A lithium-ion battery pack multi-sensor fault diagnosis method based on statistical analysis

Abstract

This invention discloses a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis, specifically including the following steps: acquiring target sensors; extracting features from the output of the target sensors to obtain statistical features; performing feature mapping based on principal component analysis on the statistical features to obtain the target feature space; and detecting fault samples in the target feature space based on the α-Hull algorithm to complete the multi-sensor fault diagnosis of lithium-ion battery packs based on statistical analysis. This invention is applicable to scenarios where multiple sensors fail simultaneously. It achieves accurate sensor fault detection without establishing a battery model or setting diagnostic thresholds, avoiding the challenge of determining thresholds. Furthermore, it can easily isolate current, voltage, and temperature sensors that fail simultaneously.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery fault diagnosis technology, and in particular relates to a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis. Background Technology

[0002] Lithium-ion batteries, with their high power density and long cycle life, have been widely used in electric vehicles and grid energy storage. However, lithium-ion battery safety incidents characterized by thermal runaway occur frequently, raising public concerns about their safety. Therefore, in practical operation, lithium-ion battery packs require a high-efficiency Battery Management System (BMS) for state monitoring, thermal management, equalization management, fault diagnosis, and prediction to ensure safe and efficient operation. A precise and efficient BMS heavily relies on accurate data acquisition from voltage, current, and temperature sensors. In practical applications, inherent defects, aging, and harsh operating environments can cause sensor malfunctions such as bias, gain, drift, and decreased accuracy. The most direct impact of sensor failure is that the BMS cannot obtain accurate battery operating conditions, leading to erroneous control commands and accelerated battery system degradation. Therefore, effective sensor fault diagnosis methods are crucial for improving battery operational safety.

[0003] Existing sensor fault diagnosis methods for lithium-ion battery packs are all model-based. They estimate the battery state using equivalent circuit models or electrothermal coupling models, and then calculate a residual by subtracting the estimated value from the actual sensor measurement. If the residual exceeds a preset threshold, it indicates a sensor fault. Although existing diagnostic methods can detect sensor faults, they still face the following limitations:

[0004] (1) The accurate threshold is difficult to determine. In model-based sensor fault diagnosis methods, the choice of threshold directly affects the accuracy and reliability of fault detection. Too high a threshold may lead to missed detections, while too low a threshold may lead to false alarms. The accurate diagnostic threshold often requires a large number of offline tests to determine, which is often very time-consuming and costly, and is not conducive to the online application of the diagnostic method.

[0005] (2) Not applicable to scenarios where multiple sensors fail simultaneously. Existing methods for sensor fault detection and isolation assume that only one sensor fails at a time. However, this is a very strict assumption. In real-world battery packs, there are a large number of sensors, and the possibility of multiple sensors failing simultaneously cannot be ignored. When multiple sensors fail simultaneously, existing methods are prone to misdiagnosis and struggle to accurately isolate faults in different types of sensors.

[0006] Therefore, there is an urgent need for a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis to address the shortcomings of existing technologies. Summary of the Invention

[0007] The purpose of this invention is to propose a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis. This method can detect faults in multiple sensors of lithium-ion battery packs and isolate different types of faulty sensors without the need to establish a battery model or set diagnostic thresholds.

[0008] To achieve the above objectives, this invention provides a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis, specifically including the following steps:

[0009] Acquire target sensor;

[0010] Feature extraction is performed on the output of the target sensor to obtain statistical features;

[0011] The statistical features are subjected to feature mapping based on principal component analysis to obtain the target feature space;

[0012] The target feature space is detected using the α-Hull algorithm to obtain fault samples and complete multi-sensor fault diagnosis of lithium-ion battery packs based on statistical analysis.

[0013] Optionally, the target sensor includes a current sensor, a voltage sensor, and a temperature sensor.

[0014] Optionally, the statistical features include mean, variance, median, minimum, maximum, peak distance, measurement of data symmetry, measurement of data flatness, first-order autocorrelation coefficient, spectral entropy, variance fluctuation, mean fluctuation, trend strength, and seasonality strength.

[0015] Optionally, performing feature mapping based on principal component analysis on the statistical features to obtain the target feature space includes:

[0016] Obtain the data matrix of the statistical features, and calculate the covariance matrix of the data matrix;

[0017] The eigenvalues ​​of the covariance matrix are obtained through the covariance matrix, and the eigenvectors corresponding to the eigenvalues ​​are obtained. The eigenvectors corresponding to the eigenvalues ​​are the load matrix.

[0018] The data matrix is ​​decomposed using the load matrix to obtain the score matrix;

[0019] Based on the score matrix, target principal components are selected to obtain the target feature space.

[0020] Optionally, the covariance matrix of the data matrix is ​​calculated as follows:

[0021]

[0022] Where R is the covariance matrix of the data matrix V, N is the number of samples, and m is the feature dimension. This is the high-level feature space.

[0023] Optionally, the eigenvalues ​​of the covariance matrix can be obtained using the covariance matrix:

[0024] |R-λI m |=0

[0025] Where λ is the eigenvalue of the covariance matrix, I m It is an m-dimensional identity matrix.

[0026] Optionally, the data matrix can be decomposed using the load matrix to obtain the score matrix as follows:

[0027]

[0028] Where V is the data matrix, S∈R nc×q For the score matrix S, each column vector s i =Vp i p i Denotes the i-th principal component. Let m represent the residual, q represent the number of principal components satisfying q < m, and P be the load matrix.

[0029] Optionally, detecting the target feature space based on the α-Hull algorithm to obtain the fault samples includes:

[0030] The target feature space is detected using the α-Hull algorithm to obtain the sample boundary of the target feature space and obtain a sample set;

[0031] The discrete group is obtained from the sample set, and the discrete group is the fault sample.

[0032] Optionally, the α-Hull algorithm is as follows:

[0033]

[0034] Among them, C α (A) represents the α-hull of set A. Let x denote an open sphere centered at x and with radius α, and let c denote the complement of the set.

[0035] The present invention has the following beneficial effects:

[0036] This invention is applicable to scenarios where multiple sensors fail simultaneously. It can accurately detect sensor failures without establishing a battery model or setting diagnostic thresholds, avoiding the challenge of determining thresholds. It can also easily isolate current, voltage, and temperature sensors that fail simultaneously. This invention provides a method for detecting simultaneous failures of multiple sensors in a lithium-ion battery pack and isolating different types of faulty sensors without establishing a battery model or setting diagnostic thresholds. Attached Figure Description

[0037] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 This is a flowchart of a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis, according to an embodiment of the present invention.

[0039] Figure 2 This is a flowchart illustrating the process of fault detection implemented by the α-hull algorithm proposed in this embodiment of the invention.

[0040] Figure 3 The fault detection result of the voltage sensor when multiple sensors fail simultaneously, as proposed in the embodiments of the present invention;

[0041] Figure 4 This is a diagram showing the fault detection results of a temperature sensor when multiple sensors fail simultaneously, as proposed in an embodiment of the present invention. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present application will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0044] like Figure 1As shown, this embodiment proposes a multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis. The implementation process mainly includes feature extraction, feature mapping based on principal component analysis (PCA), and fault detection and isolation based on the α-Hull algorithm. First, a set of general statistical features is extracted from the measurement output of each sensor in each charge-discharge cycle to capture the global characteristics of the sensor measurements; then, PCA is used to map the high-dimensional feature space onto a two-dimensional plane; based on this, the α-Hull algorithm is used to achieve fault detection and isolation. It is worth noting that, as... Figure 1 As shown, the fault diagnosis processes for current, voltage, and temperature sensors are independent of each other and do not interfere with each other. This allows the proposed diagnostic method to easily isolate different types of sensors that have failed. The specific implementation process is as follows:

[0045] I. Feature Extraction

[0046] In this embodiment, the output of each sensor in one charge-discharge cycle is considered as a sample. To accurately detect potential sensor failures, a total of 15 general statistical features were extracted from each sample to characterize the global properties of each sensor measurement. Table 1, showing the 15 general statistical features extracted from each sample, details these extracted general statistical features. These features have been shown to have a good distribution in the feature space, meaning they point in different directions, thus minimizing redundancy in the feature space. Furthermore, these statistical features have been widely used in time series prediction, classification, and anomaly detection.

[0047] Table 1

[0048]

[0049] II. Feature Mapping Based on PCA

[0050] This embodiment utilizes PCA to map a high-dimensional feature space with linearly correlated features to a two-dimensional feature space, facilitating fault detection and result visualization. Since the fault diagnosis processes for current, voltage, and temperature sensors are independent and do not interfere with each other, for simplicity, the following explanation uses only the data matrix corresponding to the voltage sensor within the battery pack as an example to illustrate the PCA calculation process.

[0051] Treating the output of each sensor in one charge-discharge cycle as a sample, feature extraction maps each raw sample to an m=15-dimensional feature vector. For all voltage sensors within the battery pack, the data matrix after feature extraction is represented as:

[0052]

[0053] in, Let represent the feature vector of the measurement value of the i-th voltage sensor in the j-th cycle, where i = 1, 2, ..., n, j = 1, 2, ..., c, n represents the number of voltage sensors, and c represents the number of charge-discharge cycles. This is the high-level feature space.

[0054] First, calculate the covariance matrix R of the data matrix V, as shown in the following equation:

[0055]

[0056] Where N is the number of samples and m is the feature dimension.

[0057] Furthermore, by solving equation (3), the eigenvalues ​​and eigenvectors of the covariance matrix are obtained, thereby determining the load matrix:

[0058] |R-λI m |=0 (3)

[0059] Among them, I m It is an m-dimensional identity matrix.

[0060] The eigenvalues ​​λ = (λ1, λ2, ..., λ) are calculated. m ), where λ1≥λ2≥…≥λ m The eigenvector matrix P corresponding to the eigenvalues. m =(p1,p2,…,p m The load matrix of V is called the load vector, and each column of the load matrix P is called the load vector. The two are orthogonal to each other.

[0061] Therefore, the data matrix V can be decomposed into the following form:

[0062]

[0063] Where S∈R nc×q For the score matrix S, each column vector s i =Vp i p i Let represent the i-th principal component, which is essentially the projection of the data matrix V along the load vector direction. Let m represent the residual, and q represent the number of principal components, satisfying q < m. The first two principal components in the principal component subspace are selected to represent the original data matrix.

[0064] Similarly, the same PCA-based feature mapping is performed on the data matrices corresponding to the current sensor and temperature sensor as on the voltage sensor.

[0065] III. Fault Detection and Isolation Based on α-Hull Algorithm

[0066] α-convex hull, or simply α-Hull, is a generalized convex hull studied in various research fields (primarily computational geometry). In layman's terms, a hull can be viewed as a geometric structure used to characterize the shape of a set. Unlike convex hulls, α-Hull can reconstruct non-convex sets. The mathematical definition of α-Hull is as follows: If... Let x represent an open sphere centered at x and with radius α. Given a set A ∈ R... d A c and Let A and C represent the complement and boundary of A, respectively. For α > 0, if A = C α If (A), then the set A∈R is called. d It is α-convex, where

[0067]

[0068] C α (A) is called the α-Hull of A, and c denotes the complement of the set. Therefore, if A is α-convex, any point in the complement of A can be separated from A by an open sphere of radius α.

[0069] The α-Hull algorithm is applied to the first two principal components obtained in the second step to obtain the boundaries of normal samples in two-dimensional space. The following example illustrates the implementation principle and process of the α-Hull algorithm using the example of the voltage sensor of the 8th cell in an 8-cell battery pack failing during the 29th-30th charge-discharge cycle. After PCA dimensionality reduction, all samples in the resulting two-dimensional space can be plotted as follows... Figure 2 The biplot shown is illustrated here. Each data point represents a sample in the dataset. As you can see, most samples are represented by black dots, positioned close to each other, indicating similar principal component scores. The α-Hull algorithm functions similarly to clustering, finding the set of points that cluster together based on the principal component scores of each sample. Samples outside this set are marked as outliers (i.e., faulty samples) and labeled with triangles.

[0070] In this embodiment, the alphahull package in the existing R software can be used to calculate α-Hull.

[0071] This invention was tested on a battery pack consisting of eight cylindrical 21700 lithium-ion battery cells connected in series to verify the effectiveness of the technology. The battery pack used in the experiment was equipped with a current sensor to collect the battery pack current. Each battery cell was equipped with a voltage sensor and a thermocouple to collect the terminal voltage and surface temperature of each cell, respectively. The battery pack was subjected to 0.5C constant current charging and dynamic stress test (DST) discharging. A total of 30 charge-discharge cycles were performed on the battery pack.

[0072] Consider a scenario where multiple sensors fail simultaneously. Given that the series-connected battery pack uses only one current sensor, simultaneous sensor failure here refers to the simultaneous failure of multiple voltage sensors and multiple temperature sensors. Specifically, at the same time, a bias fault of 0.05V and 0.1℃ is injected into the voltage and temperature sensors of cell 5, respectively; a gain fault of 1.02 is injected into the voltage and temperature sensors of cell 6; and a fault amplitude of 10 is injected into the voltage and temperature sensors of cell 7. -5 ×t and 5×10 -5 A drift fault of ×t, where t represents time; and an accuracy degradation fault with a mean of zero and a standard deviation of 0.002 and 0.02, respectively, are injected into the voltage and temperature sensors of cell 8. These faults are injected simultaneously during charge-discharge cycles 29-30.

[0073] When multiple sensors fail simultaneously, the biplots of the detection results from the voltage and temperature sensors are as follows: Figure 3 and Figure 4 As shown. The biplot can intuitively explain and visualize the results of fault detection and isolation. In the biplot, each data point represents a sample in the dataset. i_j " represents the sample number corresponding to the output of the i-th voltage sensor in the j-th cycle; similarly, "t" represents the sample number corresponding to the output of the i-th voltage sensor in the j-th cycle. i_j "" indicates the sample number corresponding to the output of the i-th temperature sensor in the j-th loop. Figure 3 and Figure 4 Points close to each other in the biplot represent samples with similar principal component scores. Normal samples are represented by black dots, and the gray irregular curve represents the boundary of normal samples obtained by the α-hull algorithm. Samples outside this boundary are detected as faulty samples and marked with triangles. Furthermore, in the biplot, each vector corresponds to a feature, and the length and direction of the vector represent the weight of the corresponding feature on the principal components.

[0074] from Figure 3It can be observed that samples v_5_29, v_5_30, v_6_29, v_6_30, v_7_29, v_7_30, v_8_29, and v_8_30 were marked as faulty samples, while all other samples were marked as normal samples. Therefore, based on the test results, it can be determined that the voltage sensors of cells 5, 6, 7, and 8 failed during the 29th-30th charge-discharge cycles. Similarly, from... Figure 4 It can be observed that samples t_5_29, t_5_30, t_6_29, t_6_30, t_7_29, t_7_30, t_8_29, and t_8_30 were marked as faulty samples, while all other samples were marked as normal samples. Therefore, based on the test results, it can be determined that the temperature sensors of cells 5, 6, 7, and 8 failed during the 29th-30th charge-discharge cycles.

[0075] The fault detection results of the voltage and temperature sensors are completely consistent with those during fault injection, indicating that the proposed method can accurately detect faults when multiple sensors fail simultaneously. Furthermore, since the fault diagnosis processes for different types of sensors (current, voltage, and temperature sensors) are independent of each other, the designed method can easily isolate different types of faulty sensors. In summary, the method designed in this invention achieves fault detection and isolation of battery pack sensors without the need to establish a battery model or set diagnostic thresholds, demonstrating promising prospects for practical engineering applications.

[0076] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis, characterized in that, Specifically, the following steps are included: Acquire target sensor; Feature extraction is performed on the output of the target sensor to obtain statistical features, including mean, variance, median, minimum, maximum, peak distance, measurement of data symmetry, measurement of data flatness, first-order autocorrelation coefficient, spectral entropy, variance fluctuation, mean fluctuation, trend intensity, and seasonality intensity. The statistical features are subjected to feature mapping based on principal component analysis to obtain the target feature space: The data matrix of the statistical features is obtained, and the covariance matrix of the data matrix is ​​calculated; The eigenvalues ​​of the covariance matrix are obtained through the covariance matrix, and the corresponding eigenvectors are obtained, where the eigenvectors are the loading matrix; The data matrix is ​​decomposed using the loading matrix to obtain the score matrix; Target principal components are selected based on the score matrix to obtain the target feature space. based on The algorithm detects the target feature space and obtains fault samples: using the... The algorithm detects the target feature space, obtains the sample boundary of the target feature space, and obtains a sample set; through the sample set, it obtains a discrete group, which is the fault sample; thus completing the multi-sensor fault diagnosis of lithium-ion battery packs based on statistical analysis, wherein... The algorithm is as follows: in, Represents a set of , Indicates a Centered on, with The opening ball with radius , Represents the complement of a set.

2. The multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis as described in claim 1, characterized in that, The target sensors include a current sensor, a voltage sensor, and a temperature sensor.

3. The multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis as described in claim 1, characterized in that, The covariance matrix of the data matrix is ​​calculated as follows: in, For data matrix The covariance matrix, For the sample size, For the feature dimension, This is the high-level feature space.

4. The multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis as described in claim 3, characterized in that, The eigenvalues ​​of the covariance matrix are obtained from the covariance matrix as follows: in, These are the eigenvalues ​​of the covariance matrix. for A dimensional unit matrix.

5. The multi-sensor fault diagnosis method for lithium-ion battery packs based on statistical analysis as described in claim 4, characterized in that, The data matrix is ​​decomposed using the load matrix to obtain the score matrix as follows: in, For data matrix, For the score matrix, Each column vector , Indicates the first principal component, Represents the residual. Represent the number of principal components and satisfy the following conditions: , This is the load matrix.

Images