Lithium battery shell toughness detection system

Abstract

The invention relates to the technical field of lithium battery detection, and discloses a lithium battery shell toughness detection system. The system comprises a multi-channel data acquisition module for acquiring multi-dimensional deformation data of a lithium battery shell under an impact load; and the signal processing and analyzing module is used for performing dynamic baseline calibration processing on the data and inputting the data into the toughness analyzing and processing layer. The toughness analysis processing layer comprises a signal preprocessing module and a feature analysis module, the feature analysis module comprises an excitation response layer, a waveform reconstruction layer and a toughness judgment layer, and time-frequency domain alignment, dynamic coupling relation modeling and multi-dimensional fusion are carried out respectively to generate a detection result. The method comprises the steps of data acquisition, calibration processing, feature extraction, result generation and the like. The method can accurately detect the toughness of the lithium battery shell, improves the detection precision and efficiency, adapts to complex working conditions, can predict the crack propagation path, and guarantees the safe and reliable operation of the lithium battery.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery detection, and particularly to a system for detecting the toughness of a lithium battery casing. Background Art

[0002] During the production and application of lithium batteries, the quality of the lithium battery casing is crucial, and its toughness is directly related to the safety and service life of the battery. With the wide application of lithium batteries in fields such as electric vehicles and energy storage systems, the requirements for the toughness of lithium battery casings are also getting higher and higher.

[0003] From the perspective of the production and manufacturing process, the current production scale of lithium batteries is continuously expanding, and the production process is becoming increasingly complex. There are differences in the material quality, processing accuracy, etc. of lithium battery casings in different batches. Traditional means of detecting casing toughness, such as manual sampling inspection, simple physical tests, etc., cannot meet the detection requirements of large scale and high precision. Manual sampling inspection is not only inefficient, but also the detection results are easily affected by subjective factors, making it difficult to ensure the accuracy and consistency of detection. Simple physical tests, such as using tools like hammers for percussion testing, can only roughly evaluate the toughness of the lithium battery casing, and cannot obtain multi-dimensional deformation data of the casing under impact loads, making it difficult to comprehensively understand the toughness characteristics of the casing.

[0004] In actual usage scenarios, lithium batteries will face various complex working conditions. During the driving process of electric vehicles, the battery will be affected by various factors such as vibration, impact, and temperature changes. If the toughness of the lithium battery casing is insufficient, cracks may appear in the casing under long-term vibration and impact, which may then lead to leakage of internal battery materials and trigger safety accidents. In energy storage systems, frequent charging and discharging of lithium batteries will generate thermal stress, which acts on the casing together with external environmental stress. If the casing has poor toughness, it will also reduce the service life of the battery and increase the maintenance cost of the energy storage system.

[0005] In addition, with the continuous progress of technology, the performance of lithium batteries is constantly improving, and the energy density is getting higher and higher. This requires the lithium battery casing to have better toughness while ensuring sufficient strength to adapt to the internal pressure changes of the battery and the influence of the external environment. However, existing detection technologies are difficult to accurately evaluate the toughness performance of the casing under such complex working conditions, and cannot provide effective data support for the design and production of lithium batteries, restricting the development of the lithium battery industry. Therefore, it is urgent to develop a system and method that can accurately detect the toughness of lithium battery casings, obtain multi-dimensional deformation data, and can adapt to the detection requirements of complex working conditions. Summary of the Invention

[0006] The purpose of the present invention is to provide a system for detecting the toughness of a lithium battery casing to solve the problems raised in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solutions: a lithium battery shell toughness detection system, the system comprising: A multi-channel data acquisition module for obtaining multi-dimensional deformation data of the target lithium battery shell under impact load, the multi-dimensional deformation data including a first detection sequence corresponding to vibration response data, a second detection sequence corresponding to stress distribution data, and a third detection sequence corresponding to deformation trajectory data, the vibration response data including a first excitation waveform generated by a multi-band excitation device and a second feedback waveform collected by a distributed sensor; A signal processing and analysis module for performing dynamic baseline calibration processing on the multi-dimensional deformation data and inputting it into a toughness analysis processing layer for feature extraction, and generating a toughness detection result of the target lithium battery shell according to the output result of the toughness analysis processing layer; The toughness analysis processing layer includes a signal preprocessing module and a feature analysis module. Among them, the signal preprocessing module is used for waveform segmentation and noise suppression of the original detection data stream, and the feature analysis module is obtained by jointly modeling based on historical waveform data and historical deformation data of multiple historical detection cycles; the feature analysis module includes an excitation response layer, a waveform reconstruction layer, and a toughness determination layer connected in sequence.

[0008] Preferably, the excitation response layer is used for performing time-frequency domain alignment processing on multiple detection sequences included in the original detection data stream to obtain time-series correlation feature data; the waveform reconstruction layer is used for modeling the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain reconstructed waveform feature data; the toughness determination layer is used for performing multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data to generate a toughness detection result.

[0009] Preferably, the modeling of the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain reconstructed waveform feature data includes: Adopting a dynamic baseline calibration algorithm to identify key deformation nodes in the time-series correlation feature data, and determining deformation correlation sequences corresponding to each detection sequence based on the load type corresponding to each key deformation node; Calculating the waveform similarity between deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences, and determining the reconstructed waveform feature data between the any two detection sequences based on the waveform similarity.

[0010] Preferably, the calculating the waveform similarity between deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences includes: When there are differences in the number of deformation nodes in the deformation correlation sequences corresponding to any two of the detection sequences, virtual node compensation is performed based on the load type corresponding to the end deformation node in the one with fewer deformation nodes, and the waveform similarity between the deformation nodes with the same load type is calculated based on the compensated data.

[0011] Preferably, the signal preprocessing module is specifically configured to: Perform equally spaced division on the waveform information included in the first detection sequence, the second detection sequence, and the third detection sequence according to a preset segmentation rule to obtain a standardized first detection sequence, a standardized second detection sequence, and a standardized third detection sequence; Perform real-time filtering on the standardized first detection sequence and the standardized second detection sequence by using a dynamic threshold adjustment method, and perform static noise reduction on the standardized third detection sequence by using a fixed threshold filtering method to generate a first processing sequence, a second processing sequence, and a third processing sequence; wherein, the first processing sequence includes a filtered first excitation waveform and a filtered second feedback waveform.

[0012] Preferably, the signal preprocessing module is further configured to: Calculate the coupling coefficient between the filtered first excitation waveform and the filtered second feedback waveform within a historical detection period; Predict the predicted deformation amount of the filtered second feedback waveform in the real-time detection period according to the coupling coefficient and the load characteristics of the filtered first excitation waveform in the real-time detection period; Generate target excitation response data according to the filtered second feedback waveform and its predicted deformation amount, and use the detection sequence corresponding to the target excitation response data as the first processing sequence.

[0013] Preferably, the waveform reconstruction layer specifically includes: A feature association unit, configured to perform a load path analysis on each detection sequence included in the time series association feature data respectively to extract a corresponding stress propagation chain from each detection sequence; A waveform matching unit, configured to perform dynamic mapping on the stress propagation chain extracted from each detection sequence and the corresponding time series feature data to generate reconstructed waveform feature data.

[0014] Preferably, the waveform reconstruction layer further includes: A harmonic suppression unit, configured to perform harmonic component elimination processing on the reconstructed waveform feature data.

[0015] Preferably, the toughness determination layer specifically includes: The multi-dimensional fusion unit includes multiple feature fusion nodes, and each feature fusion node is connected to each detection sequence in the reconstructed waveform feature data and the time-series correlation feature data through an associated configuration; The dynamic association optimization unit is used to adjust the associated configuration through a dynamic association optimization algorithm to minimize the error between the toughness detection result and the actual deformation data; The failure mode recognition unit is used to predict the crack propagation path based on the reconstructed waveform feature data and the time-series correlation feature data, and generate a toughness detection result.

[0016] Preferably, the dynamic threshold adjustment method specifically includes: Generating an adaptive filtering threshold based on the noise spectrum distribution in the real-time detection environment; Using a sliding window mechanism to perform segmented filtering processing on the standardized first detection sequence.

[0017] Preferably, the harmonic component elimination processing includes: Extracting the components in the reconstructed waveform feature data whose frequencies are integer multiples of the fundamental frequency, and filtering out the extracted components from the original data after phase inversion and superposition.

[0018] Preferably, the crack propagation path prediction includes: Establishing a crack growth rate model including the stress concentration coefficient; Calculating the crack bifurcation probability according to the stress gradient distribution in the reconstructed waveform feature data.

[0019] Preferably, the present invention further includes a method for detecting the toughness of a lithium battery housing, and the method includes: Obtaining multi-dimensional deformation data of the target lithium battery housing under impact load, where the multi-dimensional deformation data includes a first detection sequence corresponding to vibration response data, a second detection sequence corresponding to stress distribution data, and a third detection sequence corresponding to deformation trajectory data; Performing dynamic baseline calibration processing on the multi-dimensional deformation data to generate a standard detection data stream; Inputting the standard detection data stream into the toughness analysis processing layer for feature extraction; Generate the toughness detection result of the target lithium battery shell according to the output result of the toughness analysis and processing layer; the toughness analysis and processing layer includes a signal preprocessing module and a feature analysis module, wherein the signal preprocessing module is used to segment the waveform and suppress the noise of the standard detection data stream, and the feature analysis module is obtained by jointly modeling based on the historical waveform data and historical deformation data of multiple historical detection cycles; the feature analysis module includes an excitation response layer, a waveform reconstruction layer, and a toughness determination layer connected in sequence, and the excitation response layer is used to perform time-frequency domain alignment processing on multiple detection sequences included in the standard detection data stream to obtain time-series correlation feature data; the waveform reconstruction layer is used to model the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain reconstructed waveform feature data; the toughness determination layer is used to perform multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data to generate a toughness detection result; Among them, the modeling of the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain the reconstructed waveform feature data includes: Use the dynamic baseline calibration algorithm to identify the key deformation nodes in the time-series correlation feature data, and determine the deformation correlation sequence corresponding to each detection sequence based on the load type corresponding to each key deformation node; Calculate the waveform similarity between the deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences, and determine the reconstructed waveform feature data between the any two detection sequences based on the waveform similarity; the calculation of the waveform similarity between the deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences includes: When there is a difference in the number of deformation nodes in the deformation correlation sequences corresponding to the any two detection sequences, perform virtual node compensation based on the load type corresponding to the terminal deformation node in the one with fewer deformation nodes, and calculate the waveform similarity between the deformation nodes with the same load type based on the compensated data.

[0020] Preferably, the dynamic correlation optimization algorithm specifically includes: Establish an error distribution matrix between the toughness detection result and the actual deformation data; Iteratively adjust the weight parameters in the correlation configuration by the gradient descent method until the error converges.

[0021] Compared with the prior art, the beneficial effects of the present invention are: In terms of detection accuracy, the system obtains multi-dimensional deformation data of the target lithium battery shell under impact load through a multi-channel data acquisition module, covering vibration response data, stress distribution data, and deformation trajectory data. These rich data can comprehensively reflect the changes in the mechanical properties of the shell during the impact process. For example, the first excitation waveform and the second feedback waveform in the vibration response data accurately record the dynamic response of the shell under the excitation. The signal processing and analysis module performs dynamic baseline calibration on the multi-dimensional deformation data, effectively eliminating errors and noise interference during data acquisition and ensuring the accuracy of the data. The feature analysis module in the toughness analysis and processing layer performs joint modeling based on the historical waveform data and historical deformation data of multiple historical detection cycles, making the detection results more accurate and reliable. Compared with traditional detection methods, it can more accurately evaluate the toughness of the lithium battery shell.

[0022] In terms of detection efficiency, the system adopts multi-channel data acquisition and parallel processing technology. The multi-channel data acquisition module can simultaneously collect data in multiple dimensions, greatly shortening the data acquisition time. The signal processing and analysis module and each module in the toughness analysis and processing layer cooperate with each other to achieve rapid data processing and analysis. Taking the excitation response layer as an example, it can perform time-frequency domain alignment processing on multiple detection sequences contained in the original detection data stream to efficiently obtain time-series correlation feature data. The entire detection process realizes automation and intelligence, reduces manual intervention, and greatly improves the detection efficiency, meeting the rapid detection requirements in the large-scale lithium battery production process.

[0023] From the perspective of adaptability, the present invention fully considers the complex working conditions in actual applications. The multi-band excitation device can simulate impact loads of various different frequencies and intensities, making the detection environment closer to the working conditions of lithium batteries in actual use. Each module in the toughness analysis and processing layer processes and analyzes different types of data, and can effectively handle complex detection requirements. The waveform reconstruction layer models the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence, and can accurately reflect the mechanical response of the shell under complex loads. This adaptability to complex working conditions makes the detection results more valuable in practical applications and can provide strong support for the design, production, and quality control of lithium batteries.

[0024] In terms of fault prediction and prevention, the failure mode recognition unit in the resilience determination layer predicts the crack propagation path based on the reconstructed waveform feature data and the time-series correlation feature data. By establishing a crack growth rate model that includes the stress concentration factor and calculating the crack bifurcation probability according to the stress gradient distribution in the reconstructed waveform feature data, potential safety hazards in the lithium battery housing can be detected in advance. This helps the manufacturer adjust the production process in a timely manner and improve product quality. It also enables users to take preventive measures in advance to avoid battery failure and safety accidents caused by housing rupture, ensuring the safe use and reliable operation of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is the working principle diagram of the lithium battery housing resilience detection system according to the present invention; Figure 2 is the flowchart for the waveform reconstruction layer to determine the reconstructed waveform feature data; Figure 3 is the flowchart for processing the difference in the number of nodes when calculating the waveform similarity; Figure 4 is the flowchart for data processing by the signal preprocessing module; Figure 5 is the flowchart for supplementing the first processing sequence generated by the signal preprocessing module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The technical solutions in the embodiments 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 a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0027] Please refer to Figures 1 - 5 , the present invention provides a lithium battery housing resilience detection system, and the specific implementation steps are as follows: The system mainly consists of a multi-channel data acquisition module and a signal processing and analysis module. When detecting the resilience of the lithium battery housing, the multi-channel data acquisition module takes effect first. This module generates a specific first excitation waveform through a multi-band excitation device, and this excitation waveform is applied to the target lithium battery housing. At the same time, the distributed sensors collect the second feedback waveform after the housing is excited, and these two waveforms constitute the vibration response data. In addition, the multi-channel data acquisition module also acquires the stress distribution data and the deformation trajectory data, which respectively correspond to the second detection sequence and the third detection sequence. Thus, the multi-dimensional deformation data of the target lithium battery housing under impact load is completely obtained.

[0028] After obtaining the multi-dimensional deformation data, the signal processing and analysis module starts to work. It first performs dynamic baseline calibration on these data to eliminate the biases and interferences in the data. The processed data analysis is input into the toughness analysis and processing layer, which includes a signal preprocessing module and a feature analysis module. The signal preprocessing module segments the waveform and suppresses noise in the original detection data stream, while the feature analysis module is obtained by jointly modeling based on the historical waveform data and historical deformation data of multiple historical detection cycles. After feature extraction, the toughness detection result of the target lithium battery shell can be generated according to the output result of the toughness analysis and processing layer.

[0029] The present invention will be further described in detail below with specific embodiments. Embodiment 1

[0030] This embodiment mainly elaborates on the working processes of the excitation response layer, waveform reconstruction layer, and toughness determination layer, which will be described in detail below with specific examples.

[0031] After receiving the original detection data stream, the excitation response layer performs time-frequency domain alignment processing on multiple detection sequences included therein. This process unifies different detection sequences in the time and frequency dimensions to obtain time-series correlation feature data. For example, the first detection sequence corresponding to the vibration response data and the second detection sequence corresponding to the stress distribution data can more clearly show their correlation in time after time-frequency domain alignment processing.

[0032] The waveform reconstruction layer models the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence. First, a dynamic baseline calibration algorithm is used to identify the key deformation nodes in the time-series correlation feature data, and these nodes represent the positions where significant deformations occur in the lithium battery shell under impact loads. Then, based on the load type corresponding to each key deformation node, the deformation correlation sequences corresponding to each detection sequence are determined. Next, the waveform similarity between the deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences is calculated. When there is a difference in the number of deformation nodes in the deformation correlation sequences corresponding to two detection sequences, virtual node compensation is performed based on the load type of the end deformation node in the one with fewer deformation nodes, and then the waveform similarity is calculated based on the compensated data. According to these waveform similarities, the reconstructed waveform feature data between any two detection sequences can be determined.

[0033] The toughness determination layer performs multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data. Through multi-dimensional fusion, data from different dimensions are comprehensively analyzed to generate the toughness detection result. This process can more comprehensively and accurately evaluate the toughness of the lithium battery shell.

[0034] Suppose there is a new type of lithium battery, whose outer shell adopts new materials and design processes. To ensure the reliability of the outer shell during actual use, it is necessary to accurately detect its toughness.

[0035] The multi-channel data acquisition module starts to work. A series of excitations with different frequencies and intensities are applied to the outer shell of the lithium battery through a multi-band excitation device to generate a first excitation waveform. For example, to simulate the bumps and collisions that may occur during transportation, the excitation frequency is set to vary from 5 Hz at low frequency to 50 Hz at high frequency, and the intensity is adjusted accordingly. The distributed sensors collect the feedback signals of the outer shell in real time to form a second feedback waveform. At the same time, the stress distribution data and the deformation trajectory data are collected to obtain a second detection sequence and a third detection sequence respectively.

[0036] After receiving these original detection data streams, the excitation response layer starts to perform time-frequency domain alignment processing on the multiple detection sequences contained therein. For example, the change of the vibration response data in the first detection sequence on the time axis may not be completely synchronized with the stress distribution data in the second detection sequence. The excitation response layer will unify these data in the time and frequency dimensions, and display the vibration response data and the stress distribution data on the same time scale. After the time-frequency domain alignment processing, it is clearly found that when the vibration response has a peak at a certain time point, the stress distribution also has obvious changes at a similar time point, thus obtaining time-related time-series correlation feature data.

[0037] The waveform reconstruction layer starts to work. It uses a dynamic baseline calibration algorithm to identify the key deformation nodes in the time-series correlation feature data. Taking a certain corner area of the outer shell of the lithium battery as an example, when it is subjected to a specific excitation, the deformation amount of this area changes greatly, and the corresponding node of this area is identified as a key deformation node. Based on the load type corresponding to each key deformation node, the deformation correlation sequences corresponding to each detection sequence are determined. Suppose the load type corresponding to the key deformation node is impact force, then in the first detection sequence (vibration response data), the second detection sequence (stress distribution data), and the third detection sequence (deformation trajectory data), find all the deformation nodes related to the impact force to form their respective deformation correlation sequences.

[0038] Next, calculate the waveform similarity between the deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences. For example, when comparing the deformation correlation sequences of the first detection sequence and the second detection sequence, it is found that there is a difference in the number of deformation nodes in the two sequences. The deformation correlation sequence of the first detection sequence has 10 nodes, while the deformation correlation sequence of the second detection sequence has only 8 nodes. At this time, based on the load type (impact force) corresponding to the end deformation node in the second detection sequence, virtual node compensation is performed, and 2 virtual nodes are added at the end of the second detection sequence to make the number of nodes in the two sequences the same. Then, through a specific calculation method, calculate the waveform similarity between the deformation nodes with the same load type (impact force). Based on these waveform similarities, determine the reconstructed waveform feature data between the first detection sequence and the second detection sequence.

[0039] The toughness determination layer performs multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data. For example, comprehensively analyze the casing deformation trend reflected in the reconstructed waveform feature data and the stress changes, vibration responses, etc. at different time points in the time-series correlation feature data. Through multi-dimensional fusion, judge the toughness of the lithium battery casing under the current detection conditions. If, during a certain period, the vibration response is strong and the stress is concentrated, and at the same time the reconstructed waveform features show a large potential deformation risk of the casing, then it is determined that the toughness of the casing is relatively weak in this area and time period, and it may be necessary to further improve the design or strengthen the material properties, thereby generating a complete toughness detection result. Embodiment 2

[0040] The signal preprocessing module equally divides the waveform information included in the first detection sequence, the second detection sequence, and the third detection sequence according to the preset segmentation rules, which can convert the complex waveform information into a standardized format for subsequent processing. After division, a standardized first detection sequence, a standardized second detection sequence, and a standardized third detection sequence are obtained.

[0041] For the standardized first detection sequence and the standardized second detection sequence, a dynamic threshold adjustment method is used for real-time filtering. This method generates an adaptive filtering threshold based on the noise spectrum distribution in the real-time detection environment, and then uses a sliding window mechanism to perform segment-by-segment filtering on the standardized first detection sequence. This dynamic filtering method can better adapt to different detection environments and effectively remove noise. For the standardized third detection sequence, a fixed threshold filtering method is used for static noise reduction. After these processes, a first processed sequence, a second processed sequence, and a third processed sequence are generated, where the first processed sequence includes the filtered first excitation waveform and the filtered second feedback waveform.

[0042] In addition, the signal preprocessing module also calculates the coupling coefficient between the filtered first excitation waveform and the filtered second feedback waveform within the historical detection period. Based on this coupling coefficient and the load characteristics of the filtered first excitation waveform in the real-time detection period, it predicts the predicted deformation amount of the filtered second feedback waveform in the real-time detection period. Finally, it generates target excitation response data based on the filtered second feedback waveform and its predicted deformation amount, and uses the detection sequence corresponding to this data as the first processing sequence.

[0043] Taking a new lithium battery designed for a certain brand of drone as an example, during the production process, it is necessary to strictly detect the toughness of its shell. At the detection site, the multi-channel data acquisition module has obtained multi-dimensional deformation data of the target lithium battery shell under impact loads. Among them, the first detection sequence corresponding to the vibration response data, the second detection sequence corresponding to the stress distribution data, and the third detection sequence corresponding to the deformation trajectory data become the input data of the signal preprocessing module.

[0044] The signal preprocessing module first equally divides the waveform information contained in these three detection sequences according to the preset segmentation rules. Assuming that the waveform is divided every 0.01 seconds as preset, the purpose of this is to convert the complex and continuously changing waveform information into small pieces of data with the same time interval, achieving standardized processing. After division, a standardized first detection sequence, a standardized second detection sequence, and a standardized third detection sequence are obtained. This standardized processing makes subsequent data processing more regular and comparable, just like cutting materials of different lengths and shapes into small pieces of the same specification for further processing.

[0045] For the standardized first detection sequence and the standardized second detection sequence, the signal preprocessing module uses the dynamic threshold adjustment method for real-time filtering. In the actual detection environment, there are various interference factors, such as electromagnetic interference generated by other electronic devices near the detection equipment. These interferences will make the collected data contain noise. The signal preprocessing module generates an adaptive filtering threshold based on the noise spectrum distribution in the real-time detection environment. For example, if there is strong electromagnetic interference with a frequency of 60Hz in the detection environment, the module will calculate a suitable filtering threshold according to this interference frequency and its intensity, and this threshold can effectively filter out the noise signals near 60Hz.

[0046] The sliding window mechanism is adopted to perform segment-by-segment filtering on the standardized first detection sequence. The size of the sliding window can be set according to the actual situation. Suppose the window size is set to contain 10 data points. The window starts from the starting position of the sequence and moves one data point each time, and the data within the window is filtered. As the window slides, the entire first detection sequence can be effectively filtered. Through this dynamic filtering method, it can better adapt to the noise changes at different times, removing noise while retaining the useful signals to the greatest extent.

[0047] For the standardized third detection sequence, due to its relatively stable data characteristics, a fixed threshold filtering method is used for static noise reduction. For example, according to historical experience and previous tests, a fixed filtering threshold of 0.5 is determined (this threshold is determined according to the amplitude range and noise level of the data). Each data point in the third detection sequence is judged. If the amplitude of the data point is less than 0.5, it is considered a noise point and corresponding processing is performed, so as to achieve the purpose of removing noise.

[0048] After the above filtering process, a first processed sequence, a second processed sequence, and a third processed sequence are generated, where the first processed sequence contains the filtered first excitation waveform and the filtered second feedback waveform.

[0049] In addition, the signal preprocessing module also calculates the coupling coefficient between the filtered first excitation waveform and the filtered second feedback waveform in the historical detection period. Suppose in the previous 100 detections, the data of the first excitation waveform and the second feedback waveform were recorded each time. Through a specific calculation method (such as calculating the correlation between the two waveforms), the coupling coefficient between the two waveforms in the historical detection period is obtained as 0.8. This coupling coefficient reflects the degree of association between the two waveforms.

[0050] According to this coupling coefficient and the load characteristics of the filtered first excitation waveform in the real-time detection period, the predicted deformation amount of the filtered second feedback waveform in the real-time detection period is predicted. For example, it is known that the load intensity of the first excitation waveform in the current real-time detection period has increased by 20%. Combining with the coupling coefficient of 0.8, it is predicted that the deformation amount of the second feedback waveform will increase by 16% (calculation method: 20%×0.8).

[0051] Finally, target excitation response data is generated according to the filtered second feedback waveform and its predicted deformation amount, and the detection sequence corresponding to this data is used as the first processed sequence. Through this series of operations, the signal preprocessing module effectively improves the data quality and provides a reliable data basis for more accurate analysis of the toughness of the lithium battery shell in the future. Embodiment 3

[0052] This embodiment focuses on explaining the internal structure and working principle of the waveform reconstruction layer. The waveform reconstruction layer specifically includes a feature correlation unit, a waveform matching unit, and a harmonic suppression unit.

[0053] The feature correlation unit performs a load path analysis on each detection sequence contained in the time-sequence correlation feature data. Through the analysis, the corresponding stress propagation chain is extracted from each detection sequence. For example, in the detection sequence corresponding to the stress distribution data, the feature correlation unit can find the propagation path of stress on the lithium battery housing, determine the stress concentration area and the propagation direction.

[0054] The waveform matching unit dynamically maps the stress propagation chain extracted from each detection sequence to the corresponding time-sequence feature data. Through this mapping, reconstructed waveform feature data is generated. This dynamic mapping method can more accurately reflect the correlation relationship between different detection sequences, thereby obtaining more accurate reconstructed waveform feature data.

[0055] The harmonic suppression unit performs harmonic component elimination processing on the reconstructed waveform feature data. Specifically, it extracts the components in the reconstructed waveform feature data whose frequencies are integer multiples of the fundamental frequency, and after phase-inverting and superimposing these extracted components, filters them from the original data. By eliminating the harmonic components, the reconstructed waveform feature data can be made more pure, reducing interference and improving the accuracy of subsequent analysis.

[0056] Suppose it is necessary to perform a toughness detection on the housing of a certain model of square lithium battery, which is widely used in the electric vehicle battery pack, and its toughness directly affects the safety and service life of the battery pack.

[0057] After completing multi-channel data acquisition and preliminary signal processing, the data enters the toughness determination layer. The multi-dimensional fusion unit in the toughness determination layer starts to work. This unit contains multiple feature fusion nodes. Taking three of these feature fusion nodes as an example, the first node is responsible for fusing some features of the vibration response data (from the first detection sequence) and the stress distribution data (from the second detection sequence); the second node mainly correlates the relevant features of the stress distribution data and the deformation trajectory data (from the third detection sequence); the third node focuses on integrating the features of the vibration response data and the deformation trajectory data.

[0058] These feature fusion nodes are connected to each detection sequence in the reconstructed waveform feature data and the time-sequence correlation feature data through specific correlation configurations. For example, the first feature fusion node is connected to the response features in a specific frequency band of the vibration response data and the changing features of the high-stress area in the stress distribution data. When there are abnormal fluctuations in the vibration response in a certain high-frequency band and there are rapid changes in the stress distribution in a specific area, the first feature fusion node will synthesize this information and initially judge the toughness of this area.

[0059] Next, the dynamic association optimization unit comes into play. It adjusts the association configuration through the dynamic association optimization algorithm. First, an error distribution matrix of the toughness detection results and the actual deformation data is established. Suppose after a preliminary detection, it is found that the detection results of a certain area show good toughness of the shell, but in the actual simulation usage test, slight deformation occurs in this area. This generates an error, and these errors are sorted according to the corresponding detection areas and features to form an error distribution matrix.

[0060] Then, the weight parameters in the association configuration are iteratively adjusted by the gradient descent method until the error converges. For example, for the area with the above-mentioned error, if it is found that the influence weight of the vibration response data on the toughness judgment is too high, while the weight of the stress distribution data is relatively low, the gradient descent method will gradually reduce the weight of the vibration response data and increase the weight of the stress distribution data at the same time. After multiple iterations, the error between the toughness detection results and the actual deformation data gradually decreases until it converges to an acceptable range.

[0061] Finally, the failure mode recognition unit predicts the crack propagation path based on the reconstructed waveform feature data and the time-series association feature data. It first establishes a crack growth rate model including the stress concentration factor. Suppose according to the material properties and structural characteristics of the lithium battery shell, the calculation method of the stress concentration factor is determined, as well as the relationship between the crack growth rate and the stress intensity factor range , material constant , exponent is (this formula is only for example illustration, and the actual model will be more complex). Among them, represents the crack growth rate, with the unit of mm / cycle; and are material-related constants determined through material tests; is the stress intensity factor range, which reflects the influence of stress changes on crack propagation.

[0062] The crack bifurcation probability is calculated based on the stress gradient distribution in the reconstructed waveform feature data. If in a certain area of the shell, the reconstructed waveform feature data shows a large stress gradient change, through a specific algorithm combined with the stress concentration factor and the crack growth rate model, it is calculated that the crack bifurcation probability in this area is relatively high. Combining this information, a toughness detection result is generated, such as judging that the toughness of the lithium battery shell is weak in some key areas, there is a high risk of crack propagation, and improvements need to be made in the production process or material selection. Example 4

[0063] This embodiment details the composition and working mode of the toughness determination layer. The toughness determination layer mainly consists of a multi-dimensional fusion unit, a dynamic association optimization unit, and a failure mode recognition unit.

[0064] The multi-dimensional fusion unit includes multiple feature fusion nodes, and each feature fusion node is connected to each detection sequence in the reconstructed waveform feature data and the time-series association feature data through association configuration. These feature fusion nodes will fuse the data of different detection sequences, comprehensively analyze the data from multiple dimensions, and provide more comprehensive information for subsequent determination.

[0065] The dynamic association optimization unit will adjust the association configuration through the dynamic association optimization algorithm. The specific operation is to first establish an error distribution matrix between the toughness detection result and the actual deformation data, and then iteratively adjust the weight parameters in the association configuration through the gradient descent method until the error converges. In this way, the association configuration can be continuously optimized, making the toughness detection result closer to the actual situation and improving the detection accuracy.

[0066] The failure mode recognition unit will predict the crack propagation path based on the reconstructed waveform feature data and the time-series association feature data. It first establishes a crack growth rate model including the stress concentration coefficient, and then calculates the crack bifurcation probability according to the stress gradient distribution in the reconstructed waveform feature data. Through these calculations and analyses, a toughness detection result is generated, which can predict the possible failure modes of the lithium battery shell and provide an important basis for product quality assessment.

[0067] Suppose a toughness detection is performed on the shell of a square lithium battery, which is mainly used in electric vehicles, and the toughness of its shell is directly related to the safety and stability of the battery under complex working conditions.

[0068] The multi-channel data acquisition module first applies impact loads that may be encountered during the driving of an electric vehicle to the shell of this lithium battery, such as the impact forces generated by sudden braking and bumpy roads. Different frequencies and intensities of excitation are generated through the multi-band excitation device, and the distributed sensors collect vibration response data, stress distribution data, and deformation trajectory data, respectively forming the first detection sequence, the second detection sequence, and the third detection sequence. After the dynamic baseline calibration and preliminary processing by the signal processing and analysis module, these data enter the toughness determination layer.

[0069] The multi-dimensional fusion unit of the toughness determination layer includes multiple feature fusion nodes. Each feature fusion node is connected to each detection sequence in the reconstructed waveform feature data and the time-series correlation feature data. For example, there is a feature fusion node dedicated to fusing the first detection sequence (vibration response data) and the vibration-related part of the reconstructed waveform feature data. It receives data from both parts simultaneously, combining information such as vibration frequency and amplitude in the vibration response data with the waveform features in the reconstructed waveform feature data that reflect the impact of vibration on the outer shell. Through this fusion, data can be comprehensively analyzed from multiple dimensions. For example, it can be understood which parts of the outer shell are prone to large deformations at a specific vibration frequency, providing a more comprehensive basis for accurately evaluating the toughness of the outer shell in the follow-up.

[0070] The dynamic association optimization unit starts to play a role. It first establishes an error distribution matrix between the toughness detection result and the actual deformation data. Assume that in previous multiple detections, some samples of known actual deformation data have been accumulated. The toughness detection result obtained from this detection is compared with these actual deformation data, and an error distribution matrix is constructed with the error of each data point as an element. For example, at a corner of the outer shell, the detection result shows a deformation amount of 0.5 mm, while the actually measured deformation amount is 0.6 mm, then the error at this position is 0.1 mm, and all similar error values are organized in matrix form.

[0071] Then, the dynamic association optimization unit iteratively adjusts the weight parameters in the association configuration through the gradient descent method. The gradient descent method is a commonly used optimization algorithm. It calculates the gradient of the error function with respect to the weight parameters and gradually adjusts the weight parameters along the opposite direction of the gradient, making the error gradually decrease. For example, the current value of a certain weight parameter is 0.8. After calculating the gradient, according to the rules of the gradient descent method, it is adjusted to 0.75. This process is repeated continuously until the error converges. After multiple iterative adjustments, the association configuration is optimized, making the toughness detection result closer to the actual situation and improving the accuracy of the detection.

[0072] The failure mode recognition unit predicts the crack propagation path based on the reconstructed waveform feature data and the time-series correlation feature data. It first establishes a crack growth rate model that includes the stress concentration factor. For example, according to the material properties and structural characteristics of the square lithium battery outer shell, the relationship between the stress concentration factor and the crack growth rate is determined. Assume the stress concentration factor is K and the crack growth rate is v, and a functional relationship between the two is established through experimental data and theoretical analysis. 。

[0073] Next, the crack bifurcation probability is calculated based on the stress gradient distribution in the reconstructed waveform feature data. In a certain area of the outer shell, the reconstructed waveform feature data shows a large stress gradient, indicating that the stress changes violently in this area. By analyzing the stress gradient distribution and combining the previously established model and related algorithms, the probability of crack bifurcation in this area is calculated. If the calculated probability is high, it means that crack bifurcation is likely to occur in this area during subsequent use, resulting in the failure of the outer shell. Finally, based on these analysis results, the toughness detection result is generated, providing an important reference for evaluating the quality and reliability of the lithium battery outer shell, so as to timely discover potential problems and take corresponding improvement measures. Example 5

[0074] The multi-channel data acquisition module obtains multi-dimensional deformation data of the target lithium battery outer shell under impact load, including the first detection sequence corresponding to the vibration response data, the second detection sequence corresponding to the stress distribution data, and the third detection sequence corresponding to the deformation trajectory data.

[0075] The signal processing and analysis module performs dynamic baseline calibration processing on the multi-dimensional deformation data to generate a standard detection data stream. Then, the standard detection data stream is input into the toughness analysis and processing layer. In the toughness analysis and processing layer, the signal preprocessing module first performs waveform segmentation and noise suppression on the standard detection data stream to generate the first processing sequence, the second processing sequence, and the third processing sequence.

[0076] The feature analysis module starts to work. The excitation response layer performs time-frequency domain alignment processing on multiple detection sequences included in the standard detection data stream to obtain time-series correlation feature data. The waveform reconstruction layer models the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain the reconstructed waveform feature data. In this process, the feature correlation unit, the waveform matching unit, and the harmonic suppression unit work together. Finally, the toughness determination layer performs multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data. The multi-dimensional fusion unit, the dynamic correlation optimization unit, and the failure mode recognition unit work together to generate the toughness detection result of the target lithium battery outer shell. Through such a complete process, the toughness detection of the lithium battery outer shell can be completed efficiently and accurately.

[0077] Suppose a lithium battery manufacturing enterprise develops a new model of lithium battery. To ensure that the toughness of its outer shell meets the quality standards, the detection system and method of the present invention are used for detection.

[0078] In the production workshop, the staff place the lithium batteries to be tested on the detection equipment. The multi-channel data acquisition module starts to operate, and the multi-band excitation device applies a series of impact loads simulating actual usage scenarios to the outer shell of the lithium battery. For example, to simulate the drop impact that the lithium battery may receive during transportation, it is first excited with an equivalent impact at a height of 0.5 meters, and then the vibration impacts that may be received in different temperature environments (such as high temperature of 45°C and low temperature of -20°C) are simulated, and the frequency range is set between 20Hz and 100Hz. The distributed sensors quickly collect various data of the outer shell under these impact loads. Among them, the first detection sequence corresponding to the vibration response data records the vibration conditions of the outer shell at different times; the second detection sequence corresponding to the stress distribution data presents the stress changes of each part of the outer shell during the impact; the third detection sequence corresponding to the deformation trajectory data accurately depicts the deformation process of the outer shell.

[0079] After collecting the multi-dimensional deformation data, the signal processing and analysis module is started. It first performs dynamic baseline calibration processing on these data. Due to factors such as electromagnetic interference in the detection environment and small errors of the equipment itself, the original data may have deviations. The dynamic baseline calibration processing is like "calibrating the zero point" of the data, removing these interferences and deviations, and generating a standard detection data stream. For example, by comparing the historical detection data and the real-time environmental parameters, the data is adjusted to ensure the accuracy of subsequent analysis.

[0080] The standard detection data stream is input into the toughness analysis and processing layer. In the toughness analysis and processing layer, the signal preprocessing module plays a role. It equally divides the waveform information included in the first detection sequence, the second detection sequence, and the third detection sequence according to the preset segmentation rules. Assuming that it is preset to divide every 0.05 seconds, in this way, the complex waveform data is transformed into standardized small segment data, which is convenient for subsequent processing. Then, different filtering methods are used to suppress the noise of these standardized sequences. For the standardized first detection sequence and the second detection sequence, according to the noise spectrum distribution in the real-time detection environment, the filtering threshold is dynamically adjusted for real-time filtering. For example, if there is strong electromagnetic interference at 50Hz in the detection environment, the filtering threshold is adjusted accordingly to remove the noise in this frequency band. For the standardized third detection sequence, a fixed threshold filtering method is used for static noise reduction. After these processes, the first processing sequence, the second processing sequence, and the third processing sequence are generated.

[0081] After that, the feature analysis module starts to work. The excitation response layer performs time-frequency domain alignment processing on multiple detection sequences included in the standard detection data stream. For example, the vibration response data in the first detection sequence and the stress distribution data in the second detection sequence may originally have differences in time and frequency. After the time-frequency domain alignment processing, the corresponding relationship between the vibration response and the stress change in time can be clearly seen, so as to obtain the time-series correlation feature data.

[0082] The waveform reconstruction layer models the dynamic coupling relationship between the timing correlation feature data corresponding to each detection sequence. Taking the vibration response data and the stress distribution data as examples, the dynamic baseline calibration algorithm is used to identify the key deformation nodes in the timing correlation feature data. During a single impact process, it is found that the deformation amount of a node in a certain corner of the shell changes significantly, and this node is the key deformation node. Based on the load type corresponding to the key deformation node, the deformation correlation sequences corresponding to each detection sequence are determined, and then the waveform similarity between the deformation nodes with the same load type in any two detection sequences (such as the sequences corresponding to the vibration response data and the stress distribution data) is calculated. When the number of deformation nodes in the two sequences is different, the similarity is calculated after virtual node compensation according to the rules, and finally the reconstructed waveform feature data is obtained.

[0083] The toughness determination layer performs multi-dimensional fusion based on the reconstructed waveform feature data and the timing correlation feature data. Multiple feature fusion nodes in the multi-dimensional fusion unit synthesize and analyze data of different dimensions. The dynamic association optimization unit continuously adjusts the association configuration by establishing an error distribution matrix and using the gradient descent method to make the detection result more accurate. The failure mode recognition unit predicts the failure mode of the shell by establishing a crack growth rate model and calculating the crack bifurcation probability. Finally, the toughness detection result of the target lithium battery shell is generated. If the detection result shows that the toughness of the shell is weak in some areas, the manufacturing enterprise can specifically improve the production process or optimize the shell design to ensure the quality and safety of the lithium battery.

[0084] It should be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variation thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0085] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made therein without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lithium battery shell toughness detection system, characterized in that, Including: A multi-channel data acquisition module for obtaining multi-dimensional deformation data of the target lithium battery shell under impact load. The multi-dimensional deformation data includes a first detection sequence corresponding to vibration response data, a second detection sequence corresponding to stress distribution data, and a third detection sequence corresponding to deformation trajectory data. The vibration response data includes a first excitation waveform generated by a multi-band excitation device and a second feedback waveform collected by a distributed sensor; A signal processing and analysis module for performing dynamic baseline calibration processing on the multi-dimensional deformation data and inputting it into a toughness analysis processing layer for feature extraction, and generating a toughness detection result of the target lithium battery shell according to the output result of the toughness analysis processing layer; The toughness analysis processing layer includes a signal preprocessing module and a feature analysis module. Among them, the signal preprocessing module is used for waveform segmentation and noise suppression of the original detection data stream, and the feature analysis module is obtained by jointly modeling based on historical waveform data and historical deformation data of multiple historical detection cycles; The feature analysis module includes an excitation response layer, a waveform reconstruction layer, and a toughness determination layer connected in sequence.

2. The lithium battery case toughness detection system according to claim 1, wherein The excitation response layer is used for time-frequency domain alignment processing of multiple detection sequences included in the original detection data stream to obtain time-series correlation feature data; the waveform reconstruction layer is used for modeling the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain reconstructed waveform feature data; the toughness determination layer is used for multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data to generate a toughness detection result.

3. The lithium battery shell toughness detection system according to claim 2, characterized in that The modeling of the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain reconstructed waveform feature data includes: Using a dynamic baseline calibration algorithm to identify key deformation nodes in the time-series correlation feature data, and determining deformation correlation sequences corresponding to each detection sequence based on the load type corresponding to each key deformation node; Calculating the waveform similarity between deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences, and determining the reconstructed waveform feature data between the any two detection sequences based on the waveform similarity.

4. The lithium battery case toughness detection system according to claim 3, characterized in that The calculation of the waveform similarity between deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences includes: When there is a difference in the number of deformation nodes in the deformation correlation sequences corresponding to the any two detection sequences, virtual node compensation is performed based on the load type corresponding to the end deformation node in the one with fewer deformation nodes, and the waveform similarity between the deformation nodes with the same load type is calculated based on the compensated data.

5. The lithium battery case toughness detection system according to claim 1, characterized in that The signal preprocessing module is specifically used for: Equally spacing the waveform information included in the first detection sequence, the second detection sequence, and the third detection sequence according to a preset segmentation rule to obtain a standardized first detection sequence, a standardized second detection sequence, and a standardized third detection sequence; Perform real-time filtering on the standardized first detection sequence and the standardized second detection sequence using a dynamic threshold adjustment method, and perform static noise reduction on the standardized third detection sequence using a fixed threshold filtering method to generate a first processed sequence, a second processed sequence, and a third processed sequence; wherein, the first processed sequence includes a filtered first excitation waveform and a filtered second feedback waveform.

6. The lithium battery case toughness detection system according to claim 5, wherein The signal preprocessing module is further configured to: Calculate the coupling coefficient between the filtered first excitation waveform and the filtered second feedback waveform in the historical detection period; Predict the predicted deformation amount of the filtered second feedback waveform in the real-time detection period according to the coupling coefficient and the load characteristics of the filtered first excitation waveform in the real-time detection period; Generate target excitation response data according to the filtered second feedback waveform and its predicted deformation amount, and use the detection sequence corresponding to the target excitation response data as the first processed sequence.

7. The lithium battery shell toughness detection system according to claim 2, characterized in that, The waveform reconstruction layer specifically includes: A feature correlation unit, configured to perform load path analysis on each detection sequence included in the time series correlation feature data respectively, so as to extract a corresponding stress propagation chain from each detection sequence; A waveform matching unit, configured to perform dynamic mapping on the stress propagation chain extracted from each detection sequence and the corresponding time series feature data to generate reconstructed waveform feature data.

8. The lithium battery case toughness detection system according to claim 7, wherein, The waveform reconstruction layer further includes: A harmonic suppression unit, configured to perform harmonic component elimination processing on the reconstructed waveform feature data.

9. The lithium battery shell toughness detection system according to claim 2, characterized in that, The toughness determination layer specifically includes: A multi-dimensional fusion unit, including a plurality of feature fusion nodes, and each feature fusion node is connected to each detection sequence in the reconstructed waveform feature data and the time series correlation feature data through associated configuration; A dynamic association optimization unit, configured to adjust the associated configuration through a dynamic association optimization algorithm to minimize the error between the toughness detection result and the actual deformation data; A failure mode recognition unit, configured to predict a crack propagation path based on the reconstructed waveform feature data and the time series correlation feature data to generate a toughness detection result.

10. The lithium battery case toughness detection system according to claim 5, wherein, The dynamic threshold adjustment method specifically includes: Generate an adaptive filtering threshold based on the noise spectrum distribution in the real-time detection environment; Perform segmented filtering processing on the standardized first detection sequence by using a sliding window mechanism.

11. The lithium battery case toughness detection system according to claim 8, wherein The harmonic component elimination processing includes: Extract the components in the reconstructed waveform feature data whose frequencies are integer multiples of the fundamental frequency, and filter out the extracted components from the original data after phase inversion and superposition.

12. The lithium battery case toughness detection system according to claim 9, characterized in that, The crack propagation path prediction includes: Establish a crack growth rate model including a stress concentration coefficient; Calculate the crack bifurcation probability according to the stress gradient distribution in the reconstructed waveform feature data.

13. A method for detecting the toughness of a lithium battery shell, characterized in that, Includes: Obtain multi-dimensional deformation data of the target lithium battery shell under impact load, where the multi-dimensional deformation data includes a first detection sequence corresponding to vibration response data, a second detection sequence corresponding to stress distribution data, and a third detection sequence corresponding to deformation trajectory data; Perform dynamic baseline calibration processing on the multi-dimensional deformation data to generate a standard detection data stream; Input the standard detection data stream into the toughness analysis processing layer for feature extraction; Generate the toughness detection result of the target lithium battery housing according to the output result of the toughness analysis and processing layer; the toughness analysis and processing layer includes a signal preprocessing module and a feature analysis module, wherein the signal preprocessing module is used for waveform segmentation and noise suppression of the standard detection data stream, and the feature analysis module is obtained by jointly modeling based on the historical waveform data and historical deformation data of multiple historical detection cycles; the feature analysis module includes an excitation response layer, a waveform reconstruction layer, and a toughness determination layer connected in sequence. The excitation response layer is used for time-frequency domain alignment processing of multiple detection sequences included in the standard detection data stream to obtain time-series correlation feature data; the waveform reconstruction layer is used for modeling the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain reconstructed waveform feature data; the toughness determination layer is used for multi-dimensional fusion based on the reconstructed waveform feature data and the time-series correlation feature data to generate a toughness detection result; Among them, the modeling of the dynamic coupling relationship between the time-series correlation feature data corresponding to each detection sequence to obtain the reconstructed waveform feature data includes: Use the dynamic baseline calibration algorithm to identify the key deformation nodes in the time-series correlation feature data, and determine the deformation correlation sequence corresponding to each detection sequence based on the load type corresponding to each key deformation node; Calculate the waveform similarity between the deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences, and determine the reconstructed waveform feature data between the any two detection sequences based on the waveform similarity; the calculation of the waveform similarity between the deformation nodes with the same load type in the deformation correlation sequences corresponding to any two detection sequences includes: When there is a difference in the number of deformation nodes in the deformation correlation sequences corresponding to the any two detection sequences, perform virtual node compensation based on the load type corresponding to the terminal deformation node in the one with fewer deformation nodes, and calculate the waveform similarity between the deformation nodes with the same load type based on the compensated data.

14. The method for detecting the toughness of the lithium battery casing according to claim 13, wherein The dynamic association optimization algorithm specifically includes: Establish an error distribution matrix between the toughness detection result and the actual deformation data; Iteratively adjust the weight parameters in the association configuration by the gradient descent method until the error converges.

Images