Lithium battery internal micro-gas generation detection method and system based on neural network matching bubble confirmation coefficient

Abstract

This invention discloses a method and system for detecting micro-gas generation inside lithium batteries based on neural network matching bubble confirmation coefficients. The method includes the following steps: scanning the lithium battery surface based on ultrasonic penetration method and bottom wave height method, and calculating the maximum attenuation C respectively. A Maximum attenuation C B Based on the maximum attenuation C A Maximum attenuation C B Simulation tests yielded the corresponding bubble confirmation coefficients Sf and Rf, a training set was constructed, and a neural network model was trained. The maximum attenuation C was obtained by scanning imaging of the lithium battery under test using the ultrasonic penetration-bottom wave height method. A Maximum attenuation C B The optimal bubble confirmation coefficients Sf and Rf are obtained based on the trained neural network model, and ultrasonic imaging of internal gas-producing lithium batteries is performed based on these optimal bubble confirmation coefficients Sf and Rf. This invention achieves more comprehensive and accurate imaging and judgment of the micro-gas-producing state inside lithium batteries.

Description

Technical Field

[0001] This invention relates to the field of lithium battery testing technology, specifically to a method and system for detecting micro-gas generation inside lithium batteries based on neural network matching bubble confirmation coefficients. Background Technology

[0002] Because lithium batteries are often completely covered by metal, they are essentially black boxes, making it impossible to directly assess their internal condition from a macroscopic perspective. Internal defects are also difficult to accurately detect using traditional methods such as internal resistance testing and charge-discharge cycle testing. This is especially true for minute defects within lithium battery materials, such as microbubbles and microcracks, which are difficult to detect and pose significant safety risks. Ultrasonic waves, with their excellent penetration and directionality, can provide relatively clear detection results for subtle changes in material structure. Therefore, ultrasonic non-destructive testing technology is gaining increasing attention.

[0003] Traditional ultrasonic testing has certain limitations. First, ultrasonic testing is difficult to detect defects close to the area below the test area, resulting in a blind zone. To prevent the scattering of ultrasonic waves, there are also certain requirements for the surface of the object being tested, often requiring the use of a coupling agent. At the same time, for multi-layered and porous structures, scattering is often more complex and difficult to handle.

[0004] For example, the patent publication number CN113533523A describes a method and apparatus for detecting internal defects in lithium batteries that integrates linear and nonlinear ultrasonic features. This method uses ultrasonic transmitting and receiving probes at both ends of the lithium battery to perform ultrasonic penetration detection to obtain the transmission coefficient. The ultrasonic signal is then processed using Fourier transform to obtain a nonlinear coefficient (the ratio of the second harmonic amplitude to the first harmonic amplitude). The presence of internal defects is determined by checking whether the obtained nonlinear transmission coefficient falls within the confidence interval of a normal lithium battery coefficient. However, the ultrasonic penetration detection method used in this method is prone to errors due to diffraction at the edges of micro-gas formations within the battery. Furthermore, the fixed window width of the Fourier transform used to process the ultrasonic signal cannot address the processing of non-steady-state signals, meaning it cannot effectively reduce noise and extract features from the ultrasonic signal. Consequently, the detection fails to produce an image of gas formation within the lithium battery, and the boundaries and areas of the bubbles cannot be displayed.

[0005] For example, a roller array ultrasonic sensor for detecting the internal state of a lithium-ion battery, as disclosed in patent publication number CN115524401A, uses the defect echo method to detect and image internal defects of the lithium battery by receiving returned ultrasonic waves. However, the defect echo method has a detection blind zone for lithium batteries and cannot detect micro-gas generated inside the lithium battery near the surface. Since no specific judgment method is used, it is impossible to make a fast and accurate judgment on the detection results. Summary of the Invention

[0006] To overcome the shortcomings and deficiencies of existing technologies, this invention provides a method for detecting micro-gas production inside lithium batteries based on a neural network-matched bubble confirmation coefficient. This invention employs an ultrasonic detection method combining ultrasonic penetration and bottom wave height methods, specifically an ultrasonic penetration-bottom wave height method. This method enables faster and more accurate detection of micro-gas production inside lithium batteries, improving the detection accuracy. An optimal coefficient selection method based on a BP neural network is used to match the optimal confirmation coefficient of internal gas production in the lithium battery corresponding to the signal amplitude information obtained from the scan. The optimal coefficient is used to create an ultrasonic imaging map of the internal gas production in the lithium battery. Furthermore, a bubble imaging area boundary correction algorithm based on ultrasonic detection of internal gas production in the lithium battery is used to obtain a clear boundary for the area of ​​the internal gas-producing bubbles, achieving a more comprehensive and accurate imaging and judgment of the internal micro-gas production state of the lithium battery.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a method for detecting micro-gas generation inside lithium batteries based on neural network matching bubble confirmation coefficients, comprising the following steps:

[0009] Based on the ultrasonic penetration method, the peak-to-peak value of the ultrasonic signal is obtained and a feature matrix A of the two-dimensional peak-to-peak position of the transmitted wave is generated.

[0010] The difference between the maximum and minimum values ​​within the characteristic matrix A is used to obtain the maximum attenuation C of the peak-to-peak value of the transmission method ultrasound. A ;

[0011] Based on the bottom wave height method, the peak-to-peak value of the ultrasonic signal of the bottom wave height method is obtained and a feature matrix B of the two-dimensional peak-to-peak value position of the bottom wave is generated.

[0012] The difference between the maximum and minimum values ​​within the characteristic matrix B is used to obtain the maximum attenuation C of the peak-to-peak value of the ultrasonic wave using the bottom wave height method. B ;

[0013] Based on the maximum attenuation C A Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Sf, based on the maximum attenuation C. B Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Rf;

[0014] Based on the maximum attenuation C A Maximum attenuation C B A training set is constructed using the bubble confirmation coefficients Sf and Rf. This training set is then input into the neural network model for training. The maximum decay C is then calculated. A Maximum attenuation C BAs input variables, bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are used as the true values ​​of the corresponding output values. The training is iterated until the training stopping condition is met, and the trained neural network model is obtained.

[0015] The maximum attenuation C of the lithium battery under test was obtained by scanning imaging based on the ultrasonic penetration-bottom wave height method. A Maximum attenuation C B The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model.

[0016] Ultrasonic imaging of internally generated gas lithium batteries is performed based on the optimal bubble confirmation coefficients Sf and Rf.

[0017] As a preferred technical solution, the method based on the maximum attenuation C A Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Sf, based on the maximum attenuation C. B To conduct a simulation test and obtain the corresponding bubble confirmation coefficient Rf, the specific steps include:

[0018] A model of a lithium battery is created, which includes the basic components of a lithium battery and hydrogen bubbles with adjustable volume.

[0019] The projected area S of hydrogen bubbles on the two-dimensional plane of the lithium battery in the calculation model;

[0020] Based on the ultrasonic penetration-bottom wave height method, scanning imaging of lithium batteries was performed, the imaging bubble area S1 was calculated, and the maximum attenuation C under the current hydrogen bubble volume was calculated. A Maximum attenuation C B And set the initial values ​​of bubble confirmation coefficient Sf and bubble confirmation coefficient Rf;

[0021] Determine whether the projected area S is equal to the current imaging bubble area S1. If they are equal, use the current bubble confirmation coefficient Sf and bubble confirmation coefficient Rf as the corresponding bubble confirmation coefficient Sf and corresponding bubble confirmation coefficient Rf. If they are not equal, adjust the values ​​of bubble confirmation coefficient Sf and bubble confirmation coefficient Rf according to the set step size until the projected area S is equal to the current imaging bubble area S1.

[0022] As a preferred technical solution, calculating the imaging bubble area S1 specifically includes:

[0023] The lithium battery is scanned and imaged using the ultrasonic penetration-bottom wave height method to obtain the imaging matrix P. The number of pixels containing bubbles in the imaging matrix P is calculated, and the area of ​​the imaging bubble is obtained by multiplying the number of pixels by the size of the corresponding pixels.

[0024] As a preferred technical solution, based on the maximum attenuation CA Maximum attenuation C B The training set is constructed using bubble confirmation coefficients Sf and Rf, specifically including:

[0025] By changing the size of the hydrogen bubbles in the model, multiple sets of maximum attenuation C were obtained. A The corresponding bubble confirmation coefficient Sf, and the maximum attenuation C B The corresponding bubble confirmation coefficient Rf is used to construct the training set.

[0026] As a preferred technical solution, the training set is input into the neural network model for training, specifically including:

[0027] The neural network model employs a backpropagation (BP) neural network with a supervised learning approach. When a pair of learning patterns is provided to the BP neural network, the activation values ​​of neurons in the input layer are propagated through each hidden layer to the output layer. The output layer neurons output the network response corresponding to the input pattern. The error between the output value and the actual value of the BP neural network is calculated using a loss function. Following the principle of reducing error, the connection weights are adjusted layer by layer from the output layer through the hidden layers back to the input layer. The gradient between the error and each weight value is calculated through backpropagation, and the weight values ​​are updated accordingly. Finally, the maximum decay C is fitted. A Maximum attenuation C B The nonlinear relationship between the bubble confirmation coefficient Sf and the bubble confirmation coefficient Rf.

[0028] As a preferred technical solution, the training stop conditions include the error signal meeting a set range or the training count reaching a set number.

[0029] As a preferred technical solution, ultrasonic imaging of lithium-ion batteries with gas generation inside the battery is performed based on the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf. The method also includes a boundary correction step for the gas generation area inside the lithium-ion battery, specifically comprising:

[0030] The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model.

[0031] Traverse the feature matrix A to obtain the maximum value Amax of feature matrix A, and traverse the feature matrix B to obtain the maximum value Bmax of feature matrix B.

[0032] If the value of a point on the feature matrix A is less than the product of the bubble confirmation coefficient Sf and the maximum value Amax, and the value of the feature matrix B at that point is greater than the product of the bubble confirmation coefficient Rf and the maximum value Bmax, then an imaging matrix P is generated and assigned the value a at that point, and the value b at the other points, where a >> b.

[0033] Imaging of internally generated gas lithium batteries using ultrasound based on imaging matrix P.

[0034] As a preferred technical solution, the steps for generating the imaging matrix P include:

[0035] The feature matrix A and feature matrix B have the same number of rows and columns. During the traversal, nested loops are performed with the same row number i and column number j. The values ​​of row number i and column number j during the loop correspond to the same position point in feature matrix A and feature matrix B. Logical AND is used to perform conditional judgment to generate the value of imaging matrix P.

[0036] The present invention also provides a lithium battery internal micro-gas generation detection system based on neural network matching bubble confirmation coefficient, comprising: an ultrasonic scanning control module, a maximum attenuation calculation module, a bubble confirmation coefficient acquisition module, a training set construction module, a network training module, an optimal bubble confirmation coefficient output module, and an imaging module;

[0037] The ultrasonic scanning control module is used to scan the lithium battery surface based on the ultrasonic penetration method and the bottom wave height method, respectively.

[0038] The maximum attenuation calculation module is used to calculate the maximum attenuation C respectively. A Maximum attenuation C B ;

[0039] Maximum attenuation C A The calculations include:

[0040] Based on the peak-to-peak value of the ultrasonic signal obtained by the ultrasonic transmission method, a feature matrix A is generated to show the two-dimensional peak-to-peak position of the transmitted wave. The difference between the maximum and minimum values ​​within feature matrix A is calculated to obtain the maximum attenuation C of the ultrasonic peak-to-peak value obtained by the transmission method. A ;

[0041] Maximum attenuation C B The calculations include:

[0042] The peak-to-peak value of the ultrasonic signal is obtained based on the bottom wave height method, and a feature matrix B is generated to determine the two-dimensional peak-to-peak position of the bottom wave. The difference between the maximum and minimum values ​​in the feature matrix B is calculated to obtain the maximum attenuation C of the ultrasonic peak-to-peak value obtained by the bottom wave height method. B ;

[0043] The bubble confirmation coefficient acquisition module is used to obtain the maximum attenuation C. A Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Sf, based on the maximum attenuation C. B Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Rf;

[0044] The training set construction module is used to construct the training set based on the maximum decay C. A Maximum attenuation C B Construct a training set using bubble confirmation coefficients Sf and Rf;

[0045] The network training module is used to input the training set into the neural network model for training, with a maximum decay C. A Maximum attenuation C B As input variables, bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are used as the true values ​​of the corresponding output values. The training is iterated until the training stopping condition is met, and the trained neural network model is obtained.

[0046] The optimal bubble confirmation coefficient output module is used to obtain the maximum attenuation C obtained by scanning imaging of the lithium battery under test based on the ultrasonic penetration-bottom wave height method. A Maximum attenuation C B And based on the trained neural network model, the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained;

[0047] The imaging module is used for ultrasonic imaging of internally generated gas lithium batteries based on the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf.

[0048] As a preferred technical solution, a gas-generating area boundary correction module is also provided. This module is used to correct the gas-generating area boundary of the ultrasonically imaged lithium battery, specifically including:

[0049] The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model.

[0050] Traverse the feature matrix A to obtain the maximum value Amax of feature matrix A, and traverse the feature matrix B to obtain the maximum value Bmax of feature matrix B.

[0051] If the value of a point on the feature matrix A is less than the product of the bubble confirmation coefficient Sf and the maximum value Amax, and the value of the feature matrix B at that point is greater than the product of the bubble confirmation coefficient Rf and the maximum value Bmax, then an imaging matrix P is generated and assigned the value a at that point, and the value b at the other points, where a >> b.

[0052] Imaging of internally generated gas lithium batteries using ultrasound based on imaging matrix P.

[0053] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0054] (1) This invention combines the advantages of the penetration method and the bottom wave height method, with the ultrasonic penetration method as the main method and the bottom wave height method as the supplement. Compared with the single ultrasonic penetration method and bottom wave height method, this invention can not only obtain more accurate data on the micro-gas generated inside the lithium battery and improve the detection accuracy of micro-gas, but also avoid various errors caused by ultrasonic diffraction phenomena and ultrasonic detection blind zones, thereby achieving accurate judgment of the size of micro-gas generated inside the lithium battery.

[0055] (2) The present invention uses a BP neural network to match the optimal internal gas production confirmation coefficient of the lithium battery corresponding to the signal amplitude information obtained by scanning, and uses the optimal coefficient to make an ultrasonic internal gas production lithium battery imaging map, so as to realize a more comprehensive and accurate imaging and judgment of the micro gas production state inside the lithium battery.

[0056] (3) The present invention is based on the bubble imaging area boundary correction algorithm of ultrasonic detection of gas generation in lithium battery to obtain a clear boundary of the gas generation bubble area in lithium battery, which further improves the accuracy of gas generation detection and imaging and clearly reflects the gas generation status of lithium battery. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the method for detecting micro-gas generation inside a lithium battery based on neural network matching bubble confirmation coefficients according to the present invention.

[0058] Figure 2 (a) is a schematic diagram of ultrasonic testing for lithium batteries with no internal defects using the ultrasonic penetration testing method.

[0059] Figure 2 (b) is a schematic diagram of ultrasonic testing for small defects inside lithium batteries using the ultrasonic penetration testing method.

[0060] Figure 2 (c) is a schematic diagram of ultrasonic testing for lithium batteries with large internal defects using the ultrasonic penetration testing method.

[0061] Figure 3 (a) is a schematic diagram of ultrasonic testing for lithium batteries with no internal defects, using the bottom wave height detection method;

[0062] Figure 3 (b) is a schematic diagram of ultrasonic testing for lithium batteries with internal defects using the bottom wave height detection method;

[0063] Figure 4 This is a schematic diagram of the multi-layer architecture of the BP neural network of the present invention;

[0064] Figure 5 This is a schematic diagram of the BP neural network structure of the present invention;

[0065] Figure 6 This is a schematic diagram of a traditional ultrasonic scanning imaging method for gas generation inside a lithium battery.

[0066] Figure 7 This is a schematic diagram of the imaging effect of the algorithm based on the boundary correction of the gas production area inside the lithium battery according to the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0068] Example 1

[0069] like Figure 1 As shown, this embodiment provides a method for detecting micro-gas production inside a lithium battery based on a neural network-matched bubble confirmation coefficient. This method is based on ultrasonic detection using the ultrasonic penetration-bottom wave height method. Ultrasonic waves are generated by an ultrasonic probe and allowed to pass through the battery. The ultrasonic waves propagate inside the battery, and changes in the micro-gas production state inside the battery will lead to changes in acoustic characteristics, affecting the propagation characteristics of the ultrasonic waves at the corresponding locations. When different ultrasonic scanning methods are used to detect lithium batteries, they will also affect the propagation law of ultrasonic waves in different ways. By combining the ultrasonic penetration method and the bottom wave height method, an ultrasonic scanning bubble confirmation coefficient is defined to locate the internal gas production. The bubble area boundary is corrected and the exact internal gas production bubble area is presented by combining the characteristics of the ultrasonic signal amplitude changes of both methods. The optimal internal gas production bubble confirmation coefficient corresponding to the signal amplitude information obtained by the BP neural network is matched, and the ultrasonic imaging of the internal gas production lithium battery is performed using the optimal coefficient.

[0070] The method specifically includes the following steps:

[0071] S1: The ultrasonic signal amplitude data of the ultrasonic transmission method is obtained by scanning the lithium battery surface based on the ultrasonic transmission method, and the ultrasonic signal amplitude data of the bottom wave height method is obtained by scanning the lithium battery surface based on the bottom wave height method.

[0072] In this embodiment, a piezoelectric ultrasonic probe is used. Based on the piezoelectric effect, an electrical signal is converted into mechanical vibration. A control system, such as a microcontroller, generates an excitation signal that changes the electric field of the energy storage element in the probe. The piezoelectric crystal then converts electrical energy into mechanical energy. Depending on the type of battery used, the ultrasonic frequency range can be selected from 500kHz to 20MHz. Ultrasonic waves are received at the appropriate location and transmitted to a host computer.

[0073] like Figure 2 (a) Figure 2 (b) and Figure 2As shown in (c), the ultrasonic penetration method is used to scan the surface of the lithium battery. Focused scanning is used instead of the traditional wide-beam fixed-point scanning. The ultrasonic penetration method uses a dual-probe mode with one transmitter and one receiver. The two probes are placed on both sides of the lithium battery. The energy change after the pulse wave or continuous wave penetrates the workpiece is used to detect the workpiece. When the size of the micro-gas generated inside the lithium battery is smaller than the beam width of the detector, the ultrasonic penetration method alone has low measurement sensitivity. The ultrasonic penetration method alone has certain deficiencies in the detection sensitivity of the micro-gas generated inside the lithium battery.

[0074] like Figure 3 (a) and Figure 3 As shown in (b), this embodiment also uses the bottom wave height method for assistance to achieve more complete detection of micro-gas generation inside the lithium battery and reduce the false negative rate. Ultrasonic waves are generated by the ultrasonic probe and pass through the workpiece to be tested. When the sound wave encounters a defect inside the workpiece, it will be reflected at the interface between the intact part of the workpiece and the defect. The nature and location of the material defect are determined by analyzing the reflected ultrasonic signal to determine the condition and location of the defect.

[0075] S2: Selecting bubble confirmation coefficients Sf and Rf based on a BP neural network;

[0076] In this embodiment, the quality of bubble imaging in the area correction algorithm depends on the appropriate selection of bubble confirmation coefficients Sf and Rf. This embodiment uses a backpropagation (BP) neural network to select the bubble confirmation coefficients Sf and Rf. Figure 4 As shown, the BP network employs a supervised learning approach. When a pair of learning patterns is provided to the BP network, the activation values ​​of neurons in the input layer propagate through each hidden layer to the output layer. The output layer neurons then output the network response corresponding to the input pattern. Following the principle of reducing error (the difference between the expected and actual output), the connection weights are adjusted layer by layer from the output layer through the hidden layers back to the input layer. The accuracy of the network's response to the input pattern continuously improves with backpropagation training until the error signal meets the allowable range or the training iterations reach the pre-designed number of iterations. Through training on a large dataset, the nonlinear relationship between the input and output variables can be fitted.

[0077] The larger the bubble volume, the greater its thickness and its projection on the two-dimensional scanning plane of the lithium battery. A larger bubble thickness will affect the maximum attenuation C of the ultrasonic PP value. A and C B (That is, the difference between the maximum and minimum values ​​in the bubble PP value matrix, C) A and C B These represent the maximum PP value attenuation for the penetration method and the bottom wave height method, respectively. For each bubble volume, there is a corresponding pair of optimal confirmation coefficients Sf and Rf, while C... A and CB It is clearly related to Sf and Rf, but highly nonlinear. By repeatedly conducting simulation experiments, the value at a specific C can be obtained. A and C B The optimal values ​​of Sf and Rf are found at these values, which change the bubble volume and thus C. A and C B The dataset is obtained by taking the values ​​of Sf and Rf and extracting the corresponding optimal values. Let C be the dataset. A and C B As input variables, the optimal Sf and Rf are used as approximations (true values) and input into the network for learning, to obtain C. A and C B The nonlinear relationship between Sf and Rf allows for the prior calculation of C in this test when detecting gas generation within a lithium battery. A and C B The optimal Sf and Rf values ​​are then selected for imaging. The specific steps are as follows:

[0078] Step 1: Model the lithium battery using COMSOL software. The model contains the basic components of a lithium battery and a hydrogen bubble with an adjustable volume.

[0079] Step 2: Calculate the projected area S of the bubble in the two-dimensional plane of the lithium battery in the model;

[0080] Step 3: Use the ultrasonic penetration-bottom wave height method to scan and image the lithium battery, and calculate the area S1 of the imaging bubble at this time;

[0081] In the area boundary correction algorithm, an imaging matrix P is generated. The value of P is 'a' at the location where a bubble is considered to be present, and the value of P is 'b' at the location where there is no bubble. By counting the number of 'a' values ​​in P, the number of pixels containing bubbles can be obtained. Since the size of the pixels is known, the calculated area of ​​the bubble can be obtained by multiplying the two values.

[0082] Step 4: Calculate the difference C between the maximum and minimum values ​​in the characteristic matrix A of the ultrasonic penetration-bottom wave height method. A And the difference C between the maximum and minimum values ​​in feature matrix B. B ;

[0083] Step 5: If S1 = S, then record the values ​​of Sf and Rf selected in Step 3, as well as the C calculated in Step 4. A C B If the values ​​of S1 and S are not equal, adjust the values ​​of Sf and Rf until S1 and S are equal.

[0084] In this embodiment, when the set values ​​of Sf and Rf change, the bubble area obtained from imaging also changes accordingly. When the optimal values ​​of Sf and Rf are unknown, initial values ​​of Sf and Rf are customized and self-tuned. The optimal values ​​of Sf and Rf are obtained when the bubble area obtained from imaging matches the bubble area set in the simulation model. The optimal values ​​of Sf and Rf are then compared with the calculated C... A and C B Record it and use it as the training set for the neural network;

[0085] In this embodiment, the initial values ​​of Sf and Rf can preferably be any values ​​within the range of 0.8-0.9. The values ​​of Sf and Rf are adjusted according to a set step size, which can be calculated by the bisection method, so that the bubble area obtained by imaging continuously approaches the bubble area set by the simulation model.

[0086] Step 6: Change the internal gas production volume in the model, and repeat steps 2, 3, 4, and 5 above to obtain dataset L. Dataset L contains C. A C B The values ​​of Sf and RF were calculated, and 200 datasets were collected, with 80% used as the training set and 20% as the test set.

[0087] Step 7: Input the training set data into the neural network for training. The i-th C in the dataset L Ai C Bi The value is used as an input variable, Sf i With Rf i To obtain the true value of the output, a backpropagation (BP) neural network is constructed using the Newff function to train the network model. The hidden layer activation function is the Tansig function, and the output layer function is the Purelin function. The number of training iterations is set to 15,000, meaning that if the network has not converged after 15,000 iterations, the iteration stops. The learning rate is set to 0.01, the momentum factor is set to 0.95, and the minimum performance gradient is set to 0.00001.

[0088] Step 8: Input the test set data into the trained neural network to verify the stability of the neural network model. If the maximum relative error is less than 5% and the average relative error is less than 4%, the model is considered to have reliable prediction performance, and the neural network training is completed.

[0089] The trained neural network can be used for the coefficient matching part of the ultrasonic penetration-bottom wave height method. During the imaging process, the difference C between the maximum and minimum values ​​in feature matrix A and feature matrix B of the lithium battery penetration scanning result can be calculated first. A and C B Then, the optimal bubble confirmation coefficients Sf and Rf are matched by the neural network, thus obtaining the image with the best imaging effect.

[0090] In this embodiment, the BP neural network training process is as follows:

[0091] Initialization steps: Initialize and assign values ​​to each weight in the network.

[0092] Data creation and transmission steps: Use the training set portion of the dataset L above as the training set for the neural network, where C in dataset L... Ai C Bi The value is used as an input variable for forward propagation, while Sf i With Rf i The true value of the output value is used for correction.

[0093] Error estimation steps: Calculate the error between the output value and the actual value of the BP neural network using the loss function.

[0094] The steps for updating weight values ​​are as follows: backpropagation is used to calculate the gradient between the error and each weight value, and this gradient is then used to update each weight value.

[0095] The training process involves repeating the data establishment and transmission steps, the error estimation steps, and the weight value update steps until the obtained error value reaches the expected value, thereby successfully training the model to fit the relationship between the micro-gas generation state inside the lithium battery and the characteristic value of the ultrasonic signal.

[0096] like Figure 5 As shown, a BP neural network can be divided into three parts: an input layer, a hidden layer, and an output layer. i For input data (C in L) Ai C Bi Value as input variable x 1i and x 2i ), w i b is the weight value. i Here, f is the bias, and h is the activation function. i y represents the value obtained after the data passes through the hidden layer. i This is the output value. The formula for calculating data from the input layer to the hidden layer during forward propagation in the network is:

[0097] (1)

[0098] (2)

[0099] The formula for calculating data from the hidden layer to the output layer is:

[0100] (3)

[0101] (4)

[0102] During backpropagation, the error between the output value and the actual value is first calculated, i.e., the defined loss function l. Then, the gradient between the error and each weight value is calculated and used to update the weight values. For the weight values ​​corresponding to the hidden layers (e.g., ... Figure 5 w in i The calculation formula is:

[0103] (5)

[0104] In the formula Let w be the learning rate of the neural network. For network layers far from the final output layer, the gradient can be calculated using the chain rule. For example, for w1, the update formula is:

[0105] (6)

[0106] In the formula, l is the loss function;

[0107] The constructed BP neural network contains one hidden layer with 5 neurons. The activation function of the hidden layer is the tansig function, and its formula is:

[0108] (7)

[0109] In the formula: x is the input value.

[0110] The activation function of the output layer uses a pureline function (such as...). It is a linear transfer function. When calculating the error during forward propagation, i.e., the loss function l, the mean squared error (MSE) is used, and its calculation formula is:

[0111] (8)

[0112] S3: Ultrasonic imaging of lithium batteries with internal gas generation based on an algorithm for correcting the boundary of the gas generation area inside the lithium battery.

[0113] In this embodiment, the advantages of the penetration method and the bottom wave height method are combined to propose a bubble imaging area boundary correction algorithm for ultrasonic detection of gas generation inside lithium batteries. The purpose of the correction is that in traditional lithium battery imaging images, there is often a phenomenon of blurred internal gas boundaries. The dark part in the imaging effect image, that is, the part with lower signal amplitude, does not mean that there is a bubble at that location. This error is caused by the diffraction phenomenon of ultrasonic waves at the bubble boundary.

[0114] The implementation process of the algorithm for correcting the boundary of gas generation area inside a lithium battery includes:

[0115] Step 1: Calculate the signal PP value (peak-to-peak value) at the point obtained by the ultrasonic signal value obtained by the two-dimensional surface scanning of the lithium battery using the penetration method, and generate the two-dimensional PP value location feature matrix A of the penetration wave;

[0116] Step 2: Iterate through the feature matrix A to find the maximum value Amax and the minimum value Amin, and calculate the difference between them to obtain C. A ;

[0117] Step 3: Calculate the signal PP value at the point obtained by scanning the two-dimensional surface of the lithium battery using the bottom wave height method, and generate the two-dimensional bottom wave PP value position feature matrix B;

[0118] Step 4: Iterate through the feature matrix B to find the maximum value Bmax and the minimum value Bmin, and calculate the difference between them to obtain C. B ;

[0119] Step 5: Use the trained neural network to match the optimal penetrating wave bubble to confirm the attenuation coefficient Sf and the bottom wave bubble to confirm the attenuation coefficient Rf;

[0120] Step 6: Traverse and judge if the value of a point in feature matrix A is less than the product of Sf and Amax and the value of feature matrix B at that point is greater than the product of Rf and Bmax, then generate imaging matrix P and assign the value a at that point, and assign the value b to the other points, where a >> b.

[0121] In this embodiment, the number of rows and columns of feature matrix A and feature matrix B are equal. During the traversal, nested loops with the same row number i and column number j can be used to map the values ​​of i and j during the loop to the same position point in A and B. Then, a logical AND is used to perform conditional judgment to generate the imaging matrix P value.

[0122] In this embodiment, when the value of a (i.e., the value at which the bubble is considered to exist) is much greater than the value of b (i.e., the value at which the bubble is considered not to exist), the imaging color distinction between the regions with the value of a and the regions with the value of b in the image is higher, which can achieve a better bubble area boundary discrimination effect.

[0123] Step 7: Image formation is performed using the imaging matrix P.

[0124] like Figure 6 , Figure 7 As shown, a traditional imaging image and an imaging effect image of the present invention are obtained. The dark area in the imaging effect image of the present invention is the area where the bubble is confirmed. This imaging image combines the imaging advantages of transmission wave and bottom wave and obtains a bubble image with a clear area boundary through the bubble confirmation coefficients Sf and Rf. The detection effect of gas generation in lithium battery is better than that of traditional imaging images.

[0125] This invention employs an ultrasonic detection method based on the ultrasonic penetration-bottom wave height method, combining ultrasonic penetration and bottom wave height methods. This method enables faster and more accurate detection of micro-gas production inside lithium batteries, improving the detection accuracy. It utilizes an optimal coefficient selection method based on a BP neural network, matching the optimal internal gas production confirmation coefficient corresponding to the signal amplitude information obtained from the scan using the BP neural network. The optimal coefficient is then used to create an ultrasonic imaging map of the internal gas production lithium battery. Furthermore, a bubble imaging area boundary correction algorithm based on ultrasonic detection of internal gas production in lithium batteries is used to obtain a clear boundary for the internal gas production bubble area, achieving a more comprehensive and accurate imaging and judgment of the internal micro-gas production state of the lithium battery.

[0126] Example 2

[0127] This embodiment provides a lithium battery internal micro-gas generation detection system based on neural network matching bubble confirmation coefficient, which implements the lithium battery internal micro-gas generation detection method based on neural network matching bubble confirmation coefficient in Embodiment 1 above. The system includes: an ultrasonic scanning control module, a maximum attenuation calculation module, a bubble confirmation coefficient acquisition module, a training set construction module, a network training module, an optimal bubble confirmation coefficient output module, and an imaging module.

[0128] In this embodiment, the ultrasonic scanning control module is used to scan the lithium battery surface based on the ultrasonic penetration method and the bottom wave height method, respectively.

[0129] In this embodiment, the maximum attenuation calculation module is used to calculate the maximum attenuation C. A Maximum attenuation C B ;

[0130] In this embodiment, the maximum attenuation C A The calculations include:

[0131] Based on the peak-to-peak value of the ultrasonic signal obtained by the ultrasonic transmission method, a feature matrix A is generated to show the two-dimensional peak-to-peak position of the transmitted wave. The difference between the maximum and minimum values ​​within feature matrix A is calculated to obtain the maximum attenuation C of the ultrasonic peak-to-peak value obtained by the transmission method. A ;

[0132] In this embodiment, the maximum attenuation C B The calculations include:

[0133] The peak-to-peak value of the ultrasonic signal is obtained based on the bottom wave height method, and a feature matrix B is generated to determine the two-dimensional peak-to-peak position of the bottom wave. The difference between the maximum and minimum values ​​in the feature matrix B is calculated to obtain the maximum attenuation C of the ultrasonic peak-to-peak value obtained by the bottom wave height method. B ;

[0134] In this embodiment, the bubble confirmation coefficient acquisition module is used to obtain the maximum attenuation C. ASimulation tests were conducted to obtain the corresponding bubble confirmation coefficient Sf, based on the maximum attenuation C. B Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Rf;

[0135] In this embodiment, the training set construction module is used to construct the training set based on the maximum decay C. A Maximum attenuation C B Construct a training set using bubble confirmation coefficients Sf and Rf;

[0136] In this embodiment, the network training module is used to input the training set into the neural network model for training, with a maximum decay C. A Maximum attenuation C B As input variables, bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are used as the true values ​​of the corresponding output values. The training is iterated until the training stopping condition is met, and the trained neural network model is obtained.

[0137] In this embodiment, the optimal bubble confirmation coefficient output module is used to obtain the maximum attenuation C obtained by scanning imaging of the lithium battery under test based on the ultrasonic penetration-bottom wave height method. A Maximum attenuation C B And based on the trained neural network model, the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained;

[0138] In this embodiment, the imaging module is used to perform ultrasonic imaging of internally generated gas lithium batteries based on the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf.

[0139] In this embodiment, a gas-generating area boundary correction module is also provided. This module is used to correct the gas-generating area boundary of the ultrasonic imaging of the gas-generating lithium battery, specifically including:

[0140] The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model.

[0141] Traverse the feature matrix A to obtain the maximum value Amax of feature matrix A, and traverse the feature matrix B to obtain the maximum value Bmax of feature matrix B.

[0142] If the value of a point on the feature matrix A is less than the product of the bubble confirmation coefficient Sf and the maximum value Amax, and the value of the feature matrix B at that point is greater than the product of the bubble confirmation coefficient Rf and the maximum value Bmax, then an imaging matrix P is generated and assigned the value a at that point, and the value b at the other points, where a >> b.

[0143] Imaging of internally generated gas lithium batteries using ultrasound based on imaging matrix P.

[0144] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A method for detecting micro-gas generation inside a lithium battery based on neural network matching bubble confirmation coefficient, characterized in that, Includes the following steps: Based on the ultrasonic penetration method, the peak-to-peak value of the ultrasonic signal is obtained and a feature matrix A of the two-dimensional peak-to-peak position of the transmitted wave is generated. The difference between the maximum value and the minimum value in the characteristic matrix A is calculated to obtain the maximum attenuation C of the peak-to-peak value of the penetrating method ultrasonic wave A ; Based on the bottom wave height method, the peak-to-peak value of the ultrasonic signal of the bottom wave height method is obtained and a feature matrix B of the two-dimensional peak-to-peak value position of the bottom wave is generated. The difference between the maximum value and the minimum value in the characteristic matrix B is calculated to obtain the maximum attenuation C of the bottom wave height method ultrasonic peak-to-peak value B ; Based on the maximum attenuation C A Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Sf, based on the maximum attenuation C. B To conduct a simulation test and obtain the corresponding bubble confirmation coefficient Rf, the specific steps include: A model of a lithium battery is created, which includes the basic components of a lithium battery and hydrogen bubbles with adjustable volume. The projected area S of hydrogen bubbles on the two-dimensional plane of the lithium battery in the calculation model; Based on the ultrasonic penetration-bottom wave height method, scanning imaging of lithium batteries was performed, the imaging bubble area S1 was calculated, and the maximum attenuation C under the current hydrogen bubble volume was calculated. A Maximum attenuation C B And set the initial values ​​of bubble confirmation coefficient Sf and bubble confirmation coefficient Rf; Determine whether the projected area S is equal to the current imaging bubble area S1. If they are equal, use the current bubble confirmation coefficient Sf and bubble confirmation coefficient Rf as the corresponding bubble confirmation coefficient Sf and corresponding bubble confirmation coefficient Rf. If they are not equal, adjust the values ​​of bubble confirmation coefficient Sf and bubble confirmation coefficient Rf according to the set step size until the projected area S is equal to the current imaging bubble area S1. Based on the maximum attenuation C A Maximum attenuation C B A training set is constructed using the bubble confirmation coefficients Sf and Rf. This training set is then input into the neural network model for training. The maximum decay C is then calculated. A Maximum attenuation C B As input variables, bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are used as the true values ​​of the corresponding output values. The training is iterated until the training stopping condition is met, and the trained neural network model is obtained. The training set is input into the neural network model for training, specifically including: The neural network model employs a backpropagation (BP) neural network with a supervised learning approach. When a pair of learning patterns is provided to the BP neural network, the activation values ​​of neurons in the input layer are propagated through each hidden layer to the output layer. The output layer neurons output the network response corresponding to the input pattern. The error between the output value and the actual value of the BP neural network is calculated using a loss function. Following the principle of reducing error, the connection weights are adjusted layer by layer from the output layer through the hidden layers back to the input layer. The gradient between the error and each weight value is calculated through backpropagation, and the weight values ​​are updated accordingly. Finally, the maximum decay C is fitted. A Maximum attenuation C B The nonlinear relationship between the bubble confirmation coefficient Sf and the bubble confirmation coefficient Rf; The maximum attenuation C of the lithium battery under test was obtained by scanning imaging based on the ultrasonic penetration-bottom wave height method. A Maximum attenuation C B The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model. Ultrasonic imaging of internally generated lithium batteries based on optimal bubble confirmation coefficients Sf and Rf. It also includes a step for correcting the boundary of the gas generation area within the lithium battery, specifically including: The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model. Traverse the feature matrix A to obtain the maximum value Amax of feature matrix A, and traverse the feature matrix B to obtain the maximum value Bmax of feature matrix B. If the value of a point on the feature matrix A is less than the product of the bubble confirmation coefficient Sf and the maximum value Amax, and the value of the feature matrix B at that point is greater than the product of the bubble confirmation coefficient Rf and the maximum value Bmax, then an imaging matrix P is generated and assigned the value a at that point, and the value b at the other points, where a >> b. Imaging of internally generated gas lithium batteries using ultrasound based on imaging matrix P.

2. The method for detecting micro-gas generation inside a lithium battery based on neural network matching bubble confirmation coefficients according to claim 1, characterized in that, The calculation of the imaging bubble area S1 specifically includes: The lithium battery is scanned and imaged using the ultrasonic penetration-bottom wave height method to obtain the imaging matrix P. The number of pixels containing bubbles in the imaging matrix P is calculated, and the area of ​​the imaging bubble is obtained by multiplying the number of pixels by the size of the corresponding pixels.

3. The method for detecting micro-gas generation inside a lithium battery based on neural network matching bubble confirmation coefficients according to claim 1, characterized in that, Based on the maximum attenuation C A Maximum attenuation C B The training set is constructed using bubble confirmation coefficients Sf and Rf, specifically including: By changing the size of the hydrogen bubbles in the model, multiple sets of maximum attenuation C were obtained. A The corresponding bubble confirmation coefficient Sf, and the maximum attenuation C B The corresponding bubble confirmation coefficient Rf is used to construct the training set.

4. The method for detecting micro-gas generation inside a lithium battery based on neural network matching bubble confirmation coefficients according to claim 1, characterized in that, Training stop conditions include the error signal meeting the set range or the training count reaching the set number of times.

5. The method for detecting micro-gas generation inside a lithium battery based on neural network matching bubble confirmation coefficients according to claim 1, characterized in that, The steps for generating the imaging matrix P include: The feature matrix A and feature matrix B have the same number of rows and columns. During the traversal, nested loops are performed with the same row number i and column number j. The values ​​of row number i and column number j during the loop correspond to the same position point in feature matrix A and feature matrix B. Logical AND is used to perform conditional judgment to generate the value of imaging matrix P.

6. A micro-gas generation detection system inside a lithium battery based on neural network matching bubble confirmation coefficient, characterized in that, The method for detecting micro-gas production inside a lithium battery based on matching bubble confirmation coefficient using a neural network, as described in any one of claims 1-5, comprises: an ultrasonic scanning control module, a maximum attenuation calculation module, a bubble confirmation coefficient acquisition module, a training set construction module, a network training module, an optimal bubble confirmation coefficient output module, and an imaging module. The ultrasonic scanning control module is used to scan the lithium battery surface based on the ultrasonic penetration method and the bottom wave height method, respectively. The maximum attenuation calculation module is used to calculate the maximum attenuation C respectively. A Maximum attenuation C B ; Maximum attenuation C A The calculations include: Based on the peak-to-peak value of the ultrasonic signal obtained by the ultrasonic transmission method, a feature matrix A is generated to show the two-dimensional peak-to-peak position of the transmitted wave. The difference between the maximum and minimum values ​​within feature matrix A is calculated to obtain the maximum attenuation C of the ultrasonic peak-to-peak value obtained by the transmission method. A ; Maximum attenuation C B The calculations include: The peak-to-peak value of the ultrasonic signal is obtained based on the bottom wave height method, and a feature matrix B is generated to determine the two-dimensional peak-to-peak position of the bottom wave. The difference between the maximum and minimum values ​​in the feature matrix B is calculated to obtain the maximum attenuation C of the ultrasonic peak-to-peak value obtained by the bottom wave height method. B ; The bubble confirmation coefficient acquisition module is used to obtain the maximum attenuation C. A Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Sf, based on the maximum attenuation C. B Simulation tests were conducted to obtain the corresponding bubble confirmation coefficient Rf; The training set construction module is used to construct the training set based on the maximum decay C. A Maximum attenuation C B Construct a training set using bubble confirmation coefficients Sf and Rf; The network training module is used to input the training set into the neural network model for training, with a maximum decay C. A Maximum attenuation C B As input variables, bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are used as the true values ​​of the corresponding output values. The training is iterated until the training stopping condition is met, and the trained neural network model is obtained. The optimal bubble confirmation coefficient output module is used to obtain the maximum attenuation C obtained by scanning imaging of the lithium battery under test based on the ultrasonic penetration-bottom wave height method. A Maximum attenuation C B And based on the trained neural network model, the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained; The imaging module is used to perform ultrasonic imaging of internally generated lithium batteries based on the optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf. A gas-producing area boundary correction module is also provided. This module is used to correct the gas-producing area boundary of the ultrasonic imaging of the gas-producing lithium battery, specifically including: The optimal bubble confirmation coefficient Sf and bubble confirmation coefficient Rf are obtained based on the trained neural network model. Traverse the feature matrix A to obtain the maximum value Amax of feature matrix A, and traverse the feature matrix B to obtain the maximum value Bmax of feature matrix B. If the value of a point on the feature matrix A is less than the product of the bubble confirmation coefficient Sf and the maximum value Amax, and the value of the feature matrix B at that point is greater than the product of the bubble confirmation coefficient Rf and the maximum value Bmax, then an imaging matrix P is generated and assigned the value a at that point, and the value b at the other points, where a >> b. Imaging of internally generated gas lithium batteries using ultrasound based on imaging matrix P.

Images