map set representing the characteristic frequency response intensity of each candidate fault category.

[0141] The candidate key activation map generation module is used to filter out the frequency activation maps with activation intensity greater than a preset activation threshold along the frequency dimension for each candidate defect category to obtain the candidate key activation map set for that defect category.

[0142] The similarity calculation module is used to calculate the similarity between every two candidate key activation graphs in the candidate key activation graph set to obtain a similarity matrix;

[0143] The key activation map generation module is used to filter paired candidate key activation maps that exceed a preset similarity threshold based on the similarity matrix, retain the candidate key activation maps with higher average activation intensity, and finally obtain the key activation map set.

[0144] The core frequency band generation module is used to extract significant continuous frequency bands with activation intensities exceeding an adaptive threshold along the frequency dimension of the target key activation map to obtain the core frequency band corresponding to the key activation map; the target key activation map is any key activation map in the key activation map set.

[0145] The test result generation module is used to determine the defects of the battery top cover based on all core frequency bands and obtain the test results.

[0146] In one embodiment, the detection result generation module includes:

[0147] The intensity comprehensive value calculation module is used to count the number of frequency points and their average activation intensity in all core frequency bands to obtain an intensity comprehensive value that characterizes the scale of the anomaly.

[0148] The bandwidth span value calculation module is used to measure the coverage width of all core discrimination bandwidths on the frequency axis to obtain the bandwidth span value that characterizes the breadth of the abnormal distribution;

[0149] The weighting module is used to weight the comprehensive intensity value and the frequency band span value to obtain anomaly scores;

[0150] The judgment module is used to compare the abnormal score with the preset judgment threshold. If the abnormal score exceeds the preset judgment threshold, the airtightness of the battery top cover is judged to be unqualified.

[0151] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A method for detecting airtightness of a battery top cover, characterized by, The method comprises: applying a preset dynamic pressure sequence to a pressurized area of a detection cavity outside a battery top cover to synchronously collect a vibration response signal of the battery top cover; time-dividing the vibration response signal through a preset sliding window to obtain a set of sub-vibration signals; inputting the set of sub-vibration signals into a preset feature extraction module to obtain a set of vibration features; performing Fourier transform on each vibration feature in the set of vibration features to obtain a set of frequency graphs; performing air tightness analysis on the battery top cover according to the set of frequency graphs to obtain a detection result.

2. The battery top cover airtightness detection method of claim 1, wherein, The working principle of the preset feature extraction module comprises: performing a preset convolution operation on each sub-vibration signal in the set of sub-vibration signals to obtain a set of multi-scale time domain features; splitting the set of multi-scale time domain features according to the scale dimension to obtain a set of single-scale features; performing variational modal decomposition on a target single-scale feature, extracting a preset number of intrinsic modal components, and selecting and retaining waveforms of the preset number of intrinsic modal components to obtain a joint time domain feature representing a specific vibration mode of the scale; the target single-scale feature is any single-scale feature in the set of single-scale features; calculating the mean, variance, kurtosis and energy proportion of all joint time domain features to obtain a set of scale perception vectors; taking the scale perception vectors in the set of scale perception vectors as nodes, calculating the feature similarity between any two nodes as an edge weight, and constructing a completely connected scale relationship graph; performing iterative graph convolution propagation on the scale relationship graph, aggregating the features of each node and its first-order neighborhood nodes, and redistributing the node state according to the edge weight, and finally outputting a set of adaptive fusion weight coefficients containing global topological dependency; after weighting and summing the adaptive fusion weight coefficients and the set of multi-scale time domain features, performing time alignment, standardization processing and length normalization to obtain a set of vibration features.

3. The method of claim 2, wherein the method further comprises: The process of the preset convolution operation comprises: performing multi-channel filtering processing on the set of sub-vibration signals through a one-dimensional convolution layer to obtain a set of multi-channel time sequence features; performing step-by-step feature mapping on a target multi-channel time sequence feature through a plurality of cascaded convolution layers to obtain a deep feature set; the target multi-channel time sequence feature is any multi-channel time sequence feature in the set of multi-channel time sequence features; integrating the deep features in the deep feature set across channels to obtain multi-scale time domain features; statistically obtaining a set of multi-scale time domain features.

4. The method of claim 1, wherein the method further comprises: The air tightness analysis of the battery according to the set of frequency graphs to obtain a detection result comprises: performing weighted summation on each frequency graph in the set of frequency graphs through a preset weight vector corresponding to each defect category to generate a frequency activation graph set representing the feature frequency response intensity of each candidate failure category for each candidate defect category; for each candidate defect category, filtering out frequency activation graphs with an activation intensity greater than a preset activation threshold from the frequency activation graph set corresponding to the candidate defect category to obtain a candidate key activation graph set under the defect category; performing similarity calculation on each two candidate key activation graphs in the candidate key activation graph set to obtain a similarity matrix; According to the similarity matrix, the pair of candidate key activation maps exceeding the preset similarity threshold are screened, the candidate key activation map with higher average activation intensity is retained, and finally a key activation map set is obtained; A significant continuous frequency band corresponding to the core frequency band of the target key activation map is obtained by extracting the activation strength exceeding the adaptive threshold along the frequency dimension of the target key activation map; the target key activation map is any one of the key activation maps in the key activation map set; According to all core frequency bands, the battery top cover is judged for defects to obtain a detection result.

5. The method of claim 4, wherein the step of detecting the air tightness of the battery top cover is performed by using a pressure sensor. According to all core frequency bands, the battery top cover is judged for defects to obtain a detection result, which includes: Statistical all core frequency bands of the frequency point number and its average activation intensity to obtain the strength comprehensive value representing the abnormal scale; Measure the coverage width of all core frequency bands on the frequency axis to obtain the frequency band span value representing the distribution width of the abnormal distribution; The strength comprehensive value and the frequency band span value are weighted to obtain an abnormal score; The abnormal score is compared with a preset judgment threshold, and if the abnormal score exceeds the preset judgment threshold, the battery top cover is determined to be unqualified in air tightness.

6. A battery top cover airtightness detection device, characterized in that, The device includes: The acquisition module is configured to apply a preset dynamic pressure sequence to a pressurized area of a detection cavity covering the outside of the battery top cover, and synchronously acquire the vibration response signal of the battery top cover; The time segmentation module is configured to perform time segmentation on the vibration response signal by using a preset sliding window to obtain a set of sub-vibration signals; The vibration feature generation module is configured to input the set of sub-vibration signals into a preset feature extraction module to obtain a set of vibration features; The frequency map generation module is configured to perform Fourier transform on each vibration feature in the set of vibration features to obtain a set of frequency maps; The air tightness analysis module is configured to perform air tightness analysis on the battery top cover based on the set of frequency maps to obtain a detection result.

7. The battery top cover airtightness detection device according to claim 6, characterized in that, The working principle of the preset feature extraction module includes: Performing a preset convolution operation on each sub-vibration signal in the set of sub-vibration signals to obtain a set of multi-scale time domain features; Splitting the set of multi-scale time domain features by scale dimension to obtain a set of single-scale features; Performing variational modal decomposition on a target single-scale feature, extracting a preset number of intrinsic modal components, and performing screening and waveform retention on the preset number of intrinsic modal components to obtain a joint time domain feature representing a specific vibration mode of the scale; the target single-scale feature is any one of the single-scale features in the set of single-scale features; Calculating the mean, variance, kurtosis, and energy proportion of all joint time domain features to obtain a set of scale perception vectors; Taking the scale perception vectors in the set of scale perception vectors as nodes, calculating the feature similarity between any two nodes as edge weights, and constructing a fully connected scale relationship graph; Performing iterative graph convolution propagation on the scale relationship graph, aggregating the features of each node and its first-order neighborhood nodes, and redistributing the node state according to the edge weight, and finally outputting a set of adaptive fusion weight coefficients containing global topological dependency; After weighting and summing the adaptive fusion weight coefficients and the set of multi-scale time domain features, performing time alignment, standardization processing, and length normalization to obtain a set of vibration features.

8. The battery top cover airtightness detection device according to claim 7, characterized in that, The process of the preset convolution operation includes: The sub-vibration signal set is subjected to multi-channel filtering processing through a one-dimensional convolution layer to obtain a multi-channel time sequence feature set; Deep layer features are obtained by performing feature mapping on target multi-channel time sequence features through multiple cascaded convolution layers; the target multi-channel time sequence features are any one of the multi-channel time sequence features in the multi-channel time sequence feature set; The deep layer features in the deep layer feature set are integrated across channels to obtain multi-scale time domain features; All multi-scale time domain features are counted to obtain a multi-scale time domain feature set.

9. The battery top cover airtightness detection device according to claim 6, characterized in that, The air tightness analysis module comprises: A frequency activation module is configured to perform weighted summation on each frequency diagram in the frequency diagram set by using a preset weight vector corresponding to each defect category, and generate a frequency activation diagram set for each candidate fault category, which represents the feature frequency response intensity of the category; A candidate key activation diagram generation module is configured to filter out frequency activation diagrams with activation intensity greater than a preset activation threshold from the frequency activation diagram set corresponding to each candidate defect category along the frequency dimension to obtain a candidate key activation diagram set under the defect category; A similarity calculation module is configured to perform similarity calculation on each two candidate key activation diagrams in the candidate key activation diagram set to obtain a similarity matrix; A key activation diagram generation module is configured to filter out pairs of candidate key activation diagrams exceeding a preset similarity threshold according to the similarity matrix, and retain the candidate key activation diagram with higher average activation intensity to finally obtain a key activation diagram set; A core frequency band generation module is configured to extract a significant continuous frequency band with activation intensity exceeding an adaptive threshold from a target key activation diagram along the frequency dimension to obtain a core frequency band corresponding to the key activation diagram; the target key activation diagram is any one of the key activation diagrams in the key activation diagram set; A detection result generation module is configured to perform defect judgment on the battery cover according to all core frequency bands to obtain a detection result.

10. The battery top cover airtightness detection device according to claim 9, characterized in that, The detection result generation module comprises: An intensity comprehensive value calculation module is configured to count the frequency point number and average activation intensity of all core frequency bands to obtain an intensity comprehensive value representing the abnormality scale; A frequency band span value calculation module is configured to measure the coverage width of all core discriminant frequency bands on the frequency axis to obtain a frequency band span value representing the abnormality distribution width; A weighting module is configured to weight the intensity comprehensive value and the frequency band span value to obtain an abnormality score; A judgment module is configured to compare the abnormality score with a preset judgment threshold, and if the abnormality score exceeds the preset judgment threshold, it is determined that the battery cover air tightness is unqualified.

Images