Method and device for tracing lithium-ion battery thermal runaway causes based on particulate matter morphology

Abstract

The application discloses a lithium ion battery thermal runaway cause tracing method and device based on particle morphology, and relates to the technical field of batteries. The method comprises the following steps: receiving an electron microscopic image of the micron-level morphology of thermal runaway eruption particles; performing pretreatment and data enhancement on the electron microscopic image, and processing standardized image data by using a thermal runaway cause classification model to obtain a cause classification result; the thermal runaway cause classification model processes the standardized image data, which comprises the following steps: extracting multi-scale micron-level morphology features from the standardized image data based on a preset network; applying an attention mechanism weighting, and gradually reducing the dimension and classifying the weighted features by using a multi-layer classifier to obtain the cause classification result. The application combines scanning electron microscope imaging and deep learning, solves the problems of low precision and strong subjectivity caused by the dependence of traditional methods on macro features and expert experience, and realizes automatic and high-precision identification of causes such as overcharging, heating and needle pricking.

Description

Technical Field

[0001] This application relates to the field of battery technology, and in particular to a method and apparatus for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology. Background Technology

[0002] Lithium-ion batteries, as the core power source for new energy vehicles and large-scale energy storage systems, are seeing their application scale continuously expand and their energy density constantly improve. However, batteries are prone to thermal runaway under extreme abuse conditions such as overcharging, overheating, and nail penetration, leading to serious safety accidents such as smoke, fire, and even explosion. Therefore, accurately tracing the causes of battery thermal runaway accidents is of paramount importance for accident investigation, liability determination, product safety improvement, and the refinement of industry standards.

[0003] Currently, the identification of the causes of thermal runaway in lithium-ion batteries mainly relies on macroscopic examination and electrochemical analysis of battery debris after an accident. These methods typically make inferences based on macroscopic or indirect characteristics such as ablation marks, voltage curves, and changes in internal resistance, and largely depend on expert judgment. However, the thermal runaway process is violent and complex, and the battery itself is often severely damaged, leading to the loss or distortion of crucial macroscopic evidence. This results in highly subjective and uncertain analytical results, making it difficult to form objective and consistent identification conclusions.

[0004] When a battery experiences thermal runaway, internal materials are violently ejected, forming a large number of micron-sized particles. The microscopic morphology of these particles varies significantly depending on the cause. To compensate for the deficiencies of macroscopic analysis, existing technologies use scanning electron microscopes to obtain high-resolution morphological images of these micron-sized particles. However, traditional methods mainly rely on manual observation of morphological images for identification, which has inherent drawbacks such as low efficiency, susceptibility to subjective factors, and lack of quantitative standards. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology. By constructing a thermal runaway cause classification model that includes EfficientNet feature extraction, channel and spatial attention mechanisms, and a multi-layer classifier, and by adopting a data augmentation strategy that simulates SEM physical noise and particulate deformation, this method solves the technical problems of traditional thermal runaway cause identification methods that rely on macroscopic features and expert experience, have strong subjectivity, insufficient accuracy, and low efficiency. It achieves high-precision identification and tracing of different thermal runaway causes such as overcharging, heating, and needle puncture.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology, comprising: receiving an electron micrograph of the micron-level morphology of particulate matter ejected during thermal runaway of a lithium-ion battery; preprocessing the electron micrograph to obtain preprocessed image data; performing data augmentation on the preprocessed image data to obtain standardized image data; processing the standardized image data using a pre-trained thermal runaway cause classification model, wherein the thermal runaway cause classification model outputs a cause classification result for the thermal runaway of the lithium-ion battery; wherein processing the standardized image data using the pre-trained thermal runaway cause classification model includes: extracting multi-scale micron-level morphological features from the standardized image data based on a preset network; applying an attention mechanism to weight the extracted multi-scale micron-level morphological features to obtain weighted features; and progressively reducing and classifying the weighted features through a multi-layer classifier containing at least three fully connected layers to obtain the cause classification result containing cause categories and corresponding confidence scores.

[0007] Optionally, the lithium-ion battery thermal runaway cause tracing method based on particulate morphology further includes: determining the cause category of the cause classification result to obtain a determination result; the determination of the cause category of the cause classification result to obtain a determination result includes: if the confidence score of a single cause category exceeds a preset score threshold, then the cause category is determined to be a thermal runaway cause; if the confidence scores of multiple cause categories are all higher than the preset score threshold, and the difference between the highest score and the second highest score is less than a preset score difference, then it is determined to be a mixed cause.

[0008] Optionally, the lithium-ion battery thermal runaway cause tracing method based on particulate morphology further includes: generating a cause classification report of the cause classification results, wherein the cause classification report includes the judgment result, the confidence score, and the historical performance evaluation index of the thermal runaway cause classification model.

[0009] Optionally, the data augmentation of the preprocessed image data includes at least one of the following methods: simulating physical noise in the scanning electron microscope imaging process in the preprocessed image data; simulating the elastic deformation characteristics of particulate matter in the preprocessed image data.

[0010] Optionally, the preset network is an EfficientNet network.

[0011] Optionally, the preprocessing of the electron microscope image to obtain preprocessed image data includes: denoising the electron microscope image; normalizing the size of the denoised image; and standardizing the grayscale of the normalized image to obtain the preprocessed image data.

[0012] Optionally, the attention mechanism is a mechanism that combines channel attention and spatial attention in a series.

[0013] Secondly, this application provides a lithium-ion battery thermal runaway cause tracing device based on particulate morphology, comprising: an image receiving module configured to receive an electron micrograph of the micron-level morphology of particulate matter ejected during thermal runaway of a lithium-ion battery; a preprocessing module configured to preprocess the electron micrograph to obtain preprocessed image data; a data augmentation module configured to perform data augmentation on the preprocessed image data to obtain standardized image data; and a result output module configured to process the standardized image data using a pre-trained thermal runaway cause classification model, wherein the thermal runaway... The cause classification model outputs the cause classification result of the lithium-ion battery thermal runaway; wherein, the result output module includes: a feature extraction submodule, configured to extract multi-scale micron-level morphological features from the standardized image data based on a preset network; a weighting submodule, configured to apply an attention mechanism to weight the extracted multi-scale micron-level morphological features to obtain weighted features; and a dimensionality reduction and classification submodule, configured to perform progressive dimensionality reduction and classification of the weighted features through a multi-layer classifier containing at least three fully connected layers to obtain the cause classification result including cause categories and corresponding confidence scores.

[0014] Optionally, the lithium-ion battery thermal runaway cause tracing device based on particulate morphology further includes: a determination module configured to determine the cause category of the cause classification result and obtain a determination result.

[0015] Optionally, the lithium-ion battery thermal runaway cause tracing device based on particulate morphology further includes: a report generation module configured to generate a cause classification report of the cause classification result, wherein the cause classification report includes the judgment result, the confidence score, and the historical performance evaluation index of the thermal runaway cause classification model.

[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 lithium-ion battery thermal runaway cause tracing method based on particulate morphology as described 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 lithium-ion battery thermal runaway cause tracing method based on particulate morphology as described above.

[0018] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the lithium-ion battery thermal runaway cause tracing method based on particulate morphology as described above.

[0019] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method and apparatus for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology. The method involves preprocessing, data augmentation, and feature extraction of electron micrographs of the micron-level morphology of particulate matter ejected during thermal runaway from lithium-ion batteries. A thermal runaway cause classification model is then used to analyze the causes of battery thermal runaway. Through the coordinated use of feature extraction, attention mechanism weighting, and multi-level classifier classification, the technical challenge of high-precision, automated mapping from complex microscopic morphology to thermal runaway cause categories is solved. This achieves accurate and automatic identification from particulate matter images to cause categories, with a high tracing accuracy far exceeding that of traditional manual identification methods. Furthermore, after the thermal runaway cause classification model identifies the cause categories, a judgment logic based on confidence scoring is introduced, and a classification report containing quantitative data is generated. This enables clear determination of single causes and effective indication of mixed causes, providing intuitive, reliable, and quantitatively based scientific conclusions for accident investigation and liability determination. Attached Figure Description

[0020] 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.

[0021] Figure 1 This is an application environment diagram of a lithium-ion battery thermal runaway cause tracing method based on particulate morphology in one embodiment of this application. Figure 2 A flowchart illustrating a method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology, provided in an embodiment of this application. Figure 3 for Figure 1 A detailed flowchart of step S14 in the middle section; Figure 4 for Figure 1 Detailed flowchart of step S12 Figure 5 This is a schematic diagram illustrating data augmentation of preprocessed image data in one embodiment of this application; Figure 6 This is a schematic diagram of the channel attention mechanism structure in one embodiment of this application; Figure 7 This is a schematic diagram of the structure of a multi-layer classifier in one embodiment of this application; Figure 8 This is a confusion matrix heatmap for evaluating the performance of a thermal runaway cause classification model in one embodiment of this application; Figure 9 A schematic diagram of the functional modules of a lithium-ion battery thermal runaway cause tracing device based on particulate morphology provided in an embodiment of this application; Figure 10 for Figure 9 A detailed functional module diagram of the result output module; Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] 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.

[0023] 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.

[0024] The lithium-ion battery thermal runaway cause tracing method based on particulate morphology provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be set up independently, integrated into server 102, or placed in the cloud or on another server. Terminal 101 can send electron micrographs to server 102. Server 102 receives the electron micrographs, preprocesses them to obtain preprocessed image data, performs data augmentation on the preprocessed image data to obtain standardized image data, and processes the standardized image data using a pre-trained thermal runaway cause classification model. The thermal runaway cause classification model outputs the cause classification result of the lithium-ion battery thermal runaway. Server 102 can feed back the obtained cause classification result to terminal 101. Furthermore, in some embodiments, the cause classification result can also be implemented by server 102 or terminal 101 independently. For example, terminal 101 can directly process the electron micrographs to be processed, or server 102 can obtain the electron micrographs from the data storage system.

[0025] The terminal 101 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 102 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0026] In one exemplary embodiment, such as Figure 2 As shown, a method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology is provided. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is applied to... Figure 1 Let's take server 102 as an example for illustration. Figure 2 As shown, the method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology includes the following steps: S11, receiving electron micrographs of the micron-scale morphology of particulate matter ejected from the thermal runaway of lithium-ion batteries; S12, preprocess the electron microscope image to obtain preprocessed image data; S13, perform data augmentation on the preprocessed image data to obtain standardized image data; S14, The standardized image data is processed using a pre-trained thermal runaway cause classification model, and the thermal runaway cause classification model outputs the cause classification results of lithium-ion battery thermal runaway. S15, determine the category of the cause based on the cause classification results, and obtain the determination result; S16, Generate a cause classification report of the cause classification results; Among them, such as Figure 3 As shown, step S14 involves processing the standardized image data using a pre-trained thermal runaway cause classification model, including the following steps: S141, based on a preset network, extracts multi-scale micron-level morphological features from standardized image data; S142, an attention mechanism is applied to the extracted multi-scale micron-level morphological features to obtain weighted features; S143, the weighted features are progressively reduced and classified through a multi-layer classifier containing at least three fully connected layers to obtain the cause classification results containing the cause category and the corresponding confidence score.

[0027] The lithium-ion battery thermal runaway cause tracing method based on particulate morphology in this application embodiment solves the technical problems of traditional macroscopic analysis methods and manual identification, which are highly subjective, reliant on expert experience, and costly. It establishes a standardized automated analysis process from image standardization and targeted data enhancement to deep feature extraction, significantly reducing analysis costs and dependence on human experience. Furthermore, in the process of analyzing battery thermal runaway causes using a thermal runaway cause classification model, the synergistic cooperation of feature extraction, attention mechanism weighting, and multi-level classifier classification solves the problem of high-precision and automatic identification between complex microscopic morphology and thermal runaway cause categories. The technical challenges of automated mapping have been overcome, enabling accurate and automatic identification of cause categories from particulate matter images, with a high source tracing accuracy far exceeding that of traditional manual identification methods. Specifically, the applied attention mechanism weighting addresses the technical challenge of complex backgrounds and difficulty in focusing effective morphological features in electron microscopy images. The attention mechanism allows the network to adaptively focus on the most discriminative morphological features for identifying thermal runaway causes, effectively suppressing interference from irrelevant background noise and significantly improving the quality of feature representation and the model's discriminative ability. Finally, by introducing a judgment logic based on confidence scores and generating classification reports containing quantitative data, clear determination of single causes and effective indication of mixed causes are achieved, providing intuitive, credible, and quantitatively based scientific conclusions for accident investigation and liability determination.

[0028] In practice, the collection of particulate matter ejected from the thermal runaway of lithium-ion batteries adopts a multi-stage filtration mesh system. Specifically, the first-stage coarse filter can have a pore size of 100 micrometers, the second-stage fine filter can have a pore size of 50 micrometers, and the bottom layer uses conductive carbon tape to fix the particulate matter.

[0029] In practice, a scanning electron microscope (SEM) is used to perform high-resolution imaging of the collected micron-sized particles. SEM is a powerful scientific instrument that uses a focused electron beam to scan the sample surface point by point, thereby obtaining high-resolution, large-depth-of-field microscopic morphological information. During the imaging process, the particle samples are first conductively treated using a sputtering gold plating process, with the gold layer thickness controlled to 5-10 nanometers. Then, the SEM operating parameters are set, including an accelerating voltage of 5-15 kilovolts, a working distance of 8-12 millimeters, and a magnification of 500-5000 times. Finally, images of the particles are acquired from different fields of view, with at least 50 images obtained for each sample, each with a resolution of no less than 2048×2048 pixels, ensuring the capture of the fine morphological features of the particles.

[0030] In specific implementation, such as Figure 4 As shown, the steps for preprocessing electron microscope images to obtain preprocessed image data include: S121, Denoising the electron microscope image; S122, Perform size normalization on the denoised image; S123, perform grayscale normalization on the size-normalized image to obtain preprocessed image data.

[0031] The denoising process employs a nonlocal mean filtering algorithm, which effectively removes random noise while preserving edge features; the size normalization process adjusts all images to a preset pixel size, specifically 224×224 pixels; and the grayscale normalization process uses the Z-score normalization method to ensure that the image grayscale value distribution has zero mean and unit variance, which can eliminate brightness differences under different imaging conditions.

[0032] In specific implementation, such as Figure 5 As shown, data augmentation methods for preprocessed image data can include: geometric transformation, simulating physical noise, simulating elastic deformation features, optical adjustment, and online adjustment.

[0033] Geometric transformations include rotation, flipping, scaling, and shearing of the original electron micrograph. Rotation can be a random angle from -15° to 15° or a 90° multiplier angle. Flipping includes horizontal, vertical, and diagonal rotations. Scaling can be 0.8-1.2 times.

[0034] The simulated physical noise includes shot noise, charging effect, and scanning distortion. The electron count in the complete statistical model for shot noise follows a Poisson distribution, and its noise variance formula is: The signal-to-noise ratio formula is: In the formula, I Here, t is the signal current, t is the detection time, and e is the elementary charge. It's the detector efficiency. It is the average number of electrons detected. It is an average signal. This is the standard deviation of the noise. For the electric field distribution model of the charging effect, the potential distribution formula is: This formula calculates the value at the sample surface point. At that point, the total electrostatic potential generated by the charge distributed throughout the surface. In the formula, It describes every point on the sample surface. How much charge has accumulated on it, σ s This refers to the area irradiated by the electron beam. These are the coordinates indicating the charge source point. It is the straight-line distance between the observation point and the source point. It is a constant in Coulomb's law. It is a double integral. The formula for the electron beam deflection angle is: In the formula, This refers to the electric field component perpendicular to the electron beam, where e is the elementary charge. It is electronic quality. v It is the electron velocity. L It is electrons in an electric field The effective path length during flight and under its influence. A parameterized model for scan distortion is provided, including linear and nonlinear distortions (barrel / pincushion); the matrix equation for linear distortion is... In the formula This refers to the actual pixel coordinates after distortion. This refers to the original ideal pixel coordinates, and the six parameters (m) of the middle 3×3 matrix. 11 m 12 m 13 m 21 m 22 m 23 The distortion type is determined by these factors; the formula for nonlinear distortion (barrel / pincushion) is: In the formula, It refers to the radial distance from a point in the original image to the center of the image. It is the new radial distance after the distortion. 、 、 It is the distortion coefficient. It is the dominant term, describing the basic distortion pattern. and It is a higher-order correction term used to fit complex distorted curves more accurately.

[0035] By co-simulating shot noise, charging effect and scanning distortion, the inherent imaging defects of electron microscopy images from multiple physical sources are solved, and the generalization ability and robustness of the thermal runaway cause classification model on real and complex SEM image data are improved, thus ensuring high accuracy and high reliability in tracing the cause of thermal runaway.

[0036] The simulated elastic deformation characteristics are image deformations using the moving least squares method, where the weighting function is... In the formula It refers to the distance from pixel x to The Euclidean distance, h refers to the smoothing parameter, and the shape function is... In the formula, This refers to constructing a locally weighted coordinate system at point x. and Similar, but only for a specific control point.

[0037] Using the moving least squares method to simulate elastic deformation produces continuous, smooth, and physically reasonable deformation, forcing the model to focus on the more essential microscopic morphological features that remain unchanged under elastic deformation, thereby improving the model's discriminative ability.

[0038] Optical adjustment mainly refers to the transformation of the pixel values ​​of an image itself, such as brightness, contrast, and saturation adjustment; online adjustment refers to the real-time, random data augmentation transformation of the input image in each training batch.

[0039] Specifically, advanced data augmentation techniques such as MixUp, CutMix, and Mosaic can also be used to achieve data augmentation.

[0040] In practice, by performing data augmentation on the preprocessed image data, 30 augmented samples with different changes can be generated from the original image, indicating the generalization ability of the thermal runaway cause classification model.

[0041] It is understood that in step S14, the thermal runaway cause classification model adopts an extensible framework design, which facilitates the addition of new cause categories. In this embodiment of the application, the cause categories may include four types, namely overcharging, heating, needle puncture, and normal.

[0042] It is understandable that in step S14, the cause classification result output by the thermal runaway cause classification model includes the cause category and the corresponding confidence score.

[0043] It is understandable that in step S15, the cause category of the cause classification result is determined to obtain the determination result, which includes: If the confidence score of a single cause category exceeds a preset score threshold, then that cause category is determined to be a cause of thermal runaway.

[0044] If multiple incentive categories have confidence scores higher than a preset score threshold, and the difference between the highest and second-highest scores is less than a preset score difference, then it is determined to be a mixed incentive.

[0045] In practice, the preset scoring threshold can be 0.8.

[0046] Understandably, in step S16, the cause classification report includes the judgment result, confidence score, and historical performance evaluation indicators of the thermal runaway cause classification model. The historical performance evaluation indicators include, for example, accuracy, recall, and F1 score, to provide macro-level evidence of the reliability of the thermal runaway cause classification model itself and to provide further credibility support for accident analysis.

[0047] In a specific implementation, in step S141, the preset network can be an EfficientNet network, specifically an EfficientNet-B0 network. A composite scaling strategy can be used to balance the network's depth, width, and resolution. The EfficientNet-B0 network can include seven MBConv module groups, each MBConv module consisting of a 1×1 dilated convolution, a depthwise separable convolution, a Squeeze-and-Excitation module, and a 1×1 projective convolution. Specifically, using this network, through progressive downsampling, the input 224×224×3 image can be gradually converted into a 7×7×1280 high-dimensional feature representation, effectively capturing multi-level features of particles from local texture to overall shape.

[0048] In the MBConv module mentioned above, the formula for dilated convolution is: And the expansion rate is calculated using the following formula: In the formula, This refers to a 1×1 convolution. BN This refers to batch normalization. This refers to the activation function. This refers to the number of output channels of the 1×1 convolution. Number of input channels The ratio; the formula for depthwise separable convolution is In the formula, This refers to using a k×k convolution kernel; the Squeeze-and-Excitation module formula includes... In the formula, This refers to the feature map Global average pooling, which will pool the values ​​of each channel. The spatial information is compressed into a single numerical value, and the vector z captures the global distribution of each channel. It is a fully connected layer that serves to reduce dimensionality. This refers to the introduction of nonlinearity. This refers to comparing the learned weights s with the original feature map. Channel-by-channel multiplication; the formula for projective convolution is... In the formula, It refers to another 1×1 convolution.

[0049] Furthermore, when the input and output dimensions match, residual connections are also applied in each MBConv module. The formula for residual connections is: In the formula, This is the final output of the MBConv module. It refers to the feature map obtained after all operations within the module (dilated convolution, depthwise separable convolution, SE module, projective convolution). xThis refers to the original feature map input to the MBConv module. This refers to the input feature map. x The number of channels, This refers to the output feature map. The number of channels, This refers to the step size of the MBConv module.

[0050] Therefore, by using the EfficientNet-B0 network based on a composite scaling strategy as the backbone and leveraging the core MBConv module for multi-level feature extraction, the technical challenge of efficiently and accurately capturing and fusing multi-scale morphological features of particulate matter from complex SEM images was solved, providing the most crucial feature foundation for achieving high-precision tracing of the causes of thermal runaway. Specifically, the MBConv module's internal components address the issue of improving feature richness with low computational cost through 1×1 dilated convolutions, the problem of high computational cost and parameter redundancy in standard convolutions through depthwise separable convolutions, the problem of equal importance among feature channels and the inability to adaptively focus on key information through the Squeeze-and-Excitation module, and the problem of connecting high-dimensional features within the module with the dimensions of the external network through 1×1 projective convolutions. When dimensions match, conditional residual connections address the gradient vanishing and network degradation problems during network training. This ensures the stability of network training and protects the microscopic morphological information extracted from SEM images from being destroyed by the deep network.

[0051] In specific implementation, step S142 employs a combined channel attention and spatial attention mechanism. Channel attention obtains channel descriptors through global average pooling, which are then processed through at least two fully connected layers to generate channel weights, with a compression ratio of 16. Spatial attention, through the fusion of max pooling and average pooling features, generates a spatial weight map via a 7×7 convolution. The two attention mechanisms work synergistically, enabling the network to adaptively focus on the most discriminative morphological features for identifying thermal runaway causes, while suppressing interference from irrelevant background noise.

[0052] The aforementioned channel attention mechanism can be as follows: Figure 6 The diagram shown illustrates the structure of the channel attention mechanism. The input feature of the channel attention mechanism is F∈ℝ^(H×W×C), where F refers to the input feature map, ℝ refers to the real number domain, the feature map height H=7, the feature map width W=7, and the number of feature map channels C=1280. Then, global average pooling is used to compress the global spatial information, which is then used as the input to subsequent fully connected layers. Next, the first fully connected (FC) layer is used for dimensionality reduction, as shown in the formula... In the formula This refers to the weight matrix of the first fully connected layer, and z refers to the global descriptor. Refers to the bias vector of the first fully connected layer. This refers to the output vector of the first fully connected layer, and the weights in the first FC layer dimensionality reduction are initialized using He initialization, as shown in the formula. Next, ReLU activation is used, with the activation function being... In the formula It refers to the output vector after the ReLU activation function. This refers to the ReLU function, which sets all negative input values ​​to 0 while keeping positive input values ​​unchanged, thus enhancing the network's expressive power. Next, it enters the second fully connected (FC) layer for dimensionality increase, as shown in the formula: In the formula, This refers to the weight matrix of the second fully connected layer. This refers to the bias vector of the second fully connected layer. This refers to the original output vector of the second fully connected layer; then, the Sigmoid function is applied for activation, and the activation function is... In the formula, This refers to the channel attention weight vector; then feature recalibration is performed, using the formula: In the formula, This refers to the weighted output feature map, with dimensions equal to the original features. F Same, i.e., identity mapping, These refer to the height, width, and channel index of the feature map, respectively; finally, the weighted feature is obtained. It is understandable that the input 1280-dimensional feature F, after... Figure 6 The channel attention mechanism in the code ultimately outputs a 1280-dimensional weighted feature. .

[0053] In specific implementation, the structure diagram of the multi-layer classifier in step S143 can be as follows: Figure 7As shown, the multi-layer classifier can map the extracted high-dimensional features to the category of thermal runaway causes. The multi-layer classifier employs a progressive dimensionality reduction structure, containing four fully connected layers. The calculation rule for the number of parameters in each layer is: number of parameters = (input dimension × output dimension) + (output dimension). Specifically, the 1280-dimensional features are used as the input to the first fully connected layer (FC1), with an output of 512 dimensions and a parameter count of 1280 × 512 + 512 = 655872. The activation function is ReLU, and the dropout rate is 0.3. Next, the second fully connected layer (FC2) has an input of 512 dimensions and an output of 256 dimensions, with a parameter count of 512 × 256 + 256 = 131328. The activation function is ReLU, and the dropout rate is 0.3. Following this, the third fully connected layer (FC3) has an input of 256 dimensions and an output of... The first fully connected layer has 128 dimensions, with 256 × 128 + 128 = 32896 parameters. The activation function is ReLU, and the dropout rate is 0.2. Next, the fourth fully connected layer (FC4) takes 128 dimensions as input and outputs 4 dimensions, with 128 × 4 + 4 = 516 parameters. This fourth fully connected layer acts as the output layer, mapping the 128-dimensional features to the final four categories: overfilling, heating, needle prick, and normal. The output layer's activation function is Softmax, which generates probability distributions for each category, representing the confidence level of the cause category. Furthermore, each fully connected layer undergoes normalization to accelerate the training process.

[0054] Therefore, by adopting a progressive dimensionality reduction structure and integrating batch normalization, ReLU activation, and Dropout, the multi-layer classifier solves the problem of accurately and robustly mapping high-dimensional abstract morphological features to thermal runaway cause categories, providing objective and quantitative decision-making basis for accident tracing, and ensuring the model's excellent generalization ability under small sample conditions.

[0055] In practical implementation, a comprehensive evaluation of the performance of thermal runaway cause classification models can be conducted using a 4-class confusion matrix. Specifically, for example... Figure 8 The figure shows a confusion matrix heatmap for evaluating the performance of the thermal runaway cause classification model. The total sample size is 240, and the cause categories include overfilling, heating, missing counts, and normal. The figure shows that the accuracy for overfilling, missing counts, and normal categories is 100%, while the accuracy for heating is 90%. This indicates that the thermal runaway cause classification model performs excellently in identifying each category. Furthermore, the overall accuracy of the thermal runaway cause classification model reaches 92.9%. The reliability and significance of the thermal runaway cause classification model's performance can be verified using the McNemar test and Wilson score intervals. The Wilson score interval verification process is as follows: 0.929 represents the observed accuracy, 0.071 represents the observed error rate, 240 represents the total number of test samples, and 1.96 represents the standard normal distribution score corresponding to the 95% confidence interval. This verification process provides a reliable fluctuation range [89.7%, 96.1%] for the overall accuracy of the thermal runaway cause classification model in the example reaching 92.9%. This fluctuation range is high and narrow, strongly demonstrating the stability and reliability of the thermal runaway cause classification model. It shows that the thermal runaway cause classification model can continuously provide high-precision tracing results in actual deployment, rather than being a coincidence.

[0056] For example, such as Figure 9 As shown, a lithium-ion battery thermal runaway cause tracing device based on particulate morphology is provided, comprising: an image receiving module 21, a preprocessing module 22, a data augmentation module 23, a result output module 24, a judgment module 25, and a report generation module 26.

[0057] Image receiving module 21 is configured to receive electron micrographs of the micron-scale morphology of particulate matter ejected from a lithium-ion battery during thermal runaway. Preprocessing module 22 is configured to perform preprocessing on the electron microscope image to obtain preprocessed image data; Data augmentation module 23 is configured to perform data augmentation on the preprocessed image data to obtain standardized image data; The output module 24 is configured to process the standardized image data using a pre-trained thermal runaway cause classification model, and the thermal runaway cause classification model outputs the cause classification result of the lithium-ion battery thermal runaway.

[0058] The determination module 25 is configured to determine the cause category of the cause classification result and obtain the determination result. The report generation module 26 is configured to generate a cause classification report of the cause classification results, the cause classification report including the determination results, the confidence score and the historical performance evaluation index of the thermal runaway cause classification model.

[0059] In specific implementation, such as Figure 10 As shown, the result output module 24 includes: feature extraction submodule 241, weighting submodule 242, and dimensionality reduction and classification submodule 243.

[0060] Feature extraction submodule 241 is configured to extract multi-scale micron-level morphological features from the standardized image data based on a preset network; The weighting submodule 242 is configured to perform an attention mechanism weighting on the extracted multi-scale micron-level topography features to obtain weighted features; The dimensionality reduction and classification submodule 243 is configured to perform progressive dimensionality reduction and classification of the weighted features through a multi-layer classifier containing at least three fully connected layers, to obtain the cause classification result containing the cause category and corresponding confidence score.

[0061] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 11 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores electron micrographs. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology.

[0062] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0063] 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.

[0064] 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.

[0065] 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.

[0066] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0067] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0068] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0069] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0070] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A particle morphology-based lithium-ion battery thermal runaway causation traceback method, characterized in that, include: Electron micrographs of the micron-scale morphology of particulate matter ejected during thermal runaway from lithium-ion batteries were obtained. The electron microscope image is preprocessed to obtain preprocessed image data; The preprocessed image data is augmented to obtain standardized image data; The standardized image data is processed using a pre-trained thermal runaway cause classification model, and the thermal runaway cause classification model outputs the cause classification result of the thermal runaway of the lithium-ion battery. The step of processing the standardized image data using a pre-trained thermal runaway cause classification model includes: Multi-scale micron-level morphological features are extracted from the standardized image data based on a preset network; the preset network is the EfficientNet-B0 network; the EfficientNet-B0 network includes 7 MBConv module groups, each MBConv module consists of a 1×1 dilated convolution, a depthwise separable convolution, a Squeeze-and-Excitation module and a 1×1 projective convolution. The extracted multi-scale micron-level morphological features are weighted by an attention mechanism to obtain weighted features; the attention mechanism is a mechanism of channel attention and spatial attention in series; wherein, channel attention obtains channel descriptors through global average pooling and generates channel weights through at least two fully connected network layers; spatial attention generates a spatial weight map through the fusion of features of max pooling and average pooling and through 7×7 convolution. The weighted features are progressively reduced and classified using a multi-layer classifier containing at least three fully connected layers to obtain the cause classification result, which includes the cause category and the corresponding confidence score.

2. The particle morphology-based lithium-ion battery thermal runaway causation traceback method according to claim 1, wherein, The method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology further includes: determining the cause category of the cause classification result to obtain the determination result; The determination of the cause category based on the cause classification results, to obtain the determination result, includes: If the confidence score of a single cause category exceeds the preset score threshold, then that cause category is determined to be a cause of thermal runaway. If multiple cause categories have confidence scores higher than the preset score threshold, and the difference between the highest and second-highest scores is less than a preset score difference, then it is determined to be a mixed cause.

3. The particle morphology-based lithium-ion battery thermal runaway causation traceback method according to claim 2, wherein, The lithium-ion battery thermal runaway cause tracing method based on particulate morphology further includes: generating a cause classification report of the cause classification results, wherein the cause classification report includes the judgment result, the confidence score, and the historical performance evaluation index of the thermal runaway cause classification model.

4. The particle morphology based lithium-ion battery thermal runaway causation traceback method of claim 1, wherein, The data augmentation of the preprocessed image data includes at least one of the following methods: The preprocessed image data is used to simulate physical noise during the scanning electron microscope imaging process; The preprocessed image data is used to simulate the elastic deformation characteristics of particulate matter.

5. The method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology according to claim 1, characterized in that, The preset network is the EfficientNet network.

6. The method for tracing the causes of thermal runaway in lithium-ion batteries based on particulate morphology according to claim 1, characterized in that, The preprocessing of the electron microscope image to obtain preprocessed image data includes: The electron micrographs are then denoised. The size of the denoised image is normalized. The image after size normalization is subjected to grayscale normalization to obtain the preprocessed image data.

7. A lithium-ion battery thermal runaway cause tracing device based on particulate morphology, characterized in that, include: The image receiving module is configured to receive electron micrographs of the micron-scale morphology of particulate matter ejected from a lithium-ion battery during thermal runaway. The preprocessing module is configured to perform preprocessing on the electron microscope image to obtain preprocessed image data; The data augmentation module is configured to perform data augmentation on the preprocessed image data to obtain standardized image data; The result output module is configured to process the standardized image data using a pre-trained thermal runaway cause classification model, and the thermal runaway cause classification model outputs the cause classification result of the lithium-ion battery thermal runaway. The result output module includes: The feature extraction submodule is configured to extract multi-scale micron-level topographic features from the standardized image data based on a preset network; the preset network is an EfficientNet-B0 network; the EfficientNet-B0 network includes 7 MBConv module groups, each MBConv module consists of a 1×1 dilated convolution, a depthwise separable convolution, a Squeeze-and-Excitation module and a 1×1 projective convolution. The weighting submodule is configured to apply an attention mechanism to the extracted multi-scale micron-level topographic features to obtain weighted features; the attention mechanism is a mechanism of channel attention and spatial attention in series; wherein, channel attention obtains channel descriptors through global average pooling and generates channel weights through at least two fully connected network layers; spatial attention generates a spatial weight map through the fusion of features of max pooling and average pooling and through 7×7 convolution. The dimensionality reduction and classification submodule is configured to perform progressive dimensionality reduction and classification of the weighted features through a multi-layer classifier containing at least three fully connected layers, to obtain the cause classification result containing the cause category and corresponding confidence score.

8. The lithium-ion battery thermal runaway cause tracing device based on particulate morphology according to claim 7, characterized in that, The lithium-ion battery thermal runaway cause tracing device based on particulate morphology further includes: a determination module configured to determine the cause category of the cause classification result and obtain a determination result.

9. The lithium-ion battery thermal runaway cause tracing device based on particulate morphology according to claim 8, characterized in that, The lithium-ion battery thermal runaway cause tracing device based on particulate morphology further includes: a report generation module configured to generate a cause classification report of the cause classification result, wherein the cause classification report includes the judgment result, the confidence score, and the historical performance evaluation index of the thermal runaway cause classification model.

Images