A cell sealing welding abnormality monitoring method and system based on concentric circle feature extraction and dynamic clustering

Abstract

The application discloses a kind of based on concentric circle feature extraction and dynamic clustering cell seal weld abnormality monitoring method and system, it is related to lithium ion cell manufacturing technical field, including: to historical abnormal image carries out concentric circle feature extraction (global and local), based on feature extraction result clustering generates abnormal class and Other class;To the incremental abnormal image of the nth incremental period acquisition, carries out concentric circle feature extraction;If 0 < n ≤ m, then based on feature extraction result incremental abnormal image is classified to abnormal class and Other class;If n > m, then based on the clustering center of abnormal class is updated after the incremental abnormal image of past m incremental periods, classification is carried out;Meanwhile, the number of abnormal images in Other class is counted every several incremental periods;If, then Other class is clustered and class is updated.The application can realize the automatic classification and dynamic update of seal weld abnormality, support subsequent root cause analysis.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery cell manufacturing technology, and in particular to a method and system for monitoring abnormalities in battery cell sealing welding based on concentric circle feature extraction and dynamic clustering. Background Technology

[0002] In the sealing welding process of lithium-ion battery cell manufacturing, the welding quality of the liquid injection port (such as the integrity of the solder mark, absence of blast points and warping) directly determines the cell's sealing performance. Figure 1 As shown, welding abnormalities can lead to electrolyte leakage and moisture intrusion, causing cell capacity degradation and decreased safety performance. Currently, the industry primarily employs three technical approaches to manage abnormalities in sealing weld images:

[0003] 1. Traditional manual visual inspection method: This method relies on quality inspectors taking images of the injection port using industrial cameras to manually determine whether there are abnormalities such as blowholes or warped nails. This method is inefficient, highly subjective, and lacks accuracy in identifying subtle differences such as "upper left arc-shaped abnormality" or "upper right arc-shaped abnormality," and it cannot form a standardized abnormality classification.

[0004] 2. Fixed Template Matching Scheme: A pre-set template of normal injection port weld marks is used to determine anomalies by calculating the similarity (e.g., Euclidean distance) between the image to be detected and the template. However, this scheme only focuses on global features and cannot distinguish between local differences such as "single explosion point" and "multiple explosion points," or "upper left arc shape" and "upper right arc shape," resulting in insufficient coverage of anomaly categories.

[0005] 3. Static Clustering Detection Scheme: This scheme uses algorithms such as K-means and DBSCAN to cluster historical anomaly images and establish a fixed anomaly category library. However, this scheme has two major limitations: first, it does not design a specific feature extraction method for the circular weld mark area of ​​the injection port, resulting in low feature discrimination; second, it lacks an incremental clustering mechanism, so newly emerging anomaly types (such as novel knock-off nails) cannot be identified, requiring manual retraining of the model, resulting in poor adaptability.

[0006] As battery cell manufacturing evolves towards higher speed and consistency, the aforementioned technical approaches can no longer meet the requirements for anomaly control, which involve "rapid detection + accurate classification + dynamic updates". Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for monitoring abnormalities in battery cell sealing welds based on concentric circle feature extraction and dynamic clustering. Through concentric circle feature extraction and dynamic clustering, the automatic classification and dynamic updating of sealing weld abnormalities can be realized, supporting subsequent root cause analysis and meeting the abnormality control requirements of "rapid detection + accurate classification + dynamic updating".

[0008] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0009] In a first aspect, the present invention provides a method for monitoring abnormalities in battery cell sealing welds based on concentric circle feature extraction and dynamic clustering, comprising:

[0010] Concentric circle features were extracted from historical abnormal images of abnormal injection port weld marks acquired during historical periods.

[0011] Based on the feature extraction results, the historical abnormal images are clustered to generate an abnormal class and an Other class.

[0012] Concentric circle feature extraction is performed on the incremental abnormal image of the abnormal injection port weld mark obtained in the nth incremental cycle.

[0013] If 0 < n ≤ m, then based on the feature extraction results, the incremental abnormal image is classified into the abnormal class and the Other class;

[0014] If n > m, then the cluster center of the anomaly class is updated based on the incremental anomaly images of the past m incremental periods, and the incremental anomaly images are classified into the anomaly class and the Other class based on the feature extraction results.

[0015] The concentric circle feature extraction includes:

[0016] Preprocess the abnormal images of the abnormal injection port weld marks input;

[0017] The preprocessed abnormal image is subjected to extraction of the injection port weld area, and the extracted injection port weld area image is normalized.

[0018] Global and local features are extracted from the normalized image of the injection port weld area.

[0019] This invention provides a method for monitoring abnormalities in battery cell sealing welding based on concentric circle feature extraction and dynamic clustering. By "precise extraction of concentric circle regions + global-local feature fusion", it effectively filters background noise and accurately captures subtle differences in abnormalities, improves the accuracy of abnormality classification, reduces the probability of misjudgment and missed detection, and significantly reduces the risk of battery cell sealing failure caused by welding abnormalities.

[0020] Optional, also includes:

[0021] Every Each incremental cycle counts the number of abnormal images in the Other category. ;

[0022] like , If a preset upper limit is set for the number of abnormal images in the Other class, then concentric circle features are extracted from the abnormal images in the Other class.

[0023] Based on the feature extraction results, clustering operations are performed on the abnormal images to generate new abnormal classes and new Other classes;

[0024] The similarity between the new anomaly class and the existing anomaly class is calculated based on the feature extraction results.

[0025] If there are anomaly classes whose similarity to the new anomaly class is less than the similarity threshold, then merge them.

[0026] If there is no anomaly class with a similarity less than the similarity threshold to the new anomaly class, then judge the past. Does an exception class with no increment exist in each increment cycle? This refers to the multiplier factor;

[0027] If an exception class exists that does not have an increment, it will be replaced with the new exception class.

[0028] The dynamic clustering provided by this invention includes full clustering, incremental clustering, and class updating, solving the pain point of traditional static clustering's inability to identify new anomalies. Incremental clustering ensures that newly acquired anomaly images are quickly classified, and class updates ensure that the anomaly category library covers actual anomaly types for a long time. The anomaly identification lag time is greatly shortened, eliminating the need for frequent manual retraining of the model and adapting to the dynamic management needs of high-speed production lines.

[0029] Optionally, the clustering operation on the historical anomaly images based on the feature extraction results includes:

[0030] Step S1, Initialization Take a historical anomalous image as the cluster. Cluster centers;

[0031] Step S2: Calculate the clusters based on the feature vectors. The error between cluster centers and historical anomaly images that are not clustered;

[0032] If there are errors below the first threshold, the historical anomaly image corresponding to the smallest error will be classified into a cluster. and update the clusters. Find the cluster centers and return to step S2;

[0033] If there is no error below the first threshold, then let Take a historical anomaly image that is not clustered as the cluster. Find the cluster centers and return to step S2;

[0034] Step S3: After clustering is completed, merge clusters with fewer than the second threshold number of historical abnormal images into a single cluster. ;

[0035] Step S4, each of the clusters The cluster corresponds to an exception class. Corresponding to the Other class;

[0036] Wherein, the updating of the cluster The cluster centers are:

[0037] The clusters The mean image of all historical anomalous images in the dataset is used as its cluster center, and the cluster is selected. The mean of the feature vectors of all historical abnormal images is used as the feature vector of the cluster center.

[0038] This invention uses clustering operations to generate outlier and other classes, without relying on predefined labels. It is suitable for exploratory analysis of data with unknown categories and can reveal groups, distributions or outliers in the data that were not known beforehand, thus solving the pain point of traditional static clustering that "new anomalies cannot be identified".

[0039] Optionally, the preprocessing includes performing grayscale processing and Gaussian filtering on the abnormal image of the input abnormal injection port weld mark.

[0040] In this invention, grayscale processing converts a color image into a single-channel grayscale image. This helps reduce data dimensionality, enabling feature extraction algorithms to handle brightness variations more efficiently and avoiding interference from color information. Gaussian filtering, through weighted averaging using convolution kernels, effectively filters out high-frequency noise and minute details, preventing these irrelevant factors from causing misjudgments during feature extraction, thereby ensuring that the extracted features are more stable and representative.

[0041] Optionally, the step of extracting the injection port weld area from the preprocessed abnormal image includes:

[0042] The preprocessed abnormal image is then used to detect the circular region of the injection port and the coordinates of its center using Hough circle transform.

[0043] By fitting historical normal images of normal injection port weld marks obtained from historical time periods, the inner and outer radii of the injection port weld mark area are obtained.

[0044] A mask for the injection port solder area is generated based on the center coordinates, the inner radius, and the outer radius;

[0045] The injection port weld mark area mask is multiplied by the preprocessed abnormal image to obtain the injection port weld mark area.

[0046] This invention is based on the circular structure of the injection port. Taking the center of the image as the center, it automatically calculates the inner radius (injection hole radius) and the outer radius (weld stamp radius) to extract the annular region of interest (ROI). This improves the background noise suppression rate and the extraction accuracy of local anomalies (such as arc-shaped anomalies), thus preparing for subsequent concentric circle feature extraction.

[0047] Optionally, the feature extraction results of the global feature extraction include:

[0048] Global grayscale mean :

[0049]

[0050] Global grayscale variance :

[0051]

[0052] Global grayscale entropy :

[0053]

[0054] In the formula, Pixels within the injection port soldering area grayscale value, The pixel size of the injection port solder area; The grayscale value of the area within the injection port welding mark is The probability of;

[0055] The local feature extraction includes:

[0056] The area of ​​the injection port is evenly divided into multiple sub-regions along the circumference with the center as the center.

[0057] Extract the maximum grayscale value for each sub-region. Minimum value mean and variance .

[0058] This invention addresses the problem of insufficient feature discrimination by combining global and local feature extraction. Global feature extraction is fast but susceptible to interference and cannot handle local matching. Local feature extraction is robust but may lack overall constraints and perform poorly when a global scene understanding is required. By integrating global and local feature extraction, a balance is struck between speed and accuracy.

[0059] Optionally, the feature extraction results of the global feature extraction are used to calculate the global error between two anomalous images. :

[0060]

[0061] In the formula, The feature extraction results are for the global feature extraction of the two abnormal images. These represent the global grayscale mean. Global grayscale variance and global grayscale entropy ;

[0062] The feature extraction results of the local feature extraction are used to calculate the local error between the two abnormal images. :

[0063]

[0064]

[0065] In the formula, For the two abnormal images, the first Feature extraction results of local feature extraction of each sub-region These represent the maximum grayscale values ​​of the sub-regions, respectively. Minimum value mean and variance , For the first Local errors in individual sub-regions This represents the number of sub-regions.

[0066] Based on the global error and the local error Weighted calculation of the error between two anomalous images :

[0067]

[0068] In the formula, All are weighting coefficients. .

[0069] This invention achieves accurate classification by comprehensively considering the errors in global and local feature extraction results and accurately matching similar anomalies.

[0070] Secondly, the present invention provides a battery cell sealing weld anomaly monitoring system based on concentric circle feature extraction and dynamic clustering, comprising:

[0071] The offline processing module is configured to extract concentric circle features from historical abnormal images of abnormal injection port weld marks acquired over historical time periods; and to perform clustering operations on the historical abnormal images based on the feature extraction results to generate an abnormal class and an Other class.

[0072] The online processing module is configured to extract concentric circle features from the incremental abnormal image of the abnormal injection port weld mark acquired in the nth incremental cycle.

[0073] If 0 < n ≤ m, then based on the feature extraction results, the incremental abnormal image is classified into the abnormal class and the Other class;

[0074] If n > m, then the cluster center of the anomaly class is updated based on the incremental anomaly images of the past m incremental periods, and the incremental anomaly images are classified into the anomaly class and the Other class based on the feature extraction results.

[0075] The concentric circle feature extraction includes:

[0076] Preprocess the abnormal images of the abnormal injection port weld marks input;

[0077] The preprocessed abnormal image is subjected to extraction of the injection port weld area, and the extracted injection port weld area image is normalized.

[0078] Global and local features are extracted from the normalized image of the injection port weld area.

[0079] Optionally, an optimization processing module is also included, configured as follows:

[0080] Every Each incremental cycle counts the number of abnormal images in the Other category. ;

[0081] like , If a preset upper limit is set for the number of abnormal images in the Other class, then concentric circle features are extracted from the abnormal images in the Other class.

[0082] Based on the feature extraction results, clustering operations are performed on the abnormal images to generate new abnormal classes and new Other classes;

[0083] The similarity between the new anomaly class and the existing anomaly class is calculated based on the feature extraction results.

[0084] If there are anomaly classes whose similarity to the new anomaly class is less than the similarity threshold, then merge them.

[0085] If there is no anomaly class with a similarity less than the similarity threshold to the new anomaly class, then judge the past. Does an exception class with no increment exist in each increment cycle? This refers to the multiplier factor;

[0086] If an exception class exists that does not have an increment, it will be replaced with the new exception class.

[0087] Thirdly, the present invention provides an electronic device, including a processor and a storage medium;

[0088] The storage medium is used to store instructions;

[0089] The processor is configured to operate according to the instructions to perform the steps according to the method described above.

[0090] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0091] Fifthly, the present invention provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0092] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0093] This invention provides a method and system for monitoring abnormalities in battery cell sealing welding based on concentric circle feature extraction and dynamic clustering. By "precise extraction of concentric circle regions + global-local feature fusion", it effectively filters background noise and accurately captures subtle differences in abnormalities, improves the accuracy of abnormality classification, reduces the probability of misjudgment and missed detection, and significantly reduces the risk of battery cell sealing failure caused by welding abnormalities.

[0094] Dynamic clustering, comprising full clustering, incremental clustering, and class updates, addresses the pain point of traditional static clustering's inability to identify new anomalies. Incremental clustering ensures rapid classification of newly acquired anomaly images, and class updates ensure that the anomaly category library consistently covers actual anomaly types. This significantly reduces anomaly identification lag time, eliminates the need for frequent manual model retraining, and adapts to the dynamic management needs of high-speed production lines. Attached Figure Description

[0095] Figure 1 This is a schematic diagram of an abnormal weld mark at the injection port provided in the background art of this invention;

[0096] Figure 2 This is a flowchart illustrating the cell sealing welding anomaly monitoring method provided in an embodiment of the present invention;

[0097] Figure 3 This is a schematic diagram of the process for updating exception classes and other classes provided in an embodiment of the present invention;

[0098] Figure 4 This is a schematic diagram of the process of dividing the ring on the injection port welding area into 8 sub-regions along the circumference, according to an embodiment of the present invention.

[0099] Figure 5This is a schematic diagram of the process for clustering historical abnormal images based on feature extraction results, provided in an embodiment of the present invention.

[0100] Figure 6 This is a schematic diagram of the full clustering process provided in an embodiment of the present invention;

[0101] Figure 7 This is a schematic diagram of the incremental clustering process provided in an embodiment of the present invention;

[0102] Figure 8 This is a schematic diagram of the timed category update process provided in an embodiment of the present invention. Detailed Implementation

[0103] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0104] Example 1:

[0105] like Figure 2 As shown, this embodiment of the invention provides a method for monitoring abnormalities in battery cell sealing welds based on concentric circle feature extraction and dynamic clustering, including the following steps:

[0106] Step S10: Extract concentric circle features from historical abnormal images of abnormal injection port weld marks acquired during historical periods;

[0107] Step S20: Based on the feature extraction results, perform clustering operations on historical abnormal images to generate abnormal classes and Other classes;

[0108] Step S30: Initialize n=1;

[0109] Step S40: Extract concentric circle features from the incremental abnormal image of the abnormal injection port weld mark obtained in the nth incremental cycle.

[0110] If 0 < n ≤ m, then based on the feature extraction results, the incremental abnormal images are classified into the abnormal class and the Other class;

[0111] If n > m, then the cluster center of the anomaly class is updated based on the incremental anomaly images of the past m incremental periods, and the incremental anomaly images are classified into the anomaly class and the Other class based on the feature extraction results.

[0112] Step S50, n=n+1, return to step S40.

[0113] Based on the above-mentioned methods for monitoring abnormalities in battery cell sealing welds, such as Figure 3 As shown, the embodiments of the present invention further include:

[0114] Step S11, every Each incremental cycle counts the number of abnormal images in the Other category. ;

[0115] Step S21, if , If a preset upper limit is set for the number of abnormal images in the Other class, then concentric circle features are extracted from the abnormal images in the Other class.

[0116] Step S31: Based on the feature extraction results, perform clustering operations on the abnormal images to generate new abnormal classes and new Other classes;

[0117] Step S41: Calculate the similarity between the new anomaly class and the existing anomaly class based on the feature extraction results.

[0118] If there are anomaly classes whose similarity to the new anomaly class is less than the similarity threshold, then merge them.

[0119] If there is no anomaly class with a similarity less than the similarity threshold to the new anomaly class, then judge the past. Does an exception class with no increment exist in each increment cycle? This refers to the multiplier factor;

[0120] If an exception class exists that does not have an increment, it will be replaced with the new exception class.

[0121] Specifically, in this embodiment, considering the actual usage scenario, the incremental period is set to daily. The incremental cycle is set to monthly. Each incremental cycle is set to 3 months.

[0122] Specifically, in this embodiment, the concentric circle feature extraction in steps S10 and S40 includes:

[0123] (1) Preprocess the abnormal images of the abnormal injection port weld marks.

[0124] All abnormal images were acquired using an industrial CCD camera, installed behind the sealing and welding station, to vertically capture images of the injection port and obtain RGB images.

[0125] Preprocessing includes grayscale conversion and Gaussian filtering of the abnormal weld marks on the input injection port. Grayscale conversion uses a weighted average method: Gray = 0.299R + 0.587G + 0.114B, where R, G, and B represent the three primary color channels: red, green, and blue, respectively. Gaussian filtering removes high-frequency noise using a 5×5 convolution kernel and a standard deviation of 1.2.

[0126] Grayscale conversion transforms a color image into a single-channel grayscale image, which helps reduce data dimensionality, allowing feature extraction algorithms to process brightness variations more efficiently and avoiding interference from color information. Gaussian filtering, on the other hand, uses convolutional kernels to perform weighted averaging, effectively filtering out high-frequency noise and minute details, preventing these irrelevant factors from causing misjudgments during feature extraction, thus ensuring that the extracted features are more stable and representative.

[0127] The pre-processed abnormal images can be indexed by "cell number - equipment number - welding time" and stored in the image database for later use.

[0128] (2) Extract the injection port weld area from the preprocessed abnormal image and normalize the extracted injection port weld area image.

[0129] (2.1) For the preprocessed abnormal image, Hough circle transform is used to detect the circular region of the injection port and its center coordinates; combined with the actual application scenario, the parameters of Hough circle transform are set as follows: minimum radius 1.5mm (injection hole radius), maximum radius 2.5mm (injection hole + solder inner layer radius), and center coordinates. Positioning accuracy ±1 pixel.

[0130] (2.2) Fit historical normal images of normal injection port weld marks acquired during historical periods to obtain the inner and outer radii of the injection port weld mark area; combine with actual usage scenarios, collect 1000 historical normal images and fit the inner radius. and outer radius , Radius of the injection hole, inner radius and outer radius The circular area formed is the core area of ​​the concentric circles, and also the area where the injection port is soldered.

[0131] (2.3) Based on the coordinates of the center of the circle , inner radius Outer radius Generate a mask for the injection port welding area, where the grayscale value of pixels within the circular area is 1, and the grayscale value of pixels outside the circular area is 0.

[0132] (2.4) Multiply the mask of the injection port weld mark area with the preprocessed abnormal image to obtain the injection port weld mark area.

[0133] (2.5) The size and pixel normalization of the solder area of ​​the liquid injection port are performed. Based on the actual usage scenario, it is scaled to 256×256 pixels to eliminate the size difference of the liquid injection port of different cells. The formula is:

[0134]

[0135] In the formula, The image dimensions before and after normalization;

[0136]

[0137] In the formula, These represent the maximum and minimum grayscale values ​​of pixels in the 256×256 pixel injection port solder joint area. The pixels in the 256×256 pixel injection port solder area grayscale value, for The normalized result.

[0138] (3) Perform global feature extraction and local feature extraction on the normalized image of the injection port welding area.

[0139] (3.1) The feature extraction results of global feature extraction include:

[0140] Global grayscale mean :

[0141]

[0142] Global grayscale variance :

[0143]

[0144] Global grayscale entropy :

[0145]

[0146] In the formula, Pixels within the injection port soldering area grayscale value, The pixel size of the injection port solder joint area is determined based on the above normalization results. ; The grayscale value of the area within the injection port welding mark is The probability of.

[0147] Local feature extraction includes:

[0148] The area of ​​the injection port is evenly divided into multiple sub-regions along the circumference with the center as the center.

[0149] Extract the maximum grayscale value for each sub-region. Minimum value mean and variance .like Figure 4As shown, in combination with actual usage scenarios, the ring on the injection port welding area is evenly divided into 8 sub-areas along the circumference, spaced at 45° intervals, numbered 1-8.

[0150] Specifically, in this embodiment, step S20 involves clustering historical anomaly images based on feature extraction results, such as... Figure 5 As shown, it includes:

[0151] Step S1, Initialization Take a historical anomalous image as the cluster. Cluster centers;

[0152] Step S2: Calculate clusters based on feature vectors The error between cluster centers and historical anomaly images that are not clustered;

[0153] If there are errors below the first threshold, the historical anomaly image corresponding to the smallest error will be classified into a cluster. and update the clusters. Find the cluster center and return to step S2; based on the actual use scenario and the normal classification accuracy of 95%, determine the first threshold as 0.6;

[0154] If there is no error below the first threshold, then let Take a historical anomaly image that is not clustered as the cluster. Find the cluster centers and return to step S2;

[0155] Step S3: After clustering is completed, merge clusters with fewer than the second threshold number of historical abnormal images into a single cluster. ;

[0156] Step S4, for each cluster Corresponding to an exception class, a cluster Corresponding to the Other class;

[0157] Among them, updating clusters The cluster centers are:

[0158] Clustering The mean image of all historical anomalous images in the data is used as its cluster center, and the clusters are selected. The mean of the feature vectors of all historical abnormal images is used as the feature vector of the cluster center.

[0159] Specifically, in this embodiment, the error calculation in step S20 depends on the following calculation method:

[0160] The feature extraction results from global feature extraction are used to calculate the global error between two anomalous images. :

[0161]

[0162] In the formula, The feature extraction results are for the global feature extraction of the two abnormal images. These represent the global grayscale mean. Global grayscale variance and global grayscale entropy ;

[0163] The feature extraction results from local feature extraction are used to calculate the local error between two anomalous images. :

[0164]

[0165]

[0166] In the formula, For the two abnormal images, the first Feature extraction results of local feature extraction of each sub-region These represent the maximum grayscale values ​​of the sub-regions, respectively. Minimum value mean and variance , For the first Local errors in individual sub-regions This represents the number of sub-regions.

[0167] Based on global error and local error Weighted calculation of the error between two anomalous images :

[0168]

[0169] In the formula, All are weighting coefficients. .

[0170] Specifically in this embodiment, in step S41, the similarity between the new anomaly class and the anomaly class is calculated based on the feature extraction results. Similarly, the error calculation method is used to calculate the error between the cluster centers of the new anomaly class and the anomaly class, and the similarity is represented by the magnitude of the error.

[0171] Specifically, in this embodiment, step S40 classifies the incremental abnormal images into anomaly and other categories based on the feature extraction results, including:

[0172] Calculate the error between the incremental anomaly image and the cluster center of each anomaly class, and take the minimum error. If the minimum error is lower than the third threshold, the incremental anomaly image is classified into the corresponding anomaly class; otherwise, the incremental anomaly image is classified into the Other class.

[0173] In summary, the cell sealing welding anomaly monitoring method provided in this embodiment of the invention can be divided into three stages. Steps S10 and S20 constitute the first stage, as follows: Figure 6 As shown, full clustering is performed. Step S40 is the second stage, as follows: Figure 7 As shown, incremental clustering is performed. Steps S11-S41 constitute the third stage, as follows: Figure 8 As shown, update the exception class and the Other class.

[0174] After the cell sealing weld anomaly monitoring method provided in this embodiment of the invention achieves automated classification and dynamic updating of sealing weld anomalies, subsequent root cause analysis may include:

[0175] (1) The efficiency of root cause investigation has been greatly improved, reducing production line downtime costs.

[0176] An anomaly database is built to achieve a full-link association between "anomaly type - process parameter - handling record". Engineers can quickly query historical process parameter deviations by anomaly type. Combined with root cause suggestions automatically generated by the system, troubleshooting time is greatly reduced and maintenance efficiency is significantly improved. At the same time, real-time alarms and workstation pause functions triggered by serious anomalies (multiple explosion points, knocked-out screws) prevent the generation of defective products in batches and reduce downtime losses.

[0177] (2) Data supports process optimization and improves the consistency of cell manufacturing.

[0178] The anomaly database has accumulated data on "anomaly type - process parameter - quality result" over a long period. Statistical analysis can identify the direction for process parameter optimization (e.g., "when the welding energy is controlled at 200-230J, the proportion of single-explosion anomalies decreases from 12% to 3%"). Data-driven process adjustments have improved the consistency of cell sealing weld quality (weld stamp roundness) from 88% to over 96%, providing a key guarantee for the overall performance stability of the cell.

[0179] Example 2

[0180] This invention provides a battery cell sealing weld anomaly monitoring system based on concentric circle feature extraction and dynamic clustering, comprising:

[0181] The offline processing module is configured to extract concentric circle features from historical abnormal images of abnormal injection port weld marks acquired in historical time periods; and to perform clustering operations on the historical abnormal images based on the feature extraction results to generate abnormal classes and Other classes.

[0182] The online processing module is configured to extract concentric circle features from the incremental abnormal image of the abnormal injection port weld mark acquired in the nth incremental cycle.

[0183] If 0 < n ≤ m, then based on the feature extraction results, the incremental abnormal images are classified into the abnormal class and the Other class;

[0184] If n > m, then the cluster center of the anomaly class is updated based on the incremental anomaly images of the past m incremental periods, and the incremental anomaly images are classified into the anomaly class and the Other class based on the feature extraction results.

[0185] The optimization processing module is configured as follows:

[0186] Every Each incremental cycle counts the number of abnormal images in the Other category. ;

[0187] like , If a preset upper limit is set for the number of abnormal images in the Other class, then concentric circle features are extracted from the abnormal images in the Other class.

[0188] Based on the feature extraction results, clustering operations are performed on the abnormal images to generate new abnormal classes and new Other classes;

[0189] The similarity between the new anomaly class and the existing anomaly class is calculated based on the feature extraction results.

[0190] If there are anomaly classes whose similarity to the new anomaly class is less than the similarity threshold, then merge them.

[0191] If there is no anomaly class with a similarity less than the similarity threshold to the new anomaly class, then judge the past. Does an exception class with no increment exist in each increment cycle? This refers to the multiplier factor;

[0192] If an exception class exists that does not have an increment, it will be replaced with the new exception class.

[0193] The concentric circle feature extraction includes:

[0194] Preprocess the abnormal images of the abnormal injection port weld marks input;

[0195] The injection port weld area is extracted from the preprocessed abnormal image, and the extracted injection port weld area image is normalized.

[0196] Global and local features were extracted from the normalized image of the injection port weld area.

[0197] Example 3

[0198] Based on the cell sealing welding anomaly monitoring method provided in Embodiment 1, this embodiment of the invention provides an electronic device, including a processor and a storage medium;

[0199] Storage media are used to store instructions;

[0200] The processor is used to perform operations according to instructions to execute the steps according to the method described above.

[0201] Example 4

[0202] Based on the cell sealing welding anomaly monitoring method provided in Embodiment 1, this embodiment of the invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above method.

[0203] Example 5

[0204] Based on the cell sealing welding anomaly monitoring method provided in Embodiment 1, this embodiment of the invention provides a computer program product, including a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0205] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0206] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0207] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0208] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0209] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring abnormality of cell seal welding based on concentric circle feature extraction and dynamic clustering, characterized in that, include: Concentric circle features were extracted from historical abnormal images of abnormal injection port weld marks acquired during historical periods. Based on the feature extraction results, the historical abnormal images are clustered to generate an abnormal class and an Other class. Concentric circle feature extraction is performed on the incremental abnormal image of the abnormal injection port weld mark obtained in the nth incremental cycle. If 0 < n ≤ m, then based on the feature extraction results, the incremental abnormal image is classified into the abnormal class and the Other class; If n > m, then the cluster center of the anomaly class is updated based on the incremental anomaly images of the past m incremental periods, and the incremental anomaly images are classified into the anomaly class and the Other class based on the feature extraction results. The concentric circle feature extraction includes: Preprocess the abnormal images of the abnormal injection port weld marks input; The preprocessed abnormal image is subjected to extraction of the injection port weld area, and the extracted injection port weld area image is normalized. Global and local features are extracted from the normalized image of the injection port weld area.

2. The cell seam inspection method based on concentric circle feature extraction and dynamic clustering of claim 1, wherein, Also includes: Every other increment period, count the number of abnormal images in the Other class ;​ If , is the preset upper limit of the number of abnormal images in the Other category, the concentric circle feature extraction is performed on the abnormal images in the Other category. Based on the feature extraction results, clustering operations are performed on the abnormal images to generate new abnormal classes and new Other classes; The similarity between the new anomaly class and the existing anomaly class is calculated based on the feature extraction results. If there are anomaly classes whose similarity to the new anomaly class is less than the similarity threshold, then merge them. If there is no anomaly class similar to the new anomaly class with a similarity less than a similarity threshold, it is determined whether there is an anomaly class without an increment in the past increment period, is a rate coefficient; If an exception class exists that does not have an increment, it will be replaced with the new exception class.

3. The cell seam inspection method based on concentric circle feature extraction and dynamic clustering of claim 1, wherein, The clustering operation of the historical anomaly images based on the feature extraction results includes: Step S1, initialization , take a historical abnormal image as the clustering center of a clustering cluster ; Step S2, calculating the clustering centers of the clustering clusters based on the feature vectors of the un-clustered historical abnormal images; If there are errors below the first threshold, the historical anomaly image corresponding to the smallest error will be classified into a cluster. and update the clusters. Find the cluster centers and return to step S2; If there is no error below the first threshold, then let Take a historical anomaly image that is not clustered as the cluster. Find the cluster centers and return to step S2; Step S3, after clustering is completed, merging the cluster clusters whose number of historical abnormal images is lower than the second threshold value into a cluster cluster ; Step S4, each of the cluster groups corresponding to an exception class, the cluster group corresponding to the Other class; The updating the cluster centers of the cluster clusters comprises: The updating the cluster centers of the cluster clusters comprises: The cluster center of the cluster The mean image of all historical abnormal images in the cluster The mean of the feature vectors of all historical abnormal images in the cluster 4. The cell seam inspection method based on concentric circle feature extraction and dynamic clustering of claim 1, wherein, The preprocessing includes grayscale processing and Gaussian filtering of the abnormal image of the abnormal injection port weld mark.

5. The method of cell seam weld anomaly monitoring based on concentric circle feature extraction and dynamic clustering of claim 1, wherein, The step of extracting the injection port weld area from the preprocessed abnormal image includes: The preprocessed abnormal image is then used to detect the circular region of the injection port and the coordinates of its center using Hough circle transform. By fitting historical normal images of normal injection port weld marks obtained from historical time periods, the inner and outer radii of the injection port weld mark area are obtained. A mask for the injection port solder area is generated based on the center coordinates, the inner radius, and the outer radius; The injection port weld mark area mask is multiplied by the preprocessed abnormal image to obtain the injection port weld mark area.

6. The method of cell seam weld anomaly monitoring based on concentric circle feature extraction and dynamic clustering of claim 1, wherein, The feature extraction results of the global feature extraction include: Global gray mean : global gray scale variance : global gray level entropy : In the formula, Pixels within the injection port soldering area grayscale value, The pixel size of the injection port solder area; The grayscale value of the area within the injection port welding mark is The probability of; The local feature extraction includes: The area of ​​the injection port is evenly divided into multiple sub-regions along the circumference with the center as the center. Extract the maximum grayscale value for each sub-region. Minimum value mean and variance .

7. The cell seam weld anomaly monitoring method based on concentric circle feature extraction and dynamic clustering of claim 6, wherein, The feature extraction result of the global feature extraction is used to calculate a global error between two abnormal images : In the formula, is a feature extraction result of global features of two abnormal images, respectively represent global gray mean , global gray variance , and global gray entropy ; The feature extraction result of the local feature extraction is used to calculate local errors between two abnormal images : In the formula, For the two abnormal images, the first Feature extraction results of local feature extraction of each sub-region These represent the maximum grayscale values ​​of the sub-regions, respectively. Minimum value mean and variance , For the first Local errors in individual sub-regions Number of sub-regions; Based on the global error and the local error Weighted calculation of the error between two anomalous images : In the formula, All are weighting coefficients. .

8. A cell seal weld anomaly monitoring system based on concentric circle feature extraction and dynamic clustering, characterized in that, include: The offline processing module is configured to extract concentric circle features from historical abnormal images of abnormal injection port weld marks acquired over historical time periods. Based on the feature extraction results, the historical abnormal images are clustered to generate an abnormal class and an Other class. The online processing module is configured to extract concentric circle features from the incremental abnormal image of the abnormal injection port weld mark acquired in the nth incremental cycle. If 0 < n ≤ m, then based on the feature extraction results, the incremental abnormal image is classified into the abnormal class and the Other class; If n > m, then the cluster center of the anomaly class is updated based on the incremental anomaly images of the past m incremental periods, and the incremental anomaly images are classified into the anomaly class and the Other class based on the feature extraction results. The concentric circle feature extraction includes: Preprocess the abnormal images of the abnormal injection port weld marks input; The preprocessed abnormal image is subjected to extraction of the injection port weld area, and the extracted injection port weld area image is normalized. Global and local features are extracted from the normalized image of the injection port weld area.

9. The concentric feature extraction and dynamic clustering based cell seai weld anomaly monitoring system of claim 8, wherein, It also includes an optimization processing module, configured as follows: Every other increment period, count the number of abnormal images in the Other class ;​ If , is the preset upper limit of the number of abnormal images in the Other category, the abnormal images in the Other category are subjected to concentric circle feature extraction. Based on the feature extraction results, clustering operations are performed on the abnormal images to generate new abnormal classes and new Other classes; The similarity between the new anomaly class and the existing anomaly class is calculated based on the feature extraction results. If there are anomaly classes whose similarity to the new anomaly class is less than the similarity threshold, then merge them. If there is no anomaly class with a similarity less than the similarity threshold to the new anomaly class, then judge the past. Does an exception class with no increment exist in each increment cycle? This refers to the multiplier factor; If an exception class exists that does not have an increment, it will be replaced with the new exception class.

10. An electronic device, comprising: Including processor and storage media; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

11. A computer readable storage medium having stored thereon a computer program, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

12. A computer program product comprising computer programs / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-7.

Images