A can body pour detection system

Abstract

This application provides a can tipping detection system, relating to the field of aluminum can production technology, including: a data acquisition module; a point cloud fitting module; a bottom contact analysis module for performing bottom contact analysis based on a 3D model of the aluminum can to determine the bottom contact ratio and position; a region division module for determining multiple conveying regions; a tipping probability prediction module for predicting the tipping probability of multiple conveying regions and generating multiple tipping risk coefficients; and a warning zone setting module for removing the aluminum can from the conveyor belt before it reaches the first warning zone. This application solves the technical problem in existing aluminum can production processes where the inability to accurately predict tipping trends leads to the inability to promptly detect and handle potentially tipping cans, resulting in a high risk of production line downtime and impacting production progress and efficiency. It reduces the probability of production line downtime and improves the overall production efficiency of aluminum cans.

Description

Technical Field

[0001] This application relates to the field of beverage can manufacturing technology, and in particular to a can tipping detection system. Background Technology

[0002] On existing aluminum can production lines, conveyor belts are typically used to transport cans from one process to the next. However, during production, various factors such as unstable conveyor belt speed, equipment malfunction, and operational errors can cause cans to tip over. Failure to promptly detect and address these potential tipping cans can lead to production line downtime, consequently impacting the production schedule and efficiency of aluminum can production.

[0003] Currently, in the existing beverage can production process, the inability to accurately predict the can's tipping trend leads to the inability to detect and handle potentially tipping cans in a timely manner, resulting in a significant risk of production line downtime and affecting the progress and efficiency of beverage can production. Summary of the Invention

[0004] The purpose of this application is to provide a can tipping detection system to solve the technical problem in the existing beverage can production process where the inability to accurately predict the tipping trend of cans leads to the inability to detect and handle potentially tipping cans in a timely manner, resulting in a high risk of production line downtime and affecting the production progress and efficiency of beverage cans.

[0005] In view of the above problems, this application provides a can tipping detection system, comprising: a data acquisition module, used to collect real-time data on a target aluminum can on a conveyor belt based on a dual-source monitoring network, and acquire a dual-source monitoring dataset, wherein the dual-source monitoring network is deployed at the starting position of the conveyor belt, and the dual-source monitoring dataset includes an image set and a point cloud set; a point cloud fitting module, used to perform contour recognition using the image set to obtain the aluminum can contour, and to fit the point cloud set based on the aluminum can contour to generate a three-dimensional model of the aluminum can; and a bottom contact analysis module, used to perform bottom contact analysis in three-dimensional space based on the three-dimensional model of the aluminum can to determine the bottom contact features, wherein the bottom contact features include The system includes: a bottom contact ratio and bottom contact position; a region division module, used to read the topology of the conveyor belt structure and divide the conveyor belt into regions according to the tilt angle and tilt direction to determine multiple conveying regions; a tipping probability prediction module, used to predict the tipping probability of the multiple conveying regions based on the attribute characteristics of the target can, the surface roughness of the conveyor belt, the bottom contact ratio, and the bottom contact position, generating multiple tipping risk coefficients; and a warning region setting module, used to set the conveying region corresponding to the tipping risk coefficient greater than a preset threshold as a warning region. If the warning region is not 0, the target can is removed from the conveyor belt by a robotic arm before it reaches the first warning region.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] By monitoring and acquiring image and point cloud data of aluminum cans on the conveyor belt, the outline of the cans is determined based on image data analysis. Then, the point cloud data is fitted based on the can outline to construct a 3D model of the cans, which improves the accuracy of the 3D model construction. Next, in 3D space, bottom contact analysis is performed on the 3D model of the cans and the horizontal plane to determine the bottom contact ratio and position. Furthermore, based on the topology of the conveyor belt structure, the conveyor belt is divided into regions according to the tilt angle and tilt direction to determine multiple conveying areas. Finally, based on the attribute characteristics of the target aluminum cans and the surface roughness of the conveyor belt, ... The bottom contact ratio and bottom contact position are used to construct a tipping prediction model based on machine learning. This model predicts the tipping probability of multiple conveying areas and generates multiple tipping risk coefficients. Furthermore, conveying areas with tipping risk coefficients exceeding a preset threshold are designated as warning zones. Finally, before the target can reaches the first warning zone, a robotic arm removes the target can from the conveyor belt. This improves the accuracy and reliability of can tipping trend prediction, enabling rapid and accurate identification and handling of potentially tipping cans, reducing production line downtime, and ultimately improving the overall production efficiency of cans.

[0008] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of the structure of a tank tilting detection system according to this application.

[0011] Figure 2 This is a schematic diagram of the process for obtaining the outline of an aluminum can in a can tipping detection system according to this application.

[0012] Explanation of reference numerals in the attached figures:

[0013] Data acquisition module 11, point cloud fitting module 12, bottom contact analysis module 13, area division module 14, tipping probability prediction module 15, and early warning area setting module 16. Detailed Implementation

[0014] This application provides a can tipping detection system, which solves the technical problem in existing aluminum can production processes where the inability to accurately predict can tipping trends leads to the inability to promptly detect and handle potentially tipping cans, resulting in significant production line downtime risks and impacting production progress and efficiency. The system improves the accuracy and reliability of can tipping trend prediction, thereby enabling rapid and accurate identification and handling of potentially tipping cans, reducing the probability of production line downtime, and ultimately improving the overall production efficiency of aluminum cans.

[0015] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0016] For examples, please refer to the appendix. Figure 1 This application provides a tank tipping detection system, comprising:

[0017] The data acquisition module 11 is used to collect data on the target cans on the conveyor belt in real time based on the dual-source monitoring network and obtain the dual-source monitoring dataset. The dual-source monitoring network is deployed at the beginning of the conveyor belt, and the dual-source monitoring dataset includes an image set and a point cloud set.

[0018] Specifically, firstly, a dual-source monitoring network is constructed. This network is a monitoring system combining visual inspection and laser scanning technologies. Industrial cameras and laser scanners deployed at the beginning of the conveyor belt collect data simultaneously, acquiring both image information and 3D point cloud data of the cans on the conveyor belt. The starting position of the conveyor belt is, by default, parallel to the horizontal plane. Next, the dual-source monitoring network performs real-time data collection on the target cans on the conveyor belt, obtaining a dual-source monitoring dataset. This dataset contains monitoring data for the same can, including both image and point cloud datasets.

[0019] The point cloud fitting module 12 is used to perform contour recognition using the image set, obtain the contour of the aluminum can, and fit the point cloud set based on the aluminum can contour to generate a three-dimensional model of the aluminum can.

[0020] Specifically, the outline features of the target aluminum can are extracted based on the image set, wherein the image set includes image data from multiple different acquisition angles. First, the image data from different angles are fused and processed to obtain a complete image of the aluminum can. Then, the outline features of the complete image of the aluminum can are extracted based on the edge detection algorithm and the convolutional neural network, and the outline of the aluminum can is obtained after fusion.

[0021] Next, using the outline of the aluminum can as a constraint, a point cloud fitting algorithm is used to iteratively fit the point cloud set, obtaining the point cloud fitting result with the highest fitting accuracy under the expected conditions. Finally, based on the point cloud fitting result, a 3D model is constructed using CAD software. By constructing a 3D model of the aluminum can based on the fusion of image data and point cloud data, where point cloud data provides high-precision 3D information of the aluminum can surface and image data provides surface features and outline information, the advantages of both types of data can be combined to provide more comprehensive, accurate, and robust data, thereby further improving the accuracy and reliability of the 3D model construction of the aluminum can.

[0022] Bottom contact analysis module 13 is used to perform bottom contact analysis based on the three-dimensional model of the can in three-dimensional space to determine bottom contact characteristics, wherein the bottom contact characteristics include bottom contact ratio and bottom contact position.

[0023] Specifically, firstly, a 3D space is constructed using 3D modeling software, and then the 3D model of the aluminum can is imported into this space. Next, a horizontal plane is created in the 3D space as the analysis benchmark, where the horizontal plane represents the initial state of the conveyor belt. The 3D model of the aluminum can is aligned with the horizontal plane, ensuring that the bottom is in contact with it. Further, a bottom contact analysis is performed between the 3D model of the aluminum can and the horizontal plane to determine the bottom contact characteristics. These characteristics include the bottom contact ratio and the bottom contact position. The bottom contact ratio refers to the ratio of the area of ​​the bottom of the aluminum can in contact with the horizontal plane to the total area of ​​the bottom of the aluminum can; the bottom contact position refers to the distribution of contact points between the bottom of the aluminum can and the horizontal plane. By obtaining the bottom contact ratio and bottom contact position, data support is provided for subsequent aluminum can tipping prediction.

[0024] The region division module 14 is used to read the topology of the conveyor belt structure, divide the conveyor belt into regions according to the tilt angle and tilt direction, and determine multiple conveying regions.

[0025] Specifically, on a can production line, to more effectively monitor and control the cans on the conveyor belt, it is necessary to divide the conveyor belt into zones to facilitate monitoring and management based on different tilt angles and directions. First, the conveyor belt topology is read. This topology includes the structural features of the entire can production line, such as tilt angles, tilt directions, and connection relationships. Then, based on the conveyor belt topology, the conveyor belt is divided into zones according to tilt angles and directions. Zones with the same tilt direction and tilt angle within the same threshold are grouped into a single conveyor zone, thus defining multiple conveyor zones. By defining multiple conveyor zones, the granularity of the conveyor belt analysis can be improved, thereby enhancing the precision and accuracy of can tipping prediction.

[0026] The tipping probability prediction module 15 is used to predict the tipping probability of the multiple conveying areas based on the attribute characteristics of the target can, the surface roughness of the conveyor belt, the bottom contact ratio, and the bottom contact position, and generate multiple tipping risk coefficients.

[0027] Specifically, firstly, multiple prediction units are constructed based on various prediction operators, including BP neural networks, support vector machines, and random decision forests. Then, using the attribute features of the target can and the surface roughness of the conveyor belt as constraints, a sample dataset is obtained through source information retrieval based on the Industrial Internet. Next, the multiple prediction units are trained under supervision using the sample dataset to obtain multiple convergent prediction units. Finally, based on the principle of ensemble learning, the multiple convergent prediction units are fused to generate a tipping prediction model. The input data of the tipping prediction model includes the bottom contact ratio, bottom contact position, tilt angle, and tilt direction, while the output data is the tipping risk coefficient.

[0028] Then, using a tipping prediction model, the tipping probability is predicted for each of the multiple transport areas, generating multiple tipping risk coefficients. The tipping risk coefficient represents the tipping probability of the can; the higher the tipping probability, the higher the tipping risk coefficient.

[0029] The warning zone setting module 16 is used to set the conveyor area corresponding to the tipping risk coefficient that is greater than the preset threshold as the warning zone. If the warning zone is not 0, the target can is removed from the conveyor belt by a robotic arm before the target can reaches the first warning zone.

[0030] Specifically, multiple tipping risk coefficients are assessed based on preset risk thresholds, which can be set based on production experience and experimental data. Conveyor areas with tipping risk coefficients exceeding these thresholds are designated as warning zones. If a warning zone is not zero, a robotic arm removes the target can from the conveyor belt before it reaches the first warning zone. The first warning zone is the closest warning zone to the can. The robotic arm system precisely locates the target can and performs a gripping action to remove it from the conveyor belt. Simultaneously, operational safety is ensured during the gripping action to avoid damage to the can or the production line.

[0031] By providing a can tipping detection system, this invention addresses the technical problem in existing can production processes where the inability to accurately predict can tipping trends leads to the failure to promptly detect and handle potentially tipping cans, resulting in significant production line downtime and impacting production progress and efficiency. The system monitors and acquires image and point cloud data of cans on a conveyor belt. Based on image data analysis, the can outline is determined, and then the point cloud data is fitted to this outline to construct a 3D model of the can, improving the accuracy of the 3D model construction. Next, in 3D space, bottom contact analysis is performed on the 3D model of the can and the horizontal plane to determine the bottom contact ratio and position. Furthermore, based on the conveyor belt's structural topology, the conveyor belt is divided into regions according to its tilt angle and direction, identifying multiple conveying areas. Finally, based on the target can's attribute characteristics, conveyor belt surface roughness, and other factors, the system... The bottom contact ratio and bottom contact position are used to construct a tipping prediction model based on machine learning. This model predicts the tipping probability of multiple conveying areas and generates multiple tipping risk coefficients. Furthermore, conveying areas with tipping risk coefficients exceeding a preset threshold are designated as warning zones. Finally, before the target can reaches the first warning zone, a robotic arm removes the target can from the conveyor belt. This improves the accuracy and reliability of can tipping trend prediction, enabling rapid and accurate identification and handling of potentially tipping cans, reducing production line downtime, and ultimately improving the overall production efficiency of cans.

[0032] Furthermore, this application includes:

[0033] The dual-source monitoring network includes an image acquisition array and a point cloud scanning device, and the image acquisition angles include at least a top-down angle and a side-down angle.

[0034] Specifically, the dual-source monitoring network includes an image acquisition array and a point cloud scanning device. This design provides omnidirectional monitoring of aluminum cans, enabling more accurate identification and handling of potential spill risks. The image acquisition angles include at least a top-down view and a side view. The top-down view captures images of the can from above, clearly showing the top and sides, while the side view captures images of the can from the side, clearly showing the sides, top, and bottom. This means that images of the can can be captured from multiple angles, resulting in more comprehensive information.

[0035] Further, obtain the outline of the aluminum can, as shown in the attached image. Figure 2 As shown, this application includes:

[0036] The image set is localized, fused, and inpainted according to the acquisition angle to obtain a three-dimensional image of the can; the three-dimensional image of the can is edge-tracked and contour-fitted based on an edge detection algorithm to obtain a first contour; the three-dimensional image of the can is feature-extracted through a contour analysis model to output a second contour, wherein the contour analysis model is constructed based on a convolutional neural network; the first contour and the second contour are fitted to obtain the contour of the can.

[0037] Specifically, firstly, the image set is located and fused according to the acquisition angle. Image registration technology is used to align images from different angles to the same coordinate system. Then, the registered images are fused to create a single image containing information from multiple angles. Further, image restoration is performed on the single image, such as denoising and enhancement processing on the fused image to improve image quality. Image restoration is performed on hidden or occluded areas, such as using multi-view information to estimate the shape and texture of hidden areas. Finally, a 3D image of the aluminum can is obtained.

[0038] First, an edge detection algorithm is selected. Commonly used edge detection algorithms include the Canny edge detector and the Sobel operator. The appropriate algorithm can be selected according to the actual scenario. Next, edge tracking is performed on the 3D image of the can based on the edge detection algorithm, that is, tracking along the detected edges to obtain complete edge information. Then, curve fitting algorithms such as the least squares method are used to fit the tracked edges to obtain a smooth contour curve. The fitted contour is then refined to remove discontinuous edge points, resulting in a more accurate contour. Finally, key contour features, such as the length, width, and center point of the contour, are extracted from the fitted contour to construct the first contour.

[0039] Convolutional neural networks (CNNs) are deep learning models suitable for processing data such as images and videos. Through multiple convolutional and pooling layers, they can automatically learn the features of the data. A contour analysis model is constructed based on a CNN. The input data for the contour analysis model is an image of an aluminum can, and the output data is the contour of the aluminum can. Sample training data is retrieved to train the contour analysis model until it converges. Then, the contour analysis model extracts features from the 3D image of the aluminum can and outputs a second contour.

[0040] Further, based on a contour fitting algorithm, the first and second contours are fitted. For example, the least squares method is used for contour fitting. First, the coordinates of the first and second contours are used as the input data for fitting. Then, the fitting model is initialized according to the selected fitting algorithm. Finally, the model parameters are iteratively adjusted to achieve the highest best match between the fitting result and the actual contour. Finally, based on the optimal fitting result, the contour of the can is reconstructed using CAD software to obtain the can contour. By extracting contour features based on edge detection algorithms and convolutional neural networks, and fitting the extracted contour feature structure to obtain the can contour, the advantages of the two algorithms can be combined, reducing contour extraction errors and further improving the accuracy and reliability of obtaining the can contour.

[0041] Furthermore, a three-dimensional model of the aluminum can is generated, which includes:

[0042] Using the outline of the aluminum can as a constraint, the point cloud set is randomly fitted once to obtain a first fitting result, and a first fitting degree is calculated, where the first fitting degree is the ratio of the number of point clouds falling within the outline of the aluminum can to the total number of point clouds in the point cloud set; the point cloud set is then randomly fitted a second time using the outline of the aluminum can as a constraint to obtain a second fitting result and a second fitting degree; the fitting is iterated until a predetermined number of times is met, and the fitting result with the highest fitting degree is output as the point cloud fitting result. Based on the point cloud fitting result, a 3D model is generated to produce a 3D model of the aluminum can.

[0043] Specifically, using the outline of the aluminum can as a constraint, a first random fitting is performed on the point cloud set, that is, the point cloud data is randomly placed within the outline of the aluminum can to obtain a first fitting result. Based on the first fitting result, a first degree of fit is calculated, where the first degree of fit is the ratio of the number of point clouds falling within the outline of the aluminum can to the total number of point clouds in the point cloud set. The higher the degree of fit, the better the point cloud fitting effect. Then, using the outline of the aluminum can as a constraint again, a second random fitting is performed on the point cloud set to obtain a second fitting result and a second degree of fit. The same method is used for iterative fitting until a predetermined number of times is met, and the fitting result with the highest degree of fit is output as the point cloud fitting result. Finally, a 3D model is generated based on the point cloud fitting result to produce a 3D model of the aluminum can.

[0044] Furthermore, this application defines multiple transmission areas, including:

[0045] Based on the conveyor belt structure topology, the conveyor belt is divided once according to the tilt direction to generate multiple conveying sections; the multiple conveying sections are then divided a second time according to the tilt angle to obtain multiple conveying areas.

[0046] Specifically, based on the conveyor belt structure topology, the conveyor belt is first divided according to the tilt direction, that is, adjacent conveyor belts belonging to different tilt directions are divided to obtain multiple conveying intervals; then, the multiple conveying intervals are further divided according to the tilt angle, that is, the same conveying interval is divided again according to the tilt angle to obtain multiple conveying areas.

[0047] Furthermore, this application further divides the multiple transmission intervals according to the tilt angle, including:

[0048] A first transmission interval is selected from the plurality of transmission intervals, and the tilt angles of multiple regions within the first transmission interval are obtained. The tilt angles of the first region and the second region are selected, and the deviation between the tilt angles of the first region and the second region is calculated to obtain a first angle difference. The first region and the second region are any two adjacent regions within the first transmission interval. If the first angle difference is less than or equal to a predetermined angle threshold, the first region and the second region are clustered into the same region. If the first angle difference is greater than the predetermined angle threshold, a first dividing line is set between the first region and the second region. Iterative division is performed to determine multiple dividing lines. The first transmission interval is divided using multiple dividing lines to determine multiple first transmission regions, which are then added to the plurality of transmission regions.

[0049] Specifically, firstly, any one of the multiple transmission intervals is randomly selected and designated as the first transmission interval. Then, the tilt angles of multiple regions within the first transmission interval are obtained. Further, the first transmission interval is divided into multiple regions according to a preset step size, and a first region and a second region are randomly selected from these regions, where the first region and the second region are any two adjacent regions within the first transmission interval. Next, the tilt angles of the first region and the second region are obtained, and a deviation calculation is performed between the two tilt angles to obtain a first angle difference, where the first angle difference is the absolute value of the difference between the tilt angle of the first region and the tilt angle of the second region.

[0050] Next, the first angle difference is judged based on a predetermined angle threshold. If the first angle difference is less than or equal to the predetermined angle threshold, the first region and the second region are clustered into the same region. If the first angle difference is greater than the predetermined angle threshold, a first dividing line is set between the first region and the second region. The same method is used for iterative division until the first transmission interval is divided and multiple dividing lines are determined. Finally, the first transmission interval is divided using multiple dividing lines to determine multiple first transmission regions, which are added to the multiple transmission regions to obtain the multiple transmission regions.

[0051] Furthermore, this application includes predicting the tilt probability for the plurality of transmission areas, comprising:

[0052] Constrained by the properties of the target aluminum can and the surface roughness of the conveyor belt, a sample dataset is obtained based on industrial internet retrieval. The sample data includes the bottom contact ratio, bottom contact position, tilt angle, tilt direction, and tipping risk coefficient of the sample. N prediction operators are configured, and supervised learning and cross-validation are performed on each of the N prediction operators using the sample dataset to obtain N convergent prediction units, which are used to construct a tipping prediction model. The output of the tipping prediction model is the average of the outputs of the N convergent prediction units. The tipping probability is predicted for the multiple conveying areas using the tipping prediction model, and multiple tipping risk coefficients are output.

[0053] Specifically, firstly, using the target aluminum can's attribute characteristics and the conveyor belt surface roughness as retrieval constraints, a source information retrieval is performed based on the Industrial Internet to obtain a sample dataset. This sample data includes the sample bottom contact ratio, sample bottom contact position, sample tilt angle, sample tilt direction, and sample tipping risk coefficient. Next, N prediction operators are configured. These prediction operators are algorithms for tipping prediction and can be selected according to the scenario, such as BP neural networks, support vector machines, and random decision forests.

[0054] Next, the sample dataset is divided into N equal parts, and selected N times with replacement to construct the first dataset. This process is repeated N times to obtain N datasets. Then, supervised learning is performed on N prediction operators using each of the N datasets, for example, using backpropagation and a loss function for supervised training. After a predetermined number of training iterations, the datasets are interacted to validate the N prediction operators, resulting in N convergent prediction units that meet the expected convergence conditions. Finally, based on the principle of ensemble learning, a tilt prediction model is constructed from the N convergent prediction units, where the output of the tilt prediction model is the average of the outputs of the N convergent prediction units.

[0055] Finally, the tipping probability is predicted for each of the multiple transport areas using the tipping prediction model, outputting multiple tipping risk coefficients. By constructing a tipping prediction model based on the principle of ensemble learning and multiple prediction operators, the advantages of multiple prediction operators can be combined to improve the accuracy and robustness of the tipping prediction model.

[0056] Furthermore, to obtain the surface roughness of the conveyor belt, this application includes:

[0057] Constrained by the material properties and service life of the conveyor belt, a second sample dataset is retrieved, which includes sample texture features, sample shape features, sample defect features, and sample surface roughness. The roughness prediction unit is trained under supervision using the second sample dataset to obtain a roughness prediction unit that meets the expected convergence conditions. A roughness analysis model is constructed by combining the roughness prediction unit with the surface feature extraction unit, where the feature extraction indicators of the surface feature extraction unit include texture features, shape features, and defect features. The conveyor belt image is collected and input into the roughness analysis model, and the surface roughness of the conveyor belt is output.

[0058] Specifically, the material properties and service life of the conveyor belt are used as constraints. The material properties include material type, tensile strength, elastic modulus, etc. The service life is related to the aging degree of the conveyor belt. Information retrieval is performed based on the Industrial Internet to obtain a second sample dataset. The second sample data includes sample texture features, sample shape features, sample defect features, and sample surface roughness.

[0059] Next, a surface feature extraction unit is constructed based on a convolutional neural network. The feature extraction metrics of the surface feature extraction unit include texture features, shape features, and defect features. A roughness prediction unit is constructed based on a backpropagation neural network, and the roughness prediction unit is trained under supervision using the second sample dataset to obtain a roughness prediction unit that meets the expected convergence conditions. A roughness analysis model is constructed based on the roughness prediction unit and the surface feature extraction unit. Finally, images of the conveyor belt are acquired and input into the roughness analysis model, and the surface roughness of the conveyor belt is output.

[0060] Furthermore, before outputting the surface roughness of the conveyor belt, this application includes:

[0061] The ambient temperature and humidity of the area where the conveyor belt is located are collected. Based on the ambient temperature and humidity, the roughness of the conveyor belt is analyzed to determine the environmental influence coefficient. The surface roughness of the conveyor belt is compensated according to the environmental influence coefficient.

[0062] Specifically, firstly, ambient temperature and humidity data are collected in real time using temperature and humidity sensors installed near the conveyor belt. Next, a mathematical model is established to analyze the relationship between surface roughness and temperature and humidity, and the impact of ambient temperature and humidity on the conveyor belt surface roughness is analyzed using this model. Furthermore, the ambient temperature and humidity are input into the mathematical model for roughness influence analysis to determine the environmental influence coefficient. Finally, the conveyor belt surface roughness is compensated based on the environmental influence coefficient, i.e., the conveyor belt surface roughness is optimized and adjusted according to the environmental influence coefficient, thereby further improving the accuracy and rationality of the conveyor belt surface roughness.

[0063] In summary, the tank tipping detection system provided in this application has the following technical advantages:

[0064] 1. By monitoring and acquiring image and point cloud data of aluminum cans on the conveyor belt, the outline of the cans is determined based on image data analysis. Then, the point cloud data is fitted based on the can outline to construct a 3D model of the cans, which can improve the accuracy of the 3D model construction. Next, in 3D space, bottom contact analysis is performed on the 3D model of the cans and the horizontal plane to determine the bottom contact ratio and position. Furthermore, based on the topology of the conveyor belt structure, the conveyor belt is divided into regions according to the tilt angle and tilt direction to determine multiple conveying areas. Finally, based on the attribute characteristics of the target aluminum cans and the surface roughness of the conveyor belt... The bottom contact ratio and bottom contact position are used to construct a tipping prediction model based on machine learning. This model predicts the tipping probability of multiple conveying areas and generates multiple tipping risk coefficients. Furthermore, conveying areas with tipping risk coefficients exceeding a preset threshold are designated as warning zones. Finally, before the target can reaches the first warning zone, a robotic arm removes the target can from the conveyor belt. This improves the accuracy and reliability of can tipping trend prediction, enabling rapid and accurate identification and handling of potentially tipping cans, reducing production line downtime, and ultimately improving the overall production efficiency of cans.

[0065] 2. A 3D model of an aluminum can is constructed by fusing image data and point cloud data. Point cloud data provides high-precision 3D information about the surface of the aluminum can, while image data provides surface features and contour information. This method combines the advantages of both types of data to provide more comprehensive, accurate, and robust data, thereby further improving the accuracy and reliability of the 3D model construction of the aluminum can.

[0066] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0067] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A can body pour-out detection system characterized by, include: The data acquisition module is used to collect data on the target cans on the conveyor belt in real time based on the dual-source monitoring network and obtain the dual-source monitoring dataset. The dual-source monitoring network is deployed at the beginning of the conveyor belt, and the dual-source monitoring dataset includes an image set and a point cloud set. The point cloud fitting module is used to perform contour recognition using the image set, obtain the contour of the aluminum can, and fit the point cloud set based on the aluminum can contour to generate a three-dimensional model of the aluminum can. The bottom contact analysis module is used to perform bottom contact analysis based on the three-dimensional model of the can in three-dimensional space to determine the bottom contact characteristics, wherein the bottom contact characteristics include the bottom contact ratio and the bottom contact position. The region division module is used to read the topology of the conveyor belt structure, divide the conveyor belt into regions according to the tilt angle and tilt direction, and determine multiple conveying regions. The tipping probability prediction module is used to predict the tipping probability of the multiple conveying areas based on the attribute characteristics of the target can, the surface roughness of the conveyor belt, the bottom contact ratio, and the bottom contact position, and generate multiple tipping risk coefficients. The warning zone setting module is used to set the conveyor area corresponding to the dumping risk coefficient that is greater than the preset threshold as the warning zone. If the warning zone is not 0, the target can will be removed from the conveyor belt by a robotic arm before the target can reaches the first warning zone. The process of predicting the tilt probability for the multiple transmission areas includes: Constrained by the properties of the target aluminum can and the surface roughness of the conveyor belt, a sample dataset is obtained based on industrial internet retrieval. The sample data includes the bottom contact ratio of the sample, the bottom contact position of the sample, the tilt angle of the sample, the tilt direction of the sample, and the tipping risk coefficient of the sample. Configure N prediction operators, and use the sample dataset to perform supervised learning and cross-validation on the N prediction operators respectively to obtain N convergent prediction units to construct a tilt prediction model, wherein the output of the tilt prediction model is the average of the output results of the N convergent prediction units; The tipping prediction model is used to predict the tipping probability of the multiple transport areas respectively, and multiple tipping risk coefficients are output. The process of obtaining the surface roughness of the conveyor belt includes: Constrained by the material properties and service life of the conveyor belt, a second sample dataset is retrieved, which includes sample texture features, sample shape features, sample defect features and sample surface roughness. The roughness prediction unit is trained under supervision using the second sample dataset to obtain a roughness prediction unit that meets the expected convergence condition. A roughness analysis model is constructed by combining it with the surface feature extraction unit. The feature extraction index of the surface feature extraction unit includes texture features, shape features and defect features. The roughness analysis model is input by acquiring images of the conveyor belt and outputs the surface roughness of the conveyor belt.

2. A pour detection system for a can body as defined in claim 1, wherein The dual-source monitoring network includes an image acquisition array and a point cloud scanning device, and the image acquisition angles include at least a top-down angle and a side-down angle.

3. The tank tipping detection system according to claim 2, characterized in that, Obtain the outline of the aluminum can, including: The image set is located, fused, and insulated according to the acquisition angle to obtain a three-dimensional image of the aluminum can; The edge detection algorithm is used to perform edge tracking and contour fitting on the three-dimensional image of the can to obtain the first contour. The contour analysis model is used to extract features from the three-dimensional image of the can and output a second contour. The contour analysis model is constructed based on a convolutional neural network. The contour of the can is obtained by fitting the first contour and the second contour.

4. The tank tipping detection system according to claim 1, characterized in that, Generate a 3D model of the aluminum can, including: Using the outline of the aluminum can as a constraint, the point cloud set is randomly fitted once to obtain a first fitting result, and the first fitting degree is calculated, wherein the first fitting degree is the ratio of the number of point clouds falling within the outline of the aluminum can to the total number of point clouds in the point cloud set. Using the outline of the aluminum can as a constraint, the point cloud is subjected to a second random fitting to obtain a second fitting result and a second fitting degree. Perform iterative fitting until a predetermined number of times is met, and output the fitting result with the highest fitting degree as the point cloud fitting result. Based on the point cloud fitting result, perform 3D modeling to generate the 3D model of the can.

5. The tank tipping detection system according to claim 1, characterized in that, Multiple transport areas have been identified, including: Based on the conveyor belt structure topology, the conveyor belt is divided once according to the inclined direction to generate multiple conveying sections; The multiple transmission intervals are further divided according to the tilt angle to obtain multiple transmission areas.

6. The tank tipping detection system according to claim 5, characterized in that, The multiple transmission intervals are further divided according to the tilt angle, including: Select a first transmission interval from the plurality of transmission intervals, and obtain the tilt angles of multiple regions within the first transmission interval; Select the tilt angle of the first region and the tilt angle of the second region, and calculate the deviation between the tilt angle of the first region and the tilt angle of the second region to obtain the first angle difference, wherein the first region and the second region are any two adjacent regions within the first transmission interval. If the first angle difference is less than or equal to a predetermined angle threshold, the first region and the second region are clustered into the same region; if the first angle difference is greater than the predetermined angle threshold, a first dividing line is set between the first region and the second region. The process involves iterative division to determine multiple dividing lines, which are then used to divide the first transmission interval into multiple first transmission areas, which are then added to the multiple transmission areas.

7. The tank tilting detection system according to claim 1, characterized in that, The output of the conveyor belt surface roughness also includes, prior to: The ambient temperature and humidity of the area where the conveyor belt is located are collected, and the roughness influence of the conveyor belt is analyzed based on the ambient temperature and humidity to determine the environmental influence coefficient. The surface roughness of the conveyor belt is compensated based on the environmental impact coefficient.

Images