A method and system for monitoring battery mechanical damage during electric vehicle transportation based on acoustic emission spectrum analysis

Abstract

This invention discloses a method and system for monitoring mechanical damage to batteries during electric vehicle transportation based on acoustic emission spectrum analysis. The method includes: performing multi-scale decomposition of the original signal sequence using wavelet transform to separate high-frequency transient components and low-frequency background noise, obtaining a denoised elastic wave signal; calculating time-domain features, including peak amplitude and duration, based on the denoised elastic wave signal, and combining this with frequency-domain features such as the dominant frequency component to obtain a comprehensive feature vector; if the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is judged as a potential damage event, and damage-related subsets are extracted from the feature vector to obtain damage candidate features; training the damage candidate features using a support vector machine classifier to obtain damage type labels; and fusing transportation environment data with the damage type labels, using sliding window analysis to track signal change trends and determine the degree of damage evolution. This invention achieves accurate identification, classification, and dynamic monitoring of battery pack transportation damage.

Description

Technical Field

[0001] This invention belongs to the field of information technology, specifically relating to a method and system for monitoring mechanical damage to batteries during the transportation of electric vehicles based on acoustic emission spectrum analysis. Background Technology

[0002] As the core of clean energy transportation, the safety and reliability of battery packs in electric vehicles are crucial to the industry's development. Battery packs can suffer mechanical damage during transportation due to external forces such as bumps and collisions, affecting not only battery performance but also potential safety hazards. Ensuring the structural integrity of battery packs in complex transportation environments is a critical issue that the electric vehicle industry urgently needs to address. Currently, battery pack damage monitoring technologies largely rely on manual inspection or simple vibration sensors. These methods struggle to capture minute damage in real time and cannot accurately distinguish damage types, leading to the overlooking of potential risks. The main drawback of existing monitoring methods is that they often only detect visible damage, such as obvious deformation of the outer shell, while lacking the ability to effectively identify latent damage such as internal microcracks or loose connections. This limitation stems from insufficient extraction of signal characteristics by monitoring methods, making it difficult to accurately reflect the complex physical changes inside the battery pack. Furthermore, traditional methods typically require shutdown for inspection, making them unsuitable for the dynamic and continuous monitoring needs during transportation. The core technical challenge lies in how to capture transient elastic wave signals induced by external forces during transportation and extract effective information reflecting damage characteristics from them. Transient elastic wave signals, due to their short and complex characteristics, are highly susceptible to environmental noise interference, leading to blurred signal features and making accurate analysis difficult. For example, during transport, minor fractures may occur in the electrode plates inside the battery pack. The signals generated by this damage are weak and mixed, making the extraction of key features a primary challenge. This challenge further leads to another technical difficulty: how to quickly and accurately identify damage types in dynamic environments, such as distinguishing between outer shell cracks and loose internal welds. Because the signal characteristics of different damage types differ subtly, traditional analysis methods struggle to achieve efficient classification. Therefore, how to acquire and analyze transient elastic wave signals in real time during transport to accurately distinguish different types of mechanical damage inside the battery pack has become a critical issue that urgently needs to be addressed. Only by solving this problem can the safety and reliability of battery packs during transport be ensured, providing technical support for the large-scale development of the electric vehicle industry. Failure to detect such latent damage may lead to performance degradation of the battery pack in subsequent use, or even serious consequences such as short circuits or thermal runaway. For example, during long-distance transport, continuous bumps may cause internal welds to loosen in the battery pack. This damage is difficult to detect initially but may cause unstable current during vehicle operation, affecting the safety of the entire vehicle. Current technologies are unable to distinguish the signal characteristics of these subtle damages in real time during dynamic transportation scenarios, leading to problems being overlooked and eventually causing larger failures. Therefore, how to capture and analyze transient elastic wave signals in real time during transportation in a non-invasive manner to accurately identify different types of mechanical damage inside the battery pack has become a key issue in the safety management of electric vehicle batteries. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention provides a method and system for monitoring battery mechanical damage during electric vehicle transportation based on acoustic emission spectrum analysis. Through the efficient integration of signal processing and machine learning, it achieves accurate identification, classification, and dynamic monitoring of battery pack transportation damage, significantly improving transportation safety and equipment reliability.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for monitoring battery mechanical damage during electric vehicle transportation based on acoustic emission spectrum analysis, the method comprising:

[0006] Transient elastic wave signals generated during transportation are collected by an array of piezoelectric sensors deployed on the battery pack. A synchronous triggering mechanism is used to capture instantaneous vibration data caused by bumps and collisions, and the original signal sequence is obtained.

[0007] Wavelet transform is used to perform multi-scale decomposition on the original signal sequence to separate high-frequency transient components and low-frequency background noise, resulting in a denoised elastic wave signal.

[0008] The time-domain features are calculated based on the denoised elastic wave signal, and combined with the frequency-domain features to obtain a comprehensive feature vector;

[0009] If the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is judged as a potential damage event, and then damage-related subsets are extracted from the feature vector to obtain damage candidate features.

[0010] Damage type labels are obtained by training the damage candidate features using a support vector machine classifier.

[0011] By integrating damage type labels with transportation environment data, a sliding window analysis is used to track signal change trends and determine the degree of damage evolution.

[0012] An alarm signal is generated based on the degree of damage evolution. If the alarm signal lasts for more than a set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions.

[0013] Preferably, the method of using wavelet transform to perform multi-scale decomposition of the original signal sequence to separate high-frequency transient components and low-frequency background noise to obtain the denoised elastic wave signal includes:

[0014] Wavelet transform is used to perform multi-scale decomposition on the original signal sequence to obtain high-frequency transient components and low-frequency background components, thus obtaining the decomposed signal components.

[0015] Based on the decomposed signal components, calculate the energy characteristics of the high-frequency transient components and determine the significance of the transient components;

[0016] If the energy characteristics of a high-frequency transient component exceed a preset threshold, the component is retained; otherwise, it is attenuated to obtain an adjusted high-frequency component.

[0017] For the low-frequency background component, Fourier transform is used to extract its frequency characteristics and determine the distribution characteristics of low-frequency background noise.

[0018] If the frequency characteristics of the low-frequency background component conform to the preset noise model, the low-frequency component is smoothed by a filter to obtain the smoothed low-frequency component.

[0019] By combining the adjusted high-frequency components and the smoothed low-frequency components, the signal sequence is reconstructed to obtain the denoised elastic wave signal.

[0020] Based on the denoised elastic wave signal, its signal-to-noise ratio is calculated, and it is determined whether the signal quality meets the preset requirements to obtain the final signal processing result.

[0021] Preferably, the method for calculating the time-domain features based on the denoised elastic wave signal and combining them with the frequency-domain features to obtain the comprehensive feature vector includes:

[0022] The denoised elastic wave signal is obtained, and the peak amplitude and duration are calculated through time-domain analysis to obtain the time-domain feature set;

[0023] Based on the time-domain feature set, the frequency-domain features are extracted using the Fast Fourier Transform to determine the dominant frequency component;

[0024] From the main frequency components, the main frequency point with the highest amplitude is obtained to obtain the frequency domain feature set;

[0025] For time-domain feature sets and frequency-domain feature sets, if the feature dimensions are consistent, the features are fused by concatenating vectors to generate a comprehensive feature vector;

[0026] If the dimensionality of the comprehensive feature vector is too high, principal component analysis is used to reduce the dimensionality and obtain an optimized feature vector.

[0027] Based on the optimized feature vectors, the signal category is determined by the support vector machine classification, and the classification result is obtained.

[0028] From the classification results, the signal category labels are extracted to generate the final feature description.

[0029] Preferably, if the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is determined to be a potential damage event. The method for extracting a damage-related subset from the feature vector to obtain damage candidate features includes:

[0030] If the peak amplitude of the comprehensive feature vector exceeds the preset threshold, it is determined to be a potential damage event through amplitude analysis, and a preliminary damage judgment result is obtained.

[0031] From the preliminary damage assessment results, a damage-related subset is separated using feature extraction methods to obtain candidate damage features;

[0032] Based on the candidate damage features, the support vector machine algorithm is used for classification to determine the damage type and obtain the damage classification result.

[0033] Based on the damage classification results, cluster analysis was used to group the candidate damage features, resulting in damage feature groups.

[0034] If the damage feature grouping meets the preset damage pattern matching conditions, the damage location is determined by pattern recognition technology to obtain the damage localization result.

[0035] From the damage localization results, spatial distribution information related to the damage is extracted to obtain the damage spatial features;

[0036] By using the spatial characteristics of the damage, a decision tree algorithm is employed to assess the severity of the damage and obtain the damage severity level.

[0037] Preferably, methods for obtaining damage type labels by training a support vector machine classifier on damage candidate features include:

[0038] Obtain the structural acceleration response at the measurement point;

[0039] Calculate the transfer function between structural acceleration responses at adjacent measurement points;

[0040] Calculate the normalized wavelet packet energy impairment feature and select the first value of the transfer function amplitude. The principal components constitute one-dimensional damage characteristics;

[0041] Normalized wavelet packet energy damage features and one-dimensional damage features are used as sample sets, and damage location is used as the corresponding sample label. A support vector machine classification algorithm is used to identify the damage location. Normalized wavelet packet energy damage features and one-dimensional damage features are used as sample sets, and damage degree is used as the corresponding sample label. A support vector machine regression algorithm is used to identify the damage degree.

[0042] Preferably, the method for determining the degree of damage evolution by fusing damage type labels with transportation environment data and using sliding window analysis to track signal change trends includes:

[0043] By fusing multi-source features, battery status data and environmental parameters are integrated to construct a spatiotemporally complete battery damage dataset.

[0044] By utilizing big data analytics algorithms, we can uncover the cross-influence of various factors and establish a big data analytics model that simultaneously characterizes the dynamic process over time and the superimposed effects of multiple causes.

[0045] The sliding time window technique is used to analyze the trend of damage data over a continuous time period. The least squares method is used to fit the damage evolution curve. The damage development rate and acceleration are calculated based on the slope and second derivative of the curve. The damage evolution law parameters are obtained for predictive analysis.

[0046] Preferred big data analysis models that simultaneously characterize the dynamic process over time and the superimposed influence of multiple causes include: the spatiotemporal geographic weighted regression model GTWR and the dependent variable spatiotemporal geographic weighted regression model;

[0047] The GTWR model incorporates the spatiotemporal characteristics of the data into the regression model, addressing the non-stationarity issues in both space and time. Its calculation formula is as follows:

[0048] ;

[0049] in, For the first The dependent variable for each sample point; It is the first The spatiotemporal coordinates of each sample point; In the first The first sample point at the sample point One independent variable value; It is the first The intercept of the regression equation for each sample point; In the first The first sample point at the sample point regression coefficients of the independent variables; For random error term, The number of sample points;

[0050] The multi-dependent variable spatiotemporal geographic weighted regression model further extends the GTWR model by considering multiple dependent variables. Its formula is as follows:

[0051] ;

[0052] in, Indicates the first The first sample point One dependent variable; It is the first The dependent variable in the th ... Intercept term for each sample point; It is the first The dependent variable in the th... The first sample point The regression coefficients of the independent variables; It is a random error term. This represents the number of sample points.

[0053] This invention also provides a battery mechanical damage monitoring system for electric vehicles during transportation based on acoustic emission spectrum analysis. The system is used to implement the aforementioned method and includes: a signal acquisition module, a signal processing module, a feature extraction module, an event judgment module, a damage classification module, an evolution analysis module, and an alarm generation module.

[0054] The signal acquisition module is used to acquire transient elastic wave signals generated during transportation through a piezoelectric sensor array deployed on the battery pack, and to capture instantaneous vibration data caused by bumps and collisions using a synchronous triggering mechanism to obtain the original signal sequence.

[0055] The signal processing module is used to perform multi-scale decomposition on the original signal sequence using wavelet transform, separating high-frequency transient components and low-frequency background noise to obtain a denoised elastic wave signal.

[0056] The feature extraction module is used to calculate time-domain features based on the denoised elastic wave signal and combine them with frequency-domain features to obtain a comprehensive feature vector.

[0057] The event judgment module is used to determine a potential damage event if the peak amplitude of the comprehensive feature vector exceeds a preset threshold, and then extract a damage-related subset from the feature vector to obtain damage candidate features.

[0058] The damage classification module is used to train the damage candidate features using a support vector machine classifier to obtain damage type labels.

[0059] The evolution analysis module is used to fuse transportation environment data with damage type labels, and to use sliding window analysis to track signal change trends and determine the degree of damage evolution.

[0060] The alarm generation module is used to generate alarm signals based on the degree of damage evolution. If the duration of the alarm signal exceeds a set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions.

[0061] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0062] This invention discloses a method for acquiring transient elastic wave signals during battery pack transportation using a piezoelectric sensor array. This addresses the challenge of real-time monitoring and classification of battery pack damage during transport. Specifically, it tackles the difficulty in accurately identifying and tracking the evolution of potential damage caused by bumps and collisions. A synchronous triggering mechanism is employed to capture vibration data. Wavelet transform is used for multi-scale decomposition to remove noise, extracting combined time and frequency domain feature vectors. A support vector machine classifier is then used to distinguish damage types such as outer shell cracks and loose internal welds. A sliding window analysis is used to integrate transportation environment data to track damage evolution trends. When the damage level exceeds a certain threshold, a remote transmission module is activated to send an alarm to the monitoring center, enabling real-time intervention. This invention achieves accurate identification, classification, and dynamic monitoring of battery pack transportation damage through the efficient integration of signal processing and machine learning, significantly improving transportation safety and equipment reliability. Attached Figure Description

[0063] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a schematic diagram of a method for monitoring battery mechanical damage during electric vehicle transportation based on acoustic emission spectrum analysis, according to an embodiment of the present invention.

[0065] Figure 2 This is a schematic diagram illustrating the classification of data using a Support Vector Machine (SVM) according to an embodiment of the present invention. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0068] Example 1

[0069] like Figure 1 As shown, this invention provides a method for monitoring battery mechanical damage during electric vehicle transportation based on acoustic emission spectrum analysis, the method comprising:

[0070] Transient elastic wave signals generated during transportation are collected by an array of piezoelectric sensors deployed on the battery pack. A synchronous triggering mechanism is used to capture instantaneous vibration data caused by bumps and collisions, and the original signal sequence is obtained.

[0071] Wavelet transform is used to perform multi-scale decomposition on the original signal sequence to separate high-frequency transient components and low-frequency background noise, resulting in a denoised elastic wave signal.

[0072] The time-domain features are calculated based on the denoised elastic wave signal, and combined with the frequency-domain features to obtain a comprehensive feature vector;

[0073] If the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is judged as a potential damage event, and then damage-related subsets are extracted from the feature vector to obtain damage candidate features.

[0074] Damage type labels are obtained by training the damage candidate features using a support vector machine classifier.

[0075] By integrating damage type labels with transportation environment data, a sliding window analysis is used to track signal change trends and determine the degree of damage evolution.

[0076] An alarm signal is generated based on the degree of damage evolution. If the alarm signal lasts for more than a set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions.

[0077] In this embodiment, the method for acquiring the original signal sequence by collecting transient elastic wave signals generated during transportation through a piezoelectric sensor array deployed on the battery pack and capturing instantaneous vibration data caused by bumps and collisions using a synchronous triggering mechanism includes:

[0078] The piezoelectric sensor is bonded to key locations on the battery pack, such as the outer surface or internal structural components, to ensure that the sensor can effectively detect vibrations experienced by the battery pack during transportation. The sensor placement should take into account the structural characteristics and stress conditions of the battery pack to obtain comprehensive and accurate vibration signals.

[0079] Utilizing the piezoelectric effect of piezoelectric sensors, when a battery pack is subjected to bumps or impacts, the sensor generates an electrical charge signal due to the force applied. These signals reflect the vibration of the battery pack. The data acquisition equipment needs to have a high sampling rate and high-precision analog-to-digital conversion capabilities to ensure accurate capture of transient elastic wave signals.

[0080] Choose a suitable trigger signal source, such as an accelerometer or a laser sensor. When the battery pack is subjected to bumps or impacts, the trigger signal source will detect the vibration and generate a trigger signal, which is used to synchronously start the data acquisition of the piezoelectric sensor array.

[0081] A synchronous triggering mechanism ensures that all piezoelectric sensors begin acquiring data at the same time, thereby achieving synchronous capture of instantaneous vibration data. This requires the acquisition system to have precise time synchronization capabilities to guarantee the consistency of data acquired by each sensor in time.

[0082] In this embodiment, the method of using wavelet transform to perform multi-scale decomposition on the original signal sequence to separate high-frequency transient components and low-frequency background noise, and obtaining the denoised elastic wave signal includes:

[0083] Wavelet transform is used to perform multi-scale decomposition on the original signal sequence to obtain high-frequency transient components and low-frequency background components, thus obtaining the decomposed signal components.

[0084] Based on the decomposed signal components, calculate the energy characteristics of the high-frequency transient components and determine the significance of the transient components;

[0085] If the energy characteristics of a high-frequency transient component exceed a preset threshold, the component is retained; otherwise, it is attenuated to obtain an adjusted high-frequency component.

[0086] For the low-frequency background component, Fourier transform is used to extract its frequency characteristics and determine the distribution characteristics of low-frequency background noise.

[0087] If the frequency characteristics of the low-frequency background component conform to the preset noise model, the low-frequency component is smoothed by a filter to obtain the smoothed low-frequency component.

[0088] By combining the adjusted high-frequency components and the smoothed low-frequency components, the signal sequence is reconstructed to obtain the denoised elastic wave signal.

[0089] Based on the denoised elastic wave signal, its signal-to-noise ratio is calculated, and it is determined whether the signal quality meets the preset requirements to obtain the final signal processing result.

[0090] In this embodiment, the method for calculating the time-domain features based on the denoised elastic wave signal and combining them with the frequency-domain features to obtain the comprehensive feature vector includes:

[0091] The denoised elastic wave signal is obtained, and the peak amplitude and duration are calculated through time-domain analysis to obtain the time-domain feature set;

[0092] Based on the time-domain feature set, the frequency-domain features are extracted using the Fast Fourier Transform to determine the dominant frequency component;

[0093] From the main frequency components, the main frequency point with the highest amplitude is obtained to obtain the frequency domain feature set;

[0094] For time-domain feature sets and frequency-domain feature sets, if the feature dimensions are consistent, the features are fused by concatenating vectors to generate a comprehensive feature vector;

[0095] If the dimensionality of the comprehensive feature vector is too high, principal component analysis is used to reduce the dimensionality and obtain an optimized feature vector.

[0096] Based on the optimized feature vectors, the signal category is determined by the support vector machine classification, and the classification result is obtained.

[0097] From the classification results, the signal category labels are extracted to generate the final feature description.

[0098] In this embodiment, if the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is determined to be a potential damage event. The method for extracting a damage-related subset from the feature vector to obtain damage candidate features includes:

[0099] If the peak amplitude of the comprehensive feature vector exceeds the preset threshold, it is determined to be a potential damage event through amplitude analysis, and a preliminary damage judgment result is obtained.

[0100] From the preliminary damage assessment results, a damage-related subset is separated using feature extraction methods to obtain candidate damage features;

[0101] Based on the candidate damage features, the support vector machine algorithm is used for classification to determine the damage type and obtain the damage classification result.

[0102] Based on the damage classification results, cluster analysis was used to group the candidate damage features, resulting in damage feature groups.

[0103] If the damage feature grouping meets the preset damage pattern matching conditions, the damage location is determined by pattern recognition technology to obtain the damage localization result.

[0104] From the damage localization results, spatial distribution information related to the damage is extracted to obtain the damage spatial features;

[0105] By using the spatial characteristics of the damage, a decision tree algorithm is employed to assess the severity of the damage and obtain the damage severity level.

[0106] In this embodiment, the method for obtaining damage type labels by training a support vector machine classifier on damage candidate features includes:

[0107] Damage features are constructed by calculating the transfer function of the structure, and then the SVM classification algorithm is used to identify the damage location, while the SVM regression algorithm is used to identify the degree of damage.

[0108] In this invention, the transfer function (TF) is used as the primary data for damage identification. The transfer function is the ratio of the frequency response functions (FRF) at two spatial locations on the structure and can be estimated in a nonparametric preprocessing step.

[0109] Assuming the input degrees of freedom (DOF) There is a single force at that point The transfer function is:

[0110] ;

[0111] in and They are respectively in degrees of freedom and degrees of freedom The Fourier transform of the measured response, and and They are degrees of freedom and degrees of freedom Elements of the frequency response function, ω is the angular frequency.

[0112] It can be seen that when a structure is subjected to a single-point excitation, the transfer function depends only on the location of the force, and not on the nature or amplitude of the force signal. This eliminates the influence of the input signal (excitation force).

[0113] The transfer function is highly sensitive to damage because it is a ratio of functions with many peaks and troughs. Since the dynamic response of a structure changes when damage occurs, it is natural to expect that the transfer function will also change. Changes in the transfer function represent changes in structural properties, and it can be inferred that changes in the transfer function indicate damage to the structure.

[0114] Wavelet analysis is a multi-resolution analysis method for signals in both the time and frequency domains, and it has the ability to describe the local features of signals. Wavelet packet decomposition (WPD) can be considered an extension of discrete wavelet analysis. It does not require prior knowledge of the frequency band, but instead filters the signal adaptively. Discrete wavelet analysis only decomposes the approximation coefficients of the previous stage using high-pass and low-pass filters, while wavelet packet decomposition (WPD) decomposes both detail coefficients and approximation coefficients simultaneously, allowing important information in high-frequency components to be preserved.

[0115] The original signal can be represented as the sum of wavelet packet decomposition (WPD) components, that is:

[0116] ;

[0117] in It's a time lag. It is a wavelet packet decomposition (WPD) component signal, which can be represented by a linear combination of wavelet packet functions.

[0118] To examine the health of a structure, it is crucial to obtain indicators sensitive to structural damage. Typically, the measured vibration signal is decomposed into component signals using wavelet packet decomposition (WPD), and then the component energy is calculated. Wavelet packet component energy is a suitable tool for identifying and characterizing specific phenomena in the time-frequency domain.

[0119] The wavelet packet component energy of a signal is defined as:

[0120] ;

[0121] When the mother wavelet is orthogonal, the total signal energy can be expressed as the sum of the wavelet packet component energies:

[0122] ;

[0123] Then, the energy ratio of each wavelet coefficient can be written as:

[0124] ;

[0125] in, The value corresponds to a specific coefficient. The ratio of energy to total energy.

[0126] The normalized wavelet packet component energy vector of a signal is defined as:

[0127] ;

[0128] When the structure is damaged in different locations The value of the wavelet packet energy vector (WPEV) will differ from that of the wavelet packet energy vector; therefore, in this invention, WPEV can be used as a damage feature to identify damage to the structure.

[0129] Damage characteristics based on principal component analysis (PCA):

[0130] Principal Component Analysis (PCA) is a multivariate analysis method that can be considered a linear data compression technique, widely used in image processing, flow visualization, and order reduction models. Through orthogonal projection, The original set of variables in dimensional space is transformed into A new set of uncorrelated variables in dimensional space, the so-called principal components, such that... .

[0131] Under the assumption of a stationary random process, the time-frequency (TF) amplitude is used as the analysis signal and it is divided into... Groups, each group has Several frequency points form a matrix. Therefore, the centered matrix can be obtained using the following formula. :

[0132] ;

[0133] in, It is a matrix The List, yes The mean.

[0134] The correlation matrix can be defined as:

[0135] ;

[0136] By definition, principal components are correlation matrices. Eigenvalues ​​and related eigenvectors:

[0137] ;

[0138] in, and They are the first Eigenvalues ​​and eigenvectors. Normalized matrix. The principal components are given by the following formula:

[0139] ;

[0140] in, It is the first There are one principal component. Any two principal components are uncorrelated, and The variance is .because The first principal component, namely the highest eigenvalue and its associated eigenvector, represents the direction and magnitude of the greatest variability in the original data. The next principal component, orthogonal to the first component, represents the second most important contribution to the original data, and so on.

[0141] The contribution rate is given by the following formula:

[0142] ;

[0143] Contribution rate The larger the value, the stronger the ability of the corresponding principal components to synthesize the original data information.

[0144] The cumulative contribution rate is given by the following formula:

[0145] ;

[0146] To improve the reliability of the selected principal components, the cumulative contribution rate It should be greater than 85%. (From the previous...) The damage feature PC composed of principal components can be formed as follows:

[0147] ;

[0148] Generally, the first few principal components contain most of the information; analyzing these components can reflect the overall variation characteristics of the original data. Damage identification can be achieved in a low-dimensional space composed of the first few principal components, thus reducing the analytical difficulty.

[0149] By applying kernel functions (such as linear, polynomial, or radial basis functions) as inner products of functions that map data to a high-dimensional feature space, support vector machines can also be applied to nonlinear classification and regression analysis.

[0150] The entire process of support vector machines classifying data is as follows: Figure 2 As shown. The training dataset is considered as... Each of them Describes the input space of the sample, and for There are corresponding target values ,in This corresponds to the size of the training data.

[0151] If the training data is linearly separable, then the boundary can be represented as follows:

[0152] ;

[0153] Where vector Define boundaries, It is a dimension of The input vector, It is a scalar threshold, and ( ) represents the dot product. At the boundary where the support vectors lie, the equations for classes A and B are as follows:

[0154] ;

[0155] ;

[0156] For inseparable data, the soft boundary of a support vector machine is defined as:

[0157] ;

[0158] And the optimized solution will be:

[0159] ;

[0160] ;

[0161] in, It is a slack variable. This is an error penalty. By solving the dual optimization problem using the Lagrange multiplier algorithm, the nonlinear decision function will produce:

[0162] ;

[0163] ;

[0164] ;

[0165] in, The number of training samples. These are Lagrange multipliers (dual variables). For the input sample, Indicates the first training samples, These are the parameters of the kernel function. It is a radial basis function, which is defined as a kernel function. By using this kernel function, high-dimensional data can be analyzed.

[0166] The main idea behind regression problems is to determine an approximation that accurately, or at least closely, approaches the expected value to determine or predict future target values. Generally, the linear estimate of a regression function takes the form:

[0167] ;

[0168] in The goal is to find and The value of is such that it can be determined by minimizing the regression risk. Value:

[0169] ;

[0170] ;

[0171] ;

[0172] ;

[0173] in It's about precision. , It is a slack variable. It is a constant. For the target value, This represents the number of training samples.

[0174] By using the Lagrange multiplier algorithm to solve the dual optimization problem, the nonlinear decision function will produce...

[0175] ;

[0176] ;

[0177] where ;

[0178] in For Lagrange multipliers, This represents the number of training samples.

[0179] This invention proposes a novel structural damage identification method based on transfer function (TF) and support vector machine (SVM). Two damage features (wavelet packet energy damage value (WPEV) and principal component (PC)) are used as inputs to the SVM to identify the damage location and degree. The proposed strategy is based on the following four steps:

[0180] (1) Obtain the structural acceleration response at the measurement point;

[0181] (2) Calculate the transfer function (TF): Calculate the transfer function between the acceleration responses at adjacent measurement points and use it as the study signal;

[0182] (3) Constructing Damage Features (WPEV, PC): Calculate the normalized wavelet packet energy damage feature (WPEV). Similarly, select the first value of the transfer function amplitude... The cumulative contribution rate of the principal components is greater than that of the principal components. To form a one-dimensional damage feature (PC);

[0183] (4) Structural damage identification based on damage features (WPEV, PC) and support vector machine: The two damage features (WPEV, PC) are used as sample sets, and the damage location is used as the corresponding sample label; the classification algorithm of support vector machine is used to identify the damage location. Then, the two damage features (WPEV, PC) are used as sample sets, and the damage degree is used as the corresponding sample label; the regression algorithm of support vector machine is used to identify the damage degree.

[0184] In this embodiment, the method for determining the degree of damage evolution by fusing damage type labels with transportation environment data and using sliding window analysis to track signal change trends includes:

[0185] By fusing multi-source features, battery status data is integrated with environmental parameters (such as weather information, temperature, humidity, etc.) to construct a spatiotemporally complete battery damage dataset.

[0186] By utilizing big data analytics algorithms, we can uncover the cross-influence of various factors and establish a big data analytics model that simultaneously characterizes the dynamic process over time and the superimposed effects of multiple causes.

[0187] The sliding time window technique is used to analyze the trend of damage data over a continuous time period. The least squares method is used to fit the damage evolution curve. The damage development rate and acceleration are calculated based on the slope and second derivative of the curve. The damage evolution law parameters are obtained for predictive analysis.

[0188] Among them, big data analysis models that simultaneously characterize the dynamic process of time dimension and the superimposed influence of multiple causes include: the spatiotemporal geographic weighted regression model (GTWR) and the dependent variable spatiotemporal geographic weighted regression model;

[0189] The GTWR model incorporates the spatiotemporal characteristics of the data into the regression model, addressing the issues of spatial and temporal nonstationarity. Its calculation formula is as follows:

[0190] ;

[0191] in, For the first The dependent variable for each sample point; It is the first The spatiotemporal coordinates of each sample point; In the first The first sample point at the sample point One independent variable value; It is the first The intercept of the regression equation for each sample point; In the first The first sample point at the sample point regression coefficients of the independent variables; This is the random error term.

[0192] The multi-dependent variable spatiotemporal geo-weighted regression model further extends the GTWR model (geographically weighted regression model) by considering multiple dependent variables, thus providing a more comprehensive characterization of spatiotemporal dynamic processes and multiple factors.

[0193] Due to the cumulative effect, the formula is:

[0194] ;

[0195] in, Indicates the first The first sample point One dependent variable; It is the first The dependent variable in the th... Intercept term for each sample point; It is the first The dependent variable in the th... The first sample point The regression coefficients of the independent variables; It is a random error term.

[0196] In this embodiment, an alarm signal is generated based on the degree of damage evolution. If the duration of the alarm signal exceeds a set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions. The method includes:

[0197] If the degree of damage evolution exceeds a preset threshold, the degree of damage evolution is determined by a comparison algorithm to obtain a trigger signal;

[0198] Based on the trigger signal, an alarm signal is generated using a signal generation module, and the alarm signal status is determined.

[0199] The duration of the alarm signal is obtained through the timing module to determine whether the duration exceeds the set period;

[0200] If the duration exceeds the set period, an alarm signal will be sent to the monitoring center via the remote transmission module to obtain transmission confirmation;

[0201] Based on the transmission confirmation, obtain real-time intervention instructions from the monitoring center and determine the content of the instructions;

[0202] The real-time intervention commands are processed by the command parsing module to obtain the execution parameters;

[0203] Based on the execution parameters, the control module adjusts the system's operating status to achieve real-time intervention.

[0204] Example 2

[0205] The present invention also provides a battery mechanical damage monitoring system for electric vehicles during transportation based on acoustic emission spectrum analysis. The system is used to implement the method described in Embodiment 1. The system includes: a signal acquisition module, a signal processing module, a feature extraction module, an event judgment module, a damage classification module, an evolution analysis module, and an alarm generation module.

[0206] The signal acquisition module is used to acquire transient elastic wave signals generated during transportation through a piezoelectric sensor array deployed on the battery pack. It uses a synchronous triggering mechanism to capture instantaneous vibration data caused by bumps and collisions to obtain the original signal sequence.

[0207] The signal processing module is used to perform multi-scale decomposition of the original signal sequence using wavelet transform, separating high-frequency transient components and low-frequency background noise to obtain the denoised elastic wave signal.

[0208] The feature extraction module is used to calculate the time-domain features based on the denoised elastic wave signal and combine them with the frequency-domain features to obtain a comprehensive feature vector.

[0209] The event judgment module is used to judge a potential damage event if the peak amplitude of the comprehensive feature vector exceeds a preset threshold, and then extract the damage-related subset from the feature vector to obtain damage candidate features.

[0210] The damage classification module is used to train the damage candidate features using a support vector machine classifier to obtain damage type labels.

[0211] The evolution analysis module is used to fuse transportation environment data with damage type labels, and to use sliding window analysis to track signal change trends and determine the degree of damage evolution.

[0212] The alarm generation module is used to generate alarm signals based on the degree of damage evolution. If the duration of the alarm signal exceeds the set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions.

[0213] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for monitoring mechanical damage to batteries during the transportation of electric vehicles based on acoustic emission spectrum analysis, characterized in that, The method includes: Transient elastic wave signals generated during transportation are collected by an array of piezoelectric sensors deployed on the battery pack. A synchronous triggering mechanism is used to capture instantaneous vibration data caused by bumps and collisions, and the original signal sequence is obtained. Wavelet transform is used to perform multi-scale decomposition on the original signal sequence to separate high-frequency transient components and low-frequency background noise, resulting in a denoised elastic wave signal. The time-domain features are calculated based on the denoised elastic wave signal, and combined with the frequency-domain features to obtain a comprehensive feature vector; If the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is judged as a potential damage event, and then damage-related subsets are extracted from the feature vector to obtain damage candidate features. Damage type labels are obtained by training the damage candidate features using a support vector machine classifier. By integrating damage type labels with transportation environment data, a sliding window analysis is used to track signal change trends and determine the degree of damage evolution. An alarm signal is generated based on the degree of damage evolution. If the alarm signal lasts for more than a set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions.

2. The method according to claim 1, characterized in that, Methods for using wavelet transform to perform multi-scale decomposition of the original signal sequence, separating high-frequency transient components and low-frequency background noise, to obtain the denoised elastic wave signal include: Wavelet transform is used to perform multi-scale decomposition on the original signal sequence to obtain high-frequency transient components and low-frequency background components, thus obtaining the decomposed signal components. Based on the decomposed signal components, calculate the energy characteristics of the high-frequency transient components and determine the significance of the transient components; If the energy characteristics of a high-frequency transient component exceed a preset threshold, the component is retained; otherwise, it is attenuated to obtain an adjusted high-frequency component. For the low-frequency background component, Fourier transform is used to extract its frequency characteristics and determine the distribution characteristics of low-frequency background noise. If the frequency characteristics of the low-frequency background component conform to the preset noise model, the low-frequency component is smoothed by a filter to obtain the smoothed low-frequency component. By combining the adjusted high-frequency components and the smoothed low-frequency components, the signal sequence is reconstructed to obtain the denoised elastic wave signal. Based on the denoised elastic wave signal, its signal-to-noise ratio is calculated, and it is determined whether the signal quality meets the preset requirements to obtain the final signal processing result.

3. The method according to claim 1, characterized in that, Methods for calculating time-domain features from denoised elastic wave signals and combining them with frequency-domain features to obtain a comprehensive feature vector include: The denoised elastic wave signal is obtained, and the peak amplitude and duration are calculated through time-domain analysis to obtain the time-domain feature set; Based on the time-domain feature set, the frequency-domain features are extracted using the Fast Fourier Transform to determine the dominant frequency component; From the main frequency components, the main frequency point with the highest amplitude is obtained to obtain the frequency domain feature set; For time-domain feature sets and frequency-domain feature sets, if the feature dimensions are consistent, the features are fused by concatenating vectors to generate a comprehensive feature vector; If the dimensionality of the comprehensive feature vector is too high, principal component analysis is used to reduce the dimensionality and obtain an optimized feature vector. Based on the optimized feature vectors, the signal category is determined by the support vector machine classification, and the classification result is obtained. From the classification results, the signal category labels are extracted to generate the final feature description.

4. The method according to claim 1, characterized in that, If the peak amplitude of the comprehensive feature vector exceeds a preset threshold, it is judged as a potential damage event. Then, a damage-related subset is extracted from the feature vector to obtain damage candidate features. Methods include: If the peak amplitude of the comprehensive feature vector exceeds the preset threshold, it is determined to be a potential damage event through amplitude analysis, and a preliminary damage judgment result is obtained. From the preliminary damage assessment results, a damage-related subset is separated using feature extraction methods to obtain candidate damage features; Based on the candidate damage features, the support vector machine algorithm is used for classification to determine the damage type and obtain the damage classification result. Based on the damage classification results, cluster analysis was used to group the candidate damage features, resulting in damage feature groups. If the damage feature grouping meets the preset damage pattern matching conditions, the damage location is determined by pattern recognition technology to obtain the damage localization result. From the damage localization results, spatial distribution information related to the damage is extracted to obtain the damage spatial features; By using the spatial characteristics of the damage, a decision tree algorithm is employed to assess the severity of the damage and obtain the damage severity level.

5. The method according to claim 1, characterized in that, Methods for obtaining damage type labels by training a support vector machine classifier on damage candidate features include: Obtain the structural acceleration response at the measurement point; Calculate the transfer function between structural acceleration responses at adjacent measurement points; Calculate the normalized wavelet packet energy impairment feature and select the first value of the transfer function amplitude. The principal components constitute one-dimensional damage characteristics; Normalized wavelet packet energy damage features and one-dimensional damage features are used as sample sets, and damage location is used as the corresponding sample label. A support vector machine classification algorithm is used to identify the damage location. Normalized wavelet packet energy damage features and one-dimensional damage features are used as sample sets, and damage degree is used as the corresponding sample label. A support vector machine regression algorithm is used to identify the damage degree.

6. The method according to claim 1, characterized in that, Methods for determining the degree of damage evolution by fusing damage type labels with transportation environment data and employing sliding window analysis to track signal change trends include: By fusing multi-source features, battery status data and environmental parameters are integrated to construct a spatiotemporally complete battery damage dataset. By utilizing big data analytics algorithms, we can uncover the cross-influence of various factors and establish a big data analytics model that simultaneously characterizes the dynamic process over time and the superimposed effects of multiple causes. The sliding time window technique is used to analyze the trend of damage data over a continuous time period. The least squares method is used to fit the damage evolution curve. The damage development rate and acceleration are calculated based on the slope and second derivative of the curve. The damage evolution law parameters are obtained for predictive analysis.

7. The method according to claim 6, characterized in that, Big data analysis models that simultaneously characterize dynamic processes over time and the cumulative effects of multiple contributing factors include: the spatiotemporal geographic weighted regression model GTWR and the dependent variable spatiotemporal geographic weighted regression model; The GTWR model incorporates the spatiotemporal characteristics of the data into the regression model, addressing the non-stationarity issues in both space and time. Its calculation formula is as follows: ； in, For the first The dependent variable for each sample point; It is the first The spatiotemporal coordinates of each sample point; In the first The first sample point at the sample point One independent variable value; It is the first The intercept of the regression equation for each sample point; In the first The first sample point at the sample point regression coefficients of the independent variables; For random error term, The number of sample points; The multi-dependent variable spatiotemporal geographic weighted regression model further extends the GTWR model by considering multiple dependent variables. Its formula is as follows: ； in, Indicates the first The first sample point One dependent variable; It is the first The dependent variable in the th... Intercept term for each sample point; It is the first The dependent variable in the th... The first sample point The regression coefficients of the independent variables; It is a random error term. The number of sample points.

8. A battery mechanical damage monitoring system for electric vehicles during transportation based on acoustic emission spectrum analysis, the system being used to implement the method described in any one of claims 1-7, characterized in that, The system includes: a signal acquisition module, a signal processing module, a feature extraction module, an event judgment module, a damage classification module, an evolution analysis module, and an alarm generation module; The signal acquisition module is used to acquire transient elastic wave signals generated during transportation through a piezoelectric sensor array deployed on the battery pack, and to capture instantaneous vibration data caused by bumps and collisions using a synchronous triggering mechanism to obtain the original signal sequence. The signal processing module is used to perform multi-scale decomposition on the original signal sequence using wavelet transform, separating high-frequency transient components and low-frequency background noise to obtain a denoised elastic wave signal. The feature extraction module is used to calculate time-domain features based on the denoised elastic wave signal and combine them with frequency-domain features to obtain a comprehensive feature vector. The event judgment module is used to determine a potential damage event if the peak amplitude of the comprehensive feature vector exceeds a preset threshold, and then extract a damage-related subset from the feature vector to obtain damage candidate features. The damage classification module is used to train the damage candidate features using a support vector machine classifier to obtain damage type labels. The evolution analysis module is used to fuse transportation environment data with damage type labels, and to use sliding window analysis to track signal change trends and determine the degree of damage evolution. The alarm generation module is used to generate alarm signals based on the degree of damage evolution. If the duration of the alarm signal exceeds a set period, the remote transmission module is activated to send it to the monitoring center to obtain real-time intervention instructions.

Images