A real-time monitoring method and system for a radiator production line

Abstract

The present invention relates to the field of image processing technology, and in particular to a real-time monitoring method and system for a radiator production line. The method comprises: collecting a surface image of the radiator, determining the arrangement direction of the fins based on the distribution of the image gradient direction; calculating the edge point probability of each pixel in the image according to the arrangement direction; in the iteration of the agglomerative hierarchical clustering algorithm, dynamically determining the merging criteria for each round by combining the spatial proximity of the clusters to be merged and the discrete degree of the edge point probability of the pixels within the cluster; after obtaining a number of clusters through clustering algorithm processing, screening out the edge area clusters; based on the spatial distribution law of the pixels within the edge area clusters, identifying potential damage by detecting abnormal morphology of the fin distribution. The present invention can effectively avoid the problem of mismerging the edges of adjacent fins, accurately segment the edge area clusters, and improve recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a real-time monitoring method and system for a radiator production line. Background Art

[0002] A radiator is a heat exchange device that cools equipment by expanding its heated surface area and accelerating heat transfer to the surrounding environment. Its core heat dissipation element is a densely packed array of fins. These fins maximize contact area with the air and, combined with the airflow path design, improve heat dissipation efficiency, making them a key component in determining radiator performance.

[0003] During the manufacturing process, stamping and assembly processes often cause structural damage to the fins, such as falling, sticking, and deformation. This damage can disrupt the regularity of the fin array, affecting the heat sink's airflow efficiency and heat exchange area, and therefore becomes a core indicator of production line quality control.

[0004] In recent years, image processing-based radiator fin inspection methods have gained increasing attention. Damage to radiator fins is often concentrated at the edges, and inspection typically requires identifying these fin edges from images. Related techniques often employ agglomerative hierarchical clustering algorithms for segmentation. These algorithms use pixels as initial clusters and iteratively merge them based on grayscale similarity until a predetermined number of clusters are obtained.

[0005] However, this algorithm has limitations: the spacing between heat sink fins is typically small, and the grayscale gradients at adjacent fin edges vary little. Given the close proximity of adjacent fins, it is easy to mistakenly merge adjacent fin edge regions, or fin edge and non-edge regions, during the later stages of iterative merging. This mismerging can distort edge contours, leading to false detections in subsequent damage identification and failing to meet the accuracy requirements of real-time quality control on the production line. Summary of the Invention

[0006] In order to solve the technical problem that the above-mentioned agglomerative hierarchical clustering algorithm may cause distortion of the fin edge contour and misdetection of the heat sink, the present invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a method for real-time monitoring of a radiator production line, the method comprising the steps of:

[0008] Collect a surface image of a radiator; determine the fin arrangement direction of the radiator based on the gradient direction distribution of the surface image; calculate the edge point probability of each pixel in the surface image according to the fin arrangement direction, wherein the edge point probability is the possibility that the pixel is located in the edge area of ​​the fin; process the pixel points through an agglomerative hierarchical clustering algorithm to obtain a number of clusters, wherein, in each iterative round, the merging criterion of the corresponding round is determined according to the spatial proximity between the clusters to be merged and the discrete degree of the edge point probability of each pixel point within the clusters; from the several clusters, a predetermined number of clusters are screened out according to the mean value of the edge point probability of the pixel points in each cluster, and recorded as edge area clusters; based on the spatial distribution regularity of the pixel points in the edge area clusters, potential damage to the fins is identified by detecting abnormal morphology of the fin distribution.

[0009] The present invention realizes the identification of fin edge damage of radiator through multi-link collaborative optimization. First, the fin arrangement direction is determined based on the gradient direction, and the edge point probability of the pixel point is calculated in a targeted manner, which improves the initial accuracy of edge positioning; secondly, the improved agglomerative hierarchical clustering algorithm combines spatial proximity and edge probability discreteness to dynamically adjust the merging criteria, effectively avoiding the mismerging of adjacent fin edges and accurately segmenting edge area clusters; then, a predetermined number of clusters are screened according to the average edge probability to match the number characteristics of the fins, ensuring that all real edges are covered and noise interference is eliminated; finally, damage is identified based on the spatial distribution law of edge clusters, which can accurately capture structural defects such as lodging and adhesion. The overall solution is adaptable to different fin specifications and has strong anti-interference ability, realizing efficient processing of the entire process from image acquisition to defect identification, and improving the accuracy and reliability of real-time monitoring of the production line.

[0010] Preferably, the similarity metric of the agglomerative hierarchical clustering algorithm is constructed by fusing the edge point probability difference and spatial distance of each pixel point.

[0011] Preferably, the similarity metric satisfies the relationship: ;in, is the similarity between any two pixels between the clusters to be merged, is the absolute difference in the edge point probabilities of the two pixels between the clusters to be merged, is the spatial distance between the two pixels between the clusters to be merged, is the natural exponential function, is the standard normalization function.

[0012] The similarity measurement formula of the present invention uses the absolute difference of edge point probabilities to accurately focus on the differences in fin edge features, and uses spatial distance to constrain spatial proximity relationships, synergistically improving the accuracy of fin edge cluster division during clustering, reducing mismerging, and building a reliable data foundation for subsequent defect identification, adapting to the real-time monitoring needs of the production line.

[0013] Preferably, the determination of the merging criterion depends on the merging distance threshold; when the clusters to be merged are adjacent in space, the merging distance threshold ; When the clusters to be merged are not adjacent in space, the merging distance threshold ;in, It is The merging distance threshold of the clusters to be merged in the round, It is The minimum spatial distance between clusters to be merged in a round, It is The standard deviation of the edge point probabilities of all pixels in the cluster to be merged in this round, It is The maximum standard deviation of the edge point probabilities of all pixels in the cluster to be merged in the round and the merged clusters in its historical rounds, It is The average value of the standard deviation of the edge point probabilities of all pixels in the merged clusters in the historical rounds of the round, It is a small value used to prevent the denominator from being zero.

[0014] The proposed merging criterion dynamically adjusts the merging distance threshold based on spatial proximity. When the clusters to be merged are adjacent, the threshold is optimized by combining the historical and current standard deviations of the cluster edge probabilities, taking into account both spatial correlation and feature discreteness. When the clusters to be merged are not adjacent, a minimum spatial distance constraint is used to avoid erroneous merging. This precisely adapts to the needs of fin edge clustering, effectively distinguishing true fin edges from noise and interference from adjacent fins, and improving the accuracy of edge region cluster extraction.

[0015] Preferably, the spatial distribution regularity of pixel points within the edge area cluster includes: grouping the pixel points within the edge area cluster according to a coordinate axis parallel to the arrangement direction of the fins, with each group defined as a row; constructing a distance sequence of adjacent pixels in each row of pixel points; and characterizing the spatial distribution regularity of pixel points within the edge area cluster according to the degree of discreteness of the distance sequence of adjacent pixels in each row of pixel points.

[0016] Preferably, the spatial distribution regularity of the pixels in the edge region cluster is quantified by jump periodicity, and the jump periodicity satisfies the relationship: ;in, It is The jump periodicity of the row pixel distribution, It is In the distance sequence of adjacent pixels in a row, data points and The absolute difference of the data points; It is The average value of the absolute difference between all two adjacent data points in the distance sequence of adjacent pixels in a row, It is The total number of data points in the distance sequence of adjacent pixels in a row, is the natural exponential function, is the absolute value symbol.

[0017] This method quantifies the spatial distribution of pixels within edge clusters using a jump periodicity metric. Using a formula, the deviation between the absolute difference in distances between adjacent pixels and the mean is converted into a jump periodicity metric using a natural exponent. This method accurately captures the fluctuation characteristics of the pixel distance sequence at the fin edge, effectively identifying distribution anomalies caused by fin damage such as lodging and fracture, providing a reliable basis for subsequent location of defective areas.

[0018] Preferably, the identification of potential damage to the fin also includes: obtaining the row with the highest jump periodicity in the edge area cluster, and recording the average of the absolute differences between each pair of adjacent data points in the distance sequence of adjacent pixel points corresponding to the highest row as the target jump distance; traversing each row of adjacent pixel points in the edge area cluster, when the absolute difference between the distances of adjacent pixel points and the absolute difference between the target jump distance is greater than a preset fault tolerance distance, marking the row where the corresponding pixel point is located as a defective area.

[0019] Preferably, determining the arrangement direction of the fins of the radiator includes: calculating the gradient direction angle of each pixel point in the surface image, and counting the frequency of occurrence of each gradient direction angle; and determining whether the arrangement direction of the fins is horizontal or vertical based on the average of the gradient direction angles with the highest and second highest occurrence frequencies.

[0020] Preferably, the calculation of the edge point probability of each pixel point in the surface image as an edge point includes: for any pixel point, obtaining the grayscale value of its adjacent pixel point in the direction perpendicular to the fin arrangement direction; and obtaining the edge point probability of the pixel point by using the difference between the grayscale value of the pixel point and the grayscale value of its adjacent pixel point.

[0021] This method focuses on the fin arrangement direction and extracts the grayscale differences of adjacent pixels perpendicular to that direction to calculate the edge probability of each pixel. Leveraging the structural characteristics of the fins, it specifically targets potential edge areas, reduces interference from irrelevant directions, and improves the accuracy of initial edge point screening. The grayscale difference quantification method aligns with visual characteristics, providing reliable basic data for subsequent clustering and damage identification.

[0022] In the second aspect, the present invention provides a real-time monitoring system for a radiator production line. The real-time monitoring system for a radiator production line includes a memory and a processor. The memory stores computer program instructions. When the computer program instructions are executed by the processor, a real-time monitoring method for a radiator production line according to the first aspect of the present invention is implemented.

[0023] By adopting the above technical solution, a real-time monitoring method for a radiator production line in the first aspect of the present invention is generated into a computer program and stored in a memory so as to be loaded and executed by a processor, thereby making a terminal device based on the memory and the processor for easy use.

[0024] The beneficial effects of the present invention are as follows: The present invention achieves precise identification of radiator fin edge damage through the collaboration of multiple links: Fin arrangement orientation is determined based on gradient direction and pixel edge point probability is calculated to improve initial positioning accuracy; an improved agglomerative hierarchical clustering algorithm is used to optimize cluster division by integrating edge point probability differences with spatial distance constraints; merging criteria are dynamically adjusted based on spatial proximity and probability discreteness, doubly avoiding the mismerging of adjacent fin edges and accurately segmenting edge region clusters; and finally, defects such as lodging and adhesion are captured based on the distribution patterns of edge clusters. The solution is adaptable to fins of various specifications and has strong anti-interference capabilities, achieving efficient processing of the entire process from image acquisition to defect identification, improving the accuracy and reliability of real-time monitoring of production lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flow chart of a real-time monitoring method for a radiator production line provided by an embodiment of the present invention;

[0026] Figure 2 This is a structural block diagram of a real-time monitoring system for a radiator production line provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] A first aspect of the present invention provides a real-time monitoring method for a radiator production line. Figure 1 As shown, the method includes steps S100 to S600:

[0028] Step S100: collecting a surface image of the radiator.

[0029] It should be noted that radiators typically use an array of fins that are densely arranged at equal intervals. This regular structure maximizes the heat exchange area and ensures airflow efficiency, and is the core design that ensures its heat exchange performance and operating efficiency. Once the fin structure is damaged, the actual contact area between the radiator surface and the fluid will be significantly reduced. This means that the number of fluid molecules involved in heat exchange will decrease, which will directly lead to a decrease in heat exchange capacity, and thus seriously affect the overall heat dissipation effect of the radiator. Therefore, after the radiator fins are assembled, it is necessary to collect images of the radiator fin area to assist in analyzing whether there is any structural damage.

[0030] Specifically, sampling points were set up along the assembly line, and image acquisition equipment was deployed to capture images of the radiator surface, ensuring that the detailed features of each fin area on the radiator surface were clearly captured. Furthermore, to reduce the impact of environmental noise on image quality, the captured images were filtered and denoised. Furthermore, to reduce the computational complexity of subsequent image processing, the images were grayscaled.

[0031] At this point, the surface image of the radiator was acquired for subsequent analysis.

[0032] Step S200: determining the arrangement direction of the fins of the heat sink based on the gradient direction distribution of the surface image.

[0033] It should be noted that after actual production and assembly, the fins may appear in a horizontal or vertical distribution in the surface images of the heat sink collected during the transmission process. Fins with different distribution directions have different geometric characteristics, which provides an analytical basis for extracting the edge direction of the fin area and helps improve the accuracy of edge extraction. Because the gradient direction is perpendicular to the edge direction, the main direction of the fin edge can be identified by extracting image gradient information. Based on the correspondence between the edge direction and the fin arrangement, the horizontal or vertical arrangement of the fins can be determined.

[0034] Specifically, calculate the gradient direction angle of each pixel in the current image, count the frequency of occurrence of each gradient direction angle, and extract the gradient direction angle corresponding to the highest frequency and the second highest frequency. If the mean of these two gradient direction angles is less than , it is determined that the gradient direction angle in the image is concentrated in the vertical direction; based on the characteristic that the gradient direction is perpendicular to the edge direction, the main direction of the fin edge is horizontal, corresponding to the horizontally arranged fins. If the average of these two gradient direction angles is greater than or equal to , it is determined that the gradient direction angle in the image is concentrated in the horizontal direction; similarly, the main direction of the fin edge is the vertical direction, corresponding to the longitudinally arranged fins.

[0035] At this point, the fin arrangement direction of the radiator is obtained.

[0036] Step S300 : Calculate the edge point probability of each pixel in the surface image according to the fin arrangement direction. The edge point probability is the possibility that the pixel is located in the edge area of ​​the fin.

[0037] It should be noted that since structural damage to fins during manufacturing is typically concentrated at the edge, it's necessary to analyze the likelihood of each pixel being an edge point. Fins are mostly made of metals like copper and aluminum, arranged in a dense, equidistant array. The background areas between them have low brightness, resulting in significant grayscale differences between the fin edge and non-fin areas. Based on this, the probability of each pixel being an edge point can be calculated by quantifying the differences in the positional distribution characteristics between edge points and adjacent non-fin areas, combined with the distribution direction of the fins in the current image.

[0038] Specifically, taking the longitudinal fins as an example, since the fins are distributed longitudinally, the pixel point of each fin in the image will show obvious grayscale difference characteristics with the corresponding pixel points in the adjacent row on its row.

[0039] According to the above logic, for any pixel in each row, the probability formula of the pixel being an edge point can be constructed by the grayscale difference between the pixel's row and the pixels in the same column in the two adjacent rows. The edge point probability of the pixel satisfies the relationship:

[0040] ;

[0041] in, It is Rank The edge point probability of a pixel, It is Rank The gray value of a pixel, It is Rank The gray value of a pixel, Is relative to the The row offset value of the row is -1 and 1, corresponding to the The previous and next row of the row, is the standard normalization function, is the absolute value symbol.

[0042] In this formula, Reflect the Rank The sum of the grayscale deviations between the pixel point and the pixel points in the same column in the two adjacent rows. The larger the value, the Rank The more significant the grayscale jump is at the pixel point, the The greater the probability that a pixel is an edge point.

[0043] In addition, in the actual implementation process, Rank The adjacent pixels of a pixel are not limited to The pixel points of the previous and next rows of a row can also be set according to the specific scenario. For example, when there is a wide transition area at the edge of the fin, it can be extended to two adjacent rows, such as OK, OK, OK, The edge point probability of the pixel is calculated by summing the grayscale differences of more rows to reduce local noise interference.

[0044] At this point, the edge point probability of each pixel in the surface image is obtained.

[0045] Step S400: Process each pixel point through an agglomerative hierarchical clustering algorithm to obtain a number of clusters, wherein, in each iterative round, the merging criterion of the corresponding round is determined according to the spatial proximity between the clusters to be merged and the discrete degree of the edge point probability of each pixel point within them.

[0046] It should be noted that structural damage to fins caused by actual processes is often concentrated in the edge area, requiring segmentation of each fin's edge area to accurately locate the defect. Because the probability of pixels in the fin edge area is significantly higher than in other areas, and the pixels along a single edge are continuously distributed, an agglomerative hierarchical clustering algorithm can be used for segmentation.

[0047] The agglomerative hierarchical clustering algorithm is a bottom-up clustering algorithm that uses pixel points as initial clusters and iteratively merges clusters based on the minimum distance between cluster features. It does not require a preset number of clusters, is sensitive to local features, can adapt to the grayscale difference between the fin edge and the background, and does not rely on the prior number of edge clusters. It is suitable for the segmentation needs of fins of various specifications.

[0048] Specifically, the edge point probability and position information of the pixel point are used as clustering features: the edge point probability reflects whether the pixel point belongs to the edge area, and the position information reflects the continuous distribution characteristics of the edge. According to the above logic, the similarity between the clusters to be merged satisfies the relationship:

[0049] ;

[0050] in, is the similarity between any two pixels between the clusters to be merged, is the absolute difference in the probability of edge points of any two pixels between the clusters to be merged, is the spatial distance between any two pixels between the clusters to be merged, is the natural exponential function, is the standard normalization function.

[0051] In this formula, It reflects the probability of any two pixels being at the edge of the clusters to be merged and the degree of spatial proximity. The smaller the value, the more likely the two pixels are to be in the same region, i.e., the higher the similarity between the two pixels. The larger the value, the more likely the two pixels are to be in different regions, i.e., the lower the similarity between the two pixels.

[0052] Based on this, the clustering process is based on the similarity Instead of the traditional inter-cluster distance metric of the agglomerative hierarchical clustering algorithm, the method iteratively merges pixel clusters with high similarity to ultimately achieve accurate segmentation of the fin edge area and reduce mismerging caused by close edge positions and similar probabilities.

[0053] The above describes how to quantify the inter-cluster distance metric of the agglomerative hierarchical clustering algorithm, and the following describes the merging criterion of the agglomerative hierarchical clustering algorithm.

[0054] It should be noted that, given the proximity of two edges between adjacent fins and similar pixel edge probability, they can be mistakenly merged into the same cluster during clustering iterations, resulting in inaccurate segmentation of individual edges. Therefore, it is necessary to combine the continuous distribution characteristics of pixels and the edge characteristics within clusters to quantify the merge distance threshold for each round to reduce misclassification.

[0055] Specifically, we first select any pair of adjacent pixel points and calculate the Euclidean distance of their position information as the connectivity index distance. The connectivity index distance is used to determine whether the clusters to be merged are spatially adjacent. Secondly, for each round of two clusters to be merged, we extract the minimum spatial distance between the two clusters as the core parameter to measure the degree of spatial proximity between clusters. Based on the above logic, the merging distance threshold satisfies the relationship:

[0056] ;

[0057] in, It is The merging distance threshold of the clusters to be merged in the round, It is The minimum spatial distance between clusters to be merged in a round, It is The standard deviation of the edge point probabilities of all pixels in the cluster to be merged in this round, It is The maximum standard deviation of the edge point probability of all pixels in the cluster to be merged in the round and the merged clusters in its historical rounds. The meaning of historical rounds is: assuming the current round is the 10th round, the historical rounds refer to the previous 9 rounds. It is The average value of the standard deviation of the edge point probabilities of all pixels in the merged clusters in the historical rounds of the round, is the connectivity index distance, It is a small value used to prevent the denominator from being 0. It can be set to 0.01 or set as required.

[0058] In this formula, when When , it indicates that the two clusters to be merged are not adjacent in space and do not meet the continuous distribution characteristics, which means that the two clusters to be merged cannot be the same edge area of ​​the same fin. At this time, it is necessary to take the inverse to make the merging distance threshold significantly smaller than , thereby reducing the probability of merging and avoiding accidental merging.

[0059] when When , it indicates that the two clusters to be merged are adjacent in space, which conforms to the continuous distribution feature. However, considering that the actual non-fin edge area is distributed in the middle of the fin edge area, this means that there may be However, the two clusters are actually different areas. Therefore, it is necessary to introduce the standard deviation indicator of the edge point probability of all pixels in the two clusters to be merged for further judgment:

[0060] if , indicating that the edge point probability stability of all pixels in the clusters to be merged in the current round is significantly lower than the historical average level, which means that the two clusters to be merged in the current round are more likely to be the edge areas of different fins, that is, they contain pixels in other areas of the single edge of different fins. In this case, it is necessary to reduce Avoid merging. If , which means that the edge area of ​​the same fin in the cluster to be merged in the current round may be larger. In this case, it is necessary to increase or maintain Merging is allowed.

[0061] In the clustering iteration, if the similarity of all pixel pairs in the cluster to be merged is the minimum , then merge and enter the next round of iteration; if the minimum value , then stop merging. Through the above improvements, multiple clusters can be obtained after agglomerative hierarchical clustering processing.

[0062] Step S500 : Filtering a predetermined number of clusters from a plurality of clusters according to the mean of the edge point probabilities of the pixels in each cluster, and recording them as edge region clusters.

[0063] It should be noted that the mean probability of edge points in a cluster can intuitively reflect its edge properties: the higher the mean, the more consistent the pixels in the cluster are with edge characteristics, and the greater the likelihood that they are the true fin edge area. The smaller the mean, the lower the mean, the lower the mean of clusters in non-edge areas due to the low proportion of edge points. Therefore, this mean can be used to filter edge area clusters.

[0064] Specifically, for each cluster, the edge point probability mean of all pixels in the cluster is calculated and used as the probability measure of the cluster being the edge area of ​​the fin. After sorting all clusters from large to small in probability, the top The clusters are edge area clusters of the fins. Indicates the number of fins on the radiator. A fin usually has two edges, the front The clusters just match the complete edge number of all fins, which can not only cover all edges but also eliminate non-edge noise clusters through high probability screening.

[0065] So far, several edge region clusters have been obtained.

[0066] Step S600 : Based on the spatial distribution regularity of the pixels in the edge region cluster, potential damage of the fins is identified by detecting abnormal distribution patterns of the fins.

[0067] It's important to note that structural damage to fins caused by processes like stamping and assembly is often concentrated at the edges. Furthermore, the jump distances at the edges of individual fins exhibit a regular distribution. Therefore, by grouping the segmented clusters of individual fin edge regions by row and evaluating the jump periodicity of the pixel distribution within each row, we can assist in identifying potential damage areas.

[0068] Specifically, we extract all pixel points from the fin edge region clusters and group them by row. For each row of pixels, we construct a sequence of Euclidean distances between adjacent pixel positions. We then calculate the sum of the deviations between the absolute differences between each adjacent Euclidean distance in this sequence and the overall mean to quantify the spatial distribution regularity of the pixels within the edge region clusters, i.e., the jump periodicity. This jump periodicity satisfies the relationship:

[0069] ;

[0070] in, It is The jump periodicity of the row pixel distribution, It is In the distance sequence of adjacent pixels in a row, data points and The absolute difference of the data points; It is The average of the absolute differences between all two adjacent data points in the distance sequence of adjacent pixels in a row; It is The total number of data points in the distance sequence of adjacent pixels in a row, is the natural exponential function, is the absolute value symbol.

[0071] In this formula, This value reflects the jump characteristics of the spatial distance changes between adjacent pixels in the current row. A smaller value indicates that the absolute differences in the Euclidean distance sequence between adjacent pixels in the current row fluctuate less around their mean. This indicates that the pixel distribution in this row exhibits a stable jump-like periodicity, indicating no significant structural damage in the corresponding area. A larger value indicates that there are local jumps in the sequence that deviate significantly from the mean, and that the pixel distribution exhibits a non-periodic mutation, indicating potential damage such as lodging, adhesion, or deformation.

[0072] In addition, for the distance sequence of adjacent pixels, the distance can be calculated by Euclidean distance. Row pixels: , then the distance sequence of its adjacent pixels is: , the absolute difference sequence of the distance sequence: .

[0073] It should be noted that structural damage to heat sinks is rare on actual production lines and often occurs only on a few fins. Therefore, the defective area can be located by extracting the transition distance corresponding to the row with the most stable transition periodicity as a benchmark.

[0074] Specifically, first obtain the row with the highest jump periodicity, whose jump rule in the edge region cluster best conforms to the normal fin edge characteristics, and record the average of the absolute differences between two adjacent data points in the distance sequence of adjacent pixels in this row as the target jump distance.

[0075] Traverse the adjacent pixels in each row of the edge area cluster. If the absolute difference between the Euclidean distance of each pair of adjacent pixels in each row and the absolute difference between the target jump distance is greater than the tolerance distance, it means that any pair of adjacent pixels in the current row may have potential lodging, adhesion or deformation damage. After marking the physical area corresponding to the row as a defective area with potential damage, the system will record the location coordinates, damage type and severity information of the defect in real time, and simultaneously feed back this data to the production line monitoring terminal, triggering corresponding early warning prompts. Operators can quickly locate the problem fins based on the real-time defect information pushed, and stop the machine for maintenance or adjust production parameters in time, thereby realizing real-time monitoring of the entire process of the radiator production line from defect identification, information feedback to exception handling, and ensuring product quality stability and production efficiency.

[0076] The second aspect of this embodiment provides a real-time monitoring system for a radiator production line, such as Figure 2 As shown, the real-time monitoring system for a radiator production line includes a memory and a processor. The memory stores computer program instructions. When the computer program instructions are executed by the processor, a real-time monitoring method for a radiator production line according to the first aspect of the present invention is implemented.

[0077] The radiator production line real-time monitoring system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface. Their settings and functions are known in the art and will not be described in detail here.

[0078] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium may be any suitable magnetic storage medium or magneto-optical storage medium, such as resistive random access memory, dynamic random access memory, static random access memory, enhanced dynamic random access memory, high bandwidth memory, hybrid memory cube, etc., or any other medium that can be used to store the required information and can be accessed by an application, module, or both. Any such computer storage medium may be part of, accessible to, or connectable to the device.

[0079] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A real-time monitoring method for a radiator production line, characterized in that: Including steps: Collecting surface images of the radiator; Determine the fin arrangement direction of the radiator based on the gradient direction distribution of the surface image; According to the fin arrangement direction, the edge point probability of each pixel in the surface image is calculated. The edge point probability is the probability that the pixel is located in the edge area of ​​the fin; Each pixel is processed by an agglomerative hierarchical clustering algorithm to obtain several clusters. In each iteration, the merging criterion of the corresponding round is determined based on the spatial proximity between the clusters to be merged and the discrete degree of the edge point probability of each pixel within them. The determination of the merging criterion depends on the merging distance threshold. When the clusters to be merged are adjacent in space, the merging distance threshold ; When the clusters to be merged are not adjacent in space, the merging distance threshold ; It is The merging distance threshold of the clusters to be merged in the round, It is The minimum spatial distance between clusters to be merged in a round, It is The standard deviation of the edge point probabilities of all pixels in the cluster to be merged in this round, It is The maximum standard deviation of the edge point probabilities of all pixels in the cluster to be merged in the round and the merged clusters in its historical rounds, It is The average value of the standard deviation of the edge point probabilities of all pixels in the merged clusters in the historical rounds of the round, It is a small value used to prevent the denominator from being 0; From a number of clusters, a predetermined number of clusters are selected based on the mean of the edge point probabilities of the pixels in each cluster, and are recorded as edge area clusters; Based on the spatial distribution regularity of pixel points within the edge area cluster, the potential damage of the fins is identified by detecting the abnormal morphology of the fin distribution.

2. The real-time monitoring method for a radiator production line according to claim 1, characterized in that: The similarity metric of the agglomerative hierarchical clustering algorithm is constructed by fusing the edge point probability difference and spatial distance of each pixel point.

3. The real-time monitoring method for a radiator production line according to claim 2, characterized in that: The similarity measure satisfies the relationship: ; in, is the similarity between any two pixels between the clusters to be merged, is the absolute difference in the edge point probabilities of the two pixels between the clusters to be merged, is the spatial distance between the two pixels between the clusters to be merged, is the natural exponential function, is the standard normalization function.

4. The real-time monitoring method for a radiator production line according to claim 1, characterized in that: The spatial distribution regularity of the pixels within the edge region cluster includes: Grouping the pixels in the edge region cluster according to a coordinate axis parallel to the arrangement direction of the fins, with each group defined as a row; Construct a distance sequence between adjacent pixels in each row of pixels; The spatial distribution regularity of pixels within the edge region cluster is characterized according to the discrete degree of the distance sequence between adjacent pixels in each row of pixels.

5. The real-time monitoring method for a radiator production line according to claim 4, characterized in that: The spatial distribution regularity of the pixels in the edge region cluster is quantified by the jump periodicity, and the jump periodicity satisfies the relationship: ;in, It is The jump periodicity of the row pixel distribution, It is In the distance sequence of adjacent pixels in a row, data points and The absolute difference of the data points; It is The average value of the absolute difference between all two adjacent data points in the distance sequence of adjacent pixels in a row. It is The total number of data points in the distance sequence of adjacent pixels in a row, is the natural exponential function, is the absolute value symbol.

6. The real-time monitoring method for a radiator production line according to claim 5, characterized in that: The identifying of potential damage to the fins further includes: Obtaining the row with the highest jump periodicity within the edge region cluster, and recording the average of the absolute differences between two adjacent data points in the distance sequence of adjacent pixel points corresponding to the highest row as the target jump distance; Traverse each row of adjacent pixels in the edge area cluster. When the absolute difference between the distances between adjacent pixels and the absolute difference between the target jump distance is greater than the preset fault tolerance distance, mark the row where the corresponding pixel is located as a defective area.

7. The real-time monitoring method for a radiator production line according to claim 1, characterized in that: Determining the arrangement direction of the fins of the radiator includes: Calculating the gradient direction angle of each pixel in the surface image and counting the occurrence frequency of each gradient direction angle; The arrangement direction of the fins is determined to be a horizontal arrangement or a vertical arrangement according to the average of the gradient direction angles with the highest and second highest occurrence frequencies.

8. The real-time monitoring method for a radiator production line according to claim 1, characterized in that: The calculating the edge point probability of each pixel point in the surface image being an edge point includes: For any pixel, obtain the grayscale value of its adjacent pixel in the direction perpendicular to the arrangement direction of the fins; The edge point probability of the pixel is obtained by using the difference between the grayscale value of the pixel and the grayscale values ​​of its adjacent pixels.

9. A real-time monitoring system for a radiator production line, characterized in that: The radiator production line real-time monitoring system includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, a radiator production line real-time monitoring method according to any one of claims 1 to 8 is implemented.

Images