Lithium ion battery pack fault detection and positioning method based on cross-scale information fusion

Abstract

The application discloses a lithium ion battery pack fault detection and positioning method based on cross-scale information fusion, wherein the lithium ion battery pack comprises a plurality of series modules, and each series module comprises a plurality of parallel battery monomers; the method comprises the following steps: acquiring voltage data of the series modules, extracting normalized inter-module deviation based on inter-module similarity analysis; acquiring temperature data of the parallel battery monomers, constructing dominant statistical quantity and residual statistical quantity through space-time feature decomposition; combining the deviation, dominant and residual statistical quantities, designing a multi-scale fusion index to realize fault detection; and designing a hierarchical positioning method based on inter-module similarity degradation and space-time feature contribution to realize fault positioning. The application extracts features from battery system data by using cross-scale information fusion, constructs fault detection statistical quantity according to the extracted voltage features and temperature features to realize rapid fault detection, and realizes spatial positioning of the fault according to inter-module similarity degradation and space-time feature contribution.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery pack fault diagnosis, and more specifically, to a method for fault detection and localization of lithium-ion battery packs based on cross-scale information fusion. Background Technology

[0002] Lithium-ion battery packs, with their superior energy density, ultra-long cycle life, and continuously decreasing manufacturing costs, have become the mainstream energy storage solution in electric vehicles (EVs), grid systems, and portable electronic products. As the global electric vehicle market experiences unprecedented growth, ensuring battery reliability and safety has become increasingly critical. Despite significant technological breakthroughs, battery pack thermal runaway and internal short circuits (ISCs) remain unresolved safety challenges, especially in automotive applications where battery packs must cope with dynamic operating conditions and varying thermal environments, making this issue even more pronounced.

[0003] Early internal anomalies in lithium-ion battery packs, if not identified and addressed promptly, can escalate into catastrophic failures, disrupting system operation and posing significant risks to equipment integrity and operational safety. The inherent spatiotemporal coupling effects of distributed thermal systems can cause localized faults to propagate rapidly throughout the system domain, triggering cascading failures and potentially dangerous thermal runaway events. For example, internal short circuits (ISCs) in lithium-ion batteries are among the most concerning faults, capable of generating substantial amounts of heat and developing into thermal runaway within a short timeframe. Therefore, developing robust fault detection and localization methods for lithium-ion battery packs is crucial for ensuring the reliable and safe operation of modern industrial processes.

[0004] Existing fault detection methods for lithium-ion battery packs can be categorized into three types: model-based methods, signal processing techniques, and machine learning-based methods. Model-based methods use equivalent circuits or physics-based models to identify fault-induced deviations, but these methods face challenges due to large system complexity, parameter drift during aging, and high computational costs. Signal processing techniques apply transforms to extract fault features from sensor data, but require specialized knowledge and are sensitive to measurement noise. Machine learning-based methods utilize advanced algorithms to identify fault modes without explicit physical models, but typically require large training datasets and lack interpretability for root cause analysis.

[0005] Information fusion methods have emerged as a promising solution for enhancing fault detection by integrating complementary information from lumped variables (voltage and current) and distributed variables (temperature field). However, existing fusion schemes face key challenges: a lack of unified weight optimization strategies, inability to address precise spatiotemporal localization in complex distributed domains, and a lack of an integrated framework for simultaneous rapid detection and precise spatial localization. Most existing methods operate independently on lumped or distributed variables, limiting their performance across diverse operational scenarios.

[0006] In summary, existing methods suffer from the following technical problems: model-based methods are highly dependent on partial differential equations, making it difficult to obtain accurate physical models for complex lithium-ion battery packs, and they also suffer from parameter drift and high computational costs during the aging process; traditional information fusion methods lack a unified weight optimization strategy and cannot adaptively determine the optimal fusion weights for different information sources; existing methods cannot simultaneously achieve a balance between rapid detection and accurate spatial positioning, resulting in a contradiction between detection sensitivity and spatial positioning accuracy; most existing methods target lumped or distributed variables independently, failing to overcome single-modal constraints and limiting performance in diverse operating scenarios; and there is a lack of an integrated framework capable of handling accurate spatiotemporal positioning in complex distributed domains, failing to provide hierarchical positioning capabilities from coarse to fine granular.

[0007] Therefore, it is necessary to propose a cross-scale information fusion method for rapid fault detection and accurate spatial positioning of lithium-ion battery packs. Summary of the Invention

[0008] To address the problems of existing model-based lithium-ion battery pack fault diagnosis methods being highly dependent on physical models, and data-based fault diagnosis methods being unable to adapt to cross-scale information fusion and achieve rapid detection and precise location, this invention provides a lithium-ion battery pack fault detection and location method based on cross-scale information fusion. Through cross-scale information fusion, rapid fault detection and precise spatial location of lithium-ion battery packs can be achieved.

[0009] According to a first aspect of the present invention, a method for fault detection and localization of a lithium-ion battery pack based on cross-scale information fusion is provided, wherein the lithium-ion battery pack includes Ns series modules, each series module including Np parallel battery cells, and the method includes: S1. Obtain the voltage data of the series modules, and extract the normalized inter-module deviation based on the inter-module similarity analysis; S2. Obtain the temperature data of the parallel battery cells, and construct the dominant statistics and residual statistics through spatiotemporal feature decomposition; S3. By combining the normalized inter-module bias, dominant statistics, and residual statistics, a multi-scale fusion index is designed to achieve fault detection; S4. A hierarchical localization method is designed based on the degradation of inter-module similarity and the contribution of spatiotemporal features to achieve fault localization.

[0010] In the above scheme, step S1 includes: S11. Obtain the voltage data of the series module; S12. Based on the acquired voltage data, calculate the inter-module similarity coefficient between adjacent module voltages using a recursive algorithm; S13. Based on the minimum similarity coefficient between the two modules, calculate the similarity degradation index, convert the similarity degradation index into a quantitative deviation from normal behavior, take the maximum value of the deviation in the training data as a scaling reference, and finally obtain the normalized inter-module deviation.

[0011] In the above scheme, step S11 further includes: A controlled square wave perturbation is applied to the voltage data of each series module:

[0012] In the formula, For the first i Voltage data of each series module; The voltage data after applying a controlled square wave disturbance; The disturbance coefficient; The disturbance period; For time; Step S12 includes: The similarity between modules is designed to quantify the linear relationship between the voltages of adjacent modules:

[0013] In the formula, The similarity coefficient between modules; W The length of the sliding window; Derive the sliding window length using a recursive algorithm. W Upper Module i and modules j The inter-module similarity coefficient is defined as a recursive variable as follows:

[0014]

[0015]

[0016]

[0017]

[0018] in, It is the sum of cross products. and It is voltage and, and It is the sum of squares of voltages; Step S13 includes: The average value of the minimum similarity coefficient between two modules is taken as the similarity degradation index. : Similarity degradation index Converted into quantification and normal behavior deviation ; Take the maximum value of the bias in the training data. As a scaling reference, the final normalized inter-module deviation is obtained. .

[0019] In the above scheme, step S2 includes: S21. Obtain the temperature data of the parallel battery cells and construct a temperature data matrix. Where N represents the number of samples, and M = Np × Ns represents the total number of parallel battery cells in the lithium-ion battery pack, i.e. the total number of temperature data. S22. Identification of dominant thermal modes through spatiotemporal feature decomposition:

[0020] In the formula, The load matrix; This is the score matrix; T denotes the transpose; S23. Construct the dominant statistic and residual statistic:

[0021]

[0022] In the formula, The dominant statistic; For residual statistics; This is a column vector of temperature data for all parallel battery cells at time t, with a size of M×1; for The inverse matrix; It is an identity matrix.

[0023] In the above scheme, step S3 includes: S31. Combine normalized inter-module deviations Dominant statistics and residual statistics Design multi-scale fusion indicators :

[0024] In the formula, , and To integrate weights, satisfy ; Dominant statistic The upper control limit; For residual statistics The upper control limit; S32. The optimal fusion weights are determined using a genetic algorithm, with the objective function as follows:

[0025] in, FAR stands for False Alarm Rate; ADR stands for Anomaly Detection Rate. The false alarm rate (FAR) quantifies the number of normal samples that are incorrectly classified as faults:

[0026] In the formula, This represents the number of normal samples that are mistakenly detected as abnormal samples under normal circumstances. This represents the total number of normal samples. Examples of faults correctly identified by the Anomaly Detection Rate (ADR) measurement:

[0027] In the formula, This represents the number of abnormal samples detected under abnormal circumstances. This represents the total number of abnormal samples. Optimal fusion weights Determined as:

[0028] S33. Determine the detection threshold ,like If a fault is detected, the fault time will be recorded as follows: Otherwise, it's normal.

[0029] In the above scheme, the detection threshold The method for determining it is as follows:

[0030] That is, extract multi-scale fusion metrics from the training data. The 99.9 percentile.

[0031] In the above scheme, step S4 includes: S41. Achieve coarse-grained module fault location based on inter-module similarity degradation, that is, analyze the inter-module similarity degradation pattern between adjacent module pairs to identify faulty modules; S42. Fine-grained single-unit fault location is achieved based on the contribution functions of dominant and residual statistics. That is, within the identified faulty modules, the faulty single unit is located by analyzing the spatial contribution of dominant and residual statistics.

[0032] In the above scheme, step S41 includes: After fault detection, calculate each adjacent module pair within the time window ( i , i The number of low-relevance samples (+1) is:

[0033] In the formula, For adjacent module pairs ( i , i The number of low-correlation samples (+1) ,when hour, i +1=1; The length of the time window; The similarity coefficient between modules; It is a similarity threshold; It is an indicator function; it equals 1 when the condition is met and 0 when the condition is not met. Then the fault module Identified as:

[0034] In the formula, For the series module that failed; when i When =1, ; make The largest i That is This allows for the identification of faulty modules; Step S42 includes: Leading contribution Quantification of the contribution to anomalies:

[0035]

[0036] In the formula, The score for the dominant component k; Load matrix No. k load vectors The j One element; For the time of failure Time j Temperature data of individual parallel battery cells; To correspond to the dominant component k eigenvalues; For the time of failure A column vector of temperature data for all parallel battery cells; Residual contribution The contribution of measurements to residual thermal modes that cannot be accounted for by the dominant components:

[0037]

[0038] In the formula, It is a residual; Design a comprehensive contribution rate by combining the dominant contribution and residual contribution. :

[0039] In the formula, and For weighted parameters; Then the faulty unit Identified as:

[0040] In the formula, For the faulty parallel battery cell; ; make The largest j That is This allows for the identification of faulty individual units.

[0041] According to a second aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the lithium-ion battery pack fault detection and location method based on cross-scale information fusion as described in any one of the first aspects.

[0042] According to a third aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the lithium-ion battery pack fault detection and location method based on cross-scale information fusion as described in any one of the first aspects.

[0043] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: This invention provides a fault detection and localization method for lithium-ion battery packs based on cross-scale information fusion. It utilizes cross-scale information fusion to extract features from lithium-ion battery pack data, and constructs fault detection statistics based on the extracted voltage and temperature features to achieve rapid fault detection. Spatial fault localization is achieved based on inter-module similarity degradation and spatiotemporal feature contributions. This method is a data-driven approach, independent of precise physical models, and therefore suitable for fault detection and localization in practical lithium-ion battery packs. Furthermore, this method achieves a balance between rapid detection and precise localization. Attached Figure Description

[0044] Figure 1This is a schematic flowchart illustrating a lithium-ion battery pack fault detection and localization method based on cross-scale information fusion according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a lithium-ion battery pack structure according to an embodiment of the present invention; Figure 3 This is a UDDS test current curve diagram according to an embodiment of the present invention; Figure 4 This is an anomaly detection result diagram under fault #5 condition according to an embodiment of the present invention; wherein, Figure 4 (a) in the figure is the anomaly detection result diagram of the present invention. Figure 4 Figures (b), (c), and (d) in the diagram represent the anomaly detection results of the traditional method. Figure 5 This is a diagram showing the anomaly location result under fault #1 condition according to an embodiment of the present invention; wherein, Figure 5 (a) in the figure is the module-level fault location result diagram under fault #1 condition. Figure 5 (b) in the figure is the fault location result of a single unit under fault #1 condition; Figure 6 This is an anomaly location result diagram under fault #5 condition according to an embodiment of the present invention; wherein, Figure 6 (a) in the figure shows the module-level fault location result under fault #5. Figure 5 (b) in the figure is the fault location result of a single unit under fault #5 condition; Figure 7 This is a schematic block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0045] The following detailed description is merely exemplary in nature and is not intended to limit the disclosed technology or its application and use. Furthermore, it is not intended to be bound by any express or implied theory presented in the foregoing technical fields, background art, or the following detailed description.

[0046] In the following detailed description of the embodiments, numerous specific details are set forth in order to provide a more thorough understanding of the disclosed technology. However, it will be apparent to those skilled in the art that the disclosed technology can be practiced without these specific details. In other instances, well-known features have not been described in detail to avoid unnecessarily complicating the description.

[0047] Terms such as "comprising" and "including" indicate that, in addition to the units and steps that are directly and explicitly stated in the specification, the technical solution of the present invention does not exclude the presence of other units and steps that are not directly or explicitly stated. Terms such as "first" and "second" do not indicate the order of the units in terms of time, space, size, etc., but are merely used to distinguish the units.

[0048] This invention provides a fault detection and localization method for lithium-ion battery packs based on cross-scale information fusion. It utilizes cross-scale information fusion to extract features from system data and constructs fault detection statistics based on the extracted voltage and temperature features to achieve rapid fault detection. Spatial fault localization is achieved based on inter-module similarity degradation and spatiotemporal feature contributions. This method is a purely data-driven approach, independent of precise physical models, and therefore suitable for fault detection and localization in practical lithium-ion battery packs. Furthermore, this method achieves a balance between rapid detection and precise localization.

[0049] Next, one or more embodiments of the present invention will be described in detail with reference to the accompanying drawings.

[0050] Please refer to the attached diagram below. Figure 1 This is a schematic flowchart illustrating a lithium-ion battery pack fault detection and localization method based on cross-scale information fusion according to one or more embodiments of the present invention. Figure 1 As shown in the figure, the lithium-ion battery pack fault detection and localization method based on cross-scale information fusion according to an embodiment of the present invention includes the following steps: S1. Based on inter-module similarity analysis, feature extraction is performed on the collected lithium-ion battery pack voltage data to obtain the normalized inter-module deviation; S2. Analyze the distributed temperature data based on the spatiotemporal characteristic decomposition, and construct the dominant statistics and residual statistics; S3. A multi-scale fusion index is designed based on a weighted fusion strategy optimized by a genetic algorithm to achieve fault detection; S4. A hierarchical localization method is designed based on the degradation of inter-module similarity and the contribution of spatiotemporal features to achieve fault localization.

[0051] In some embodiments, the specific process of step S1 includes: S11. Square wave perturbation processing; S12. Module similarity calculation based on recursive algorithm; S13. Module similarity normalization processing.

[0052] Specifically, step S11 is as follows: like Figure 2 As shown, consider a lithium-ion battery pack consisting of Ns series-connected modules, each containing Np parallel-connected battery cells. The system is equipped with a hierarchical sensor network, including voltage sensors. (i∈{1,2,...,Ns}) Measuring lumped electrical properties, temperature sensor (j∈{1,2,...,Ns×Np}) captures the distributed thermal field of the system domain.

[0053] To enhance the signal-to-noise ratio and eliminate static bias, a controlled square wave perturbation is applied to the voltage of each module:

[0054] Where A represents the disturbance coefficient, which should be small enough to avoid affecting system operation; H is the disturbance period; i Indicates the module number. This preprocessing step filters out low-frequency noise and measurement drift while preserving fault-related voltage dynamics. For the first i Voltage data of each series module over time t Change is a time series.

[0055] Specifically, step S12 is as follows: The similarity between modules is designed to quantify the linear relationship between the voltages of adjacent modules:

[0056] in It is the similarity coefficient between modules, representing the similarity between modules; the numerator is the measurement covariance, and the denominator is normalized by the product of standard deviations.

[0057] For efficient computation, a recursive algorithm is used to derive the sliding window length. W Upper Module i and j The similarity coefficient between them; the recursive variables are defined as follows:

[0058]

[0059]

[0060]

[0061]

[0062] in, It is the sum of cross products. and It is voltage and, and It is the sum of the squares of the voltages.

[0063] These recursive formulas avoid recalculating the entire window at each time step, reducing the computational complexity from O(W) to O(1).

[0064] Specifically, step S13 is as follows: First, calculate the set of similarities between adjacent modules. :

[0065] When a fault occurs, the affected module simultaneously causes a degradation in the inter-module similarity between two adjacent module pairs. The similarity coefficients are then sorted.

[0066] in .

[0067] Capturing similarity degradation using the average of two minimum values:

[0068] In the formula, This is an indicator of similarity degradation.

[0069] Similarity degradation index Converted into quantified and normal behavioral deviations:

[0070] in This indicates the deviation between modules.

[0071] The maximum value of the inter-module deviation in the training data is used as the scaling reference:

[0072] The normalized inter-module deviation is then calculated as follows:

[0073] In the formula, This is to normalize the inter-module bias. It can be used for fault detection, under normal circumstances <1, during a fault ≥1.

[0074] In some embodiments, the specific process of step S2 includes: S21. Temperature data preprocessing; S22. Spatiotemporal feature decomposition; S23. Construction of dominant statistics and residual statistics.

[0075] Specifically, step S21 is as follows: Temperature measurements provide spatiotemporal information about the system's thermal state. Spatiotemporal eigenvalue decomposition extracts dominant and residual statistics from distributed temperature data. The temperature data matrix is ​​represented as follows: Where N and M represent the number of samples and the number of sensors, respectively; matrix X is normalized as: ,in μ σ and σ represent the mean and standard deviation of the temperature data, respectively.

[0076] Specifically, step S22 is as follows: Identifying dominant thermal modes through discrete spatiotemporal separation:

[0077] Load matrix Includes the first n feature vectors It captures 99.9% of the system's energy.

[0078] Specifically, step S23 is as follows: Dominant and residual statistics can be constructed:

[0079]

[0080] The dominant statistic ξ measures the deviation within the dominant thermal model; the residual statistic δ(t) captures information not considered by the dominant component.

[0081] In some embodiments, the specific process of step S3 includes: S31. Construction of multi-scale indicators; S32. Genetic Algorithm Weight Optimization; S33. Determine the detection threshold.

[0082] Specifically, step S31 is as follows: Multi-scale indicators combine inter-module similarity, dominant statistics, and residual statistics:

[0083] Constraints:

[0084] in and yes and The upper control limit of the statistic can be derived through kernel density estimation. Each term in the equation represents a normalization index, where a value less than 1 indicates a normal condition and a value greater than 1 indicates a fault condition.

[0085] Specifically, step S32 is as follows: Genetic algorithms determine the optimal fusion weights based on performance metrics. Anomaly detection rate (ADR) measures correctly identified fault instances.

[0086] False Alarm Rate (FAR) quantifies the number of normal samples that are misclassified as faults:

[0087] Optimize the objective function by combining ADR and FAR:

[0088] The optimal weights are determined as follows:

[0089] in .

[0090] Specifically, step S33 is as follows: Detection threshold for the multi-scale index F(t):

[0091] This corresponds to the 99.9 percentile of the multiscale index during normal operation.

[0092] when The fault was detected at the time of recording. .

[0093] In some embodiments, the specific process of step S4 includes: S41. Coarse-grained module fault location; S42. Fine-grained individual fault location.

[0094] Specifically, step S41 is as follows: The fault location strategy employs a hierarchical approach, locating faults from the module level down to the individual cell level. The two-stage framework leverages inter-module similarity and thermal contribution analysis to achieve high-precision fault location within lithium-ion battery packs.

[0095] The first stage identifies the affected modules by analyzing the degradation patterns of inter-module similarity between adjacent module pairs. After fault detection, each adjacent module pair within the time window ( i , i The number of low-relevance samples (+1) is calculated as follows:

[0096] in ;when At that time, i+1=1; The similarity threshold is determined statistically from the training data, and W is the length of the time window, which may differ from the sliding window length for inter-module similarity. This equation captures the duration of correlation degradation. It is an indicator function; it equals 1 when the condition is met and 0 when the condition is not met.

[0097] Fault Module It can be identified as:

[0098] when iWhen =1, When the exception module i When identified, the similarity between modules and They both decreased simultaneously. Then, and The value will increase and can be used to identify faulty modules. In contrast, noise or measurement errors typically only affect one aspect of the similarity between modules.

[0099] Specifically, step S42 is as follows: The contribution functions of dominant and residual statistics are used to locate specific faulty cells within the identified modules. Dominant contribution Quantification of the contribution to anomalies:

[0100]

[0101] in It is the dominant component k The score, It is the k-th load vector The j One element, It is time Time from sensor j Normalized measurement value, It corresponds to the dominant component. k The equation captures the eigenvalues ​​of each sensor that deviate from the normal thermal pattern. It helps in detecting anomalous thermal behavior in the presence of normal thermal coupling patterns.

[0102] Residual contribution The contribution of measurements to residual thermal modes that cannot be accounted for by the dominant components:

[0103] in It is a sensor j The residual (prediction error) at the location is calculated as follows:

[0104] This residual-based statistic can effectively detect faults that deviate from historical thermal behavior patterns.

[0105] ξ Contribution and δ Contribution combined with overall design contribution :

[0106] in and They are ξ (t) and δ (t) weighting parameters of the statistic. (Through...) Normalization ensures that the weights of the overall contribution are expressed in units.

[0107] Faulty unit The location identified as having the greatest thermal contribution:

[0108] in .

[0109] The advantages of the hierarchical localization framework are as follows: the first stage focuses only on the faulty module rather than analyzing all individual units, thereby reducing computational complexity; the combination of inter-module similarity analysis and heat-based individual unit identification makes accurate individual unit localization possible.

[0110] More specifically, this invention implements a solution using the thermal process fault detection and location of lithium-ion battery packs as an example.

[0111] Lithium-ion battery packs are widely used in portable devices, medical devices, and electric vehicles. Localized heat sources caused by internal short circuits not only affect battery life but also increase the risk of thermal runaway in the battery system. Therefore, early detection and localization of internal short circuit anomalies are crucial. The following performance metrics are defined for performance evaluation and comparison: Anomaly detection rate:

[0112] False alarm rate:

[0113] Fault detection delay: Delay = td - tf in, N ta and N ns These represent the total number of abnormal samples and the total number of normal samples, respectively. N da This indicates the number of abnormal samples detected under abnormal circumstances; N fa This represents the number of normal samples that were mistakenly detected as abnormal under normal circumstances; td is the moment when the abnormality was first detected and an alarm was issued; tf is the moment when the abnormality actually occurred.

[0114] Experiments verified the use of a 6S4P lithium-ion battery configuration, such as... Figure 2As shown, the equipment required for the experimental environment includes a battery pack equipped with an embedded system capable of monitoring voltage and temperature data, a charging and discharging device to simulate current changes, and a temperature-controlled chamber for the battery pack. Table 1 shows six fault scenarios with different fault locations, resistance, current curves, and occurrence times. The fault location zf is denoted as zf=[mf,cf]. Data was collected at 1Hz for 2000 time points.

[0115] Table 1 Fault Condition Setting Table

[0116] like Figure 3 As shown, the Urban Dynamometer Driving Schedule (UDDS) represents a standardized urban driving pattern with dynamic current curves. This experiment verifies the stability of the method under fluctuating conditions. To simulate real-world scenarios, Gaussian white noise (Voltage: SNR=90dB; Temperature: SNR=50dB) was added to the measurements. Genetic algorithm parameters were set empirically: population size of 30 individuals (3-5% of training data), 50 generations, crossover rate of 0.8, and mutation rate of 0.1.

[0117] The main algorithm parameters are listed in Table 2. The method is evaluated using three key performance metrics: Anomaly Detection Rate (ADR), False Alarm Rate (FAR), and Detection Latency. Detection latency is the difference between the time to first fault detection and the time of fault occurrence. These metrics comprehensively evaluate the method's sensitivity, stability, and response speed. Fast response is crucial for preventing the spread of thermal runaway in battery systems.

[0118] Table 2 Main Algorithm Parameters

[0119] like Figure 4 As shown in Table 3, the cross-scale information fusion method demonstrates superior performance by combining inter-module similarity, dominant statistics, and residual statistics. As shown in Table 3, inter-module similarity with weight w3 becomes more dominant under dynamic condition #5, while temperature-based indicators contribute more significantly under steady-state conditions.

[0120] Table 3. Performance of Genetic Algorithm Optimization and Detection

[0121] Correlation analysis methods Statistics and the SPE statistics of the traditional PCA method were used to compare with the proposed cross-scale information fusion method. Table 4 shows that the cross-scale information fusion method achieves 97.8-99.9% ADR in all scenarios. Except for the complex UDDS dynamic case where the FAR is 0.4%, the FAR is 0.1% in other cases, with detection latency of 2-23 seconds. In challenging UDDS scenarios, the fusion method maintains 97.8% ADR, while individual methods show significant degradation. However, The statistic only achieves an ADR of 56.7% and an FAR of 11.7%. The ADR of the voltage correlation coefficient method is very unstable, ranging from 21.3% to 83.7%. The traditional T^2 statistical method reduces the ADR under dynamic conditions, while the traditional SPE statistical method produces a higher FAR.

[0122] Table 4. Comparison of the overall performance of the detection methods

[0123] like Figure 5 and Figure 6 As shown, the hierarchical fault location strategy comprises two stages: module-level identification and unit-level identification. The proposed method achieves excellent location accuracy in all test scenarios. Table 5 shows that the proposed method achieves accurate module and unit-level location in all scenarios. Faulty modules are identified through inter-module similarity analysis between adjacent module pairs (e.g., (5, 6) and (6, 1)). Figure 6 (a) and Figure 7 (a) in the diagram illustrates the module-level identification process.

[0124] Table 5 Fault location performance results

[0125] Individual cell localization demonstrates the ability to accurately identify faulty cells within an identified faulty module. Spatiotemporal feature contribution analysis based on optimized weights successfully identified specific faulty cells, locating cells 3 and 4. Experimental results show the stability of the proposed method under diverse fault types, locations, and operating conditions.

[0126] Experimental results demonstrate that the proposed cross-scale information fusion method outperforms all evaluation scenarios. This method consistently maintains an ADR (Advanced Distance Rate) between 97.8% and 99.9%, while keeping the FAR (Failure Accuracy) below 0.4% even under dynamic and complex conditions. The proposed method significantly outperforms traditional detection methods using a single statistic. Under steady-state conditions, the fault detection latency can be as fast as 2 seconds, and under dynamic conditions, it can be stably maintained within 23 seconds. The fault detection latency under steady-state conditions is generally maintained within 5 seconds, achieving rapid fault detection response. The proposed fusion framework achieves a balance between stability and sensitivity, with an average fault detection latency of only 7.0 seconds, demonstrating a balance between response speed and diagnostic reliability. Its adaptive weighting mechanism automatically adjusts the contributions of different statistics according to the system state, providing enhanced adaptability. The hierarchical localization framework can stably and accurately locate fault coordinates in all scenarios, effectively solving the localization problem of serial and parallel systems, and exhibiting high localization stability and accuracy.

[0127] In summary, this invention proposes a fault detection and localization method for lithium-ion battery packs based on cross-scale information fusion. This method includes: monitoring voltage anomalies through recursive correlation calculation and similarity analysis between square wave perturbation design modules; extracting dominant and residual statistics of distributed temperature using spatiotemporal feature decomposition; constructing a multi-scale index integrating inter-module similarity, dominant statistics, and residual statistics based on a weighted fusion strategy optimized by a genetic algorithm; and employing a hierarchical localization framework, first identifying fault-affected modules through inter-module similarity, and then locating specific battery cells using the extracted spatiotemporal features. The fault detection and localization method for lithium-ion battery packs provided by this invention utilizes cross-scale information fusion to extract features from system data, constructs fault detection statistics based on extracted voltage and temperature features to achieve rapid fault detection, and achieves spatial fault localization based on inter-module similarity degradation and spatiotemporal feature contributions.

[0128] Figure 7 This is a schematic block diagram of an electronic device according to one or more embodiments of the present invention. The electronic device 200 includes a memory 210, a processor 220, and a computer program 230 stored in the memory 210 and executable on the processor 220. Execution of the computer program 230 causes the execution of a lithium-ion battery pack fault detection and localization method based on cross-scale information fusion as described in the above method embodiments. Exemplarily, the electronic device 200 may be a controller or a computer.

[0129] Alternatively, the present invention can also be implemented as a computer-readable storage medium storing a program for causing a computer to execute the lithium-ion battery pack fault detection and location method based on cross-scale information fusion described in the above-described method embodiments. Here, various types of computer-readable storage media can be used, such as disks (e.g., magnetic disks, optical disks, etc.), cards (e.g., memory cards, optical cards, etc.), semiconductor memory (e.g., ROM, non-volatile memory, etc.), and tapes (e.g., magnetic tape, cassette tape, etc.).

[0130] Where applicable, the various embodiments provided by the present invention may be implemented using hardware, software, or a combination of hardware and software. Furthermore, where applicable, without departing from the scope of the invention, the various hardware and / or software components described herein may be combined into composite components comprising software, hardware, and / or both. Where applicable, without departing from the scope of the invention, the various hardware and / or software components described herein may be divided into sub-components comprising software, hardware, or both. Additionally, where applicable, it is contemplated that software components may be implemented as hardware components, and vice versa.

[0131] Software (such as program code and / or data) according to the invention can be stored on one or more computer-readable storage media. It is also contemplated that the software identified herein can be implemented using one or more networked and / or otherwise general-purpose or special-purpose computers and / or computer systems. Where applicable, the order of the various steps described herein can be changed, combined into compound steps, and / or divided into sub-steps to provide the features described herein.

[0132] The embodiments and examples presented herein are provided to best illustrate embodiments of the invention and its particular applications, thereby enabling those skilled in the art to practice and use the invention. However, those skilled in the art will understand that the above description and examples are provided merely for ease of illustration and example. The descriptions presented are not intended to cover all aspects of the invention or to limit the invention to the precise forms disclosed.

Claims

1. A method for fault detection and localization of lithium-ion battery packs based on cross-scale information fusion, characterized in that, The lithium-ion battery pack includes Ns series modules, and each series module includes Np parallel battery cells. The method includes: S1. Obtain the voltage data of the series modules, and extract the normalized inter-module deviation based on the inter-module similarity analysis; Step S1 includes: S11. Obtain the voltage data of the series module; S12. Based on the acquired voltage data, calculate the inter-module similarity coefficient between adjacent module voltages using a recursive algorithm; S13. Based on the minimum similarity coefficient between the two modules, calculate the similarity degradation index, convert the similarity degradation index into a quantitative deviation from normal behavior, take the maximum value of the deviation in the training data as a scaling reference, and finally obtain the normalized inter-module deviation. S2. Obtain the temperature data of the parallel battery cells, and construct the dominant statistics and residual statistics through spatiotemporal feature decomposition; Step S2 includes: S21. Obtain the temperature data of the parallel battery cells and construct a temperature data matrix. Where N represents the number of samples, and M=Np×Ns represents the total number of parallel battery cells in the lithium-ion battery pack, i.e. the total number of temperature data. S22. Identification of dominant thermal modes through spatiotemporal feature decomposition: In the formula, The load matrix; This is the score matrix; T denotes the transpose; S23. Construct the dominant statistic and residual statistic: In the formula, The dominant statistic; For residual statistics; This is a column vector of temperature data for all parallel battery cells at time t, with a size of M×1; for The inverse matrix; It is the identity matrix; S3. By combining the normalized inter-module bias, dominant statistics, and residual statistics, a multi-scale fusion index is designed to achieve fault detection; S4. A hierarchical localization method is designed based on the degradation of inter-module similarity and the contribution of spatiotemporal features to achieve fault localization; Step S4 includes: S41. Implementing coarse-grained module fault localization based on inter-module similarity degradation, including: After fault detection, calculate each adjacent module pair within the time window ( i , i The number of low-relevance samples (+1) is: In the formula, For adjacent module pairs ( i , i The number of low-correlation samples (+1) ,when hour, i +1=1; The length of the time window; The similarity coefficient between modules; It is a similarity threshold; It is an indicator function; it equals 1 when the condition is met and 0 when the condition is not met. Then the fault module Identified as: In the formula, is a failed series module; when i = 1, ; causing the maximum i that is thereby identifying a faulty module; S42. Fine-grained single-unit fault localization based on the contribution function of dominant statistics and residual statistics, including: dominant contribution quantifying the contribution of anomalies: In the formula, As the dominant component k The score; Load matrix No. k load vectors The j One element; For the time of failure Time j Temperature data of individual parallel battery cells; To correspond to the dominant component k eigenvalues; For the time of failure A column vector of temperature data for all parallel battery cells; Residual contribution Measuring the contribution of residual thermal patterns that cannot be accounted for by the dominant components: In the formula, is a residual error; Designing a comprehensive contribution degree combining a dominant contribution and a residual contribution : wherein and are weighting parameters; then the faulty cell identified as: In the formula, a faulty parallel battery cell; ; causing the largest j i.e. thereby identifying the faulty cell.

2. The lithium-ion battery pack fault detection and localization method based on cross-scale information fusion according to claim 1, characterized in that, Step S11 also includes: A controlled square wave perturbation is applied to the voltage data of each series module: wherein is the voltage data of the nth i serial module; is the voltage data after applying the controlled square wave perturbation; is the perturbation coefficient; is the perturbation period; is the time; Step S12 includes: The similarity between modules is designed to quantify the linear relationship between the voltages of adjacent modules: In the formula, is the inter-module similarity coefficient; W is the sliding window length; Deriving sliding window length using a recursive algorithm W upper module i and module j between the modules, the recursive variable is defined as follows: wherein, is the cross product sum, and is the voltage sum, and is the voltage square sum; Step S13 includes: Based on the minimum similarity coefficient between two modules, the average value is taken as the similarity degradation index : degrading the similarity index converted to a quantification of deviation from normal behavior ; taking the maximum value of the bias in the training data as a scaling reference, resulting in a normalized inter-module bias .

3. The lithium-ion battery pack fault detection and localization method based on cross-scale information fusion according to claim 1, characterized in that, Step S3 includes: S31. Incorporating Normalized Inter-Module Bias , Dominant Statistics and Residual Statistics , Designing Multiscale Fusion Metrics : wherein , and are fusion weights satisfying ; is an upper control limit for the dominant statistic ; is an upper control limit for the residual statistic ; S32. The optimal fusion weights are determined using a genetic algorithm, with the objective function as follows: wherein, ; FAR is false alarm rate; ADR is abnormality detection rate; The false alarm rate (FAR) quantifies the number of normal samples that are incorrectly classified as faults: In the formula, This represents the number of normal samples that are mistakenly detected as abnormal samples under normal circumstances. This represents the total number of normal samples. Examples of faults correctly identified by the Anomaly Detection Rate (ADR) measurement: In the formula, This represents the number of abnormal samples detected under abnormal circumstances. This represents the total number of abnormal samples. Optimal fusion weights Determined as: S33. Determine the detection threshold ,like If a fault is detected, the fault time will be recorded as follows: Otherwise, it's normal.

4. The lithium-ion battery pack fault detection and localization method based on cross-scale information fusion according to claim 3, characterized in that, Detection threshold The method for determining it is as follows: That is, extract multi-scale fusion metrics from the training data. The 99.9 percentile.

5. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the lithium-ion battery pack fault detection and location method based on cross-scale information fusion as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the lithium-ion battery pack fault detection and location method based on cross-scale information fusion as described in any one of claims 1 to 4.

Images