Vibrating screen blockage identification and cleaning method and system based on 3D laser scanning point cloud

Abstract

The present invention proposes a method and system for identifying and cleaning blockages in a vibrating screen based on three-dimensional laser scanning point clouds, which relates to the technical fields of point cloud processing and vibrating screens. Three-dimensional laser scanning is used to obtain point cloud data of the vibrating screen before and after startup. The point cloud is voxelized and input into a trained blockage identification and classification neural network model. The identification and classification of the blockage identification and classification neural network model are used to obtain the blockage identification and classification results corresponding to each voxel under voxelization. Finally, according to the identification and classification results, the blockage of the vibrating screen is cleaned. This process adopts three-dimensional laser scanning and point cloud processing, combined with deep learning neural networks. Compared with mechanical vibration methods, it can identify blockages in a targeted manner. Compared with manual blockage identification methods, it is more efficient, which is conducive to improving production efficiency, clearing blockages more thoroughly, and avoiding product quality defects.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud processing and vibrating screens, and more specifically, to a method and system for identifying and cleaning blockages in a vibrating screen based on three-dimensional laser scanning point clouds. Background Art

[0002] A vibrating screen is a device that uses the reciprocating vibration generated by the excitation of a vibrator to work. It uses mechanical vibration and the gravity of objects to achieve the purpose of filtering and screening material powders. It is widely used in many industries.

[0003] However, during operation, the vibrating screen mesh can easily become clogged with material. A clogged vibrating screen loses its normal screening function, impacting processing efficiency. This is particularly true for underwater strand pelletizing lines, where the high water content and poor fluidity of pellets increase the likelihood of clogging the mesh. When switching from one batch of material to another, the screen mesh must be cleaned thoroughly. Otherwise, pellets stuck in the mesh could become loose and mixed into the next batch, causing product quality defects.

[0004] The prior art discloses a solution for a single-axis vibrating screen to avoid clogging of the screen holes. The solution drives the docking block to move in the groove through a moving mechanism, so that the connecting rod connected to the docking block contacts the cam provided at the end and the protrusion. The cam rotates along the inclined surface of the wedge-shaped protrusion, driving the connecting rod to rotate in the docking block and driving the torsion spring to rotate, storing potential energy. When the cam passes the highest point of the protrusion, the torsion spring releases the potential energy, causing the knocking arm on the connecting rod to swing, knocking the screen on the vibrating screen body, and achieving the purpose of cleaning the clogged screen. However, this method of cleaning the blockage by knocking on the screen is a holistic blockage cleaning method, which does not have a targeted cleaning effect on the blockage. There are cases where a small number of blockages are not cleaned thoroughly, affecting the production and application of the screen, and easily causing the mixing of granular materials of different products, affecting product quality. In addition, the device has a complex structure and is installed on the vibrating screen. Under the long-term vibration of the vibrating screen, it is easy to cause failures such as component falling off or fatigue damage, affecting production efficiency. If this mechanical knocking and vibration method is not adopted, and the blockage of the screen is directly identified by the naked eye, and then the blockage is targetedly cleaned using a cleaning device, the blockage identification accuracy is not high, the blockage cleaning is not thorough, and it is time-consuming and labor-intensive, which is not conducive to improving production efficiency. Summary of the Invention

[0005] In order to solve the problem of incomplete cleaning in the current method of cleaning the vibrating screen mesh blockage, the present invention proposes a vibrating screen mesh blockage identification and cleaning method and system based on three-dimensional laser scanning point cloud, which can specifically identify and efficiently clean blockages, thereby improving production efficiency, clearing blockages more thoroughly, and avoiding product quality defects.

[0006] In order to achieve the above technical effects, the technical solutions of the present invention are as follows:

[0007] In a first aspect, the present application proposes a method for identifying and cleaning a vibrating screen mesh blockage based on a three-dimensional laser scanning point cloud, which is characterized by comprising the following steps:

[0008] Before starting up, use 3D laser scanning to obtain the original state point cloud data of the vibrating screen mesh;

[0009] After startup, 3D laser scanning is used to obtain the new state point cloud data of the vibrating screen mesh;

[0010] The new state point cloud data is registered with the original state point cloud data, and both the original state point cloud data and the new state point cloud data after registration are voxelized;

[0011] The registered original state point cloud data and the new state point cloud data are input into the trained congestion recognition and classification neural network model to obtain the congestion recognition and classification results corresponding to each voxel under voxelization;

[0012] According to the identification and classification results, the blockage of the vibrating screen is cleared.

[0013] Preferably, after obtaining the original state point cloud data of the vibrating screen mesh by three-dimensional laser scanning, the following steps are also included:

[0014] Use the original point cloud data to build the original 3D model of the vibrating screen;

[0015] Compare the aperture of each mesh of the vibrating screen on the original three-dimensional model with the preset aperture value, and record the number of meshes whose aperture exceeds the preset aperture value;

[0016] Calculate the ratio of the number of sieve holes whose aperture exceeds the preset aperture value to the total number of sieve holes in the screen, and determine whether the ratio exceeds the preset value of the screen replacement standard. If so, replace the vibrating screen with a new one; otherwise, start the machine and use 3D laser scanning to obtain the new state point cloud data of the vibrating screen.

[0017] According to the above technical means, before cleaning the sieve hole blockage, if it is clear that the vibrating screen mesh has reached the scrap standard, the vibrating screen mesh blockage cleaning will no longer be performed, thereby improving production efficiency.

[0018] Preferably, a preprocessing operation is performed on the obtained original state point cloud data and the new state point cloud data of the vibrating screen mesh; the preprocessing operation includes: filtering to remove noise points, and also includes: downsampling processing;

[0019] When both the original state point cloud data and the new state point cloud data after registration are voxelized, the original state point cloud data and the new state point cloud data after registration are mapped to each voxel respectively, a unique identifier is assigned to each voxel, and whether there is point cloud data in the voxel is recorded.

[0020] Preferably, the congestion recognition and classification neural network model is an improved DGCNN model. The improved DGCNN model is based on the original network architecture of the DGCNN model, removes the spatial conversion network, adds connection layer features, freezes the feature extraction network, and retrains the separator using the trained feature vector; during the training of the improved DGCNN model, the simulated annealing algorithm is used to optimize the learning rate of the trainer;

[0021] The trained improved DGCNN model classifies each point cloud data based on the feature relationship between the point cloud data and several neighboring point cloud data, outputs the probability distribution of the point cloud data on each recognition and classification result, and takes the recognition and classification result with the highest probability as the recognition and classification result of the point cloud data. The recognition and classification results include normal and blocked.

[0022] According to the mapping relationship between each point cloud data and voxel, the blockage identification and classification results corresponding to each voxel under voxelization are obtained.

[0023] Based on the above technical means, the improved DGCNN model uses the simulated annealing algorithm to optimize the learning rate of the trainer compared to the original traditional DGCNN model. The learning rate is not fixed and it is not easy to fall into the local optimal solution under large parameters, thereby increasing the probability of finding the global optimal solution of the objective function, and ultimately improving the precision and accuracy of the DGCNN model and the prediction accuracy.

[0024] Preferably, the vibrating screen mesh blockage identification and cleaning method based on three-dimensional laser scanning point cloud further comprises: grading the blockage situation according to the identification and classification results;

[0025] If the number of blocked mesh holes accounts for less than 10% of the total number of mesh holes, the blockage level is mild.

[0026] If the number of blocked screen holes accounts for greater than or equal to 10% and less than or equal to 50% of the total number of screen holes, the blockage level is moderate;

[0027] If the number of blocked holes in the screen accounts for more than 50% of the total number of holes in the screen, the blockage level is high;

[0028] After the blockage is graded, voxels with the same blockage level are dyed with the same color to visualize the blockage.

[0029] According to the above technical means, after the blockage situation is graded, it is convenient to take different degrees of cleaning measures for the blockage of the vibrating screen according to the blockage classification results, so as to achieve targeted cleaning, which is conducive to cleaning the blockage more thoroughly and avoiding product quality defects.

[0030] Preferably, according to the identification and classification results, the process of clearing the blockage of the vibrating screen includes:

[0031] Use a manipulator to hold the screen cleaning device, and based on the visualization of the blockage situation, press the screen cleaning device into the blocked screen holes to clear the blockage;

[0032] After cleaning is completed, use the image acquisition device to take a picture of the screen, and confirm the cleaning status of the screen blockage based on the captured image. If the screen holes are not blocked after cleaning, continue the production of the next batch of materials. If there are still blockages in the screen holes that have not been thoroughly cleaned, remove the entire screen from the vibrating screen and install the spare screen on the vibrating screen to continue the production of the next batch of materials.

[0033] During production, the whole screen removed from the vibrating screen is cleaned by using a screen cleaning device to remove the blockage in the screen holes that have not been thoroughly cleaned.

[0034] According to the above technical means, the cleaning of the blockage of the sieve holes is more targeted, which is conducive to more thorough cleaning of the blockage. Moreover, during production, the sieve cleaning device is used to clean the blockage of the sieve holes that has not been thoroughly cleaned, which is conducive to improving production efficiency.

[0035] Preferably, when the integral screen is to be removed from the vibrating screen, the lock between the integral screen and the vibrating screen is released, and the manipulator clamps the integral screen away from the installation position of the vibrating screen; when the spare screen is to be installed on the vibrating screen, the manipulator clamps the spare screen to the installation position of the vibrating screen and locks the relative position of the integral screen and the vibrating screen.

[0036] In a second aspect, the present application proposes a vibrating screen mesh blockage identification and cleaning system based on a three-dimensional laser scanning point cloud. The system is used to implement the vibrating screen mesh blockage identification and cleaning method based on a three-dimensional laser scanning point cloud, and includes a three-dimensional laser scanning unit, a registration voxelization unit, a blockage identification unit, a central processing unit, and a screen mesh cleaning device;

[0037] The three-dimensional laser scanning unit is used to obtain the original state point cloud data of the vibrating screen mesh by three-dimensional laser scanning before starting the machine; and obtain the new state point cloud data of the vibrating screen mesh by three-dimensional laser scanning after starting the machine;

[0038] A registration voxelization unit is used to register the new state point cloud data with the original state point cloud data, and voxelize both the registered original state point cloud data and the new state point cloud data;

[0039] A congestion identification unit is used to receive the registered original state point cloud data and the new state point cloud data, and output the congestion identification and classification results corresponding to each voxel in voxelization;

[0040] a central processing unit, configured to receive and analyze the blockage classification result corresponding to each voxel in the voxelization output by the blockage recognition unit, and output a screen mesh blockage cleaning instruction to the screen cleaning device;

[0041] The screen cleaning device is used to clean the clogged screen holes.

[0042] Preferably, the system further comprises a congestion situation grading unit; the congestion situation grading unit grades the congestion situation according to the identification and classification result, and outputs the congestion situation grading result.

[0043] According to the above technical means, after the blockage situation is graded, it is convenient to take different degrees of cleaning measures for the blockage of the vibrating screen according to the blockage classification results, so as to achieve targeted cleaning, which is conducive to cleaning the blockage more thoroughly and avoiding product quality defects.

[0044] Preferably, it also includes a manipulator for clamping and moving the spare screen or the screen cleaning device according to the instruction of the central processing unit;

[0045] The screen cleaning device includes a cleaning needle and a base plate. The cleaning needle is arranged on the base plate. The position distribution of the cleaning needle on the base plate is consistent with the position distribution of the sieve holes on the screen. The diameter of the cleaning needle is smaller than the diameter of the sieve holes.

[0046] According to the above technical means, the position distribution of the cleaning needles on the substrate is consistent with the position distribution of the screen holes on the screen. The cleaning needles can be pressed into the blocked holes at one time, and according to the blockage classification results, different degrees of cleaning measures can be taken for the blockage of the vibrating screen. Compared with the mechanical vibration-type overall blockage cleaning, the use of the cleaning needles included in the screen cleaning device can also achieve more targeted blockage cleaning.

[0047] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0048] The present invention proposes a method and system for identifying and cleaning blockages in a vibrating screen based on three-dimensional laser scanning point clouds. The point cloud data of the vibrating screen before and after startup is obtained through three-dimensional laser scanning. After the point cloud is voxelized, it is input into a trained blockage identification and classification neural network model. The identification and classification of the blockage identification and classification neural network model is used to obtain the blockage identification and classification results corresponding to each voxel under voxelization. Finally, the blockage in the vibrating screen is cleaned according to the identification and classification results. This process adopts three-dimensional laser scanning and point cloud processing, combined with deep learning neural networks. Compared with mechanical vibration methods, it can identify blockages in a targeted manner. Compared with manual blockage identification methods, it is more efficient, which is conducive to improving production efficiency, clearing blockages more thoroughly, and avoiding product quality defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic diagram showing a flow chart of a method for identifying and cleaning a vibrating screen mesh blockage based on a three-dimensional laser scanning point cloud, according to an embodiment of the present invention;

[0050] Figure 2 A diagram showing the overall structure of removing the entire screen from the vibrating screen by using a manipulator to clamp the entire screen, as proposed in an embodiment of the present invention;

[0051] Figure 3 A basic architecture diagram of a traditional DGCNN model proposed in an embodiment of the present invention is shown;

[0052] Figure 4 A diagram showing the basic architecture of the improved DGCNN model proposed in an embodiment of the present invention;

[0053] Figure 5 A schematic diagram showing the structure of the locking mechanism of the present invention;

[0054] Figure 6 A schematic diagram showing the installation and removal of the screen proposed in an embodiment of the present invention by plugging and unplugging the guide rails;

[0055] Figure 7 A schematic diagram showing a structure of a vibrating screen blockage identification and cleaning system based on three-dimensional laser scanning point cloud proposed in an embodiment of the present invention;

[0056] Figure 8 Another structural schematic diagram of a vibrating screen blockage identification and cleaning system based on three-dimensional laser scanning point cloud proposed in an embodiment of the present invention;

[0057] In the figure: 100, vibrating screen; 110, screen; 120, guide rail; 130, slider; 200, manipulator; 300, screen cleaning device; 400, locking mechanism; 410, clamping part; 420, buckle part; 430, cylinder; 440, transmission mechanism. DETAILED DESCRIPTION

[0058] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0059] In order to better illustrate this embodiment, some parts of the drawings may be omitted, enlarged, or reduced, and do not represent the actual size;

[0060] It is understandable to those skilled in the art that descriptions of certain well-known contents may be omitted in the drawings.

[0061] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0062] The positional relationships described in the drawings are for illustrative purposes only and should not be construed as limiting this patent;

[0063] Example 1

[0064] This embodiment proposes a method for identifying and cleaning the blockage of the vibrating screen 110 based on a three-dimensional laser scanning point cloud. The flow chart of the method is shown in FIG. Figure 1 As shown, the following steps are included:

[0065] S1: Before starting up, use 3D laser scanning to obtain the original state point cloud data of the vibrating screen mesh 110;

[0066] S2: After the production is completed, the new state point cloud data of the vibrating screen 110 is obtained by using three-dimensional laser scanning;

[0067] S3: registering the new state point cloud data with the original state point cloud data, and voxelizing both the registered original state point cloud data and the new state point cloud data;

[0068] S4: Inputting the registered original state point cloud data and the new state point cloud data into the trained congestion recognition and classification neural network model to obtain the congestion recognition and classification results corresponding to each voxel under voxelization;

[0069] S5: According to the identification and classification results, the blockage of the vibrating screen 110 is cleared.

[0070] In this embodiment, startup refers to starting the vibrating screen 100. Three-dimensional laser scanning can directly obtain point cloud data of the surface of the vibrating screen mesh 110. This non-contact measurement is fast, highly accurate, and realistic, and is not affected by the complexity of the surface of the vibrating screen mesh 110. After obtaining the point cloud data, the point cloud data can be used to reconstruct a three-dimensional model. Therefore, after obtaining the original state point cloud data of the vibrating screen mesh 110 using three-dimensional laser scanning, this embodiment also includes the following steps:

[0071] S101: Construct an original three-dimensional model of the vibrating screen mesh 110 using the original point cloud data.

[0072] S102: Compare the aperture of each mesh of the vibrating screen 110 on the original three-dimensional model with a preset aperture value, and record the number of meshes whose aperture exceeds the preset aperture value.

[0073] S103: Calculate the ratio of the number of sieve holes whose aperture exceeds the preset aperture value to the total number of sieve holes in the screen 110, and determine whether the ratio exceeds the preset value of the scrap standard for the screen 110. If so, scrap the vibrating screen 110 and replace it with a new one; otherwise, start the machine and use three-dimensional laser scanning to obtain the new state point cloud data of the vibrating screen 110.

[0074] Among them, the aperture preset value is determined according to the vibrating screen mesh 110 actually used in production, and the scrap standard preset value is set to 50%. When the ratio is greater than 50%, it means that the aperture of more than 50% of the sieve holes has become larger, and the screen mesh 110 has reached the scrap standard and needs to be replaced in time. If it is clear that the vibrating screen mesh 110 has reached the scrap standard, the vibrating screen mesh 110 will no longer be cleaned for blockage, which is beneficial to improving production efficiency.

[0075] In this embodiment, considering the noise and outliers in the point cloud data due to various reasons when acquiring point cloud data, it is necessary to perform preprocessing operations on the original state point cloud data and the new state point cloud data of the acquired vibrating screen mesh 110. The preprocessing operations of this embodiment include: filtering to remove noise points, and also include: downsampling processing.

[0076] Then, this embodiment uses the ICP (Iterative Closest Point) algorithm to align the new state point cloud data with the original state point cloud data to ensure that the point cloud data results of the two 3D laser scans are in the same coordinate system.

[0077] When voxelizing the original state point cloud data and the new state point cloud data after alignment, an appropriate voxel size is set to ensure that the details of the screen 110 can be reflected without being too complicated. The original state point cloud data and the new state point cloud data after alignment are respectively mapped to each voxel, a unique identifier is assigned to each voxel, and whether there is point cloud data in the voxel is recorded.

[0078] In this embodiment, the congestion recognition and classification neural network model is an improved DGCNN (Dynamic Graph Convolutional Neural Network, DGCNN for short) model. The traditional DGCNN model is Figure 3As shown, the model architecture includes model architectures for classification (top branch) and segmentation (bottom branch). Among them, classification takes "n points" as input, calculates an edge feature set of size k for each point in the EdgeConv layer, summarizes the features in each feature set, and calculates the EdgeConv response for the corresponding point. The output features of the last EdgeConv layer are aggregated globally to form a one-dimensional global descriptor, which is used to generate classification scores for c categories. Segmentation extends classification by connecting the one-dimensional global descriptor of each point and all EdgeConv outputs (as local descriptors). It outputs the classification score of each point for p semantic labels. For ease of illustration, it is drawn Figure 3 The two arcs with arrows shown represent feature concatenation.

[0079] See also Figure 3 , also includes a point cloud transformation module, which aims to align the input point set pairs to the standard space by applying an estimated 3x3 matrix. To estimate the 3x3 matrix, a tensor is used, which contains the coordinates of each point and the coordinate differences between k adjacent points. The EdgeConv module takes a tensor of shape nxf as input and calculates the edge features of each point by applying a multi-layer perceptron (MLP), whose number of neuron layers is defined as {a1, a2, ..., a n}, and generate shape nxa after aggregating adjacent edge features n Tensor of .

[0080] at present Figure 3 The traditional DGCNN model shown in the figure has a large size due to the spatial conversion network, and the similarity of the deep features in the neighborhood cannot generate valuable edges well. In addition, there are too many parameters in training, making it difficult to find the best parameter, and only a local optimal solution may be obtained. The improved DGCNN model proposed in this embodiment is Figure 3 Based on the original network architecture of the traditional DGCNN model shown in the figure, the spatial transformation network is removed, the connection layer features are added, the feature extraction network is frozen, and the separator is retrained using the trained feature vectors. The basic architecture of the model is shown in Figure 4, including an input layer connected in sequence, two consecutive graph convolution layers and local convolution layers, a sorting pooling layer, two consecutive fully connected layers and an output layer. In addition, in the source code trainer of the DGCNN model, the learning rate used is a fixed value. Although global shape information can be extracted by stacking EdgeConv modules or using them in a loop, due to the fixed learning rate, this method is difficult to obtain a global optimal solution, resulting in a higher probability of obtaining a local optimal solution. If the global optimal solution can be obtained, the model classification accuracy will be higher. Therefore, in the process of training the improved DGCNN model, the simulated annealing algorithm is used to optimize the learning rate of the trainer to update the learning rate. If the solution state of the optimized global optimal solution is better than the solution state before optimization, the optimized learning rate will be used as the learning rate for training the improved DGCNN model.

[0081] Before using the trained improved DGCNN model for recognition and classification, point cloud features are extracted for each point cloud data. In point cloud data, features can be geometric attributes such as the position, color, normal direction, curvature, etc. of the point, or higher-level features, such as the relationship between the point and the surrounding points. DGCNN captures these relationships by constructing a k-nearest neighbor graph and extracts high-level feature representations through multi-layer graph convolution operations. The node feature matrix and adjacency matrix are input into the input layer. The node feature matrix has a shape of n×F, where n is the number of nodes and F represents the feature dimension. The adjacency matrix has a shape of n×n, representing the connection relationship between nodes. In the first graph convolution layer and local convolution layer, the graph convolution layer performs a linear transformation on the node feature matrix, transforming it from n×F to n×H1. The adjacency matrix is ​​used for graph convolution to obtain n×H1 features. The local convolution layer calculates the similarity between nodes and uses the softmax function to calculate the weights of neighboring nodes. The features of neighboring nodes are then weighted and aggregated according to the weights to obtain n×H1 features. In the second graph convolution layer and the local convolution layer, the graph convolution layer performs a linear transformation on the node feature matrix, transforming it from n×F to n×H1. The adjacency matrix is ​​used to perform graph convolution operations to obtain n×H2 features. The local convolution layer calculates the similarity between nodes and uses the softmax function to calculate the weights of neighboring nodes. The features of neighboring nodes are then weighted and aggregated according to the weights to obtain n×H2 features. In the sorting pooling layer, the n×H2 features are sorted to obtain k*H2 features. The first k node features after sorting are then selected, resulting in k×H2 features. In the fully connected layer, the pooled features are flattened. Specifically, the sorted and pooled k×H2 features are flattened into a one-dimensional vector 1×(k×H2). The fully connected layer performs feature transformation to obtain 1×m features. Finally, the prediction results are output in the output layer.

[0082] When training the improved DGCNN model, the labels "normal" and "blocked" are set. The trained improved DGCNN model identifies and classifies each point cloud data point based on the feature relationship between the point cloud data point and several neighboring point clouds. The model outputs the probability distribution of each recognition and classification result for the point cloud data point, and the recognition and classification result with the highest probability is used as the recognition and classification result for the point cloud data point. The recognition and classification results include "normal" and "blocked". Finally, based on the mapping relationship between each point cloud data point and voxel, the corresponding blockage recognition and classification result for each voxel is obtained.

[0083] Compared with the original traditional DGCNN model, the improved DGCNN model proposed in this embodiment uses a simulated annealing algorithm to optimize the learning rate of the trainer. The learning rate is not fixed, and it is not easy to fall into the local optimal solution under large parameters, thereby increasing the probability of finding the global optimal solution of the objective function, and ultimately improving the precision and accuracy of the DGCNN model and the prediction accuracy.

[0084] In this embodiment, based on the identification and classification results, the process of clearing the blockage of the vibrating screen 110 includes:

[0085] S401: Using the robot 200 to hold the screen cleaning device 300, based on the visualization of the blockage situation, the screen cleaning device 300 is pressed into the blocked screen hole to clear the blockage;

[0086] S402: After cleaning is completed, the screen 110 is photographed using an image acquisition device. The cleaning status of the screen holes is confirmed based on the photographed image. If the screen holes are not blocked after cleaning, the production of the next batch of materials is continued. If the screen holes are still blocked due to incomplete cleaning, the entire screen 110 is removed from the vibrating screen 100 and a spare screen 110 is installed on the vibrating screen 100, and the production of the next batch of materials is continued;

[0087] S403: During production, the integral screen 110 removed from the vibrating screen 100 is cleaned by using the screen cleaning device 300 to remove the blockage in the screen holes that have not been thoroughly cleaned.

[0088] The above cleaning process makes the cleaning of the blockage of the sieve holes more targeted, which is conducive to more thorough cleaning of the blockage. Moreover, during production, the sieve cleaning device 300 is used to clean the blockage of the sieve holes that has not been thoroughly cleaned, which is conducive to improving production efficiency.

[0089] In this embodiment, when the integral screen 110 is to be removed from the vibrating screen 100, the lock between the integral screen 110 and the vibrating screen 100 is released, and the manipulator 200 clamps the integral screen 110 and detaches it from the installation position of the vibrating screen 100; when the spare screen 110 is to be installed on the vibrating screen 100, the manipulator 200 clamps the spare screen 110 to the installation position of the vibrating screen 100 and locks the relative position of the integral screen 110 and the vibrating screen 100.

[0090] like Figure 2 As shown, the installation position of the vibrating screen 100 is the upper surface of the vibrating screen 100. During installation, the spare screen 110 is moved to the surface of the vibrating screen 100 by the manipulator 200; the locking between the integral screen 110 and the vibrating screen 100 is achieved by the locking mechanism 400, and a plurality of evenly arranged buckles 420 are set on the edges in the length direction of the integral screen 110 and the spare screen 110. The locking mechanism 400 includes cylinders 430 and clamping parts 410, which are equal in number to the buckles 420 and opposite in position. The extension and retraction of the cylinder 430 can realize the buckling connection between the clamping part 410 and the buckle 420 and the release of the connection between the clamping part 410 and the buckle 420; in this embodiment, the extension and retraction direction of the cylinder 430 is vertical, and the movement direction of the clamping part 410 is horizontal. A transmission mechanism 440 is also connected between the cylinder 430 and the clamping part 410 to convert vertical linear motion into horizontal linear motion. The transmission mechanism 440 may be specifically a three-link structure, one end of the three-link is hinged, the other end of one of the links is hinged to the block, and the other ends of the other two links are hinged to different positions of the cylinder 430, such as Figure 5 shown.

[0091] As another specific embodiment, Figure 6 As shown, the integral screen 110 and the spare screen 110 can also be installed and removed by plugging and unplugging the guide rail 120. Specifically, the guide rails 120 are provided on two opposite sides of the vibrating screen 100 along the length direction, and sliders 130 are provided on two opposite sides of the integral screen 110 and the spare screen 110 along the length direction, and the sliders 130 are slidably connected to the guide rails 120. The locking between the integral screen 110 and the vibrating screen 100 is achieved by a locking mechanism 400, which may include locking holes provided on the integral screen 110 and the sliders 130 of the spare screen 110, a locking member provided at the end of the guide rail 120, a spring connected between the locking member and the guide rail 120, and when the locking hole slides to a position corresponding to the locking member, the locking member springs into the locking hole to lock the integral screen 110 or the spare screen 110 to the vibrating screen 100.

[0092] Example 2

[0093] In this embodiment, in addition to being the same as the method proposed in Example 1, the method for identifying and cleaning the blockage of the vibrating screen 110 based on the three-dimensional laser scanning point cloud proposed in this embodiment also includes: grading the blockage situation according to the identification and classification results. Specifically, if the number of blocked sieve holes in the screen 110 accounts for less than 10% of the total number of sieve holes, the blockage situation level is mild blockage. If the number of blocked sieve holes in the screen 110 accounts for greater than or equal to 10% and less than or equal to 50% of the total number of sieve holes in the screen 110, the blockage situation level is moderate blockage. If the number of blocked sieve holes in the screen 110 accounts for greater than 50% of the total number of sieve holes in the screen 110, the blockage situation level is severe blockage.

[0094] After grading the blockage, different levels of cleaning measures are taken for the vibrating screen mesh according to the blockage level. For example, when using a screen cleaning device to press into the blocked mesh to clear the blockage, different levels of cleaning measures can be taken based on the blockage classification results. In the case of mild blockage, the screen cleaning device is pressed into the blocked mesh once to clear the blockage. In the case of moderate blockage, the screen cleaning device is pressed into the blocked mesh twice to clear the blockage. In actual implementation, the specific number of times is determined by taking images to confirm the cleaning status of the mesh blockage. In the case of severe blockage, the screen can be directly replaced.

[0095] In addition, after the blockage situation is graded, in the specific implementation, the voxels of the same blockage situation level can also be dyed with the same color, such as using green to indicate mild blockage, yellow to indicate mild blockage, and red to indicate severe blockage. The blockage situation can be visualized through the display of colors, and a multi-piece integrated screen cleaning device that can match the shape and size of the vibrating screen 110 can be designed to perform targeted cleaning of blockages of different levels after visualization, which is conducive to clearing the blockage more thoroughly and avoiding product quality defects.

[0096] Example 3

[0097] like Figure 7 As shown, this embodiment proposes a vibration screen mesh 110 blockage identification and cleaning system based on three-dimensional laser scanning point cloud, and the system is used to implement the vibration screen mesh 110 blockage identification and cleaning method based on three-dimensional laser scanning point cloud described in Example 1 and Example 2, including a three-dimensional laser scanning unit, a voxelization unit, a blockage identification unit, a central processing unit and a screen cleaning device 300.

[0098] The three-dimensional laser scanning unit is used to obtain the original state point cloud data of the vibrating screen mesh 110 using three-dimensional laser scanning before the vibrating screen 100 is turned on; after the vibrating screen 100 is turned on, the three-dimensional laser scanning unit is used to obtain the new state point cloud data of the vibrating screen mesh 110 using three-dimensional laser scanning. The registration voxelization unit is used to align the new state point cloud data with the original state point cloud data, and voxelize both the registered original state point cloud data and the new state point cloud data. The blockage identification unit is used to receive the registered original state point cloud data and the new state point cloud data, and output the blockage identification and classification results corresponding to each voxel under voxelization. The central processing unit is used to receive and analyze the blockage identification and classification results corresponding to each voxel under voxelization output by the blockage identification unit, and output the blockage cleaning instructions for the screen mesh 110 to the screen cleaning device 300. The screen cleaning device 300 is used to clean the blockage of the screen mesh 110.

[0099] In this embodiment, the congestion recognition unit encapsulates a congestion recognition and classification neural network model, which is a pre-trained improved DGCNN model. Based on the original DGCNN model's network architecture, the improved DGCNN model removes the spatial transformation network, adds connected hierarchical features, freezes the feature extraction network, and retrains the separator using the trained feature vectors. During the training of the improved DGCNN model, a simulated annealing algorithm is used to optimize the learning rate of the trainer. The trained improved DGCNN model classifies each point cloud data point based on the feature relationship between the point cloud data point and several neighboring point clouds. It outputs a probability distribution for each recognition and classification result for the point cloud data point, and selects the recognition and classification result with the highest probability as the recognition and classification result for the point cloud data point. The recognition and classification results include normal and congestion. Based on the mapping relationship between each point cloud data point and voxel, the congestion recognition and classification result corresponding to each voxel is obtained.

[0100] like Figure 8 As shown, the vibration screen mesh 110 blockage identification and cleaning system based on three-dimensional laser scanning point cloud proposed in this embodiment also includes a blockage situation grading unit; the blockage situation grading unit grades the blockage situation according to the identification and classification results, and outputs the blockage situation grading results, including: if the number of blocked mesh holes in the screen mesh 110 accounts for less than 10% of the total number of mesh holes, the blockage situation level is light blockage. If the number of blocked mesh holes in the screen mesh 110 accounts for greater than or equal to 10% and less than or equal to 50% of the total number of mesh holes in the screen mesh 110, the blockage situation level is moderate blockage. If the number of blocked mesh holes in the screen mesh 110 accounts for greater than 50% of the total number of mesh holes in the screen mesh 110, the blockage situation level is severe blockage.

[0101] In this embodiment, a robot arm 200 is further included, which is used to clamp and move the spare screen 110 or the screen cleaning device 300 according to the instructions of the central processing unit. The screen cleaning device 300 includes a cleaning needle and a base plate. The cleaning needle is disposed on the base plate, and the position distribution of the cleaning needle on the base plate is consistent with the position distribution of the screen holes on the screen 110. The diameter of the cleaning needle is slightly smaller than the diameter of the screen hole to ensure that the cleaning needle can be inserted into the screen hole for cleaning. In this embodiment, the cleaning needle can be connected to the base plate by threading or welding.

[0102] In this embodiment, the position distribution of the cleaning needles on the substrate is consistent with the position distribution of the sieve holes on the screen. When performing blockage cleaning, the cleaning needles can be pressed into the blocked holes at one time. On the other hand, the blockage grading unit can output the blockage grading result. When the screen cleaning device is clamped by a manipulator and the screen cleaning device is pressed into the blocked sieve holes to clean the blockage of the sieve holes, different degrees of cleaning measures can be taken for the blockage of the vibrating screen according to the blockage grading result output by the blockage grading unit. For example, in the case of light blockage, the screen cleaning device is pressed into the blocked sieve holes once to clean the lightly blocked sieve holes. In the case of moderate blockage, the screen cleaning device is pressed into the blocked sieve holes twice to clean the moderately blocked sieve holes. In specific implementation, the number of times the screen cleaning device is pressed into the blocked sieve holes is determined based on the cleaning condition of the sieve hole blockage confirmed by taking an image. In the case of severe blockage, the screen can be directly replaced.

[0103] In actual implementation, the system may also include a visualization unit, which receives the blockage classification result output by the blockage classification unit by using the visualization unit, and dyes the voxels of the same blockage level with the same color. After the visualization unit visualizes the blockage according to the coloring result, it designs a screen cleaning device that can match the shape and size of the vibrating screen mesh 110, and performs targeted cleaning on the blockages of different levels after visualization. For example, the substrate of the screen cleaning device is designed to be as large as the shape and size of the screen 110 in six equal parts, because the screen cleaning device with the same shape and size as the screen 110 in six equal parts is relatively light, which is convenient for the robot to operate and convenient for cleaning the blockage areas of different levels after visualization, especially when the sieve holes in a small area of ​​the screen are blocked, it is more convenient to clean and achieve targeted cleaning. Moreover, the number of cleaning needles on the substrate of the screen cleaning device with the same shape and size in six equal parts is small, which reduces the difficulty of aligning with the sieve holes, thereby reducing the precision requirements for the robot.

[0104] The embodiments are provided merely to illustrate the present invention and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications may be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the claims.

Claims

1. A method for identifying and cleaning vibrating screen blockage based on three-dimensional laser scanning point cloud, characterized in that: The following steps are involved: Before starting up, use 3D laser scanning to obtain the original state point cloud data of the vibrating screen mesh; After production is completed, 3D laser scanning is used to obtain the new state point cloud data of the vibrating screen mesh; Performing preprocessing operations on the acquired original state point cloud data and new state point cloud data of the vibrating screen mesh; The pre-processing operation includes: filtering to remove noise points, and also includes: downsampling processing; When the registered original state point cloud data and the new state point cloud data are both voxelized, the registered original state point cloud data and the new state point cloud data are respectively mapped to each voxel, a unique identifier is assigned to each voxel, and whether there is point cloud data in the voxel is recorded; The new state point cloud data is registered with the original state point cloud data, and both the original state point cloud data and the new state point cloud data after registration are voxelized; The registered original state point cloud data and the new state point cloud data are input into the trained congestion recognition and classification neural network model to obtain the congestion recognition and classification results corresponding to each voxel under voxelization; The congestion recognition and classification neural network model is an improved DGCNN model. Based on the original network architecture of the DGCNN model, the improved DGCNN model removes the spatial transformation network, adds connection layer features, freezes the feature extraction network, and retrains the separator using the trained feature vectors. During the training of the improved DGCNN model, a simulated annealing algorithm is used to optimize the learning rate of the trainer. The trained improved DGCNN model classifies each point cloud data based on the feature relationship between the point cloud data and several neighboring point cloud data, outputs the probability distribution of the point cloud data on each recognition and classification result, and takes the recognition and classification result with the highest probability as the recognition and classification result of the point cloud data. The recognition and classification results include normal and blocked. According to the mapping relationship between each point cloud data and voxel, the blockage identification and classification results corresponding to each voxel under voxelization are obtained; According to the identification and classification results, the blockage of the vibrating screen is cleared.

2. The method for identifying and cleaning vibrating screen blockage based on three-dimensional laser scanning point cloud according to claim 1 is characterized in that: After obtaining the original state point cloud data of the vibrating screen mesh using 3D laser scanning, the following steps are also included: Use the original point cloud data to build the original 3D model of the vibrating screen; Compare the aperture of each mesh of the vibrating screen on the original three-dimensional model with the preset aperture value, and record the number of meshes whose aperture exceeds the preset aperture value; Calculate the ratio of the number of sieve holes whose aperture exceeds the preset aperture value to the total number of sieve holes in the screen, and determine whether the ratio exceeds the preset value of the screen replacement standard. If so, replace the vibrating screen with a new one; otherwise, start the machine and use 3D laser scanning to obtain the new state point cloud data of the vibrating screen.

3. The method for identifying and cleaning vibrating screen blockage based on three-dimensional laser scanning point cloud according to claim 1, characterized in that: Also includes: Based on the identification and classification results, the congestion situation is graded; If the number of blocked sieve holes accounts for less than 10% of the total number of sieve holes, the blockage level is mild. If the number of blocked screen holes accounts for greater than or equal to 10% and less than or equal to 50% of the total number of screen holes, the blockage level is moderate; If the number of blocked holes in the screen accounts for more than 50% of the total number of holes in the screen, the blockage level is high.

4. The method for identifying and cleaning vibrating screen blockage based on three-dimensional laser scanning point cloud according to claim 3 is characterized in that: According to the identification and classification results, the process of cleaning the vibrating screen blockage includes: Use the robot to hold the screen cleaning device, and based on the blockage classification results, press the screen cleaning device into the blocked screen holes to clean the blockage of the screen holes; After cleaning is completed, use the image acquisition device to take a picture of the screen, and confirm the cleaning status of the screen blockage based on the captured image. If the screen holes are not blocked after cleaning, continue the production of the next batch of materials. If there are still blockages in the screen holes that have not been thoroughly cleaned, remove the entire screen from the vibrating screen and install the spare screen on the vibrating screen to continue the production of the next batch of materials. During production, the whole screen removed from the vibrating screen is cleaned by using a screen cleaning device to remove the blockage in the screen holes that have not been thoroughly cleaned.

5. The method for identifying and cleaning vibrating screen blockage based on three-dimensional laser scanning point cloud according to claim 4 is characterized in that: When the integral screen is to be removed from the vibrating screen, the lock between the integral screen and the vibrating screen is released, and the manipulator clamps the integral screen and separates it from the installation position of the vibrating screen; when the spare screen is to be installed on the vibrating screen, the manipulator clamps the spare screen to the installation position of the vibrating screen and locks the relative position of the integral screen and the vibrating screen.

6. A vibrating screen blockage identification and cleaning system based on 3D laser scanning point cloud, characterized in that: The system is used to implement the vibration screen blockage identification and cleaning method based on three-dimensional laser scanning point cloud according to any one of claims 1 to 5, comprising a three-dimensional laser scanning unit, a registration voxelization unit, a blockage identification unit, a central processing unit and a screen cleaning device; The 3D laser scanning unit is used to obtain the original state point cloud data of the vibrating screen mesh by 3D laser scanning before starting the machine; and obtain the new state point cloud data of the vibrating screen mesh by 3D laser scanning after production is completed; A registration voxelization unit is used to register the new state point cloud data with the original state point cloud data, and voxelize both the registered original state point cloud data and the new state point cloud data; A congestion identification unit is used to receive the registered original state point cloud data and the new state point cloud data, and output the congestion identification and classification results corresponding to each voxel in voxelization; a central processing unit, configured to receive and analyze the blockage classification result corresponding to each voxel in the voxelization output by the blockage recognition unit, and output a screen mesh blockage cleaning instruction to the screen cleaning device; The screen cleaning device is used to clean the clogged screen holes.

7. The vibrating screen blockage identification and cleaning system based on three-dimensional laser scanning point cloud according to claim 6 is characterized in that: It also includes a congestion situation grading unit; the congestion situation grading unit grades the congestion situation according to the identification and classification results, and outputs the congestion situation grading result.

8. The vibrating screen blockage identification and cleaning system based on three-dimensional laser scanning point cloud according to claim 7 is characterized in that: It also includes a robot arm for clamping and moving the spare screen or the screen cleaning device according to the instruction of the central processing unit; The screen cleaning device includes a cleaning needle and a base plate. The cleaning needle is arranged on the base plate. The position distribution of the cleaning needle on the base plate is consistent with the position distribution of the sieve holes on the screen. The diameter of the cleaning needle is smaller than the diameter of the sieve holes.

Images