Battery abnormality detection method and system for battery swap cabinet based on thermal image

Abstract

The application discloses a battery abnormality detection method and system based on thermal images, and relates to the field of battery safety monitoring.The method comprises the following steps: collecting surface thermal image sequences in the battery compartment through an infrared thermal imager arranged on the battery swap cabinet, which has a time sequence and a charging stage label; extracting temperature field features in the time and space dimensions from the thermal image sequences, and identifying abnormal states and abnormal areas of the battery based on the temperature field features; obtaining real-time battery management system (BMS) data of the battery, verifying the correlation between the abnormal states or abnormal areas and the BMS data, and determining the abnormality level and the location of the faulty battery unit according to the correlation verification result.The technical problems of inaccurate and untimely detection and inaccurate fault positioning of the existing battery swap cabinet battery abnormality detection are solved, and the technical effects of improving the accuracy, timeliness and positioning accuracy of battery abnormality detection are achieved.

Description

Technical Field

[0001] This application relates to the field of battery safety monitoring, and in particular to a method and system for detecting abnormal batteries in battery swapping cabinets based on thermal images. Background Technology

[0002] Thermal runaway of lithium batteries during charging in battery swapping cabinets is a core risk posing a significant safety hazard, potentially leading to fires and other major accidents. Real-time and accurate anomaly detection is crucial for ensuring public safety. Currently, the industry primarily relies on electrical parameters such as voltage and current provided by Battery Management Systems (BMS) for threshold alarms, or uses traditional point-temperature sensors to monitor the temperature at a few fixed locations. However, BMS alarms cannot pinpoint the physical location of faulty cells and are insensitive to slowly developing faults such as early internal short circuits. Point-temperature sensors, due to their limited deployment, have monitoring blind spots and struggle to comprehensively capture anomalies such as localized overheating on the battery surface, resulting in delayed or missed warnings.

[0003] At present, the battery anomaly detection technology in battery swapping cabinets suffers from technical problems such as inaccurate and untimely detection, as well as inaccurate fault location. Summary of the Invention

[0004] This application provides a method and system for detecting battery anomalies in battery swapping cabinets based on thermal images. It employs an infrared thermal imager to acquire a sequence of thermal images of the battery compartment surface with time-series and charging stage labels. Spatiotemporal temperature field features are extracted from the thermal image sequence to identify abnormal battery states and regions. Real-time BMS data is acquired, and the correlation between the abnormal state or region and the data is verified to determine the anomaly level and locate the faulty battery unit. These technical means solve the technical problems of inaccurate and untimely detection and inaccurate fault location in existing battery anomaly detection methods for battery swapping cabinets, achieving the technical effect of improving the accuracy, timeliness, and precision of battery anomaly detection.

[0005] This application provides a method for detecting battery anomalies in a battery swapping cabinet based on thermal images, comprising: acquiring a sequence of surface thermal images inside the battery compartment using an infrared thermal imager installed in the battery swapping cabinet, which has a time series and charging stage labels; extracting spatiotemporal temperature field features from the thermal image sequence, and identifying abnormal states and abnormal areas of the battery based on the temperature field features; acquiring real-time battery management system (BMS) data of the battery, verifying the correlation between the abnormal state or abnormal area and the BMS data, and determining the anomaly level and locating the faulty battery cell based on the correlation verification results.

[0006] In a possible implementation, the spatiotemporal temperature field features are extracted from the thermal image sequence. Before this, the following processing is performed: noise filtering and non-uniformity correction preprocessing is applied to the original thermal image to obtain high-quality basic temperature field data; the preprocessed thermal image is pixel-level registered with the synchronously acquired visible light image to establish a precise correspondence between each pixel in the thermal image and the physical location of the battery; based on the registration relationship, an adaptive contrast enhancement algorithm is used for key parts of the battery to highlight the preset granular temperature difference features of the key parts of the battery, and thermal image sequence optimization and enhancement processing is performed.

[0007] In a possible implementation, the spatiotemporal temperature field features are extracted from the thermal image sequence, and the following processing is performed: based on the registered and enhanced thermal image sequence, the highest temperature, average temperature, temperature standard deviation, and maximum temperature difference between different cells in each battery cell region are calculated to obtain spatial temperature field features; during the constant current charging stage, the temperature rise rate of the battery cell region is calculated using a sliding window method, and the temperature change trend is analyzed by cumulative summation to obtain temporal temperature field features; the spatial temperature field features and temporal temperature field features are merged to obtain the spatiotemporal temperature field features.

[0008] In a possible implementation, the abnormal state or abnormal region is correlated with BMS data, and the following processing is performed: the abnormal region in the thermal image is mapped and matched with the cell location of the BMS voltage abnormality, and a first verification result is obtained by comparing the spatial location overlap with a set threshold; a cross-correlation algorithm is used to analyze the time series of temperature change and voltage fluctuation, and when there is a significant correlation and the temperature change lags behind the voltage change, the causal relationship of the fault is confirmed, and a second verification result is obtained; a feature library containing typical faults such as internal short circuit, poor connection, and lithium plating is established, and the matching degree between the abnormal region and each fault mode is calculated by multi-feature weighted fusion to obtain a third verification result; the correlation verification result is determined based on the fusion result of the first verification result, the second verification result, and the third verification result.

[0009] In a possible implementation, a correlation verification result is determined based on the fusion result of the first verification result, the second verification result, and the third verification result, and the following processing is performed: configuring the fusion weight of each verification result; weighting and fusing the first verification result, the second verification result, and the third verification result according to the fusion weight to obtain a fusion correlation evaluation value; and determining the correlation verification result based on the fusion correlation evaluation value.

[0010] In a possible implementation, based on the temperature field characteristics, abnormal states and abnormal regions of the battery are identified, and the following processing is performed: the battery surface is divided into multiple regularly arranged temperature monitoring units; based on the temperature field characteristics, the average temperature of each monitoring unit is calculated; the average temperature of any monitoring unit is compared with a first absolute temperature threshold, and if it exceeds the threshold, the area where the corresponding monitoring unit is located is determined to be an abnormal region.

[0011] In a possible implementation, the following processing is also performed: at a single time point, the statistical distribution of temperature values ​​of all monitoring units is calculated, including the global temperature standard deviation and range; when the global temperature standard deviation or range exceeds a preset consistency threshold, it is determined that the battery has an abnormal temperature distribution and the abnormal area is located.

[0012] In possible implementations, to locate abnormal areas, the following processing is also performed: for each monitoring unit, extract the temperature time series during a complete charging process; calculate the temperature rise rate of each series or the fitting residual with the standard charging temperature rise curve; when the temperature rise rate or fitting residual of the monitoring unit exceeds the abnormal threshold, it is determined that there is an abnormality in the area where the corresponding monitoring unit is located.

[0013] In a possible implementation, abnormal states and abnormal regions of the battery are identified, and the following processing is performed: a spatiotemporal feature vector is constructed, wherein the spatiotemporal feature vector contains at least the real-time temperature, temperature rise rate, and temperature gradient with surrounding units of each monitoring unit; the spatiotemporal feature vector is input into a pre-trained anomaly classification model, and the anomaly classification model outputs the anomaly probability of each monitoring unit; adjacent monitoring units with anomaly probabilities exceeding a threshold are clustered to obtain the boundary of the abnormal region, and the abnormal region is located.

[0014] This application also provides a battery anomaly detection system for battery swapping cabinets based on thermal images, including: a surface thermal image sequence acquisition module, used to acquire surface thermal image sequences inside the battery compartment using an infrared thermal imager installed in the battery swapping cabinet, which has time series and charging stage labels; an anomaly identification module, used to extract spatiotemporal temperature field features from the thermal image sequence, and identify abnormal states and abnormal areas of the battery based on the temperature field features; and a correlation verification module, used to acquire real-time battery management system (BMS) data of the battery, perform correlation verification between the abnormal state or abnormal area and the BMS data, and determine the anomaly level and locate the faulty battery cell based on the correlation verification results.

[0015] The proposed method and system for battery anomaly detection in battery swapping cabinets based on thermal images first acquires a sequence of surface thermal images of the battery compartment using an infrared thermal imager installed in the swapping cabinet. These images include time-series data and charging stage labels. Then, spatiotemporal temperature field features are extracted from the thermal image sequence. Based on these features, abnormal battery states and regions are identified. Finally, real-time battery management system (BMS) data is acquired, and the abnormal states or regions are correlated with the BMS data. Based on the correlation verification results, the anomaly level is determined, and the faulty battery cell is located. This method achieves the technical effect of improving the accuracy, timeliness, and precision of battery anomaly detection. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the battery anomaly detection method for battery swapping cabinets based on thermal images provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of a battery anomaly detection system for a battery swapping cabinet based on thermal images, provided in an embodiment of this application.

[0019] Figure labeling: Surface thermal image sequence acquisition module 10, anomaly identification module 20, correlation verification module 30. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a method for detecting battery anomalies in battery swapping cabinets based on thermal images, such as... Figure 1 As shown, the method includes:

[0024] Step S100: Collect a series of surface thermal images inside the battery compartment using an infrared thermal imager installed in the battery swapping cabinet. The images have a time series and charging stage labels.

[0025] Specifically, an infrared thermal imager is an imaging device that detects the infrared radiation emitted by an object and converts it into a visible image of temperature distribution, with colors in the image representing temperature levels. One or more infrared thermal imagers are fixedly installed inside or in front of each battery compartment, and these imagers connect to a central processing server via Ethernet or Wi-Fi. Raw thermal data is acquired from the infrared thermal imager at a fixed frame rate, such as 1 frame / second or 1 frame / 5 seconds, through a data acquisition service. Each data point is a two-dimensional matrix, and each value in the matrix represents the absolute temperature value of a pixel. The acquisition time for each frame of the thermal image is recorded, such as year, month, day, hour, minute, second, and millisecond, forming a time series.

[0026] The data acquisition service communicates in real time with the main controller of the battery swapping cabinet or the battery BMS via communication interfaces such as CAN bus or RS485 to obtain the current charging status information of the battery. The charging stages are divided into resting, constant current charging, constant voltage charging, and charging complete. Based on the current and voltage data provided by the BMS, the data acquisition service tags each frame of the thermal image with a corresponding charging stage label; for example, the label "CC" represents constant current charging. The heat generation mechanisms and temperature rise patterns differ in different charging stages, and the charging stage labels are used for accurate anomaly detection.

[0027] Step S200: Extract the spatiotemporal temperature field features from the thermal image sequence, and identify the abnormal state and abnormal region of the battery based on the temperature field features.

[0028] Specifically, the spatiotemporal dimension refers to the simultaneous inclusion of spatial and temporal information; spatial refers to the location on the battery surface, and temporal refers to the moment in the charging process. Temperature field characteristics refer to quantifiable indicators calculated from the temperature distribution image that characterize its state, such as maximum temperature and average temperature rise rate. Abnormal state refers to the overall health status conclusion of the battery, such as "normal," "abnormal," or "severely abnormal." Abnormal region refers to specific locations on the battery surface identified as exhibiting overheating or irregular temperature rise.

[0029] The raw thermal image sequence is transformed into quantifiable features. A series of image preprocessing and computer vision techniques are used to improve data quality and establish a mapping between temperature and physical location. Features describing the spatial distribution and temporal variation of temperature are extracted from the purified data. Based on these features, multi-criteria decision-making or machine learning models are employed to determine whether the battery is abnormal and where the abnormality occurs.

[0030] In one possible implementation, before extracting the spatiotemporal temperature field features from the thermal image sequence, step S200 further includes step S210, which preprocesses the original thermal image with noise filtering and non-uniformity correction to obtain high-quality basic temperature field data. Specifically, standard filters in digital image processing are used for noise filtering. For example, a median filter is used, specifically by traversing each pixel of the image with a 3×3 or 5×5 sliding window, sorting the temperature values ​​of all pixels within the window, and replacing the value of the original center pixel with the median value. Through noise filtering, random and isolated erroneous pixel values ​​in the image are eliminated, improving image quality.

[0031] Non-uniformity correction refers to correcting errors caused by inconsistencies in the response characteristics of individual pixels in an infrared thermal imager, ensuring that the temperature displayed at all points in the image is consistent when measuring objects at the same temperature. The specific operation involves placing a blackbody source with a uniform temperature in front of the infrared thermal imager lens. The processor inside the imager acquires a thermal image of the blackbody. Because the blackbody has a uniform temperature, the response differences between different pixels in the image represent the non-uniformity error. The processor calculates a correction coefficient matrix, and all subsequently acquired image data is multiplied by this matrix to eliminate the response deviations of individual pixels.

[0032] Step S220 involves pixel-level registration of the preprocessed thermal image with the synchronously acquired visible light image to establish a precise correspondence between each pixel in the thermal image and the physical location of the battery. Specifically, a feature-based image registration algorithm, such as SIFT (Scale Invariant Feature Transform), is used for pixel-level registration. The SIFT algorithm is run on both the visible light and thermal images to detect key points in the images, such as the battery's corners, screws, and labels. A descriptor, describing the surrounding image information, is calculated for each key point. Matching methods such as the k-nearest neighbor algorithm are used to find the key point pairs with the most similar descriptors in the visible light and thermal images. Using the successfully matched point pairs, a homography matrix is ​​estimated using the RANSAC (Random Sampling Consensus) algorithm. This matrix defines how to perform a perspective transformation on the thermal image to align it with the visible light image. Using the calculated homography matrix, a geometric transformation, such as an affine transformation or perspective transformation, is performed on the thermal image to ensure that each pixel corresponds one-to-one with its physical location in the visible light image.

[0033] Step S230: Based on the registration relationship, an adaptive contrast enhancement algorithm is used for key parts of the battery to highlight the preset granularity temperature difference features of these key parts, performing thermal image sequence optimization and enhancement processing. Specifically, the adaptive contrast enhancement refers to adjusting the contrast according to the characteristics of local areas of the image, rather than a globally uniform adjustment. The preset granularity refers to the scale of the temperature difference of interest, for example, whether to focus on the temperature difference of the entire battery pack or a small temperature difference of a few degrees Celsius within a single cell. Image enhancement is performed using limited contrast adaptive histogram equalization, dividing the image into several small patches, such as 8×8. Histogram equalization is performed on each patch, and the histogram is cropped during the calculation of the transform function to limit the magnification and avoid excessive noise enhancement. After processing each patch, bilinear interpolation is used to stitch adjacent patches together to eliminate block artifacts. By limiting contrast adaptive histogram equalization, the small temperature difference between key battery parts, such as the center and edge of the cell, becomes more apparent in the image.

[0034] In one possible implementation, temperature field features in the spatiotemporal dimension are extracted from the thermal image sequence. Step S200 further includes step S240, which, based on the registered and enhanced thermal image sequence, calculates the highest temperature, average temperature, temperature standard deviation, and maximum temperature difference between different cells for each battery cell region to obtain spatial temperature field features. Specifically, spatial temperature field features are quantitative indicators describing the spatial distribution of temperature at a certain moment. Based on the registered visible light image, image segmentation algorithms, such as the deep learning-based semantic segmentation model U-Net or simple template matching, are used to identify the contour of each battery cell and assign it a unique ID. For each frame of thermal image, the pixel region corresponding to each battery cell ID is traversed, and the temperature values ​​of all pixels in that region are read. The highest temperature of this cell is obtained by finding the maximum value among all temperature values ​​in that region. The average temperature of this cell is calculated by summing the temperature values ​​of all pixels and dividing by the total number of pixels. Based on the calculated average temperature, the square root of the average of the squares of the differences between the temperature values ​​of all pixels and the average temperature is further calculated to obtain the temperature standard deviation, which reflects the uniformity of temperature distribution. After processing all individual battery cells, the average temperature of each cell is collected and listed. By identifying the maximum and minimum values ​​in this list and subtracting the minimum value from the maximum, the maximum temperature difference between cells, representing the extreme temperature difference within the entire battery pack, is calculated.

[0035] In step S250, during the constant current charging stage, the temperature rise rate of the battery cell region is calculated using a sliding window method, and the temperature change trend is analyzed using cumulative sum to obtain the time-series temperature field characteristics. Specifically, the time-series temperature field characteristics are quantitative indicators describing the temperature change trend over time. The sliding window is a fixed-size data buffer; as new data is added, the oldest data is removed to process the real-time data stream. The cumulative sum is a sequence analysis algorithm that detects small-amplitude drifts in the process mean by accumulating deviations.

[0036] Based on the charging stage label in step S100, select thermal image data labeled "constant current charging". Define a time window, such as containing the most recent 10 data points, and use a linear regression algorithm to fit the temperature change curve over time within this window. The slope of the fitted line is the temperature rise rate. The window slides as new data points are added, enabling real-time calculation. Maintain a cumulative sum variable. For each new temperature rise value, calculate its deviation from the expected normal temperature rise. The cumulative sum variable = previous cumulative sum variable + (actual temperature rise - expected normal temperature rise). When the value of the cumulative sum variable exceeds a preset threshold, it indicates a small but continuous abnormal change in the temperature rise trend.

[0037] Step S260: Merge the spatial temperature field features and the temporal temperature field features to obtain the spatiotemporal temperature field features. Specifically, create a feature vector for each battery cell. For example, at time point t, the feature vector of a certain battery cell could be: [highest temperature] t average temperature t Temperature standard deviation t rate of temperature rise t Cumulative sum t At the same time, global features such as maximum temperature difference t It will also be added. This vector is the temperature field feature of the spatiotemporal dimension, which integrates spatial and temporal information.

[0038] In one possible implementation, based on the temperature field characteristics, abnormal states and abnormal areas of the battery are identified. Step S200 further includes step S270, dividing the battery surface into multiple regularly arranged temperature monitoring units; step S280, calculating the average temperature of each monitoring unit based on the temperature field characteristics; and step S290, comparing the average temperature of any monitoring unit with a first absolute temperature threshold. If the threshold is exceeded, the area where the corresponding monitoring unit is located is determined to be an abnormal area.

[0039] Specifically, the entire battery surface image is meshed. For example, the image is divided into 32×32 pixel squares, each square being a temperature monitoring unit. The average temperature of all pixels within each square is calculated. A fixed safety threshold, namely the first absolute temperature threshold, is set as a fixed upper limit based on the battery materials and safety standards. All squares are iterated through; if the average temperature of any square exceeds the first absolute temperature threshold, that square is marked as abnormal. All marked squares together constitute the abnormal region.

[0040] In one possible implementation, step S200 further includes step S2100, calculating the statistical distribution of temperature values ​​of all monitoring units at a single time point, including the global temperature standard deviation and range; step S2110, when the global temperature standard deviation or range exceeds a preset consistency threshold, determining that the battery has an abnormal temperature distribution and locating the abnormal area.

[0041] Specifically, at each specific detection moment, the average temperature values ​​of all monitoring units are aggregated to form a dataset containing temperature data from all units. The global temperature standard deviation of this dataset is calculated; this value reflects the dispersion of all monitoring unit temperatures relative to the average temperature. The highest and lowest temperature values ​​in this dataset are identified, and the range is calculated by subtracting the minimum value from the maximum. This range reflects the maximum fluctuation range of the battery surface temperature. These two statistics are compared with preset safety thresholds. For example, the standard deviation threshold is set to 5°C, and the range threshold to 15°C. If the calculated global temperature standard deviation is greater than 5°C, or the range is greater than 15°C, either condition is met, and the battery is determined to have an abnormal temperature distribution. In this case, the abnormal area refers to those parts where the temperature significantly deviates from the overall average level. A statistical distribution-based method is used for location: first, the average temperature of all monitoring units is calculated, and then the global temperature standard deviation is calculated. A judgment boundary is set; for example, the value of "overall average temperature plus twice the standard deviation" is used as the high-temperature anomaly threshold. By iterating through all monitoring units, any unit whose average temperature exceeds the calculated high-temperature anomaly threshold will be identified as a high-temperature anomaly area.

[0042] In one possible implementation, to locate the abnormal area, step S200 further includes step S2120, extracting the temperature time series during a complete charging process for each monitoring unit; step S2130, calculating the temperature rise rate of each series or the fitting residual with the standard charging temperature rise curve; step S2140, when the temperature rise rate or fitting residual of the monitoring unit exceeds the abnormal threshold, determining that there is an abnormality in the area where the corresponding monitoring unit is located.

[0043] Specifically, after a charge is completed, the temperature data of each monitoring unit throughout the charging process is extracted from the database to form a temperature-time curve. A standard temperature rise curve is pre-trained using a large amount of normal battery charging data; this is an idealized or averaged curve showing the temperature change over time during battery charging under normal operating conditions—the baseline. A curve fitting algorithm, such as the least squares method, is used to fit the measured curve of each unit to the standard curve. The difference between the measured point and the corresponding point on the standard curve is calculated—the residual—and the sum of squares or root mean square error (RMSE) of the residuals is calculated. If the RMSE value of a certain unit is significantly higher than that of other units, for example, exceeding three standard deviations of the average RMSE of normal units, then that unit is considered abnormal.

[0044] In one possible implementation, identifying abnormal states and abnormal regions of the battery, step S200 further includes step S2150, constructing a spatiotemporal feature vector, wherein the spatiotemporal feature vector at least includes the real-time temperature, temperature rise rate, and temperature gradient with surrounding units of each monitoring unit; step S2160, inputting the spatiotemporal feature vector into a pre-trained anomaly classification model, and the anomaly classification model outputting the anomaly probability of each monitoring unit; step S2170, clustering adjacent monitoring units whose anomaly probabilities exceed a threshold to obtain the boundary of the abnormal region and locate the abnormal region.

[0045] Specifically, the anomaly classification model is a machine learning model that, after training, can automatically distinguish between normal and anomalous patterns. Clustering is used to merge discrete anomaly points into continuous anomalous regions. A feature vector is constructed for each monitoring unit, for example: [unit real-time temperature, unit temperature rise rate, temperature gradient between the unit and its four neighboring units above / below / left / right]. Using an Isolation Forest or Support Vector Machine model pre-trained with a large amount of labeled normal / anomaly data, the feature vector is input into the model, and the model outputs a value between 0 and 1, representing the probability that the unit is an anomaly. The locations of all units with probabilities exceeding a set threshold are identified on the image, and the DBSCAN spatial clustering algorithm is used to aggregate these discrete anomaly points into one or more continuous anomalous regions, calculating the contour boundaries of these regions.

[0046] Step S300: Obtain real-time battery management system (BMS) data of the battery, verify the correlation between the abnormal state or abnormal area and the BMS data, and determine the abnormal level and locate the faulty battery cell based on the correlation verification results.

[0047] Specifically, BMS data refers to the electrical parameters monitored by the battery management system, such as voltage, current, and SOC. Correlation verification refers to the process of comparing thermal imaging with abnormal signals from electrical sources to confirm whether they point to the same fault source.

[0048] The electrical parameters of the BMS are acquired in real time through a data interface, such as a CAN bus reader. The thermal anomaly conclusion output in step S200 is then compared with the electrical anomaly conclusion of the BMS through cross-modal correlation analysis. This analysis considers multiple dimensions, including spatial location, time series, and fault mode, to arrive at a comprehensive reliability score. Based on this score, the anomaly level is determined, such as "Level 1 Warning" or "Level 2 Alarm," and the most likely faulty individual battery or battery module is located.

[0049] In one possible implementation, the abnormal state or abnormal region is correlated with BMS data. Step S300 further includes step S310, which maps and matches the abnormal region in the thermal image with the cell location of the BMS voltage abnormality. A first verification result is obtained by comparing the spatial overlap with a set threshold. Specifically, voltage data for each battery cell is obtained from the BMS, and voltage abnormality conditions are set, such as a cell voltage less than 2.5V or a cell voltage difference greater than 0.1V from the average voltage. The BMS cell number corresponding to the area marked as abnormal on the thermal image is checked to see if it also triggers a voltage abnormality. The matching degree is calculated as: Matching degree = (Number of cells with simultaneous thermal and voltage abnormalities) / (Total number of cells with thermal abnormalities). If the matching degree exceeds a threshold, such as 60%, the verification is considered successful.

[0050] Step S320: The cross-correlation algorithm is used to analyze the time series of temperature changes and voltage fluctuations. When a significant correlation exists and the temperature change lags behind the voltage change, the causal relationship of the fault is confirmed, and a second verification result is obtained. Specifically, the cross-correlation function is used for calculation. The cross-correlation algorithm is a method to measure the similarity of two time series at different time offsets. Temperature and voltage series within the same time period are selected and normalized. The correlation coefficients of the temperature and voltage series at different time lags are calculated, i.e., the moving average voltage series. The correlation coefficient with the temperature series is calculated, and the lag value that maximizes the correlation coefficient is found. If the maximum correlation coefficient is very high, such as >0.8, and the lag value is positive, meaning the voltage change occurs before the temperature change, this conforms to the physical law that "electrical faults, such as internal short circuits leading to local overheating," thus confirming the causal relationship of the fault.

[0051] Step S330: Establish a feature library containing typical faults such as internal short circuits, poor connections, and lithium plating. Calculate the matching degree between abnormal regions and each fault mode through multi-feature weighted fusion to obtain the third verification result. Specifically, based on historical fault data, laboratory analysis reports, and domain expert knowledge, define a digital feature vector template for each known typical fault mode. Each template consists of multiple key feature dimensions. For example: Internal short circuit feature = [temperature rise rate: extremely high, temperature distribution: local hot spot, voltage correlation: high], Poor connection feature = [temperature rise rate: moderate, temperature distribution: hot spot at the connection point, voltage correlation: low], Lithium plating feature = [temperature rise rate: low or abnormally slow, temperature distribution: overall uniformity or specific low / high temperature zone, correlation with charging rate / low temperature environment: high, capacity decay rate: fast]. Simultaneously, quantify each feature dimension. The features exhibited by the currently detected abnormal battery region are scored according to the same dimensions and quantification standards as the feature library, forming a current abnormal feature vector. Algorithms such as cosine similarity are used to calculate the similarity between the current abnormal feature vector and each fault template vector in the feature library. Cosine similarity determines the consistency of direction by calculating the cosine of the angle between two vectors; the closer the value is to 1, the higher the matching degree. Since different features have varying importance in diagnosing different faults, a preset weight is assigned to each feature dimension. For example, when diagnosing internal short circuits, the weight of "temperature rise rate" is much higher than that of "temperature distribution location." The formula for calculating the overall matching degree score is: Overall Matching Degree = (Feature 1 Similarity × Weight 1) + (Feature 2 Similarity × Weight 2) + ... Finally, three overall matching degree scores are obtained for the current abnormal area and fault modes such as "internal short circuit," "poor connection," and "lithium plating." The fault mode with the highest score is the most likely cause of the fault.

[0052] Step S340: Determine the correlation verification result based on the fusion result of the first verification result, the second verification result, and the third verification result. Step S340 further includes step S341: configuring the fusion weights of each verification result; and step S342: weighting and fusing the first verification result, the second verification result, and the third verification result according to the fusion weights to obtain a fusion correlation evaluation value, and determining the correlation verification result based on the fusion correlation evaluation value.

[0053] Specifically, weights are assigned to the three verification methods based on expert experience or machine learning optimization. For example, W1 (spatial verification) = 0.3, W2 (temporal verification) = 0.4, and W3 (pattern matching) = 0.3. The outputs of each verification method, such as the matching degree of S310, the correlation coefficient of S320, and the matching degree score of S330, are normalized to between 0 and 1, and then the following calculation is made: Fusion Score = W1 × Score 1 + W2 × Score 2 + W3 × Score 3. A final threshold is set. If the fusion score is greater than this threshold, the thermal anomaly is confirmed as a real fault, and the anomaly level is determined based on the score and the matched fault mode, such as warning, severe, or critical.

[0054] This application embodiment uses an infrared thermal imager of the battery swapping cabinet to collect a sequence of thermal images of the battery compartment surface with time series and charging stage labels. The spatiotemporal temperature field features are extracted from the thermal image sequence to identify abnormal battery states and areas. Real-time BMS data is obtained, and the correlation between the abnormal state or area and the data is verified to determine the abnormality level and locate the faulty battery unit. These technical means solve the technical problems of inaccurate and untimely detection and inaccurate fault location in existing battery swapping cabinets, and achieve the technical effect of improving the accuracy, timeliness and accuracy of battery abnormality detection.

[0055] In the above text, refer to Figure 1 This paper describes in detail a battery anomaly detection method for battery swapping cabinets based on thermal images according to embodiments of the present invention. Next, reference will be made to... Figure 2 A thermal imaging-based battery anomaly detection system for battery swapping cabinets according to an embodiment of the present invention is described.

[0056] The battery anomaly detection system for battery swapping cabinets based on thermal images according to embodiments of the present invention addresses the technical problems of inaccurate, untimely, and inaccurate fault location in existing battery anomaly detection methods for battery swapping cabinets, thereby improving the accuracy, timeliness, and precision of battery anomaly detection. The battery anomaly detection system for battery swapping cabinets based on thermal images includes: a surface thermal image sequence acquisition module 10, an anomaly identification module 20, and a correlation verification module 30.

[0057] The surface thermal image sequence acquisition module 10 is used to acquire surface thermal image sequences inside the battery compartment using an infrared thermal imager installed in the battery swapping cabinet. The sequences have time series and charging stage labels. The anomaly identification module 20 is used to extract spatiotemporal temperature field features from the thermal image sequences and identify abnormal states and abnormal areas of the battery based on the temperature field features. The correlation verification module 30 is used to acquire real-time battery management system (BMS) data of the battery, perform correlation verification between the abnormal states or abnormal areas and the BMS data, and determine the anomaly level and locate the faulty battery cell based on the correlation verification results.

[0058] The specific configuration of the anomaly identification module 20 is described in detail below: As mentioned above, before extracting the spatiotemporal temperature field features from the thermal image sequence, the anomaly identification module 20 may further include: a preprocessing unit for performing noise filtering and non-uniformity correction preprocessing on the original thermal image to obtain high-quality basic temperature field data; a pixel-level registration unit for performing pixel-level registration between the preprocessed thermal image and the synchronously acquired visible light image to establish a precise correspondence between each pixel in the thermal image and the physical location of the battery; and an enhancement processing unit for performing thermal image sequence optimization and enhancement processing based on the registration relationship, using an adaptive contrast enhancement algorithm to highlight the preset granular temperature difference features of the key parts of the battery.

[0059] The anomaly identification module 20, which extracts spatiotemporal temperature field features from the thermal image sequence, may further include: a spatial temperature field feature acquisition unit for calculating the highest temperature, average temperature, temperature standard deviation, and maximum temperature difference between different cells for each battery cell region based on the registered and enhanced thermal image sequence, to obtain spatial temperature field features; a temporal temperature field feature acquisition unit for calculating the temperature rise rate of the battery cell region in a sliding window manner during the constant current charging stage, and for analyzing the temperature change trend by accumulating and summing, to obtain temporal temperature field features; and a temperature field feature merging unit for merging the spatial temperature field features and the temporal temperature field features to obtain the spatiotemporal temperature field features.

[0060] The specific configuration of the correlation verification module 30 is described in detail below: As mentioned above, the correlation verification module 30 verifies the correlation between the abnormal state or abnormal region and the BMS data. The correlation verification module 30 may further include: a first verification unit for mapping and matching the abnormal region of the thermal image with the cell location of the abnormal voltage in the BMS, and obtaining a first verification result based on the comparison of the spatial location overlap with a set threshold; a second verification unit for analyzing the time series of temperature change and voltage fluctuation using a cross-correlation algorithm, and confirming the causal relationship of the fault when there is a significant correlation and the temperature change lags behind the voltage change, and obtaining a second verification result; a third verification unit for establishing a feature library containing typical faults such as internal short circuit, poor connection, and lithium plating, and calculating the matching degree between the abnormal region and each fault mode through multi-feature weighted fusion to obtain a third verification result; and a verification result fusion unit for determining the correlation verification result based on the fusion result of the first verification result, the second verification result, and the third verification result.

[0061] Specifically, based on the fusion result of the first verification result, the second verification result, and the third verification result, a correlation verification result is determined. The verification result fusion unit may further include: a fusion weight configuration subunit for configuring the fusion weight of each verification result; and a weighted fusion subunit for weighting and fusing the first verification result, the second verification result, and the third verification result according to the fusion weight to obtain a fusion correlation evaluation value, and determining the correlation verification result based on the fusion correlation evaluation value.

[0062] The abnormal identification module 20, which identifies abnormal states and abnormal regions of the battery based on the temperature field characteristics, may further include: a temperature monitoring division unit for dividing the battery surface into multiple regularly arranged temperature monitoring units; an average temperature calculation unit for calculating the average temperature of each monitoring unit based on the temperature field characteristics; and an abnormal region determination unit for comparing the average temperature of any monitoring unit with a first absolute temperature threshold, and determining that the area where the corresponding monitoring unit is located is an abnormal region if the threshold is exceeded.

[0063] The anomaly identification module 20 may further include: a statistical distribution calculation unit for calculating the statistical distribution of temperature values ​​of all monitoring units at a single time point, including the global temperature standard deviation and range; and an anomaly area location unit for determining that the battery has an uneven temperature distribution anomaly and locating the anomaly area when the global temperature standard deviation or range exceeds a preset consistency threshold.

[0064] The anomaly identification module 20, which locates abnormal areas, may further include: a temperature time series extraction unit for extracting the temperature time series during a complete charging process for each monitoring unit; a fitting residual calculation unit for calculating the temperature rise rate of each sequence or the fitting residual with the standard charging temperature rise curve; and an anomaly determination unit for determining that there is an anomaly in the area where the corresponding monitoring unit is located when the temperature rise rate or fitting residual of the monitoring unit exceeds the anomaly threshold.

[0065] The abnormality identification module 20, which identifies abnormal states and abnormal regions of the battery, may further include: a spatiotemporal feature vector construction unit for constructing spatiotemporal feature vectors, wherein the spatiotemporal feature vectors at least include the real-time temperature, temperature rise rate, and temperature gradient with surrounding units of each monitoring unit; an abnormality classification unit for inputting the spatiotemporal feature vectors into a pre-trained abnormality classification model, and the abnormality probability of each monitoring unit output by the abnormality classification model; and a clustering unit for clustering adjacent monitoring units whose abnormality probabilities exceed a threshold to obtain the boundary of the abnormal region and locate the abnormal region.

[0066] The battery anomaly detection system for battery swapping cabinets based on thermal images provided in this embodiment of the invention can execute the battery anomaly detection method for battery swapping cabinets based on thermal images provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0067] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0068] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for detecting battery anomalies in battery swapping cabinets based on thermal images, characterized in that, include: The infrared thermal imager installed in the battery swapping cabinet collects a series of surface thermal images inside the battery compartment, which have time series and charging stage labels. The spatiotemporal temperature field features are extracted from the thermal image sequence, and based on the temperature field features, the abnormal state and abnormal areas of the battery are identified. Obtain real-time battery management system (BMS) data of the battery, verify the correlation between the abnormal state or abnormal area and the BMS data, and determine the abnormal level and locate the faulty battery cell based on the correlation verification results. The process of verifying the correlation between the abnormal state or abnormal region and the BMS data includes: The abnormal areas in the thermal image are mapped and matched with the cell locations where the BMS voltage is abnormal. The first verification result is obtained by comparing the spatial overlap with a set threshold. The cross-correlation algorithm is used to analyze the time series of temperature change and voltage fluctuation. When there is a significant correlation and the temperature change lags behind the voltage change, the causal relationship of the fault is confirmed and a second verification result is obtained. A feature library containing typical faults such as internal short circuit, poor connection, and lithium plating was established. The matching degree between abnormal regions and each fault mode was calculated by multi-feature weighted fusion to obtain the third verification result. Based on the fusion result of the first verification result, the second verification result, and the third verification result, the correlation verification result is determined; The correlation verification result is determined based on the fusion result of the first verification result, the second verification result, and the third verification result, including: Configure the fusion weights for each verification result; The first verification result, the second verification result, and the third verification result are weighted and fused according to the fusion weight to obtain a fusion correlation evaluation value, and the correlation verification result is determined according to the fusion correlation evaluation value.

2. The method for detecting battery anomalies in a battery swapping cabinet based on thermal images according to claim 1, characterized in that, Extracting spatiotemporal temperature field features from the thermal image sequence, prior to: The original thermal image is preprocessed with noise filtering and non-uniformity correction to obtain high-quality basic temperature field data; The preprocessed thermal image and the synchronously acquired visible light image are registered at the pixel level to establish a precise correspondence between each pixel in the thermal image and the physical location of the battery. Based on the registration relationship, an adaptive contrast enhancement algorithm is adopted for key parts of the battery to highlight the pre-defined granular temperature difference features of key parts of the battery and to perform thermal image sequence optimization and enhancement processing.

3. The method for detecting abnormal batteries in a battery swapping cabinet based on thermal images according to claim 2, characterized in that, Extracting spatiotemporal temperature field features from the thermal image sequence includes: Based on the registered and enhanced thermal image sequence, the highest temperature, average temperature, temperature standard deviation, and maximum temperature difference between different cells in each cell region are calculated to obtain the spatial temperature field characteristics. During the constant current charging stage, the temperature rise rate of the battery cell region is calculated using a sliding window method, and the temperature change trend is analyzed by cumulative sum to obtain the time-series temperature field characteristics. The spatial temperature field features and the temporal temperature field features are combined to obtain the temperature field features in the spatiotemporal dimension.

4. The method for detecting battery anomalies in a battery swapping cabinet based on thermal images according to claim 3, characterized in that, Based on the temperature field characteristics, abnormal states and abnormal regions of the battery are identified, including: The battery surface is divided into multiple regularly arranged temperature monitoring units; Based on the temperature field characteristics, the average temperature of each monitoring unit is calculated; The average temperature of any monitoring unit is compared with a first absolute temperature threshold. If the threshold is exceeded, the area where the corresponding monitoring unit is located is determined to be an abnormal area.

5. The method for detecting abnormal batteries in a battery swapping cabinet based on thermal images according to claim 4, characterized in that, Also includes: At a single point in time, calculate the statistical distribution of temperature values ​​for all monitoring units, including the global temperature standard deviation and range; When the global temperature standard deviation or range exceeds a preset consistency threshold, it is determined that the battery has an abnormal temperature distribution, and the abnormal area is located.

6. The method for detecting battery anomalies in a battery swapping cabinet based on thermal images according to claim 5, characterized in that, Locating abnormal areas also includes: For each monitoring unit, extract the temperature time series during a complete charging process; Calculate the temperature rise rate or the fitting residual to the standard charging temperature rise curve for each sequence; When the temperature rise rate or fitting residual of the monitoring unit exceeds the abnormal threshold, it is determined that there is an abnormality in the area where the corresponding monitoring unit is located.

7. The method for detecting abnormal batteries in a battery swapping cabinet based on thermal images according to claim 6, characterized in that, Identify abnormal battery conditions and abnormal areas, including: Construct a spatiotemporal feature vector, which includes at least the real-time temperature, temperature rise rate, and temperature gradient with surrounding units for each monitoring unit; The spatiotemporal feature vector is input into a pre-trained anomaly classification model, which outputs the anomaly probability of each monitoring unit. Cluster adjacent monitoring units whose anomaly probability exceeds the threshold to obtain the boundary of the anomaly region and locate the anomaly region.

8. A battery anomaly detection system for battery swapping cabinets based on thermal imaging, characterized in that, The system is used to implement the battery anomaly detection method for battery swapping cabinets based on thermal images as described in any one of claims 1-7, and the system comprises: The surface thermal image sequence acquisition module is used to acquire surface thermal image sequences inside the battery compartment using an infrared thermal imager installed in the battery swapping cabinet. The sequence has time series and charging stage labels. An anomaly detection module is used to extract spatiotemporal temperature field features from the thermal image sequence, and based on the temperature field features, identify abnormal states and abnormal regions of the battery. The correlation verification module is used to acquire real-time battery management system (BMS) data of the battery, verify the correlation between the abnormal state or abnormal area and the BMS data, and determine the abnormal level and locate the faulty battery cell based on the correlation verification results.

Images