Method and system for detecting health state of power battery pack

Abstract

The invention relates to the technical field of power batteries, in particular to a power battery pack health state detection method and system. The power battery pack health state detection method comprises the following steps: acquiring open-circuit voltage change data and terminal voltage fluctuation data of a battery under micro current pulse excitation, recording a change curve of voltage along with time through high-precision time sequence acquisition equipment, and obtaining an initial data set of voltage dynamic response; acquiring magnetic field intensity distribution and magnetic field gradient change data of the surface of the battery in a weak alternating magnetic field, and acquiring spatial magnetic field information through a magnetic sensor array to obtain a three-dimensional distribution diagram reflecting magnetic field nonuniformity; through a three-dimensional distribution diagram of magnetic field intensity distribution and magnetic field gradient change, a defect detection algorithm is adopted to analyze a mapping relation between magnetic field non-uniformity and internal micro cracks, and correlation distribution of defect positions and internal resistance growth rates is determined.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries, and particularly to a method and system for detecting the state of health of a power battery pack. Background Art

[0002] The detection of the state of health of a power battery pack is a crucial research direction in the new energy field, which is directly related to the safety, performance stability, and service life of electric vehicles. With the widespread application of electric vehicles, the assessment of the state of health of the battery pack is not only the key to technological innovation but also the cornerstone for ensuring the efficient operation of the energy system. Accurately detecting the state of health of the battery can effectively prevent potential failures, extend the battery life, and promote the sustainable development of green energy. However, the research in this field still faces many challenges and there is an urgent need to break through the existing technical bottlenecks.

[0003] Currently, many methods for detecting the state of health mainly rely on single-dimensional monitoring means, such as indirectly inferring through changes in voltage or temperature. Although these methods are easy to operate, they often lack sufficient accuracy and reliability under complex working conditions. For example, traditional methods are difficult to comprehensively capture the subtle changes inside the battery caused by aging or defects, resulting in a deviation between the evaluation result and the actual state. In addition, existing solutions mostly focus on the measurement of surface parameters, and the excavation of the characteristics of deep electrochemical reactions inside the battery is insufficient, which limits the further improvement of detection technology.

[0004] In the detection of the state of health of a power battery pack, the core challenge focuses on how to accurately characterize the variation law of electromagnetic variables and their correlation with battery aging. Electrical variables such as open-circuit voltage, terminal voltage, internal resistance, and polarization voltage, as well as magnetic variables such as magnetic field strength, magnetic field gradient, and distribution uniformity, are key indicators reflecting the state of health of the battery. However, the dynamic responses of these variables under different working conditions are complex, and the mapping relationship between the non-uniformity of the magnetic field distribution and internal structure defects is not yet fully clear. This makes it difficult to establish an evaluation model that takes into account multiple variables and adapts to multiple scenarios. At the same time, existing technologies are still insufficient in integrating electromagnetic variables to improve detection accuracy, resulting in difficulty in accurately identifying potential risks during the battery aging process.

[0005] Therefore, how to analyze the change characteristics of electrical and magnetic variables during the charge and discharge process, reveal the correlation between the non-uniformity of the magnetic field distribution and internal defects of the battery, and build a reliable state of health evaluation model on this basis has become the key problem to be solved urgently in this research. Summary of the Invention

[0006] To solve the technical problem of how to analyze the variation characteristics of electrical and magnetic variables during the charging and discharging process, reveal the correlation between the magnetic field distribution non-uniformity and internal defects of the battery, and build a reliable health state assessment model on this basis, the present invention provides a method for detecting the health state of a power battery pack.

[0007] A method for detecting the health state of a power battery pack provided by the present invention mainly includes:

[0008] Obtain the open-circuit voltage change data and terminal voltage fluctuation data of the battery under a micro-current pulse excitation, record the voltage change curve over time through a high-precision timing acquisition device, and obtain the initial data set of the voltage dynamic response;

[0009] Extract the specific values of the internal resistance growth rate and polarization voltage offset from the initial data set of the voltage dynamic response, calculate the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determine the offset of the polarization voltage under high-temperature conditions;

[0010] Obtain the magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field, collect the spatial magnetic field information through a magnetic sensor array, and obtain a three-dimensional distribution map reflecting the magnetic field non-uniformity;

[0011] Integrate the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology. Adopt a method where the data input dimension includes multi-source data of voltage and magnetic field, determine the fusion weight distribution, and dynamically adjust the ratio according to the data reliability to obtain a comprehensive evaluation parameter after multi-dimensional feature dimensionality reduction;

[0012] If the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, then analyze the change trend of the voltage dynamic response baseline under high-temperature conditions through a convolutional neural network, judge the occurrence probability of the risk of micro-cracks inside the battery, and obtain a risk warning result;

[0013] According to the requirements of the risk warning result and the working condition classification label to distinguish high-temperature scenarios, adopt an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the accuracy improvement index;

[0014] Through the three-dimensional distribution map of the magnetic field intensity distribution and magnetic field gradient change, adopt a defect detection algorithm to analyze the relationship between the magnetic field non-uniformity and the mapping of internal micro-cracks, and determine the correlation distribution between the defect location and the internal resistance growth rate;

[0015] Obtain the correlation distribution between the defect location and the internal resistance growth rate and the accuracy improvement index, map the heterogeneous data to a unified dimension through parameter standardization processing, adopt an outlier rejection rule to filter out noise interference, and generate a final health state assessment report including the risk of micro-cracks.

[0016] The present invention provides a system for detecting the state of health of a power battery pack, mainly including:

[0017] A voltage data acquisition module, which is used to obtain the open-circuit voltage change data and terminal voltage fluctuation data of the battery under the excitation of a tiny current pulse, record the voltage change curve over time through a high-precision timing acquisition device, and obtain the initial data set of the voltage dynamic response;

[0018] An internal resistance polarization analysis module, which is used to extract the specific values of the internal resistance growth rate and polarization voltage offset from the initial data set of the voltage dynamic response, calculate the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determine the offset of the polarization voltage under high-temperature working conditions;

[0019] A magnetic field data acquisition module, which is used to obtain the magnetic field intensity distribution and magnetic field gradient change data on the surface of the battery under a weak alternating magnetic field, collect the spatial magnetic field information through a magnetic sensor array, and obtain a three-dimensional distribution map reflecting the magnetic field inhomogeneity;

[0020] A data fusion module, which is used to integrate the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology, adopt a method in which the data input dimension includes multi-source data of voltage and magnetic field, determine the fusion weight distribution, dynamically adjust the ratio according to the data reliability, and obtain the comprehensive evaluation parameters after dimensionality reduction of multi-dimensional features;

[0021] A risk assessment module, which is used to, if the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, analyze the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network, judge the occurrence probability of the risk of tiny cracks inside the battery, and obtain a risk warning result;

[0022] An accuracy optimization module, which is used to distinguish the requirements of high-temperature scenarios according to the risk warning result and the working condition classification label, adopt an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the accuracy improvement index;

[0023] A defect correlation analysis module, which is used to analyze the relationship between the magnetic field inhomogeneity and the mapping of tiny cracks inside through the three-dimensional distribution map of the magnetic field intensity distribution and magnetic field gradient change, and determine the correlation distribution between the defect position and the internal resistance growth rate;

[0024] A health assessment module, which is used to obtain the correlation distribution between the defect position and the internal resistance growth rate and the accuracy improvement index, map the heterogeneous data to a unified dimension through parameter standardization processing, filter out noise interference using the outlier rejection rule, and generate a final health state assessment report including the risk of tiny cracks.

[0025] Optionally, to obtain the open - circuit voltage change data and terminal voltage fluctuation data of the battery under the excitation of a micro - current pulse, record the voltage change curve over time through a high - precision timing acquisition device to obtain the initial data set of voltage dynamic response, including:

[0026] Record the voltage of the battery under the excitation of a micro - current pulse through the acquisition device to obtain the initial data set of open - circuit voltage and terminal voltage fluctuation;

[0027] Use a timing device to perform high - precision acquisition of the voltage change in the initial data set to obtain a time curve;

[0028] For the dynamic response in the time curve, extract the fluctuation characteristics of the open - circuit voltage and determine the voltage change trend;

[0029] If the voltage change trend exceeds the preset threshold, analyze the frequency components of the terminal voltage fluctuation through Fourier transform to obtain the frequency distribution;

[0030] According to the frequency distribution, judge the periodic characteristics in the dynamic response and obtain the influence range of the pulse excitation;

[0031] Extract the voltage change law corresponding to the micro - current from the influence range to obtain the adjusted data set;

[0032] Classify the adjusted data set through a support vector machine to judge the voltage stability of the battery.

[0033] Optionally, to extract the specific values of the internal resistance growth rate and polarization voltage offset from the initial data set of voltage dynamic response, calculate the growth trend of the internal resistance with the charge - discharge cycle through an impedance analysis algorithm, and determine the offset of the polarization voltage under high - temperature conditions, including:

[0034] Extract the voltage dynamic data from the initial data set through a preset voltage acquisition tool to obtain the specific values of the internal resistance growth rate and polarization voltage;

[0035] Use an impedance analysis algorithm to process the extracted specific values and calculate the change trend of the internal resistance with the charge - discharge cycle;

[0036] Judge whether the internal resistance growth rate exceeds the preset threshold according to the change trend. If it exceeds, record the abnormal points;

[0037] Obtain the voltage dynamic data under high - temperature conditions and calculate the offset of the polarization voltage;

[0038] Analyze the correlation between the polarization voltage and high - temperature conditions through the offset to determine the distribution range of the offset;

[0039] For the offset within the distribution range, use a machine - learning regression algorithm to predict the long - term trend of the polarization voltage;

[0040] Extract key features from the prediction results and determine the coupling relationship between internal resistance growth and polarization voltage offset.

[0041] Optionally, the method for obtaining the magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field includes: collecting spatial magnetic field information through a magnetic sensor array to obtain a three-dimensional distribution map reflecting magnetic field inhomogeneity, including:

[0042] Collect magnetic field information on the battery surface under an alternating magnetic field through a magnetic sensor array, obtain spatial magnetic field intensity and magnetic field gradient data, and obtain a preliminary magnetic field data set;

[0043] Process the preliminary magnetic field data set using a preset grid division method, determine the spatial distribution characteristics of the magnetic field intensity and magnetic field gradient, and generate structured magnetic field data;

[0044] Calculate the structured magnetic field data through an interpolation algorithm, obtain continuous magnetic field intensity distribution and magnetic field gradient change, and obtain a smoothed magnetic field data set;

[0045] Construct a three-dimensional coordinate system based on the smoothed magnetic field data set, and generate a three-dimensional distribution model reflecting the magnetic field intensity and gradient change;

[0046] If the magnetic field inhomogeneity in the three-dimensional distribution model exceeds a preset threshold, analyze the inhomogeneous region through a convolutional neural network, determine the position of the abnormal magnetic field distribution, and obtain abnormal distribution data;

[0047] Adjust the three-dimensional distribution model for the abnormal distribution data, obtain the optimized magnetic field inhomogeneity distribution, and obtain the final three-dimensional distribution map;

[0048] Process the final three-dimensional distribution map through a visualization tool to generate a complete spatial distribution image reflecting the magnetic field inhomogeneity on the battery surface.

[0049] Optionally, the method for integrating the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology includes: adopting a data input dimension including multi-source data of voltage and magnetic field, determining the fusion weight distribution and dynamically adjusting the ratio according to data reliability, and obtaining a comprehensive evaluation parameter after multi-dimensional feature dimensionality reduction, including:

[0050] Obtain the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through a multi-source data acquisition tool, and obtain an initial multi-dimensional data set;

[0051] Adopt a principal component analysis algorithm to extract the distribution characteristics of the open-circuit voltage, polarization voltage, and magnetic field intensity from the initial multi-dimensional data set, and obtain a dimensionality-reduced feature set;

[0052] For the dimension-reduced feature set, judge the reliability of each data source. If the reliability is higher than the preset threshold, assign a higher weight to determine the dynamic weight ratio.

[0053] Through the weighted average method, fuse the dimension-reduced feature set and the dynamic weight ratio to obtain the comprehensive feature parameters.

[0054] After obtaining the comprehensive feature parameters, use the support vector machine algorithm to judge the change trend of the parameters and determine the classification boundary.

[0055] According to the classification boundary, analyze the distribution characteristics of the comprehensive feature parameters to obtain the evaluation result.

[0056] Through the evaluation result, adjust the weight distribution ratio in data fusion to obtain the optimized comprehensive evaluation parameters.

[0057] Optionally, if the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, analyze the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network, judge the occurrence probability of the risk of micro-cracks inside the battery, and obtain the risk warning result, including:

[0058] By collecting battery operation data, obtain the comprehensive evaluation parameter value, and judge through the preset threshold comparison logic to determine the parameter exceeding status.

[0059] If the parameter exceeding status is established, process the voltage dynamic response data through a convolutional neural network to obtain the baseline change trend.

[0060] According to the baseline change trend, use the feature extraction technology under high-temperature working conditions to judge the potential distribution characteristics of micro-cracks.

[0061] Analyze the distribution characteristics of micro-cracks through a probability judgment model to obtain the risk occurrence probability value.

[0062] If the risk occurrence probability value exceeds the safe range, generate warning result data through logical comparison.

[0063] According to the warning result data, use the data fusion technology to integrate the comprehensive evaluation parameter and the voltage dynamic response to obtain the final risk assessment output.

[0064] Through the final risk assessment output, determine the dynamic monitoring result of the risk of micro-cracks inside the battery.

[0065] Optionally, according to the requirements of the risk warning result and the working condition classification label to distinguish high-temperature scenarios, use an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the accuracy improvement index, including:

[0066] Distinguish high-temperature scenarios through risk warning results and working condition classification labels, obtain the data fluctuation characteristics corresponding to high-temperature scenarios, and determine the fluctuation amplitude range;

[0067] Use an adaptive filtering window to adjust the smoothing range according to the fluctuation amplitude to obtain a smoothed data sequence;

[0068] Extract dynamic response characteristics from the smoothed data sequence and judge the response baseline offset;

[0069] Adjust the data sampling frequency through the offset to obtain optimized sampling data;

[0070] Calculate the accuracy improvement index for the optimized sampling data and determine the detection deviation rate;

[0071] If the detection deviation rate exceeds the preset threshold, adjust the size of the adaptive filtering window and re-obtain the smoothed data sequence;

[0072] Update the dynamic response baseline according to the adjusted smoothed data sequence to obtain the final detection deviation rate.

[0073] Optionally, for the three-dimensional distribution map of magnetic field intensity distribution and magnetic field gradient change, use a defect detection algorithm to analyze the relationship between magnetic field inhomogeneity and internal micro-crack mapping, and determine the correlation distribution between defect location and internal resistance growth rate, including:

[0074] Collect three-dimensional magnetic field intensity distribution data through a magnetic field sensor to obtain an initial magnetic field distribution map;

[0075] Process the initial magnetic field distribution map using a gradient calculation method to obtain a three-dimensional distribution map of gradient change;

[0076] Apply a defect detection algorithm to the three-dimensional distribution map of gradient change to identify magnetic field inhomogeneous regions and determine the distribution positions of micro-cracks;

[0077] Extract characteristic parameters from the distribution positions of micro-cracks to obtain spatial coordinate data of defect locations;

[0078] Collect internal resistance growth data through a resistance measurement device and judge the corresponding relationship between internal resistance growth and defect location;

[0079] Use a regression analysis algorithm to process the spatial coordinate data of defect locations and internal resistance growth data to determine the change trend of the correlation distribution;

[0080] Generate a three-dimensional visualization map according to the change trend of the correlation distribution to obtain the mapping result between defect location and internal resistance growth rate.

[0081] Optionally, to obtain the correlation distribution between the defect location and the internal resistance growth rate and the accuracy improvement index, heterogeneous data is mapped to a unified dimension through parameter standardization processing, and noise interference is filtered using an outlier rejection rule to generate a final health status assessment report including the risk of micro-cracks, including:

[0082] Map heterogeneous data to a unified dimension through parameter standardization processing to obtain a normalized data set;

[0083] Filter noise interference in the normalized data set using an outlier rejection rule to obtain a clean data set;

[0084] Extract the characteristic values of the defect location and internal resistance growth from the clean data set to determine the characteristic pairs;

[0085] Calculate the correlation distribution between the characteristic pairs through correlation analysis to obtain a distribution matrix;

[0086] If the correlation value in the distribution matrix exceeds the preset threshold, judge the risk of micro-cracks through the support vector machine algorithm to obtain a risk classification result;

[0087] Calculate the corresponding relationship between internal resistance growth and accuracy improvement for the risk classification result to obtain a health status quantification index;

[0088] Generate an assessment result including the risk of micro-cracks according to the health status quantification index.

[0089] On the other hand, the present application also provides a health status detection system for a power battery pack, and the system includes:

[0090] A voltage data acquisition module for obtaining the open-circuit voltage change data and terminal voltage fluctuation data of the battery under a small current pulse excitation, recording the voltage change curve over time through a high-precision timing acquisition device, and obtaining an initial data set of the voltage dynamic response;

[0091] An internal resistance polarization analysis module for extracting the specific values of the internal resistance growth rate and polarization voltage offset from the initial data set of the voltage dynamic response, calculating the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determining the offset of the polarization voltage under high-temperature conditions;

[0092] A magnetic field data acquisition module for obtaining the magnetic field intensity distribution and magnetic field gradient change data on the surface of the battery under a weak alternating magnetic field, collecting spatial magnetic field information through a magnetic sensor array, and obtaining a three-dimensional distribution map reflecting the magnetic field inhomogeneity;

[0093] A data fusion module, which is used to integrate open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology. By adopting a method where the data input dimension includes multi-source data of voltage and magnetic field, it determines the fusion weight distribution and dynamically adjusts the ratio according to data reliability, and obtains a comprehensive evaluation parameter after dimensionality reduction of multi-dimensional features;

[0094] A risk assessment module, which is used to analyze the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network if the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, judge the occurrence probability of the risk of micro-cracks inside the battery, and obtain a risk warning result;

[0095] An accuracy optimization module, which is used to distinguish the requirements of high-temperature scenarios according to the risk warning result and the working condition classification label, adopt an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the accuracy improvement index;

[0096] A defect correlation analysis module, which is used to analyze the relationship between magnetic field inhomogeneity and internal micro-crack mapping through a three-dimensional distribution map of magnetic field intensity distribution and magnetic field gradient change, and determine the correlation distribution between the defect position and the internal resistance growth rate;

[0097] A health assessment module, which is used to obtain the correlation distribution between the defect position and the internal resistance growth rate and the accuracy improvement index, map heterogeneous data to a unified dimension through parameter standardization processing, filter out noise interference using the outlier rejection rule, and generate a final health status assessment report including the risk of micro-cracks.

[0098] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0099] The present invention discloses a method for detecting the health status of a power battery pack, specifically a method for realizing early warning of the risk of micro-cracks in the battery based on electromagnetic feature fusion. This method obtains the voltage dynamic response data of the battery under a small current pulse excitation and the magnetic field distribution data under a weak alternating magnetic field, and uses electromagnetic variable fusion technology to integrate multi-source information to obtain a comprehensive evaluation parameter. When the parameter exceeds the preset threshold, a convolutional neural network is used to analyze the change trend of the voltage response baseline to judge the risk of internal micro-cracks. Further, by combining the analysis of magnetic field inhomogeneity, the correlation distribution between the defect position and the internal resistance growth rate is determined, and finally a health status assessment report including the risk of micro-cracks is generated. The present invention realizes the early detection and risk warning of internal micro-cracks in the battery through the fusion of multi-dimensional features of voltage and magnetic field, and improves the accuracy and reliability of the battery safety assessment. Description of the Drawings

[0100] Figure 1It is a flowchart of a method for detecting the state of health of a power battery pack according to the present invention.

[0101] Figure 2 It is a schematic diagram of a method and system for detecting the state of health of a power battery pack according to the present invention.

[0102] Figure 3 It is another schematic diagram of a method and system for detecting the state of health of a power battery pack according to the present invention.

[0103] Figure 4 It is a structural schematic diagram of a method and system for detecting the state of health of a power battery pack according to the present invention. Detailed implementation manners

[0104] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0105] As Figures 1-4 , a method and system for detecting the state of health of a power battery pack in this embodiment may specifically include:

[0106] S101. Obtain the open-circuit voltage change data and terminal voltage fluctuation data of the battery under a small current pulse excitation, record the voltage change curve over time through a high-precision timing acquisition device, and obtain the initial data set of the voltage dynamic response.

[0107] Record the voltage of the battery under a small current pulse excitation through a collection device to obtain the initial data set of the open-circuit voltage and terminal voltage fluctuation. Use a timing device to perform high-precision acquisition of the voltage change in the initial data set to obtain a time curve. For the dynamic response in the time curve, extract the fluctuation characteristics of the open-circuit voltage, and determine the voltage change trend. If the voltage change trend exceeds a preset threshold, analyze the frequency components of the terminal voltage fluctuation through Fourier transform to obtain the frequency distribution. According to the frequency distribution, judge the periodic characteristics in the dynamic response, and obtain the influence range of the pulse excitation. Extract the voltage change rule corresponding to the small current from the influence range to obtain the adjusted data set. Classify the adjusted data set through a support vector machine to judge the voltage stability of the battery.

[0108] Exemplarily, when obtaining the open-circuit voltage change data and terminal voltage fluctuation data of the battery under a micro-current pulse excitation, it is first necessary to design a micro-current pulse excitation source, whose current amplitude is usually set to 0.1 mA, the pulse width is 10 ms, and the repetition frequency is 1 Hz. Through a high-precision timing acquisition device, such as a 16-bit precision data acquisition card, the voltage change curve over time is recorded at a rate of 1000 sampling points per second. The collected data is processed by low-pass filtering, and the cut-off frequency is set to 100 Hz to remove high-frequency noise. Next, the fast Fourier transform (FFT) algorithm is used to perform frequency-domain analysis on the filtered data to extract the main frequency components of the voltage dynamic response. By calculating the root mean square (RMS) value of the voltage change, the amplitude of the terminal voltage fluctuation is evaluated, and usually the RMS value is within 0.5 mV. Finally, the processed data is stored in CSV format for subsequent analysis and modeling. The whole process is implemented through an automated script to ensure the continuity and consistency of data acquisition and processing.

[0109] S102. Extract the specific values of the internal resistance growth rate and polarization voltage offset from the initial dataset of the voltage dynamic response, calculate the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determine the offset of the polarization voltage under high-temperature conditions.

[0110] Extract the voltage dynamic data from the initial dataset through a preset voltage acquisition tool to obtain the specific values of the internal resistance growth rate and polarization voltage. Use an impedance analysis algorithm to process the extracted specific values and calculate the change trend of the internal resistance with the charge and discharge cycles. Judge whether the internal resistance growth rate exceeds a preset threshold according to the change trend. If it exceeds, record the abnormal points. Obtain the voltage dynamic data under high-temperature conditions and calculate the offset of the polarization voltage. Analyze the correlation between the polarization voltage and high-temperature conditions through the offset to determine the distribution range of the offset. For the offset within the distribution range, use a machine learning regression algorithm to predict the long-term trend of the polarization voltage. Extract the key features from the prediction results and judge the coupling relationship between the internal resistance growth and the polarization voltage offset.

[0111] Exemplarily, to extract the specific values of the internal resistance growth rate and polarization voltage offset from the initial dataset of the voltage dynamic response, first perform filtering and denoising processing on the original voltage data through data preprocessing techniques to ensure the accuracy of the data.

[0112] For example, the Kalman filtering algorithm is used to smooth the voltage data and remove high-frequency noise. Then, the least squares method is adopted to fit the voltage curve to extract the specific values of the internal resistance growth rate and the polarization voltage offset. Assuming the initial internal resistance is 0.1 ohm, after 100 charge-discharge cycles, the internal resistance increases to 0.15 ohm, and the internal resistance growth rate is 0.0005 ohm per cycle. Through impedance analysis algorithms such as electrochemical impedance spectroscopy (EIS) analysis, the growth trend of the internal resistance with the charge-discharge cycles is calculated. The specific algorithm includes establishing an equivalent circuit model and fitting the EIS data through the nonlinear least squares method to obtain the curve of the internal resistance varying with the number of cycles.

[0113] For example, after 500 cycles, the internal resistance increases to 0.25 ohm, and the growth trend conforms to a linear model. Determine the offset of the polarization voltage under high-temperature conditions. By comparing the voltage curves at room temperature and high temperature, calculate the difference in the polarization voltage. Assume that the polarization voltage is 3.7 volts at 25°C and shifts to 3.65 volts at 60°C, with an offset of 0.05 volts. Through the above methods, the specific values of the internal resistance growth rate and the polarization voltage offset can be accurately extracted, and their trends with the charge-discharge cycles and temperature changes can be analyzed.

[0114] S103. Obtain the magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field, collect the spatial magnetic field information through a magnetic sensor array, and obtain a three-dimensional distribution map reflecting the magnetic field inhomogeneity.

[0115] Collect the magnetic field information on the battery surface under an alternating magnetic field through a magnetic sensor array, obtain the spatial magnetic field intensity and magnetic field gradient data, and obtain a preliminary magnetic field data set. Process the preliminary magnetic field data set using a preset grid division method to determine the spatial distribution characteristics of the magnetic field intensity and magnetic field gradient, and generate structured magnetic field data. Calculate the structured magnetic field data through an interpolation algorithm to obtain a continuous magnetic field intensity distribution and magnetic field gradient change, and obtain a smoothed magnetic field data set. Construct a three-dimensional coordinate system based on the smoothed magnetic field data set to generate a three-dimensional distribution model reflecting the magnetic field intensity and gradient changes. If the magnetic field inhomogeneity in the three-dimensional distribution model exceeds a preset threshold, analyze the inhomogeneous region through a convolutional neural network, determine the position of the abnormal magnetic field distribution, and obtain the abnormal distribution data. Adjust the three-dimensional distribution model according to the abnormal distribution data to obtain an optimized magnetic field inhomogeneity distribution, and obtain the final three-dimensional distribution map. Process the final three-dimensional distribution map through a visualization tool to generate a complete spatial distribution image reflecting the magnetic field inhomogeneity on the battery surface.

[0116] Exemplarily, in order to obtain the magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field, it is first necessary to design a highly sensitive magnetic sensor array. For example, 16 TMR sensors are arranged in a 4×4 square matrix, and the sensor spacing is set to 5 mm to ensure a spatial resolution of 0.1 mT. The sensor array synchronously collects data through the SPI interface at a sampling frequency of 1 kHz. The output of each sensor is transmitted to the FPGA for real-time preprocessing after being converted by a 24-bit ADC, including baseline correction (using moving average filtering with a window length of 100 points) and noise suppression (using wavelet transform threshold denoising, selecting the db4 wavelet basis and a decomposition level of 5). The preprocessed data reconstructs the spatial magnetic field distribution through a three-dimensional interpolation algorithm (such as Kriging interpolation with a Gaussian-type variogram model and a nugget value of 0.05) to generate a three-dimensional grid model with a resolution of 0.5 mm. For magnetic field gradient calculation, the central difference method is used to calculate the magnetic field change rate between adjacent grid points. For example, in the x-direction gradient

[0117] ΔBx / Δx = (B(x + 0.5) - B(x - 0.5)) / 1.0, in millimeters, and the second-order gradient is calculated simultaneously to quantify local non-uniformity. To analyze the magnetic field anomaly region, threshold conditions (such as the absolute value of the gradient exceeding 0.3 mT / mm) are set to mark potential defect positions, and morphological opening operations are combined to eliminate artifacts caused by noise. Finally, the magnetic field intensity and gradient distributions are rendered through a three-dimensional visualization platform (such as the VTK library), and HSV color mapping (intensity corresponding to brightness and gradient direction corresponding to hue) is used to achieve multi-parameter fusion display. At the same time, the maximum gradient value (such as 0.45 mT / mm) and spatial standard deviation (such as 0.12) are output as quantization indicators. The entire process is integrated through the LabVIEW system to achieve a closed-loop processing from data acquisition to analysis.

[0118] S104. Integrate the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology. Adopt a data input dimension that includes multi-source data of voltage and magnetic field, and determine that the fusion weight allocation dynamically adjusts the ratio according to data reliability to obtain a comprehensive evaluation parameter after multi-dimensional feature dimensionality reduction.

[0119] Through a multi-source data acquisition tool, data on open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field strength distribution are obtained to form an initial multi-dimensional dataset. Using the principal component analysis algorithm, the distribution characteristics of open-circuit voltage, polarization voltage, and magnetic field strength are extracted from the initial multi-dimensional dataset to obtain a reduced-dimensional feature set. For the reduced-dimensional feature set, the reliability of each data source is judged. If the reliability is higher than a preset threshold, a higher weight is assigned to determine the dynamic weight ratio. Through the weighted average method, the reduced-dimensional feature set and the dynamic weight ratio are fused to obtain a comprehensive feature parameter. After obtaining the comprehensive feature parameter, the support vector machine algorithm is used to judge the change trend of the parameter and determine the classification boundary. Based on the classification boundary, the distribution characteristics of the comprehensive feature parameter are analyzed to obtain an evaluation result. Through the evaluation result, the weight distribution ratio in data fusion is adjusted to obtain an optimized comprehensive evaluation parameter.

[0120] Exemplarily, in the electromagnetic variable fusion technology, first, data on open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field strength distribution are collected through sensors. For example, the open-circuit voltage changes from 3.2V to 3.5V, the internal resistance growth rate increases from 0.1Ω / h to 0.15Ω / h, the polarization voltage offset increases from 0.02V to 0.03V, and the magnetic field strength distribution data shows a gradient change from 0.5T to 0.8T in space. These data are integrated through a multi-source data input module, and the principal component analysis (PCA) algorithm is used to reduce the dimension of the voltage and magnetic field data to extract the main feature vectors. For example, the original 10-dimensional data is reduced to 3 dimensions by PCA, retaining 95% of the variance information. Then, the fusion weights are dynamically adjusted according to the data reliability. For example, the reliability of the open-circuit voltage data is 0.9, the internal resistance growth rate is 0.8, the polarization voltage offset is 0.85, and the magnetic field strength distribution is 0.75. The comprehensive evaluation parameter is calculated by the weighted average method, with the weight distribution being 0.3 for the open-circuit voltage, 0.25 for the internal resistance growth rate, 0.25 for the polarization voltage offset, and 0.2 for the magnetic field strength distribution, and finally the comprehensive evaluation parameter is 0.87. This parameter can be used for the evaluation of the battery health state. For example, when the comprehensive evaluation parameter is lower than 0.8, the system will issue a warning signal indicating that the battery needs maintenance or replacement. The whole process ensures the accuracy and real-time nature of the evaluation result through automated data processing and analysis.

[0121] S105. If the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, then the change trend of the voltage dynamic response baseline under high-temperature working conditions is analyzed through a convolutional neural network, and the occurrence probability of the risk of micro-cracks inside the battery is judged to obtain a risk warning result.

[0122] By collecting battery operation data, obtaining comprehensive evaluation parameter values, performing comparison logic judgment through preset thresholds, and determining the parameter exceeding status. If the parameter exceeding status is established, the voltage dynamic response data is processed by a convolutional neural network to obtain the baseline change trend. According to the baseline change trend, feature extraction technology under high-temperature conditions is used to judge the potential distribution characteristics of microcracks. The distribution characteristics of microcracks are analyzed through a probability judgment model to obtain the risk occurrence probability value. If the risk occurrence probability value exceeds the safe range, warning result data is generated through logical comparison. According to the warning result data, data fusion technology is used to integrate the comprehensive evaluation parameters and voltage dynamic response to obtain the final risk assessment output. Through the final risk assessment output, the dynamic monitoring result of the microcrack risk inside the battery is determined.

[0123] Exemplarily, when the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, the system first compares the parameters such as voltage, temperature, and current collected in real time with the preset threshold through a preset threshold comparison algorithm. For example, when the voltage fluctuation exceeds ±5% or the temperature exceeds 60°C, the system determines that the threshold is exceeded. Subsequently, the system calls a convolutional neural network (CNN) model, which has been trained to analyze the change trend of the voltage dynamic response baseline under high-temperature conditions.

[0124] Specifically, the CNN model extracts features in the voltage signal through multiple convolutional and pooling operations, and combines historical data under high-temperature conditions to predict the change trend of the voltage dynamic response.

[0125] For example, when the CNN model predicts that the voltage dynamic response baseline shows a continuous downward trend at high temperature and the downward amplitude exceeds 3%, the system will further analyze the occurrence probability of the microcrack risk inside the battery. By introducing a Bayesian probability model, the system combines the prediction result of the CNN model and the historical failure data of the battery to calculate the occurrence probability of the microcrack risk. For example, when the probability exceeds 80%, the system generates a risk warning result and automatically triggers the warning mechanism to notify relevant personnel for processing. The whole process is automatically completed through information technology to ensure the accuracy and timeliness of evaluation and warning.

[0126] S106. According to the requirements of the high-temperature scenario differentiated by the risk warning result and the working condition classification label, an adaptive filtering window is used to adjust the smoothing range according to the data fluctuation amplitude, the data sampling frequency is optimized for the dynamic response baseline, and the detection deviation rate reflected by the accuracy improvement index is determined.

[0127] Distinguish high-temperature scenarios based on risk warning results and operating condition classification labels, obtain the data fluctuation characteristics corresponding to high-temperature scenarios, and determine the fluctuation amplitude range. Use an adaptive filtering window to adjust the smoothing range according to the fluctuation amplitude to obtain a smoothed data sequence. Extract dynamic response characteristics from the smoothed data sequence and judge the response baseline offset. Adjust the data sampling frequency through the offset to obtain optimized sampled data. Calculate the accuracy improvement index for the optimized sampled data to determine the detection deviation rate. If the detection deviation rate exceeds the preset threshold, adjust the size of the adaptive filtering window and re-obtain the smoothed data sequence. Update the dynamic response baseline according to the adjusted smoothed data sequence to obtain the final detection deviation rate.

[0128] Exemplarily, in the risk warning of high-temperature scenarios, first divide the temperature data into two categories: steady state (fluctuation within ±2°C) and transient (fluctuation exceeding ±5°C) through the operating condition classification label. For steady-state data, use a moving average filter with a window size of 10, while for transient data, use an adaptive Kalman filter, and its process noise covariance Q is dynamically adjusted according to the real-time fluctuation amplitude. For example, when the temperature gradient reaches 3°C / min, Q is adjusted from 0.1 to 0.5. The data sampling frequency is dynamically optimized through FFT analysis. When the detected main frequency components are concentrated in 0 - 0.2 Hz, the sampling rate is reduced from 1 Hz to 0.5 Hz to reduce redundant data, and cubic spline interpolation is used to ensure time alignment. The accuracy improvement index is achieved by calculating the root mean square error (RMSE) between the predicted value and the measured value within the sliding window. When the RMSE within a 30-second window exceeds 1.5°C, baseline correction is triggered, and the recursive least squares method with a forgetting factor is used to update the model parameters. The forgetting factor λ is set to 0.95 to balance the weights of new and old data. For the calculation of the detection deviation rate, a probability statistical method based on the confidence interval is introduced. When the temperature reading exceeds the 95% confidence interval (μ ± 2σ) of the historical data and lasts for more than 5 sampling points, it is determined as an effective deviation and the number of deviations is recorded. The final deviation rate is output as (number of deviations / total number of samples) × 100%. At the same time, the Pearson correlation coefficient is used to verify the correlation between the warning result and the equipment failure record, and it is required to reach above 0.7 to confirm the effectiveness of the model.

[0129] S107. Through the three-dimensional distribution map of the magnetic field strength distribution and the magnetic field gradient change, use the defect detection algorithm to analyze the relationship between the magnetic field inhomogeneity and the mapping of internal micro-cracks, and determine the associated distribution of the defect position and the internal resistance growth rate.

[0130] Collect three-dimensional magnetic field intensity distribution data through a magnetic field sensor to obtain an initial magnetic field distribution map. Use a gradient calculation method to process the initial magnetic field distribution map and obtain a three-dimensional distribution map of gradient changes. For the three-dimensional distribution map of gradient changes, apply a defect detection algorithm to identify the magnetic field non-uniform region and determine the distribution position of micro-cracks. Extract characteristic parameters from the distribution position of micro-cracks to obtain the spatial coordinate data of the defect position. Collect internal resistance growth data through a resistance measurement device and judge the corresponding relationship between the internal resistance growth and the defect position. Use a regression analysis algorithm to process the spatial coordinate data of the defect position and the internal resistance growth data to determine the change trend of the associated distribution. According to the change trend of the associated distribution, generate a three-dimensional visualization map and obtain the mapping result of the defect position and the internal resistance growth rate.

[0131] Exemplarily, establish a magnetic field intensity distribution model through three-dimensional finite element simulation, set the background magnetic field intensity to 50 mT, use COMSOL Multiphysics software to calculate the spatial gradient distribution, and set the mesh size to 0.1 mm×0.1 mm×0.1 mm. When there is a crack with a depth of 0.5 mm and a width of 20 μm, a magnetic field intensity mutation will occur within 5 mm of the crack edge, and the gradient value will rise from 0.3 T / m in the normal region to 1.8 T / m. Use an improved Canny edge detection algorithm to process the magnetic field gradient data, set the Gaussian filter kernel σ = 0.5, and the double thresholds are 30% and 60% of the gradient amplitude respectively, and the defect contour can be accurately extracted. Import the detected defect position coordinates into ANSYS Workbench for multi-physics coupling analysis. When the stress concentration coefficient at the crack tip reaches 3.2, there is an exponential relationship between the corresponding internal resistance growth rate: when the crack length increases from 100 μm to 500 μm, the internal resistance rises from the initial value of 2.1 mΩ to 8.7 mΩ, and the correlation coefficient of the fitting curve R = 2.1e^(0.0047L) reaches 0.98. By establishing a three-dimensional convolutional neural network model, the input layer receives magnetic field tensor data of 256×256×256, uses a 3×3×3 convolutional kernel and a ReLU activation function, and the output layer realizes defect probability prediction through a Sigmoid function, achieving a positioning accuracy of 94.3% on the test set, and the position error is controlled within the range of ±0.15 mm. Combining with X-ray tomography verification, it is found that the spatial matching degree between the magnetic field abnormal region and the real defect reaches 91.2%, which verifies the effectiveness of this method.

[0132] S108. Obtain the associated distribution and accuracy improvement index of the defect position and the internal resistance growth rate, map heterogeneous data to a unified dimension through parameter standardization processing, use an outlier rejection rule to filter out noise interference, and generate a final health status assessment report including the risk of micro-cracks.

[0133] Heterogeneous data is mapped to a unified dimension through parameter standardization to obtain a normalized data set. The noise interference in the normalized data set is filtered by using an outlier rejection rule to obtain a clean data set. The eigenvalue of the defect position and the internal resistance growth is extracted from the clean data set to determine the feature pair. The correlation distribution between the feature pairs is calculated through correlation analysis to obtain a distribution matrix. If the correlation value in the distribution matrix exceeds a preset threshold, the risk of microcracks is judged by the support vector machine algorithm to obtain a risk classification result. The corresponding relationship between the internal resistance growth and the accuracy improvement is calculated for the risk classification result to obtain a health status quantization index. An evaluation result including the risk of microcracks is generated according to the health status quantization index.

[0134] Exemplarily, when obtaining the correlation distribution between the defect position and the internal resistance growth rate, first, the position data of the internal defects of the battery are collected by a high-precision sensor, and the change in the internal resistance is measured in combination with the electrochemical impedance spectroscopy (EIS).

[0135] For example, during the cyclic charge and discharge process of a certain battery, the internal resistance at defect position A increases from an initial value of 0.05 Ω to 0.08 Ω, while the internal resistance at position B increases from 0.06 Ω to 0.12 Ω. Through linear regression analysis, it is found that the internal resistance growth rate at position A is 0.003 Ω / cycle, and at position B is 0.006 Ω / cycle, indicating that the defect at position B has a greater impact on the battery performance. To improve the accuracy, the least squares method is used to fit the data, and the fitting error is controlled within ±0.001 Ω. Then, through parameter standardization, the defect position coordinates and the internal resistance growth rate are mapped to the range of 0 to 1. For example, the coordinates of position A (x = 10, y = 20) are standardized to (0.2, 0.4), and the internal resistance growth rate of 0.003 Ω / cycle is standardized to 0.3. To filter out noise interference, the 3σ criterion is used to reject outliers. For example, in a certain measurement, the internal resistance growth rate reaches 0.015 Ω / cycle, which is significantly deviated from the mean value of 0.0045 Ω / cycle and is determined as an outlier and rejected. Finally, combined with a microcrack detection algorithm, such as edge detection technology based on image processing, microcracks with a width less than 0.1 mm are identified near the defect position, and a final health status evaluation report is generated. The report shows that the defect risk level at position B is high, and it is recommended to repair it first.

[0136] The present invention provides a health status detection system for a power battery pack, mainly including:

[0137] A voltage data acquisition module, configured to obtain the open-circuit voltage change data and the terminal voltage fluctuation data of the battery under a micro-current pulse excitation, record the voltage change curve over time through a high-precision timing acquisition device, and obtain an initial data set of the voltage dynamic response;

[0138] The internal resistance polarization analysis module is used to extract the specific values of the internal resistance growth rate and the polarization voltage offset from the initial dataset of the voltage dynamic response, calculate the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determine the offset of the polarization voltage under high-temperature working conditions;

[0139] The magnetic field data acquisition module is used to obtain the magnetic field intensity distribution and the magnetic field gradient change data on the battery surface under a weak alternating magnetic field, collect the spatial magnetic field information through a magnetic sensor array, and obtain a three-dimensional distribution map reflecting the magnetic field non-uniformity;

[0140] The data fusion module is used to integrate the open-circuit voltage change, the internal resistance growth rate, the polarization voltage offset, and the magnetic field intensity distribution data through electromagnetic variable fusion technology. By adopting a data input dimension that includes multi-source data of voltage and magnetic field, it determines the fusion weight allocation and dynamically adjusts the ratio according to the data reliability, and obtains the comprehensive evaluation parameters after dimensionality reduction of multi-dimensional features;

[0141] The risk assessment module is used to, if the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, analyze the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network, judge the occurrence probability of the risk of micro-cracks inside the battery, and obtain the risk warning result;

[0142] The precision optimization module is used to distinguish the requirements of high-temperature scenarios according to the risk warning result and the working condition classification label, adopt an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the precision improvement index;

[0143] The defect correlation analysis module is used to, through the three-dimensional distribution map of the magnetic field intensity distribution and the magnetic field gradient change, analyze the relationship between the magnetic field non-uniformity and the mapping of internal micro-cracks by using a defect detection algorithm, and determine the correlation distribution between the defect position and the internal resistance growth rate;

[0144] The health assessment module is used to obtain the correlation distribution between the defect position and the internal resistance growth rate and the precision improvement index, map the heterogeneous data to a unified dimension through parameter standardization processing, filter out the noise interference by using the outlier rejection rule, and generate the final health status assessment report including the risk of micro-cracks.

[0145] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting the state of health of a power battery pack, characterized in that, The method includes: Obtain the open-circuit voltage change data and terminal voltage fluctuation data of the battery under the excitation of a micro-current pulse. Record the voltage change curve over time through a high-precision timing acquisition device to obtain the initial data set of the voltage dynamic response; Extract the specific values of the internal resistance growth rate and polarization voltage offset from the initial data set of the voltage dynamic response. Calculate the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determine the offset of the polarization voltage under high-temperature working conditions; Obtain the magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field. Collect the spatial magnetic field information through a magnetic sensor array to obtain a three-dimensional distribution map reflecting the magnetic field inhomogeneity; Integrate the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology. Adopt a method in which the data input dimension includes multi-source data of voltage and magnetic field, and determine that the fusion weight distribution dynamically adjusts the ratio according to the data reliability to obtain the comprehensive evaluation parameter after multi-dimensional feature dimensionality reduction; If the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, analyze the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network, judge the occurrence probability of the risk of micro-cracks inside the battery, and obtain the risk warning result; According to the risk warning result and the requirements of the working condition classification label to distinguish high-temperature scenarios, adopt an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the accuracy improvement index; Through the three-dimensional distribution map of the magnetic field intensity distribution and magnetic field gradient change, adopt a defect detection algorithm to analyze the relationship between the magnetic field inhomogeneity and the mapping of internal micro-cracks, and determine the correlation distribution between the defect location and the internal resistance growth rate; Obtain the correlation distribution between the defect location and the internal resistance growth rate and the accuracy improvement index. Map the heterogeneous data to a unified dimension through parameter standardization processing, and filter out noise interference by adopting an outlier rejection rule to generate a final health status evaluation report including the risk of micro-cracks; 2. The method according to claim 1, wherein The obtaining of the open-circuit voltage change data and terminal voltage fluctuation data of the battery under the excitation of a micro-current pulse, and recording the voltage change curve over time through a high-precision timing acquisition device to obtain the initial data set of the voltage dynamic response includes: Record the voltage of the battery under the excitation of a micro-current pulse through a collection device to obtain the initial data set of the open-circuit voltage and terminal voltage fluctuation; Adopt a timing device to perform high-precision acquisition of the voltage change in the initial data set to obtain a time curve; For the dynamic response in the time curve, extract the fluctuation characteristics of the open-circuit voltage and determine the voltage change trend; If the voltage change trend exceeds the preset threshold, analyze the frequency components of the terminal voltage fluctuation through Fourier transform to obtain the frequency distribution; According to the frequency distribution, judge the periodic characteristics in the dynamic response and obtain the influence range of the pulse excitation; Extract the voltage change law corresponding to the micro-current from the influence range to obtain the adjusted data set; Classify the adjusted data set through a support vector machine to judge the voltage stability of the battery.

3. The method according to claim 1, characterized in that, Specific values of the internal resistance growth rate and polarization voltage offset are extracted from the initial dataset of the voltage dynamic response, and the growth trend of the internal resistance with charge and discharge cycles is calculated through an impedance analysis algorithm to determine the offset of the polarization voltage under high-temperature conditions, including: Extract voltage dynamic data from the initial dataset through a preset voltage acquisition tool to obtain specific values of the internal resistance growth rate and polarization voltage; Use an impedance analysis algorithm to process the extracted specific values and calculate the change trend of the internal resistance with charge and discharge cycles; Judge whether the internal resistance growth rate exceeds a preset threshold according to the change trend. If it exceeds, record the abnormal point; Obtain voltage dynamic data under high-temperature conditions and calculate the offset of the polarization voltage; Analyze the correlation between the polarization voltage and the high-temperature condition through the offset to determine the distribution range of the offset; For the offset within the distribution range, use a machine learning regression algorithm to predict the long-term trend of the polarization voltage; Extract key features from the prediction results and judge the coupling relationship between the internal resistance growth and the polarization voltage offset.

4. The method according to claim 1, characterized in that, The magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field are obtained, and spatial magnetic field information is collected through a magnetic sensor array to obtain a three-dimensional distribution map reflecting magnetic field non-uniformity, including: Collect magnetic field information on the battery surface under an alternating magnetic field through a magnetic sensor array to obtain spatial magnetic field intensity and magnetic field gradient data, and obtain a preliminary magnetic field dataset; Process the preliminary magnetic field dataset using a preset grid division method to determine the spatial distribution characteristics of the magnetic field intensity and magnetic field gradient, and generate structured magnetic field data; Calculate the structured magnetic field data through an interpolation algorithm to obtain a continuous magnetic field intensity distribution and magnetic field gradient change, and obtain a smooth magnetic field dataset; Construct a three-dimensional coordinate system based on the smooth magnetic field dataset to generate a three-dimensional distribution model reflecting the magnetic field intensity and gradient changes; If the magnetic field non-uniformity in the three-dimensional distribution model exceeds a preset threshold, analyze the non-uniformity region through a convolutional neural network to judge the abnormal distribution position of the magnetic field and obtain abnormal distribution data; Adjust the three-dimensional distribution model for the abnormal distribution data to obtain an optimized magnetic field non-uniformity distribution and obtain the final three-dimensional distribution map; Process the final three-dimensional distribution map through a visualization tool to generate a complete spatial distribution image reflecting the magnetic field non-uniformity on the battery surface.

5. The method according to claim 1, characterized in that, The open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data are integrated through an electromagnetic variable fusion technology. In a way that the data input dimension includes multi-source data of voltage and magnetic field, the fusion weight allocation is determined and the ratio is dynamically adjusted according to the data reliability to obtain a comprehensive evaluation parameter after multi-dimensional feature dimensionality reduction, including: Obtain open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through a multi-source data acquisition tool to obtain an initial multi-dimensional dataset; Use a principal component analysis algorithm to extract the distribution characteristics of the open-circuit voltage, polarization voltage, and magnetic field intensity from the initial multi-dimensional dataset to obtain a dimensionality-reduced feature set; For the dimensionality-reduced feature set, judge the reliability of each data source. If the reliability is higher than a preset threshold, assign a higher weight to determine the dynamic weight ratio; By using the weighted average method, fuse the feature set after dimensionality reduction and the dynamic weight ratio to obtain the comprehensive feature parameters; After obtaining the comprehensive feature parameters, use the support vector machine algorithm to judge the change trend of the parameters and determine the classification boundary; According to the classification boundary, analyze the distribution characteristics of the comprehensive feature parameters to obtain the evaluation result; Based on the evaluation result, adjust the weight distribution ratio in data fusion to obtain the optimized comprehensive evaluation parameters.

6. The method according to claim 1, characterized in that, If the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, then analyze the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network, judge the occurrence probability of the risk of micro-cracks inside the battery, and obtain the risk warning result, including: Collect the battery operation data to obtain the value of the comprehensive evaluation parameter, and perform a comparison logic judgment through the preset threshold to determine the parameter exceeding status; If the parameter exceeding status is established, process the voltage dynamic response data through a convolutional neural network to obtain the baseline change trend; According to the baseline change trend, adopt the feature extraction technology under high-temperature working conditions to judge the potential distribution characteristics of micro-cracks; Analyze the distribution characteristics of micro-cracks through the probability judgment model to obtain the risk occurrence probability value; If the risk occurrence probability value exceeds the safe range, generate warning result data through logical comparison; According to the warning result data, use the data fusion technology to integrate the comprehensive evaluation parameter and the voltage dynamic response to obtain the final risk assessment output; Determine the dynamic monitoring result of the risk of micro-cracks inside the battery through the final risk assessment output.

7. The method according to claim 1, characterized in that, According to the requirements of distinguishing high-temperature scenarios based on the risk warning result and the working condition classification label, adopt an adaptive filtering window to adjust the smoothing range according to the data fluctuation amplitude, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the accuracy improvement index, including: Distinguish high-temperature scenarios through the risk warning result and the working condition classification label, obtain the data fluctuation characteristics corresponding to the high-temperature scenario, and determine the fluctuation amplitude range; Adopt an adaptive filtering window to adjust the smoothing range according to the fluctuation amplitude to obtain the smoothed data sequence; Extract the dynamic response characteristics from the smoothed data sequence and judge the response baseline offset; Adjust the data sampling frequency through the offset to obtain the optimized sampling data; Calculate the accuracy improvement index for the optimized sampling data to determine the detection deviation rate; If the detection deviation rate exceeds the preset threshold, adjust the size of the adaptive filtering window and re-obtain the smoothed data sequence; Update the dynamic response baseline according to the adjusted smoothed data sequence to obtain the final detection deviation rate.

8. The method according to claim 1, characterized in that, Through the three-dimensional distribution map of the magnetic field intensity distribution and the magnetic field gradient change, adopt the defect detection algorithm to analyze the relationship between the magnetic field inhomogeneity and the mapping of internal micro-cracks, and determine the associated distribution of the defect position and the internal resistance growth rate, including: Collect the three-dimensional magnetic field intensity distribution data through a magnetic field sensor to obtain the initial magnetic field distribution map; Process the initial magnetic field distribution map using the gradient calculation method to obtain the three-dimensional distribution map of the gradient change; For the three-dimensional distribution map of the gradient change, apply the defect detection algorithm to identify the magnetic field inhomogeneous region and determine the distribution position of micro-cracks; Extract characteristic parameters from the distribution positions of micro-cracks to obtain the spatial coordinate data of the defect positions; Collect the internal resistance growth data through a resistance measurement device to judge the corresponding relationship between the internal resistance growth and the defect positions; Use a regression analysis algorithm to process the spatial coordinate data of the defect positions and the internal resistance growth data to determine the change trend of the associated distribution; Generate a three-dimensional visualization diagram according to the change trend of the associated distribution to obtain the mapping result of the defect positions and the internal resistance growth rate.

9. The method according to claim 1, wherein The associated distribution and accuracy improvement index of the defect positions and the internal resistance growth rate are obtained. The heterogeneous data is mapped to a unified dimension through parameter standardization processing, and the noise interference is filtered by using the outlier rejection rule to generate a final health status assessment report including micro-crack risks, including: Map the heterogeneous data to a unified dimension through parameter standardization processing to obtain a normalized data set; Use the outlier rejection rule to filter the noise interference in the normalized data set to obtain a clean data set; Extract the characteristic values of the defect positions and the internal resistance growth from the clean data set to determine the characteristic pairs; Calculate the associated distribution between the characteristic pairs through correlation analysis to obtain a distribution matrix; If the associated value in the distribution matrix exceeds the preset threshold, judge the micro-crack risk through a support vector machine algorithm to obtain a risk classification result; Calculate the corresponding relationship between the internal resistance growth and the accuracy improvement for the risk classification result to obtain a health status quantification index; Generate an evaluation result including micro-crack risks according to the health status quantification index.

10. A battery pack health state detection system, characterized in that, The system includes: A voltage data acquisition module for obtaining the open-circuit voltage change data and terminal voltage fluctuation data of the battery under a small current pulse excitation, recording the voltage change curve over time through a high-precision timing acquisition device to obtain an initial data set of the voltage dynamic response; An internal resistance polarization analysis module for extracting the specific values of the internal resistance growth rate and the polarization voltage offset from the initial data set of the voltage dynamic response, calculating the growth trend of the internal resistance with the charge and discharge cycles through an impedance analysis algorithm, and determining the offset of the polarization voltage under high-temperature working conditions; A magnetic field data acquisition module for obtaining the magnetic field intensity distribution and magnetic field gradient change data on the battery surface under a weak alternating magnetic field, collecting the spatial magnetic field information through a magnetic sensor array to obtain a three-dimensional distribution map reflecting the magnetic field inhomogeneity; A data fusion module for integrating the open-circuit voltage change, internal resistance growth rate, polarization voltage offset, and magnetic field intensity distribution data through electromagnetic variable fusion technology, using a data input dimension that includes multi-source data of voltage and magnetic field, determining the fusion weight distribution and dynamically adjusting the ratio according to the data reliability to obtain a comprehensive evaluation parameter after multi-dimensional feature dimensionality reduction; A risk assessment module for, if the evaluation parameter output of the comprehensive evaluation parameter exceeds the preset threshold comparison logic, analyzing the change trend of the voltage dynamic response baseline under high-temperature working conditions through a convolutional neural network, judging the occurrence probability of micro-crack risks inside the battery to obtain a risk warning result; The precision optimization module is used to distinguish the requirements of high-temperature scenarios according to the risk warning results and operating condition classification labels, adjust the smoothing range according to the data fluctuation amplitude by using an adaptive filtering window, optimize the data sampling frequency for the dynamic response baseline, and determine the detection deviation rate reflected by the precision improvement index; The defect correlation analysis module is used to analyze the relationship between magnetic field inhomogeneity and internal micro-crack mapping by using a defect detection algorithm through a three-dimensional distribution map of magnetic field intensity distribution and magnetic field gradient change, and determine the correlation distribution between the defect position and the internal resistance growth rate; The health assessment module is used to obtain the correlation distribution between the defect position and the internal resistance growth rate and the precision improvement index, map heterogeneous data to a unified dimension through parameter standardization processing, filter out noise interference by using outlier rejection rules, and generate a final health status assessment report including the risk of micro-cracks.

Images