A method for measuring abnormality of battery string system housing potential and distinguishing electric leakage

Abstract

The application discloses a battery string system shell potential abnormality measurement and electric leakage discrimination method, relates to the field of battery safety detection, and comprises the following steps: collecting shell potentials of each shell in a battery string system, determining a target shell according to the shell potentials, applying local thermal disturbance to a region corresponding to the target shell, collecting the shell potentials of the target shell in a sampling period before disturbance, a sampling period during disturbance and a sampling period after disturbance, respectively, constructing an abnormality category feature vector to input an abnormality category discrimination model, and obtaining a discrimination result of a surface conductive medium pseudo electric leakage abnormality or a real electric leakage abnormality. The application solves the problem that real electric leakage and pseudo electric leakage are difficult to distinguish in shell potential abnormality.

Description

Technical Field

[0001] This invention relates to the field of battery safety testing, and more specifically, to a method for measuring abnormal casing potential and identifying leakage current in a battery series system. Background Technology

[0002] Battery series systems are widely used in new energy vehicles, electrochemical energy storage, and other scenarios. These systems typically consist of multiple cells connected in series. During operation, it is crucial to monitor not only the voltage, current, and temperature of individual cells, but also the potential relationship between the cell casing, electrode nodes, and the system reference point. This is because abnormal casing potential is often related to changes in insulation state, leakage current, structural coupling, and the conductivity of the environmental medium. Inaccurate assessment can easily affect fault warnings and subsequent maintenance decisions. Existing publications include CN110687467A, "A Method for Detecting Leakage Current in Lithium Batteries," which proposes charging the battery under test and collecting time-voltage curves, then iteratively adjusting the charging current based on the curve slope to obtain the battery leakage current detection value; and CN117491813A, "A Method for Detecting Insulation Anomalies in Power Battery Systems for New Energy Vehicles," which proposes extracting parameters such as insulation resistance, voltage, current, and temperature from the original data and using cluster analysis to determine the insulation anomalies of the power battery system. The aforementioned disclosure indicates that existing technologies have already focused on battery leakage current detection and power battery system insulation anomaly detection, and mainly revolve around electrical parameters such as voltage, current, and insulation resistance for judgment.

[0003] However, in a battery series system, an abnormal casing potential is not the same as a true leakage current anomaly. For the same abnormal casing potential, it could be caused by internal insulation degradation or the formation of a true leakage path within the cell, or by surface conductive media such as condensation, coolant residue, dirt, or dust absorbing moisture forming a temporary conductive path on the outer surface of the casing. These two types of situations may appear similar in static measurement results. If judgment is based solely on the casing potential, node potential, or system insulation parameters at a single moment, it is easy to misjudge a surface pseudo-leakage as a true leakage current. At the same time, the casing sampling channel itself may also have poor contact, oxidation, partial breaks, or moisture leakage, distorting the measured casing potential and further increasing the difficulty of judgment.

[0004] Furthermore, even when a genuine leakage anomaly already exists, it is difficult to accurately identify the cell most relevant to the anomaly based solely on the static casing potential distribution. This is because there may be electrical or conductive coupling between adjacent cells, connectors, and support structures in a series structure. A genuine anomaly source may cause multiple adjacent casings to exhibit anomalies simultaneously, leading to confusion between the anomaly source and the affected casings. Therefore, how to further distinguish between genuine leakage anomalies and pseudo-leakage anomalies caused by surface conductive media after a casing potential anomaly occurs, and how to improve the accuracy of identifying associated abnormal cells while ensuring the reliability of the measurement link, has become a technical problem that needs to be solved in the field of casing potential detection for battery series systems. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for measuring abnormal potential and identifying leakage current in the casing of a battery series system, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for measuring abnormal casing potential and identifying leakage current in a battery series system includes the following steps:

[0008] Collect the casing potential of each casing in the battery series system; determine the target casing based on the casing potential, wherein the target casing is the casing whose casing potential exceeds the abnormal threshold;

[0009] A local thermal disturbance is applied to the corresponding region of the target shell; the shell potential of the target shell is collected during the sampling periods before the disturbance, during the disturbance, and after the disturbance.

[0010] An anomaly category feature vector is constructed based on the shell potential of the target shell during the sampling periods before the disturbance, during the disturbance, and after the disturbance. The anomaly category feature vector is then input into the anomaly category discrimination model to obtain the discrimination result that the anomaly corresponding to the target shell is either a pseudo leakage anomaly of the surface conductive medium or a real leakage anomaly.

[0011] Preferably, the casing potential of each casing in the battery series system is collected, and the node potential of the corresponding cell in each casing is collected;

[0012] Determining the target shell based on the shell potential includes calculating the potential difference between the shell potential of the candidate shell and the corresponding node potential of the candidate shell, and calculating the shell potential difference between the candidate shell and the preceding shell adjacent in the series direction, and the shell potential difference between the candidate shell and the following shell adjacent in the series direction.

[0013] When the potential difference value is greater than the first preset abnormal threshold, and at least one of the shell potential difference value between the candidate shell and the previous shell and the shell potential difference value between the candidate shell and the next shell is greater than the second preset abnormal threshold, the candidate shell is determined as the target shell.

[0014] Preferably, applying local thermal disturbance to the area corresponding to the target housing includes controlling the operation of a thermal disturbance unit corresponding to the target housing to raise the temperature of the area corresponding to the target housing from the initial temperature to the target temperature and maintain it for a first preset duration.

[0015] The shell potential is collected during the sampling period before the start of the local thermal disturbance within a second preset duration before the start of the disturbance; the shell potential is collected during the sampling period during the disturbance within a first preset duration; and the shell potential is collected during the sampling period after the end of the local thermal disturbance within a third preset duration.

[0016] Preferably, the anomaly category feature vector includes the average shell potential before the disturbance, the average shell potential during the disturbance, the average shell potential after the disturbance, a first potential difference between the average shell potential before the disturbance and the average shell potential during the disturbance, a second potential difference between the average shell potential during the disturbance and the average shell potential after the disturbance, the temperature change of the region corresponding to the target shell, and the recovery time required for the shell potential of the target shell to recover to no greater than the anomaly threshold within the sampling period after the disturbance.

[0017] The anomaly classification model is a machine learning model trained using labeled pseudo leakage anomaly samples and real leakage anomaly samples of surface conductive medium; the classification result includes pseudo leakage anomaly labels and real leakage anomaly labels of surface conductive medium.

[0018] Preferably, before acquiring the casing potential of each casing in the battery series system, the casing sampling channel corresponding to each casing is self-verified. The self-verification includes, under the first reference bias state, controlling the switching unit to connect the casing sampling channel to the first reference bias node through the first preset resistor, and acquiring the first channel response value of the casing sampling channel under the first reference bias state.

[0019] In the second reference bias state, the switching unit is controlled to connect the housing sampling channel to the second reference bias node through the second preset resistor, and the second channel response value of the housing sampling channel in the second reference bias state is collected;

[0020] A channel health feature vector is constructed based on the first channel response value and the second channel response value, and the channel health feature vector is input into the channel health discrimination model to obtain the discrimination result of whether the shell sampling channel is a valid shell sampling channel.

[0021] Preferably, the channel health feature vector includes a first deviation value between the first channel response value and the reference potential corresponding to the first reference bias node, a second deviation value between the second channel response value and the reference potential corresponding to the second reference bias node, a response difference between the first channel response value and the second channel response value, a first stabilization time required for the first channel response value to reach a stable state, and a second stabilization time required for the second channel response value to reach a stable state.

[0022] The channel health discrimination model is a machine learning model trained using labeled valid shell sampling channel samples and failed shell sampling channel samples; the discrimination result includes valid shell sampling channel labels and failed shell sampling channel labels.

[0023] Preferably, the shell potential is collected only from the shell corresponding to the shell sampling channel that is determined to be a valid shell sampling channel by the channel health discrimination model;

[0024] When the channel health discrimination model determines that the housing sampling channel is a failed housing sampling channel, it records the abnormal channel state of the housing corresponding to the housing sampling channel and prohibits the execution of the target housing determination step based on the housing potential of the housing corresponding to the housing sampling channel.

[0025] Preferably, after the anomaly classification model determines the anomaly corresponding to the target housing as a real leakage anomaly, potential perturbations are sequentially applied to the candidate associated cells corresponding to the target housing. During each potential perturbation application, the perturbation response values ​​of the target housing, the preceding housing adjacent to the target housing in the series direction, and the following housing adjacent to the target housing in the series direction are collected. A location feature vector is constructed based on the perturbation response values, and the location feature vector is input into the associated abnormal cell classification model to obtain the associated abnormal cell classification result.

[0026] Preferably, the candidate associated cells include the cell corresponding to the target housing, the preceding cell adjacent to the target housing cell in the series direction, and the following cell adjacent to the target housing cell in the series direction; applying potential perturbations to the candidate associated cells sequentially includes controlling the conduction of the equalization branch or perturbation branch corresponding to the candidate associated cells, so that the node potential of the candidate associated cells to which potential perturbations are applied changes by a fourth preset potential amplitude relative to before the potential perturbation is applied, and maintains a fourth preset duration.

[0027] Preferably, the positioning feature vector includes the target response amplitude of the target housing under each potential disturbance, the response amplitude of the first adjacent housing in the series direction of the target housing under each potential disturbance, the response amplitude of the second adjacent housing in the series direction of the target housing under each potential disturbance, the target response duration required for the target response amplitude to reach its peak value, the response duration of the first adjacent housing required for the first adjacent housing response amplitude to reach its peak value, and the response duration of the second adjacent housing required for the second adjacent housing response amplitude to reach its peak value; the associated abnormal cell discrimination model is a machine learning model trained using labeled associated abnormal cell samples; the associated abnormal cell discrimination result includes the cell label corresponding to the target housing, the label of the previous cell, the label of the next cell, and an undetermined label; when the associated abnormal cell discrimination model outputs an undetermined label, only the discrimination result of the actual leakage abnormality is maintained and the associated abnormal cell is not output.

[0028] The advantage of this invention over existing technologies lies in its focus on the difficulty of distinguishing between true and false anomalies in casing potential anomalies. By applying local thermal disturbance to the corresponding area of ​​the target casing and collecting casing potential data during the sampling periods before, during, and after the disturbance, an anomaly category feature vector is constructed and input into the anomaly category discrimination model. This allows for the differentiation between pseudo-leakage anomalies in the surface conductive medium and true leakage anomalies. The rationale is that the conductivity state formed by the surface conductive medium is more sensitive to local temperature changes; short-term thermal disturbances alter its surface conductivity conditions. In contrast, true leakage paths are located inside the battery cell or at insulation failure sites, and their responses to short-term surface thermal disturbances do not exhibit the same characteristics. Based on this response difference, this invention is no longer limited to a single static potential value but utilizes the dynamic response process to complete anomaly discrimination, thereby solving the problem of easy misjudgment of casing potential anomalies and improving the specificity of leakage current discrimination.

[0029] Building upon the above, this invention further introduces a self-verification mechanism for the housing sampling channel. Before formally acquiring the housing potential, the response values ​​of the housing sampling channel are acquired under both the first and second reference bias states, and a channel health feature vector is constructed and input into the channel health discrimination model. This allows for the determination of the sampling channel's validity before deciding whether to use the corresponding housing potential for subsequent discrimination. The underlying principle is that a healthy sampling channel exhibits predictable responses to a known reference bias, while sampling channels with poor contact, oxidation, partial breaks, or moisture leaks will show abnormal characteristics in terms of deviation values, response differences, and stabilization times. Based on this, this invention solves the problem of misjudgment introduced by the measurement link itself, enabling subsequent housing potential anomaly discrimination to be based on more reliable measurements.

[0030] After determining that a leakage current anomaly is genuine, this invention further applies potential perturbations to candidate associated cells sequentially and collects the perturbation response values ​​of the target casing and its adjacent casings. This constructs a location feature vector input to the associated abnormal cell discrimination model, yielding the associated abnormal cell discrimination result. This part utilizes another electrical response law: cells or conduction paths closer to the genuine anomaly source typically exhibit more direct and significant characteristics in response amplitude and timing under controlled perturbation, while adjacent casings only affected by coupling show weaker or more delayed responses. Based on this difference, this invention further solves the problem of accurately locating associated abnormal cells in static anomaly distributions, making the entire scheme a continuous discrimination process consisting of measurement reliability judgment, genuine / false anomaly discrimination, and associated anomaly identification. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the overall steps of the battery series system casing potential anomaly measurement and leakage current detection method of the present invention;

[0032] Figure 2 This is a schematic diagram of the local physical structure and thermal disturbance application unit of the battery series system of the present invention;

[0033] Figure 3 This is a time-series waveform diagram of temperature and time when a local thermal disturbance is applied according to the present invention;

[0034] Figure 4 This is the input-output architecture diagram of the anomaly category discrimination model of the present invention;

[0035] Figure 5 This is the input / output architecture diagram of the channel health discrimination model of the present invention;

[0036] Figure 6 This is the input / output architecture diagram of the abnormal battery cell discrimination model of the present invention;

[0037] Figure 7 This is a schematic diagram of the sampling channel self-certification circuit principle of the present invention;

[0038] Figure 8 This is a diagram showing the arrangement of applying the cell potential disturbance according to the present invention. Detailed Implementation

[0039] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0040] This invention provides a method for measuring abnormal casing potential and identifying leakage current in a battery series system, applicable to battery series systems composed of multiple cells connected in series. Each cell has a corresponding casing, and the system includes a casing potential sampling circuit, a node potential sampling circuit, a thermal disturbance unit, a control unit, and a model calculation unit. The casing potential sampling circuit is used to collect the casing potential of each casing relative to the system reference potential, and the node potential sampling circuit is used to collect the node potential of the corresponding cell for each casing. The system reference potential can be selected as the negative terminal reference point of the battery series system or as a common reference node in the isolated sampling domain. The model calculation unit can be integrated into the battery management controller or configured in a host controller that is communicatively connected to it.

[0041] The core idea of ​​this invention is to first identify the target shell with an abnormal shell potential from among the shells, then apply a local thermal disturbance only to the corresponding area of ​​the target shell, and collect the shell potential during the sampling periods before, during, and after the disturbance. Based on these three data segments, an anomaly category feature vector is constructed and input into an anomaly category discrimination model. Finally, it is determined whether the anomaly corresponding to the target shell is a pseudo-leakage anomaly of the surface conductive medium or a true leakage anomaly. Figure 1 As shown, the main process of the entire method can be summarized as follows: casing potential acquisition, target casing determination, application of local thermal disturbance, time-series potential sampling, and anomaly classification. This design is based on the fact that an abnormal battery casing potential does not necessarily mean that a real leakage path has formed inside the cell. Surface conductive media such as condensate film, coolant residue, dirt, and damp dust can also cause casing potential anomalies. Judging solely based on the static casing potential at a given moment can easily confuse two different types of anomalies. This invention classifies anomalies by applying localized thermal stimulation to the target casing, utilizing the difference that surface conductive media are more sensitive to thermal changes while real leakage paths do not have the same thermal response characteristics.

[0042] In one embodiment, the control unit first controls the casing potential sampling circuit to collect the casing potential of each casing in the battery series system. The casing potential sampling circuit is preferably a high input impedance sampling circuit to reduce the impact of the measurement process on the casing potential itself. The input impedance can be set to 10MΩ to 100MΩ, the analog-to-digital converter accuracy can be set to 12-bit to 16-bit, and the sampling period can be set to 5ms to 20ms. A sampling period that is too long will average out the detailed changes during thermal disturbance, while a sampling period that is too short will increase the computational burden on the controller; therefore, it is preferable to select a period within the above range. In one embodiment, the node potential of the corresponding cell for each casing can also be collected synchronously. The node potential here can be one of the positive node potential of a single cell, the negative node potential of a single cell, or the node potential of a connecting piece; preferably, it is the node potential of the single cell closest to the casing location and with stable sampling conditions. The purpose of introducing the node potential is to provide a reference for subsequent judgment on whether a casing deviates from the potential relationship that its corresponding cell should have.

[0043] In one embodiment, after acquiring the shell potential and corresponding node potential of each shell, the control unit executes a target shell determination process. The target shell is not simply defined as the shell with the largest absolute shell potential value, but rather as the shell whose shell potential exceeds an abnormal threshold and exhibits local abrupt change characteristics. This is because cells at different locations in a series system are inherently at different potential levels; if only absolute values ​​are used for judgment, differences in series positions can easily be misjudged as abnormal. Therefore, this invention employs a hierarchical screening method. First, candidate shells whose shell potential exceeds the abnormal threshold are selected from all shells; then, a relative difference judgment is performed on the candidate shells to determine the target shell.

[0044] In one embodiment, the potential difference between the candidate housing's housing potential and the node potential of the corresponding cell is first calculated. Then, the housing potential difference between the candidate housing and the preceding housing in the series direction, and the housing potential difference between the candidate housing and the following housing in the series direction are calculated respectively. When the potential difference between the candidate housing and its corresponding node is greater than a first preset anomaly threshold, and at least one of the housing potential differences between the candidate housing and the preceding housing and the following housing is greater than a second preset anomaly threshold, the candidate housing is determined as the target housing. The former is used to determine whether the housing deviates from the normal potential relationship of its own single cell, and the latter is used to determine whether the housing has abnormally prominent features in a local area. When both conditions are met, it indicates that the housing is more worthy of being identified as a key target.

[0045] In one embodiment, the anomaly threshold, the first preset anomaly threshold, and the second preset anomaly threshold can be obtained through calibration. During calibration, data on the casing potential and node potential under different temperatures, states of charge, and operating currents can be collected on a normal battery pack, and the normal fluctuation range can be statistically analyzed. If a fixed threshold method is used, the anomaly threshold can be set to 50mV to 300mV, the first preset anomaly threshold can be set to 30mV to 200mV, and the second preset anomaly threshold can be set to 20mV to 150mV. If the system is sensitive to different operating conditions, a threshold table for different operating conditions can also be used. The purpose of setting these parameters in this way is to ensure that the selection of the target casing can cover real anomalies while avoiding false triggering caused by fluctuations in normal operating conditions.

[0046] After identifying the target housing, the control unit applies a localized thermal disturbance to the corresponding area of ​​the target housing. For example... Figure 2 As shown, the thermal disturbance unit can be arranged in the vicinity of the target casing. The thermal disturbance unit can be a thin-film resistance heating element, a PTC micro-heater, an integrated micro-heating element, or a temperature sampling device with auxiliary heating function. The emphasis here on localized thermal disturbance is to ensure that the thermal stimulus primarily acts on the surface area of ​​the target casing, rather than altering the thermal field of the entire battery pack. This is because the present invention aims to utilize the principle that the conductivity state of the surface conductive medium easily changes after localized heating, rather than observing changes in the internal insulation of the cell through overall heating. Heating the entire system over a large area would not only cause significant disturbance but also introduce additional variables, reducing the specificity of the judgment.

[0047] In one embodiment, a thermal disturbance unit is controlled to operate, causing the temperature of the corresponding area of ​​the target housing to rise from an initial temperature to a target temperature, and maintaining this temperature for a first preset duration. Figure 3 As shown, local thermal disturbance may include a heating phase and a holding phase. The target temperature should not be set too high, otherwise it may easily cause changes in the thermal state of the battery cell itself; nor should it be set too low, otherwise the state change of the surface conductive medium will not be obvious. In one embodiment, the increase in target temperature relative to the initial temperature can be set to 3°C to 15°C, preferably 5°C to 10°C. The first preset duration can be set to 3s to 30s, preferably 5s to 15s. The heating rate can be set to 0.5°C / s to 3°C / s. This parameter range is adopted to ensure that observable changes in the conductive state of the surface condensate film, damp liquid layer, etc., occur without causing strong thermal disturbance to the entire package.

[0048] During the thermal disturbance, the control unit collects the shell potential of the target shell during three sampling periods: before the disturbance, during the disturbance, and after the disturbance. The pre-disturbance sampling period acquires the baseline shell potential before the thermal disturbance is applied; the during-disturbance sampling period acquires the change in shell potential under thermal stimulation; and the post-disturbance sampling period acquires the recovery characteristics after the thermal stimulation ends. All three sampling periods must be retained because the discrimination in this invention is not based on the instantaneous value at a single moment, but on the dynamic response trajectory of the shell potential before and after the thermal disturbance. In one embodiment, the second preset duration can be set to 1 to 20 seconds, preferably 2 to 10 seconds, and the third preset duration can be set to 1 to 30 seconds, preferably 3 to 15 seconds. To ensure the comparability of the three data segments, the sampling periods for the three sampling periods are preferably kept consistent.

[0049] In one embodiment, to reduce the impact of noise, the average shell potential before the disturbance, the average shell potential during the disturbance, and the average shell potential after the disturbance can be calculated separately. The average value is obtained by averaging multiple samples. The average value is used instead of a single point value because the shell potential fluctuates greatly at a single point when affected by factors such as sampling noise and switching interference, while the average value can more stably reflect the overall characteristics of the stage.

[0050] In one embodiment, an anomaly category feature vector is constructed based on the above sampling results. The anomaly category feature vector includes the average shell potential before the disturbance, the average shell potential during the disturbance, the average shell potential after the disturbance, a first potential difference between the average shell potential before and during the disturbance, a second potential difference between the average shell potential during and after the disturbance, the temperature change of the corresponding region of the target shell, and the recovery time required for the shell potential of the target shell to recover to no greater than the anomaly threshold within the sampling period after the disturbance. Here, the recovery time refers to the time from the end of the local thermal disturbance to the first time the target shell potential falls back to no greater than the anomaly threshold and remains continuously for at least three sampling cycles. If it does not recover to no greater than the anomaly threshold within the entire sampling period after the disturbance, the recovery time is recorded as the third preset duration. The reason for introducing the recovery time is that pseudo-leakage anomalies of the surface conductive medium not only exhibit potential changes during thermal disturbance but also often show a relatively clear decline or recovery process after the thermal disturbance ends, while real leakage anomalies usually do not have the same recovery pattern.

[0051] like Figure 4As shown, the anomaly classification model receives anomaly category feature vectors and outputs either a pseudo-leakage anomaly label or a true leakage anomaly label for the surface conductive medium. In one embodiment, the anomaly classification model can employ a multilayer perceptron. The input layer has a dimension of 7, corresponding to the aforementioned 7 features. The first hidden layer can have 16 neurons, the second hidden layer can have 8 neurons, and the output layer has 2 neurons, corresponding to the two types of labels respectively. The hidden layer activation function can be ReLU, and the output layer can use Softmax. During model inference, the label with the higher output probability is taken as the final classification result. A machine learning model is used instead of a single fixed rule because the difference between pseudo-leakage anomalies and true leakage anomalies in the surface conductive medium is not reflected in a single indicator, but rather in the combination of multiple temporal features and temperature features. The model can learn this combination pattern, thereby improving the classification accuracy.

[0052] In one embodiment, the training samples for the anomaly classification model can be constructed experimentally. Pseudo-leakage anomaly samples of the surface conductive medium can be collected after conditions such as the formation of a condensation film, low-concentration conductive liquid residue, or a thin layer of damp dust on the shell surface. Real leakage anomaly samples can be collected after setting a known leakage path between the cell polarity node and the shell under safe and controlled conditions. During training, the sampled three segments of shell potential and corresponding temperature data are extracted into the aforementioned seven features to constitute the training samples. The training set, validation set, and test set can be divided in a ratio of 70%, 15%, and 15%, respectively. The loss function is the cross-entropy loss function, the optimizer is Adam, the learning rate is set to 0.001, the batch size is set to 16 to 128, and the number of training rounds is set to 100 to 500. For scenarios with a small sample size, support vector machines or gradient boosting trees can also be used to implement anomaly classification; this invention does not limit this.

[0053] When the anomaly classification model outputs a pseudo-leakage anomaly label for the surface conductive medium, it indicates that the anomaly of the target casing is more likely caused by a temporary conductive path on the casing surface. In this case, a pseudo-leakage anomaly status can be output for reference in subsequent maintenance procedures. When the anomaly classification model outputs a true leakage anomaly label, it indicates that the target casing anomaly did not exhibit typical surface conductive medium response characteristics before and after the local thermal disturbance, and the process of identifying associated abnormal cells should continue. At this point, the main process of implementing local thermal disturbance around the target casing and classifying true and false leakage based on the dynamic response of the casing potential during three sampling periods has been completed.

[0054] In a further embodiment, to improve the reliability of the casing potential sampling in the main process described above, the casing sampling channel corresponding to each casing can be self-verified before collecting the casing potential of each casing in the battery series system. Here, the casing sampling channel refers to the complete measurement link from the casing sampling point to the analog-to-digital converter input terminal, including sampling contacts, wires, connectors, conditioning circuits, and input ports. The purpose of setting the channel self-verification before formal sampling is to eliminate abnormal interference from the sampling link itself. Because if the sampling channel has problems such as poor contact, oxidation, partial breakage, or moisture leakage in the wiring harness, even if the subsequent thermal disturbance and model discrimination logic are correct, incorrect conclusions may still be drawn due to distorted input data.

[0055] like Figure 7 As shown, in one embodiment, the channel self-verification includes two test states: a first reference bias state and a second reference bias state. In the first reference bias state, the control switching unit connects the housing sampling channel to the first reference bias node via a first preset resistor and acquires the first channel response value of the housing sampling channel in the first reference bias state. In the second reference bias state, the control switching unit connects the housing sampling channel to the second reference bias node via a second preset resistor and acquires the second channel response value of the housing sampling channel in the second reference bias state. The first and second reference bias nodes can be provided by a precision reference source, a resistor divider network, or an isolated reference source. The first and second preset resistors are used to inject known reference conditions into the sampling channel with minimal disturbance, making the channel health status measurable. If the resistance value is too low, it may excessively affect the original channel state; if the resistance value is too high, it will result in an insufficient response. In one embodiment, the first and second preset resistors can be set to 100kΩ to 5MΩ, preferably 470kΩ to 2MΩ.

[0056] In one embodiment, a channel health feature vector is constructed based on the first channel response value and the second channel response value. The channel health feature vector includes a first deviation value between the first channel response value and the reference potential corresponding to the first reference bias node, a second deviation value between the second channel response value and the reference potential corresponding to the second reference bias node, a response difference between the first and second channel response values, a first stabilization time required for the first channel response value to reach a stable state, and a second stabilization time required for the second channel response value to reach a stable state. Here, the stabilization time refers to the time elapsed from the application of the reference bias until the response enters a stable range. In one embodiment, the stable range can be defined as the response fluctuation not exceeding 2% of the final stable value within five consecutive sampling periods. If the stable range is not entered within a predetermined test duration, the corresponding stabilization time is recorded as the predetermined test duration. The stabilization time feature is introduced because healthy channels typically respond quickly and stably, while channels with moisture leakage or abnormal contact resistance will exhibit significant tailing.

[0057] like Figure 5 As shown, the channel health discrimination model is used to classify channel health feature vectors. In one embodiment, the model can also adopt a multilayer perceptron structure. The input layer has a dimension of 5, the first hidden layer has 8 neurons, the second hidden layer has 4 neurons, and the output layer has 2 neurons, corresponding to the labels of valid and failed housing sampling channels, respectively. In the training samples, the valid housing sampling channel samples come from measured channel data under normal assembly, normal insulation, and normal environmental conditions; the failed housing sampling channel samples can be constructed by artificially introducing fault states such as increased contact resistance, partial breakage of sampling lines, oxidation of connection points, and moisture leakage on the surface of the wiring harness. The training method can be similar to that of the anomaly category discrimination model, using the cross-entropy loss function, the Adam optimizer, and normalization preprocessing. After the model is trained, inference is performed on each housing sampling channel one by one before formal measurement. If the channel health discrimination model determines that a housing sampling channel is a valid housing sampling channel, it is allowed to participate in housing potential acquisition; if it is determined to be a failed housing sampling channel, the channel abnormality state of the corresponding housing is recorded, and the target housing determination step is prohibited based on the housing potential of the corresponding housing of the housing sampling channel. This allows us to distinguish between channel malfunctions and casing leakage malfunctions, preventing channel failures from interfering with the main process judgment.

[0058] In a further embodiment, after the anomaly classification model determines that the anomaly corresponding to the target casing is a real leakage anomaly, the associated abnormal cell classification can continue to be performed to further determine the location of the cell most relevant to the anomaly. The purpose of this arrangement is to first determine the nature of the anomaly, and then perform fine-tuning under the premise of a real anomaly, avoiding meaningless perturbation tests on pseudo-leakage anomalies.

[0059] like Figure 8 As shown, in one embodiment, candidate associated cells are first determined. These candidate associated cells include the cell corresponding to the target housing, the preceding cell adjacent to the target housing cell in the series direction, and the following cell adjacent to the target housing cell in the series direction. These three candidates are chosen based on the fact that the impact of real anomalies in a series system on the target housing and its adjacent housings typically concentrates first on the few cells with the closest electrical connections. Perturbing all cells individually from the outset would not only be inefficient but would also introduce too many irrelevant responses.

[0060] In one embodiment, potential perturbations are sequentially applied to candidate associated cells. These perturbations can be implemented through a balancing branch or an independent perturbation branch. If the system already has a single-cell balancing branch, the balancing branch of the corresponding candidate associated cell can be directly controlled to conduct, causing a controlled change in the cell node potential relative to before the perturbation was applied. If the system lacks a reusable balancing branch, a high-impedance perturbation branch can be set, applying a small potential perturbation to the corresponding node through a switching unit. The fourth preset potential amplitude can be set to 10mV to 150mV, preferably 20mV to 80mV, and the fourth preset duration can be set to 0.2s to 5s, preferably 0.5s to 2s. If the potential perturbation amplitude is set too small, the casing response will be insignificant; if set too large, it may cause unnecessary impact on the original system state. A recovery interval is preferably allowed between each two candidate associated cells, which can be set to 0.5s to 10s, to avoid residual effects of the previous perturbation on subsequent tests.

[0061] During each potential disturbance application, the control unit synchronously acquires the disturbance response values ​​of the target shell, the shell preceding it in the series direction, and the shell following it in the series direction. These disturbance response values ​​can be understood as the change in shell potential relative to the baseline before the disturbance during the application of the potential disturbance. The reason for acquiring not only the target shell but also its preceding and following shells is that the actual location of the anomaly source differs, and the response amplitude and timing relationships among the three will also differ; comprehensive comparison is more beneficial for localization.

[0062] In one embodiment, a positioning feature vector is constructed based on the above response results. The positioning feature vector includes the target response amplitude of the target shell under each potential perturbation, the response amplitude of the first adjacent shell under each potential perturbation of the preceding shell, the response amplitude of the second adjacent shell under each potential perturbation of the subsequent shell, the target response duration required for the target response amplitude to reach its peak value, the response duration of the first adjacent shell required for the first adjacent shell response amplitude to reach its peak value, and the response duration of the second adjacent shell required for the second adjacent shell response amplitude to reach its peak value. The response amplitude reflects the strength of the response, and the response duration reflects the speed of the response. Combining these two types of features allows for a better distinction between directly related anomalies and indirect effects caused by coupling.

[0063] like Figure 6As shown, the associated abnormal battery cell discrimination model is used to output the battery cell label corresponding to the target casing, the label of the previous battery cell, the label of the next battery cell, or an undetermined label based on the location feature vector. In one embodiment, the associated abnormal battery cell discrimination model can use a classification network with an input layer dimension of 6, two hidden layers, and an output layer dimension of 4. The first hidden layer can have 12 neurons, the second hidden layer can have 8 neurons, and the output layer uses Softmax. Training samples can be obtained through controlled experiments, that is, constructing real leakage states at the battery cell corresponding to the target casing, the previous battery cell, and the next battery cell, respectively, and then collecting the response data of the target casing, the previous casing, and the next casing according to the same potential perturbation process, extracting the aforementioned 6 location features, and labeling the corresponding abnormal battery cell. For complex samples that do not meet the criteria of a clear single location, they are labeled as undetermined labels. The cross-entropy loss function and the Adam optimizer can also be used during training. When the model outputs, if the label with the highest probability is an undetermined label, only the judgment result of the actual leakage current anomaly is retained and the associated abnormal cell is not output; if the output is any of the other three labels, the corresponding cell is output as the associated abnormal cell.

[0064] In summary, the specific execution sequence of this invention can be summarized as follows: In the preferred embodiment, the shell sampling channel is first self-verified, then the shell potential and corresponding node potential of each shell are collected, and the target shell is determined; subsequently, local thermal perturbation is implemented around the target shell, and the shell potentials during the sampling periods before, during, and after the perturbation are used to construct an anomaly category feature vector to classify pseudo-leakage anomalies and real leakage anomalies of the surface conductive medium; when the classification result is a real leakage anomaly, the associated abnormal cells are then identified through sequential potential perturbation of candidate associated cells and multi-shell response analysis. Through the above sequence arrangement, the measurement reliability judgment, anomaly nature judgment, and anomaly location judgment can form a continuous and complete diagnostic process.

[0065] The above description is only a preferred embodiment 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 measuring abnormal casing potential and identifying leakage current in a battery series system, characterized in that, Includes the following steps: Collect the casing potential of each casing in the battery series system; determine the target casing based on the casing potential, wherein the target casing is the casing whose casing potential exceeds the abnormal threshold; Apply localized thermal disturbance to the corresponding region of the target shell; The shell potential of the target shell is collected during the sampling periods before, during, and after the disturbance. An anomaly category feature vector is constructed based on the shell potential of the target shell during the sampling periods before the disturbance, during the disturbance, and after the disturbance. The anomaly category feature vector is then input into the anomaly category discrimination model to obtain the discrimination result that the anomaly corresponding to the target shell is either a pseudo leakage anomaly of the surface conductive medium or a real leakage anomaly.

2. The method for measuring abnormal casing potential and identifying leakage current in a battery series system according to claim 1, characterized in that, Collect the casing potential of each casing in the battery series system, and collect the node potential of the corresponding cell in each casing; Determining the target shell based on the shell potential includes calculating the potential difference between the shell potential of the candidate shell and the corresponding node potential of the candidate shell, and calculating the shell potential difference between the candidate shell and the preceding shell adjacent in the series direction, and the shell potential difference between the candidate shell and the following shell adjacent in the series direction. When the potential difference value is greater than the first preset abnormal threshold, and at least one of the shell potential difference value between the candidate shell and the previous shell and the shell potential difference value between the candidate shell and the next shell is greater than the second preset abnormal threshold, the candidate shell is determined as the target shell.

3. The method for measuring abnormal casing potential and identifying leakage current in a battery series system according to claim 1, characterized in that, Applying a local thermal disturbance to the area corresponding to the target housing includes controlling the operation of a thermal disturbance unit corresponding to the target housing to raise the temperature of the area corresponding to the target housing from the initial temperature to the target temperature and maintain it for a first preset duration. The shell potential is collected during the sampling period before the start of the local thermal disturbance within a second preset duration before the start of the disturbance; the shell potential is collected during the sampling period during the disturbance within a first preset duration; and the shell potential is collected during the sampling period after the end of the local thermal disturbance within a third preset duration.

4. The method for measuring abnormal casing potential and identifying leakage current in a battery series system according to claim 1, characterized in that, The anomaly category feature vector includes the average shell potential before the disturbance, the average shell potential during the disturbance, the average shell potential after the disturbance, the first potential difference between the average shell potential before the disturbance and the average shell potential during the disturbance, the second potential difference between the average shell potential during the disturbance and the average shell potential after the disturbance, the temperature change of the region corresponding to the target shell, and the recovery time required for the shell potential of the target shell to recover to no greater than the anomaly threshold within the sampling period after the disturbance. The anomaly classification model is a machine learning model trained using labeled pseudo leakage anomaly samples and real leakage anomaly samples of surface conductive medium; the classification result includes pseudo leakage anomaly labels and real leakage anomaly labels of surface conductive medium.

5. The method for measuring abnormal casing potential and determining leakage current in a battery series system according to claim 1, characterized in that, Before acquiring the casing potential of each casing in the battery series system, the casing sampling channel corresponding to each casing is self-verified. The self-verification includes, under the first reference bias state, controlling the switching unit to connect the casing sampling channel to the first reference bias node through the first preset resistor, and acquiring the first channel response value of the casing sampling channel under the first reference bias state. In the second reference bias state, the switching unit is controlled to connect the housing sampling channel to the second reference bias node through the second preset resistor, and the second channel response value of the housing sampling channel in the second reference bias state is collected; A channel health feature vector is constructed based on the first channel response value and the second channel response value, and the channel health feature vector is input into the channel health discrimination model to obtain the discrimination result of whether the shell sampling channel is a valid shell sampling channel.

6. The method for measuring abnormal casing potential and determining leakage current in a battery series system according to claim 5, characterized in that, The channel health feature vector includes a first deviation value between the first channel response value and the reference potential corresponding to the first reference bias node, a second deviation value between the second channel response value and the reference potential corresponding to the second reference bias node, a response difference between the first channel response value and the second channel response value, a first stabilization time required for the first channel response value to reach a stable state, and a second stabilization time required for the second channel response value to reach a stable state. The channel health discrimination model is a machine learning model trained using labeled valid shell sampling channel samples and failed shell sampling channel samples; the discrimination result includes valid shell sampling channel labels and failed shell sampling channel labels.

7. The method for measuring abnormal casing potential and determining leakage current in a battery series system according to claim 5, characterized in that, Shell potential is collected only for shell sampling channels that are determined to be valid shell sampling channels by the channel health discrimination model. When the channel health discrimination model determines that the housing sampling channel is a failed housing sampling channel, it records the abnormal channel state of the housing corresponding to the housing sampling channel and prohibits the execution of the target housing determination step based on the housing potential of the housing corresponding to the housing sampling channel.

8. The method for measuring abnormal casing potential and determining leakage current in a battery series system according to claim 4, characterized in that, After the anomaly classification model determines the anomaly corresponding to the target housing as a real leakage anomaly, potential perturbations are sequentially applied to the candidate associated cells corresponding to the target housing. During each potential perturbation application, the perturbation response values ​​of the target housing, the preceding housing adjacent to the target housing in the series direction, and the following housing adjacent to the target housing in the series direction are collected. A location feature vector is constructed based on the perturbation response values, and the location feature vector is input into the associated abnormal cell classification model to obtain the associated abnormal cell classification result.

9. The method for measuring abnormal casing potential and determining leakage current in a battery series system according to claim 8, characterized in that, The candidate associated cells include the cell corresponding to the target housing, the preceding cell adjacent to the target housing in the series direction, and the following cell adjacent to the target housing in the series direction; potential perturbations are sequentially applied to the candidate associated cells, including controlling the conduction of the equalization branch or perturbation branch corresponding to the candidate associated cells, so that the node potential of the candidate associated cells to which potential perturbations are applied changes by a fourth preset potential amplitude relative to before the potential perturbation is applied, and maintains a fourth preset duration.

10. The method for measuring abnormal casing potential and determining leakage current in a battery series system according to claim 8, characterized in that, The positioning feature vector includes the target response amplitude of the target shell under each potential disturbance, the response amplitude of the first adjacent shell of the preceding shell in the series direction under each potential disturbance, the response amplitude of the second adjacent shell of the following shell in the series direction under each potential disturbance, the target response duration required for the target response amplitude to reach its peak value, the first adjacent shell response duration required for the first adjacent shell response amplitude to reach its peak value, and the second adjacent shell response duration required for the second adjacent shell response amplitude to reach its peak value. The associated abnormal cell discrimination model is a machine learning model trained using labeled associated abnormal cell samples; the associated abnormal cell discrimination result includes the cell label corresponding to the target casing, the previous cell label, the next cell label, and an undetermined label; when the associated abnormal cell discrimination model outputs an undetermined label, it only retains the discrimination result of the real leakage abnormality and does not output the associated abnormal cell.

Images