Assembly welding positioning method and device for automobile parts

Abstract

The present application relates to the technical field of assembly welding, and specifically discloses a method and device for assembling welding positioning of automobile parts, which uses a camera to collect the welding state image between the target muffler and the target partition, uses the image recognition and analysis algorithm based on computer vision to perform welding state mask segmentation on the welding state image, and then extracts the welding state features of the segmented welding detection area state mask image, so as to realize intelligent detection of whether there are defects in the welding state between the target muffler and the target partition by optimizing the neighborhood importance according to the extracted welding detection state features. In this way, tiny defects such as fine cracks and internal pores can be captured, thereby improving the accuracy of defect identification. And the detection speed is greatly improved, while maintaining consistency and stability, reducing the probability of missed detection or false detection, to ensure that each welding point is strictly quality controlled.

Description

Technical Field

[0001] The present application relates to the technical field of assembly welding, and more specifically, to an assembly welding positioning method and device for automobile parts. Background Art

[0002] In the field of automobile manufacturing, the welding quality of the muffler and the baffle is crucial to the safety and performance of the car. With the upgrading of the automobile industry, the market's requirements for product quality and production efficiency are constantly increasing. In the traditional automobile parts assembly and welding process, the welding of parts such as mufflers and baffles usually relies on manual experience and a relatively extensive positioning method. This not only leads to low welding efficiency, but also makes it difficult to ensure the position accuracy and consistency of each welding. Once the welding position deviates, it will affect the silencing effect of the muffler and may even cause the entire component structure to be unstable.

[0003] In this regard, patent CN113182717B proposes an automobile silencer assembly welding device and processing technology, which first installs and fixes three partitions at a predetermined interval on a dedicated clamping mechanism, inserts the first, second and third silencer pipes into the fixed partitions in turn, and ensures that each silencer pipe is correctly positioned and firmly immobilized through the abutment mechanism, and then starts the manipulator to perform welding operations on the contact points between each silencer pipe and the partition until all welding points are completed. In this way, the use of the abutment mechanism and the manipulator for automated positioning and welding can ensure the accurate position of the welding point, making the connection between the silencer pipe and the partition more secure, thereby avoiding the instability of manual positioning.

[0004] After welding is completed, it is necessary to conduct welding quality inspection on each welding part of the silencer to ensure that there are no defects in the welding part of the silencer, maintain its structural stability and sealing, and ensure the silencer effect. However, traditional welding quality inspection is mainly based on manual visual inspection, supplemented by simple measuring tools. This inspection method is not only highly subjective, but also difficult to accurately identify minor welding defects such as fine cracks and internal pores. In addition, manual inspection is inefficient and cannot meet the rhythm of large-scale production, which is prone to missed inspections or false inspections.

[0005] Therefore, an optimized assembly welding positioning solution for automotive parts is desired. Summary of the invention

[0006] The present application provides an assembly welding positioning method and device for automobile parts, which can capture tiny defects such as fine cracks and internal pores, thereby improving the accuracy of defect identification. It also greatly improves the detection speed, while maintaining consistency and stability, reducing the probability of missed detection or false detection, to ensure that each welding point is strictly quality controlled.

[0007] In a first aspect, a method for assembling, welding and positioning automobile parts is provided, comprising: placing three partitions in a clamping mechanism and arranging the three partitions at a predetermined interval; respectively passing a first muffler pipe, a second muffler pipe and a third muffler pipe through the three partitions, and respectively positioning and abutting and fixing the three muffler pipes by using an abutting mechanism; starting a manipulator to respectively weld the first muffler pipe, the second muffler pipe and the third muffler pipe to the three partitions to obtain an automobile muffler package; wherein the method for assembling, welding and positioning automobile parts further comprises: performing welding quality inspection on the automobile muffler package;

[0008] The welding quality inspection of the automobile silencer package includes:

[0009] The welding state image between the target muffler pipe and the target partition is collected by a camera;

[0010] Performing welding state mask segmentation on the welding state image to obtain a welding detection area state mask image;

[0011] Extracting welding state features from the welding detection area state mask image to obtain a welding detection state feature map;

[0012] The welding detection state feature map is subjected to feature optimization based on neighborhood importance to obtain the welding detection state optimized feature map, including: performing feature decoupling and feature flattening on the welding detection state feature map to obtain a set of welding detection state local feature vectors; arranging the set of welding detection state local feature vectors in descending order to obtain a descending sequence of welding detection state local feature vectors; and performing feature filtering and reshaping on the descending sequence of welding detection state local feature vectors to obtain the welding detection state optimized feature map;

[0013] Based on the welding detection state optimization characteristic diagram, a detection result is obtained for indicating whether there is a defect in the welding state between the target muffler pipe and the target partition plate.

[0014] In a second aspect, a device for assembling and welding automobile parts is provided, comprising:

[0015] A welding state image acquisition module, used for acquiring a welding state image between a target muffler pipe and a target partition through a camera;

[0016] A welding state mask segmentation module, used for performing welding state mask segmentation on the welding state image to obtain a welding detection area state mask image;

[0017] A welding state feature extraction module, used for performing welding state feature extraction on the welding detection area state mask image to obtain a welding detection state feature map;

[0018] A welding detection state optimization module, used for performing feature optimization based on neighborhood importance on the welding detection state feature map to obtain a welding detection state optimized feature map, comprising: a welding detection state feature flattening unit, used for performing feature decoupling and feature flattening on the welding detection state feature map to obtain a set of welding detection state local feature vectors; a descending order arrangement unit, used for performing descending order arrangement on the set of welding detection state local feature vectors to obtain a descending sequence of welding detection state local feature vectors; a feature filtering unit, used for performing feature filtering and reshaping on the descending sequence of the welding detection state local feature vectors to obtain the welding detection state optimized feature map;

[0019] The welding detection result determination unit is used to obtain a detection result indicating whether there is a defect in the welding state between the target muffler pipe and the target partition plate based on the welding detection state optimization characteristic diagram.

[0020] The present application provides a method and device for assembling welding positioning of automobile parts, which uses a camera to collect the welding state image between the target muffler and the target partition, uses the image recognition and analysis algorithm based on computer vision to perform welding state mask segmentation on the welding state image, and then extracts the welding state features of the segmented welding detection area state mask image, so as to realize intelligent detection of whether there are defects in the welding state between the target muffler and the target partition by optimizing the neighborhood importance according to the extracted welding detection state features. Compared with traditional manual visual inspection, the present application can capture tiny defects such as fine cracks and internal pores, thereby improving the accuracy of defect identification. And it greatly improves the detection speed, while maintaining consistency and stability, reducing the probability of missed detection or false detection, to ensure that each welding point is strictly quality controlled. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments of the present application are briefly introduced below. Obviously, the drawings described below only relate to some embodiments of the present application, and are not intended to limit the present application.

[0022] Figure 1 This is a schematic flow chart of an assembly welding positioning method for automobile parts according to an embodiment of the present application.

[0023] Figure 2 This is a schematic flow chart of S4 in the assembly welding positioning method of automobile parts according to an embodiment of the present application.

[0024] Figure 3 This is a schematic flow chart of S44 in the assembly welding positioning method of automobile parts according to an embodiment of the present application.

[0025] Figure 4 This is a schematic flow chart of S443 in the assembly welding positioning method of automobile parts according to an embodiment of the present application.

[0026] Figure 5 This is a schematic block diagram of an assembly welding positioning device for automobile parts according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present application.

[0028] Based on patent CN113182717B, this application proposes a method for assembling and welding positioning of automobile parts, such as Figure 1 As shown, the assembly welding positioning method of the automobile parts comprises: S1, placing three partitions in a clamping mechanism, and arranging the three partitions at a predetermined interval; S2, respectively passing the first silencer pipe, the second silencer pipe and the third silencer pipe through the three partitions, and respectively positioning and abutting and fixing the three silencer pipes by using abutment mechanisms; S3, starting a manipulator to weld the first silencer pipe, the second silencer pipe and the third silencer pipe to the three partitions respectively to obtain an automobile silencer package; S4, performing welding quality inspection on the automobile silencer package.

[0029] Exemplarily, in step S1, three partitions are placed in a clamping mechanism, and the three partitions are arranged at predetermined intervals. It should be understood that during the assembly and welding process of the automobile silencer package, the main purpose of placing the three partitions in the clamping mechanism and arranging them at predetermined intervals is to ensure the precise positioning and stability of the internal structure of the silencer package. Specifically, this is done to reduce the possibility of heat deformation of parts during the welding process, thereby improving the product qualification rate and production efficiency. By using a clamping mechanism to clamp and fix the partitions, the distance between the partitions can be accurately controlled to ensure the consistency and accuracy of subsequent welding operations.

[0030] Exemplarily, in step S2, the first silencer pipe, the second silencer pipe and the third silencer pipe are respectively passed through the three partitions, and the three silencer pipes are respectively positioned and abutted and fixed by abutment mechanisms. It should be understood that the first silencer pipe, the second silencer pipe and the third silencer pipe are respectively passed through the three partitions, and the silencer pipes are positioned and abutted and fixed by abutment mechanisms in order to ensure the precise position and stability of the various components of the silencer package during the welding process. This process is crucial to ensuring the quality of welding, because only when the relative position between the silencer pipe and the partition is accurately fixed can the displacement or misalignment caused by thermal deformation or other factors during the welding process be avoided, thereby improving the product qualification rate and production efficiency.

[0031] Exemplarily, in step S3, the manipulator is started to weld the first muffler pipe, the second muffler pipe and the third muffler pipe to the three partitions respectively to obtain the automobile muffler package. It should be understood that starting the manipulator to weld the first muffler pipe, the second muffler pipe and the third muffler pipe to the three partitions respectively to obtain the automobile muffler package is a key step to ensure the assembly quality of the muffler package. The main purpose of using a manipulator for welding is to improve welding efficiency and accuracy, while reducing the uncertainty caused by manual operation, thereby significantly improving the product qualification rate and production efficiency.

[0032] Exemplarily, in step S4, the welding quality of the automobile silencer package is inspected. It should be understood that the welding quality inspection of the automobile silencer package is to ensure that each welding point is strictly quality controlled, so as to ensure the performance and safety of the final product. In the field of automobile manufacturing, the welding quality of the silencer pipe and the partition is directly related to the noise reduction effect, structural stability and overall safety of the vehicle. The traditional manual visual inspection method has the problems of strong subjectivity and difficulty in identifying subtle defects (such as fine cracks, internal pores, etc.), and is inefficient and prone to missed detection or false detection. Therefore, the introduction of advanced automated detection technology has become the key to improving product quality and production efficiency.

[0033] Accordingly, in the welding quality inspection of the automobile silencer package, the technical concept of the present application is to collect the welding state image between the target silencer pipe and the target partition through a camera, use the image recognition and analysis algorithm based on computer vision to perform welding state mask segmentation on the welding state image, and then extract the welding state features of the segmented welding detection area state mask image, so as to realize the intelligent detection of whether there are defects in the welding state between the target silencer pipe and the target partition by optimizing the neighborhood importance according to the extracted welding detection state features. Compared with traditional manual visual inspection, the present application can capture tiny defects such as fine cracks and internal pores, thereby improving the accuracy of defect identification. And it greatly improves the detection speed, while maintaining consistency and stability, reducing the probability of missed detection or false detection, to ensure that each welding point is strictly quality controlled.

[0034] In one embodiment, Figure 2 As shown, the welding quality inspection of the automobile silencer package includes: S41, collecting the welding state image between the target silencer pipe and the target partition through a camera; S42, performing welding state mask segmentation on the welding state image to obtain a welding detection area state mask image; S43, performing welding state feature extraction on the welding detection area state mask image to obtain a welding detection state feature map; S44, performing feature optimization based on neighborhood importance on the welding detection state feature map to obtain a welding detection state optimized feature map; S45, based on the welding detection state optimized feature map, obtaining a detection result for indicating whether there is a defect in the welding state between the target silencer pipe and the target partition.

[0035] Exemplarily, in step S41, a welding state image between the target muffler and the target partition is collected by a camera. It should be understood that the purpose of collecting the welding state image between the target muffler and the target partition by a camera is to achieve accurate and efficient detection of welding quality. In one embodiment, the camera is installed in a fixed position to ensure that its viewing angle covers the entire welding area and can quickly capture high-quality images after welding is completed. In order to obtain the best image quality, the camera is usually equipped with an appropriate light source to ensure the clarity and contrast of the welding area. At the same time, the camera is connected to the control system, and the shooting time can be set by programming to ensure that the image is collected at the most appropriate time. The collected image will be transmitted to the image processing unit in real time, and after preprocessing, it will be analyzed by a special algorithm. In this way, not only can the welding quality be monitored in real time, but also the test results can be fed back to the production line, and the welding parameters can be adjusted in time to prevent unqualified products from flowing into the next process.

[0036] Exemplarily, in step S42, the welding state image is subjected to welding state mask segmentation to obtain a welding detection area state mask image. It should be understandable. Considering that in the entire welding state image, not all parts are directly related to welding quality detection, which may contain complex background information, such as the shape and color of the muffler and partition, and the surrounding environment. These background information may interfere with the judgment of the quality of the welding area. Therefore, in order to be able to focus on the area related to the welding part, that is, to focus on the weld itself and eliminate unnecessary background interference, in the technical solution of the present application, the welding state image is passed through a welding state mask segmentor based on a mask area convolutional neural network to obtain a welding detection area state mask image.

[0037] That is, in one embodiment, the welding state image is subjected to welding state mask segmentation to obtain the welding detection area state mask image, including: passing the welding state image through a welding state mask segmenter based on a mask area convolutional neural network to obtain the welding detection area state mask image. It should be understood that the mask area convolutional neural network is a deep learning model that extends the commonly used Faster R-CNN target detection framework and increases the pixel-level segmentation capability. This means that in addition to being able to identify and locate objects in an image, a compact segmentation mask for each object can also be generated, that is, each pixel in the image is marked as belonging to a specific object. In this way, the weld area and the background can be accurately separated, thereby providing more accurate data for subsequent analysis. Specifically, the mask area convolutional neural network first performs target detection like Faster R-CNN to find the area of ​​interest that may exist in the image, such as the welding area. For each detected target (i.e., welding area), the mask area convolutional neural network generates a binary mask to represent the specific position of the target in the image. This mask is at the pixel level, which means it can define the boundaries of the welding area very accurately, providing data support for subsequent more accurate testing of welding quality.

[0038] Exemplarily, in step S43, the welding state feature extraction is performed on the welding detection area state mask image to obtain a welding detection state feature map. It should be understood that, considering that the welding detection area state mask image has been focused on the welding area, the feature extraction of the welding area still has a certain complexity. The shape, texture, fusion and other features of the weld are multi-dimensional and interrelated, and welding defects may appear at different scales, from tiny pores to larger cracks.

[0039] In one embodiment, the welding detection area state mask image is subjected to welding state feature extraction to obtain a welding detection state feature map, including: passing the welding detection area state mask image through a welding state feature extractor based on an expanded convolutional neural network model to obtain the welding detection state feature map. Based on this, in the technical solution of the present application, the welding detection area state mask image is passed through a welding state feature extractor based on an expanded convolutional neural network model to obtain a wider range of context information while maintaining a higher resolution, thereby obtaining a welding detection state feature map. That is, the expanded convolutional neural network model can effectively capture these complex features through its special convolution kernel structure. Compared with traditional convolutional neural networks, the expanded convolution can expand the receptive field without increasing the number of parameters, that is, it can obtain a wider range of image information. For example, for a longer crack defect in a weld, the expanded convolution can better capture its continuity features without losing part of the information due to the limited receptive field like ordinary convolution. In addition, the expanded convolution can better process multi-scale local detail features and global overall feature information, ensuring that various types of defects can be effectively identified to form a more comprehensive feature representation.

[0040] Exemplarily, in step S44, the welding detection state feature map is optimized based on the neighborhood importance to obtain the welding detection state optimized feature map. It should be understood that each feature point in the welding detection state feature map does not exist in isolation, but is interrelated with the feature points of the surrounding neighborhood. In the welding area, a feature change in a small area may imply a specific welding defect or quality condition. For example, a porosity defect may appear as an abnormal change in the feature value in a local area on the feature map, and these changes are not only reflected in the defect center, but also affect its surrounding neighborhood. Therefore, in order to capture the dependency between each local feature, so as to fully consider the impact of the features in the neighborhood on the target feature point, so as to more accurately describe the welding state, the present application performs feature optimization based on the neighborhood importance on the welding detection state feature map to obtain the welding detection state optimized feature map. In particular, the method uses the feature neighborhood activity to evaluate the importance and degree of interaction of the features in its surrounding area (i.e., the semantic neighborhood) in real time, thereby highlighting the key features and making the subsequent processing flow more efficient and accurate.

[0041] In one embodiment, Figure 3As shown, the feature optimization based on neighborhood importance is performed on the welding detection state feature map to obtain the welding detection state optimized feature map, including: S441, feature decoupling and feature flattening of the welding detection state feature map to obtain a set of welding detection state local feature vectors; S442, arranging the set of welding detection state local feature vectors in descending order to obtain a descending sequence of welding detection state local feature vectors; S443, feature filtering and reshaping the descending sequence of the welding detection state local feature vectors to obtain the welding detection state optimized feature map.

[0042] In one embodiment, feature decoupling and feature flattening are performed on the welding detection state feature map to obtain a set of local feature vectors of the welding detection state, including: feature decoupling the welding detection state feature map along the channel dimension of the welding detection state feature map to obtain a set of local feature matrices of the welding detection state. Specifically, the process can be expressed by a formula as follows:

[0043]

[0044] in, is the welding detection state characteristic diagram, is a feature decoupling operation, , , and is the first, second, and third local feature matrix of the welding detection state. and A local feature matrix of welding detection state.

[0045] Each welding detection state local feature matrix in the set of welding detection state local feature matrices is feature flattened to obtain the set of welding detection state local feature vectors. Specifically, the process can be expressed by the formula:

[0046]

[0047] in, is the feature flattening operation, , , and are the first, second, and third local feature vectors in the set of welding detection states. and The local feature vector of welding detection state.

[0048] Specifically, first, the feature decoupling of the welding detection state feature map is performed along the channel dimension to obtain a set of welding detection state local feature matrices. Here, the "channel dimension" refers to each channel in the feature map representing different types of feature information (for example, different convolutional layer outputs or different color channels). Through feature decoupling, these complex multidimensional information can be decomposed into multiple independent but related local feature matrices. Each local feature matrix represents a part of the welding area, and together they constitute a complete welding state feature representation. The effect of this decoupling operation is that it can separate different types of feature information so that each local feature matrix can more clearly reflect specific aspects, such as weld shape, texture, or fusion. This is very important for identifying minor defects such as fine cracks or internal pores, because these defects are often reflected in specific types of features. In this way, feature decoupling helps to improve the understanding and capture of welding area details. Next, each local feature matrix in the set of welding detection state local feature matrices is feature flattened and converted into a set of welding detection state local feature vectors. Flattening converts a two-dimensional local feature matrix into a one-dimensional feature vector. This ensures that the elements in each local feature vector represent different but related feature dimensions, thus forming a compact and effective feature expression. The flattened local feature vector not only retains the spatial structure information of the original feature matrix, but also facilitates subsequent algorithm processing.

[0049] In one embodiment, arranging the set of the welding detection state local feature vectors in descending order to obtain a descending sequence of the welding detection state local feature vectors includes: performing importance measurement on each welding detection state local feature vector in the set of the welding detection state local feature vectors to obtain a set of welding detection state local feature importance score values. Specifically, the process can be expressed by a formula as follows:

[0050]

[0051]

[0052] in, and They are The corresponding weight matrix and bias vector, is matrix multiplication, is the weight score vector, is the set of importance score values ​​of local features of welding inspection status, , , and They are the first, second, and third in the set of local feature importance scores of welding detection status. and Importance score of local features of welding inspection status;

[0053] Based on the set of importance score values ​​of the welding detection state local features, the set of the welding detection state local feature vectors is arranged in descending order to obtain a descending sequence of the welding detection state local feature vectors. Specifically, the process can be expressed by the formula:

[0054]

[0055] in, Based on The set of the welding detection state local feature matrices is arranged in descending order, , , and They are the first, second, and third in the descending sequence of the local feature vectors of the welding detection state. and The local feature vector of welding detection state.

[0056] Specifically, first, by calculating the importance score of the local feature vector of each welding inspection state, the system is able to quantify the contribution of each feature to the overall welding quality inspection task. They can evaluate the role of each feature vector in predicting welding defects. In this process, features that are better at capturing subtle changes such as fine cracks and internal pores will receive higher scores. The effect of the importance metric is that it not only identifies the features that are most critical to the task, but also provides a clear view of the relative importance of the features. This step helps to filter out redundant or irrelevant features, reduce the amount of data for subsequent processing, improve computational efficiency, and provide a scientific basis for feature selection.

[0057] Next, based on the obtained set of local feature importance scores of the welding inspection state, all local feature vectors are sorted in descending order from high to low to form a descending sequence. This sorting process itself does not change the content of the features, but reorganizes the order of the features so that the most important features are at the front. In this way, a part of the top-ranked feature vectors can be directly selected for subsequent analysis to achieve feature sparsification. This method not only reduces the computational burden and improves the speed and efficiency of model training, but also focuses on analyzing the most critical information to avoid being disturbed by secondary or insignificant data.

[0058] In one embodiment, Figure 4As shown, the descending sequence of the local feature vectors of the welding detection state is subjected to feature filtering and reshaping to obtain the optimized feature graph of the welding detection state, including: S4431, based on the neighborhood features of each local feature vector of the welding detection state in the descending sequence of the local feature vectors of the welding detection state, the feature neighborhood activity of each local feature vector of the welding detection state is calculated to obtain a sequence of the local feature neighborhood activity of the welding detection state. Specifically, the process can be expressed by the formula:

[0059]

[0060] in, yes Middle The eigenvalues ​​at the positions, yes The number of eigenvalues ​​in , and They are and The corresponding weighted average value of welding detection status, yes The corresponding local feature neighborhood activity of welding detection status.

[0061] S4432, based on the sequence of the neighborhood activity of the local feature of the welding detection state, feature filtering is performed on the descending sequence of the local feature vector of the welding detection state to obtain a descending sequence of the local feature vector of the welding detection state after filtering. Specifically, the process can be expressed by the formula:

[0062]

[0063]

[0064] in, For Perform feature selection, is the preset threshold, , , and They are the first, second, and third in the descending sequence of the local feature vectors of the welding detection state after filtering. and The local feature vector of welding detection state after filtering.

[0065] S4433, reshaping the descending sequence of the filtered local feature vectors of the welding detection state to obtain the welding detection state optimization feature map. Specifically, the process can be expressed by the formula:

[0066]

[0067] in, For the reshape operation, It is the welding detection state optimization characteristic diagram.

[0068] In one embodiment, based on the neighborhood features of each welding detection state local feature vector in the descending sequence of the welding detection state local feature vectors, the feature neighborhood activity of each welding detection state local feature vector is calculated to obtain a sequence of welding detection state local feature neighborhood activity, including: based on the welding detection state weighted average value of each welding detection state local feature vector in the descending sequence of the welding detection state local feature vectors and the subsequent welding detection state local feature vector, the feature neighborhood activity of each welding detection state local feature vector is calculated to obtain the sequence of welding detection state local feature neighborhood activity.

[0069] Specifically, first, by calculating the feature neighborhood activity of the local feature vectors of each welding detection state, the strength of the association between each feature and its surrounding features can be evaluated. Feature neighborhood activity reflects the degree of interaction between each local feature and the surrounding features, emphasizing the spatial relationship between features, rather than the importance of a single feature. In the welding area, the change of one feature may affect multiple features around it, so considering the neighborhood features can more comprehensively understand the existence of welding defects and their scope of influence. For example, there may be defects such as fine cracks or pores at the welding joint, which not only manifest as anomalies at a specific location, but also cause changes in the feature values ​​of the surrounding area. By calculating the feature neighborhood activity, the system can capture the spatial distribution characteristics of such local feature changes, thereby more accurately locating and identifying potential welding defects. This method deepens the understanding of the feature space and helps to discover problems that are difficult to detect with a single feature alone.

[0070] Next, based on the obtained neighborhood activity sequence of local features of welding detection status, the local feature vectors in descending order are filtered to retain the most valuable features. This process does not rely solely on the importance score of individual features, but takes into account the relationship between the feature and its neighbors. The features selected in this way can not only represent the original data well, but also better capture the patterns and regularities in the data. The effect of feature filtering is that it further optimizes the feature set, making it more compact and more expressive. The selected high-activity features usually contain more key information about welding defects, which helps to improve the performance of subsequent classification or detection models. In addition, reducing the number of unnecessary features also reduces the computational complexity, improves the processing speed, and makes the entire detection process more efficient.

[0071] Finally, the descending sequence of the filtered local feature vectors of the welding detection state is reshaped to restore the spatial structure of the original feature map to form an optimized feature map of the welding detection state. The reshaping process involves reorganizing the one-dimensional feature vector into a multi-dimensional feature map to ensure consistency with the original input format. This is crucial to maintaining the consistency and compatibility of the model architecture. The reshaped feature map not only retains the selected high-quality feature information, but also maintains the original spatial structure, which is convenient for subsequent tasks.

[0072] Exemplarily, in step S45, based on the welding detection state optimization characteristic map, a detection result is obtained for indicating whether the welding state between the target muffler pipe and the target partition has defects. In one embodiment, based on the welding detection state optimization characteristic map, a detection result is obtained for indicating whether the welding state between the target muffler pipe and the target partition has defects, including: passing the welding detection state optimization characteristic map through a welding quality detector based on a classifier to obtain the detection result, and the detection result is used to indicate whether the welding state between the target muffler pipe and the target partition has defects. That is, the welding detection state optimization characteristic map obtained by performing neighborhood importance optimization on the welding detection state characteristic map is classified, so as to use the powerful classification decision-making ability of the classifier to judge whether the welding state belongs to the defective or non-defective category, thereby obtaining the detection result, so as to clarify whether the welding state between the target muffler pipe and the target partition has defects, which is helpful to timely discover unqualified products in the production process.

[0073] In one embodiment, the classifier-based welding quality detector can use a softmax classifier. The softmax classifier is usually used for multi-classification problems, but it can also be well applied to binary classification tasks, and the probability distribution characteristics of its output help provide more intuitive and reliable detection results. Specifically, the softmax classifier can map the input features to a probability distribution, thereby assigning a probability value to each category (such as qualified welds and defective welds). This probabilistic output not only provides a clear classification decision, but also helps to evaluate the model's confidence in the classification of a specific sample.

[0074] In a specific embodiment, the welding detection state optimization feature map is expanded into a welding detection state optimization feature vector according to a row vector or a column vector; the welding detection state optimization feature vector is sent to a softmax classifier for final quality judgment. The softmax classifier maps each welding detection state optimization feature vector to a probability distribution on two categories (qualified and unqualified) by learning the patterns in the training data. For each welding point to be tested, the classifier outputs two probability values, which respectively represent the possibility that the welding point belongs to the qualified or unqualified category. The sum of these two probability values ​​is equal to 1, so their sizes can be directly compared to make the final classification decision.

[0075] In a preferred example, the step of passing the welding detection state optimization feature map through a classifier-based welding quality detector to obtain a detection result includes:

[0076] Determine the mean of the characteristic values ​​corresponding to the welding detection state optimization characteristic diagram and the standard deviation of the eigenvalues ;

[0077] The first welding detection state optimization fair target feature map is obtained by multiplying the welding detection state optimization feature map and the feature value mean point subtraction feature map with the feature value standard deviation:

[0078]

[0079] in, represents dot product, Indicates point reduction, represents the welding detection state optimization characteristic diagram, represents a first welding inspection state optimization fairness target characteristic map;

[0080] The point-by-point subtraction feature map of the welding detection state optimization feature map and the feature value standard deviation is multiplied by the feature value mean to obtain a second welding detection state optimization fair target feature map:

[0081]

[0082] in, represents a second welding inspection state optimization fairness target feature map;

[0083] After multiplying the bitwise inverse of the second welding detection state optimization fair target feature map by the first welding detection state optimization fair target feature map, taking the bitwise logarithm with base 2 to obtain the welding detection state optimization information correction feature map:

[0084]

[0085] in, represents the bit-by-bit inverse of the second welding detection state optimization fairness target feature map, that is, calculating the inverse of each eigenvalue of the second welding detection state optimization fairness target feature map, Represents the welding inspection state optimization information correction characteristic diagram;

[0086] The eigenvalue mean Divide by the standard deviation of the eigenvalue The square root of the quotient is multiplied by the weight hyperparameter, and then the optimized welding detection state optimization feature map is obtained by adding the welding detection state optimization information correction feature map points:

[0087]

[0088] in, Indicates point addition, represents the weight hyperparameter, An optimized characteristic diagram representing an optimized welding inspection state;

[0089] The optimized welding detection state optimization characteristic diagram is passed through a welding quality detector based on a classifier to obtain a detection result.

[0090] Here, since the welding detection state feature map represents the local dilated convolution associated image semantic features of the mask-segmented welding state image, when it is subjected to feature gating enhancement guided by grid energy significance, the feature population attributes of different local image semantics will have differences in the fairness of guided gating enhancement based on grid energy significance, thereby affecting the enhanced feature distribution inclusiveness of the welding detection state optimization feature map and reducing the accuracy of the detection results obtained by the classifier-based welding quality detector.

[0091] Therefore, considering the attribute level fairness differences of the data population corresponding to the local spatial fusion features of the welding detection state optimization feature map, in order to improve the inclusiveness of the welding detection state optimization feature map under the feature distribution diversity, the cross probability value constraint based on the welding detection state optimization feature map is used as the interactive fairness target representation to correct the interactive propagation of the group feature information of the welding detection state optimization feature map, and the unified statistical feature response interaction based on the welding detection state optimization feature map is used as the feature distribution multi-level fairness target bias to achieve the robust distribution fairness unified representation of the welding detection state optimization feature map, form a fair collaboration paradigm under the feature distribution framework of the welding detection state optimization feature map, and improve the accuracy of the detection results obtained by the classifier-based welding quality detector.

[0092] In summary, the assembly welding positioning method of automobile parts according to the embodiment of the present application is explained, which uses a camera to collect the welding state image between the target muffler and the target partition, uses the image recognition and analysis algorithm based on computer vision to perform welding state mask segmentation on the welding state image, and then extracts the welding state features of the segmented welding detection area state mask image, so as to realize intelligent detection of whether there are defects in the welding state between the target muffler and the target partition according to the neighborhood importance optimization of the extracted welding detection state features. Compared with traditional manual visual inspection, the present application can capture tiny defects such as fine cracks and internal pores, thereby improving the accuracy of defect identification. And it greatly improves the detection speed, while maintaining consistency and stability, reducing the probability of missed detection or false detection, to ensure that each welding point is strictly quality controlled.

[0093] Figure 5 Schematic block diagram of the assembly welding positioning device of automobile parts according to the embodiment of the present application. Figure 5 As shown, the assembly welding positioning device 100 for automobile parts includes: a welding state image acquisition module 110, which is used to acquire the welding state image between the target muffler and the target partition through a camera; a welding state mask segmentation module 120, which is used to perform welding state mask segmentation on the welding state image to obtain a welding detection area state mask image; a welding state feature extraction module 130, which is used to perform welding state feature extraction on the welding detection area state mask image to obtain a welding detection state feature map; a welding detection state optimization module 140, which is used to perform feature optimization based on neighborhood importance on the welding detection state feature map to obtain a welding detection state optimized feature map; a welding detection result determination unit 150, which is used to obtain a detection result indicating whether there is a defect in the welding state between the target muffler and the target partition based on the welding detection state optimized feature map.

[0094] In one embodiment, the welding detection state optimization module includes: a welding detection state feature flattening unit, used to perform feature decoupling and feature flattening on the welding detection state feature map to obtain a set of welding detection state local feature vectors; a descending order arrangement unit, used to perform descending order arrangement on the set of welding detection state local feature vectors to obtain a descending sequence of welding detection state local feature vectors; a feature filtering unit, used to perform feature filtering and reshaping on the descending sequence of the welding detection state local feature vectors to obtain the welding detection state optimization feature map;

[0095] In one embodiment, the welding state mask segmentation module is used to: pass the welding state image through a welding state mask segmentor based on a mask region convolutional neural network to obtain the welding detection area state mask image.

[0096] Here, those skilled in the art will appreciate that the specific operations of the various modules and units in the above-mentioned assembly welding positioning device for automobile parts have been described in detail above. Figures 1 to 4 The description of the assembly welding positioning method of automobile parts has been described in detail, and therefore, its repeated description will be omitted.

[0097] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the devices, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, program segment or a part of the code contains at least one executable instruction for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0098] The present application uses specific words to describe the embodiments of the present application. For example, "first / second embodiment", "one embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of the present application. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned in different positions in this specification does not necessarily refer to the same embodiment. In addition, some features, structures or characteristics in one or more embodiments of the present application can be appropriately combined.

[0099] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined as such herein.

[0100] The above is an explanation of the present application and should not be considered as a limitation thereof. Although several exemplary embodiments of the present application are described, it will be readily understood by those skilled in the art that many modifications may be made to these exemplary embodiments without departing from the novel teachings and advantages of the present application. Therefore, all of these modifications are intended to be included within the scope of the present application. It should be understood that the foregoing is an explanation of the present application and should not be considered to be limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the definition of the present application. The present application is defined by its description and equivalents.

Claims

1. A method for assembling and welding automobile parts, comprising: Placing three partitions in the clamping mechanism and arranging the three partitions at a predetermined interval; The first silencer pipe, the second silencer pipe and the third silencer pipe are respectively passed through the three partitions, and the three silencer pipes are respectively positioned and abutted and fixed by using abutment mechanisms; the manipulator is started to weld the first silencer pipe, the second silencer pipe and the third silencer pipe to the three partitions respectively to obtain an automobile silencer package; the assembly welding positioning method of automobile parts further includes: performing welding quality inspection on the automobile silencer package; The welding quality inspection of the automobile silencer package includes: The welding state image between the target muffler pipe and the target partition is collected by a camera; Performing welding state mask segmentation on the welding state image to obtain a welding detection area state mask image; Extracting welding state features from the welding detection area state mask image to obtain a welding detection state feature map; The welding detection state feature map is subjected to feature optimization based on neighborhood importance to obtain the welding detection state optimized feature map, including: performing feature decoupling and feature flattening on the welding detection state feature map to obtain a set of welding detection state local feature vectors; arranging the set of welding detection state local feature vectors in descending order to obtain a descending sequence of welding detection state local feature vectors; and performing feature filtering and reshaping on the descending sequence of welding detection state local feature vectors to obtain the welding detection state optimized feature map; Based on the welding detection state optimization characteristic diagram, a detection result is obtained for indicating whether there is a defect in the welding state between the target muffler pipe and the target partition plate; The method of arranging the set of the welding detection state local feature vectors in descending order to obtain a descending sequence of the welding detection state local feature vectors includes: Performing importance measurement on each welding detection state local feature vector in the set of welding detection state local feature vectors to obtain a set of welding detection state local feature importance score values; Based on the set of importance score values ​​of the welding detection state local features, the set of the welding detection state local feature vectors is arranged in descending order to obtain a descending sequence of the welding detection state local feature vectors; The step of filtering and reshaping the descending sequence of the local feature vectors of the welding detection state to obtain the welding detection state optimization feature map includes: Based on the neighborhood features of each welding detection state local feature vector in the descending sequence of the welding detection state local feature vectors, calculating the feature neighborhood activity of each welding detection state local feature vector to obtain a sequence of welding detection state local feature neighborhood activity; Based on the sequence of the neighborhood activity of the local feature of the welding detection state, feature filtering is performed on the descending sequence of the local feature vectors of the welding detection state to obtain a descending sequence of the local feature vectors of the welding detection state after filtering; The descending sequence of the filtered local feature vectors of the welding detection state is reshaped to obtain the welding detection state optimization feature map.

2. The assembly welding positioning method of automobile parts according to claim 1, characterized in that: Performing welding state mask segmentation on the welding state image to obtain a welding detection area state mask image includes: passing the welding state image through a welding state mask segmentor based on a mask area convolutional neural network to obtain the welding detection area state mask image.

3. The assembly welding positioning method of automobile parts according to claim 2, characterized in that: The welding detection area state mask image is subjected to welding state feature extraction to obtain a welding detection state feature map, including: the welding detection area state mask image is passed through a welding state feature extractor based on an expanded convolutional neural network model to obtain the welding detection state feature map.

4. The assembly welding positioning method of automobile parts according to claim 3, characterized in that: Performing feature decoupling and feature flattening on the welding detection state feature map to obtain a set of local feature vectors of the welding detection state, including: Performing feature decoupling on the welding detection state feature map along the channel dimension of the welding detection state feature map to obtain a set of welding detection state local feature matrices; Each welding detection state local feature matrix in the set of welding detection state local feature matrices is feature flattened to obtain the set of welding detection state local feature vectors.

5. The assembly welding positioning method of automobile parts according to claim 4, characterized in that: Based on the neighborhood features of each welding detection state local feature vector in the descending sequence of the welding detection state local feature vectors, the feature neighborhood activity of each welding detection state local feature vector is calculated to obtain a sequence of welding detection state local feature neighborhood activity, including: based on the welding detection state weighted average value of each welding detection state local feature vector in the descending sequence of the welding detection state local feature vectors and the subsequent welding detection state local feature vector, the feature neighborhood activity of each welding detection state local feature vector is calculated to obtain the sequence of welding detection state local feature neighborhood activity.

6. The assembly welding positioning method of automobile parts according to claim 5, characterized in that: Based on the welding detection state optimization characteristic diagram, a detection result is obtained to indicate whether there are defects in the welding state between the target silencer pipe and the target partition plate, including: passing the welding detection state optimization characteristic diagram through a welding quality detector based on a classifier to obtain the detection result, and the detection result is used to indicate whether there are defects in the welding state between the target silencer pipe and the target partition plate.

7. An assembly welding positioning device for automobile parts, used to perform the assembly welding positioning method for automobile parts according to claim 1, characterized in that: include: A welding state image acquisition module, used for acquiring a welding state image between a target muffler pipe and a target partition through a camera; A welding state mask segmentation module, used for performing welding state mask segmentation on the welding state image to obtain a welding detection area state mask image; A welding state feature extraction module, used for performing welding state feature extraction on the welding detection area state mask image to obtain a welding detection state feature map; A welding detection state optimization module, used for performing feature optimization based on neighborhood importance on the welding detection state feature map to obtain a welding detection state optimized feature map; wherein the welding detection state optimization module comprises: a welding detection state feature flattening unit, used for performing feature decoupling and feature flattening on the welding detection state feature map to obtain a set of welding detection state local feature vectors; a descending order arrangement unit, used for performing descending order arrangement on the set of welding detection state local feature vectors to obtain a descending sequence of welding detection state local feature vectors; a feature filtering unit, used for performing feature filtering and reshaping on the descending sequence of the welding detection state local feature vectors to obtain the welding detection state optimized feature map; The welding detection result determination unit is used to obtain a detection result indicating whether there is a defect in the welding state between the target muffler pipe and the target partition plate based on the welding detection state optimization characteristic diagram.

8. The assembly welding positioning device for automobile parts according to claim 7, characterized in that: The welding state mask segmentation module is used to: pass the welding state image through a welding state mask segmentor based on a mask area convolutional neural network to obtain the welding detection area state mask image.

Images