Battery lithium precipitation diagnosis method based on multi-dimensional feature clustering

Abstract

The application discloses a battery lithium precipitation diagnosis method based on multi-dimensional feature clustering and belongs to the field of battery fault detection; the method comprises the following steps: S1, collecting massive charge-discharge data of the battery under different cycle conditions and uploading the massive data platform; S2, extracting original data in the massive data platform and performing pretreatment; S3, extracting multi-dimensional core feature parameters of the battery; S4, establishing an unsupervised learning clustering analysis lithium precipitation diagnosis model of intelligent optimization parameters, and realizing graded diagnosis of the battery lithium precipitation; the application realizes self-adaptive graded diagnosis of lithium ion battery lithium precipitation faults based on battery operation data, in combination with a battery multi-dimensional feature extraction method and a clustering analysis algorithm. Compared with existing diagnosis technologies, the application can realize graded diagnosis of lithium ion battery lithium precipitation faults, and is more accurate than a single-dimensional feature lithium precipitation diagnosis method.

Description

Technical Field

[0001] This invention belongs to the field of battery fault detection, specifically relating to a battery lithium plating diagnosis method based on multidimensional feature clustering. Background Technology

[0002] To promote carbon peaking and carbon neutrality, the electrification transformation of the transportation sector is accelerating, leading to the booming development of the electric vehicle industry. Lithium-ion batteries, with their high energy density, long cycle life, and lack of memory effect, have become the most commonly used power batteries in the electric vehicle industry. However, with the development and application of all-weather batteries and high-rate fast charging technology, extreme conditions such as low temperatures, overcharging and over-discharging, and high-rate charging can cause various battery failures, including increased internal resistance, internal short circuits, aging, and thermal runaway. These failures can reduce battery life and even lead to fires and explosions, threatening people's lives and property. Among the many causes of these failures, lithium plating on the negative electrode is one of the most critical factors.

[0003] Lithium plating is the process by which lithium ions, in the form of metallic lithium, deposit on the surface of the negative electrode. Mild lithium plating is relatively common during battery charging, and the deposited lithium usually re-enters the negative electrode during subsequent resting, having little impact on the battery. However, severe lithium plating reduces the porosity of the negative electrode and decreases the number of cyclic lithium ions, leading to increased internal resistance and reduced usable capacity. In severe cases, the deposited lithium accumulates, forming lithium dendrites that puncture the separator, causing internal short circuits and directly affecting battery safety and performance, potentially leading to serious safety accidents. Therefore, fault diagnosis of lithium plating in lithium-ion batteries is particularly important.

[0004] Existing lithium plating diagnostic methods offer the advantages of being intuitive and accurate, enabling both online and offline lithium plating diagnosis. However, their proper application within a BMS system remains a significant challenge. Physical disassembly diagnostic methods require high-precision instruments, resulting in high costs and operational difficulties, and many are invasive procedures. While electrochemical diagnostic methods are non-destructive and highly sensitive, they are limited to analyzing external battery characteristics and cannot reveal the actual lithium plating situation inside the battery.

[0005] Currently, several solutions have emerged for the electrochemical diagnosis of lithium plating faults in lithium-ion batteries. For example, Chinese patent CN116794535A proposes a method and device for detecting lithium plating using graphite dot collection based on micro-switching voltage. By collecting micro-switching voltages at different discharge depths, a lithium plating peak appears when the micro-switching voltage curve fluctuates. Another example is Chinese patent CN117192405A, which proposes an online impedance measurement and lithium plating detection method for energy storage lithium-ion batteries. This method determines the lithium plating SOC of the lithium-ion battery by obtaining the impedance value change curve with SOC at different frequencies.

[0006] The above-mentioned methods can all achieve non-destructive lithium plating diagnosis, but they still have certain diagnostic errors. Fluctuations in the micro-switched voltage curve are due to lithium metal re-intercalation; when the amount of lithium plating is small, this fluctuation disappears, making accurate diagnosis of lithium plating impossible. Furthermore, while electrochemical impedance spectroscopy (EIS) measurements are highly accurate, changes in contact, pressure, and environmental conditions can all affect the measurement results. Therefore, current lithium plating diagnostic methods still have certain limitations. Summary of the Invention

[0007] To address the technical problems existing in the background art, this invention aims to provide a battery lithium plating diagnosis method based on multidimensional feature clustering. Based on large-scale battery operation data, this method combines multidimensional feature extraction and unsupervised learning clustering analysis to diagnose lithium plating in batteries. The proposed method extracts multidimensional high-order features from the original data to comprehensively diagnose lithium plating, eliminating diagnostic errors from single-dimensional features and thus improving the accuracy of lithium plating diagnosis. Simultaneously, it utilizes intelligent optimization algorithms to achieve adaptive hierarchical diagnosis of lithium plating faults.

[0008] To solve the technical problem, the technical solution of the present invention is as follows:

[0009] A battery lithium plating diagnostic method based on multidimensional feature clustering, the method comprising:

[0010] The battery operation data from the cloud big data platform is preprocessed, and the original dataset for the feature extraction model is obtained through the objective weighting method.

[0011] Battery electrochemical features were extracted from multiple dimensions, a mapping relationship between raw data and high-order features was established, and feature parameters strongly related to lithium plating were screened out and dimensionality was reduced using correlation analysis and principal component analysis to obtain multi-dimensional composite lithium plating features.

[0012] A clustering algorithm model for intelligent optimization parameters is established, taking battery lithium plating characteristics as input and whether lithium plating occurs and the degree of lithium plating as output. The battery capacity degradation rate, which can be used to characterize lithium plating, is marked in the extracted feature space to determine the accuracy of the lithium plating diagnosis results.

[0013] Furthermore, prior to step S1, the method further includes:

[0014] Under various operating conditions, the battery is subjected to charge and discharge tests, including conventional battery performance tests and lithium plating tests, and the resulting battery charge and discharge data is uploaded to a cloud-based big data platform.

[0015] Furthermore, the conventional battery performance tests include capacity testing, hybrid pulse power characteristic testing, federal urban driving condition testing, and electrochemical impedance spectroscopy testing; the battery lithium plating tests include: low-temperature charging lithium plating-induced experiment, high-rate charging-induced lithium plating experiment, and overcharging-induced lithium plating experiment.

[0016] Furthermore, the battery operating data includes voltage, current, and impedance data.

[0017] Furthermore, in step S1:

[0018] The data preprocessing method includes data cleaning and data filling;

[0019] The objective weighting method adopts one of the following: the coefficient of variation method, the entropy weight method, or the CRITIC weight method.

[0020] Furthermore, the battery electrochemical characteristics in step S2 include: battery equivalent circuit model parameters, voltage plateau characteristics, capacity increment curve parameters, electrochemical impedance spectroscopy parameters, and coulombic efficiency.

[0021] Furthermore, step S2 specifically includes:

[0022] S201: Extracting multi-dimensional electrochemical core parameters of the battery: Comparing the performance changes of the battery before and after lithium plating, and extracting key characteristic parameters;

[0023] S202: Conduct correlation analysis: Analyze battery characteristics and capacity degradation rate using multiple correlation analysis methods, and comprehensively select the lithium plating-strongly correlated characteristics;

[0024] S203: Perform principal component analysis: Use principal component analysis to reduce the dimensionality of the data, simplify the data analysis process and reduce computational complexity.

[0025] Furthermore, in step S3, the intelligent optimization algorithm employs one of the following: particle swarm optimization, simulated annealing, and genetic algorithm.

[0026] Furthermore, in step S3, the clustering method employs one of K-means clustering, DBSCAN clustering, AffinityPropagation clustering, and Agglomerative Clustering.

[0027] Furthermore, in step S3, establishing the lithium plating diagnostic model includes the following steps:

[0028] Set the fitness function and use intelligent optimization algorithms to find the parameters required in the clustering algorithm;

[0029] An optimized clustering algorithm was used to diagnose lithium plating in batteries.

[0030] Capacity degradation rate, which can be used to characterize the degree of lithium plating, is labeled into the feature space to determine the accuracy of clustered lithium plating diagnostic results.

[0031] Compared with the prior art, the advantages of the present invention are as follows:

[0032] This invention effectively utilizes measurement signals with different dimensions from multidimensional electrical sensors for comprehensive diagnosis, overcoming common error problems in single-dimensional diagnosis. Employing an unsupervised learning clustering algorithm, it can diagnose lithium plating based on actual data without requiring a training dataset. By setting fitness functions in different intelligent optimization algorithms and obtaining the optimal solution for clustering algorithm parameters, adaptive hierarchical diagnosis of lithium-ion battery lithium plating faults can be achieved. Distinguishing between severely and mildly lithium-plated batteries provides guidance for subsequent battery use, preventing further lithium plating. This invention has high application value for battery health monitoring and fault diagnosis. Attached Figure Description

[0033] Figure 1 A flowchart of the overall method of the present invention;

[0034] Figure 2 The lithium plating feature extraction diagram of the present invention;

[0035] Figure 3 The lithium plating diagnostic results diagram of the present invention;

[0036] Figure 4 Scanning electron microscope images of the present invention. Detailed Implementation

[0037] The specific implementation of the present invention is described below with reference to embodiments:

[0038] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0039] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0040] Example 1:

[0041] This implementation case uses lithium-ion battery cells with the parameters described in Table 1.

[0042] Table 1 Experimental lithium-ion battery parameters

[0043]

[0044] like Figure 1 As shown, a battery lithium plating diagnostic method based on multidimensional feature clustering includes the following steps:

[0045] S1: Collect massive amounts of battery charge and discharge data under different operating conditions, including conventional battery performance tests and lithium plating tests, and upload the data to the cloud big data platform;

[0046] Routine battery performance tests include capacity testing, hybrid pulse power characteristic testing, and electrochemical impedance spectroscopy testing.

[0047] The lithium plating test is a low-temperature charging-induced test, in which the battery is charged and discharged three times at different rate gradients of 0℃, -5℃ and -10℃ to induce lithium plating to different degrees.

[0048] The massive amount of operational data includes terminal voltage, load current, and impedance information.

[0049] S2: As Figure 2 As shown, the battery operating data is preprocessed.

[0050] Data preprocessing methods include data cleaning and data filling.

[0051] Data cleaning refers to deleting data with low signal-to-noise ratio and low utilization value from massive amounts of operational data, such as data on charging and discharging conditions with minute fluctuations in current.

[0052] Data filling refers to filling in missing values ​​in massive amounts of operational data, using interpolation methods to fill in missing sensor data.

[0053] The objective weighting method uses the coefficient of variation (COP) method. The negative COP is set as the inconsistency of battery characteristic parameters before lithium plating, and the positive COP is set as the rate of change of battery characteristic parameters before and after lithium plating. The score for each sampling point is calculated using the COP method, and the data with the highest score is selected as the original dataset for the feature extraction model. Specifically:

[0054]

[0055] Among them, Score i For the score of the i-th sampling point, ω i Let r be the weight of the j-th indicator. ij The element in the i-th row and j-th column of the standardized data matrix.

[0056] S3: Specifically includes the following sub-steps:

[0057] S301: Extracting multi-dimensional electrochemical core parameters of the battery, such as Figure 2 As shown

[0058] (1) Extracting Battery Equivalent Circuit Model Parameters: Few available features exist both inside and outside the battery, and extraction is difficult. It is necessary to establish an equivalent circuit model and use the model parameters as features to describe the internal electrochemical processes of the battery. Considering the requirements of the onboard processor's accuracy and computational complexity, a first-order RC equivalent circuit is adopted, and its state-space equation is:

[0059]

[0060] U T =U OCV,k (SOC)-I k *R0-U 1,k

[0061] In the formula, z represents SOC, U ocv U is the open-circuit voltage. T U is the terminal voltage, U1 is the polarization voltage, I is the operating current, and C is the battery rated capacity; R0 is the ohmic internal resistance, R1 is the polarization internal resistance, and C1 is the polarization capacitance. The parameters to be identified in the model include R0, R1, and C1.

[0062] (2) Extracting equivalent circuit parameters of electrochemical impedance spectroscopy: The battery impedance spectrum information is measured by using an electrochemical workstation, and the complex impedance of each single point in a specific frequency range is analyzed. This can distinguish between charge transfer, diffusion and SEI interface processes, and realize the function of obtaining internal information from externally measurable data.

[0063] (3) Extracting parameters from the capacity increment curve: The capacity increment curve can transform the battery's first-order phase transition voltage plateau into a peak. Parameters such as the peak value, position, and area contained within the peak can be used to analyze the battery capacity degradation mode and mechanism. ΔQ / ΔV is used instead of dQ / dV for calculation:

[0064]

[0065] In the formula, dQ represents the capacity differential, dV represents the voltage differential, and Q i+1 Let Q be the capacity value at position i+1. i Let V be the capacity value at position i. i+1 V is the voltage value at position i+1. i Let be the voltage value at position i.

[0066] S302: Correlation Analysis: Capacity degradation rate directly represents the degree of lithium plating. Therefore, a correlation analysis between the parameters and the capacity degradation rate is required. Pearson and Spearman correlation coefficients are used for calculation. Parameters with correlation coefficients greater than 0.7 are extracted as lithium plating characteristics. The specific calculation formula is as follows:

[0067]

[0068]

[0069] In the formula: X and Y represent the characteristic and capacity decay sample populations, respectively, x i y i Each represents a corresponding sample.

[0070] S303: Perform principal component analysis: Further perform principal component analysis on the feature parameter matrix to reduce the dimension of the feature parameter matrix and reduce computational complexity while retaining most of the data information.

[0071] S4: The specific steps for establishing a lithium plating diagnostic model are as follows:

[0072] S401: Particle swarm optimization (PSO) is selected as the intelligent optimization algorithm. The fitness function is set to whether the battery classification is correct. Specifically, the capacity degradation rate of severely lithium-plated clusters exceeds the set severe lithium plating threshold; the capacity degradation rate of mildly lithium-plated clusters is between the mild and severe lithium plating thresholds; and the capacity degradation rate of non-lithium-plated clusters is below the mild lithium plating threshold. A correct classification is recorded as 1, and an incorrect classification as 0. The cumulative sum is the fitness value. In this example, the mild and severe lithium plating thresholds are set to 2% and 7.5%, respectively.

[0073] S402: Use DBSCAN clustering (density-based clustering analysis algorithm) to diagnose lithium plating in batteries.

[0074] S403: Capacity degradation rates, which can be used to characterize the degree of lithium plating, are labeled into the feature space to determine the accuracy of clustered lithium plating diagnostic results. Diagnostic results are as follows: Figure 3 As shown, the studied lithium-ion batteries were divided into three categories in the feature space. The first category showed high clustering, while the second and third categories exhibited significant deviations. To further confirm the accuracy of the graded lithium plating results, the capacity degradation rate was used for verification. The results showed that the first category, with a capacity degradation rate below 2%, was classified as non-lithium-plated batteries; the second category, with a capacity degradation rate above 7.5%, was classified as severely lithium-plated batteries; and the third category, with a capacity degradation rate between 2% and 7.5%, was classified as mildly lithium-plated batteries. The grading diagnosis results were correct and consistent with experimental expectations.

[0075] The battery was further disassembled, and physical testing methods, including scanning electron microscopy (SEM), were used to observe the morphology and coverage area of ​​lithium plating, visually verifying the accuracy of the lithium plating diagnostic model proposed in this invention. Figure 4As can be seen, graphite particles on the negative electrode can be directly observed in normal battery images. No lithium metal dendrites are present. In mildly lithium-ion batteries, lithium metal partially covers the graphite but is still observable. In severely lithium-ion batteries, lithium metal completely covers the negative electrode surface, exhibiting a large, moss-like morphology. Therefore, the accuracy of the model is verified, and graded diagnosis of lithium-ion battery lithium plating is achieved.

[0076] In summary, this invention proposes a battery lithium plating diagnosis method based on multidimensional feature clustering. Utilizing large amounts of battery operational data, and based on multidimensional feature extraction and unsupervised machine learning algorithms, an adaptive grading diagnosis model for lithium-ion batteries is established. This method can detect whether lithium plating has occurred and the degree of plating. It overcomes the diagnostic error problem of single-dimensional features, improving diagnostic accuracy. Furthermore, the clustering algorithm is an unsupervised machine learning algorithm, requiring no training set, making it more convenient for practical applications.

[0077] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0078] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A battery lithium plating diagnostic method based on multidimensional feature clustering, characterized in that, The method includes: Step S1: Preprocess the battery operation data from the cloud big data platform and obtain the original dataset for the feature extraction model using the objective weighting method; Step S2: Extract battery electrochemical features from multiple dimensions, establish the mapping relationship between the original data and high-order features, use correlation analysis and principal component analysis to screen out feature parameters that are strongly related to lithium plating and perform dimensionality reduction to obtain multi-dimensional composite lithium plating features; Step S3: Establish a clustering algorithm model for intelligent optimization parameters, taking battery lithium plating characteristics as input, and whether lithium plating has occurred and the degree of lithium plating as output. Mark the battery capacity degradation rate that can be used for lithium plating characterization in the extracted feature space to determine the accuracy of the lithium plating diagnosis results. The battery electrochemical characteristics in step S2 include: battery equivalent circuit model parameters, voltage plateau characteristics, capacity increment curve parameters, electrochemical impedance spectroscopy parameters, and coulombic efficiency. Step S2 specifically includes: S201: Extracting multi-dimensional electrochemical core parameters of the battery: Comparing the performance changes of the battery before and after lithium plating, and extracting key characteristic parameters; S202: Conduct correlation analysis: Analyze battery characteristics and capacity degradation rate using multiple correlation analysis methods, and comprehensively select the lithium plating-strongly correlated characteristics; S203: Perform principal component analysis: Use principal component analysis to reduce the dimensionality of the data, simplify the data analysis process and reduce computational complexity.

2. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 1, characterized in that, Prior to step S1, the method further includes: Under various operating conditions, the battery is subjected to charge and discharge tests, including conventional battery performance tests and lithium plating tests, and the resulting battery charge and discharge data is uploaded to a cloud-based big data platform.

3. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 2, characterized in that, The battery's routine performance tests include capacity testing, hybrid pulse power characteristic testing, federal urban driving condition testing, and electrochemical impedance spectroscopy testing; the battery lithium plating tests include: low-temperature charging lithium plating-induced experiment, high-rate charging-induced lithium plating experiment, and overcharging-induced lithium plating experiment.

4. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 1, characterized in that, The battery operating electrical data includes: voltage, current, and impedance data.

5. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 1, characterized in that, In step S1: The data preprocessing method includes data cleaning and data filling; The objective weighting method adopts one of the following: the coefficient of variation method, the entropy weight method, or the CRITIC weight method.

6. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 1, characterized in that, In step S3, the intelligent optimization employs one of the following algorithms: particle swarm optimization, simulated annealing, and genetic algorithm.

7. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 1, characterized in that, In step S3, the clustering algorithm employs one of K-means clustering, DBSCAN clustering, Affinity Propagation clustering, and Agglomerative Clustering.

8. The battery lithium plating diagnostic method based on multidimensional feature clustering according to claim 1, characterized in that, In step S3, establishing the lithium plating diagnostic model includes the following steps: Set the fitness function and use intelligent optimization algorithms to find the parameters required in the clustering algorithm; An optimized clustering algorithm was used to diagnose lithium plating in batteries. Capacity degradation rate, which can be used to characterize the degree of lithium plating, is labeled into the feature space to determine the accuracy of clustered lithium plating diagnostic results.

Images