A lithium battery pack offline damage assessment method, system, electronic device and storage medium

Abstract

The present invention discloses a method, system, electronic device, and storage medium for offline damage assessment of lithium battery packs, relating to the field of damage diagnosis technology. The method includes: utilizing an improved Pearson correlation coefficient curve of lithium batteries under the influence of damage of different severity levels (I, II, III) to quantify battery voltage synchronization into an improved Pearson correlation coefficient sequence; calling a MATLAB toolbox to convert the correlation coefficient sequence into a recursive graph image using a recursive graph transformation; constructing a training data set based on the recursive graph image corresponding to each damage severity level and the recursive graph image in the undamaged state; using the training data set to train an M-RVM model; inputting the recursive graph image corresponding to the lithium battery pack to be diagnosed into the trained model, and using the trained model to output damage diagnosis results. The present invention can improve damage detection accuracy, damage classification accuracy, and damage severity grading accuracy while reducing the amount of computation.

Description

Technical Field

[0001] The present invention relates to the technical field of battery damage diagnosis, and in particular to a method, system, electronic device and storage medium for offline damage assessment of a lithium battery pack. Background Art

[0002] Lithium-ion batteries (Li-ion) have long played an irreplaceable role in energy storage applications for grid-connected power systems and electric vehicles, thanks to their long cycle life, high energy and power density, and negligible self-discharge. However, the inherent fragility of Li-ion batteries makes components such as insulating separators, electrodes, and electrolytes susceptible to damage, compromising their proper operation. Without timely maintenance, minor damage can easily escalate into systemic damage. The safety of Li-ion battery systems has garnered widespread attention from both industry and academia. Therefore, the development of effective diagnostic technologies for various Li-ion battery damages is imperative.

[0003] Existing lithium battery damage diagnosis methods can be classified into three types based on the main technologies used: behavioral mechanism modeling, prior knowledge accumulation, and abnormal signal analysis. Their characteristics are as follows:

[0004] 1. By integrating an observer or filter, the low-frequency impedance characteristics are used to reveal the correlation between the separator conductivity and short-circuit resistance based on a customized P2D model.

[0005] 2. Explore the electrochemical principles, obtain damage evidence by modeling the residuals of voltage, temperature and state of charge (SoC), construct a pseudo-distributed battery model, extract damage features, and use relevance vector machines to quantitatively evaluate the internal short circuit (ISC) state.

[0006] 3. To isolate damage in reconfigurable components, a structural analysis is performed based on an electrothermal coupling model, and a minimal sensor subset with optimal damage diagnosis reliability is pursued. An augmented state space model is established that combines traditional battery states and short current states to detect ISC.

[0007] Existing solutions are based on customized P2D models (pseudo-two-dimensions) and use low-frequency impedance characteristics to reveal the correlation between separator conductivity and short-circuit resistance. Although they have achieved admirable performance, they use a large number of partial differentials, which leads to high computational complexity. A large number of partial differential operations are used to describe the particle migration dynamics, resulting in huge computational overhead. Alternatively, they use pseudo-distributed battery models or expanded state-space models, which are overly dependent on model accuracy. If simplifications such as order reduction or linear approximation are used, the model accuracy will be greatly reduced, resulting in a significant reduction in damage diagnosis performance. As a result, the damage detection accuracy (the ability to distinguish between damage and normal without damage), damage classification accuracy, and damage severity grading accuracy are greatly reduced.

[0008] In summary, how to reduce the amount of computation while improving the accuracy of damage detection, damage classification, and damage severity grading has become an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0009] The purpose of the present invention is to provide a lithium battery pack offline damage assessment method, system, electronic device and storage medium, which can reduce the amount of calculation while improving the damage detection accuracy, damage classification accuracy and damage severity grading accuracy.

[0010] To achieve the above object, the present invention provides the following solutions:

[0011] A method for offline damage assessment of a lithium battery pack, the method comprising:

[0012] Obtaining improved Pearson correlation coefficient curves for the constructed lithium battery pack under the influence of various types of damage at each damage severity level; each of the improved Pearson correlation coefficient curves is obtained by applying offline current excitation to the constructed lithium battery pack to simulate different types of damage, and simulating different damage severity levels for each type of damage; the lithium battery pack includes a plurality of new lithium batteries connected in series; the improved Pearson correlation coefficient is obtained by improving the Pearson correlation coefficient using recursive adaptation, forgetting, and noise suppression;

[0013] For each of the improved Pearson correlation coefficient curves, the voltage synchronization between adjacent lithium batteries in the lithium battery pack is quantified into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve;

[0014] Calling the Matlab toolbox, for each of the improved Pearson correlation coefficient sequences, converting the improved Pearson correlation coefficient sequence into a recursion graph image in the form of recursion graph transformation;

[0015] A training data set is constructed based on the recursion graph images corresponding to each type of damage at each damage severity level and the recursion graph image corresponding to no damage; the training data set includes a plurality of the recursion graph images and damage diagnosis results corresponding to the recursion graph images; the damage diagnosis results include no damage and internal short circuit damage, external short circuit damage, and overheating damage at different damage severity levels; the different damage severity levels include minor damage, moderate damage, and severe damage;

[0016] Using the training data set to train the M-RVM model to obtain a trained M-RVM model;

[0017] Obtaining a recursive graph image corresponding to the lithium battery pack to be diagnosed;

[0018] The recursive graph image corresponding to the lithium battery pack to be diagnosed is input into the trained M-RVM model, and the trained M-RVM model is used to output a damage diagnosis result.

[0019] Optionally, the step of obtaining a recursion graph image corresponding to the lithium battery pack to be diagnosed further includes:

[0020] Obtaining an improved Pearson correlation coefficient curve corresponding to the lithium battery pack to be diagnosed;

[0021] quantifying the voltage synchronization between adjacent lithium batteries in the lithium battery group to be diagnosed into an improved Pearson correlation coefficient sequence corresponding to the lithium battery group to be diagnosed using the improved Pearson correlation coefficient curve corresponding to the lithium battery group to be diagnosed;

[0022] The improved Pearson correlation coefficient sequence corresponding to the lithium battery pack to be diagnosed is converted into a recursion graph image corresponding to the lithium battery pack to be diagnosed in the form of recursion graph transformation.

[0023] Optionally, the recursion graph image corresponding to the no-damage condition is obtained by detecting an improved Pearson correlation coefficient curve of the lithium battery pack without the influence of damage without simulating damage to the lithium battery pack, and using the improved Pearson correlation coefficient curve to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence, and converting the improved Pearson correlation coefficient sequence into a recursion graph image in the form of a recursion graph transformation.

[0024] The present invention also provides the following solution:

[0025] A lithium battery pack offline damage assessment system, the system comprising:

[0026] An improved Pearson correlation coefficient curve acquisition module is used to obtain improved Pearson correlation coefficient curves for a constructed lithium battery pack under the influence of various types of damage at each damage severity level; each improved Pearson correlation coefficient curve is obtained by applying offline current excitation to the constructed lithium battery pack to simulate different types of damage and different damage severity levels for each type of damage; the lithium battery pack includes multiple new lithium batteries connected in series; the improved Pearson correlation coefficient is obtained by improving the Pearson correlation coefficient using recursive adaptation, forgetting, and noise suppression;

[0027] a voltage synchronization quantification module, configured to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve for each of the improved Pearson correlation coefficient curves;

[0028] A recursion graph transformation module is used to call a MATLAB toolbox and, for each of the improved Pearson correlation coefficient sequences, convert the improved Pearson correlation coefficient sequence into a recursion graph image using a recursion graph transformation;

[0029] a training data set construction module, configured to construct a training data set based on the recursion graph images corresponding to various types of damage at each damage severity level and the recursion graph image corresponding to no damage; the training data set comprising a plurality of the recursion graph images and damage diagnosis results corresponding to the recursion graph images; the damage diagnosis results comprising no damage and internal short circuit damage, external short circuit damage, and overheating damage at different damage severity levels; the different damage severity levels comprising minor damage, moderate damage, and severe damage;

[0030] An M-RVM model training module is used to train the M-RVM model using the training data set to obtain a trained M-RVM model;

[0031] A recursion graph image acquisition module is used to acquire a recursion graph image corresponding to the lithium battery pack to be diagnosed;

[0032] The damage diagnosis module is used to input the recursive graph image corresponding to the lithium battery pack to be diagnosed into the trained M-RVM model, and output the damage diagnosis result using the trained M-RVM model.

[0033] Optionally, the system further comprises:

[0034] A module for obtaining a Pearson correlation coefficient curve to be diagnosed, used to obtain an improved Pearson correlation coefficient curve corresponding to the lithium battery pack to be diagnosed;

[0035] a voltage synchronization quantification module to be diagnosed, configured to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery group to be diagnosed into an improved Pearson correlation coefficient sequence corresponding to the lithium battery group to be diagnosed using an improved Pearson correlation coefficient curve corresponding to the lithium battery group to be diagnosed;

[0036] The Pearson correlation coefficient sequence conversion module for diagnosis is used to convert the improved Pearson correlation coefficient sequence corresponding to the lithium battery pack for diagnosis into a recursion graph image corresponding to the lithium battery pack for diagnosis in the form of recursion graph transformation.

[0037] Optionally, the recursion graph image corresponding to the no-damage condition is obtained by detecting an improved Pearson correlation coefficient curve of the lithium battery pack without the influence of damage without simulating damage to the lithium battery pack, and using the improved Pearson correlation coefficient curve to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence, and converting the improved Pearson correlation coefficient sequence into a recursion graph image in the form of a recursion graph transformation.

[0038] The present invention also provides the following solution:

[0039] An electronic device includes a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the offline damage assessment method for a lithium battery pack.

[0040] The present invention also provides the following solution:

[0041] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the offline damage assessment method for a lithium battery pack.

[0042] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0043] The offline damage assessment method, system, electronic device and storage medium of lithium battery pack disclosed in the present invention quantify the voltage synchronization between batteries into an IPCC (improved Pearson correlation coefficient sequence) sequence, in which inferences unrelated to damage, such as load dynamics and noise, can be effectively shielded, thereby retaining key information; calling the matlab toolbox and using the RP transformation (recursive graph transformation) to convert the time series (improved Pearson correlation coefficient sequence) into an RP image (recursive graph image), can intuitively visualize the hidden cross-time autocorrelation features as image textures, avoiding a large amount of useless data calculations, and in the process of converting the time series into the RP image In the process, Matlab is used to implement it, and there is no need to use a large number of partial differentials, thus reducing the amount of calculation; RP images contain rich texture details and can be used as fault evidence. RP images of different faults have unique textures, and the M-RVM model can effectively distinguish them. A real high-precision damage dataset based on RP images is constructed. Using this dataset to train the M-RVM model can obtain a higher-precision model. The trained M-RVM model can accurately identify non-damage and damage, and accurately diagnose the damage type and damage severity level, thereby improving the damage detection accuracy, damage classification accuracy, and damage severity grading accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a flow chart of Example 1 of the offline damage assessment method for lithium battery packs of the present invention;

[0046] Figure 2 This is a specific flow chart of the offline damage assessment method for lithium battery packs based on RP images of the present invention;

[0047] Figure 3 A physical view and schematic diagram of the experimental device of the present invention;

[0048] Figure 4 Schematic diagram of the cyclic load curve of the present invention;

[0049] Figure 5 A schematic diagram showing the comparison of battery voltage during injury;

[0050] Figure 6 Schematic diagram of the IPCC curve affected by different types and levels of damage;

[0051] Figure 7Schematic diagram of RP images affected by different types and levels of damage;

[0052] Figure 8 Schematic diagram of the damage diagnosis model of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] The purpose of the present invention is to provide a lithium battery pack offline damage assessment method, system, electronic device and storage medium, which can reduce the amount of calculation while improving the damage detection accuracy, damage classification accuracy and damage severity grading accuracy.

[0055] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] Example 1

[0057] Figure 1 This is a flow chart of Example 1 of the offline damage assessment method for lithium battery packs of the present invention. Figure 1 As shown, this embodiment provides a method for offline damage assessment of a lithium battery pack, comprising the following steps:

[0058] Step 101: Obtain an improved Pearson correlation coefficient curve for the constructed lithium battery pack under the influence of various types of damage at each damage severity level; each improved Pearson correlation coefficient curve is obtained by applying offline current excitation to the constructed lithium battery pack, simulating different types of damage, and simulating different damage severity levels for each type of damage; the lithium battery pack includes multiple new lithium batteries connected in series; the improved Pearson correlation coefficient is obtained by improving the Pearson correlation coefficient using recursive adaptation, forgetting, and noise suppression.

[0059] Step 102: For each improved Pearson correlation coefficient curve, the voltage synchronization between adjacent lithium batteries in the lithium battery pack is quantified into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve.

[0060] Step 103: calling the Matlab toolbox, and for each improved Pearson correlation coefficient sequence, converting the improved Pearson correlation coefficient sequence into a recursion graph image using a recursion graph transformation.

[0061] Step 104: Construct a training dataset based on the recursion graph images corresponding to each type of damage at each damage severity level, as well as the recursion graph image corresponding to no damage. The training dataset includes multiple recursion graph images and damage diagnosis results corresponding to the recursion graph images. The damage diagnosis results include no damage and internal short circuit damage, external short circuit damage, and overheating damage at different damage severity levels. The different damage severity levels include minor damage, moderate damage, and severe damage.

[0062] In step 104, the recursion graph image corresponding to no damage is obtained by detecting the lithium battery pack without simulating damage, and obtaining an improved Pearson correlation coefficient curve under the influence of no damage to the lithium battery pack. The improved Pearson correlation coefficient curve is used to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence, and the improved Pearson correlation coefficient sequence is converted into a recursion graph image in the form of a recursion graph transformation.

[0063] Step 105: Use the training data set to train the M-RVM model to obtain a trained M-RVM model.

[0064] Step 106: Obtain a recursion graph image corresponding to the lithium battery pack to be diagnosed.

[0065] Before step 106, the following steps are also included:

[0066] Obtain an improved Pearson correlation coefficient curve corresponding to the lithium battery pack to be diagnosed.

[0067] The voltage synchronization between adjacent lithium batteries in the lithium battery group to be diagnosed is quantified into an improved Pearson correlation coefficient sequence corresponding to the lithium battery group to be diagnosed using the improved Pearson correlation coefficient curve corresponding to the lithium battery group to be diagnosed.

[0068] The improved Pearson correlation coefficient sequence corresponding to the lithium battery pack to be diagnosed is converted into a recursion graph image corresponding to the lithium battery pack to be diagnosed in the form of recursion graph transformation.

[0069] Step 107: Input the recursive graph image corresponding to the lithium battery pack to be diagnosed into the trained M-RVM model, and use the trained M-RVM model to output a damage diagnosis result.

[0070] The technical solution of the present invention is described below with a specific embodiment:

[0071] The offline damage assessment method for lithium battery packs of the present invention is an offline damage assessment method for lithium battery packs based on RP images. The specific flow chart is as follows: Figure 2 As shown, the steps of the offline damage assessment method for lithium battery packs used in the present invention are as follows:

[0072] Step 1: Connect multiple lithium batteries (lithium-ion batteries) in series to construct a lithium-ion battery pack (lithium battery pack).

[0073] Figure 3 The physical view and schematic diagram of the experimental device of the present invention are shown in FIG. Figure 3 As shown, four new lithium-ion batteries (NCM, 3.7V, 4Ah) are connected in series to form a lithium-ion battery pack. The cascade voltage of the batteries is synchronously measured using a custom-designed circuit board (voltage acquisition unit), with a dedicated LTC6811 chip serving as the core processor. The lithium-ion batteries are ternary lithium batteries (NCM NI-CO-Mnbattery). An electronic load, a controllable DC power supply, and several relays work together to simulate load excitation on the components. An STM32-based module (i.e., local control unit) switches charging and discharging, uploads terminal data, and triggers damage. An RS-232 bus is used for data and command transmission between devices. A LabView-based application running on the host computer is designed for global control and data management. The signal sampling frequency is set to 5Hz. Figure 4 This is a schematic diagram of the cyclic load curve of the present invention, as shown in Figure 4 The multi-stage load state is shown as a load excitation cycle.

[0074] Step 2: Artificially simulate the behavior of damaging the battery.

[0075] The damage state is simulated because there is no problem with the battery itself in step one, but the damage needs to be measured, so artificial simulation of the damage is required. The battery cells are subjected to multiple abuse operations to induce the three types of damage shown in Table 1, and the three levels of each type (i.e., damage severity levels) are simulated. The three types are external short circuit (ESC), overheating (OHT), and internal short circuit (ISC). The three levels are I: minor damage, II: medium damage, and III: severe damage. Among them, level I is mild (Minor), level II is medium (Medium), and level III is serious (Serious). The damage configuration is shown in Table 1.

[0076] Table 1 Damage configuration

[0077]

[0078] The internal resistance in Table 1 refers to the internal resistance measured after the battery is overcharged, and the distance refers to the distance from the spray gun nozzle to the battery surface.

[0079] The simulation of ISC damage is to change the state of charge (SOC) percentage and the internal resistance value. The simulation of ESC damage is to change the short-circuit resistance value. The simulation of OHT damage is to change the heating temperature and distance.

[0080] Measuring voltage can be used as a measure of damage because there is a voltage difference between damaged and healthy cells. Lithium battery damage can affect ion insertion and extraction rates, solid electrolyte interphase (SEI) thickness, and electrolyte conductivity, manifesting macroscopically as changes in overpotential and internal resistance. Therefore, monitoring electrical synchrony (voltage synchrony) between cells can provide evidence for fault diagnosis.

[0081] The Pearson correlation coefficient (PCC) can reflect the relationship between voltages. It is a linear correlation coefficient and the most commonly used correlation coefficient, cov(X,Y), used to reflect the degree of linear correlation between two variables, X and Y. A larger absolute value indicates a stronger correlation. Here, it is used to observe the correlation between the voltages of battery cells for correlation analysis.

[0082] To detect system anomalies online, it is necessary to introduce recursive and noise suppression mechanisms to PCC. Furthermore, to reveal damage-related information from lengthy non-damage data, rather than being overwhelmed by it, a forget window needs to be added to PCC to balance damage sensitivity and data retention. This is known as the Impoving Pearson Correlation Coefficient (IPCC). The IPCC (Improved Pearson Correlation Coefficient) curve quantifies the electrical synchronization between units as a Pearson correlation coefficient sequence, effectively shielding damage-independent inferences such as load dynamics and noise.

[0083] Improved Pearson correlation coefficient:

[0084] PCC is the most commonly used index to measure the correlation between two variables. For discrete variables X and Y, the PCC expression is:

[0085]

[0086]

[0087] At the same time, in order to detect system anomalies online, it is necessary to introduce recursion and noise suppression mechanisms. In addition, to reveal damage-related information, a forgetting window is required to balance damage sensitivity and data retention. The PCC formula is improved to:

[0088]

[0089] in:

[0090] Here, P, Q, R, S, and T are abbreviations for convenience and have no meaning. The specific forms are as follows:

[0091]

[0092] in v x,i and v y,i is the original signal (e.g. battery voltage), It is an auxiliary square wave that eliminates disordered oscillations caused by measurement errors and noise in a stable state, w is the forgetting window length, and recursive calculation is performed. In the present invention, w is consistent with the length and width of the image described later, The sampling interval is set to 6. IPCC can effectively mitigate false alarms caused by load fluctuations while retaining key information.

[0093] The above formulas are all existing formulas. The meanings of the various parameters involved in the formulas are all recorded in existing public documents and therefore will not be repeated here.

[0094] The Pearson correlation coefficient is both an upgrade of the Euclidean distance (that is, it provides processing steps for different variable value ranges, and the differences in the dimensions of different variables are removed in the calculation process), and an improvement of the cosine similarity when dimension values ​​are missing.

[0095] In order to observe the damage of the battery, it is necessary to artificially apply damage operations. Table 1 shows the specific operations of the damage configuration.

[0096] Step 3: Obtain the IPCC curve, which is used as the image source for step 4 in subsequent applications.

[0097] In the previous step 2, multiple groups of abuse operations are performed on the battery cells to induce OHT and ISC damage. The voltage of the adjacent batteries is collected by the voltage acquisition unit. Figure 5 As shown, Figure 5 A comparison diagram of battery voltage during damage. Figure 5 Part (a) shows the voltage of the healthy battery and the OHT battery. Figure 5 Part (b) shows the voltage of healthy battery and ISC. Taking OHT and ISC damage as examples, Figure 5The terminal voltages of two adjacent cells (one intact, the other damaged) are depicted in Figure 1, where the dashed box indicates the duration of damage activation. It can be seen that in both cases, the voltage of the damaged cell gradually deviates from that of the healthy cell. However, traditional threshold-based methods do not provide sufficient evidence to reveal damage details, as voltage differences are easily overwhelmed by high dynamic loads, sensor temperature drift, and EMI distortion. Since these adverse factors typically interfere with all voltage levels simultaneously, IPCC can accurately suppress them. From an electrochemical perspective, OHT typically has a direct impact on the electrolyte, lithiation and delithiation, thereby affecting particle migration resistance; ISC caused by separator rupture typically reduces static internal resistance without significantly changing dynamic internal resistance. These unique impedance changes, manifested as distinct responses in the frequency domain, can be preserved by IPCC and then materialized as image textures. Overall, IPCC demonstrates both sensitivity and reliability in damage detection. Figure 6 Schematic diagram of the IPCC curve affected by different types and levels of damage. Figure 6 The IPCC curves obtained by detection under the influence of different types and levels of damage are shown. Figure 6 Part (a) shows the IPCC curve obtained without damage. Figure 6 Part (b) shows the IPCC curve obtained under the influence of ESC damage. Figure 6 Part (c) shows the IPCC curve obtained under the influence of ISC damage. Figure 6 Part (d) shows the IPCC curve obtained under the influence of OHT damage.

[0098] Step 4: Use RP transformation to convert the time series into RP image.

[0099] RP is a recurrence visualization method proposed based on the Poincare theorem. It draws prior knowledge from the internal structure of time series, explains the similarity and information content of time series, and enables predictability analysis of signals. It is an important method for analyzing periodicity, chaos, and non-stationarity in time series. RP is essentially a time-to-time signal processing method used to encode cyclical behavior in time series into a two-dimensional image for representation. Time series data can be classified based on unique recurring behaviors, such as periodicity and irregular periodicity.

[0100] Figure 7 Schematic diagram of RP images affected by different types and levels of damage. Figure 7 Part (a) shows the RP image affected by ISC damage. Figure 7 Part (b) shows the RP image affected by ESC damage. Figure 7Part (c) shows the RP image affected by OHT damage. In order to compare the texture differences between the images, Figure 7 Figure 2 shows nine sets of RP images collected based on vertical type and horizontal grade. Images within the same beam have the same damage type and grade, but are subject to different loads. After load calibration, images of the same location from different beams are derived. Observations show that these RP images contain rich textural details, providing evidence of faulting. This is why the RP transform was used instead of directly using the IPCC curves.

[0101] Steps 1 through 4 are all intended to obtain a realistic damage dataset for training the mRVM model (M-RVM model). Based on these steps, the present invention obtains a relatively realistic training dataset, i.e., a realistic, high-precision damage dataset. The high accuracy of the present invention is due to the relatively realistic training dataset obtained through steps 1 through 4. Using this dataset to train the M-RVM model yields a highly accurate model, thereby achieving precise damage classification.

[0102] Step 5: Train the M-RVM model and obtain damage diagnosis (damage detection) results.

[0103] M-RVM (multi-class relevance vector machine) is a multi-class relevance vector machine; the RVM algorithm is a machine learning model based on the Bayesian framework. It obtains relevant vectors (RVs) and weights by maximizing the marginal likelihood, and then performs machine learning through the M-RVM algorithm (M-RVM model) to train the model samples. During the training iteration process, the posterior distribution of most parameters tends to zero, and the samples corresponding to the non-negligible parameters are called RVs, which reflect the main information of all samples. The number of RVs undergoes a three-phase change of increase-flat-decrease, which means that a highly sparse model is obtained. The model can diagnose the damage level of the battery in the battery pack, such as No fault (no damage), Minor, Medium, Serious, and classify its damage type into ISC, ESC, OHT. The process is as follows Figure 8 As shown, Figure 8 This is a schematic diagram of the damage diagnosis model of the present invention. A total of 10,000 signal samples, covering ten states (including one undamaged state and the nine damaged states shown in Table 1), were input into the M-RVM algorithm for machine learning model training. As shown in Table 2, 10,000 RPs were generated using this data. The image sets were then split at an 8:2 ratio for training and testing, respectively, as shown in Table 2.

[0104] Table 2 Image samples and labels of different damage configurations used for model training and testing

[0105] Label Damage status train test A No damage 800 200 B External short circuit damage (I) 800 200 C External short circuit damage (II) 800 200 D External short circuit damage (III) 800 200 E Internal short circuit damage (I) 800 200 F Internal short circuit damage (II) 800 200 G Internal short circuit damage (III) 800 200 H Overheat Damage (I) 800 200 I Overheating damage (II) 800 200 J Overheating damage (III) 800 200

[0106] The training dataset is three-quarters of the 10,000 data samples mentioned above, which are used for model training, and the rest are used for method testing. The feature samples are input and the damage detection and separation matrix is ​​output.

[0107] Damage Detection:

[0108] We mixed 800 samples from the "test" dataset in Table 2 (including A) with another 600 uniformly randomly selected samples from the three damage types. Only two ISC samples on the RP were misclassified as ESC and undamaged, respectively. Detection accuracy was perfect, correctly identifying all undamaged samples. This means that RP images of different slices have unique textures, and the mRVM algorithm (M-RVM model) can effectively distinguish them.

[0109] Injury classification:

[0110] The damage classification performance was verified. Referring to the test data in Table 2, three datasets of each damage type were mixed with the undamaged dataset: "B+C+D+A," "E+F+G+A," and "H+I+J+A," so that each test set contained 800 samples. Table 3 summarizes the results, showing that a significant portion of the damaged samples were misdiagnosed. The RP score for ESC samples was 61%. This result is predictable from the RP mechanism: the uniform partitioning scheme obfuscates the IPCC data for ESC, which are mostly concentrated near the minimum with minimal fluctuation. For the ISC case, the RP score achieved an average accuracy of 63.3%, with slightly higher accuracy at 51.9% for Level II (row "F"). This can be explained by the considerable randomness of the destructive effects of high temperature on battery structure and materials, and the inability to fully demonstrate the consequences of overheating when the defective battery is subjected to dynamic loading. The overall damage classification accuracy for RP was 58.7%, which is generally consistent with the training results. Table 3 shows the damage classification matrix for RP, with reference to the labels in Table 2.

[0111] Table 3 Damage classification matrix on RP

[0112]

[0113]

[0114] The bold data in Table 3 represent successful classification (i.e., the type of injury was successfully diagnosed). Except for the bold data, data 0, and accuracy data (including 61.0%, 63.3%, 51.9%, and 58.7%), the remaining data represent unsuccessful diagnosis of the type of injury, and zero represents no sample.

[0115] In summary, the offline damage assessment method for lithium-ion battery packs proposed in this paper demonstrates significant advantages in damage type detection (damage detection). Although OHT's performance is significantly inferior in terms of damage severity classification, with subtle texture differences between different damage severities difficult to discern, this method, while less than satisfactory, shows promise for further optimization.

[0116] The present invention provides an offline damage assessment method for lithium battery packs, a method for lithium battery pack damage diagnosis. It aims to construct a novel, intelligent, damage diagnosis framework for series battery packs. First, the electrical system quantifies the synchronization between adjacent cells in real time to indicate lithium battery anomalies. The improved PCC (IPCC) sequence is then converted into a recurrence plot (RP) using the RP transformation. The recurrence plot (RP) is then used to convert the correlation coefficient sequence into an RP image (using a MATLAB toolbox). The texture of the RP image reflects detailed information about the system status. Finally, the M-RVM algorithm is used to integrate feature evidence to detect damage occurrence and determine the damage type and severity. This improves both the overall damage detection accuracy (the ability to distinguish damaged segments from normal, undamaged segments) and the damage classification accuracy, specifically the damage severity classification (grades I, II, and III), in large battery pack damage diagnosis.

[0117] The present invention constructs an intelligent diagnostic framework for common damages of lithium battery packs, including ESC, ISC, and OHT. First, the Pearson correlation coefficient (PCC) is improved through recursive adaptation, forgetting, and noise suppression. Then, the synchronization between adjacent batteries is quantified in real time through the electrical system to indicate lithium battery abnormalities. The improved PCC (IPCC) sequence is then converted into an RP image through RP transformation. Thus, the timeline fluctuations representing different damages can be regarded as image textures. Finally, the M-RVM algorithm is used to integrate feature evidence to detect the occurrence of damage and make judgments on the type and severity of the damage, thereby improving the overall damage detection accuracy and damage grading accuracy in large battery pack damage diagnosis.

[0118] The key points of the present invention are:

[0119] 1. IPCC datasets and images of faults of different types and levels.

[0120] 2. Structural difference data and images of RP under different fault states.

[0121] 3. Use the sparse M-RVM model (M-RVM model) to diagnose damage type (detection) and damage level (evaluation).

[0122] Compared with the prior art, the advantages of the present invention are:

[0123] 1. The electrical synchrony between batteries is quantified as a sequence of correlation coefficients (IPCC), where damage-irrelevant inferences (such as load dynamics and noise) can be effectively shielded.

[0124] 2. Using the RP transform as an example, the time series is converted into RP images. This can intuitively visualize the hidden cross-temporal autocorrelation features as image textures, avoiding a large amount of useless data calculations. (After conversion to image textures, the MATLAB toolbox can be directly called for identification.) The present invention uses MATLAB to implement the process of converting the time series into RP images. Because it does not use a large number of partial differentials, the amount of calculation is small, greatly reducing the amount of computation.

[0125] 3. Current methods for battery pack damage assessment use behavioral mechanism modeling combined with prior knowledge to estimate internal battery parameters. However, these methods use inaccurate battery models, have insufficient data, and fail to consider the electrochemical reaction mechanisms within the battery. The actual process is more complex than anticipated. The present invention uses real-time measurement data from the battery pack, avoiding errors in battery model estimation. Furthermore, the data is fully obtained in real time from actual measurements, making it sufficiently comprehensive. The resulting data fully encompasses the internal electrochemical reaction mechanisms of the battery, resulting in higher accuracy.

[0126] 4. By applying different physical damages to the battery, including thermal and electrical abuse, damage consequences are induced and a real high-precision damage dataset is obtained.

[0127] In addition, other more complex experimental algorithms can also achieve better damage identification effects to some extent. However, due to the complexity of their models and algorithms, they are not suitable for engineering implementation of existing damage identification. Therefore, the present invention is a relatively superior solution.

[0128] Example 2

[0129] In order to execute the method corresponding to the above embodiment 1 and achieve the corresponding functions and technical effects, a lithium battery pack offline damage assessment system is provided below, which includes:

[0130] An improved Pearson correlation coefficient curve acquisition module is used to obtain improved Pearson correlation coefficient curves for a constructed lithium battery pack under the influence of various types of damage at each damage severity level; each improved Pearson correlation coefficient curve is obtained by applying offline current excitation to the constructed lithium battery pack, simulating different types of damage, and simulating different damage severity levels for each type of damage; the lithium battery pack includes multiple new lithium batteries connected in series; the improved Pearson correlation coefficient is obtained by improving the Pearson correlation coefficient using recursive adaptation, forgetting, and noise suppression.

[0131] The voltage synchronization quantification module is used to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve for each improved Pearson correlation coefficient curve.

[0132] The recursive graph transformation module is used to call the matlab toolbox and, for each improved Pearson correlation coefficient sequence, convert the improved Pearson correlation coefficient sequence into a recursive graph image in the form of recursive graph transformation.

[0133] The training dataset construction module is used to construct a training dataset based on the recursion graph images corresponding to various types of damage at each damage severity level, as well as the recursion graph images corresponding to no damage. The training dataset includes multiple recursion graph images and the damage diagnosis results corresponding to the recursion graph images. The damage diagnosis results include no damage and internal short circuit damage, external short circuit damage, and overheating damage at different damage severity levels. The different damage severity levels include minor damage, moderate damage, and severe damage.

[0134] Among them, the recursion graph image corresponding to no damage is obtained by detecting the lithium battery pack without simulating damage, and the improved Pearson correlation coefficient curve under the influence of no damage is obtained. The improved Pearson correlation coefficient curve is used to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence, and the improved Pearson correlation coefficient sequence is converted into a recursion graph image in the form of recursion graph transformation.

[0135] The M-RVM model training module is used to train the M-RVM model using the training data set to obtain a trained M-RVM model.

[0136] The recursion graph image acquisition module is used to acquire the recursion graph image corresponding to the lithium battery pack to be diagnosed.

[0137] The damage diagnosis module is used to input the recursive graph image corresponding to the lithium battery pack to be diagnosed into the trained M-RVM model, and output the damage diagnosis result using the trained M-RVM model.

[0138] The lithium battery pack offline damage assessment system also includes:

[0139] The module for obtaining the Pearson correlation coefficient curve of the battery to be diagnosed is used to obtain the improved Pearson correlation coefficient curve corresponding to the lithium battery pack to be diagnosed.

[0140] The voltage synchronization quantification module to be diagnosed is used to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery group to be diagnosed into an improved Pearson correlation coefficient sequence corresponding to the lithium battery group to be diagnosed using the improved Pearson correlation coefficient curve corresponding to the lithium battery group to be diagnosed.

[0141] The Pearson correlation coefficient sequence conversion module to be diagnosed is used to convert the improved Pearson correlation coefficient sequence corresponding to the lithium battery pack to be diagnosed into a recursion graph image corresponding to the lithium battery pack to be diagnosed in the form of recursion graph transformation.

[0142] The lithium battery pack offline damage assessment system of the present invention is an RP image-based lithium battery pack offline damage assessment system, which can reduce the amount of calculation while improving the damage detection accuracy, damage classification accuracy and damage severity grading accuracy.

[0143] Example 3

[0144] A third embodiment of the present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the offline damage assessment method for a lithium battery pack of the first embodiment.

[0145] The electronic device mentioned above may be a server.

[0146] Example 4

[0147] A fourth embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the offline damage assessment method for a lithium battery pack of the first embodiment is implemented.

[0148] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0149] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A lithium battery pack offline damage assessment method, characterized in that: The method comprises: Obtaining improved Pearson correlation coefficient curves for the constructed lithium battery pack under the influence of various types of damage at each damage severity level; each of the improved Pearson correlation coefficient curves is obtained by applying offline current excitation to the constructed lithium battery pack to simulate different types of damage, and simulating different damage severity levels for each type of damage; the lithium battery pack includes a plurality of new lithium batteries connected in series; the improved Pearson correlation coefficient is obtained by improving the Pearson correlation coefficient using recursive adaptation, forgetting, and noise suppression; For each of the improved Pearson correlation coefficient curves, the voltage synchronization between adjacent lithium batteries in the lithium battery pack is quantified into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve; Calling the Matlab toolbox, for each of the improved Pearson correlation coefficient sequences, converting the improved Pearson correlation coefficient sequence into a recursion graph image in the form of recursion graph transformation; A training data set is constructed based on the recursion graph images corresponding to each type of damage at each damage severity level and the recursion graph image corresponding to no damage; the training data set includes a plurality of the recursion graph images and damage diagnosis results corresponding to the recursion graph images; the damage diagnosis results include no damage and internal short circuit damage, external short circuit damage, and overheating damage at different damage severity levels; the different damage severity levels include minor damage, moderate damage, and severe damage; Using the training data set to train the M-RVM model to obtain a trained M-RVM model; Obtaining a recursive graph image corresponding to the lithium battery pack to be diagnosed; The recursive graph image corresponding to the lithium battery pack to be diagnosed is input into the trained M-RVM model, and the trained M-RVM model is used to output a damage diagnosis result.

2. The offline damage assessment method for lithium battery pack according to claim 1, characterized in that: The step of obtaining the recursive graph image corresponding to the lithium battery pack to be diagnosed further includes: Obtaining an improved Pearson correlation coefficient curve corresponding to the lithium battery pack to be diagnosed; quantifying the voltage synchronization between adjacent lithium batteries in the lithium battery group to be diagnosed into an improved Pearson correlation coefficient sequence corresponding to the lithium battery group to be diagnosed using the improved Pearson correlation coefficient curve corresponding to the lithium battery group to be diagnosed; The improved Pearson correlation coefficient sequence corresponding to the lithium battery pack to be diagnosed is converted into a recursion graph image corresponding to the lithium battery pack to be diagnosed in the form of recursion graph transformation.

3. The offline damage assessment method for lithium battery pack according to claim 1, characterized in that: The recursion graph image corresponding to the absence of damage is obtained by detecting the lithium battery pack without simulating damage to the lithium battery pack, and quantifying the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve, and converting the improved Pearson correlation coefficient sequence into a recursion graph image using a recursion graph transformation.

4. A lithium battery pack offline damage assessment system, characterized in that: The system comprises: An improved Pearson correlation coefficient curve acquisition module is used to obtain improved Pearson correlation coefficient curves for a constructed lithium battery pack under the influence of various types of damage at each damage severity level; each improved Pearson correlation coefficient curve is obtained by applying offline current excitation to the constructed lithium battery pack to simulate different types of damage and different damage severity levels for each type of damage; the lithium battery pack includes multiple new lithium batteries connected in series; the improved Pearson correlation coefficient is obtained by improving the Pearson correlation coefficient using recursive adaptation, forgetting, and noise suppression; a voltage synchronization quantification module, configured to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve for each of the improved Pearson correlation coefficient curves; A recursion graph transformation module is used to call a MATLAB toolbox and, for each of the improved Pearson correlation coefficient sequences, convert the improved Pearson correlation coefficient sequence into a recursion graph image using a recursion graph transformation; a training data set construction module, configured to construct a training data set based on the recursion graph images corresponding to various types of damage at each damage severity level and the recursion graph image corresponding to no damage; the training data set comprising a plurality of the recursion graph images and damage diagnosis results corresponding to the recursion graph images; the damage diagnosis results comprising no damage and internal short circuit damage, external short circuit damage, and overheating damage at different damage severity levels; the different damage severity levels comprising minor damage, moderate damage, and severe damage; An M-RVM model training module is used to train the M-RVM model using the training data set to obtain a trained M-RVM model; A recursion graph image acquisition module is used to acquire a recursion graph image corresponding to the lithium battery pack to be diagnosed; The damage diagnosis module is used to input the recursive graph image corresponding to the lithium battery pack to be diagnosed into the trained M-RVM model, and output the damage diagnosis result using the trained M-RVM model.

5. The lithium battery pack offline damage assessment system according to claim 4, characterized in that: The system further comprises: A module for obtaining a Pearson correlation coefficient curve to be diagnosed, used to obtain an improved Pearson correlation coefficient curve corresponding to the lithium battery pack to be diagnosed; a voltage synchronization quantification module to be diagnosed, configured to quantify the voltage synchronization between adjacent lithium batteries in the lithium battery group to be diagnosed into an improved Pearson correlation coefficient sequence corresponding to the lithium battery group to be diagnosed using an improved Pearson correlation coefficient curve corresponding to the lithium battery group to be diagnosed; The Pearson correlation coefficient sequence conversion module for diagnosis is used to convert the improved Pearson correlation coefficient sequence corresponding to the lithium battery pack for diagnosis into a recursion graph image corresponding to the lithium battery pack for diagnosis in the form of recursion graph transformation.

6. The lithium battery pack offline damage assessment system according to claim 4, characterized in that: The recursion graph image corresponding to the absence of damage is obtained by detecting the lithium battery pack without simulating damage to the lithium battery pack, and quantifying the voltage synchronization between adjacent lithium batteries in the lithium battery pack into an improved Pearson correlation coefficient sequence using the improved Pearson correlation coefficient curve, and converting the improved Pearson correlation coefficient sequence into a recursion graph image using a recursion graph transformation.

7. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the offline damage assessment method for a lithium battery pack according to any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that It stores a computer program, which, when executed by a processor, implements the offline damage assessment method for a lithium battery pack as described in any one of claims 1 to 3.

Images