A method for constructing a lithium-ion battery lithium plating detection model

Abstract

This invention discloses a method for constructing a lithium-ion battery lithium plating detection model, relating to the field of battery testing technology. The construction steps are as follows: S1: Select multiple battery packs of the same model and specifications, and divide them into multiple experimental groups proportionally according to different experimental conditions; S2: Place all battery packs within each experimental group under their corresponding experimental conditions, and perform periodic charge-discharge cycles of constant current and constant voltage charging and discharging on all battery packs within each experimental group, recording the charge-discharge data for each constant current and constant voltage charging and discharging cycle. This invention predicts lithium plating in lithium-ion batteries by extracting IC curve features and combining them with a support vector machine binary classification algorithm. The prediction accuracy is 91.4%, which is high, and it has a good predictive effect on whether lithium plating will occur in subsequent lithium batteries during cycling.

Description

Technical Field

[0001] This invention relates to the field of battery testing technology, specifically to a method for constructing a lithium plating detection model for lithium-ion batteries. Background Technology

[0002] Currently, commercial lithium-ion batteries are widely used in 3C digital products, electric vehicles, energy storage devices, and other fields, leading to a surge in lithium-ion battery production. With the widespread use of lithium-ion batteries, their safety and electrochemical performance are particularly important, including aspects such as battery life, rate performance, safety performance, and lithium plating characteristics. Lithium plating, in particular, has been a key concern for both manufacturers and users, as it significantly impacts battery cycle performance and poses a considerable safety hazard.

[0003] Currently, there are two main methods for detecting lithium plating in batteries: destructive lithium plating detection and non-destructive lithium plating detection. As the name suggests, destructive lithium plating detection involves disassembling the battery and then combining electrode surface phenomena with other physical characteristics to determine if lithium plating has occurred. Non-destructive lithium plating detection, on the other hand, extracts charge and discharge data from the lithium battery and extracts parameters characterizing the battery's lithium plating properties. These parameters are then post-processed to evaluate the battery's lithium plating characteristics. In practical applications, non-destructive lithium plating detection is preferred; however, the accuracy of currently available non-destructive lithium plating detection methods, especially in the early stages of battery cycling, is not high. Summary of the Invention

[0004] The purpose of this invention is to provide a method for constructing a lithium-ion battery lithium plating detection model to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a lithium-ion battery lithium plating detection model, the construction steps of which are as follows:

[0006] S1: Select multiple battery packs of the same model and specifications, and divide the multiple battery packs into multiple experimental groups in equal proportions according to different experimental conditions;

[0007] S2: Place all battery packs in each experimental group under their corresponding experimental conditions, and perform periodic charge-discharge cycles of constant current and constant voltage charging and constant current discharging on all battery packs in each experimental group, and record the charge-discharge data in each constant current and constant voltage charging and constant current discharging cycle.

[0008] S3: After cycling through n constant current and constant voltage charging and constant current discharging cycles, select at least one fully charged battery pack in each experimental group for physical disassembly and observe whether lithium is deposited on the surface of the negative electrode of its cell. Record the results as "lithium deposited" and "no lithium deposited", where n is a positive integer greater than 1.

[0009] S4: Process the full-charge data for the nth and mth cycle numbers corresponding to each disassembled battery pack, extract the voltage and capacity data during the constant current phase of charging, and plot the corresponding QV curves, which are QV... n Curves and QV m The curve is obtained by differentiating the QV curve obtained above, and then the IC curve is obtained by using QV. n Curves and QV m The curve yields the corresponding IC. n Curves and IC m A curve, where n > m, and m is a positive integer;

[0010] S5: Select each battery pack IC through feature analysis. m Curves and IC n Multiple eigenvalues ​​between curves form eigenvalue groups. The eigenvalue groups obtained from z battery packs are summarized to form an eigenvalue set, where z is the number of disassembled battery packs and is a positive integer.

[0011] S6: Use support vector machines to model and analyze the above feature set, and after modeling and training, output a lithium-ion battery lithium plating detection model with high prediction accuracy.

[0012] In this preferred embodiment, in step S1, different experimental conditions are set as different experimental temperatures and different charging rates, and 5 to 15 parallel samples are set in the experimental group corresponding to each experimental condition.

[0013] In this preferred embodiment, the experimental group in step S1 comprises 20 groups, wherein the experimental temperature corresponding to each experimental group is any one of 10℃, 25℃, 35℃ and 45℃, and the charging rate corresponding to each experimental group is any one of 0.33C, 0.50C, 1.00C, 1.50C and 2.00C.

[0014] In this preferred embodiment of the technical solution, the charging cutoff rate of each battery pack in step S2 is 0.05C, and the discharge rate is 1C for constant current discharge. The frequency of recording the charging and discharging data is 0.1s.

[0015] In this preferred embodiment, n is a multiple of 100 and is less than the number of life-end cycles of the cells in the battery pack, and m is specifically in the range of 2 to 20.

[0016] In this preferred embodiment of the technical solution, the method for obtaining the IC curve from the QV curve in step S4 is as follows: linear interpolation is performed on the QV curve using cubic sampling interpolation fitting, and then the IC curve is calculated. Wherein, the range of the linear interpolation voltage must remain consistent when processing all data and can cover all characteristic values ​​of the IC curve.

[0017] In this preferred embodiment, the number of linear interpolation points is 500 to 1000 within the range of linear interpolation voltage.

[0018] In this preferred embodiment, the feature value in step S5 includes IC. m Curves and IC n Mean, minimum, maximum, median, variance, and IC of the differences between curves n The peak areas and heights of peaks 1, 2, and 3 in the curve are related to IC. m The ratio of peak area to height for peaks 1, 2, and 3 in the curve, where the starting voltage of peak 1 is the initial value of the linear interpolation voltage range, and the ending voltage is the voltage corresponding to the lowest point between peaks 1 and 2; the starting voltage of peak 2 is the voltage corresponding to the lowest point between peaks 1 and 2, and the ending voltage is the voltage corresponding to the lowest point between peaks 2 and 3; the starting voltage of peak 3 is the voltage corresponding to the lowest point between peaks 2 and 3, and the ending voltage is the ending value of the linear interpolation voltage range.

[0019] In this preferred technical solution, the specific steps for obtaining the lithium-ion battery lithium plating detection model using a support vector machine in step S6 are as follows:

[0020] S61: First, the eigenvalues ​​in the z eigenvalue sets are normalized using the eigenvalue normalization method;

[0021] S62: Then, the dataset is divided into 5 parts using the 5-fold cross-validation method, with 4 parts used as the training set and the other 1 part as the test set, and cross-validation is performed sequentially.

[0022] S63: Modeling and training are performed by optimizing the kernel function, penalty coefficient C, degree, and gamma selection within the support vector machine until a lithium-ion battery lithium plating detection model with high prediction accuracy is output.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] This lithium-ion battery lithium plating detection model construction method predicts lithium plating in lithium-ion batteries by extracting IC curve features and combining them with a support vector machine binary classification algorithm. The prediction accuracy is 91.4%, which is high. It has a good predictive effect on whether lithium plating will occur in subsequent lithium batteries during cycling.

[0025] This method can not only predict whether lithium plating will occur in a battery, but also apply the final model to commercial lithium-ion battery factories and end-use equipment for testing. Lithium plating in a battery often indicates rapid degradation of battery performance, thus alerting manufacturers and end-use equipment. Manufacturers can improve subsequent battery testing measures to improve battery cycle performance, or end-use equipment can consider repairing and replacing the battery. Attached Figure Description

[0026] Figure 1 This is a QV curve diagram from an embodiment of the present invention (experimental conditions: 25℃, 0.50C);

[0027] Figure 2 The IC curve is shown in the embodiment of the present invention (experimental conditions: 25℃, 0.50C);

[0028] Figure 3 This is a schematic diagram showing the segmentation of peaks 1, 2, and 3 in this invention;

[0029] Figure 4 This is a model confusion matrix diagram according to an embodiment of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] It should be noted that in the description of this invention, the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0032] Furthermore, it should be understood that, for ease of description, the dimensions of the various components shown in the accompanying drawings are not drawn to actual scale; for example, the thickness or width of some layers may be exaggerated relative to other layers.

[0033] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined or described in one figure, it will not need to be discussed or described in detail in the description of the subsequent figures.

[0034] Before understanding this invention, it is important to understand that current commercially available methods for non-destructive lithium plating detection in lithium-ion batteries primarily involve extracting characteristic quantities from the battery cycling process and setting thresholds to determine whether lithium plating has occurred. For example, publications CN114200322A ("A Method for Detecting Lithium Plating in Lithium-ion Batteries") and CN112240984A ("A Method and Device for Detecting Lithium Plating in Lithium-ion Batteries") both rely on setting thresholds to determine lithium plating, but this method lacks accuracy. This invention, however, employs machine learning to extract multiple characteristic quantities and model them to detect lithium plating in lithium-ion batteries.

[0035] It's important to understand that the QV curve represents the relationship between the battery cell's capacity and voltage during charging. The IC curve (capacity increment curve), on the other hand, represents the derivative of the QV curve during charging, i.e., the dq / dv curve. Each peak in the IC curve represents a voltage plateau during the cell's charging process and also reflects the phase transition of the material. The IC curve contains characteristic peaks of the positive and negative electrodes, as well as other characteristic peaks characterizing various properties. These characteristic peaks may be independent or intermingled. Therefore, analyzing the IC curve and extracting the corresponding characteristic values ​​representing lithium plating can be used to model and determine whether lithium plating has occurred in the battery.

[0036] Support Vector Machines (SVMs) are a commonly used machine learning algorithm in this field. Essentially, they are a binary classification model. Their basic model is a linear classifier with the largest margin defined in the feature space. However, when linearity is not feasible, a kernel function is used to map to a higher-dimensional space, solving for the maximum margin hyperplane to achieve linear feasibility. Therefore, for batteries, cells can only be classified into two categories: lithium-plated and non-lithium-plated. The SVM model is a good classification method for this.

[0037] Therefore, the present invention provides a technical solution: a method for constructing a lithium-ion battery lithium plating detection model, the construction steps of which are as follows:

[0038] S1: Select multiple battery packs of the same model and specifications, and divide them into multiple experimental groups in equal proportions according to different experimental conditions. In this embodiment of the invention, the battery packs are LFP / graphite system soft-pack batteries with a rated capacity of 5.5Ah and a voltage system of 2.5-3.65V. A total of 100 such soft-pack batteries are selected for backup. The 100 soft-pack batteries are grouped according to different experimental temperatures and different charging rates. The specific grouping conditions are as follows: Temperature 10℃, charging rate 0.33C; Temperature 10℃, charging rate 0.50C; Temperature 10℃, charging rate 1.00C; Temperature 10℃, charging rate 1.50C; Temperature 10℃, charging rate 2.00C; Temperature 25℃, charging rate 0.33C; Temperature 25℃, charging rate 0.50C; Temperature 25℃, charging rate 1.00C; Temperature 25℃, charging rate 1.50C; Temperature 25℃, charging rate... The charging rates are as follows: 2.00C; 35°C, charging rate 0.33C; 35°C, charging rate 0.50C; 35°C, charging rate 1.00C; 35°C, charging rate 1.50C; 35°C, charging rate 2.00C; 14°C, charging rate 0.33C; 45°C, charging rate 0.50C; 45°C, charging rate 1.00C; 45°C, charging rate 1.50C; 45°C, charging rate 2.00C. A total of 20 groups are formed, therefore each group contains 5 pouch cells as parallel samples. In other embodiments of the invention, 5 to 15 parallel samples can be selected. Specifically, it can be any one of 6, 7, 8, 9, 10, 11, 12, 13, 14, or 15.

[0039] S2: Place all battery packs within each experimental group under their corresponding experimental conditions, and perform periodic charge-discharge cycles of constant current and constant voltage charging and discharging on all battery packs within each experimental group. Record the charge-discharge data for each constant current and constant voltage charging and discharging cycle. In a specific embodiment of the present invention, the charging cutoff rate of each battery pack is 0.05C, and it is discharged using a 1C discharge rate for constant current discharge. The timing frequency for recording charge-discharge data is 0.1s.

[0040] S3: After n constant current constant voltage charging and constant current discharging cycles, at least one fully charged battery pack in each experimental group is physically disassembled, and the presence or absence of lithium plating on the negative electrode surface of the battery cells is observed. The results are recorded as "lithium plating" and "no lithium plating," respectively, where n is a positive integer greater than 1. Specifically, n can be a multiple of 100 and less than the number of cycles required to terminate the battery cell's lifespan in the battery pack. In the embodiment of this invention, n is 200.

[0041] S4: Process the full-charge data for the nth and mth cycle numbers corresponding to each disassembled battery pack, extract the voltage and capacity data during the constant current phase of charging, and plot the corresponding QV curve (e.g., ...). Figure 1 As shown), they are QV respectively. n Curves and QV m The curve is obtained by differentiating the QV curve obtained above to obtain the IC curve (e.g., Figure 2 As shown), via QV n Curves and QV m The curve yields the corresponding IC. n Curves and IC m The curve is defined as follows: n > m, where m is a positive integer. It is important to note that m ranges from 2 to 20. The reason m is not set to 1 is to avoid data distortion due to instability during the first charge of a new battery pack. In this embodiment, m is 10. The method for obtaining the IC curve from the QV curve is as follows: linear interpolation is performed on the QV curve using cubic sampling interpolation fitting, and then the IC curve is calculated. The range of the linear interpolation voltage must remain consistent when processing all data and must cover all characteristic values ​​of the IC curve. Furthermore, within the range of the linear interpolation voltage, the number of linear interpolation points is between 500 and 1000. Specifically, it can be any one of 500, 550, 600, 650, 700, 750, 800, 850, 900, 950, and 1000, or any number of adjacent numbers mentioned above. In an embodiment of the present invention, the range of the linear interpolation voltage is selected to be 2.9V to 3.5V, and the number of linear interpolation points selected within this range is 1000.

[0042] S5: Select each battery pack IC through feature analysis. m Curves and IC n Multiple eigenvalues ​​between curves form eigenvalue groups. The z eigenvalue groups obtained from z battery packs are then summarized to form an eigenvalue set, where z is the number of disassembled battery packs and is a positive integer. In an embodiment of the invention, one battery pack is selected for disassembly in each experimental group, therefore z is 20. Furthermore, the aforementioned eigenvalues ​​include IC. m Curves and IC n Mean, minimum, maximum, median, variance, and IC of the differences between curves n Peak 1 in the curve and IC m The area ratio and height ratio of peak 1 in the curve, IC n Peak 2 in the curve and IC mThe area ratio and height ratio of peak 2 in the curve and IC n Peak 3 in the curve and IC m The area ratio and height ratio of peak 3 in the curve. A detailed diagram illustrating the segmentation of peaks 1, 2, and 3 is shown below. Figure 3 As shown. The starting voltage of peak 1 is the initial value of the linear interpolation voltage range (2.9V in this embodiment), and the ending voltage is the voltage corresponding to the lowest point between peak 1 and peak 2. The starting voltage of peak 2 is the voltage corresponding to the lowest point between peak 1 and peak 2, and the ending voltage is the voltage corresponding to the lowest point between peak 2 and peak 3; the starting voltage of peak 3 is the voltage corresponding to the lowest point between peak 2 and peak 3, and the ending voltage is the ending value of the linear interpolation voltage range (3.5V in this embodiment). S6: A support vector machine is used to model and analyze the above feature set, and after modeling and training, a lithium-ion battery lithium plating detection model with high prediction accuracy is output.

[0043] Specifically, the steps for obtaining the lithium plating detection model for lithium-ion batteries using support vector machines are as follows:

[0044] S61: First, the eigenvalues ​​in the z eigenvalue sets are normalized using the eigenvalue normalization method;

[0045] S62: Then, the dataset is divided into 5 parts using the 5-fold cross-validation method, with 4 parts used as the training set and the other 1 part as the test set, and cross-validation is performed sequentially.

[0046] S63: Modeling and training are performed by optimizing the kernel function, penalty coefficient C, degree, and gamma within the support vector machine (SVM) until a lithium-ion battery lithium plating detection model with high prediction accuracy is output. Specifically, in the embodiments of this invention, the kernel functions selected for the SVM are linear kernel, polynomial kernel (where the kernel function parameter degree is set to 2 and 3), and Gaussian kernel (RBF, where the kernel function parameter gamma is set to 0.1 to 10), and the regularization parameter is set in the range of 0.1 to 1000. Finally, modeling and training are performed. After training, the SVM algorithm uses the Gaussian kernel (RBF) with a regularization parameter of 100 and a kernel function parameter gamma of 0.5, achieving a prediction accuracy of 91.4%. The specific verification results are as follows... Figure 4 As shown.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a lithium precipitation detection model of a lithium electronic battery, characterized in that, The construction steps are as follows: S1: select a plurality of battery packs of the same model and specification, and divide the plurality of battery packs into a plurality of experimental groups in accordance with different experimental conditions; S2: place all battery packs in each experimental group in the corresponding experimental condition, and perform periodic charging and discharging cycles of constant current and constant voltage charging and constant current discharging on all battery packs in each experimental group, and record the charging and discharging data in each constant current and constant voltage charging and constant current discharging cycle; S3: after n constant current and constant voltage charging and constant current discharging cycles, select at least one fully charged battery pack in each experimental group for physical disassembly, and observe whether lithium is precipitated on the negative electrode surface of the battery cell, and record the results as "lithium precipitation" and "no lithium precipitation", respectively, wherein n is a positive integer greater than 1; S4: process the full charge data of the nth cycle and the mth cycle of each disassembled battery pack, extract the voltage and capacity data during the constant current stage of the charging process, draw the corresponding Q-V curve, which is Q-Vn curve and Q-Vm curve respectively, and obtain the IC curve by differentiating the above obtained Q-V curve, and obtain the corresponding ICn curve and ICm curve through Q-Vn curve and Q-Vm curve, wherein n>m and m is a positive integer; S5: through feature analysis, select a plurality of characteristic values between the ICm curve and the ICn curve of each battery pack to form a characteristic value group, wherein z characteristic value groups obtained from z battery packs are summarized to form a characteristic value collection, and z is the number of disassembled battery packs, which is a positive integer; Wherein, the plurality of characteristic values between the battery pack ICm curve and the ICn curve include: the average value, the minimum value, the maximum value, the median value, the variance of the difference value between the ICm curve and the ICn curve, the area ratio and the height ratio of peak 1 in the ICn curve and the corresponding peak 1 in the ICm curve, the area ratio and the height ratio of peak 2 in the ICn curve and the corresponding peak 2 in the ICm curve, and the area ratio and the height ratio of peak 3 in the ICn curve and the corresponding peak 3 in the ICm curve; S6: use support vector machine to model and analyze the above characteristic value collection, and output a lithium battery lithium precipitation detection model with high prediction accuracy after modeling and training.

2. The method of claim 1, wherein the method comprises: In step S1, the different experimental conditions are set as different experimental temperatures and different charge rates, and 5 to 15 parallel samples are set in each experimental group corresponding to each experimental condition.

3. The method of claim 2, wherein the method further comprises: In step S1, the experimental group contains 20, wherein the experimental temperature corresponding to each experimental group is any one of 10℃, 25℃, 35℃ and 45℃, and the charge rate corresponding to each experimental group is any one of 0.33C, 0.50C, 1.00C, 1.50C and 2.00C.

4. The method according to claim 3, wherein, In step S2, the charge cutoff rate of each battery pack is 0.05C, and the discharge rate is 1C during constant current discharge, and the sampling frequency of the charging and discharging data is 0.1s.

5. The method according to any one of claims 1 to 4, wherein n is a multiple of 100 and n is less than the end of life cycle number of the battery cell in the battery pack, and m is specifically in the range of 2 to 20.

6. The method according to any one of claims 1 to 4, wherein The method for obtaining the IC curve from the Q-V curve in step S4 is: linear interpolation is performed on the Q-V curve by using cubic sampling interpolation fitting, and then the IC curve is calculated, wherein the range of the linear interpolation voltage must be consistent and cover all characteristic values of the IC curve when processing all data.

7. The method of claim 6, wherein the method further comprises: In the range of the linear interpolation voltage, the number of linear interpolation points is 500 to 1000. 8.The method of claim 6, wherein, In step S5, the characteristic values include the average, minimum, maximum, median, variance of the difference between the ICm curve and the ICn curve, the peak area and height of peak 1, peak 2 and peak 3 in the ICn curve, and the ratio of the peak area and height of peak 1, peak 2 and peak 3 in the ICm curve, wherein the starting voltage of peak 1 is the initial value of the range of the linear interpolation voltage, and the terminal voltage is the voltage corresponding to the lowest point between peak 1 and peak 2; the starting voltage of peak 2 is the voltage corresponding to the lowest point between peak 1 and peak 2, and the terminal voltage is the voltage corresponding to the lowest point between peak 2 and peak 3; the starting voltage of peak 3 is the voltage corresponding to the lowest point between peak 2 and peak 3, and the terminal voltage is the terminal value of the range of the linear interpolation voltage.

9. The method of claim 8, wherein the method further comprises: The specific steps of obtaining the lithium battery lithium precipitation detection model by using the support vector machine in step S6 are as follows: S61: First, the characteristic values in the z characteristic value groups are normalized by using the characteristic value normalization method; S62: Then, the data set is divided into 5 parts by using the 5-fold cross-validation method, 4 parts are used as the training set, and the other 1 part is used as the test set, and cross-validation is performed in turn; S63: The kernel function, penalty coefficient C, degree and gamma inside the support vector machine are optimized to select the modeling training, until the lithium battery lithium precipitation detection model with high prediction accuracy is output.

Images