A health grading monitoring and fault positioning method for mine emergency lithium battery pack based on single cell differential analysis

Abstract

The application discloses a mine emergency lithium battery pack health grading monitoring and fault positioning method for single body differentiation analysis, relates to the technical field of mine emergency lithium battery pack state monitoring, and collects four types of operating characteristic parameters of the terminal voltage, alternating current internal resistance, surface temperature and charge and discharge cycle number of each battery cell cycle by cycle, constructs a dynamic characteristic baseline matched with the initial characteristics, operation time and environmental cumulative damage of each single body, calculates the deviation value of the real characteristics of the single body and the exclusive baseline, combines the characteristic correlation verification results of adjacent single bodies, and removes common interference caused by the overall change of load fluctuation, electromagnetic interference and environmental temperature of the whole group. The application can greatly reduce the false positive rate and false negative rate of health evaluation, accurately identifies early hidden degradation faults of the battery cell, does not need to rely on the calculation power and communication link on the well, and can still stably operate in the communication interruption scene.

Description

Technical Field

[0001] This invention relates to the field of condition monitoring technology for emergency lithium battery packs used in mining, and in particular to a method for health grading monitoring and fault location of emergency lithium battery packs used in mining based on individual cell differentiation analysis. Background Technology

[0002] Emergency lithium battery packs for mining are core emergency power supply equipment in scenarios such as fully mechanized mining faces, emergency refuge chambers, and emergency lighting in roadways in coal mines. They play a crucial role in providing continuous power for personnel evacuation, rescue command, and safety monitoring systems after a power outage due to a disaster. With the advancement of intelligent construction in coal mines, the industry's requirements for the operational reliability and timely fault warning of emergency power supplies for mining are continuously increasing. How to accurately identify the deterioration of individual battery cells, detect potential faults in advance, and ensure the continuity of power supply has become a core requirement in this field.

[0003] Currently, mainstream monitoring solutions fall into two main categories. The first is the group-level threshold monitoring solution, which works by collecting only three group parameters: total battery voltage, total charging and discharging current, and average temperature and humidity of the battery compartment. Fixed parameter thresholds are preset, and an alarm is triggered when the group parameters exceed the threshold range. This solution has simple logic and low hardware cost, and is currently widely used in old emergency power supply equipment in existing small and medium-sized coal mines, completing basic operation monitoring without complex computing power. However, this solution can only identify extreme faults at the group level and cannot locate individual cell-level degradation and early latent faults. Furthermore, the fixed thresholds do not consider parameter drift caused by strong electromagnetic interference underground, load fluctuations, and changes in ambient temperature, resulting in high rates of missed and false alarms. It is also highly susceptible to causing a complete power outage due to the failure to detect a single degraded cell in time.

[0004] The second type is a general-purpose single-cell health monitoring solution. Its working principle involves collecting the voltage and temperature parameters of each cell and using the average parameter value of all cells in the group as a unified reference baseline. When the deviation of a single cell's parameter from the group's average baseline exceeds a set range, it is considered abnormal. This solution can achieve single-cell-level status identification, and its monitoring accuracy is significantly improved compared to group-level solutions. It has already been deployed in emergency power supplies in some newly built mines. However, this solution does not consider the initial consistency differences in the cell production process, or the differences in individual aging rates under different installation scenarios. Using a unified reference baseline easily leads to misjudging inherent parameter differences and normal aging drift as faults. Furthermore, it cannot distinguish between fluctuations in the overall group's operating conditions, local environmental disturbances, and single-cell faults, resulting in a persistently high false alarm rate. Simultaneously, most solutions rely on cloud computing power above ground for health assessment. When underground communication is interrupted, the monitoring function completely fails. The logic of directly cutting off the power supply to the entire group after a fault is triggered could actually lead to more serious safety risks in disaster scenarios.

[0005] Existing monitoring solutions have not been adapted and optimized for the special constraints of strong electromagnetic interference, unstable communication, and uninterrupted emergency power supply in underground mines. They cannot meet the high reliability requirements of emergency power supply in coal mine safety regulations. The industry urgently needs high-precision health monitoring and fault location technology adapted to the characteristics of mining scenarios. Summary of the Invention

[0006] This invention proposes a method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis, in order to solve the problems mentioned in the prior art, such as high false alarm and false alarm rates, easy failure due to communication interruption, and easy power interruption during fault handling.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis, applied to the status management of lithium iron phosphate battery packs in explosion-proof emergency power supply scenarios in coal mines, comprising the following steps: Relying on the opto-isolated sampling link that meets the intrinsic safety requirements of mining, and according to the differentiated sampling frequency that matches the characteristics of underground electromagnetic disturbance, the real-time terminal voltage, charging and discharging branch current, electrode surface temperature, 1kHz AC internal resistance parameters of each cell in the battery pack are collected synchronously, as well as the ambient temperature, ambient humidity, and vibration amplitude parameters of the mounting bracket inside the explosion-proof battery cavity. Based on the underground IEEE1588 precise clock, all sampling data are timestamped and aligned. By identifying outliers, pulse interference sampling points caused by the start and stop of the frequency converter are eliminated. The data gaps caused by communication interruption are filled in by combining the current change trend, forming a standardized time series dataset. Based on the full life cycle operation data of each cell from its first commissioning, a dedicated dynamic characteristic baseline for each cell is generated. The deviation of real-time operating parameters from the dedicated baseline is calculated, and the characteristic interference introduced by the common load fluctuations of the whole group and the overall environmental changes is eliminated to extract the individual cell's differentiated deviation characteristics. The differential deviation features are input into a pre-trained lightweight evaluation model adapted to embedded computing power, and the health status quantification value of each cell is output. The individual cell health level is divided by combining the minimum redundancy duration threshold of emergency power supply for mining. For abnormal individual units that exceed the safety threshold, the fault type label is matched by the feature deviation dimension. The common disturbance misjudgment is eliminated by the feature association verification of adjacent individual units, and the accurate location of the individual unit-level fault is completed. The classification results and location information are encrypted and uploaded to the mine power supply monitoring platform.

[0008] Furthermore, when removing outliers and filling gaps in the original sampled data, a sliding window with a length of 10 seconds is used to traverse the entire sampled sequence. Outlier sampling points exceeding 1.5 times the interquartile range threshold are identified by the interquartile range method and marked as outliers caused by electromagnetic interference. For data gaps caused by outliers and communication interruptions, weighted interpolation is performed to fill the gaps by combining the rate of change of charging and discharging current within the gap time window. Compared with the linear interpolation method with fixed weights, this method can effectively avoid parameter trend distortion under dynamic conditions of high current, and provide a highly reliable data foundation for subsequent feature calculations.

[0009] Furthermore, when constructing the individual cell-specific dynamic characteristic baseline, the baseline values ​​of each characteristic dimension of each cell are updated successively according to the following formula: ;in For the i-th cell at the current moment The baseline value of the corresponding feature dimension, These are the initial characteristic reference values ​​obtained from actual measurements under three standard operating conditions—full charge and rest, rated load discharge, and full discharge and rest—when the i-th cell is first put into use. For the first The cycle degradation coefficient of the energy-saving cell is iteratively corrected based on the capacity calibration results after each complete charge-discharge cycle. For the first Energy-saving cell deadline The total number of complete charge-discharge cycles experienced, For the first The temperature cumulative damage coefficient of the energy-saving cell is determined by the results of accelerated aging tests on batteries at different temperatures. For the first The duration of equivalent temperature damage exceeding the rated operating temperature range that the energy-saving cell has accumulated by the cutoff time t.

[0010] The dynamic baseline iterates continuously as the individual unit operates, which can completely offset the differences in initial capacity and internal resistance consistency caused by different individual unit production stages, and avoid the problem of insufficient sensitivity in identifying degraded individual units caused by using the average parameters within the group as a unified baseline.

[0011] Furthermore, when extracting the individual cell differential deviation features, for each feature dimension of each cell, the differential deviation value after removing common interference is calculated according to the following formula:

[0012] in For the first Energy-saving cells in the first Differential deviation values ​​for each feature dimension For the first Energy-saving cells in the first Real-time sampled values ​​of each feature dimension. For the first Energy-saving cells at the current moment The first working condition Baseline values ​​for each feature dimension This represents the total number of battery cells connected in series within the battery pack. The first in the same group The first energy-saving cell Real-time sampled values ​​for each feature dimension For the first The first working condition corresponding to the energy-saving cell The baseline value of each feature dimension.

[0013] The calculation covers four core features: terminal voltage deviation, AC internal resistance growth rate, electrode temperature difference deviation, and charge / discharge voltage plateau offset. The second term on the right side of the formula corresponds to the average relative deviation of the remaining cells in the group under the same feature dimension. Subtracting this from the relative deviation of the individual cells completely isolates the feature deviations caused by common operating conditions such as sudden load changes and overall rise and fall of ambient temperature, and only retains the unique feature components generated by the anomalies of the individual cells.

[0014] Furthermore, when calculating the quantitative value of an individual's health status using the lightweight assessment model, the health status result is obtained by fusing multi-dimensional differential deviation features according to the following formula: ;in For the first The health status quantification value of the energy-saving cell ranges from 0 to 1, with a value closer to 1 indicating a better health status. For the first The weight coefficients corresponding to each feature dimension are adaptively generated by ranking the feature contribution of the aging test data of mining batteries during the model training process.

[0015] The evaluation model adopts a lightweight gradient boosting tree architecture after quantization and pruning. The training samples cover the full-cycle aging test data of mining lithium iron phosphate batteries under different ambient temperatures and discharge rates, as well as manually labeled fault data from actual underground operation. The model parameters are compressed to adapt to the low computing power embedded operating environment of the underground monitoring terminal. It can directly complete the health status inference on the terminal without relying on the computing power support of the cloud above ground, effectively reducing the risk of monitoring failure caused by communication interruption.

[0016] Furthermore, when classifying the health status of individual units, the health status is divided into five levels based on the minimum redundancy duration requirements for emergency power supply in mines. Level 1 is a healthy state, capable of meeting full-power emergency discharge requirements; Level 2 is a slightly degraded state, capable of meeting rated power emergency discharge requirements; Level 3 is a moderately degraded state, unable to meet full-power discharge requirements, and needs to be included in the next inspection and maintenance list; Level 4 is a severely degraded state, unable to meet the minimum emergency power supply duration requirements, triggering a graded early warning; Level 5 is a fault state, with risks of thermal runaway and power outage, triggering an emergency alarm. The threshold values ​​for each level are dynamically adjusted based on the emergency power supply redundancy requirements of different mining faces in the mine, and different levels are matched with differentiated response procedures to avoid excessive or delayed early warnings caused by uniform thresholds. This ensures the reliability of emergency power supply while reducing the frequency of unnecessary underground inspections by maintenance personnel.

[0017] Furthermore, when matching fault type labels, five common faults in mining lithium battery packs are covered: loose terminal connection pieces, early signs of internal short circuits, abnormal degradation of internal resistance, rapid capacity reduction of active materials, and single-cell overheating. The matching logic is established based on the differentiated deviation feature dimension triggering rules corresponding to each fault type. Among them, loose terminal connection pieces correspond to positive deviations in both AC internal resistance and terminal temperature dimensions; early signs of internal short circuits correspond to a continuous negative deviation in the terminal voltage dimension; abnormal degradation of internal resistance corresponds to a continuous positive deviation in the AC internal resistance dimension; rapid capacity reduction of active materials corresponds to a sudden change in the charge / discharge voltage plateau offset dimension; and single-cell overheating corresponds to a positive deviation in the terminal temperature dimension. The matching process is verified by combining the feature change trends of five consecutive sampling periods to eliminate mismatch problems caused by instantaneous operating condition disturbances, thereby achieving accurate identification of early latent faults.

[0018] Furthermore, when conducting feature correlation verification between adjacent monomers, the correlation coefficient between the feature sequences of the abnormal monomer and its adjacent monomers is calculated according to the following formula: ;in For the first The Pearson correlation coefficient between the anomalous monomer and the differentially characterized sequences of at least three adjacent monomers, ranging from -1 to 1. The covariance calculation function, This is the differential deviation feature sequence of the i-th abnormal single entity over ten consecutive sampling periods. The correlation coefficient represents the differential deviation feature sequence within the same time window for adjacent units, where σ is the standard deviation of the corresponding feature sequence. When the correlation coefficient is higher than the set threshold, the abnormal deviation is determined to be triggered by a common disturbance in the entire group, and the alarm is canceled. When the correlation coefficient is lower than the set threshold, the abnormality is determined to be caused by a fault in the unit itself, and the fault location result is confirmed. This can effectively solve the problem of false alarms caused by voltage fluctuations in the entire group due to the start-up and shutdown of high-power equipment in the well.

[0019] Furthermore, after completing the fault location at the individual cell level, corresponding local control commands are generated based on the fault level. For cells with a level four severe degradation, the active balancing circuit built into the battery pack is triggered to perform charge and discharge balancing adjustment on the cell, controlling the voltage difference between cells within the allowable operating range. For cells with a level five fault, a non-disruptive redundant circuit switching is triggered to disconnect the faulty cell from the main power supply circuit, and the power supply load is taken over by adjacent redundant cells. The power interruption duration of the entire switching process meets the allowable interruption threshold requirements for mine emergency power supply, and will not cause power interruption for underground emergency lighting and emergency refuge facilities, ensuring that the emergency power supply function does not fail during the fault handling process.

[0020] Furthermore, during the monitoring process, the health level change trajectory of each battery cell, the full sampling data for 30 minutes before and after the fault, the feature calculation results, and the fault type determination results are stored in a chain with encryption to construct a full life cycle health record for each cell. When the monitoring terminal communicates normally with the surface monitoring platform, the record data is automatically resumed from the breakpoint to the platform database, providing data support for subsequent batch quality analysis of battery packs, operation and maintenance cycle optimization, and fault tracing. At the same time, based on the degradation trajectory analysis of the same batch of battery cells, batch quality hazards can be identified in advance, improving the overall operational reliability of the mine emergency power supply system.

[0021] Compared with existing technologies, the beneficial effects of this invention are: This invention constructs a dynamic feature baseline specific to each cell, combines differential deviation feature calculation with feature correlation verification between adjacent cells, effectively offsets the evaluation bias caused by initial consistency differences in cell production and individual aging rate differences under different installation scenarios, completely eliminates common feature interference caused by load fluctuations, electromagnetic interference, and overall changes in ambient temperature, can accurately identify early hidden degradation faults in cells, significantly reduce the false alarm rate and false negative rate of condition assessment, and effectively solve the problems of poor adaptability of existing reference baselines and insufficient accuracy of early fault identification.

[0022] This invention deploys a lightweight health assessment model with quantitative tailoring. All data processing, state reasoning, and fault handling logic can be completed on the local embedded terminal downhole, without relying on cloud computing power and communication links aboveground. Even in scenarios where downhole communication is interrupted, it can still fully realize the whole process monitoring function, effectively solving the problems of existing monitoring solutions relying on cloud computing power and monitoring function failure in scenarios of network outage, and greatly improving the monitoring reliability in extreme disaster scenarios.

[0023] This invention, through multi-level health level classification and power-off fault switching logic, can dynamically adjust the level threshold according to the emergency power supply redundancy requirements of different scenarios and match differentiated response strategies. This not only avoids ineffective operation and maintenance work caused by excessive early warning and reduces the frequency of operation and maintenance personnel going down into the well for inspection, but also allows bypassing the faulty individual through redundant circuits after fault confirmation, ensuring uninterrupted emergency power supply throughout the process. This effectively solves the problem of existing fault protection logic cutting off the entire group output and causing secondary safety risks in disaster scenarios.

[0024] This invention does not require modification to the main circuit structure and explosion-proof performance of existing mine emergency lithium battery packs. Deployment can be completed simply through firmware upgrades. It is compatible with all underground emergency lithium battery application scenarios, such as fully mechanized mining faces, emergency refuge chambers, and roadway emergency lighting. It has extremely high promotional value and industrial adaptability, and can comprehensively improve the operational reliability of mine emergency power supply systems. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the method for health classification monitoring and fault location of emergency lithium battery packs for mining proposed in this invention; Figure 2 This is a flowchart of the single-unit-level time-series data acquisition and timestamp alignment preprocessing of the present invention; Figure 3 This is a flowchart illustrating the construction of the proprietary feature baseline and the extraction of differential deviation features for this invention. Figure 4 This is a flowchart of the individual health status assessment and graded early warning response of the present invention; Figure 5 This is a flowchart of the single-unit level fault location verification and emergency response archiving process of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figures 1 to 5This invention discloses a method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis. It is applied to the full life-cycle status management of lithium iron phosphate battery packs in explosion-proof emergency power supply scenarios in coal mines. Firstly, relying on an intrinsically safe opto-isolated sampling link integrated into the built-in BMS motherboard of the explosion-proof lithium-ion battery power supply, data is collected according to differentiated sampling frequencies matched to the characteristics of underground electromagnetic disturbances and parameter change rates. Specifically, the sampling frequency for cell terminal voltage and terminal surface temperature is set to 1 Hz, the sampling frequency for charge / discharge branch current is set to 10 Hz, the sampling frequency for AC internal resistance and mounting bracket vibration amplitude is set to 0.2 Hz, and the sampling frequency for ambient temperature and humidity inside the explosion-proof battery cavity is set to 0.1 Hz.

[0028] The sampling link employs linear optocouplers to achieve electrical isolation between the high-voltage side and the control side, with a common-mode interference suppression capability of no less than 4 kV, effectively resisting electromagnetic pulse crosstalk generated by the start-up and shutdown of high-power frequency converters in the mine. During the acquisition process, the IEEE 1588 precise clock protocol supported by the underground industrial Ethernet is used to timestamp-align the data acquired by all distributed sampling nodes, with a time synchronization accuracy of no less than 1 millisecond, solving the parameter matching error problem caused by clock deviation of distributed sampling nodes. After the acquisition is completed, pulse interference sampling points caused by the start-up and shutdown of frequency converters and electrostatic discharge are first identified and removed. The data gaps caused by fiber optic jitter and communication interruptions are filled in by combining the charging and discharging current change trend, forming a standardized time-series dataset covering all individual parameter dimensions.

[0029] Subsequently, based on the full life cycle operation data of each cell from its first commissioning, a dedicated dynamic characteristic baseline for each cell is generated. The deviation of real-time operating parameters from the dedicated baseline is calculated point by point. By averaging the characteristic dimensions within the group, the characteristic interference introduced by the common load fluctuations and overall changes in ambient temperature are eliminated, and the differentiated deviation characteristics that only reflect the state of the individual cell are extracted.

[0030] The extracted differential deviation features are then input into a pre-trained lightweight evaluation model adapted to the computing power of the embedded terminal. The health status quantification value of each cell is directly output on the local terminal. Combined with the minimum redundant power supply duration threshold required for emergency power supply in mining, the refined classification of individual cell health levels is completed.

[0031] Finally, for abnormal units whose health status exceeds the safety threshold range, the corresponding fault type label is matched by the dimensional distribution of the differentiated deviation characteristics. By verifying the feature association of adjacent units, misjudgments caused by local environmental disturbances and instantaneous operating condition fluctuations are eliminated, and the fault location at the unit level is accurately located. Finally, the hierarchical monitoring results and fault location information are encrypted using the national cryptographic SM4 algorithm and uploaded to the mine surface power supply monitoring platform.

[0032] This invention also discloses the specific implementation logic for outlier removal and gap filling of the original sampled data. During the data processing, a sliding window with a length of 10 seconds is used to traverse the entire sampled sequence segment by segment. Outlier sampling points exceeding 1.5 times the interquartile range threshold are identified by the interquartile range method and marked as invalid outliers caused by electromagnetic interference.

[0033] For the identified outliers and data gaps caused by communication interruptions, weighted interpolation is performed to complete the data by combining the rate of change of charging and discharging current within the gap time window. The interpolation weights are dynamically adjusted according to the slope of current change. Under high current dynamic conditions, the weights of the current trend terms at adjacent times are increased to avoid the parameter trend distortion problem of the fixed weight linear interpolation method under dynamic conditions.

[0034] This processing logic can effectively filter out pulse interference caused by strong electromagnetic environment downhole while preserving the actual trend of sampling data changes, providing a highly reliable data foundation for subsequent feature calculations and solving the problem of inaccurate state assessment caused by many dirty points and large distortions in the original sampling data.

[0035] This invention also discloses the successive update logic for the individual cell-specific dynamic characteristic baseline. For each characteristic dimension of each cell, the baseline value is updated successively according to a unified formula after each complete charge-discharge cycle. The update formula is as follows: ;in For the first Energy-saving cells at the current moment The baseline value of the corresponding feature dimension, The initial characteristic reference value is obtained by three standard operating condition cycles of actual measurement during the ground commissioning phase before the i-th cell is put into use for the first time. The standard operating conditions cover three stages: full charge and rest for two hours, rated current constant current discharge, and full discharge and rest for two hours. The average value of the parameters from the three tests is taken as the initial reference. For the first The cycle degradation coefficient of the energy-saving cell is iteratively corrected by the ampere-hour integral capacity calibration result after each complete charge-discharge cycle, reflecting the cycle aging rate characteristics of the cell itself. The total number of complete charge-discharge cycles experienced by the i-th cell at time t is denoted by t. For the first The temperature cumulative damage coefficient of the energy-saving cell is calibrated based on the results of accelerated aging tests of lithium iron phosphate batteries. It conforms to the Arrhenius law of battery aging and reflects the degree of influence of temperature on the aging rate of individual cells. For the first Energy-saving cell deadline The cumulative duration of equivalent temperature damage exceeding the rated operating temperature range is calculated by converting the operating time at different temperatures into the equivalent damage time under normal temperature rated operating conditions.

[0036] The dynamic baseline can be continuously iterated and corrected as each cell ages, completely offsetting the differences in initial capacity, internal resistance, and voltage platform consistency between different cells during the production process. This solves the problem of traditional methods using a unified fixed baseline within the group and misjudging inherent differences between individual cells as deterioration faults.

[0037] This invention also discloses the specific extraction logic for the individual cell differential deviation features. For each feature dimension of each cell, the differential deviation value after removing common interference is calculated according to a unified formula, which is as follows: ;in For the first Energy-saving cells in the first Differential deviation values ​​for each feature dimension For the first Energy-saving cells in the first Real-time sampled values ​​of each feature dimension For the first Energy-saving cells at the current moment The first working condition Baseline values ​​for each feature dimension This represents the total number of battery cells connected in series within the battery pack. The first in the same group The first energy-saving cell Real-time sampled values ​​for each feature dimension For the first The first working condition corresponding to the energy-saving cell The baseline value of each feature dimension.

[0038] The calculation covers four core features: terminal voltage deviation, AC internal resistance growth rate, electrode temperature difference deviation, and charge / discharge voltage plateau offset. The second term on the right side of the formula corresponds to the average relative deviation of the remaining cells in the group under the same feature dimension. Subtracting this from the relative deviation of the individual cells completely isolates the feature deviations caused by common operating conditions such as sudden load changes and overall ambient temperature rises and falls. Only the unique feature components generated by the individual cell's own anomalies are retained. This solves the problem that traditional methods cannot distinguish between common operating condition fluctuations and individual cell anomalies, resulting in a high false alarm rate.

[0039] This invention also discloses the reasoning and operational logic of a lightweight health status assessment model. When calculating the quantitative value of an individual's health status, the model integrates multi-dimensional differential deviation features according to a unified formula. The calculation formula is as follows: ;in For the first The health status quantification value of the energy-saving cell ranges from 0 to 1. The closer the value is to 1, the better the health status of the individual cell. For the first The weight coefficients corresponding to each feature dimension are adaptively generated by sorting the feature gains of the aging test data of mining batteries during the model training process.

[0040] The evaluation model adopts a lightweight gradient boosting tree architecture after quantization and pruning. The training samples cover the full-cycle aging test data of mining lithium iron phosphate batteries under different ambient temperatures and discharge rates, as well as manually labeled fault data from actual underground operation. After training, the floating-point parameters of the model are quantized into 8-bit fixed-point numbers, and the model size is compressed to less than 100 kilobytes. It can run on a general-purpose embedded MCU with a computing power of tens of MIPS, and the time for a single health state inference does not exceed 2 milliseconds.

[0041] All inference processes of the model are completed on the local BMS terminal without relying on cloud computing power support, which effectively solves the problem of monitoring function failure when downhole communication is interrupted, while taking into account the accuracy of health status assessment and the computing power constraints of embedded terminals.

[0042] This invention also discloses the classification and response logic for individual unit health levels. The classification process, combined with the minimum redundancy duration requirements for emergency power supply in mines, divides the health status of individual units into five levels. Level 1 is a healthy state, with a health status quantification value higher than 0.9, meeting the emergency impact load discharge requirements of 3 times the rated power; Level 2 is a slightly degraded state, with a health status quantification value between 0.8 and 0.9, meeting the continuous emergency discharge requirements at rated power; Level 3 is a moderately degraded state, with a health status quantification value between 0.6 and 0.8, unable to meet the full-power impact discharge requirements and needing to be included in the maintenance list for the next periodic inspection; Level 4 is a severely deteriorated state, with a health status quantification value between 0.4 and 0.6, unable to meet the minimum emergency power supply duration required by coal mine safety regulations, triggering a graded early warning; Level 5 is a fault state, with a health status quantification value lower than 0.4, posing a risk of thermal runaway and power outage, triggering an emergency alarm.

[0043] The threshold for each level can be dynamically adjusted according to the emergency power supply redundancy requirements of different installation locations. For example, the threshold can be appropriately increased for core locations such as fully mechanized mining faces and emergency refuge chambers, while the threshold can be appropriately decreased for regular emergency lighting points in roadways. Different levels are matched with differentiated response processes, which solves the problem of excessive warnings or delayed warnings caused by the use of a uniform threshold in traditional methods. While ensuring the reliability of emergency power supply, it reduces the frequency of unnecessary underground inspections by operation and maintenance personnel.

[0044] This invention also discloses a matching logic for fault type labels. The matching process covers five common faults in mining lithium battery packs: loose terminal connection, early signs of internal short circuit, abnormal degradation of internal resistance, rapid capacity reduction of active material, and overheating of individual cells. The matching logic is established based on the differentiated deviation feature dimension triggering rules corresponding to each fault type.

[0045] Among them, loose electrode connection points correspond to positive deviations in both AC internal resistance and electrode temperature dimensions; early signs of internal short circuits correspond to continuous negative deviations in terminal voltage dimension; abnormal deterioration of internal resistance corresponds to continuous positive deviations in AC internal resistance dimension; a sharp drop in active material capacity corresponds to a sudden change in the charge / discharge voltage platform shift dimension; and overheating of individual cells corresponds to positive deviations in electrode temperature dimension.

[0046] The matching process is verified by combining the feature change trend of five consecutive sampling periods. The corresponding fault label matching is completed only when the feature deviation persists and the change trend conforms to the evolution law of the corresponding fault. This eliminates the mismatch problem caused by instantaneous operating condition disturbances and enables early identification of early hidden faults such as internal short circuits. It solves the problems of traditional methods that use single-point threshold to trigger alarms, cannot identify early hidden faults, and have inaccurate fault type matching.

[0047] This invention also discloses the specific logic for verifying the correlation between adjacent monomer features. The verification process calculates the Pearson correlation coefficient between the feature sequences of the abnormal monomer and its adjacent monomers according to a unified formula, which is as follows: ;in For the first The Pearson correlation coefficient between the anomalous monomer and the differentially characterized sequences of at least three adjacent monomers, ranging from -1 to 1. The covariance calculation function, For the first The differential deviation feature sequence of the abnormal single entity over ten consecutive sampling periods. These are the differential deviation feature sequences of adjacent monomers within the same time window. This represents the standard deviation of the corresponding feature sequence.

[0048] When the correlation coefficient is higher than the set threshold of 0.8, the abnormal deviation is determined to be triggered by local environmental changes or common disturbances of the whole group, and the alarm is automatically canceled; when the correlation coefficient is lower than the set threshold of 0.3, the abnormality is determined to be caused by the fault of the individual unit, and the fault location result is confirmed.

[0049] This verification logic can effectively solve the problem of false alarms caused by fluctuations in the synchronization parameters of the entire group or local multi-cell batteries due to the start-up and shutdown of high-power equipment downhole and poor local heat dissipation, and further improve the accuracy of fault location.

[0050] This invention also discloses the local power supply protection control logic after fault confirmation. After completing the fault location at the individual cell level, a corresponding local control command is generated according to the fault level. For cells with a level four severe degradation, the inductive active balancing circuit built into the battery pack is triggered to perform charge and discharge balancing adjustment on the cell, with the balancing current not less than 1 ampere, controlling the voltage difference between cells within the allowable operating range of 20 millivolts, thus slowing down the degradation rate of the cell. For cells with a level five fault, a non-disruptive redundant circuit switching is triggered. The faulty cell is bypassed from the main power supply circuit by a magnetic latching relay connected in parallel with the cell, and the redundant cells reserved in the battery pack design fill the power supply load. The power interruption time of the entire switching process does not exceed 10 milliseconds, which meets the allowable power interruption time requirements for mine emergency loads and will not cause power interruption for underground emergency lighting, personnel positioning, and emergency escape facilities.

[0051] This control logic changes the traditional fault protection method of directly cutting off the entire output. It continuously ensures that the emergency power supply function does not fail during the fault handling process, solving the problem that the traditional fault protection causes the emergency power supply to be interrupted, which in turn causes downhole safety risks.

[0052] This invention also discloses the logic for constructing and managing a single cell's full lifecycle health record. During the monitoring process, the health level change trajectory of each cell, the full sampling data for 30 minutes before and after the fault, the feature calculation results, and the fault type determination results are stored in a chain-like encrypted manner. Each data block carries a corresponding timestamp and the hash check value of the previous data block to prevent the locally stored data from being accidentally tampered with. The storage medium is a 128-megabyte industrial-grade Flash chip, which can store at least ten years of single cell operation data.

[0053] When the monitoring terminal communicates normally with the surface monitoring platform, it automatically uploads the stored archive data to the platform database using the MQTT protocol, which supports breakpoint resumption. If communication is interrupted, the data is temporarily stored locally and automatically re-uploaded after communication is restored, ensuring no monitoring data is missed. The health records provide reliable data support for subsequent batch quality analysis of battery packs, optimization of operation and maintenance cycles, and fault tracing. Simultaneously, based on the degradation trajectory analysis of cells within the same batch, it can identify potential batch quality issues in advance, solving the problem that traditional monitoring systems only store real-time data and cannot conduct full lifecycle tracing analysis. This aligns with the requirements of full lifecycle management for mining power supply equipment.

[0054] Reference Figure 1This flowchart illustrates the overall implementation steps of health grading monitoring and fault location for emergency lithium battery packs in mining. The system first establishes a single-cell-level time-series data acquisition link, synchronously acquiring cell operating parameters and battery compartment environmental parameters, and performing outlier removal and timestamp alignment preprocessing to output a standardized time-series dataset. Subsequently, a dedicated feature baseline is constructed using historical operating data throughout the single cell's lifecycle, and the differential deviation feature vector of real-time parameters relative to the baseline is calculated. Next, the deviation feature vector is input into a pre-trained evaluation model, outputting a quantitative value of the single cell's health status, and combining this with the minimum redundant power supply threshold to complete the single cell health level classification. Finally, fault type labels are matched to abnormal cells, and common interference is eliminated through feature correlation verification between adjacent cells to complete fault location. The grading monitoring results and fault location information are then uploaded to the underground monitoring platform.

[0055] Reference Figure 2 This flowchart details the raw time-series data acquisition and standardized preprocessing mechanism. The system synchronously acquires the battery cell's terminal voltage, charging / discharging current, temperature, and AC internal resistance at a set sampling frequency that meets mining electromagnetic compatibility requirements, and records the ambient temperature, humidity, and vibration amplitude of the battery compartment. In the preprocessing stage, the system uses a sliding window quartile method to accurately identify outliers in the sampled data caused by electromagnetic interference, and uses linear interpolation results from adjacent sampling times to numerically complete outliers and missing data segments caused by communication interruptions. Finally, resampling and alignment operations are performed on parameter streams with different sampling frequencies to effectively eliminate data jump errors caused by the strong electromagnetic environment underground, ensuring a high-quality standardized time-series dataset for subsequent evaluation and analysis.

[0056] Reference Figure 3 This flowchart illustrates the construction logic of the dedicated feature baseline and the extraction process of differentiated deviation features. The system uses the measured parameters of a single cell under three standard operating conditions—fully charged, fully discharged, and under rated load at rest—when it is first put into use as the initial benchmark values. Subsequently, the baseline parameters are dynamically updated by combining the capacity decay correction coefficient after each complete charge-discharge cycle, thereby effectively offsetting the impact of initial consistency differences between different cells on the condition assessment during the production process. When extracting features, the system compares the real-time sampled parameters with the parameters of the same operating condition nodes in the dedicated baseline, calculates their relative differences, and generates a differentiated deviation feature vector covering four core dimensions: single cell terminal voltage deviation, AC internal resistance growth rate, temperature rise and temperature difference deviation, and charge-discharge voltage plateau offset.

[0057] Reference Figure 4This flowchart illustrates the process of individual cell health status assessment and classification based on a pre-trained model. The system employs a lightweight gradient boosting tree architecture to establish a health status assessment model. This model is pre-trained using accelerated aging test data from the entire lifecycle of mining lithium iron phosphate batteries and labeled data from actual underground operations, incorporating individual cell-specific feature weighting factors during training. After the differentiated deviation feature vector is input into the model, the system outputs a quantitative value of the health status of each cell. Furthermore, the system combines the minimum continuous power supply duration requirements and minimum redundant power supply thresholds for mining emergency power supply to precisely classify the individual cell health levels into five levels, clearly defining the emergency power supply guarantee capabilities corresponding to each level, and automatically matching the corresponding early warning response strategy to guide on-site operation and maintenance.

[0058] Reference Figure 5 This flowchart illustrates the fault location, correlation verification, and subsequent emergency closed-loop handling logic for abnormal individual cells. For common mining lithium battery pack faults, the system uses threshold triggering logic across different dimensions of differentiated deviation characteristics for initial judgment, and combines this with verification of feature change trends over multiple consecutive sampling periods to eliminate instantaneous false alarms. To further reduce the false alarm rate, the system calculates the correlation coefficient of the abnormal individual's feature deviation values ​​with at least three adjacent individuals within the same group, strictly judging based on thresholds and eliminating interference from common operating conditions. After completing individual cell-level fault location, for critical and higher-level fault cells, the system automatically generates individual cell balancing adjustment or redundant power supply circuit switching commands to transfer the power supply load. Finally, the health change trajectory and time-series characteristics are stored in a chain to construct a full lifecycle health record for each individual cell.

[0059] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis, characterized in that, Includes the following steps: S1. Construct a single-cell-level time-series data acquisition link. According to the set sampling frequency that meets the requirements of electromagnetic compatibility for mining, synchronously collect the terminal voltage, charging and discharging current, surface temperature, AC internal resistance parameters of each cell in the lithium battery pack, as well as the ambient temperature, ambient humidity, and vibration amplitude parameters inside the battery compartment. Perform outlier removal and timestamp alignment preprocessing on the collected raw data to obtain a standardized time-series dataset. S2. Based on the full life cycle historical operation data of each cell, construct the exclusive characteristic baseline of the corresponding cell, calculate the differential deviation feature vector of real-time operation parameters relative to the exclusive baseline, and identify and eliminate the feature interference introduced by the common operating condition fluctuations within the group. S3. Input the differentiated deviation feature vector of each cell into the pre-trained health status assessment model, output the health status quantification value of each cell, and combine it with the minimum redundant power supply threshold for mine emergency power supply to complete the health level classification of each cell. S4. For abnormal units whose health level exceeds the preset safety range, the fault type label is matched by the dimensional distribution of the differentiated deviation characteristics. Common interference is eliminated by verifying the feature association of adjacent units, and the unit-level fault location is completed. Finally, the hierarchical monitoring results and fault location information are uploaded to the underground power supply monitoring platform.

2. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, When performing outlier removal and timestamp alignment preprocessing on the raw data, the sliding window quartile method is used to identify electromagnetic interference outliers in the sampled data. Outliers and missing data segments caused by communication interruptions are filled in by linear interpolation results of adjacent sampling times. Parameters at different sampling frequencies are resampled and aligned to eliminate data jump errors caused by the strong electromagnetic environment downhole.

3. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, When constructing the dedicated characteristic baseline for each individual unit, the measured parameters under three standard operating conditions—fully charged, fully discharged, and under rated load at the time of initial use—are used as the initial benchmark values. The baseline parameters are then dynamically updated in conjunction with the capacity decay correction coefficient after each complete charge-discharge cycle to offset the impact of initial consistency differences in different unit production stages on the status assessment.

4. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, The differentiated deviation feature vector includes feature values ​​in four dimensions: single-unit terminal voltage deviation, AC internal resistance growth rate, temperature rise and temperature difference deviation, and charge / discharge voltage plateau offset. Each feature value is the relative difference between the real-time sampling parameter and the node parameter of the same operating condition in the corresponding single-unit dedicated baseline.

5. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, The health status assessment model adopts a lightweight gradient boosting tree architecture. The training samples cover accelerated aging test data of lithium iron phosphate batteries for mining throughout their entire life cycle and manually labeled data from actual underground operation. Individual differential feature weighting factors are introduced during the training process.

6. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, The individual unit health level is divided into five levels, each level corresponding to a different emergency power supply guarantee capability. The level classification threshold is dynamically adjusted in combination with the minimum continuous power supply duration requirement that the mine emergency power supply must meet. Each level is matched with a corresponding early warning response strategy.

7. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, When matching fault type labels, it covers five common faults of mining lithium battery packs: internal short circuit, abnormal internal resistance degradation, capacity drop, loose connection of connecting piece, and abnormal temperature. The fault type is initially determined by threshold triggering logic of different dimensions of the differentiated deviation features. Then, it is combined with the feature change trend verification of multiple consecutive sampling periods to eliminate misjudgments caused by instantaneous operating condition fluctuations.

8. The method for health grading monitoring and fault location of mining emergency lithium battery packs based on single-cell differential analysis according to claim 7, characterized in that, When eliminating common interference by verifying the feature correlation between adjacent units, the correlation coefficient of the feature deviation value between the abnormal unit and at least three adjacent units in the same group is calculated. If the correlation coefficient is higher than the set threshold, it is determined to be a false alarm triggered by common operating condition interference. If the correlation coefficient is lower than the threshold, it is confirmed to be a fault of the unit itself, thereby reducing the false alarm rate.

9. The method for health grading monitoring and fault location of emergency lithium battery packs for mining based on individual cell differentiation analysis according to claim 1, characterized in that, After completing the fault location at the individual unit level, for abnormal units that are determined to be critical faults or above, corresponding unit balancing adjustment or redundant power supply circuit switching commands are generated to transfer the power supply load borne by the faulty unit to the redundant power supply circuit without interrupting the emergency power supply function.

10. The method for health grading monitoring and fault location of mining emergency lithium battery packs based on single-cell differential analysis according to claim 1, characterized in that, The health level change trajectory of each battery cell and the preset time sequence characteristic data before and after the fault are stored in a chain to construct a full life cycle health record for each cell. This provides data support for subsequent balanced maintenance and batch replacement decisions of the battery pack, and adapts to the management requirements of regular inspection of mining power supply equipment.

Images