Transform-based battery pack thermal fault analysis method and device

Abstract

The invention discloses a Transform-based battery pack thermal fault analysis method and device, and the method comprises the steps: carrying out the full-coverage real-time capturing of the surface of a battery pack through a thermal imager, and obtaining a real-time thermal imaging video of the surface of the battery pack at a set visual angle; extracting the real-time thermal imaging video through a Backbone feature extraction network to obtain surface thermal features; the method comprises the following steps: carrying out embedded coding by utilizing the surface thermal characteristics of a Transform coder, and decoding the surface thermal characteristics subjected to embedded coding by utilizing a Transform decoder to obtain a decoding output result of the Transform decoder; the decoding output result of the Transform decoder is subjected to multi-task head processing to complete fault diagnosis of the thermal abnormal battery, the fault diagnosis result of the thermal abnormal battery is output, and the fault diagnosis result comprises the fault type, the fault area, the fault bounding box and the confidence coefficient of the fault battery. The method is high in accuracy and high in efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of battery thermal fault diagnosis, and in particular relates to a battery pack thermal fault analysis method and device based on Transformer. Background Art

[0002] At present, there are two main methods for thermal management of lithium-ion batteries, namely model-based methods and data-driven methods. First, the model-based method is to analyze by establishing a thermophysical model of lithium-ion batteries (such as an electrochemical coupling model and an electrothermal coupling model). The method of establishing a physical model can effectively explore the inherent physical laws of lithium-ion batteries and accurately characterize the operating status of the battery at any time. The second is a data-driven method. This method refers to a method in which the system autonomously learns the mapping relationship between input and output from a large number of data and establishes a mathematical analysis model. For example, the voltage, current, SOC, SOH and other indicators of the battery pack are selected for analysis. At present, most data-driven methods use deep learning models for processing. In addition, the fault diagnosis of power equipment through infrared thermal imagers has taken shape. As a non-contact temperature measurement method, it can reduce the amount of data processed by the system. Combined with the corresponding image processing method, it can reduce the dependence on technicians during the detection process.

[0003] In the prior art, the method of establishing a physical model can effectively explore the inherent physical laws of lithium-ion batteries and accurately describe the operating state of the battery at any time. However, the process of establishing a physical model is often very complicated, requiring a large amount of experimental data observation and mechanism analysis. In addition, a good model depends on accuracy parameter identification, but some core parameters are often difficult to obtain on site. At the same time, different lithium battery materials correspond to different models, and it is difficult to cover lithium-ion batteries of all materials with an accurate physical model. At present, most data-driven methods use deep learning models for processing. The data-driven method does not rely on the precise physical model of the battery, but only analyzes data such as the voltage and current of the battery operation. This greatly simplifies the modeling process of lithium-ion batteries, but also depends more on the quality of the algorithm. In addition, whether it is a model-based method or a data-driven method, it is necessary to place a corresponding number of sensors inside the battery pack according to the needs of collecting signals. If you want to achieve the positioning of the cell body in the battery pack, you need more sensors for monitoring, which will inevitably cause redundancy. Summary of the invention

[0004] To this end, the present invention provides a method and device for analyzing thermal failure of a battery pack based on Transformer, which solves the problem that traditional technologies are difficult to accurately and efficiently analyze thermal failure of a battery pack.

[0005] In order to achieve the above object, the present invention provides the following technical solution: a battery pack thermal failure analysis method based on Transformer, comprising the following steps:

[0006] The thermal imager is used to capture the entire surface of the battery pack in real time, thereby obtaining a real-time thermal imaging video of the surface of the battery pack at a set viewing angle;

[0007] Extracting the real-time thermal imaging video of the battery pack surface through a Backbone feature extraction network to obtain surface thermal features in the real-time thermal imaging video of the battery pack surface;

[0008] Using a Transformer encoder to embed and encode the surface thermal features in the real-time thermal imaging video of the battery pack surface, and using a Transformer decoder to decode the embedded and encoded surface thermal features in the real-time thermal imaging video of the battery pack surface to obtain a decoding output result of the Transformer decoder;

[0009] The decoded output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis of the thermal abnormality battery, and the fault diagnosis result of the thermal abnormality battery is output. The fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery.

[0010] As a preferred solution of the Transformer-based battery pack thermal failure analysis method, during the process of fully covering and real-time capturing the battery pack surface with a thermal imager, the thermal imager is controlled to change the viewing angle according to a "M-shaped" orientation on the battery pack surface, and the battery pack is simulated using Ansys Fluent software to simulate the heat generation of the battery pack when a single cell fails, thereby obtaining a real-time thermal imaging video of the battery pack surface.

[0011] As a preferred solution of the Transformer-based battery pack thermal failure analysis method, the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, and the Backbone feature extraction network adopts a ResNet50 residual network.

[0012] As a preferred solution of the battery pack thermal fault analysis method based on Transformer, in the process of embedding the surface thermal features in the real-time thermal imaging video of the battery pack surface by using the Transformer encoder, the surface thermal features in the real-time thermal imaging video of the battery pack surface are added to the position code through the embedding layer operation;

[0013] Assuming the input surface thermal feature value is f, the vector obtained after the embedding layer operation is t∈[1,T], where T is the total number of frames of the input real-time thermal imaging video, and the size of each frame of the real-time thermal imaging video is H×W; the real-time thermal imaging video is divided into N blocks.

[0014] As a preferred solution of the battery pack thermal fault analysis method based on Transformer, after the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded in the position code, the input matrix A of the Transformer encoder is:

[0015] A=F+PE(pos,i)

[0016] In the formula, the position code PE (pos, i) adopts sine and cosine position code, and the calculation method is:

[0017]

[0018] Where pos is the specified block of the input vector F; d is the dimension of the output after position encoding; i is used to traverse the dimension; the odd positions in F are transformed by the cosine function, and the even positions are transformed by the sine function.

[0019] As a preferred solution for the battery pack thermal failure analysis method based on Transformer, the query vector Q, key vector K and numerical vector V are obtained by inputting the matrix A:

[0020]

[0021] Where W Q , W K , W V are three trainable parameter matrices.

[0022] As a preferred solution of the battery pack thermal fault analysis method based on Transformer, the decoded output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis process of the thermal abnormal battery, and the multi-task head includes a bounding box task head, a classification task head and a segmentation task head;

[0023] The bounding box task head consists of a three-layer feedforward neural network; the classification task head consists of a linear layer; and the segmentation task head consists of four convolutional layers and one dynamic convolutional layer.

[0024] The present invention also provides a battery pack thermal failure analysis device based on Transformer, comprising:

[0025] A thermal imaging video acquisition module is used to capture the entire surface of the battery pack in real time through a thermal imager to obtain a real-time thermal imaging video of the battery pack surface at a set viewing angle;

[0026] A surface thermal feature extraction module, used to extract the real-time thermal imaging video of the battery pack surface through a Backbone feature extraction network to obtain the surface thermal features in the real-time thermal imaging video of the battery pack surface;

[0027] a surface thermal feature processing module, configured to embed and encode the surface thermal features in the real-time thermal imaging video of the battery pack surface using a Transformer encoder, and decode the surface thermal features in the real-time thermal imaging video of the battery pack surface after embedding and encoding using a Transformer decoder to obtain a decoding output result of the Transformer decoder;

[0028] The thermal abnormality battery fault diagnosis module is used to complete the fault diagnosis of the thermal abnormality battery by processing the decoding output result of the Transformer decoder through a multi-task head, and output the fault diagnosis result of the thermal abnormality battery. The fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery.

[0029] As a preferred solution of the Transformer-based battery pack thermal failure analysis device, in the thermal imaging video acquisition module, the thermal imager is controlled to change the viewing angle according to a "M-shaped" orientation on the surface of the battery pack, and the battery pack is simulated using Ansys Fluent software to simulate the heat generation of the battery pack when a single cell fails, thereby obtaining a real-time thermal imaging video of the battery pack surface.

[0030] As a preferred solution of the Transformer-based battery pack thermal failure analysis device, in the surface thermal feature extraction module, the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, and the Backbone feature extraction network adopts a ResNet50 residual network.

[0031] As a preferred solution of the battery pack thermal failure analysis device based on Transformer, in the surface thermal feature processing module, the surface thermal features in the real-time thermal imaging video of the battery pack surface are added to the position code through an embedding layer operation;

[0032] Assuming the input surface thermal feature value is f, the vector obtained after the embedding layer operation is t∈[1,T], where T is the total number of frames of the input real-time thermal imaging video, and the size of each frame of the real-time thermal imaging video is H×W; the real-time thermal imaging video is divided into N blocks.

[0033] As a preferred solution of the battery pack thermal failure analysis device based on Transformer, in the surface thermal feature processing module, after the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded in the position code, the input matrix A of the Transformer encoder is:

[0034] A=F+PE(pos,i)

[0035] In the formula, the position code PE (pos, i) adopts sine and cosine position code, and the calculation method is:

[0036]

[0037] Where pos is the specified block of the input vector F; d is the dimension of the output after position encoding; i is used to traverse the dimension; the odd positions in F are transformed by the cosine function, and the even positions are transformed by the sine function;

[0038] In the surface thermal feature processing module, the query vector Q, the key vector K and the value vector V are obtained by inputting the matrix A:

[0039]

[0040] Where W Q , W K , W V are three trainable parameter matrices.

[0041] As a preferred solution of the battery pack thermal fault analysis device based on Transformer, in the thermal abnormality battery fault diagnosis module, the decoded output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis process of the thermal abnormality battery, and the multi-task head includes a bounding box task head, a classification task head and a segmentation task head;

[0042] The bounding box task head consists of a three-layer feedforward neural network; the classification task head consists of a linear layer; and the segmentation task head consists of four convolutional layers and one dynamic convolutional layer.

[0043] The present invention has the following advantages: the thermal imager is used to capture the battery pack surface in full coverage in real time, so as to obtain a real-time thermal imaging video of the battery pack surface under a set viewing angle; the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, so as to obtain the surface thermal features in the real-time thermal imaging video of the battery pack surface; the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded and encoded using a Transformer encoder, and the surface thermal features in the real-time thermal imaging video of the battery pack surface after embedding and encoding are decoded using a Transformer decoder, so as to obtain the decoding output result of the Transformer decoder; the decoding output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis of the thermal abnormality battery, and the fault diagnosis result of the thermal abnormality battery is output, and the fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery. The present invention can obtain the global surface temperature of the battery pack during operation from multiple angles to complete real-time monitoring; better locate the thermal fault cell body; and effectively perform fault detection even in the presence of occlusion, with the mean average precision (mAP) and mean recall (mAR) on the test data set between 0.6-0.75, and an accuracy of 96%, close to 100%. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0045] Figure 1 A schematic flow chart of a method for analyzing thermal failure of a battery pack based on Transformer provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of a technical route of a battery pack thermal failure analysis method based on Transformer provided in an embodiment of the present invention;

[0047] Figure 3 A schematic diagram of the movement direction of a thermal imager during diagnosis of a thermally abnormal battery in a battery pack thermal fault analysis method based on Transformer provided in an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of a Backbone feature extraction network in a Transformer-based battery pack thermal failure analysis method provided in an embodiment of the present invention;

[0049] Figure 5 A schematic diagram of the Transformer architecture in the Transformer-based battery pack thermal fault analysis method provided in an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of a multi-task output head in a Transformer-based battery pack thermal failure analysis method provided in an embodiment of the present invention;

[0051] Figure 7 A schematic diagram of the mAP and mAR values ​​of the diagnostic model on the validation set during the training process of the Transformer-based battery pack thermal fault analysis method provided in an embodiment of the present invention;

[0052] Figure 8 The identification result of the faulty battery under the perspective transformation by the battery pack thermal fault analysis method based on Transformer provided in the embodiment of the present invention

[0053] Fig. 9 This is an architecture diagram of a battery pack thermal failure analysis device based on Transformer provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Example 1

[0056] See also Figure 1 and Figure 2 , an embodiment of the present invention provides a battery pack thermal fault analysis method based on Transformer, comprising the following steps:

[0057] S1. Use a thermal imager to capture the entire surface of the battery pack in real time, and obtain a real-time thermal imaging video of the surface of the battery pack at a set viewing angle;

[0058] S2, extracting the real-time thermal imaging video of the battery pack surface through a Backbone feature extraction network to obtain surface thermal features in the real-time thermal imaging video of the battery pack surface;

[0059] S3, using a Transformer encoder to embed and encode the surface thermal features in the real-time thermal imaging video of the battery pack surface, and using a Transformer decoder to decode the surface thermal features in the real-time thermal imaging video of the battery pack surface after embedding and encoding, to obtain a decoding output result of the Transformer decoder;

[0060] S4. Process the decoded output result of the Transformer decoder through a multi-task head to complete the fault diagnosis of the thermal abnormality battery, and output the fault diagnosis result of the thermal abnormality battery. The fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery.

[0061] See also Figure 3 In this embodiment, in step S1, during the process of capturing the full coverage of the battery pack surface in real time by the thermal imager, the thermal imager is controlled to change the viewing angle on the battery pack surface in a "M-shaped" orientation, and the battery pack is simulated using Ansys Fluent software to simulate the heat generation of the battery pack when a single cell fails, thereby obtaining a real-time thermal imaging video of the battery pack surface.

[0062] In this embodiment, the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, and the Backbone feature extraction network adopts a ResNet50 residual network.

[0063] See also Figure 4 Specifically, the Backbone feature extraction network is responsible for extracting features from the input real-time thermal imaging video and applying these features to subsequent tasks. In order to obtain higher-dimensional and more complex features for each frame of the image, a Backbone network with deeper layers is often required. However, deep networks suffer from gradient vanishing or gradient explosion, resulting in model degradation. Therefore, the embodiment of the present invention uses the ResNet50 residual network in the Backbone feature extraction network part to retain the initial information while avoiding gradient vanishing.

[0064] See also Figure 5 In this embodiment, the Transformer architecture is the main architecture that the present invention relies on. The feature values ​​of the input real-time thermal imaging video are first passed to the Transformer encoder, and the position code is added after the embedding layer operation, and the input video features are supplemented with position information. Then the deformable attention mechanism is calculated, and then the residual connection is performed and normalized to speed up the convergence speed during training. Finally, through two layers of linear layers, namely Figure 5The feed-forward network in performs a linear transformation on the data to obtain the final Transformer encoder output, and the Transformer decoder part implements the query decomposition.

[0065] Specifically, in the process of embedding the surface thermal features in the real-time thermal imaging video of the battery pack surface by using a Transformer encoder, the surface thermal features in the real-time thermal imaging video of the battery pack surface are added to the position code through an embedding layer operation;

[0066] Assuming the input surface thermal feature value is f, the vector obtained after the embedding layer operation is t∈[1,T], where T is the total number of frames of the input real-time thermal imaging video, and the size of each frame of the real-time thermal imaging video is H×W; the real-time thermal imaging video is divided into N blocks.

[0067] Specifically, after the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded in the position code, the input matrix A of the Transformer encoder is:

[0068] A=F+PE(pos,i)

[0069] In the formula, the position code PE (pos, i) adopts sine and cosine position code, and the calculation method is:

[0070]

[0071] Where pos is the specified block of the input vector F; d is the dimension of the output after position encoding; i is used to traverse the dimension; the odd positions in F are transformed by the cosine function, and the even positions are transformed by the sine function.

[0072] Specifically, the query vector Q, key vector K and value vector V are obtained by inputting the matrix A:

[0073]

[0074] Where W Q , W K , W V are three trainable parameter matrices. Among them, the deformable attention mechanism is calculated for the query vector Q and the key vector K, and the calculation method is:

[0075]

[0076] Where x is the input eigenvalue (including Figure 5 K and V in q for Figure 5 Q in the encoder. That is, z is calculated in the encoder qand the deformable attention mechanism of x. M represents the number of multi-head attention mechanisms. m and W m 'are all trainable weight matrices in the mth attention head; W m 'x represents Figure 5 V in; W m It is a linear transformation of the result after attention is applied to V to obtain the output results of different heads. K is the number of sampling points in the eigenvalue. mqk is the attention weight in the mth attention head, which is a scalar. q Represents z q The position coordinates of mqk Represents the position offset of the kth sampling point of the mth attention head relative to the reference point. The calculation method in the Transformer decoder is the same.

[0077] After that, residual connections are made and layer normalization is performed to speed up the convergence of training. Finally, the data is linearly transformed through the feedforward network to obtain the final Transformer encoder output.

[0078] The Transformer decoder part implements query decomposition. In the first layer decoder, the instance query is performed, that is, F = f obtained after the embedding operation in the encoder t With the initial query vector I q ( Figure 5 Q') performs deformable attention to obtain the first-layer box query t∈[1,T]. That is:

[0079]

[0080] Then the query vector is updated, and the lth (l>1) layer For the upper layer Overlay with box query.

[0081]

[0082] Among them, FC stands for frame query The output after the fully connected layer. The decoder has 6 layers in total. Figure 5 The structure of one of the decoders is shown on the right.

[0083] See also Figure 6 In this embodiment, the decoded output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis process of the thermal abnormality battery. The multi-task head includes a bounding box task head, a classification task head and a segmentation task head;

[0084] The bounding box task head includes a three-layer feedforward neural network. For each frame of input video, the output of the bounding box task head is the predicted normalized center point coordinates and the height and width (x, y, w, h) of the bounding box; the classification task head includes a linear layer. For each frame of input video, its output is the probability that the target may be a faulty battery; the segmentation task head uses dynamic convolution, including four layers of convolution and one layer of dynamic convolution.

[0085] In this embodiment, Ansys Fluent software was selected to build a 1P3S lithium iron phosphate battery pack model, and a finite element simulation of the battery was performed. The battery pack has a single cell capacity of 14.6Ah, a single cell minimum stop voltage of 3V, and a single cell maximum stop voltage of 4.3V. Among them, the heat generation of the battery when the three battery cells suddenly had internal short circuit failures within 300s of normal discharge of 1C was simulated. At the same time, 8 viewing angles were set by rotating the camera to monitor the real-time surface thermal imaging of the battery pack. During the simulation, the 1P3S battery went from a fault-free state to thermal failures in cell 1, cell 2, and cell 3 at different times. The thermal imager changes its position according to the illustrated orientation every 2s, and saves the thermal image at this moment as one frame. Table 1 shows the number of video frames obtained and their distribution:

[0086] Table 1 Frame number and distribution of real-time thermal imaging video of battery surface

[0087]

[0088] The above data will be used as a dataset to train the diagnostic network, see Figure 7 , are the mAP and mAR values ​​of the diagnostic model on the validation set during the training process. Table 2 shows the mAP and mAR values ​​of the diagnostic model on the test set. mAP75 is the mAP value calculated when the IOU is greater than 0.75 and is marked as a faulty battery; mAPL represents the mAP value when the detected faulty battery cell is larger than 32*32 pixels. mAR100 represents the AR value when 100 targets are detected per frame of each video; mARL represents the mAR value when the detected faulty battery cell is larger than 32*32 pixels. F1Score is calculated as the ratio of the product of twice the mAP and mAR to the sum of mAR and mAP.

[0089] Table 2 mAP and mAR values ​​of thermal fault diagnosis system for real-time video of lithium-ion battery pack on the test set

[0090]

[0091] As can be seen from Table 2, the diagnostic model still performs well on the test set. The mAP and mAR values ​​of the bounding box task are both over 0.6, and even over 0.75 when the IOU is greater than 0.75. The mAP and mAR values ​​of the segmentation task are both over 0.72. The F1score of the segmentation task is over 0.7, which is 0.09 higher than the F1score of the bounding box task. The segmentation task has a better effect. This proves that the trained model can still achieve good recognition and location of faulty batteries on the test set.

[0092] See also Figure 8 , which are the diagnostic results of the thermal fault diagnosis system for real-time videos of lithium-ion battery packs on the test set videos, including fault masks, bounding boxes of faulty battery packs, and confidence levels. (a)-(c) are the tracking results of thermal faulty batteries in battery cell 2 when the thermal imager moves upward; (d)-(f) are the tracking results of thermal faulty batteries in battery cell 2 when the thermal imager moves to the right; (g)-(i) are the tracking results of thermal faulty batteries in battery cell 3 when the thermal imager moves downward; (j)-(l) are the tracking results of thermal faulty batteries in battery cell 3 when the thermal imager moves to the left. Fault detection can be performed effectively even in the presence of occlusion. The mean average precision (mAP) and mean recall (mAR) of the system on the test dataset are between 0.6-0.75, with an accuracy of 96% and an accuracy close to 100%.

[0093] In summary, the present invention uses a thermal imager to capture the full coverage of the battery pack surface in real time to obtain a real-time thermal imaging video of the battery pack surface under a set viewing angle; the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network to obtain surface thermal features in the real-time thermal imaging video of the battery pack surface; the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded and encoded using a Transformer encoder, and the surface thermal features in the real-time thermal imaging video of the battery pack surface after embedded encoding are decoded using a Transformer decoder to obtain the decoding output result of the Transformer decoder; the decoding output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis of the thermal abnormality battery, and the fault diagnosis result of the thermal abnormality battery is output, and the fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery. The present invention can obtain the global surface temperature of the battery pack during operation from multiple angles to complete real-time monitoring; better locate the thermal fault cell body; and effectively perform fault detection even in the presence of occlusion, with the mean average precision (mAP) and mean recall (mAR) on the test data set between 0.6-0.75, and an accuracy of 96%, close to 100%.

[0094] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0095] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0096] Example 2

[0097] See also Fig. 9 Embodiment 2 of the present invention further provides a battery pack thermal failure analysis device based on Transformer, comprising:

[0098] The thermal imaging video acquisition module 100 is used to capture the entire surface of the battery pack in real time through a thermal imager to obtain a real-time thermal imaging video of the battery pack surface at a set viewing angle;

[0099] The surface thermal feature extraction module 200 is used to extract the real-time thermal imaging video of the battery pack surface through the Backbone feature extraction network to obtain the surface thermal features in the real-time thermal imaging video of the battery pack surface;

[0100] The surface thermal feature processing module 300 is used to embed the surface thermal features in the real-time thermal imaging video of the battery pack surface using a Transformer encoder, and decode the surface thermal features in the real-time thermal imaging video of the battery pack surface after embedding the encoding using a Transformer decoder to obtain a decoding output result of the Transformer decoder;

[0101] The thermal abnormality battery fault diagnosis module 400 is used to complete the fault diagnosis of the thermal abnormality battery by processing the decoded output result of the Transformer decoder through a multi-task head, and output the fault diagnosis result of the thermal abnormality battery, which includes the fault category, fault area, fault boundary box and confidence of the faulty battery.

[0102] In this embodiment, in the thermal imaging video acquisition module 100, the thermal imager is controlled to change the viewing angle according to a "M-shaped" orientation on the surface of the battery pack, and the battery pack is simulated using Ansys Fluent software to simulate the heat generation of the battery pack when a single cell fails, thereby obtaining a real-time thermal imaging video of the surface of the battery pack.

[0103] In this embodiment, in the surface thermal feature extraction module 200, the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, and the Backbone feature extraction network adopts a ResNet50 residual network.

[0104] In this embodiment, in the surface thermal feature processing module 300, the surface thermal features in the real-time thermal imaging video of the battery pack surface are added with position codes through an embedding layer operation;

[0105] Assuming the input surface thermal feature value is f, the vector obtained after the embedding layer operation is t∈[1,T], where T is the total number of frames of the input real-time thermal imaging video, and the size of each frame of the real-time thermal imaging video is H×W; the real-time thermal imaging video is divided into N blocks.

[0106] In this embodiment, in the surface thermal feature processing module 300, the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded in the position code, and the input matrix A of the Transformer encoder is:

[0107] A=F+PE(pos,i)

[0108] In the formula, the position code PE (pos, i) adopts sine and cosine position code, and the calculation method is:

[0109]

[0110] Where pos is the specified block of the input vector F; d is the dimension of the output after position encoding; i is used to traverse the dimension; the odd positions in F are transformed by the cosine function, and the even positions are transformed by the sine function;

[0111] In the surface thermal feature processing module 300, the query vector Q, the key vector K and the value vector V are obtained by inputting the matrix A:

[0112]

[0113] Where W Q , W K , W V are three trainable parameter matrices.

[0114] In this embodiment, in the thermal abnormality battery fault diagnosis module 400, the decoded output result of the Transformer decoder is processed by a multi-task head to complete the thermal abnormality battery fault diagnosis process, and the multi-task head includes a bounding box task head, a classification task head and a segmentation task head;

[0115] The bounding box task head consists of a three-layer feedforward neural network; the classification task head consists of a linear layer; and the segmentation task head consists of four convolutional layers and one dynamic convolutional layer.

[0116] It should be noted that the information interaction, execution process and other contents between the modules of the above-mentioned device are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown in the previous part of the present application, and will not be repeated here.

[0117] Example 3

[0118] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which the program code of the Transformer-based battery pack thermal failure analysis method is stored, and the program code includes instructions for executing the Transformer-based battery pack thermal failure analysis method of Embodiment 1 or any possible implementation thereof.

[0119] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0120] Example 4

[0121] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0122] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the Transformer-based battery pack thermal failure analysis method of Example 1 or any possible implementation thereof.

[0123] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor implemented by reading software codes stored in a memory. The memory can be integrated in the processor or can be located outside the processor and exist independently.

[0124] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0125] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0126] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made to the present invention. Therefore, these modifications or improvements made without departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A battery pack thermal failure analysis method based on Transformer, characterized in that: The following steps are involved: The thermal imager is used to capture the entire surface of the battery pack in real time, thereby obtaining a real-time thermal imaging video of the surface of the battery pack at a set viewing angle; Extracting the real-time thermal imaging video of the battery pack surface through a Backbone feature extraction network to obtain surface thermal features in the real-time thermal imaging video of the battery pack surface; Using a Transformer encoder to embed and encode the surface thermal features in the real-time thermal imaging video of the battery pack surface, and using a Transformer decoder to decode the embedded and encoded surface thermal features in the real-time thermal imaging video of the battery pack surface to obtain a decoding output result of the Transformer decoder; The decoded output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis of the thermal abnormality battery, and the fault diagnosis result of the thermal abnormality battery is output. The fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery.

2. The battery pack thermal failure analysis method based on Transformer according to claim 1, characterized in that: During the process of capturing the full coverage of the battery pack surface in real time by the thermal imager, the thermal imager is controlled to change the viewing angle according to a "M-shaped" orientation on the battery pack surface, and the battery pack is simulated using Ansys Fluent software to simulate the heat generation of the battery pack when a single cell fails, thereby obtaining a real-time thermal imaging video of the battery pack surface.

3. The battery pack thermal failure analysis method based on Transformer according to claim 1, characterized in that: The real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, wherein the Backbone feature extraction network adopts a ResNet50 residual network.

4. The method for analyzing thermal failure of a battery pack based on Transformer according to claim 3, characterized in that: In the process of embedding the surface thermal features in the real-time thermal imaging video of the battery pack surface by using a Transformer encoder, the surface thermal features in the real-time thermal imaging video of the battery pack surface are added to the position code through an embedding layer operation; Assuming the input surface thermal feature value is f, the vector obtained after the embedding layer operation is t∈[1,T], where T is the total number of frames of the input real-time thermal imaging video, and the size of each frame of real-time thermal imaging video is H×W; The real-time thermal imaging video is divided into N blocks.

5. The method for analyzing thermal failure of a battery pack based on Transformer according to claim 4, characterized in that: After the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded in the position code, the input matrix A of the Transformer encoder is: A=F+PE(pos,i) In the formula, the position code PE (pos, i) adopts sine and cosine position code, and the calculation method is: Where pos is the specified block of the input vector F; d is the dimension of the output after position encoding; i is used to traverse the dimension; the odd positions in F are transformed by the cosine function, and the even positions are transformed by the sine function.

6. The method for analyzing thermal failure of a battery pack based on Transformer according to claim 5, characterized in that: By inputting the matrix A, we can get the query vector Q, key vector K and value vector V: Where W Q , W K , W V are three trainable parameter matrices.

7. The method for analyzing thermal failure of a battery pack based on Transformer according to claim 6, characterized in that: The decoded output result of the Transformer decoder is processed by a multi-task head to complete the fault diagnosis process of the thermal abnormality battery, and the multi-task head includes a bounding box task head, a classification task head and a segmentation task head; The bounding box task head consists of a three-layer feedforward neural network; the classification task head consists of a linear layer; and the segmentation task head consists of four convolutional layers and one dynamic convolutional layer.

8. A battery pack thermal failure analysis device based on Transformer, characterized in that: include: A thermal imaging video acquisition module is used to capture the entire surface of the battery pack in real time through a thermal imager to obtain a real-time thermal imaging video of the battery pack surface at a set viewing angle; A surface thermal feature extraction module, used to extract the real-time thermal imaging video of the battery pack surface through a Backbone feature extraction network to obtain the surface thermal features in the real-time thermal imaging video of the battery pack surface; a surface thermal feature processing module, configured to embed and encode the surface thermal features in the real-time thermal imaging video of the battery pack surface using a Transformer encoder, and decode the surface thermal features in the real-time thermal imaging video of the battery pack surface after embedding and encoding using a Transformer decoder to obtain a decoding output result of the Transformer decoder; The thermal abnormality battery fault diagnosis module is used to complete the fault diagnosis of the thermal abnormality battery by processing the decoding output result of the Transformer decoder through a multi-task head, and output the fault diagnosis result of the thermal abnormality battery. The fault diagnosis result includes the fault category, fault area, fault boundary box and confidence of the faulty battery.

9. The battery pack thermal failure analysis device based on Transformer according to claim 8, characterized in that: In the thermal imaging video acquisition module, the thermal imager is controlled to change the viewing angle on the surface of the battery pack according to the "M" shape, and the battery pack is simulated using Ansys Fluent software to simulate the heat generation of the battery pack when a single cell fails, so as to obtain a real-time thermal imaging video of the battery pack surface; In the surface thermal feature extraction module, the real-time thermal imaging video of the battery pack surface is extracted through a Backbone feature extraction network, wherein the Backbone feature extraction network adopts a ResNet50 residual network; In the surface thermal feature processing module, the surface thermal features in the real-time thermal imaging video of the battery pack surface are added to the position code through an embedding layer operation; Assuming the input surface thermal feature value is f, the vector obtained after the embedding layer operation is t∈[1,T], where T is the total number of frames of the input real-time thermal imaging video, and the size of each frame of the real-time thermal imaging video is H×W; the real-time thermal imaging video is divided into N blocks.

10. The battery pack thermal failure analysis device based on Transformer according to claim 9, characterized in that: In the surface thermal feature processing module, after the surface thermal features in the real-time thermal imaging video of the battery pack surface are embedded in the position code, the input matrix A of the Transformer encoder is: A=F+PE(pos,i) In the formula, the position code PE (pos, i) adopts sine and cosine position code, and the calculation method is: Where pos is the specified block of the input vector F; d is the dimension of the output after position encoding; i is used to traverse the dimension; the odd positions in F are transformed by the cosine function, and the even positions are transformed by the sine function; In the surface thermal feature processing module, the query vector Q, the key vector K and the value vector V are obtained by inputting the matrix A: Where W Q , W K , W V are three trainable parameter matrices; In the thermal abnormality battery fault diagnosis module, the decoded output result of the Transformer decoder is processed by a multi-task head to complete the thermal abnormality battery fault diagnosis process, and the multi-task head includes a bounding box task head, a classification task head and a segmentation task head; The bounding box task head consists of a three-layer feedforward neural network; the classification task head consists of a linear layer; and the segmentation task head consists of four convolutional layers and one dynamic convolutional layer.

Images