Battery bumping damage early warning method and system

Abstract

The invention discloses a battery bumping damage early warning method and system, and relates to the technical field of battery damage early warning, and the method comprises the steps: carrying out a standard charging preprocessing cycle on a single battery, carrying out the complete charging after a preset number of times is reached, and obtaining the initial quality data; performing quasi-static extrusion damage to obtain final-state quality data, and calculating a quality change rate; collecting quality change rates, detection data, simulation data and structure states of multiple groups of batteries to construct a training data set, and training to obtain a deep learning model; battery mechanical collision state monitoring data are obtained in real time, matched feature parameters are obtained through preprocessing and feature extraction and input into the model, and a structural integrity evaluation result is generated; according to the battery bumping damage early warning method and device, early warning is conducted on potential damage in advance through the deep learning model, and potential safety hazards caused by internal damage of the battery are avoided.

Description

Technical Field

[0001] This application relates to the field of battery damage early warning technology, and in particular to a method and system for early warning of battery impact damage. Background Technology

[0002] As a core energy component of electric vehicles, energy storage systems, and portable electronic devices, the operational safety of power batteries directly affects the reliability of the entire system and user safety. In actual use, individual battery cells inevitably suffer from external forces such as drops, lateral impacts, and vibration shocks. When these forces exceed a certain threshold, they can lead to hidden structural degradation such as misalignment of internal electrodes, damage to the separator, peeling off of active materials, and deformation of the casing structure. This can result in safety hazards such as capacity decay, increased internal resistance, and a heightened risk of thermal runaway. Therefore, real-time and effective monitoring and early warning of the structural integrity of batteries under impact conditions are of great significance.

[0003] In existing technologies, safety assessments of batteries after impacts or drops primarily rely on manual visual inspection, casing deformation detection, or disassembly. These methods have significant limitations: disassembly is destructive and cannot be applied to battery systems in actual operation, and is accompanied by high cost and low efficiency; assessments based on external deformation cannot identify micro-damage to internal electrode structures, failing to effectively assess latent damage; existing diagnostic methods based on mechanical impact sensors typically rely on fixed thresholds or empirical rules, making it difficult to adapt to batteries of different specifications, material systems, and aging stages, resulting in insufficient accuracy and reliability of warnings. Traditional testing methods struggle to quantify the degree of structural damage at the battery material level and lack effective indicators to characterize changes in the integrity of the battery's internal structure.

[0004] Therefore, it is necessary to design a battery impact damage early warning method and system to solve the problems existing in the current technology. Summary of the Invention

[0005] The purpose of this application is to provide a method and system for early warning of battery impact damage, which aims to solve the problems of inaccurate damage identification, delayed early warning, and inability to quantify internal damage.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for early warning of battery impact damage, including: A standard charging pretreatment cycle is performed on the battery cell. When the standard charging pretreatment cycle reaches a preset number of charging cycles, the battery cell is fully charged to obtain the initial quality data of the battery cell. Quasi-static compression damage is performed to obtain the final mass data of the battery cell; the mass change rate is calculated based on the initial mass data and the final mass data; the mass change rate characterizes the degree of damage to the battery structural integrity. Collect the mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity status of several individual battery cells to construct a training dataset; train a neural network model based on the training dataset to generate the final deep learning model; Real-time acquisition of mechanical impact status monitoring data of the battery; conversion of the mechanical impact status monitoring data into matching feature parameters based on data processing and feature extraction; The feature parameters are input into the deep learning model to generate a battery structural integrity assessment result; based on the battery structural integrity assessment result, it is determined whether there is a risk of damage; when a risk of damage is determined, a battery impact damage warning signal is generated based on the battery structural integrity assessment result.

[0007] Furthermore, the battery cells undergo a standard charging pretreatment cycle, including: The battery cells are charged with constant current and constant voltage according to the charging procedure; the battery cells after constant current and constant voltage charging are left to stand for a first preset time; after the first preset time, the battery cells are discharged with constant current according to the discharging procedure; the battery cells after constant current discharge are left to stand for a second preset time. Repeat the standard charging pre-processing cycle until the preset number of charging cycles is reached.

[0008] Furthermore, quasi-static extrusion damage is performed to obtain the final-state mass data of the battery cells, including: At a room temperature of 25 degrees Celsius, the large surface of the battery is extruded at a speed of 1 mm / s; Monitor the changes in extrusion pressure, voltage, positive electrode temperature, negative electrode temperature, large surface temperature, and side surface temperature during the extrusion process. If any parameter exceeds its corresponding safe operating range, suspend the minor extrusion process to prevent controllable damage. Continue the minor extrusion process to prevent controllable damage only after all parameters have returned to the safe operating range.

[0009] Furthermore, when performing quasi-static extrusion damage to obtain the final mass data of a single battery cell, the process also includes: After the quasi-static extrusion damage is completed, the final state mass data is obtained; at the same time, the extrusion force signal, voltage monitoring signal, battery positive electrode temperature signal, battery negative electrode temperature signal, battery large surface temperature signal and battery side temperature signal monitored in real time during the extrusion process are obtained as quasi-static extrusion damage detection data; battery shell stress data signal and battery internal stress data signal are obtained as quasi-static extrusion damage simulation data.

[0010] Furthermore, the rate of change of mass is calculated based on the initial mass data and the final mass data, including: The difference between the initial mass data and the final mass data is calculated to obtain the mass loss value; Divide the mass loss value by the initial mass data to obtain the initial mass change rate; The initial mass change rate is normalized to generate the final mass change rate.

[0011] Furthermore, the training dataset is constructed, including: The mass change rate of several different types of battery cells under different quasi-static extrusion conditions was collected, and quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity assessment data corresponding to the mass change rate were collected. The mass change rate, quasi-static extrusion damage detection data, and quasi-static extrusion damage simulation data were correlated with the battery structural integrity status to generate correlation data. The correlation data was cleaned. The cleaned correlation data was then subjected to feature engineering to generate model training feature vectors. Based on the model training feature vectors and the battery structural integrity status, a training dataset was constructed.

[0012] Furthermore, a neural network model is trained based on the training dataset to generate the final deep learning model, including: The training dataset is sampled in a 4:1 ratio to obtain a training subset and a test subset; The neural network model is iteratively trained based on the training subset, the iteratively trained neural network model is evaluated based on the test subset, and it is determined whether to stop iterative training based on the evaluation value. If the evaluation value of the neural network model after the current iteration is less than the evaluation value of the neural network model after the previous iteration, the magnitude of the change in the gradient direction of the neural network model is reduced, and iterative training continues until the preset number of iterations is reached; if the evaluation value of the neural network model after the current iteration is greater than or equal to the evaluation value of the neural network model after the previous iteration, iterative training is stopped, and the deep learning model is obtained.

[0013] Furthermore, real-time monitoring data on the mechanical impact status of the battery is collected; based on data processing and feature extraction, the mechanical impact status monitoring data is converted into matching feature parameters, including: Real-time acquisition of mechanical impact monitoring data based on accelerometer; acquisition of battery casing deformation data based on strain sensor; recording of the timestamp and duration of the mechanical impact monitoring data; The collected mechanical impact monitoring data is filtered, and impact intensity features, impact direction features, and impact duration features are extracted. The impact intensity features, impact direction features, and impact duration features are then normalized to generate the matching feature parameters.

[0014] Furthermore, based on the battery structural integrity assessment results, it is determined whether there is a risk of damage. When a risk of damage is determined, a battery impact damage warning signal is generated based on the battery structural integrity assessment results, including: The battery structure integrity assessment result includes the probability of damage risk; the probability of damage risk is compared with a preset risk threshold; the preset risk threshold includes a first preset risk threshold and a second preset risk threshold, and the first preset risk threshold is less than the second preset risk threshold; When the probability of damage is less than or equal to the first preset risk threshold, no battery impact damage warning signal is generated. When the probability of damage is greater than the first preset risk threshold and less than the second preset risk threshold, a primary battery impact damage warning signal is generated. When the probability of damage is greater than or equal to the second preset risk threshold, an advanced battery impact damage warning signal is generated.

[0015] Secondly, this application provides a battery impact damage early warning system, applied to the aforementioned battery impact damage early warning method, including: The first acquisition unit is configured to perform a standard charging preprocessing cycle on the battery cell. When the standard charging preprocessing cycle reaches a preset number of charging cycles, the battery cell is fully charged to obtain the initial quality data of the battery cell. The second acquisition unit performs quasi-static compression damage to obtain the final mass data of the battery cells; it calculates the mass change rate based on the initial mass data and the final mass data; the mass change rate characterizes the degree of damage to the battery structural integrity. The model building unit is configured to collect the mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity status of several battery cells to build a training dataset; and to train a neural network model based on the training dataset to generate the final deep learning model. The third acquisition unit is configured to acquire real-time mechanical impact condition monitoring data of the battery; and convert the mechanical impact condition monitoring data into matching feature parameters based on data processing and feature extraction. The damage warning unit is configured to input the feature parameters into the deep learning model to generate a battery structural integrity assessment result; determine whether there is a risk of damage based on the battery structural integrity assessment result; and generate a battery impact damage warning signal based on the battery structural integrity assessment result when a risk of damage is determined.

[0016] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the battery impact damage warning method described in any one of the above.

[0017] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the battery impact damage warning method described in any one of the above descriptions.

[0018] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the battery impact damage warning method described above.

[0019] In a sixth aspect, this application provides a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run a program or instructions, and the processor executing the program or instructions implementing the steps of the battery impact damage warning method described above.

[0020] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a battery impact damage early warning method and system. Through standard charging preprocessing, controllable damage from slight compression, and temperature-vacuum combined processing, it extracts the mass change rate caused by material leakage and structural damage within the battery, forming an objective indicator reflecting the battery's structural integrity. This gives the deep learning model a clear physical meaning, improving the reliability of the early warning model. Controllable damage from slight compression obtains high-precision final-state mass, improving the realism and discriminative power of the training data and providing high-quality samples for the model. By collecting mass change rate, quasi-static compression damage detection data, quasi-static compression damage simulation data, and battery structural integrity assessment data, multi-dimensional training features are formed, helping the model learn damage patterns under different scenarios and achieving stronger generalization ability and robustness. A dynamic adjustment strategy based on evaluation values ​​is introduced during neural network training. When model performance deteriorates, the gradient update amplitude is automatically reduced, avoiding gradient explosion or oscillation, ensuring stable convergence of the model in the optimal direction, and further improving evaluation accuracy. By filtering, extracting and normalizing multi-dimensional features such as impact intensity, direction, and duration, the model can identify more complex impact conditions, including cumulative damage from weak impacts and sudden damage caused by instantaneous strong impacts, thus improving the sensitivity of early warnings. Using first and second risk thresholds to achieve tiered processing of no warning, basic warning, and advanced warning, the model can not only identify severe damage but also provide early warnings of minor potential hazards, improving the accuracy of battery system safety management. Deep learning models provide early warnings of potential damage, preventing major safety hazards such as thermal runaway, leakage, and short circuits caused by internal battery damage. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A schematic flowchart illustrating a battery impact damage early warning method according to an embodiment of this application; Figure 2 This is a schematic diagram of the functional modules of a battery impact damage early warning system provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] For this, please refer to Figure 1 As shown, this application proposes a battery impact damage early warning method, including: S100: Perform a standard charging pre-processing cycle on the battery cell. When the standard charging pre-processing cycle reaches the preset number of charging cycles, fully charge the battery cell and obtain the initial quality data of the battery cell. S200: Quasi-static extrusion damage is performed to obtain the final mass data of the battery cell; the mass change rate is calculated based on the initial mass data and the final mass data; the mass change rate characterizes the degree of damage to the battery structural integrity. S300: Collect the mass change rate of several battery cells, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity status to construct a training dataset; train a neural network model based on the training dataset to generate the final deep learning model; S400: Real-time acquisition of battery mechanical impact condition monitoring data; conversion of mechanical impact condition monitoring data into matching feature parameters based on data processing and feature extraction; S500: Input the feature parameters into the deep learning model to generate the battery structural integrity assessment result; determine whether there is a risk of damage based on the battery structural integrity assessment result; when a risk of damage is determined, generate a battery impact damage warning signal based on the battery structural integrity assessment result.

[0026] Specifically, a standardized charging pretreatment cycle is performed on the battery cells under test to fully activate the cell materials and stabilize the internal chemical reaction state, thereby ensuring the accuracy of subsequent quality measurements. The battery cells are charged under constant current and constant voltage according to a preset charging procedure, and then allowed to stand for a set period after charging to allow the internal ion distribution to reach relative stability. The cells are then discharged under constant current according to a discharge procedure, and then allowed to stand again to eliminate temperature fluctuations or internal pressure changes caused by the discharge process. This charging pretreatment cycle is repeated continuously until a preset number of cycles is reached, allowing the battery to enter a stable operating state. After reaching the preset number of cycles, the battery cells are fully charged to avoid residual charge affecting subsequent minor compression-induced damage and quality assessment. Initial mass data of the battery cells is collected using high-precision mass measurement equipment as a benchmark value for subsequent structural integrity analysis. After obtaining the initial mass data, minor compression-induced damage treatment is performed to obtain the final mass data of the battery cells. The mass change rate is calculated based on the initial mass and the final mass, where the mass change rate reflects the degree of material loss after the rupture—temperature control—vacuum treatment, and can objectively characterize the damage status of the battery's structural integrity. Normalizing the mass change rate improves the consistency of subsequent model training. After obtaining multiple sets of mass change rate data for individual cells, corresponding quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity status (e.g., whether material detachment has occurred, CT scan results of structural damage) are further collected and correlated. The correlated data is then cleaned to remove noisy or outlier samples, and model training feature vectors are generated using feature engineering methods (e.g., normalization, feature encoding, dimensionality compression). A training dataset is constructed based on these training feature vectors and the corresponding structural integrity statuses, and divided into training and testing subsets according to a set ratio. The neural network model is iteratively trained using the training subset, while the model performance is evaluated using the testing subset. When the model performance does not meet the preset target, iterative training continues, and the gradient update magnitude is dynamically adjusted based on the evaluation results, ultimately resulting in a deep learning model capable of accurately predicting battery structural damage risk. In practical battery applications, various monitoring data are collected in real time when the battery is subjected to mechanical impact, including acceleration changes, casing deformation, impact direction, and impact duration, through devices such as accelerometers and strain sensors installed in the battery pack or battery module. The timestamp and duration of the impact event are also recorded. After noise filtering and outlier processing, the collected raw impact data is used to extract key impact feature parameters, including impact intensity, impact direction, and impact duration. These feature parameters are then normalized or standardized to form matching feature parameters that can be directly input into deep learning models.The real-time generated matching feature parameters are input into the previously trained deep learning model to obtain the battery structural integrity assessment result. The assessment result may include the probability of damage risk or the specific damage level. Based on this assessment result, it is compared with a set risk threshold. When the risk probability is lower than the threshold, it is determined to be in an undamaged state. When the risk probability reaches the set range, a corresponding level of impact damage warning signal is generated, including primary warning and advanced warning, so that different levels of risk can trigger different safety response mechanisms.

[0027] The working process and principle of this application are as follows: In the calibration stage, the battery cells under test are subjected to standardized constant current and constant voltage charging pretreatment cycles to stabilize the chemical and electrothermal states of the cells. After reaching a preset number of cycles, the batteries are fully charged and subjected to slight, controllable compression to induce damage, bringing the batteries to their final state and measuring their final mass. The normalized mass change rate is calculated as a physical quantity reflecting the degree of internal material dissipation and structural damage. This mass change rate physically characterizes the mass loss caused by electrode active material shedding, electrolyte evaporation, or separator damage, and is therefore directly related to structural integrity. In the data construction stage, a large number of different types of batteries under various operating conditions and different quasi-static compression-induced conditions are collected, along with their corresponding temperature / vacuum parameters and structural integrity status marked by imaging or electrical testing. After data cleaning, outlier removal, normalization, and feature engineering (including time domain, frequency domain, and statistical features), training feature vectors are generated and used for supervised training of a neural network model. During training, training / testing partitioning, evaluation-driven iterative updates, and adaptive adjustment of the learning rate or gradient magnitude are adopted to ensure that the model has both high accuracy and good generalization ability. Online monitoring is also performed. The system uses sensors such as accelerometers and strain gauges deployed on individual battery cells or battery packs to collect triaxial acceleration, spectrum, waveform duration, and casing strain data in real time during impact events. This data is then filtered, denoised, time-synchronized, and feature extracted (e.g., peak acceleration, impact energy spectrum, impact direction vector, impact duration, and cumulative micro-impact indices). These normalized matching feature parameters are organized into an input vector with the same format as during training. This vector is then input into a pre-trained deep learning model to obtain a battery structural integrity assessment output (e.g., failure probability or graded damage level). The output is compared with a set multi-level threshold and corresponding responses are triggered: below the first threshold, the system is considered safe; above the first threshold but below the second threshold, a primary warning is issued suggesting inspection / limiting charging conditions; above the second threshold, a high-level warning is issued and can be linked to power outage, isolation, or repair procedures. The entire principle is based on the mapping relationship between mass loss as a physical label and impact characteristics, learned by a deep network. This allows for the inference of physical damage measurements obtained in controlled breach-temperature vacuum tests based solely on online measurable mechanical signals, thus achieving early, continuous, and graded warnings for hidden internal damage.

[0028] As a preferred embodiment, the specific implementation of this application is as follows: In practical applications, to construct training samples required for deep learning models, this embodiment selects three 10Ah soft-pack lithium-ion battery cells from consistent production batches as test objects. Standard pretreatment cycles are performed on the three battery samples to ensure battery state consistency. Throughout the pretreatment process, each battery is charged at a constant current of 1 / 3C provided by the manufacturer. Once the voltage reaches the termination voltage of 4.2V, it immediately enters a constant voltage stage and continues charging until the current drops to 0.05C. After charging, the batteries are left to stand for 40 minutes to allow the internal chemical system to fully stabilize. Subsequently, the batteries are discharged to 2.75V at the same 1 / 3C current, and the discharge capacity is recorded. For example, sample A's three discharge capacities are 9.82Ah, 9.88Ah, and 9.85Ah, respectively, all falling within ±10% of the rated capacity of 10Ah, meeting the conditions for continued testing. After pretreatment, the three batteries immediately enter the structural damage feature acquisition stage. The initial mass of three batteries was measured: sample A was 102.35g, sample B was 101.92g, and sample C was 102.10g, and this mass was recorded as M0. Then, a slight, controlled compression was applied to bring the batteries to their final state, and their final mass was measured. The three battery samples were then removed and their masses were measured again. This time, the mass of sample A decreased to 101.27g, sample B to 100.98g, and sample C to 101.05g, recorded as M1. Based on the difference between M0 and M1, the corresponding mass change rate for the three batteries was calculated. For example, the weight loss rate for sample A was 1.08%, while for samples B and C it was 0.93% and 1.03%, respectively. These weight loss rates directly reflect the integrity damage to the internal structure of the batteries and are used as key training label data for the final deep learning model. Finally, the breakage parameters, temperature treatment parameters, vacuum treatment records, and corresponding mass change rates of the three samples were combined to form a multi-dimensional training sample for training a deep learning model. This model was then able to learn the implicit mapping relationship between battery structural damage and mass change under different mechanical impact conditions. After the model training is complete, it can be used in vehicles, electric equipment, or energy storage systems by inputting real-time mechanical impact monitoring signals to quickly predict whether there is a potential risk of battery breakage, thereby achieving early warning of battery impact damage.

[0029] Through the above scheme, this application uses the rate of mass change of the battery under slight and controllable damage by compression as the structural damage characterization index, and constructs a unified feature space by combining mechanical impact monitoring data. This enables the model to associate the internal structural degradation mechanism of the battery with the external impact response characteristics from multiple dimensions. Through the deep learning model, it achieves high-precision identification of the degree of damage to the battery structural integrity, and can distinguish between slight impact, potential hidden damage and serious damage risk, thus achieving more reliable early warning.

[0030] This application further proposes a standard charging pretreatment cycle for individual battery cells, including: The battery cells are charged with constant current and constant voltage according to the charging procedure; the battery cells after constant current and constant voltage charging are left to stand for a first preset time; after the first preset time, the battery cells are discharged with constant current according to the discharging procedure; the battery cells after constant current discharge are left to stand for a second preset time. Repeat the standard charging pre-processing cycle until the preset number of charging cycles is reached.

[0031] Specifically, during the standard charging pretreatment cycle, the battery cells are charged using a constant current and constant voltage method according to the charging protocol. For example, for a typical lithium-ion battery, a constant current charging rate of 0.5C is used. When the battery voltage reaches 4.2V (for ternary material systems) or 3.65V (for lithium iron phosphate systems), the charging mode is switched to constant voltage until the charging current drops below 0.05C. After constant current and constant voltage charging is completed, the battery cells are placed in a constant temperature environment of 25±2℃ for a first preset resting time, preferably 30-60 minutes, to allow the internal ion distribution of the battery to reach a balanced state and eliminate local concentration gradients and thermal effects generated during charging. After the resting period, the battery cells are discharged using a constant current method based on the discharge protocol. The discharge rate is the same as the charging rate (e.g., 0.5C), and the discharge cutoff voltage is the lower limit voltage specified by the manufacturer (e.g., 2.5V-3.0V). After constant current discharge, the battery cells are placed in the same constant temperature environment for a second preset rest period of 30-60 minutes to eliminate temperature fluctuations and internal stress generated during discharge. This pre-treatment charging cycle is repeated (preferably 3-5 times) to stabilize the electrochemical state of the battery cells and eliminate interference from previous charging cycles on subsequent quality measurements. After reaching the preset number of charging cycles, a final full charge is performed to ensure the battery is at 0% SOC, providing a safety guarantee against controllable damage from minor compression.

[0032] Through the above technical solutions, this application eliminates the "memory effect" of individual battery cells and the influence of historical charging states on quality measurement results, enabling the internal chemical system of the battery to reach a stable state. The resting step in the pretreatment cycle ensures uniform ion distribution inside the battery, avoiding quality measurement deviations caused by local concentration differences. The full charging treatment reduces the risk of short circuits and thermal runaway during the breaking operation, improving experimental safety.

[0033] This application further proposes to obtain the final mass data of a single battery cell by performing quasi-static extrusion damage, including: At a room temperature of 25 degrees Celsius, the large surface of the battery is extruded at a speed of 1 mm / s; Monitor the changes in extrusion pressure, voltage, positive electrode temperature, negative electrode temperature, large surface temperature, and side surface temperature during the extrusion process. If any parameter exceeds its corresponding safe operating range, suspend the minor extrusion process to prevent controllable damage. Continue the minor extrusion process to prevent controllable damage only after all parameters have returned to the safe operating range.

[0034] Specifically, in a constant temperature laboratory environment of 25±2℃, the battery cells are placed horizontally on the material testing machine platform, ensuring that the large surface area of ​​the battery (i.e., the largest plane) is in contact with the extrusion device. A semi-circular indenter with a diameter of 10mm is used to perform quasi-static extrusion on the large surface of the battery at a constant speed of 1.0±0.1mm / s until the indenter displacement reaches the set value of 2.0±0.05mm. During the extrusion process, the following key parameters are monitored in real time: extrusion force (range 0-5kN, accuracy ±1%), battery voltage (range 0-5V, accuracy ±0.001V), battery positive electrode temperature (range -20℃ to 80℃, accuracy ±0.5℃), battery negative electrode temperature (range -20℃ to 80℃, accuracy ±0.5℃), battery large surface center temperature (range -20℃ to 80℃, accuracy ±0.5℃), and battery side center temperature (range -20℃ to 80℃, accuracy ±0.5℃). The safe operating ranges for each parameter are set as follows: the extrusion force does not exceed the critical value corresponding to the yield strength of the battery casing (usually 800-1500N, depending on the battery size and structure); the voltage is not lower than 2.5V and not higher than 4.3V; and the temperature at each point does not exceed 60℃, with a temperature difference between adjacent points not exceeding 15℃. When the monitoring system detects that any parameter exceeds the safe operating range, the extrusion process is immediately paused, maintaining the current displacement state; simultaneously, a safety monitoring protocol is activated to continuously monitor changes in all parameters. If the parameter recovers to the safe range within 5 minutes, the remaining extrusion displacement continues at the original speed; if the parameter fails to recover within 5 minutes, the extrusion operation is terminated, the abnormal situation is recorded, and the sample is marked as invalid data.

[0035] Through the above technical solution, this application constructs a dynamic safety boundary for the battery extrusion process by real-time monitoring of multi-dimensional physical parameters, avoiding safety risks such as thermal runaway and internal short circuits caused by excessive extrusion, and ensuring the safety of the experimental process.

[0036] This application further proposes that when obtaining the final-state mass data of a battery cell through quasi-static extrusion damage, it also includes: After the quasi-static extrusion damage is completed, the final state mass data is obtained; at the same time, the extrusion force signal, voltage monitoring signal, battery positive electrode temperature signal, battery negative electrode temperature signal, battery large surface temperature signal and battery side temperature signal monitored in real time during the extrusion process are obtained as quasi-static extrusion damage detection data; battery shell stress data signal and battery internal stress data signal are obtained as quasi-static extrusion damage simulation data.

[0037] Specifically, after the quasi-static extrusion operation, the battery cells were placed in a constant temperature environment of 25±2℃ for 30 minutes to eliminate residual stress and temperature gradients generated during the extrusion process, ensuring the accuracy of mass measurement. After the settling period, the battery cells were weighed using a high-precision electronic balance (accuracy 0.001g) and recorded as the final mass data M1. Simultaneously, monitoring data from the entire extrusion process were extracted from the data acquisition system, including: extrusion force-displacement curve signals (sampling rate 100Hz), voltage-time change signals (sampling rate 50Hz), and temperature monitoring signals at four key locations (positive terminal, negative terminal, center of the large surface, and center of the side surface, sampling rate 10Hz). These experimental detection data directly correspond to the physical parameters synchronously acquired during the extrusion process. Furthermore, numerical simulations under the same extrusion conditions were performed using finite element simulation software (such as ANSYS or Abaqus) to obtain data signals on the stress distribution of the battery casing and the stress signals of the internal electrode components. The experimental detection data and simulation data complement each other; the former provides the true physical response, while the latter reveals the internal mechanical mechanism. Together, they constitute a complete dataset of quasi-static extrusion-induced damage.

[0038] Through the above technical solution, this application combines high-precision mass measurement with synchronous acquisition of multi-dimensional parameters, which establishes an accurate correlation between the mass change rate and the mechanical-thermal-electric response characteristics, and solves the limitation of traditional methods that rely on only a single parameter to assess battery damage.

[0039] This application further proposes to calculate the rate of change of mass based on initial mass data and final mass data, including: The difference between the initial mass data and the final mass data is calculated to obtain the mass loss value; Divide the mass loss value by the initial mass data to obtain the initial mass change rate; The initial mass change rate is normalized to generate the final mass change rate.

[0040] The formula for calculating the rate of change in mass is:

[0041] In the formula: Let be the weightlessness rate of the i-th experimental subject, and be a percentage value. The initial mass of the i-th experimental subject is expressed in grams (g). Let be the final mass of the i-th experimental object, in grams (g).

[0042] Specifically, initial and final mass data after controlled damage from slight compression are obtained, and the mass loss value is calculated, reflecting the total amount of material lost by the battery during testing. The mass loss value is divided by the initial mass data to obtain the initial mass change rate. This initial mass change rate is then normalized to generate the final mass change rate. The normalized mass change rate η ranges from [0,1], where 0 represents no mass loss (intact structure) and 1 represents maximum mass loss (severe structural damage). A mass change rate threshold can be set based on historical data to distinguish different levels of structural integrity damage.

[0043] Through the above technical solution, this application establishes a quantitative relationship between the rate of mass change and the degree of damage to the battery structural integrity. The rate of mass change, as a core characteristic parameter, can objectively and accurately characterize the loss of internal battery materials, including electrode active material shedding, electrolyte evaporation, and separator damage. Normalization eliminates the interference of different battery types, capacities, and sizes on mass measurement, making the rate of mass change comparable across battery types.

[0044] This application further proposes constructing a training dataset, including: The mass change rate of several different types of battery cells under different quasi-static extrusion conditions was collected, and quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity assessment data corresponding to the mass change rate were collected. The mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data and battery structural integrity status were correlated to generate correlation data. The correlation data was cleaned. The cleaned correlation data was then subjected to feature engineering to generate model training feature vectors. Based on the model training feature vectors and battery structural integrity status, a training dataset was constructed.

[0045] Specifically, the training dataset was constructed by collecting diverse battery samples, including different types (e.g., pouch, cylindrical, prismatic), different chemical systems (e.g., ternary lithium, lithium iron phosphate, lithium cobalt oxide), and different capacities (e.g., 2Ah-100Ah). Mechanical impacts of varying intensity, direction, and duration were applied to each battery sample to simulate real-world impact scenarios. Impact conditions included low-energy, medium-energy, and high-energy impacts, as well as different impact directions (front, side, bottom) and different impact waveforms (half-sine wave, triangular wave, square wave). After each impact, the aforementioned controlled slight compression was applied to the battery sample to obtain mass change rate data. Simultaneously, the corresponding compression force signal, voltage monitoring signal, battery positive electrode temperature signal, battery negative electrode temperature signal, battery surface temperature signal, battery side temperature signal, battery casing stress data signal, and battery internal stress data signal were recorded. The battery structural integrity status (e.g., damaged area, separator deformation degree, internal short circuit probability) was also measured via CT scan, X-ray imaging, or electrochemical impedance spectroscopy. The rate of mass change is correlated with these parameters to form structured data records. The correlated data is then cleaned to remove outliers (such as abnormal samples with a rate of mass change exceeding 20%) and noisy data (such as erroneous readings caused by temperature sensor malfunctions). Feature engineering is then performed, including standardization, time-frequency domain transformation of impact features, and hierarchical encoding of structural integrity states (e.g., intact = 0, minor damage = 1, moderate damage = 2, severe damage = 3). Finally, the processed feature vectors are paired with structural integrity state labels to construct a complete training dataset. The dataset size is preferably 500-2000 samples to ensure sufficient training capacity for the model.

[0046] Through the above technical solutions, this application constructs a high-quality, diverse training dataset, covering various battery types and damage scenarios that may be encountered in practical applications. Data cleaning and feature engineering improve data quality and eliminate the interference of noise and outliers on model training. The careful design of feature vectors enables the deep learning model to learn the complex mapping relationship between mechanical impact features and structural integrity damage, improving the model's prediction accuracy and robustness.

[0047] This application further proposes training a neural network model based on a training dataset to generate a final deep learning model, including: The training dataset is sampled in a 4:1 ratio to obtain training and test subsets; The neural network model is iteratively trained based on the training subset, the trained neural network model is evaluated based on the test subset, and the evaluation value is used to determine whether to stop the iterative training. If the evaluation value of the neural network model after the current iteration is less than the evaluation value of the neural network model after the previous iteration, the magnitude of the change in the gradient direction of the neural network model is reduced, and iterative training continues until the preset number of iterations is reached; if the evaluation value of the neural network model after the current iteration is greater than or equal to the evaluation value of the neural network model after the previous iteration, iterative training is stopped, and a deep learning model is obtained.

[0048] Specifically, the constructed training dataset is randomly divided into a training subset and a test subset in a 4:1 ratio to ensure a balanced distribution of battery type, impact conditions, and damage severity between the two datasets. A suitable neural network architecture for structured data is selected, such as a Multilayer Perceptron (MLP) or a Lightweight Convolutional Neural Network (CNN), with the number of input layer nodes matching the feature vector dimension, and the output layer being either a single node (predicting the probability of breakage) or three nodes (predicting the damage level). The model is trained using the cross-entropy loss function and the Adam optimizer, with an initial learning rate of 0.001. In each training iteration, the model parameters are updated using the training subset, and evaluation metrics (such as accuracy, F1 score, or AUC value) are calculated using the test subset. Training continues when the evaluation metric of the current iteration is better than the previous iteration; when the evaluation metric decreases twice consecutively or the change is less than a preset threshold (e.g., 0.5%), the learning rate is reduced (e.g., multiplied by a decay factor of 0.5), and training continues. If the current iteration's evaluation metric is better than the previous one, the current learning rate is maintained; if the current evaluation metric is worse than the previous one, the learning rate is reduced to 50%-70% of its original value to avoid oscillations around the optimal solution. This process continues until the preset maximum number of iterations (e.g., 500) is reached or the evaluation metric shows no significant improvement for five consecutive iterations (improvement <0.1%). Early stopping is used during training to prevent overfitting, and the best-performing model on the validation set is saved as the final deep learning model. After model training, independent validation datasets are used to evaluate model performance to ensure its generalization ability.

[0049] Through the above technical solutions, this application achieves efficient and stable training of deep learning models. The strategy of dynamically adjusting the learning rate enables the model to converge quickly in the early stages of training and finely adjust it when it is close to the optimal solution, avoiding the slow convergence speed or oscillation problems caused by traditional fixed learning rates. The training-test separation evaluation mechanism prevents overfitting and ensures that the model has good generalization ability.

[0050] This application further proposes real-time acquisition of mechanical impact condition monitoring data of batteries; based on data processing and feature extraction, the mechanical impact condition monitoring data is converted into matching feature parameters, including: Real-time acquisition of mechanical impact monitoring data based on accelerometer; acquisition of battery casing deformation data based on strain sensor; recording of timestamps and durations of mechanical impact monitoring data generation; The collected mechanical collision monitoring data are filtered and impact intensity, impact direction and impact duration features are extracted. The impact intensity, impact direction and impact duration features are then normalized to generate matching feature parameters.

[0051] Specifically, miniature triaxial accelerometers (range ±50g, sampling rate ≥1kHz) and strain sensors (range ±5000με, resolution ≤1με) are installed at key locations in the battery pack or battery module. The accelerometers acquire triaxial acceleration data of the battery in real time when subjected to mechanical impact, while the strain sensors simultaneously record the deformation response of the battery casing. The timestamp (accuracy ±1ms) and duration (time from the start of the impact to the acceleration returning to the background noise level) of each impact event are accurately recorded. The raw data is preprocessed: first, a Butterworth low-pass filter (cutoff frequency 200Hz) is applied to eliminate high-frequency noise; then, baseline correction is performed to eliminate errors caused by sensor drift; finally, outlier detection and processing are performed to remove abnormal readings that are significantly outside the physical range. Key features are extracted from the processed data: impact intensity features include peak acceleration, impact energy (obtained by integrating the square of the acceleration), and impact impulse; impact direction features include the main impact direction vector, the proportion of acceleration components along each axis, and the direction angle; impact duration features include the effective impact duration, rise time (time from 10% peak value to 90% peak value), and decay time. The extracted features are normalized to fall within the [0,1] interval. The normalized feature parameters constitute a matching feature vector, with the same dimensions as during training, which can be directly input into the deep learning model for prediction.

[0052] Through the above technical solution, this application achieves efficient acquisition and feature extraction of mechanical impact data, transforming complex time-domain signals into concise and effective feature parameters. Filtering and normalization processes improve data quality and eliminate the influence of sensor differences and environmental noise. The extracted impact intensity, direction, and duration features reflect the key characteristics of mechanical impact and are strongly correlated with battery structural damage.

[0053] This application further proposes a method to determine the existence of damage risk based on the battery structural integrity assessment results. When a damage risk is determined to exist, a battery impact damage warning signal is generated based on the battery structural integrity assessment results, including: The evaluation result of the battery structure integrity includes the probability of breakage risk; comparing the probability of breakage risk with a preset risk threshold; the preset risk threshold includes a first preset risk threshold and a second preset risk threshold, and the first preset risk threshold is less than the second preset risk threshold; When the probability of breakage risk is less than or equal to the first preset risk threshold, no battery bump damage warning signal is generated; When the probability of breakage risk is greater than the first preset risk threshold and less than the second preset risk threshold, a primary battery bump damage warning signal is generated; When the probability of breakage risk is greater than or equal to the second preset risk threshold, a high - level battery bump damage warning signal is generated.

[0054] Specifically, the deep - learning model outputs the evaluation result of the battery structure integrity, expressed as the probability of breakage risk P (0 ≤ P ≤ 1), where P = 0 means no breakage risk and P = 1 means extremely high breakage risk. Two risk thresholds are preset: the first preset risk threshold T1 (preferably 0.2 - 0.3) and the second preset risk threshold T2 (preferably 0.6 - 0.7), and T1 < T2. When the probability of breakage risk P output by the model is received, the following judgment logic is executed: If P ≤ T1, it is determined that the battery structure is intact, no warning signal is generated, and normal monitoring continues; If T1 < P < T2, it is determined that the battery has potential structural damage, and a primary battery bump damage warning signal is generated. This signal triggers recording an event log, restricting the maximum charging power of the battery (such as reducing it to 80%) and suggesting that the user conduct a professional inspection; If P ≥ T2, it is determined that the battery has a serious structural breakage risk, and a high - level battery bump damage warning signal is generated. This signal immediately triggers a safety protection mechanism, including cutting off the main battery circuit, starting enhanced heat management for cooling, sending an emergency alert to the user and suggesting stopping use. The warning signal is transmitted to the vehicle control unit or energy storage management unit through in - vehicle communication (such as CAN bus), and corresponding warning information is displayed on the user interface. The detailed information of the warning event is also recorded, including the timestamp, impact characteristics, risk probability, and measures taken, for subsequent analysis and model optimization. To avoid false alarms, a warning confirmation mechanism is set: the primary warning is automatically解除 when the risk probability is lower than T1 in 5 consecutive monitors; the high - level warning needs to be confirmed by manual intervention to be解除.

[0055] Through the above technical solution, the present application realizes the accurate evaluation and hierarchical warning of the battery bump damage risk. The dual - threshold design can distinguish structural damages of different severity levels, avoiding false alarms or missed alarms caused by a "one - size - fits - all" warning strategy. The primary warning provides early damage hints, giving users the opportunity to take preventive measures before the problem worsens; the high - level warning ensures that serious risks are handled in a timely manner, preventing safety accidents.

[0056] Based on the same inventive concept, this application also provides a battery impact damage warning system for implementing the aforementioned system. The solution provided by this system is similar to the solution described in the above method; therefore, the specific limitations of one or more battery impact damage warning system embodiments provided below can be found in the limitations of the battery impact damage warning method described above, and will not be repeated here.

[0057] In one exemplary embodiment, such as Figure 2 As shown, a battery impact damage warning system is provided, including: The first acquisition unit 201 is configured to perform a standard charging preprocessing cycle on the battery cell. When the standard charging preprocessing cycle reaches a preset number of charging cycles, the battery cell is fully charged to obtain the initial quality data of the battery cell. The second acquisition unit 202 performs quasi-static compression damage to obtain the final mass data of the battery cell; it calculates the mass change rate based on the initial mass data and the final mass data; the mass change rate characterizes the degree of damage to the battery structural integrity. Model building unit 203 is configured to collect the mass change rate of several battery cells, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity status to build a training dataset; and train a neural network model based on the training dataset to generate the final deep learning model. The third acquisition unit 204 is configured to acquire real-time mechanical impact condition monitoring data of the battery; and convert the mechanical impact condition monitoring data into matching feature parameters based on data processing and feature extraction. The damage warning unit 205 is configured to input feature parameters into a deep learning model to generate a battery structural integrity assessment result; determine whether there is a risk of damage based on the battery structural integrity assessment result; and generate a battery impact damage warning signal based on the battery structural integrity assessment result when a risk of damage is determined.

[0058] In summary, by employing standard charging preprocessing, controlled damage from slight extrusion, and a temperature-vacuum combination process, the rate of mass change caused by material leakage and structural damage within the battery is extracted, forming an objective indicator reflecting the battery's structural integrity. This imbues the deep learning model with clear physical meaning and enhances the reliability of the early warning model. Controlled damage from slight extrusion yields high-precision final-state mass, improving the realism and discriminative power of the training data and providing high-quality samples for the model. The collection of mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structural integrity assessment data forms multi-dimensional training features, which helps the model learn damage patterns under different scenarios, achieving stronger generalization ability and robustness. The introduction of a dynamic adjustment strategy based on evaluation values ​​during neural network training automatically reduces the gradient update amplitude when model performance declines, avoiding gradient explosion or oscillation and ensuring stable convergence of the model in the optimal direction, further improving evaluation accuracy. Through filtering, extraction and normalization of multi-dimensional features such as impact intensity, direction, and duration, the model can identify more complex impact conditions, including cumulative damage from weak impacts and sudden damage caused by instantaneous strong impacts, improving the sensitivity of the early warning system. By using first and second risk thresholds to achieve tiered processing of no warning, primary warning, and advanced warning, not only can serious damage be identified, but minor potential hazards can also be alerted in advance, improving the accuracy of battery system safety management. Deep learning models provide early warnings of potential damage, avoiding major safety hazards such as thermal runaway, leakage, and short circuits caused by internal battery damage.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

[0060] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0061] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0062] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0063] In one exemplary embodiment, a chip is provided, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the steps in the above method embodiments and achieve the same technical effect, and will not be described again here to avoid repetition.

[0064] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

Claims

1. A battery impact damage early warning method, characterized in that, The method comprises the following steps: performing a standard charging pretreatment cycle on the battery monomer, and obtaining initial mass data of the battery monomer when the standard charging pretreatment cycle reaches a preset charging number; performing a quasi-static extrusion damage to obtain final mass data of the battery monomer; and calculating a mass change rate according to the initial mass data and the final mass data; the mass change rate represents a damage degree of the battery structure integrity; collecting the mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data and battery structure integrity state of a plurality of battery monomers to construct a training data set; training a neural network model according to the training data set to generate a final deep learning model; real-time collecting mechanical knock state monitoring data of the battery; and converting the mechanical knock state monitoring data into matching feature parameters based on data processing and feature extraction; inputting the feature parameters into the deep learning model to generate a battery structure integrity evaluation result; and judging whether there is a damage risk based on the battery structure integrity evaluation result, and generating a battery knock damage warning signal according to the battery structure integrity evaluation result when it is judged that there is a damage risk.

2. The battery impact damage early warning method of claim 1, wherein, The method of performing a standard charging pretreatment cycle on the battery monomer comprises the following steps: performing constant-current constant-voltage charging on the battery monomer based on a charging procedure; and placing the battery monomer after the constant-current constant-voltage charging for a first preset time; performing constant-current discharging on the battery monomer after the first preset time based on a discharging procedure; and placing the battery monomer after the constant-current discharging for a second preset time. The method of performing a standard charging pretreatment cycle on the battery monomer comprises the following steps:

3. The battery impact damage early warning method of claim 2, wherein repeating the standard charging pretreatment cycle until the preset charging number is reached. The method of performing a quasi-static extrusion damage to obtain final mass data of the battery monomer comprises the following steps: extruding the battery surface at a speed of 1 mm / s under the condition of 25 degrees room temperature; 4. The battery impact damage early warning method of claim 3, wherein, monitoring changes of extrusion force parameters, voltage parameters, battery positive electrode temperature parameters, battery negative electrode temperature parameters, battery surface temperature parameters and battery side temperature parameters during the extrusion process; when any parameter exceeds its corresponding safe operation range, the slight extrusion controllable damage is paused; and the slight extrusion controllable damage is continued until all parameters return to the safe operation range. The method of performing a quasi-static extrusion damage to obtain final mass data of the battery monomer further comprises the following steps:

5. The battery impact damage early warning method of claim 4, wherein, obtaining the final mass data after the quasi-static extrusion damage is completed; and simultaneously obtaining extrusion force signals, voltage monitoring signals, battery positive electrode temperature signals, battery negative electrode temperature signals, battery surface temperature signals and battery side temperature signals monitored in real time during the extrusion process as quasi-static extrusion damage detection data; and obtaining battery shell stress data signals and battery internal stress data signals as quasi-static extrusion damage simulation data. The method of calculating a mass change rate according to the initial mass data and the final mass data comprises the following steps: calculating a difference between the initial mass data and the final mass data to obtain a mass loss value; dividing the mass loss value by the initial mass data to obtain an initial mass change rate; 6. The battery impact damage early warning method of claim 5, wherein, performing normalization processing on the initial mass change rate to generate a final mass change rate. The method of constructing a training data set comprises the following steps: The mass change rates of several different types of battery monomers under different quasi-static extrusion conditions are collected, and the quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data and battery structure integrity evaluation corresponding to the mass change rates are collected; the mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data and battery structure integrity state are associated to generate associated data; the associated data is data cleaned; the associated data after data cleaning is subjected to feature engineering processing to generate model training feature vectors; and a training data set is constructed based on the model training feature vectors and the battery structure integrity state.

7. The battery impact damage early warning method of claim 6, wherein, According to the training data set, a neural network model is trained to generate a final deep learning model, including: The training data set is sampled according to 4:1 to obtain a training subset and a test subset; According to the training subset, the neural network model is iteratively trained, the iteratively trained neural network model is evaluated according to the test subset, and whether to stop iteration training is determined according to the evaluation value; If the evaluation value of the current iteratively trained neural network model is less than the evaluation value of the previous iteratively trained neural network model, the amplitude of the change of the neural network model in the gradient direction is reduced, and the iteration training is continued until a preset iteration number is reached; if the evaluation value of the current iteratively trained neural network model is greater than or equal to the evaluation value of the previous iteratively trained neural network model, the iteration training is stopped, and the deep learning model is obtained.

8. The battery impact damage early warning method of claim 7, wherein, Real-time collection of mechanical knock state monitoring data of the battery; Based on data processing and feature extraction, the mechanical knock state monitoring data is converted into matching feature parameters, including: Real-time collection of mechanical knock state monitoring data based on an acceleration sensor; deformation data of the battery shell is collected through a strain sensor; the timestamp and duration of the mechanical knock state monitoring data are recorded; The collected mechanical knock state monitoring data is filtered and the impact strength feature, impact direction feature and impact duration feature are extracted; the impact strength feature, impact direction feature and impact duration feature are normalized to generate the matching feature parameters.

9. The battery impact damage early warning method of claim 8, wherein, Based on the battery structure integrity evaluation result, it is judged whether there is a damage risk, and when it is determined that there is a damage risk, a battery knock damage early warning signal is generated according to the battery structure integrity evaluation result, including: The battery structure integrity evaluation result includes a damage risk probability; the damage risk probability is compared with a preset risk threshold; the preset risk threshold includes a first preset risk threshold and a second preset risk threshold, and the first preset risk threshold is less than the second preset risk threshold; When the damage risk probability is less than or equal to the first preset risk threshold, no battery knock damage early warning signal is generated; When the damage risk probability is greater than the first preset risk threshold and less than the second preset risk threshold, a primary battery knock damage early warning signal is generated; When the damage risk probability is greater than or equal to the second preset risk threshold, a high-level battery knock damage early warning signal is generated.

10. A battery impact damage early warning system applied to the battery impact damage early warning method according to any one of claims 1-9, characterized in that, including: The first acquisition unit is configured to perform a standard charging pretreatment cycle on the battery monomer, perform full charging on the battery monomer when the standard charging pretreatment cycle reaches a preset charging number, and obtain initial mass data of the battery monomer; The second acquisition unit performs quasi-static extrusion damage to obtain final mass data of the battery monomer; and calculates a mass change rate according to the initial mass data and the final mass data; the mass change rate represents a damage degree of the battery structure integrity; The model construction unit is configured to acquire the mass change rate, quasi-static extrusion damage detection data, quasi-static extrusion damage simulation data, and battery structure integrity state of a plurality of battery monomers, and construct a training data set; The neural network model is trained according to the training data set, and a final deep learning model is generated; The third acquisition unit is configured to acquire mechanical knock state monitoring data of the battery in real time; and convert the mechanical knock state monitoring data into matching feature parameters based on data processing and feature extraction; The damage early warning unit is configured to input the feature parameters into the deep learning model to generate a battery structure integrity evaluation result; judge whether there is a damage risk based on the battery structure integrity evaluation result; and generate a battery knock damage early warning signal according to the battery structure integrity evaluation result when it is determined that there is a damage risk.

Images