Accurate detection and positioning method for micro short circuit defect of lithium battery protection plate

Abstract

The invention relates to a lithium battery protection board micro short circuit defect accurate detection and positioning method in the new generation information technology field, and the method comprises the steps: employing a filtering algorithm to carry out the denoising processing of a thermal image containing a potential thermal abnormal point, and carrying out the adjustment of the denoised pixel value distribution to enhance the image contrast, obtaining a clear thermal image, namely a processed image which highlights thermal distribution details of the lithium battery protection plate; performing region segmentation on the clear thermal image by adopting a clustering algorithm, classifying segmented thermal intensity clusters, namely pixel groups with different temperature levels, and determining a high-heat region set, namely a region group with relatively high temperature on the lithium battery protection plate; and processing the feature vectors by adopting a classification layer of a neural network algorithm, comparing classification output probabilities to identify the highest matching degree, and determining a micro short circuit confirmation area, namely determining a specific area of the lithium battery protection board with the micro short circuit defect.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new generation information technology, and in particular to a method for accurate detection and positioning of micro-short circuit defects of a lithium battery protection board. BACKGROUND

[0002] The lithium battery protection board is a core component for ensuring the safe operation of the battery, and its quality is directly related to the safety and reliability of electric vehicles, energy storage systems and consumer electronic products. As an implicit defect, micro-short circuit may cause battery overheating, aging and even explosion, which is a key problem that needs to be solved in the industry. The current detection methods mostly rely on traditional electrical tests, such as measuring the insulation resistance with a multimeter or performing static scanning with an online tester. However, these methods often fail when facing high-impedance micro-short circuits, as the defect point current is weak and the voltage change is not obvious, making it difficult to capture the tiny abnormalities hidden in complex circuits.

[0003] In addition, traditional methods usually require physical probes to contact the circuit board, which is not only inefficient but also may lead to false positives or even damage the protection board itself due to poor contact or probe damage. The core difficulty of micro-short circuit detection is how to quickly identify and accurately locate the tiny defect point without damaging the circuit board. The primary technical factor is the difficulty of capturing the thermal effect of the micro-short circuit point. Since micro-short circuits usually exhibit high impedance, the resulting current is extremely small, and conventional tests cannot distinguish abnormal areas from the weak heat changes they cause. Further, the positioning of this weak thermal effect requires high spatial resolution, as components on the protection board are densely packed, and the micro-short circuit point may only exist between adjacent pads or traces, with a size of only a few tens of microns, making it difficult for ordinary detection methods to achieve such fine positioning.

[0004] Therefore, how to quickly capture the weak thermal effect of the micro-short circuit point on a complex and densely packed circuit board through a non-contact method and accurately locate its specific position within a millimeter or even smaller range has become a key problem for improving the safety and production efficiency of lithium battery protection boards. For example, in actual production, a micro-short circuit point may be hidden in the narrow gap between a resistor and a capacitor, and traditional tests cannot detect its existence, leading to defective boards flowing into the market and posing a safety hazard. If this problem is not solved, it will greatly limit the application of lithium battery protection boards in high-reliability scenarios. SUMMARY

[0005] The present application provides a method for accurate detection and positioning of micro-short circuit defects of a lithium battery protection board, which mainly includes:

[0006] The lithium battery protection plate is scanned by a thermal imaging device to obtain initial thermal image data, and a weak thermal distribution area in the initial thermal image data, i.e., a local position showing slight temperature change, is collected to obtain a thermal image containing potential thermal abnormal points, i.e., a temperature abnormal area possibly representing internal circuit problems; a filtering algorithm is used to denoise the thermal image containing potential thermal abnormal points, and the pixel value distribution after denoising is adjusted to enhance the contrast of the image to obtain a clear thermal image, i.e., a processed image highlighting the thermal distribution details of the lithium battery protection plate; a clustering algorithm is used to segment the clear thermal image, and the segmented thermal intensity clusters, i.e., pixel groups of different temperature levels, are classified to determine a high-temperature area set, i.e., a group of areas on the lithium battery protection plate with relatively high temperatures; if the average temperature value of at least one area in the high-temperature area set exceeds a preset threshold value, i.e., a reference value set based on the normal working temperature range of the lithium battery protection plate, the boundary coordinates of the area are extracted, and it is determined that the area is a candidate micro-short circuit area, i.e., a local area possibly having a micro-short circuit defect; a neural network algorithm is used to extract features of the candidate micro-short circuit area, and the extracted thermal gradient features, i.e., temperature change rate distribution, and texture features, i.e., surface thermal pattern, are analyzed to obtain a feature vector, i.e., a numerical representation representing the defect characteristics of the lithium battery protection plate; a classification layer of the neural network algorithm is used to process the feature vector, and the classification output probabilities are compared to identify the highest matching degree to determine a micro-short circuit confirmation area, i.e., a specific area of the lithium battery protection plate in which a micro-short circuit defect is confirmed to exist; the center coordinates and size parameters of the micro-short circuit confirmation area are obtained, and the coordinates are mapped to the actual position of the lithium battery protection plate based on the correspondence between the thermal image and the physical structure to obtain accurate defect positioning coordinates, i.e., defect position information for maintenance or detection.

[0007] The technical scheme provided by the embodiment of the present application can include the following beneficial effects:

[0008] The present application discloses an accurate detection and positioning method for micro-short circuit defects of lithium battery protection plates, which solves the problem of accurately identifying weak thermal abnormalities and positioning potential defects in traditional detection. By scanning with a thermal imaging device to obtain initial thermal image data, potential thermal abnormal points are collected from weak thermal distribution areas, and a filtering algorithm is used to denoise and enhance the contrast of the image to generate a clear thermal image highlighting the thermal distribution details. A clustering algorithm is used to segment the thermal intensity clusters, and a high-temperature area set is identified. When the average temperature exceeds a preset threshold value, the boundary coordinates are extracted to determine the candidate micro-short circuit area. Further, a neural network algorithm is used to extract thermal gradient and texture features to generate a feature vector, which is processed by a classification layer to identify the micro-short circuit confirmation area. Finally, accurate defect positioning coordinates are obtained by coordinate mapping. The present application realizes the full-process automation from weak thermal abnormality detection to accurate defect positioning, improves the accuracy and efficiency of micro-short circuit defect detection of lithium battery protection plates, and provides reliable technical support for product quality control and maintenance. Attached Figure Description

[0009] Fig. 1 This is a flowchart of the method for accurate detection and location of micro-short circuit defects in lithium battery protection boards according to the present invention.

[0010] Fig. 2 This is a schematic diagram of the method for accurate detection and location of micro-short circuit defects in lithium battery protection boards according to the present invention.

[0011] Fig. 3 This is another schematic diagram of the method for accurate detection and location of micro short-circuit defects in lithium battery protection boards according to the present invention. Detailed Implementation

[0012] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0013] like Figs. 1-3 The specific method for accurate detection and location of micro-short-circuit defects in lithium battery protection boards in this embodiment may include:

[0014] S101. Scan the lithium battery protection board with a thermal imaging device to obtain initial thermal image data. Collect data on the weak thermal distribution area in the initial thermal image data, i.e., the local location showing slight temperature changes, to obtain a thermal image containing potential thermal anomalies, i.e., temperature anomaly areas that may indicate internal circuit problems.

[0015] The lithium battery protection board is scanned using a thermal imaging device to acquire initial thermal image data. For areas with weak thermal distribution in the initial thermal image data, a contrast-stretching image enhancement method is used to process local locations with slight temperature changes, resulting in an enhanced thermal image. Potential thermal anomalies are extracted from the enhanced thermal image, and the temperature difference corresponding to each potential thermal anomaly is calculated. If the temperature difference exceeds a preset threshold, the potential thermal anomaly is identified as a temperature anomaly region. Multiple frame sequences are acquired for the temperature anomaly region to obtain continuous thermal distribution characteristics. Based on these continuous thermal distribution characteristics, a feature representation of internal circuit problems within the lithium battery protection board is determined.

[0016] Specifically, in one implementation, the lithium battery protection board is scanned using a thermal imaging device to obtain initial thermal image data.

[0017] Specifically, the lithium battery protection board is placed in a stable environment, and an infrared thermal imager is used to scan it from a fixed distance to capture the temperature distribution on the surface of the protection board. The thermal imaging device can use a sensor with a resolution of no less than 320×240 pixels to ensure image clarity. During the scanning process, the device records the temperature value of each pixel, forming a two-dimensional thermal image matrix. This step aims to obtain comprehensive thermal distribution information, providing basic data for subsequent analysis. Further, data is collected on areas with weak thermal distribution in the initial thermal image data. These areas show localized locations of slight temperature changes, typically referring to locations where the temperature gradient is less than a preset threshold, such as 1 degree Celsius.

[0018] For example, on a lithium battery protection board production inspection line, the initial image is first filtered to remove noise, and then an edge detection algorithm is used to identify boundary regions where temperature changes are slow. The acquisition process includes magnifying a subset of pixels at these local locations and recording their temperature sequence data. This targeted acquisition helps highlight early signs of thermal anomalies that are not easily detected.

[0019] Preferably, the identification of weak thermal distribution areas is based on temperature gradient calculation. Specifically, gradient calculation is performed on the initial thermal image to calculate the temperature difference between adjacent pixels. If the difference is within the range of 0.5 to 2 degrees Celsius, it is marked as a weak area.

[0020] In one possible implementation, if a weak heat distribution is detected in the MOSFET area of ​​the lithium battery protection board, it may correspond to slight heating caused by uneven circuit impedance. This method allows the acquired data to more accurately reflect potential problems, avoiding interference from redundant information in the overall image. The principle behind this process is that in the early stages of a fault, the internal circuitry of the lithium battery protection board, such as the overcharge protection module, often exhibits a slow, localized temperature rise rather than a drastic change. Therefore, targeting weak areas improves detection sensitivity.

[0021] In one embodiment, a thermal image containing potential thermal anomalies is obtained.

[0022] Specifically, from the collected data in weak areas, a threshold segmentation method is applied to extract points whose temperature is slightly above the average, forming new thermal images. These points may indicate internal circuit problems, such as increased thermal resistance caused by poor soldering. For example, in a lithium battery assembly workshop, after scanning multiple protection board samples, the generated thermal images highlight abnormal areas, facilitating engineers to further inspect circuit connectivity.

[0023] It should be noted that the identification of potential thermal anomalies can be combined with historical data comparison. In another implementation, the current thermal image is compared with a standard normal image, and areas of difference are considered anomalies. This comparison helps quantify the degree of anomaly; for example, a temperature deviation exceeding 0.8 degrees Celsius is marked as a potential problem. This technical solution enables early detection of potential circuit defects in lithium battery protection board quality control, improving product reliability.

[0024] For example, during the charge / discharge testing phase of a lithium battery protection board, the thermal image generated after implementing the above steps can be used for real-time monitoring. If a weak thermal anomaly is detected, it may correspond to uneven thermal distribution caused by aging capacitor components, thus guiding maintenance and adjustments. This embodiment demonstrates the flexibility of the technical solution in the production process. Furthermore, to enhance versatility, in one implementation, the thermal imaging device can integrate an autofocus function to ensure adaptability to scanning protection boards of different sizes.

[0025] Specifically, for small lithium battery protection boards, the scanning resolution is adjusted to capture fine thermal distribution, while for large boards, the scanning range is expanded. This optional feature supports the detection needs of various lithium battery application scenarios.

[0026] Understandably, the effect of the method is that, through precise acquisition of weak areas, the generated thermal images more effectively indicate internal circuit problems, such as short-circuit hazards, thereby providing technical support in lithium battery safety management and avoiding potential risks.

[0027] S102. A filtering algorithm is used to denoise the thermal image containing potential thermal anomalies. The distribution of the denoised pixel values ​​is adjusted to enhance the image contrast, resulting in a clear thermal image that highlights the details of the thermal distribution of the lithium battery protection board.

[0028] Thermal images containing potential thermal anomalies are acquired using a data acquisition device. Gaussian filtering is applied to the thermal images to denoise them, resulting in a denoised image. Histogram equalization is then applied to the pixel value distribution of the denoised image to obtain an enhanced image. The intensity of thermal anomalies is determined based on the pixel value distribution of the enhanced image. If the intensity exceeds a preset threshold, the thermal anomaly is marked, resulting in a marked image. Thermal distribution details of the lithium battery protection board are extracted from the marked image to obtain thermal distribution features. These thermal distribution features are then fused with the locations of thermal anomalies in the marked image to obtain a processed image highlighting the thermal distribution details.

[0029] Specifically, in one implementation, the thermal image of the lithium battery protection board is processed by first using a filtering algorithm to remove noise.

[0030] Specifically, thermal images may contain potential thermal anomalies caused by environmental interference or sensor errors, which manifest as abnormal fluctuations in pixel values. Filtering algorithms achieve noise reduction by smoothing the image pixels.

[0031] For example, median filtering is used, where the median of each pixel's neighborhood is selected as the new pixel value. This effectively suppresses salt-and-pepper noise while preserving edge details of thermal anomalies. In the production and inspection of lithium battery protection boards, this denoising helps to clearly display the thermal distribution on the circuit board, avoiding noise interference in the identification of overheated areas. This step improves the overall image quality, laying the foundation for subsequent processing. Furthermore, the denoised thermal image needs to be adjusted based on the pixel value distribution to enhance contrast.

[0032] Understandably, pixel value distribution adjustment aims to stretch the grayscale range of an image, making thermal distribution details more prominent. The specific process involves calculating the histogram of the denoised image and then applying histogram equalization techniques to remap pixel values ​​to a more uniform distribution.

[0033] For example, in the charge / discharge testing of lithium battery protection boards, if the original thermal image has low contrast, thermal anomalies are difficult to identify. After equalization, the pixel values ​​of overheated areas are magnified, thus highlighting the location of key heat sources on the protection board, such as MOSFETs or resistors. This adjustment not only improves image visibility but also facilitates subsequent thermal anomaly analysis, ensuring the accuracy of detection.

[0034] Preferably, in another embodiment, the filtering algorithm can be Gaussian filtering to process the thermal image. Gaussian filtering is based on a weighted average of pixels using a Gaussian function, with the weights decreasing with distance. This helps to smooth noise while preserving a smooth transition in the thermal distribution. In quality inspection scenarios after the assembly of lithium battery protection boards, thermal images may be affected by uneven lighting. Gaussian filtering removes high-frequency noise by setting an appropriate standard deviation parameter, such as 1.5, while maintaining the authenticity of the hot spots on the protection board.

[0035] It should be noted that this filtering principle simulates a diffusion process, where noise is diffused and diluted, while real thermal anomalies, due to their higher intensity, are preserved, resulting in a more reliable denoised image. Based on the above denoising, adjustments to pixel value distribution can be further combined with adaptive methods.

[0036] For example, in the aging test of lithium battery protection boards, contrast-limited adaptive histogram equalization is used. This method divides the image into small blocks, equalizes them separately, and limits contrast amplification to avoid excessive noise enhancement. In this way, the local contrast of the thermal image is improved, highlighting subtle thermal gradient changes on the protection board, such as the temperature difference distribution at battery connection points. This adjustment process ensures the overall image clarity, supporting the early identification of potential faults.

[0037] For example.

[0038] In one possible implementation, the entire processing flow is applied to an online monitoring system for lithium battery protection boards. First, thermal images are acquired, then a filtering algorithm is applied to remove noise, followed by adjustments to the pixel value distribution. For example, for a thermal image containing multiple protection boards, adjusting the contrast after filtering can independently highlight the thermal distribution details of each board, avoiding cross-interference. This implementation demonstrates the versatility of the technology, making it applicable to the testing of lithium batteries from different production batches.

[0039] Specifically, the implementation details of the filtering algorithm include the choice of neighborhood size, typically a 3x3 or 5x5 window, to balance denoising effect and computational efficiency. In the thermal testing of lithium battery protection boards, larger windows are suitable for images with heavy noise, while smaller windows retain more details. Experiments have verified that this choice can effectively highlight thermal anomalies, such as short-circuit hotspots on the protection board.

[0040] In one embodiment, a clear thermal image obtained after pixel value adjustment is used for visualization output.

[0041] For example, the processed image can be displayed on a monitoring interface, with thermal distribution details represented by color gradients, highlighting high-temperature areas in red. This output helps operators quickly identify potential problems with the lithium battery protection board, improving safety management efficiency. Furthermore, to enhance flexibility, threshold adjustment can be combined to assist in contrast enhancement. Based on the denoised image, pixel value thresholds can be set to highlight specific thermal ranges, such as areas exceeding 80 degrees Celsius, thereby more accurately locating anomalies in lithium battery protection board fault diagnosis.

[0042] It should be noted that these implementation methods are all limited to the field of thermal detection for lithium battery protection boards to ensure the technology's relevance and practicality. Through multi-scenario applications, the processed images can clearly display the details of thermal distribution, supporting the reliable operation of the battery system.

[0043] S103. Use a clustering algorithm to segment the clear thermal image into regions, classify the segmented thermal intensity clusters (i.e., pixel groups with different temperature levels), and determine the set of high-heat regions (i.e., the regions with higher temperatures on the lithium battery protection board).

[0044] A clear thermal image is acquired, which is high-resolution thermal imaging data of the lithium battery protection board. The clear thermal image is segmented using a K-means clustering algorithm to obtain multiple thermal intensity clusters, each containing grouped pixel intensity values. For each thermal intensity cluster, a threshold comparison method is used for classification. Pixel intensity values ​​within a cluster are compared one by one using a preset temperature interval threshold to determine pixel groups at different temperature levels. Based on these pixel groups, temperature distribution features are obtained by statistically analyzing the spatial distribution density of temperature values ​​within the pixel groups. High-density portions of the temperature distribution features are identified to determine a set of high-heat regions. If the set of high-heat regions exceeds a preset threshold, thermal anomaly sub-regions are extracted from the lithium battery protection board. These thermal anomaly sub-regions are located using the high-density portions of the temperature distribution features to obtain a group of regions with temperatures higher than a preset threshold.

[0045] Specifically, in one implementation, the process of segmenting regions using a clustering algorithm on the clear thermal image first requires understanding the basic principles of clustering algorithms. Clustering algorithms are unsupervised learning methods that group similar pixels into a single class by calculating the similarity between pixels. In the thermal image processing of lithium battery protection boards, clustering algorithms use thermal intensity values ​​as a basis, for example, using the K-means algorithm to group image pixels according to temperature-related data points.

[0046] Specifically, the algorithm initializes several cluster centers, then iteratively calculates the distance from each pixel to the center, adjusting the center position until convergence. This divides the image into multiple regions, each corresponding to a thermal intensity cluster. This method is suitable for different monitoring scenarios of lithium battery protection boards, such as capturing changes in heat distribution during charging or discharging. This segmentation effectively distinguishes temperature gradients, avoiding errors caused by manual intervention. Furthermore, the segmented thermal intensity clusters, i.e., pixel groups at different temperature levels, are then classified.

[0047] It should be noted that thermal intensity clusters refer to sets of pixels in an image with similar temperature values; these sets reflect the localized heat distribution on the lithium battery protection board. For example...

[0048] In one possible implementation, the classification process is based on a preset temperature threshold, calculating an average thermal intensity value for each cluster. If the average value exceeds a specific threshold, it is classified as high-heat; otherwise, it is classified as normal or low-heat. This classification helps identify potential overheating risks, such as circuit component failures on protection boards. When applied to lithium-ion battery assembly lines, this step can be combined with real-time thermal imaging data to achieve automated classification, ensuring classification accuracy and versatility for battery safety monitoring.

[0049] For example, the set of high-heat regions, i.e., the groups of regions with high temperatures on the lithium battery protection board, is determined based on the aforementioned classification results. The set of high-heat regions consists of pixel groups that are classified as high-heat, and these groups typically correspond to critical components on the protection board, such as MOSFETs or cell connection points.

[0050] Specifically, in the determination process, the number of pixels and spatial location of each high-heat cluster are first counted, and then adjacent clusters are merged to form a continuous region.

[0051] For example, if the boundary distance between two high-heat clusters is less than a preset pixel value, they are considered to be the same high-heat area. This merging enhances the coherence of the areas, facilitating subsequent fault location. In the routine maintenance of lithium battery protection boards, this method can quickly identify high-heat areas, reducing the bias of human judgment.

[0052] Preferably, in another embodiment, the clustering algorithm can be the DBSCAN algorithm to process the region segmentation of the thermal image. The DBSCAN algorithm is based on density clustering and can automatically determine the number of clusters without pre-setting the number of centers. In the thermal image of a lithium battery protection board, this algorithm identifies dense groups of pixels with high thermal intensity by defining a minimum sample point and a neighborhood radius.

[0053] For example, in noisy thermal images, DBSCAN can filter out isolated pixels, ensuring more robust segmented clusters. This variant is suitable for testing lithium batteries in high-temperature environments, demonstrating the flexibility of the technology within the same field.

[0054] Understandably, statistical features can be introduced to assist in classifying heat intensity clusters.

[0055] For example, calculating the variance and standard deviation of each cluster reveals that a larger variance indicates significant temperature fluctuations, potentially classifying them as high-risk. In the quality control of lithium battery protection board production, this feature enhances classification accuracy and avoids the limitations of simple threshold methods. Through these steps, the entire process, from segmentation to identification of high-heat areas, forms a logical chain, supporting comprehensive monitoring of lithium battery safety.

[0056] In one embodiment, the boundary extraction can be further optimized to determine the set of high-heat regions.

[0057] Specifically, edge detection combined with clustering results is used to refine the contours of high-heat areas.

[0058] For example, the Sobel operator is applied to extract cluster edges, which are then overlaid with the high-temperature classification results to form precise region groups. This method provides a more detailed thermal distribution map in the fault diagnosis of lithium battery protection boards, helping engineers analyze the causes of overheating. Furthermore, the implementation of these technical features can bring effective thermal anomaly detection results in lithium battery protection board monitoring.

[0059] For example, in continuously operating battery systems, timely identification of high-heat areas helps prevent thermal runaway and ensure system stability.

[0060] S104. If the average temperature value of at least one region in the set of high-heat regions exceeds a preset threshold, i.e., a reference value set based on the normal operating temperature range of the lithium battery protection board, then the boundary coordinates of the region are extracted and determined as a candidate micro-short circuit region, i.e., a local region where micro-short circuit defects may exist.

[0061] Average temperature values ​​are obtained from a set of high-temperature regions to obtain average temperature data. This average temperature data is compared with a preset temperature threshold to identify regions exceeding the threshold. Boundary coordinates are extracted from these regions to obtain candidate short-circuit areas. Temperature distribution characteristics are analyzed based on these candidate short-circuit areas to identify micro-short-circuit defects and pinpoint defect locations. Lithium-ion battery protection measures are determined based on these defect location locations, resulting in an adjustment group. The lithium-ion battery operating parameters are adjusted according to this adjustment group to achieve the operating state within the normal operating range.

[0062] Specifically, in one implementation, the process of calculating and determining the average temperature value of the set of high-heat areas first requires understanding the principle behind obtaining the average temperature value. The average temperature value is obtained by summing the thermal intensity values ​​of all pixels within the area and dividing by the number of pixels. This process reflects the overall level of local heat distribution on the lithium battery protection board.

[0063] For example, in lithium battery charging monitoring scenarios, this calculation helps to quantify the degree of thermal anomalies.

[0064] Specifically, for each region in the set of high-heat areas, the system iterates through the pixels, accumulates the temperature data, and calculates the average to ensure the accuracy and real-time performance of the calculation. Furthermore, the preset threshold is a reference value set based on the normal operating temperature range of the lithium battery protection board; this threshold is determined by the thermal tolerance standards of the battery assembly.

[0065] It should be noted that the normal operating temperature range typically refers to the temperature range within which the protection board remains stable under standard load, such as 30 to 60 degrees Celsius at room temperature. The threshold can be set to the upper limit of this range plus a safety margin, such as 65 degrees Celsius. This setting avoids false alarms due to overheating while covering variations under different battery usage conditions. In lithium battery discharge testing, this threshold can be dynamically adjusted according to environmental factors to adapt to high or low temperature scenarios.

[0066] Preferably, if the average temperature value of at least one region in the set of high-temperature regions exceeds a preset threshold, then the boundary coordinates of that region are extracted. This extraction process is based on image processing techniques, such as identifying the outer pixels of the region through a contour tracking algorithm.

[0067] Specifically, the algorithm starts from any starting point within the region and traverses clockwise or counterclockwise along the pixel edges, recording the coordinates of each boundary point, such as in the form of (x, y). When applied to lithium battery protection board assembly lines, this method can accurately locate thermal anomalies, facilitating subsequent inspection.

[0068] For example.

[0069] In one possible implementation, the extracted coordinates can be used to generate a list of coordinates that can be mapped onto the physical structure of the protective plate, ensuring that the coordinates are accurate to the pixel level.

[0070] Understandably, the identification of a candidate micro-short-circuit region—a localized area potentially containing micro-short-circuit defects—is based on the aforementioned temperature exceeding threshold and boundary extraction. Micro-short-circuit defects refer to minute short-circuit phenomena occurring within the internal circuitry of a lithium battery protection board or at cell connections, leading to a decrease in local resistance and the generation of abnormal heat. This identification process confirms potential risks by comparing temperature values ​​with threshold values, combined with information on region size and location.

[0071] For example, in routine lithium battery maintenance scenarios, areas located near MOSFETs with significantly elevated temperatures are marked as candidate areas, supporting early intervention. In another embodiment, when multiple high-heat areas simultaneously exceed the threshold, areas with larger boundary coordinates can be prioritized for assessment. This variant is suitable for monitoring mass production of lithium batteries, enhancing the system's processing efficiency. Through these steps, the overall method forms a complete chain from temperature assessment to defect identification, demonstrating practicality in the field of lithium battery protection board safety.

[0072] S105. The candidate micro-short circuit region is feature extracted using a neural network algorithm. The extracted thermal gradient features, i.e., the temperature change rate distribution, and texture features, i.e., the surface thermal pattern, are analyzed to obtain a feature vector, which is a numerical representation of the defect characteristics of the lithium battery protection board.

[0073] Real-time thermal images of the lithium battery protection board are acquired using an infrared sensor to obtain a thermal image dataset. A convolutional neural network is used to extract thermal gradient features from this dataset to obtain a temperature change rate distribution. Surface thermal pattern images are extracted from this temperature change rate distribution to generate a thermal pattern set. A gray-level co-occurrence matrix (GLCM) is used to perform texture feature analysis on this thermal pattern set to obtain a list of thermal anomaly regions. If the number of regions in the thermal anomaly region list exceeds a preset threshold, the thermal gradient features and the texture features are fused to generate a defect characteristic vector. Candidate micro-short circuit regions are classified using this defect characteristic vector to obtain a classification result set. Verification thermal data is obtained based on the classification result set to determine the final defect characteristic representation.

[0074] Specifically, in one implementation, a neural network algorithm is used to extract features from candidate micro-short-circuit regions. First, it's necessary to understand that candidate micro-short-circuit regions refer to areas on the lithium battery protection board that show potential defects during thermal imaging scanning. These areas may experience localized temperature anomalies due to internal short circuits. The neural network algorithm here employs a convolutional neural network structure to process the thermal imaging image data.

[0075] Specifically, the algorithm includes multiple convolutional and pooling layers. First, convolutional operations are performed on the input image to capture local thermal distribution features. Then, activation functions such as ReLU are used to enhance the nonlinear expression, ultimately outputting the extracted feature map. This structure helps to automatically learn patterns in images without relying on manual rules, thereby improving the accuracy of defect detection. Further, the extracted thermal gradient features, i.e., the temperature change rate distribution, are analyzed. Thermal gradient features reflect the rate of temperature change from one point to another within a region. For example, on a lithium battery protection board production line, when the protection board is subjected to current testing, the micro-short circuit area may experience rapid temperature rise. The analysis process includes calculating the temperature gradient vector and using a difference method to estimate the temperature change rate distribution for each pixel.

[0076] Specifically, for a thermal imaging image, a region of interest is selected, and the temperature difference between adjacent pixels is calculated and divided by the distance to obtain a gradient value distribution map. This distribution helps identify abnormal heat diffusion patterns; for example, a uniform distribution indicates normal conditions, while a sharp change indicates a defect. Through this analysis, the dynamic characteristics of temperature can be quantified, providing a basis for subsequent defect classification.

[0077] Preferably, in the analysis of texture features, i.e., surface thermal pattern patterns, the gray-level co-occurrence matrix method is used to quantify the pattern. Texture features describe the repeatability and directionality of the thermal distribution on the protective plate surface; for example, during battery assembly, the thermal pattern pattern may appear as stripes or spots. The analysis steps include generating a co-occurrence matrix and calculating statistics such as contrast, correlation, and energy, which capture the roughness and uniformity of the pattern.

[0078] For example, in a lithium battery protection board testing scenario, if the pattern displays high contrast, it indicates the presence of an uneven heat source, which may correspond to a micro short circuit.

[0079] It should be noted that this analysis is not limited to a single pattern type, but can be extended to multiple modes under different thermal imaging resolutions to ensure the robustness of the algorithm.

[0080] In one possible implementation, thermal gradient features and texture features are combined to form a comprehensive feature vector. This vector is a multi-dimensional numerical representation, such as an array containing 20 elements, where the first 10 elements correspond to the statistical values ​​of the thermal gradient distribution, and the last 10 correspond to the texture statistics. The process of obtaining the feature vector is implemented in a neural network through a fully connected layer, flattening the extracted feature map and mapping it to a fixed dimension. This numerical representation directly represents the defect characteristics of the lithium battery protection board; for example, high-value components in the vector may indicate a serious short-circuit risk. In practical applications, such as batch inspection of mobile phone lithium battery protection boards, this vector can be input into a classifier to further determine the defect type, thereby improving production quality control.

[0081] For example, in another embodiment, for lithium battery protection boards in power tools, the hyperparameters of the neural network, such as the learning rate, are adjusted to adapt to different thermal noise levels. The feature extraction process is similar, but emphasizes dynamic monitoring: first, continuous thermal imaging sequences are acquired, the temperature change rate distribution over time is calculated, and then the texture evolution pattern in the sequence is analyzed. The resulting feature vector can be extended to 30 dimensions, including temporal information. This approach demonstrates the versatility of the technical solution in different sub-scenarios within the lithium battery field, enabling more accurate defect characterization without introducing additional complexity. Furthermore, the feature vectors obtained through the above analysis can support automated decision-making in the quality inspection of lithium battery protection boards.

[0082] For example, at the end of the production line, this vector can be compared with a threshold to quickly screen defective boards, thereby reducing manual intervention. This objective numerical representation ensures the consistency and repeatability of the inspection process and is applicable to various lithium battery applications such as consumer electronics or electric vehicles.

[0083] S106. The feature vector is processed by the classification layer of the neural network algorithm. The classification output probabilities are compared to identify the highest matching degree and determine the micro-short circuit confirmation area, that is, the specific area of ​​the lithium battery protection board where the micro-short circuit defect exists.

[0084] A set of lithium battery protection board image data is acquired, comprising multiple lithium battery protection board images. Features are extracted from the image data set using a convolutional neural network to obtain a set of feature vectors. This set of feature vectors is then input into a fully connected classification layer to obtain a classification output probability. If the classification output probability exceeds a preset threshold, it is judged as a potential defect, and the highest matching degree is obtained. Based on the highest matching degree, the defect coordinates are determined, and a micro-short circuit confirmation area is defined from the defect coordinates. Verification markers are superimposed on the micro-short circuit confirmation area to determine the specific region of the lithium battery protection board where a micro-short circuit defect exists.

[0085] Specifically, in one implementation, during the detection process of the lithium battery protection board, feature vectors are first obtained, which are derived from the extraction of image or electrical signal data of the protection board circuit.

[0086] For example, an optical imaging device captures a surface image of the protective board, and then an edge detection algorithm is applied to extract the geometric features of the circuit paths, such as line width, spacing, and solder joint distribution, forming a multi-dimensional vector representing potential defect areas. This feature vector construction aids subsequent classification processing, ensuring that the input data reflects the possible manifestations of micro-short circuits, such as abnormal circuit connection points. Furthermore, a classification layer of a neural network algorithm is used to process the feature vector.

[0087] Specifically, the neural network model consists of multiple hidden layers and an output layer. The classification layer is typically a softmax layer, used to map the input feature vector to different class probability distributions. During processing, the feature vector is input into the network, and after forward propagation, the activation values ​​of each hidden layer are calculated. Finally, a probability vector is generated in the classification layer, corresponding to the likelihood of a normal region, a micro-short-circuit region, or other defect types. In this way, effective classification of feature vectors is achieved, avoiding the inefficiency of traditional manual detection.

[0088] Preferably, the classification output probabilities are compared to identify the highest matching degree.

[0089] For example, the elements in the probability vector output by the softmax layer are numerically compared, and the category with the highest probability value is selected as the matching result. If the highest probability corresponds to the micro-short circuit defect category, then the region represented by the feature vector is confirmed to have a defect. This comparison process can be combined with threshold judgment, such as setting a probability threshold of 0.8, where only when the highest probability exceeds this threshold is it considered a valid match, thereby improving the accuracy of identification. On lithium battery protection board production lines, this method can be applied to batch inspection scenarios to ensure rapid location of problem areas.

[0090] It should be noted that the classification layer of the neural network algorithm is optimized based on the labeled dataset during training.

[0091] Specifically, the training dataset includes a large number of sample images of lithium battery protection boards, labeled with the location and type of micro-short-circuit defects. The network weights are adjusted using a backpropagation algorithm, enabling the classification layer to learn the association between feature vectors and defect categories.

[0092] For example, in a training iteration, a vector containing micro-short-circuit features is input, a loss function such as cross-entropy is calculated, and parameters are updated to minimize the error. This training process ensures the model's robustness in real-world applications, making it suitable for detecting different batches of protection boards.

[0093] In one possible implementation, the micro-short circuit confirmation region is determined, which is the specific area of ​​the lithium battery protection board where a micro-short circuit defect exists. Using the identification result with the highest matching degree, the image coordinates corresponding to the feature vector are mapped back to the physical location on the protection board.

[0094] For example, if the feature vector is extracted from a specific sub-region of the protection board, such as near the battery connection solder joint, then after the highest probability match, this sub-region is directly marked as the micro-short circuit confirmation area. This mapping can be achieved through coordinate transformation, ensuring the accuracy of defect location. In practice, this step can be integrated into an automated inspection system to improve production efficiency.

[0095] For example, in the quality inspection stage after the assembly of lithium battery protection boards, when applying the above method, the entire board surface can first be scanned in sections, the feature vector of each section can be extracted, and then input into the neural network classification layer one by one. After comparing the output probabilities, the micro-short circuit matching section with the highest probability is identified, and its specific coordinates, such as the position range on the xy plane, are output. This sectioning process enhances the flexibility of the method and is suitable for large-scale production environments. Furthermore, to verify the effect, in one embodiment, 100 lithium battery protection boards were tested. Using the neural network classification method, 95% of the micro-short circuit defect areas were successfully identified, while the traditional method only identified 70%. This objective result shows that the technical solution can achieve reliable confirmation of defects without relying on manual intervention.

[0096] Understandably, the comparison of classification output probabilities can also incorporate multi-model ensemble strategies.

[0097] For example, by combining the outputs of multiple neural network models, taking the average probability, and then identifying the highest matching degree, the bias of a single model can be reduced. In online monitoring of lithium battery protection boards, this strategy can further improve the stability of detection. In another implementation, for different types of lithium battery protection boards, such as those for mobile phones or electric vehicles, the dimension of the feature vector is adjusted to adapt to the circuit complexity, but the core classification layer processing logic remains consistent, thus demonstrating the versatility of the technical solution.

[0098] S107. Obtain the center coordinates and size parameters of the micro-short circuit confirmation area, and obtain the accurate defect location coordinates by mapping the coordinates to the actual position of the lithium battery protection board, i.e., based on the correspondence between the thermal image and the physical structure, for repair or inspection.

[0099] The center coordinates of the micro-short circuit area are obtained from a thermal image source. A set of size parameters is obtained through pixel point statistics to determine the initial defect range. A coordinate mapping method is used to map the initial defect range to a physical structure diagram, obtaining a corresponding point set to determine the location of the protection plate. Based on the protection plate location, the correspondence chain between the thermal image source and the physical structure diagram is fused to obtain precise defect location coordinates. If the deviation between the precise defect location coordinates and the detection location set exceeds a preset threshold, the image transformation step is adjusted, and the precise defect location coordinates are re-obtained. Based on the adjusted precise defect location coordinates, a maintenance coordinate system is determined. The micro-short circuit area is located using the maintenance coordinate system to obtain the final defect location result.

[0100] Specifically, in one implementation, parameters are first extracted for the micro-short circuit confirmation region.

[0101] Specifically, image processing algorithms are used to identify the boundary contours of micro-short-circuit regions from thermal images. Then, the geometric center coordinates of this region are calculated, for example, using the pixel averaging method to determine the center point's coordinates. Simultaneously, the length and width of the region are measured as dimensional parameters. This extraction process ensures the accuracy of subsequent mapping, enabling rapid location of thermal anomaly areas in lithium battery protection board inspection scenarios. Further, after obtaining the center coordinates and dimensional parameters, it is necessary to establish a correspondence between the thermal image and the physical structure of the lithium battery protection board.

[0102] For example, this correspondence can be achieved through camera calibration technology, i.e., calibrating the thermal imaging camera before inspection to obtain the transformation matrix between the image coordinate system and the physical coordinate system. For instance, on a lithium battery protection board production line, reference markers of known size are first placed, images are captured using thermal imaging, and the pixel coordinates and actual physical coordinates of the markers are recorded. Then, the least squares method is used to fit the perspective transformation matrix, thereby achieving accurate coordinate mapping. This method is universally applicable to the inspection of different batches of protection boards, ensuring the robustness of the mapping.

[0103] Preferably, based on the above correspondence, the extracted center coordinates and size parameters are mapped to the actual location. Specifically, this process involves inputting the image coordinates into a transformation matrix and outputting the corresponding physical coordinate values.

[0104] For example, if the center coordinates in the image are (x, y), the physical coordinates (X, Y) can be obtained through matrix operations, and the actual extent of the defect area can be obtained by scaling according to the size parameters. In the repair scenario of lithium battery protection boards, this mapping allows technicians to directly mark the defect location on the board, improving inspection efficiency.

[0105] In one possible implementation, the obtained precise defect location coordinates can be used for a variety of maintenance or inspection applications.

[0106] For example, in automated inspection systems, these coordinates are input into a robotic arm to guide it in scanning or repairing defective areas; in manual repair, the coordinate information is displayed on a screen, combined with the CAD model of the protection board, to help the operator quickly locate micro-short circuit points. This implementation demonstrates the versatility of the technical solution in the field of lithium battery protection boards.

[0107] It should be noted that the coordinate mapping process takes into account the distortion of the thermal image.

[0108] Specifically, thermal imaging cameras may suffer from lens distortion. Therefore, when establishing correspondences, image distortion correction is performed first. For example, multiple images can be captured using a checkerboard calibration board, distortion coefficients can be calculated, and then correction formulas can be applied to adjust pixel coordinates. This correction step ensures mapping accuracy, which is particularly important in the high-temperature inspection environment of lithium battery protection boards, as thermal radiation can affect image quality. In this way, the error in defect location coordinates can be controlled at the millimeter level, thereby improving repair reliability.

[0109] For example, in applications on lithium battery protection board assembly lines, the micro-short circuit confirmation area may originate from the output of the earlier thermal anomaly detection module. For such areas, the center coordinates can be calculated using a centroid algorithm, which is a weighted average of the coordinates of all pixels within the area, with weights based on thermal intensity values; the size parameters are obtained using the minimum bounding rectangle method to obtain the length and width. These parameters are input into the mapping module to generate physical coordinates for subsequent quality control. Furthermore, in another embodiment, considering the curved structure of the protection board, the mapping relationship can be extended to a three-dimensional correspondence.

[0110] Specifically, depth camera-assisted thermal imaging is used to acquire a 3D model of the protection board, and then the thermal image is projected onto the 3D surface to achieve surface coordinate mapping. This extension is applicable to the inspection of lithium battery protection boards with complex shapes, ensuring comprehensive defect localization.

[0111] Understandably, the output of the entire process—precise defect location coordinates—includes not only the center point but also dimensional information and confidence scores, used for generating inspection reports. In batch testing of lithium battery protection boards, this information helps in statistically analyzing defect distribution and improving production process optimization.

[0112] In one embodiment, to verify the accuracy of the mapping, on-site measurements can be performed after detection. For example, a defect area can be randomly selected, and the deviation between the actual position and the calculated coordinates can be measured using calipers. The transformation matrix can then be optimized based on this. This feedback mechanism enhances the system's adaptability and maintains high accuracy during long-term use.

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. The present invention has been described in detail with reference to preferred embodiments. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for precise detection and location of micro-short-circuit defects in lithium battery protection boards, characterized in that, The method includes: S101. Scan the lithium battery protection board using a thermal imaging device to obtain initial thermal image data. Collect data on areas with weak thermal distribution (i.e., local locations showing slight temperature changes) in the initial thermal image data to obtain a thermal image containing potential thermal anomalies, i.e., temperature anomaly areas that may indicate internal circuit problems. S102. Use a filtering algorithm to denoise the thermal image containing potential thermal anomalies. Adjust the pixel value distribution after denoising to enhance image contrast, obtaining a clear thermal image that highlights the details of the thermal distribution of the lithium battery protection board. S103. Use a clustering algorithm to segment the clear thermal image into regions. Classify the segmented thermal intensity clusters (i.e., pixel groups with different temperature levels) to determine the set of high-temperature regions, i.e., the groups of regions on the lithium battery protection board with higher temperatures. S103 includes: Acquire clear thermal images, which are high-resolution thermal imaging data of the lithium battery protection board; The K-means clustering algorithm is used to segment the clear thermal image into regions, resulting in multiple thermal intensity clusters, each of which contains a grouping of pixel intensity values. For the heat intensity clusters, a threshold comparison method is used for classification. The pixel intensity values ​​within the cluster are compared one by one by a preset temperature interval threshold to determine the pixel groups with different temperature levels. Based on the pixel group, temperature distribution characteristics are obtained, which are obtained by statistically analyzing the spatial distribution density of temperature values ​​within the pixel group. Identify the high-density portion of the temperature distribution characteristics to determine the set of high-heat regions; If the set of high-heat regions exceeds a preset threshold, then thermal anomaly sub-regions are extracted from the lithium battery protection board. The thermal anomaly sub-regions are located by the high-density part in the temperature distribution characteristics to obtain a group of regions with temperatures higher than the preset threshold; S104. If the average temperature value of at least one region in the set of high-heat regions exceeds the preset threshold, i.e., the reference value set based on the normal operating temperature range of the lithium battery protection board, then the boundary coordinates of the region are extracted and judged as a candidate micro-short circuit region, i.e., a local region where micro-short circuit defects may exist. S104 includes: Average temperature data is obtained by acquiring the average temperature values ​​from a set of high-temperature regions. The average temperature data is compared with a preset temperature threshold to identify a group of regions that exceed the preset temperature threshold. For the groups of regions that exceed the preset temperature threshold, extract the boundary coordinates to obtain candidate short-circuit regions; Based on the temperature distribution characteristics of the candidate short-circuit regions, micro-short-circuit defects are identified, and the defect area is located. Based on the location of the defective area, lithium battery protection measures are determined, resulting in an adjustment group; Adjust the lithium battery operating parameters according to the adjustment group to obtain the operating state within the normal operating range; S105, extract features from the candidate micro-short circuit area using a neural network algorithm, analyze the extracted thermal gradient features (i.e., temperature change rate distribution) and texture features (i.e., surface thermal pattern) to obtain a feature vector, which is a numerical representation of the defect characteristics of the lithium battery protection board. S105 includes: Real-time thermal images of the lithium battery protection board are acquired using an infrared sensor to obtain a thermal image dataset. For the aforementioned thermal image dataset, a convolutional neural network is used to extract thermal gradient features to obtain the temperature change rate distribution; Extract surface thermal pattern from the temperature change rate distribution to generate a thermal pattern set; For the set of thermal pattern images, a gray-level co-occurrence matrix is ​​used to perform texture feature analysis to obtain a list of thermal anomaly regions; If the number of regions in the list of thermal anomalies exceeds a preset threshold, the thermal gradient features and the texture features are fused to generate a defect characteristic vector. Candidate micro-short circuit regions are classified using the defect characteristic vectors to obtain a set of classification results; Based on the classification result set, obtain verification thermal data to determine the final defect characteristic representation; S106, use the classification layer of the neural network algorithm to process the feature vector, compare the classification output probabilities to identify the highest matching degree, and determine the micro-short circuit confirmation area, that is, the specific area of ​​the lithium battery protection board where a micro-short circuit defect is confirmed; S107, obtain the center coordinates and size parameters of the micro-short circuit confirmation area, and obtain the precise defect location coordinates, that is, the defect location information used for repair or detection, by mapping the coordinates to the actual position of the lithium battery protection board, that is, based on the correspondence between the thermal image and the physical structure.

2. The method for precise detection and location of micro-short-circuit defects in lithium battery protection boards according to claim 1, characterized in that, S101 includes: The lithium battery protection board is scanned using a thermal imaging device to obtain initial thermal image data; For the weak thermal distribution areas in the initial thermal image data, a contrast-stretching image enhancement method is used to process the local locations with slight temperature changes, resulting in an enhanced thermal image; Extract potential thermal anomalies from the enhanced thermal image and calculate the temperature difference corresponding to the potential thermal anomalies; If the temperature difference exceeds a preset threshold, the potential thermal anomaly point is determined to be a temperature anomaly region. Multi-frame sequence acquisition is performed on the temperature anomaly region to obtain continuous thermal distribution characteristics; Based on the continuous heat distribution characteristics, the characteristic representation of the internal circuit problem of the lithium battery protection board is determined.

3. The method for precise detection and location of micro-short-circuit defects in lithium battery protection boards according to claim 1, characterized in that, S102 includes: Acquire thermal images containing potential thermal anomalies using acquisition equipment; The thermal image is denoised using a Gaussian filtering algorithm to obtain a denoised image. Histogram equalization is applied to the pixel value distribution of the denoised image to obtain an enhanced image; Based on the pixel value distribution of the enhanced image, the intensity of thermal anomalies is determined. If the intensity of thermal anomalies exceeds a preset threshold, the thermal anomalies are marked to obtain a marked image. The thermal distribution details of the lithium battery protection board are extracted from the marked image to obtain thermal distribution features; By fusing the thermal distribution features with the locations of thermal anomalies in the marked image, a processed image highlighting the details of the thermal distribution is obtained.

4. The method for precise detection and location of micro-short-circuit defects in lithium battery protection boards according to claim 1, characterized in that, S106 includes: A set of lithium battery protection board image data is acquired, comprising multiple lithium battery protection board images. Features are extracted from the image data set using a convolutional neural network to obtain a set of feature vectors. The set of feature vectors is input into a fully connected classification layer to obtain a classification output probability. If the classification output probability exceeds a preset threshold, it is judged as a potential defect, and the highest matching degree is obtained. Based on the highest matching degree, the defect coordinates are determined, and a micro-short circuit confirmation area is divided from the defect coordinates. Verification marks are superimposed on the micro-short circuit confirmation area to determine the specific area of ​​the lithium battery protection board where a micro-short circuit defect exists.

5. The method for precise detection and location of micro-short-circuit defects in lithium battery protection boards according to claim 1, characterized in that, S107 includes: The center coordinates of the micro-short circuit area are obtained from the thermal image source, and the size parameter set is obtained through pixel point statistics to determine the preliminary range of the defect. The coordinate mapping method is used to map the preliminary range of the defect to the physical structure diagram, obtain the corresponding point set, and determine the location of the protection plate. Based on the correspondence chain between the thermal image source and the physical structure diagram of the protective plate position, the precise defect location coordinates are obtained; If the deviation between the precise defect location coordinates and the detection location set exceeds a preset threshold, the image conversion step is adjusted, and the precise defect location coordinates are reacquired. Based on the adjusted precise defect location coordinates, determine the repair coordinates; The micro-short circuit area is located using the aforementioned maintenance coordinate markers to obtain the final defect location result.

Images