Intelligent welding defect detection method and device for battery cover plate explosion-proof valve

Abstract

The application provides an intelligent welding defect detection method and device for a battery cover plate explosion-proof valve, and relates to the technical field of intelligent welding systems, which comprises the following steps: collecting basic information of the explosion-proof valve of the battery cover plate explosion-proof valve, including attribute information of the explosion-proof valve and welding position information of the explosion-proof valve; performing global image collection to obtain a welding position image; performing feature extraction and comparison to locate a welding defect area, performing local image collection to obtain a welding defect area image; performing negative sampling to obtain a welding defect sample image set; and based on defect comparison, obtaining defect type identification results corresponding to multiple known defect types. The application solves the technical problem that the prior art can usually only detect a certain type of welding defect, and the classification ability for different types of welding defects is insufficient, resulting in missed detection of welding defects and further affecting the product quality of the intelligent welding system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent welding systems, and particularly relates to an intelligent welding defect detection method and device for a battery cover plate explosion-proof valve. BACKGROUND

[0002] The battery cover plate explosion-proof valve is an important component applied in high-pressure and dangerous gas environments, and its function is to protect the system from excessive pressure and ensure safe operation of the system. Since the working environment and safety requirements of the battery cover plate explosion-proof valve are very strict, the welding quality is crucial to its performance, and any welding defect may cause the explosion-proof valve to fail, thereby threatening the safety of the system. Therefore, the welding quality of the explosion-proof valve must be strictly detected to ensure that the quality of each welding point meets the standard.

[0003] However, the prior art lacks intelligent automatic positioning capability of welding positions and defect areas, and traditional methods require manual intervention for area calibration. In large-scale production, such manual positioning is very time-consuming and lacks real-time capability, resulting in a lack of automated and accurate defect positioning capability, thereby leading to low detection efficiency and failing to meet the needs of large-scale production. Moreover, the prior art lacks classification capability for different types of welding defects, and many existing methods can only detect a certain type of welding defect. For complex and diversified defect types such as micro cracks, pores and slag, the recognition accuracy and adaptability are poor, which reduces the accuracy of the detection method when facing multiple welding defects, thereby leading to missed detection of welding defects and affecting the product quality of the intelligent welding system. SUMMARY

[0004] The present application provides an intelligent welding defect detection method and device for a battery cover plate explosion-proof valve, aiming to solve the technical problem that the prior art can generally only detect a certain type of welding defect, lacks classification capability for different types of welding defects, leads to missed detection of welding defects, and further affects the product quality of the intelligent welding system.

[0005] In a first aspect, the application discloses an intelligent welding defect detection method for a battery cover plate explosion-proof valve, which comprises the following steps: collecting explosion-proof valve basic information of the battery cover plate explosion-proof valve, wherein the explosion-proof valve basic information comprises explosion-proof valve attribute information and explosion-proof valve welding position information; performing global image collection based on the explosion-proof valve welding position information to obtain a welding position image; performing feature extraction and comparison on the welding position image to locate a welding defect area, performing local image collection based on the welding defect area to obtain a welding defect area image; performing negative sampling based on the explosion-proof valve attribute information to obtain a welding defect sample image set, wherein the welding defect sample image set is marked with a plurality of known defect types; and performing defect comparison on the welding defect area image based on the welding defect sample image set to obtain a defect type identification result corresponding to the plurality of known defect types.

[0006] In a second aspect, the application discloses an intelligent welding defect detection device for a battery cover plate explosion-proof valve, which is used in the intelligent welding defect detection method for the battery cover plate explosion-proof valve, and comprises the following modules: a basic information collection module, which is used to collect explosion-proof valve basic information of the battery cover plate explosion-proof valve, wherein the explosion-proof valve basic information comprises explosion-proof valve attribute information and explosion-proof valve welding position information; a global image collection module, which is used to perform global image collection based on the explosion-proof valve welding position information to obtain a welding position image; a local image collection module, which is used to perform feature extraction and comparison on the welding position image to locate a welding defect area, and perform local image collection based on the welding defect area to obtain a welding defect area image; a negative sampling module, which is used to perform negative sampling based on the explosion-proof valve attribute information to obtain a welding defect sample image set, wherein the welding defect sample image set is marked with a plurality of known defect types; and a defect comparison module, which is used to perform defect comparison on the welding defect area image based on the welding defect sample image set to obtain a defect type identification result corresponding to the plurality of known defect types.

[0007] The one or more technical solutions provided in the application have at least the following beneficial effects:

[0008] By collecting the explosion-proof valve basic information, including the explosion-proof valve attribute information and the explosion-proof valve welding position information, the intelligent welding system can perform subsequent image collection according to the accurate welding position information, which helps to accurately locate the welding area to be detected, avoids invalid or redundant image collection, improves the efficiency, and the explosion-proof valve attribute information enables the system to consider specific welding requirements such as material, geometric size, process requirements, and thus ensures that the welding defect detection can be analyzed according to specific product standards; by performing global image collection according to the explosion-proof valve welding position information, the complete image of the welding position is ensured to be obtained, and the key welding area is avoided to be missed, and such global collection provides more comprehensive context information for subsequent welding defect detection, avoiding the limitations that may be caused by local image collection; the welding position image is compared and extracted for feature extraction, and the welding defect area is accurately located, and the intelligent welding system can focus on the defect area through such accurate defect positioning for subsequent local image collection, improving the efficiency and accuracy of image collection; based on the explosion-proof valve attribute information, the negative sampling is performed to obtain a welding defect sample image set, and through the negative sample, the normal and abnormal welding areas can be better distinguished, thereby reducing false positives and improving the accuracy of defect identification; based on the welding defect sample image set, the welding defect area image is compared for defects, and the intelligent welding system can identify the known defect types, and through such comparison and analysis, it can automatically determine whether the welding area meets the standard and accurately identify the defect type, and such method not only improves the accuracy of defect identification, but also can handle different types of defects, so that the intelligent welding system can adapt to complex and variable welding quality control requirements.

[0009] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the content of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0010] Figure 1 The intelligent welding defect detection method flowchart of the battery cover plate explosion-proof valve provided by the embodiment of the present application.

[0011] Figure 2 The intelligent welding defect detection device structure schematic diagram of the battery cover plate explosion-proof valve provided by the embodiment of the present application.

[0012] Explanation of reference signs: basic information collection module 10, global image collection module 20, local image collection module 30, negative sampling module 40, and defect comparison module 50. DETAILED DESCRIPTION

[0013] This application provides an intelligent welding defect detection method and device for battery cover explosion-proof valves, which solves the technical problem that the prior art can usually only detect one type of welding defect and has insufficient ability to classify different types of welding defects, resulting in missed detection of welding defects and thus affecting the product quality of the intelligent welding system.

[0014] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0015] Example 1, as Figure 1 As shown in the embodiment of this application, an intelligent welding defect detection method for a battery cover explosion-proof valve is provided, the method comprising:

[0016] Collect the basic information of the explosion-proof valve of the battery cover, wherein the basic information of the explosion-proof valve includes the attribute information of the explosion-proof valve and the welding position information of the explosion-proof valve.

[0017] The basic information of the explosion-proof valve on the battery cover is collected. This includes the valve's attribute information, such as its material and mechanical properties, dimensions and geometry, operational safety performance, and welding manufacturing process. This data helps to adapt subsequent welding defect detection methods. For example, the valve's material, such as metal or composite material, affects the welding process requirements; the welding method, such as laser welding or gas-shielded welding, affects the welding quality. The explosion-proof valve's welding location information refers to the specific welding positions on the battery cover explosion-proof valve, indicating which areas require welding inspection. By locating the welding positions, subsequent image acquisition can more accurately focus on specific areas.

[0018] Global image acquisition is performed based on the welding position information of the explosion-proof valve to obtain the welding position image.

[0019] Based on the welding position information of the explosion-proof valve, the welding area on the battery cover explosion-proof valve is determined. The welding position information of the explosion-proof valve can guide the acquisition of a global image covering the entire explosion-proof valve. For example, if the explosion-proof valve has multiple welding points, the image acquisition system automatically marks these welding positions to ensure that the image contains all welding areas that need to be detected. The welding position image is located in the global image, providing more contextual information for subsequent welding defect location.

[0020] Feature extraction and comparison are performed on the welding position image to locate the welding defect area. Based on the welding defect area, local image acquisition is performed to obtain the welding defect area image.

[0021] Key features such as edges, textures, colors, etc. in the welding position image are extracted, and then these features are compared with standard welding sample images through feature comparison. The comparison can be achieved by calculating the difference between the images, such as using image similarity measurement method. According to the comparison result, it is determined whether the welding area in the image has defects. Through the results of feature extraction and comparison, the defect area existing in the image is located, for example, if the features of a certain welding point are greatly different from the standard sample image, it is considered that the area may have defects. After locating the welding defect area, local image acquisition is carried out for the welding defect area to further obtain detailed images of the defect area. This local image acquisition can be realized by high-resolution cameras or other precise imaging equipment to capture the specific details of the welding defects, ensuring that the subsequent defect identification process is more accurate.

[0022] Based on the explosion-proof valve attribute information, negative sampling is performed to obtain a welding defect sample image set, wherein the welding defect sample image set identifies a plurality of known defect types.

[0023] Based on the explosion-proof valve attribute information, negative sampling is performed to obtain a welding defect sample image set, wherein the welding defect sample image set identifies a plurality of known defect types. Negative sampling refers to extracting samples from a large number of defect images to obtain a welding defect sample image set as negative samples, which covers multiple known welding defect types, such as pores, cracks, uneven welds, etc. Negative sampling can collect these defect images from the actual production process. Through negative sampling, a sample image set containing multiple known defect types is obtained, and the corresponding defect type labels are labeled on these sample images as the basis for training and comparison.

[0024] Based on the welding defect sample image set, the welding defect area image is compared for defects to obtain defect type identification results corresponding to the plurality of known defect types.

[0025] The key of the process is to find the most matched defect type through image feature comparison. Specifically, the defect comparison can be performed by various methods, such as feature extraction and comparison based on a convolutional neural network, or based on a traditional image matching algorithm, according to the similarity or feature difference of the images to judge the defect type. For example, if the welding defect region image is similar to a sample image of a known defect type, it is marked as the defect of the type. Based on the comparison result, the defect type recognition result corresponding to multiple known defect types, such as pores, cracks, uneven welds, etc., is output, which provides support for quality control and further repair work.

[0026] Further, the welding defect region image is compared with the welding defect sample image set to obtain the defect type recognition result corresponding to the multiple known defect types, and the method comprises:

[0027] A first welding defect sample image set of a first known defect type in the multiple known defect types is extracted, and a first defect comparison group set of the first welding defect sample image set and the welding defect region image is established; a defect similarity recognition is performed on the first defect comparison group set based on a similarity twin network to obtain a first defect similarity; when the first defect similarity meets a preset defect similarity, the first known defect type is added to the defect type recognition result; and the multiple known defect types are iterated in this way.

[0028] An analysis object is randomly selected from the multiple known defect types as a first known defect type, which facilitates subsequent traversal analysis of all known defect types. A first welding defect sample image set of the first known defect type is extracted, and these sample images represent a specific known defect type. The images in the selected first welding defect sample image set are paired with the welding defect region image to form a first defect comparison group set, so that each sample image is compared with the welding defect region image to be detected one by one to form a comparison group, and the goal is to judge whether the welding defect region contains the defect of the first known defect type through the feature similarity between the images.

[0029] The similarity twin network is a model for determining whether two input images are similar by training a neural network, which outputs a value representing the similarity by comparing the feature embeddings of the input images. The network consists of two parts, each containing a set of convolutional neural networks sharing weights. These two networks extract features from the input images and evaluate the similarity by calculating their distance. Common distance measurement methods include Euclidean distance or cosine similarity. For each set of weld defect region images to be detected and sample images, the similarity twin network calculates their similarity scores. A high similarity value indicates that the weld defect region image to be detected is similar to the sample image and may have the defect type represented by the sample; a low similarity value indicates a large difference and may not be the defect type. The first defect similarity calculated will be used as the basis for the recognition result. Each pair of images outputs a similarity score representing the similarity of the pair of images.

[0030] A preset defect similarity is set. When the similarity calculated by the similarity twin network is greater than or equal to this threshold, it indicates that the weld defect region image to be detected is very similar to the sample image of the defect type, and the defect type can be determined. If the first defect similarity meets the preset defect similarity, the first known defect type is added to the defect type recognition result, indicating that the first known defect type has been identified as a problem in the weld region.

[0031] The same process is repeated to traverse all the multiple known defect types one by one until all possible defect types are checked. Whenever a known defect type meets the preset defect similarity, the known defect type is added to the defect type recognition result. After traversing all the multiple known defect types, the complete defect type recognition result is output, listing all the defect types detected in the weld defect region.

[0032] Further, the method comprises:

[0033] The similarity twin network is established, wherein the similarity twin network includes a feature mapping layer, and the features of the feature mapping layer include edge feature overlap, texture feature overlap, color feature overlap, and local shape feature overlap. The edge feature overlap, texture feature overlap, color feature overlap, and local shape feature overlap are calculated according to a weight sharing network, and the first defect similarity is output.

[0034] A similarity twin network is constructed, which extracts features from input images through two sub-networks sharing weights, to determine the similarity between two images. The two sub-networks process the two input images respectively, and output a similarity value representing the degree of similarity between the two images by calculating the feature embedding of the two images. Each sub-network is identical in structure and shares the same convolutional layers and weights, meaning that they use the same feature extraction method when processing different input images. This shared weight design enables the network to effectively compare the similarity of two images. The feature mapping layer is used to convert image input into a high-dimensional vector that reflects the essential features of the image, i.e., feature embedding. These features are typically used to calculate the similarity between two images.

[0035] wherein the edge feature overlap degree reflects the shape and boundary structure of the edges in the image, and the edge information is very important in welding defect detection because welding defects often change the shape of the edges in the image; the texture feature overlap degree reflects the texture changes in the image, and the defect area often causes abnormal texture, so the texture feature is an important basis for defect detection; the color feature overlap degree reflects the distribution of colors in the image, and the color feature is also very important in the recognition of welding defects, especially in the heat-affected zone (e.g., overheated area or insufficiently welded area) which may cause color changes; the local shape feature overlap degree reflects the changes in local shape in the image, and defects often cause the local shape of the welding area to differ from the shape of the standard welding area, and this feature can effectively reveal shape defects.

[0036] For each feature, including edges, textures, colors, and local shapes, their overlap degrees are calculated based on their performance in the image to be detected and the sample image. The overlap degree is an indicator that measures the similarity of two images in that feature dimension. For example, a high edge feature overlap degree means that the two images are highly similar in edge structure. For each feature, their similarity is calculated, which is usually achieved by calculating the distance of the feature vector, such as Euclidean distance, cosine similarity, etc. The overlap degrees of various features are weighted and summed or aggregated in other ways to calculate the overall similarity score. The influence of each feature on the final similarity score can be adjusted by weight, for example, if the color feature is crucial to the recognition of the welding defect area, a higher weight can be given to the color feature. Finally, the first defect similarity is output, representing the similarity of the image to be detected and the sample image in a specific defect type. A high similarity indicates that the image may contain that defect type, and a low similarity indicates that the defect type is less likely.

[0037] Further, the method for feature extraction and comparison of the welding position image to locate the welding defect area comprises:

[0038] Based on the explosion-proof valve attribute information, positive sampling is performed to obtain a welding standard sample image; a feature extraction comparator is constructed based on a convolutional neural network, wherein the feature extraction comparator includes a first feature extraction channel and a second feature extraction channel that share weights; the welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel to perform welding feature extraction and loss comparison analysis, and a welding feature loss data set is obtained; and based on the welding feature loss data set, a welding defect region is located in the welding position image.

[0039] Based on the explosion-proof valve attribute information, positive sampling is performed to obtain a welding standard sample image; a feature extraction comparator is constructed based on a convolutional neural network, wherein the feature extraction comparator includes a first feature extraction channel and a second feature extraction channel that share weights; the welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel to perform welding feature extraction and loss comparison analysis, and a welding feature loss data set is obtained; and based on the welding feature loss data set, a welding defect region is located in the welding position image.

[0040] Based on the explosion-proof valve attribute information, positive sampling is performed to obtain a welding standard sample image; a feature extraction comparator is constructed based on a convolutional neural network, wherein the feature extraction comparator includes a first feature extraction channel and a second feature extraction channel that share weights; the welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel to perform welding feature extraction and loss comparison analysis, and a welding feature loss data set is obtained; and based on the welding feature loss data set, a welding defect region is located in the welding position image.

[0041] Based on the explosion-proof valve attribute information, positive sampling is performed to obtain a welding standard sample image; a feature extraction comparator is constructed based on a convolutional neural network, wherein the feature extraction comparator includes a first feature extraction channel and a second feature extraction channel that share weights; the welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel to perform welding feature extraction and loss comparison analysis, and a welding feature loss data set is obtained; and based on the welding feature loss data set, a welding defect region is located in the welding position image.

[0042] Based on the explosion-proof valve attribute information, positive sampling is performed to obtain a welding standard sample image; a feature extraction comparator is constructed based on a convolutional neural network, wherein the feature extraction comparator includes a first feature extraction channel and a second feature extraction channel that share weights; the welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel to perform welding feature extraction and loss comparison analysis, and a welding feature loss data set is obtained; and based on the welding feature loss data set, a welding defect region is located in the welding position image.

[0043] Further, the method further comprises:

[0044] An analysis of the welding requirements of the battery cover plate explosion-proof valve is performed to obtain explosion-proof valve welding requirement information, wherein the explosion-proof valve welding requirement information includes welding defect tolerance, welding strength constraints, and explosion-proof performance constraints; and the welding standard sample images are screened based on the explosion-proof valve welding requirement information.

[0045] The design and manufacturing requirements of the battery cover plate explosion-proof valve are analyzed, which will affect the quality standards and tolerance range of welding, wherein the welding defect tolerance refers to the maximum range of defects allowed to exist during welding, for example, some small cracks or pores may be acceptable in some cases, but beyond a certain threshold, this tolerance is closely related to the function and safety requirements of the explosion-proof valve; the explosion-proof valve welding strength constraint refers to the minimum strength requirement that the welding area must have, usually referring to the load-bearing capacity of the welding joint under load or pressure, the strength requirement depends on the working environment and use load of the explosion-proof valve; the explosion-proof performance of the explosion-proof valve is a very important welding requirement, which must ensure the sealing and strength of the joint during welding to prevent rupture or leakage in high pressure or high temperature environment. Through the analysis of the above welding requirements, the explosion-proof valve welding requirement information is obtained.

[0046] According to the explosion-proof valve welding requirement information, the welding standard sample images are screened, specifically, images within the defect tolerance range are selected from existing standard samples, for example, if some images contain small defects that are tolerated, these images can be used as reference samples; ensure that the selected sample images represent welding areas with sufficient strength to meet the strength requirements of the explosion-proof valve; select standard sample images that meet the explosion-proof performance requirements to ensure that the welding areas of these images have the required sealing and strength. After screening, the final standard welding images that meet all the welding requirement standards are retained, which will be used as the standard template for subsequent defect detection.

[0047] Further, the method further comprises:

[0048] The explosion-proof valve basic attribute index is obtained, wherein the explosion-proof valve basic attribute index includes material and mechanical properties, size and geometric specifications, working safety performance, and welding manufacturing process; the explosion-proof valve basic attribute index is analyzed in detail to establish an explosion-proof valve basic attribute node set; the explosion-proof valve attribute knowledge graph is constructed based on the explosion-proof valve basic attribute node set; and the explosion-proof valve attribute information is obtained by attribute traversal matching of the battery cover plate explosion-proof valve based on the explosion-proof valve attribute knowledge graph.

[0049] Obtaining the basic attribute indicators of the explosion-proof valve, which cover the physical, mechanical and technical characteristics of the explosion-proof valve, can help subsequent defect detection and analysis. The material and mechanical properties include the types of materials used in the explosion-proof valve, such as steel, stainless steel, alloy, etc., as well as strength, hardness, ductility, toughness, etc. These characteristics directly affect the welding strength and defect tolerance during the welding process. The size and geometric specifications include the shape size, thickness, shape of the welding joint, size of the weld, etc. The size and geometric specifications have important influence on the difficulty of welding, defect type and defect distribution. The working safety performance includes the requirements of the explosion-proof valve during use, such as pressure, temperature, gas tightness, etc. These requirements determine the safety standards of the welding area and the control requirements of the welding quality. The welding manufacturing process includes the technical requirements, welding method, welding material, heat treatment process, etc. These factors directly affect the welding quality and the generation of defects.

[0050] According to the obtained basic attribute information of the explosion-proof valve, the information is further refined into multiple specific attributes, such as material performance, which can be further divided into specific performance indicators such as tensile strength, elongation, hardness, etc. These refined attributes are converted into a set of explosion-proof valve basic attribute nodes, each node representing a specific attribute or attribute value, such as tensile strength, elongation, hardness, etc. In addition to individual attribute nodes, relationships between nodes can also be defined, such as the association of material attributes with welding methods, and the association of size specifications with welding strength requirements.

[0051] Using a graph database or other methods, each attribute node and its refined content of the explosion-proof valve basic attribute node set are organized into a knowledge graph. The knowledge graph is a graph structure constructed by nodes and edges, where each node represents a specific attribute or attribute value of the explosion-proof valve, and the edges represent the relationship between attributes, such as the relationship between the tensile strength of a certain material and the selection of welding process. This relationship is established in the graph. Through the construction of the graph, each attribute node can be linked to other related nodes through the relationship, for example, the welding process, tensile strength, etc. of the material of the explosion-proof valve can be queried according to the material node, which helps to analyze the welding requirements related to the material.

[0052] Using the explosion-proof valve attribute knowledge graph, based on the design, manufacturing and working requirements of the explosion-proof valve, attribute matching is performed, for example, given a material type of an explosion-proof valve, the graph can be queried to obtain the welding strength requirement, heat treatment process, defect tolerance, etc. of the material. By traversing the graph, the explosion-proof valve attribute information is obtained, which will provide detailed reference for subsequent welding defect detection.

[0053] Further, the method for acquiring the welding position image based on the welding position information of the explosion-proof valve comprises:

[0054] The welding standard sample image is subjected to grayscale processing and grayscale distribution identification to obtain a welding position grayscale interval and an image background grayscale interval; an attention constraint mechanism is constructed based on the welding position grayscale interval and the image background grayscale interval; a welding position global image is acquired based on the anti-explosion valve welding position information for global image acquisition; and the welding position global image is processed based on the attention constraint mechanism to obtain the welding position image.

[0055] The welding standard sample image is subjected to grayscale processing, i.e., the welding standard sample image is converted from a color image to a grayscale image, because the multi-dimensional information (red, green, and blue) contained in the color image is usually not as intuitive and effective as the grayscale information for positioning the welding position. The grayscale processing can reduce the computational complexity and highlight the shape and brightness information of the image. The grayscale image is obtained by converting each pixel value of the image to a grayscale value, usually using a weighted method to calculate the average of the RGB values.

[0056] The grayscale distribution in the grayscale image is analyzed to find the grayscale range of the welding area and the grayscale range of the image background. Generally, the welding area and the background have a significant difference in grayscale values, with the welding area having brighter or darker grayscale values, and the background area having relatively uniform grayscale values. The welding position grayscale interval (i.e., the grayscale value range of the welding area) and the image background grayscale interval (i.e., the grayscale value range of the background area) are found through histogram analysis or other distribution identification algorithms.

[0057] The attention mechanism is a technique that highlights the most valuable areas for a task by assigning different weights to different parts of an image. In welding defect detection, the system is expected to focus on the welding area while ignoring the background area. According to the identified grayscale intervals, the image regions are assigned different attention weights based on grayscale values. The welding position grayscale interval is assigned a higher attention weight, while the image background grayscale interval is assigned a lower weight, which means that during image processing and feature extraction, the welding position area is prioritized and the background area is suppressed. This attention constraint mechanism adds a constraint layer that focuses on the welding area and reduces the impact of the background area by modifying the weight distribution of the image, which allows more accurate extraction of relevant features of the welding position from the image.

[0058] The welding position global image acquired by the camera or image acquisition device contains all the welding areas using the anti-explosion valve welding position information. The welding position global image is usually a larger field of view image that covers multiple welding points of the entire anti-explosion valve. This welding position global image provides an image containing complete welding positions and backgrounds for subsequent processing, ensuring that the system can perform welding defect detection within the entire anti-explosion valve range.

[0059] The attention constraint mechanism is applied to the welding position global image. According to the gray interval of the welding position and the gray interval of the background, the attention is focused on the welding position area, the saliency of the welding position area is enhanced by weighting the pixels of the image, and the influence of the background area is reduced. This can ensure that the subsequent image processing is more accurately focused on the welding area. After applying the attention constraint mechanism, an image focused on the welding area is generated, which is called the welding position image. The background interference in this image has been suppressed, and the details of the welding area have been highlighted.

[0060] Further, the method further comprises:

[0061] Based on the defect type recognition result, the defect statistics are performed to obtain a defect data set. The defect position and defect feature common clustering are performed on the defect data set to generate a defect common clustering result. The welding feedback information is generated based on the defect common clustering result.

[0062] From the previous defect detection process, various defect types in the welding area have been identified, such as cracks, pores, uneven welding, etc. Based on these defect type recognition results, defect statistics are performed, including the frequency of each defect type, and recording the specific position, size, severity, etc. of each defect. The statistical defect information is integrated into a defect data set, which contains detailed data of each defect type, such as position, occurrence frequency, impact degree, etc.

[0063] Use clustering algorithms such as K-means, hierarchical clustering, etc. to perform common clustering on the defect data set, including: according to the position of the defect in the welding area, such as near the edge, center, etc., clustering, if multiple defects appear in the same area or similar area, they belong to the same class; according to the characteristics of the defect, such as shape, size, type, etc., clustering, similar defects will be clustered together, such as size, shape similar pores or cracks. Through clustering analysis, the defect common clustering result is output, where each class represents a group of defects with similar characteristics or positions. This result can help further analyze the patterns or problems that may exist in the welding process, such as frequent occurrence of defects in certain specific areas, or certain welding methods are prone to cause specific types of defects.

[0064] Based on the defect commonness clustering results, analyze which defects frequently occur at the same location or under similar features, and for these defects, further investigate possible causes such as welding parameters, operation process, material problems, etc. According to the analysis results, generate welding feedback information with strong pertinence, including: if some areas frequently have defects, the operator can be prompted to pay attention to these areas, such as increasing the welding quality inspection of this area, adjusting the welding parameters, etc.; if a certain type of defect (such as porosity or crack) frequently occurs, it can be suggested to change the welding method, use different welding materials or carry out more strict control. These welding feedback information will be delivered to the welding process personnel, operators or production line managers, helping them to take action to reduce the occurrence of defects, thereby improving the welding quality and product qualification rate.

[0065] Further, the method further comprises:

[0066] Based on the defect type identification result, the welding defect of the battery cover plate explosion-proof valve is shaped, and the explosion-proof valve product is obtained, wherein the explosion-proof valve product includes qualified explosion-proof valve product and unqualified explosion-proof valve product; the product sorting equipment is started, wherein the product sorting equipment includes sorting manipulator, first conveying belt and second conveying belt; the qualified explosion-proof valve product is sorted to the first conveying belt by the sorting manipulator and flows into the next process; the unqualified explosion-proof valve product is sorted to the second conveying belt by the sorting manipulator and flows into the recycling box.

[0067] According to the defect type identification result recognized in the foregoing, the defect of the battery cover plate explosion-proof valve is repaired or shaped, for example, after finding defects such as cracks and porosity, the problems are repaired, the welding quality is adjusted, and according to the severity and position of the defect type, the explosion-proof valve product is divided into qualified explosion-proof valve product and unqualified explosion-proof valve product, wherein the qualified explosion-proof valve product meets the welding quality requirement, the welding defect is within the tolerable range, and is suitable for subsequent processes and shipment; the unqualified explosion-proof valve product has serious defects, which may affect the safety and performance of the product, and cannot pass the quality standard, and needs to be recycled or repaired.

[0068] The product sorting equipment is started, wherein the sorting manipulator is used to grab the explosion-proof valve product and send it to the corresponding conveying belt; the first conveying belt is used to convey the qualified explosion-proof valve product and flow into the next process, such as packaging, shipment, etc.; the second conveying belt is used to convey the unqualified explosion-proof valve product and send it into the recycling box or repair area. According to the defect type identification result, the product sorting equipment is automatically started, and the sorting manipulator decides which conveying belt it will be sent to according to whether the product is qualified, so as to realize the automatic sorting process.

[0069] The sorting robot sorts out the qualified explosion-proof valve products identified as qualified from the production line and accurately sends them to the first conveying belt. The qualified explosion-proof valve products flow through the first conveying belt to the next production process, such as packaging, quality inspection, and factory delivery. This step ensures that only products meeting the quality requirements can continue to be processed.

[0070] The sorting robot sorts out the unqualified explosion-proof valve products identified as unqualified from the production line and accurately sends them to the second conveying belt. The unqualified explosion-proof valve products flow through the second conveying belt to the recycling box. The products in the recycling box will be repaired, reprocessed, or finally destroyed to ensure that unqualified products do not affect the quality of the next process or the final product.

[0071] In summary, the intelligent welding defect detection method for the battery cover plate explosion-proof valve provided by the embodiments of the present application has the following technical effects:

[0072] By collecting the explosion-proof valve basic information, including the explosion-proof valve attribute information and the explosion-proof valve welding position information, the intelligent welding system can perform subsequent image acquisition based on accurate welding position information, which helps to accurately locate the welding area to be detected and avoids ineffective or redundant image acquisition, improving efficiency. The explosion-proof valve attribute information enables the system to consider specific welding requirements such as material, geometric size, and process requirements, thereby ensuring that the welding defect detection can be analyzed according to specific product standards; by performing global image acquisition based on the explosion-proof valve welding position information, the complete image of the welding position is ensured to be obtained, avoiding missing critical welding areas. This global acquisition provides more comprehensive context information for subsequent welding defect detection, avoiding the limitations that may occur during local image acquisition; feature extraction and comparison of the welding position image accurately locates the welding defect area. The intelligent welding system focuses on the defect area for subsequent local image acquisition through this accurate defect positioning, improving the efficiency and accuracy of image acquisition; based on the explosion-proof valve attribute information, the negative sampling is performed to obtain a welding defect sample image set. Through negative samples, normal and abnormal welding areas can be better distinguished, thereby reducing false positives and improving the accuracy of defect identification; based on the welding defect sample image set, the welding defect area image is compared for defects. The intelligent welding system can identify known defect types. Through this comparative analysis, it can automatically determine whether the welding area meets the standard and accurately identify the defect type. This method not only improves the accuracy of defect identification but also handles different types of defects, making the intelligent welding system adaptable to complex and variable welding quality control requirements.

[0073] Embodiment two, based on the same inventive concept as the intelligent welding defect detection method for the battery cover plate explosion-proof valve in the preceding embodiments, such as Figure 2As shown, the battery cover plate explosion-proof valve intelligent welding defect detection device provided by the embodiment of the application comprises:

[0074] The base information acquisition module 10 is configured to acquire base information of the explosion-proof valve of the battery cover plate explosion-proof valve, wherein the base information of the explosion-proof valve comprises attribute information of the explosion-proof valve and welding position information of the explosion-proof valve. The global image acquisition module 20 is configured to acquire a welding position image based on the welding position information of the explosion-proof valve. The local image acquisition module 30 is configured to perform feature extraction and comparison on the welding position image, locate a welding defect area, acquire a welding defect area image based on the welding defect area, and perform negative sampling based on the attribute information of the explosion-proof valve to acquire a welding defect sample image set, wherein the welding defect sample image set is marked with a plurality of known defect types.

[0075] Further, the defect comparison module 50 is configured to perform the following operation steps:

[0076] The first welding defect sample image set of the first known defect type in the plurality of known defect types is extracted, and a first defect comparison group set of the first welding defect sample image set and the welding defect area image is established. The first defect similarity is acquired based on the similarity twin network. When the first defect similarity satisfies a preset defect similarity, the first known defect type is added to the defect type identification result.

[0077] Further, the defect comparison module 50 is configured to perform the following operation steps:

[0078] The similarity twin network is established, wherein the similarity twin network comprises a feature mapping layer, and the features of the feature mapping layer comprise edge feature overlap, texture feature overlap, color feature overlap, and local shape feature overlap. The edge feature overlap, the texture feature overlap, the color feature overlap, and the local shape feature overlap are calculated according to a weight sharing network, and the first defect similarity is output.

[0079] Further, the local image acquisition module 30 is configured to perform the following operation steps:

[0080] Based on the explosion-proof valve attribute information, positive sampling is performed to obtain a welding standard sample image; a feature extraction comparator is constructed based on a convolutional neural network, wherein the feature extraction comparator includes a first feature extraction channel and a second feature extraction channel sharing weights; the welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel for welding feature extraction and loss comparison analysis to obtain a welding feature loss data set; and based on the welding feature loss data set, a welding defect area is located in the welding position image.

[0081] Further, the local image acquisition module 30 is configured to perform the following operation steps:

[0082] The explosion-proof valve of the battery cover plate is subjected to welding requirement analysis to obtain welding requirement information of the explosion-proof valve, wherein the welding requirement information of the explosion-proof valve includes welding defect tolerance, welding strength constraint and explosion-proof performance constraint; and the welding standard sample image is subjected to requirement adaptation screening based on the welding requirement information of the explosion-proof valve.

[0083] Further, the basic information acquisition module 10 is configured to perform the following operation steps:

[0084] The basic attribute indexes of the explosion-proof valve are acquired, wherein the basic attribute indexes of the explosion-proof valve include material and mechanical performance, size and geometric specification, working safety performance and welding manufacturing process; attribute index refinement analysis is performed on the basic attribute indexes of the explosion-proof valve to establish a basic attribute node set of the explosion-proof valve; an attribute knowledge graph of the explosion-proof valve is constructed based on the basic attribute node set of the explosion-proof valve; and attribute traversal matching is performed on the explosion-proof valve of the battery cover plate based on the attribute knowledge graph of the explosion-proof valve to obtain the attribute information of the explosion-proof valve.

[0085] Further, the global image acquisition module 20 is configured to perform the following operation steps:

[0086] The welding standard sample image is subjected to grayscale processing and grayscale distribution recognition to obtain a welding position grayscale interval and an image background grayscale interval; an attention constraint mechanism is constructed based on the welding position grayscale interval and the image background grayscale interval; a global image of a welding position is acquired based on the welding position information of the explosion-proof valve; and the welding position image is obtained by processing the global image of the welding position based on the attention constraint mechanism.

[0087] Further, the welding feedback information generation module is further configured to perform the following operation steps:

[0088] Based on the defect type recognition result, defect statistics are performed to obtain a defect dataset; common clustering of defect positions and defect features is performed on the defect dataset to generate a defect common clustering result; and welding feedback information is generated based on the defect common clustering result.

[0089] Further, a product sorting module is further included, configured to perform the following operation steps:

[0090] Based on the defect type recognition result, welding defect shaping of the battery cover plate explosion-proof valve is performed to obtain an explosion-proof valve product, wherein the explosion-proof valve product includes qualified explosion-proof valve products and unqualified explosion-proof valve products; and a product sorting device is started, wherein the product sorting device includes a sorting manipulator, a first conveying belt and a second conveying belt; the qualified explosion-proof valve products are sorted to the first conveying belt by the sorting manipulator and flow into a next process; and the unqualified explosion-proof valve products are sorted to the second conveying belt by the sorting manipulator and flow into a recycling box.

[0091] Through the foregoing detailed description of the intelligent welding defect detection method of the battery cover plate explosion-proof valve, those skilled in the art can clearly know the intelligent welding defect detection device of the battery cover plate explosion-proof valve in the embodiment, and since the device corresponds to the method disclosed in the embodiment, the device is described relatively simply, and the related parts can be referred to the method part description.

[0092] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent welding defect detection method for battery cover explosion-proof valves, characterized in that, The method includes: Collect the basic information of the explosion-proof valve of the battery cover explosion-proof valve, wherein the basic information of the explosion-proof valve includes explosion-proof valve attribute information and explosion-proof valve welding position information; Global image acquisition is performed based on the welding position information of the explosion-proof valve to obtain the welding position image; Feature extraction and comparison are performed on the welding position image to locate the welding defect area. Based on the welding defect area, local image acquisition is performed to obtain the welding defect area image. Negative sampling is performed based on the explosion-proof valve attribute information to obtain a set of welding defect sample images, wherein the set of welding defect sample images identifies multiple known defect types; Based on the set of welding defect sample images, the images of the welding defect areas are compared to obtain the defect type identification results corresponding to the multiple known defect types.

2. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 1, characterized in that, The method for comparing the weld defect region images based on the set of weld defect sample images to obtain defect type identification results corresponding to the multiple known defect types includes: Extract a set of first welding defect sample images of the first known defect type from the plurality of known defect types, and establish a first defect comparison set between the first set of first welding defect sample images and the welding defect region images; Based on the similarity Siamese network, the first defect comparison set is used to identify the defect similarity and obtain the first defect similarity. When the first defect similarity meets the preset defect similarity, the first known defect type is added to the defect type identification result; This process continues until all known defect types have been traversed.

3. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 2, characterized in that, The method includes: The similarity twin network is established, wherein the similarity twin network includes a feature mapping layer, and the features of the feature mapping layer include edge feature overlap, texture feature overlap, color feature overlap, and local shape feature overlap; The overlap of edge features, texture features, color features, and local shape features is calculated using a weighted sharing network, and the first defect similarity is output.

4. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 1, characterized in that, The method for extracting and comparing features from the welding position image to locate the welding defect area includes: Based on the explosion-proof valve attribute information, positive sampling is performed to obtain welding standard sample images; A feature extraction and comparison device is constructed based on a convolutional neural network, wherein the feature extraction and comparison device includes a first feature extraction channel and a second feature extraction channel with shared weights; The welding standard sample image and the welding position image are respectively input into the first feature extraction channel and the second feature extraction channel to perform welding feature extraction and loss comparison analysis to obtain a welding feature loss dataset. Based on the welding feature loss dataset, the welding defect region is located in the welding position image.

5. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 4, characterized in that, The method also includes: Welding requirements analysis was performed on the explosion-proof valve of the battery cover to obtain the welding requirements information of the explosion-proof valve, wherein the welding requirements information of the explosion-proof valve includes welding defect tolerance, welding strength constraints and explosion-proof performance constraints. Based on the welding requirements information of the explosion-proof valve, the welding standard sample images are screened to meet the requirements.

6. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 1, characterized in that, The method for collecting the basic information of the explosion-proof valve of the battery cover includes: Obtain the basic attribute indicators of the explosion-proof valve, wherein the basic attribute indicators of the explosion-proof valve include material and mechanical properties, size and geometric specifications, working safety performance and welding manufacturing process; The basic attribute indicators of the explosion-proof valve are analyzed in detail to establish a set of basic attribute nodes for the explosion-proof valve. Construct an explosion-proof valve attribute knowledge graph based on the set of basic attribute nodes of the explosion-proof valve; Based on the explosion-proof valve attribute knowledge graph, attribute traversal and matching are performed on the battery cover explosion-proof valve to obtain the explosion-proof valve attribute information.

7. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 1, characterized in that, The method for acquiring a welding position image based on the welding position information of the explosion-proof valve includes: The standard welding sample images are processed to grayscale and grayscale distribution is identified to obtain the grayscale range of the welding position and the grayscale range of the image background. An attention constraint mechanism is constructed based on the grayscale range of the welding position and the grayscale range of the image background. Global image acquisition is performed based on the welding position information of the explosion-proof valve to obtain a global image of the welding position; The global image of the welding position is processed based on the attention constraint mechanism to obtain the welding position image.

8. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 1, characterized in that, The method further includes: Based on the defect type identification results, defect statistics are performed to obtain a defect dataset; Perform common clustering of defect locations and defect features on the defect dataset to generate common clustering results for defects; Welding feedback information is generated based on the clustering results of the common defects.

9. The intelligent welding defect detection method for the battery cover explosion-proof valve as described in claim 1, characterized in that, The method further includes: Based on the defect type identification results, welding defects of the battery cover explosion-proof valve are shaped to obtain explosion-proof valve products, wherein the explosion-proof valve products include qualified explosion-proof valve products and unqualified explosion-proof valve products. Start the product sorting equipment, wherein the product sorting equipment includes a sorting robot, a first conveyor belt, and a second conveyor belt; The qualified explosion-proof valve products are sorted onto the first conveyor belt by the sorting robot and flow into the next process. The sorting robot sorts the defective explosion-proof valve products onto the second conveyor belt, where they flow into the recycling bin.

10. An intelligent welding defect detection device for battery cover explosion-proof valves, characterized in that, An intelligent welding defect detection method for implementing the battery cover explosion-proof valve according to any one of claims 1-9, the device comprising: The basic information acquisition module is used to acquire the basic information of the explosion-proof valve of the battery cover explosion-proof valve, wherein the basic information of the explosion-proof valve includes explosion-proof valve attribute information and explosion-proof valve welding position information; The global image acquisition module is used to acquire global images based on the welding position information of the explosion-proof valve, and obtain welding position images. The local image acquisition module is used to extract and compare features from the welding position image, locate the welding defect area, and acquire local images based on the welding defect area to obtain the welding defect area image. The negative sampling module is used to perform negative sampling based on the explosion-proof valve attribute information to obtain a set of welding defect sample images, wherein the set of welding defect sample images identifies multiple known defect types; The defect comparison module is used to compare the welding defect area image based on the welding defect sample image set, and obtain the defect type identification results corresponding to the multiple known defect types.

Images