A graded early warning method and system for thermal runaway of lithium-ion batteries based on bulging sound signals

Abstract

This invention belongs to the field of lithium-ion battery thermal runaway early warning technology, specifically disclosing a graded early warning method and system for lithium-ion battery thermal runaway based on bulging sound signals. The method includes: acquiring a lithium-ion battery sound signal sequence and performing denoising processing; performing an inverse Fourier transform on the denoised sound signal to transmit the signal back to the time domain, obtaining a denoised continuous signal; extracting six features based on the denoised continuous signal: signal peak value, zero crossover rate, energy change rate, high-frequency energy proportion, frequency entropy, and low-frequency persistence; calculating the importance distance of each feature according to its corresponding eigenvalue in the correlation coefficient matrix; selecting the five features with the largest importance distance as the final features, and using a weighted combination number Bagging model to obtain the battery thermal runaway risk early warning result. This invention has high sensitivity and can detect lithium-ion battery thermal runaway before significant changes in current, voltage, or temperature occur.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery thermal runaway early warning technology, and in particular to a graded early warning method and system for lithium-ion battery thermal runaway based on bulging sound signals. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] During use, lithium-ion batteries experience a continuous rise in internal temperature due to overcharging, overheating, or short circuits. This intensifies chemical reactions, causing the electrolyte to decompose and release flammable gases, leading to battery expansion. As these flammable gases accumulate, the temperature continues to rise, and the internal chemical reactions become even more vigorous, resulting in self-heating. If this self-heating cannot be effectively suppressed, the battery may catch fire or even explode. Therefore, accurate and reliable thermal runaway early warning methods are crucial for ensuring the safe and reliable operation of batteries.

[0004] However, lithium-ion batteries, as highly coupled nonlinear systems, have extremely complex and difficult-to-understand thermal runaway evolution mechanisms. Furthermore, the early characteristics of thermal runaway evolution are extremely subtle and easily masked by measurement noise. Once thermal runaway evolves to the middle or late stages, the battery temperature rises rapidly, leading to fires and explosions. Therefore, early warning of thermal runaway is extremely difficult.

[0005] Existing technologies disclose methods for constructing and training neural network models using battery indicators such as temperature, strain, and gas concentration as inputs to determine whether a battery is at risk of thermal runaway. Existing technologies also disclose methods for identifying and warning of thermal runaway risk by presetting minimum temperature rise rate and temperature change rate thresholds, and combining these with real-time temperature monitoring to calculate dual rate parameters. However, these methods all rely on temperature as the primary parameter for thermal runaway warning, exhibiting significant lag. Furthermore, obtaining strain and gas concentration requires auxiliary sensing equipment, increasing the cost of early warning. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a graded early warning method and system for lithium-ion battery thermal runaway based on bulging sound signals. It extracts different time-frequency features from the bulging sound signals generated during the battery bulging process and proposes a weighted combination number (WCN) Bagging model to achieve real-time graded early warning for lithium-ion battery thermal runaway.

[0007] In some implementations, the following technical solutions are adopted:

[0008] A graded early warning method for thermal runaway of lithium-ion batteries based on bulging sound signals includes:

[0009] Acquire the audio signal sequence of a lithium-ion battery and perform noise reduction processing;

[0010] Performing an inverse Fourier transform on the denoised audio signal and transferring the signal back to the time domain yields the denoised continuous signal:

[0011] Based on the denoised continuous signal, six features are extracted respectively: signal peak value, zero crossover rate, energy change rate, high frequency energy ratio, frequency entropy, and low frequency persistence.

[0012] Calculate the importance distance of each feature based on its corresponding eigenvalue in the correlation coefficient matrix;

[0013] The five features with the largest importance distance were selected as the final features, and the weighted combination number Bagging was used.

[0014] The model yields early warning results for battery thermal runaway risk.

[0015] As a further solution, the audio signal sequence of the lithium-ion battery is denoised, specifically as follows:

[0016] The lithium-ion battery audio signal sequence is divided into multiple frames, and the different frequency components in the signal are separated by fast Fourier transform to obtain the signal spectrum.

[0017] Use silent segments to estimate the power spectrum of noise;

[0018] The power spectrum after denoising is obtained by subtracting the power spectrum of noise from the power spectrum of the signal.

[0019] As a further solution, the signal peak value The maximum transient response used to characterize the signal amplitude is specifically formulated as follows:

[0020] ;

[0021] in, X i For the first i One data point;

[0022] Or, the zero cross rate The formula used to measure the oscillation frequency of a signal is as follows:

[0023] ;

[0024] Where, Ⅱ is an indicator function, when ( x i xi+1 When ) < 0, take 1. x i x i+1 If )>0, take 0; N The total number of data points in the signal. x i For the first i The value of each data point;

[0025] Or, the rate of energy change The formula used to reflect the degree of change in signal energy over time is as follows:

[0026] ;

[0027] in, N The total number of data points in the signal. x i For the first i The value of each data point E i Is the signal at the 1st i The energy at each data point is represented as the square of the amplitude of that data point.

[0028] As a further solution, the high-frequency energy ratio The formula used to describe the relative contribution of high-frequency components in a signal is as follows:

[0029] ;

[0030] in, f c This is the starting frequency of the high-frequency range. f max The highest frequency in the signal spectrum. X ( f ) represents the signal at frequency f The spectral values ​​on the spectrum.

[0031] Alternatively, the frequency entropy measures the degree of disorder in the signal spectrum, specifically:

[0032] ;

[0033] in, X ( f ) represents the signal at frequency f Spectral values ​​on P ( f ) indicates frequency f The probability density at that location;

[0034] Low-frequency persistence describes the significance of low-frequency components in a signal, and the specific formula is as follows:

[0035] ;

[0036] in, f d This is the highest frequency in the low-frequency range. f max This represents the highest frequency of the signal.

[0037] As a further approach, the importance distance of each feature is calculated based on its corresponding eigenvalue in the correlation coefficient matrix, specifically as follows:

[0038] ;

[0039] in, Representation of features In time t The value, n This represents the total time duration. express for The probability of; Representation of features eigenvalues.

[0040] As a further approach, the weighted combination number Bagging algorithm is used to obtain the battery thermal runaway risk warning result, specifically:

[0041] ;

[0042] Three types are selected as a base classifier. z This represents a sample with the three combined features. B This represents the total number of base classifiers. For the first b A base classifier evaluates samples z Category c Probability estimation; This represents the importance distance of any of the three selected features. It represents the importance distance of any feature among all features.

[0043] As a further measure, the specific results of the battery thermal runaway risk warning are: normal operation, slow bulging, or rapid bulging.

[0044] In other embodiments, the following technical solutions are adopted:

[0045] A graded early warning system for thermal runaway of lithium-ion batteries based on bulging sound signals includes:

[0046] The data acquisition module is configured to acquire the sound signal sequence of the lithium-ion battery and perform noise reduction processing.

[0047] The signal denoising module is configured to perform an inverse Fourier transform on the denoised audio signal, transmit the signal back to the time domain, and obtain the denoised continuous signal.

[0048] The feature extraction module is configured to extract six features based on the denoised continuous signal: signal peak value, zero crossover rate, energy change rate, high-frequency energy ratio, frequency entropy, and low-frequency persistence.

[0049] The distance calculation module is configured to calculate the importance distance of each feature based on the feature values ​​corresponding to each feature in the correlation coefficient matrix.

[0050] The risk prediction module is configured to select the five features with the largest importance distance as the final features and use the weighted combination number Bagging model to obtain the battery thermal runaway risk warning result.

[0051] In other embodiments, the following technical solutions are adopted:

[0052] A terminal device includes a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide the above-described graded early warning method for thermal runaway of lithium-ion batteries based on bulging sound signals.

[0053] In other embodiments, the following technical solutions are adopted:

[0054] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the above-described method for graded early warning of thermal runaway of lithium-ion batteries based on bulging sound signals.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] (1) The present invention uses battery sound signals for battery thermal runaway early warning. The characteristics of sound signals provide real-time monitoring capability and have high sensitivity, enabling the detection of thermal runaway of lithium-ion batteries before significant changes in current, voltage or temperature occur. This method does not require damage to the battery structure and achieves non-invasive monitoring.

[0057] (2) The present invention uses the high-frequency short-time bulging sound and low-frequency oscillating bulging sound generated during the battery bulging process to train the weighted combination number Bagging model. These two sounds are representative of the two classic modes of battery bulging and are relatively easy to collect and manually label.

[0058] (3) This invention extracts six sound features based on battery sound signals. For different batteries, five of these features are selected as the final features based on importance distance. This achieves adaptive feature dimensionality reduction for different batteries and adaptive dynamic selection of thermal runaway features. Different features are used at different times according to their effectiveness, avoiding interference from low-contribution features on the early warning effect. In addition, since the algorithm complexity increases sharply with the number of features, discarding low-contribution features can greatly reduce the computational complexity without significantly affecting the accuracy of the algorithm.

[0059] (4) Conventional weighted Bagging methods use linear normalization, which is simple to calculate weights and is difficult to effectively highlight the efficient features under different time periods. This invention uses a weighted combination number Bagging model for battery thermal runaway early warning. In the classification process, weights are assigned to different features based on the importance distance of the features, which strengthens the influence of features with large importance distances. Further calculation and allocation of weights helps to increase the learning ability of the algorithm and enhance the accuracy and robustness of the algorithm.

[0060] By combining feature extraction, the number of features in the base classifier is greatly reduced, significantly alleviating the training burden. Furthermore, the lower feature count improves the reliability of the base classifier, and combined with weight allocation based on feature contribution, the decision-making process further focuses on high-quality features, avoiding interference from irrelevant features.

[0061] Other features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0062] Figure 1 This is a flowchart of a graded early warning method for thermal runaway of lithium-ion batteries based on bulging sound signals, as described in an embodiment of the present invention.

[0063] Figure 2 This is a schematic diagram of the audio signal of 5C over-rate thermal runaway in an embodiment of the present invention;

[0064] Figure 3 This is a schematic diagram of the normal signal under 5C over-rate operation in an embodiment of the present invention;

[0065] Figure 4 This is a schematic diagram of a 120% SOC overcharge with bulging signal in an embodiment of the present invention;

[0066] Figure 5 This is a schematic diagram of a 130% SOC overcharge with bulging signal in an embodiment of the present invention;

[0067] Figure 6 The audio signal and warning result for the 140% SOC overcharge condition in this embodiment of the invention;

[0068] Figure 7 The audio signal and warning result for the 5C over-rate operating condition in this embodiment of the invention;

[0069] Figure 8 This is the confusion matrix of the classification results using the weighted combination number Bagging model in this embodiment of the invention. Detailed Implementation

[0070] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0071] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0072] Example 1

[0073] In one or more embodiments, a graded early warning method for thermal runaway of lithium-ion batteries based on bulging sound signals is disclosed, combined with... Figure 1 Specifically, it includes the following processes:

[0074] S101: Acquire the sound signal sequence of the lithium-ion battery and perform noise reduction processing.

[0075] Specifically, in this embodiment, an external microphone sensor is used to acquire the sound signal sequence during the operation of the lithium-ion battery; the sound signal sequence is acquired in a non-contact and non-invasive manner, which is simple and easy to implement.

[0076] In this embodiment, the process of denoising the lithium-ion battery audio signal sequence is as follows:

[0077] For a lithium-ion battery audio signal sequence, it is divided into multiple frames based on signal characteristics and computational resources, and then a window is added to each frame. The framed signal is represented as follows:

[0078] (1)

[0079] In the formula, m It is the index of the coordinate system. w ( n ) is a window function; , L represents the audio signal after framing, the audio signal before framing, and the window function length, respectively.

[0080] The Fast Fourier Transform is used to separate the different frequency components in the signal, yielding the signal's spectrum:

[0081] (2)

[0082] In the formula, k For frequency index, x ( k , m ) is the first m Frame complex spectrum.

[0083] The power spectrum of noise is estimated using the silent segment, where the silent segment is the normal ambient noise when the battery sound is not being collected, i.e., the sound when the experiment is not running.

[0084] (3)

[0085] In the formula, The power spectrum of the silent signal. M For the estimated number of frames in the silent segment, An audio signal that is framed for a silent signal.

[0086] The power spectrum after noise is subtracted from the power spectrum of the signal to obtain the denoised power spectrum.

[0087] (4)

[0088] in, P X ( k , m () represents the power spectrum of the noisy signal. α It is a factor used to improve noise suppression. β It is a small positive number used to avoid negative power spectra after spectral reduction.

[0089] S102: Perform an inverse Fourier transform on the denoised audio signal to transmit the signal back to the time domain and obtain the denoised continuous signal.

[0090] Specifically, the denoised signal is subjected to an inverse Fourier transform to transmit the signal back to the time domain.

[0091] (5)

[0092] After windowing, the signals from each frame are superimposed according to frame shift to obtain the denoised continuous signal:

[0093] (6)

[0094] (7)

[0095] In the formula, y ( t () represents the continuous signal after noise reduction. y w ( n , m The signal is the result of window function processing. Y ( k , m () is the frequency domain signal after noise suppression. L It is the window length. n and m These represent the indices for the window and the time period, respectively.

[0096] Denoising can effectively filter out the influence of environmental noise, retain the true battery swelling signal, and facilitate subsequent feature extraction and risk classification.

[0097] S103: Based on the denoised continuous signal, extract six features respectively: signal peak value, zero crossover rate, energy change rate, high frequency energy ratio, frequency entropy, and low frequency persistence.

[0098] In this embodiment, after denoising the signal, six features are extracted: signal peak value, zero crossover rate, energy change rate, high-frequency energy ratio, frequency entropy, and low-frequency persistence.

[0099] (1) The peak value of a signal is an important indicator that reflects the maximum amplitude of the signal. It can highlight the instantaneous and violent fluctuations in the signal and help to distinguish key abnormal changes from noise. It is believed that the higher the peak value, the greater the risk. High peak values ​​tend to expand rapidly, while relatively low peak values ​​tend to be slow and bulging signals.

[0100] Therefore, this embodiment uses the peak value to characterize the maximum transient response of the signal amplitude, which can effectively distinguish between the rapidly expanding signal with a peak transient and the slowly bulging signal with a low amplitude.

[0101] (8)

[0102] In the formula, X i For the first i Data points.

[0103] (2) The zero crossover rate is used to measure the oscillation frequency of a signal. For signals in dynamic processes such as battery expansion, the zero crossover rate can help identify rapid fluctuations in the signal, thereby reflecting the instantaneous changes during the expansion process. It can be used to distinguish between high-frequency continuous slow bulging signals and low-frequency peak transient signals; specifically:

[0104] (9)

[0105] In the formula, Ⅱ is the indicator function, when (x i x i+1 When ) < 0, take 1. x i x i+1 )>0 take 0.

[0106] (3) The rate of change of energy is used to reflect the degree of change of signal energy over time, and to characterize the rapid energy accumulation and release characteristics in peak transient signals:

[0107] (10)

[0108] In the formula, N The total number of data points in the signal. x i For the first i The value of each data point E i Is the signal at the 1st i The energy at each data point is represented as the square of the amplitude of that data point.

[0109] (4) The high-frequency energy ratio is used to describe the relative contribution of high-frequency components in a signal and is suitable for identifying slow, sustained high-frequency bulge signals:

[0110] (11)

[0111] in, f c This is the starting frequency of the high-frequency range. f max It is the highest frequency in the signal spectrum.

[0112] (5) Frequency entropy measures the degree of disorder in a signal's spectrum and can effectively reflect the distribution characteristics of different frequency components in a signal. By calculating the uncertainty of the signal's frequency distribution, it can identify the complexity in the signal. Especially when dealing with peak transient signals, frequency entropy can help distinguish between stationary and fluctuating signals. Signals with high frequency entropy usually mean that the frequency components are more disordered and variable, while low-frequency signals often show a more regular spectral distribution.

[0113] Therefore, it can be used to distinguish between peak transient signals (concentrated spectrum) and high-frequency continuous signals (dispersed spectrum):

[0114] (12)

[0115] In the formula, X ( f ) represents the signal at frequency f Spectral values ​​onP ( f ) indicates frequency f The probability density at that location.

[0116] Low-frequency persistence describes the significance of low-frequency components in a signal and is used to characterize the proportion of low-frequency components in a slow-bulging signal.

[0117] (13)

[0118] In the formula, f d This is the highest frequency in the low-frequency range. f max The highest frequency of the signal. X ( f ) represents the signal at frequency f The spectral component at a given frequency represents the energy of the signal at that frequency.

[0119] S104: Calculate the importance distance of each feature based on the eigenvalues ​​corresponding to each feature in the correlation coefficient matrix.

[0120] In this embodiment, principal component analysis is used to rank the importance of the six sound features.

[0121] Based on the eigenvalues ​​corresponding to each feature in the correlation coefficient matrix [ λ 1, λ 2, λ 3, λ 4, λ 5, λ 6] Define the importance distance (IMD) for each feature:

[0122] (14)

[0123] in, Representation of features In time t The value, n This represents the total time duration. express for The probability. The numerator of formula (14) is the normalized weight of the eigenvalues, which is the weight of the eigenvalues. λ i Convert to softmax ( λ i This avoids numerical instability caused by excessively large eigenvalues. The denominator of formula (14) is... A variant of information entropy, which can amplify the effects of stable sound features.

[0124] After calculating the IMD of six features, the five features with the highest IMD in each experimental cell are selected as the final features, thus achieving adaptive feature dimensionality reduction for different cells. Notably, this method achieves adaptive dynamic selection of thermal runaway features, employing different features at different times based on their effectiveness, avoiding interference from low-contribution features on the early warning effect. Furthermore, since the algorithm complexity increases sharply with the number of features, discarding low-contribution features can significantly reduce computational complexity without substantially affecting the algorithm's accuracy.

[0125] S105: Select the five features with the largest importance distance as the final features, and use the weighted combination number Bagging model to obtain the battery thermal runaway risk warning result.

[0126] Bagging is an ensemble learning method that reduces variance and improves classification accuracy through bootstrapping and model ensemble. The weighted combination number Bagging model in this embodiment is an ensemble model that uses a weighted strategy to aggregate the predictions of all base learners in Bagging.

[0127] During model training, for the six features before dimensionality reduction, three are selected as a base classifier for training, resulting in a total of... Base classifiers. In actual use, the model only needs to select 3 of the 5 features after dimensionality reduction as a base classifier, resulting in a total of... Seed base classifier.

[0128] The specific classification formula is as follows:

[0129] (15)

[0130] in, z This represents a sample with the three combined features. B This represents the total number of base classifiers. For the first b A base classifier evaluates samples z Category c For probability estimation, the base classifier can be composed of a decision tree and an activation function. This represents the importance distance of any of the three selected features. It represents the importance distance of any feature among all features.

[0131] In this embodiment, during the classification process, through By assigning weights to different features, the influence of features with larger IMD (Intense Thermal Displacement) is amplified. It can be observed that this method reduces the number of base classifiers in the algorithm through combined extraction, significantly alleviating the training burden. Furthermore, the low number of features improves the reliability of the base classifiers, and combined with weight allocation based on feature contribution, the decision-making process further focuses on high-quality features, avoiding interference from irrelevant features. The final classification results include normal operation, slow bulging, and rapid bulging. Slow bulging represents Level 1 warning, and rapid bulging represents Level 2 warning, thus characterizing different degrees of thermal runaway risk and achieving multi-level warning.

[0132] In this embodiment, when training the weighted combination number Bagging model, an experiment on the thermal runaway evolution process of the lithium-ion battery under test is conducted. The sound signals during the thermal runaway evolution process of the battery are collected and filtered. Based on the two types of bulging audio signals generated during the thermal runaway process of the battery: high-frequency short-time bulging sound and low-frequency continuous bulging sound, six features are extracted respectively: signal peak value, zero crossover rate, energy change rate, high-frequency energy ratio, frequency entropy and low-frequency persistence, to construct a training dataset.

[0133] To verify the effectiveness of this implementation method, the following experiments were conducted:

[0134] The 5C overcharge thermal runaway test is conducted as follows: the battery under test is first charged at a constant current of 5C to 4.4V, then allowed to rest briefly before being discharged at a constant current of 0.5C to 2.8V. This process is repeated multiple times until the battery experiences thermal runaway. The 120% SOC, 130% SOC, and 140% SOC overcharge thermal runaway tests are conducted as follows: the battery under test is first charged at a constant current of 0.5C to 120% SOC, 130% SOC, and 140% SOC, respectively. After a brief rest, it is discharged at a constant current of 0.5C to 2.8V. This process is repeated multiple times until the battery experiences thermal runaway.

[0135] Overcharge thermal runaway test at 120% SOC, 130% SOC and 140% SOC: The battery under test is first charged at a constant current of 0.5 C to 120% SOC, 130% SOC and 140% SOC, then allowed to stand for a short time and discharged at a constant current of 0.5 C to 2.8V. This cycle is repeated multiple times until the battery experiences thermal runaway.

[0136] Figure 2This refers to the audio signals emitted by a battery during a 5C overcharge cycle. Battery operating conditions are categorized into three different thermal runaway levels: Level 0, Level 1, and Level 2, corresponding to normal operation, slow bulging, and rapid expansion, respectively. During normal operation, there is no overheating or electrolyte decomposition, and the battery is stable. When the battery emits a continuous, high-frequency sound indicating slow bulging, it indicates that gas is gradually forming inside the battery, suggesting minor damage. This stage is classified as a Level 1 thermal runaway warning, indicating that battery monitoring should be intensified to prevent potential safety issues. Furthermore, when the battery emits a sharp, rapid expansion sound, it indicates violent expansion inside the battery, resulting in severe damage that could lead to structural failure or even more serious thermal runaway. This situation is classified as a Level 2 thermal runaway warning, at which point close monitoring of the battery's operating status is crucial. Therefore, classifying and grading the audio signals provides an effective basis for battery status monitoring and safety warnings, ensuring timely intervention.

[0137] Figure 3 This paper demonstrates the early warning results of the present invention in the normal signal of 5C over-rate operation. The method of this embodiment can successfully identify the normal operating condition signal without misdiagnosis. Figure 4 and Figure 5 The warning results for 120% SOC overcharge and 130% overcharge conditions are shown respectively. It can be seen that the present invention can accurately identify and classify the bulging signal in most periods, with only missed diagnosis and misdiagnosis at 16s and 13s respectively.

[0138] Figure 6 and Figure 7 The audio signals and diagnostic results for thermal runaway under two conditions—140% SOC overcharge thermal runaway and 5C overrate thermal runaway—were presented. Because the amplitude of slow bulging is small and its frequency is close to the noise frequency of the charging current, slow bulging can also occur during normal battery aging, leading to some misdiagnosis. Therefore, the actual warning only uses the slow bulging characteristic as an auxiliary indicator to remind relevant personnel to pay attention to the battery's operating status. Using a weighted combination number Bagging model, rapid expansion (i.e., Level 2 warning) was first identified as the warning time. The average thermal runaway warning time under all test conditions was 18726 seconds, further validating the effectiveness of this method in achieving early warning of thermal runaway.

[0139] Figure 8 The confusion matrix of the weighted combination number Bagging model for the prediction results of 2892s of data in the dataset is shown. The prediction accuracy of rapid expansion and slow bulging is 97.3% and 90.3% respectively, which are relatively high.

[0140] Therefore, this implementation method can accurately identify rapid bulging signals and maintain high accuracy in classifying slow expansion signals, further verifying its applicability in battery thermal runaway early warning tasks and demonstrating high engineering application value.

[0141] Example 2

[0142] In one or more embodiments, a graded early warning system for thermal runaway of lithium-ion batteries based on bulging sound signals is disclosed, comprising:

[0143] The data acquisition module is configured to acquire the sound signal sequence of the lithium-ion battery and perform noise reduction processing.

[0144] The signal denoising module is configured to perform an inverse Fourier transform on the denoised audio signal, transmit the signal back to the time domain, and obtain the denoised continuous signal.

[0145] The feature extraction module is configured to extract six features based on the denoised continuous signal: signal peak value, zero crossover rate, energy change rate, high-frequency energy ratio, frequency entropy, and low-frequency persistence.

[0146] The distance calculation module is configured to calculate the importance distance of each feature based on the feature values ​​corresponding to each feature in the correlation coefficient matrix.

[0147] The risk prediction module is configured to select the five features with the largest importance distance as the final features and use the weighted combination number Bagging model to obtain the battery thermal runaway risk warning result.

[0148] The specific implementation methods of the above modules are the same as those in Example 1, and will not be described in detail again.

[0149] Example 3

[0150] In one or more embodiments, a terminal device is disclosed, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions adapted to be loaded by the processor and executed by the processor to perform the lithium-ion battery thermal runaway graded early warning method based on bulging sound signals as described in Embodiment 1.

[0151] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0152] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0153] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0154] Example 4

[0155] In one or more embodiments, a computer-readable storage medium is disclosed, wherein a plurality of instructions are stored, the instructions being adapted to be loaded by a processor of a terminal device and executed by the lithium-ion battery thermal runaway graded early warning method based on bulging sound signals as described in Embodiment 1.

[0156] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A graded early warning method for thermal runaway of lithium-ion batteries based on bulging sound signals, characterized in that, include: Acquire the audio signal sequence of a lithium-ion battery and perform noise reduction processing; Performing an inverse Fourier transform on the denoised audio signal and transferring the signal back to the time domain yields the denoised continuous signal: Based on the denoised continuous signal, six features are extracted respectively: signal peak value, zero crossover rate, energy change rate, high frequency energy ratio, frequency entropy, and low frequency persistence. Calculate the importance distance of each feature based on its corresponding eigenvalue in the correlation coefficient matrix; The five features with the largest importance distance were selected as the final features, and the weighted combination number Bagging was used. The model yields early warning results for battery thermal runaway risk; Wherein, the signal peak The maximum transient response used to characterize the signal amplitude is specifically formulated as follows: ； in, X i For the first i One data point; The zero cross rate The formula used to measure the oscillation frequency of a signal is as follows: ； Where, Ⅱ is an indicator function, when ( x i x i+1 When ) < 0, take 1. x i x i+1 If )>0, take 0; N The total number of data points in the signal. x i For the first i The value of each data point; The rate of energy change The formula used to reflect the degree of change in signal energy over time is as follows: ； in, N The total number of data points in the signal. x i For the first i The value of each data point E i Is the signal at the 1st i The energy at each data point is represented as the square of the amplitude of that data point. The high-frequency energy ratio The formula used to describe the relative contribution of high-frequency components in a signal is as follows: ； in, f c This is the starting frequency of the high-frequency range. f max The highest frequency in the signal spectrum. X ( f ) is the signal at frequency f Spectral values ​​on; The frequency entropy measures the degree of disorder in the signal spectrum, specifically: ； in, X ( f ) is the signal at frequency f Spectral values ​​on P ( f ) indicates frequency f The probability density at that location; The low-frequency persistence describes the significance of low-frequency components in the signal, and the specific formula is as follows: ； in, f d This is the highest frequency in the low-frequency range. f max The highest frequency of the signal; Based on the eigenvalues ​​of each feature in the correlation coefficient matrix, the importance distance of each feature is calculated, specifically as follows: ； in, Representation of features In time t The value, n This represents the total time duration. express for The probability of; Representation of features eigenvalues.

2. The lithium-ion battery thermal runaway graded early warning method based on bulging sound signals as described in claim 1, characterized in that, The noise reduction process for the audio signal sequence of a lithium-ion battery is as follows: The lithium-ion battery audio signal sequence is divided into multiple frames, and the different frequency components in the signal are separated by fast Fourier transform to obtain the signal spectrum. Use silent segments to estimate the power spectrum of noise; The power spectrum after denoising is obtained by subtracting the power spectrum of noise from the power spectrum of the signal.

3. The lithium-ion battery thermal runaway graded early warning method based on bulging sound signals as described in claim 1, characterized in that, Using the weighted combination number Bagging algorithm, the battery thermal runaway risk warning result is obtained as follows: ； Three types are selected as a base classifier. z This represents a sample with the three combined features. B This represents the total number of base classifiers. For the first b A base classifier evaluates samples z Category c Probability estimation; This represents the importance distance of any of the three selected features. It represents the importance distance of any feature among all features.

4. The lithium-ion battery thermal runaway graded early warning method based on bulging sound signals as described in claim 1, characterized in that, The specific results of the battery thermal runaway risk warning are: normal operation, slow bulging, or rapid bulging.

5. A graded early warning system for thermal runaway of lithium-ion batteries based on bulging acoustic signals, used to implement the graded early warning method for thermal runaway of lithium-ion batteries based on bulging acoustic signals as described in claim 1, characterized in that, include: The data acquisition module is configured to acquire the sound signal sequence of the lithium-ion battery and perform noise reduction processing. The signal denoising module is configured to perform an inverse Fourier transform on the denoised audio signal, transmit the signal back to the time domain, and obtain the denoised continuous signal. The feature extraction module is configured to extract six features based on the denoised continuous signal: signal peak value, zero crossover rate, energy change rate, high-frequency energy ratio, frequency entropy, and low-frequency persistence. The distance calculation module is configured to calculate the importance distance of each feature based on the feature values ​​corresponding to each feature in the correlation coefficient matrix. The risk prediction module is configured to select the five features with the largest importance distance as the final features and use the weighted combination number Bagging model to obtain the battery thermal runaway risk warning result.

6. A terminal device comprising a processor and a memory, the processor for implementing instructions; the memory for storing multiple instructions, characterized in that, The instructions are adapted to be loaded by a processor and executed as described in any one of claims 1-4, the lithium-ion battery thermal runaway graded early warning method based on bulging sound signals.

7. A computer-readable storage medium storing a plurality of instructions, characterized in that, The instructions are adapted to be loaded and executed by the processor of the terminal device, and are based on the lithium-ion battery thermal runaway graded early warning method according to any one of claims 1-4.

Images