Discharging method

Abstract

The invention belongs to the technical field of automatic control, and particularly relates to a discharging method which is characterized in that a planar model is verified twice in a material pile model construction process, so that the reliability of material pile surface fitting is remarkably improved, a model with high fitting degree with the real surface of a material pile can be constructed, and a reliable guarantee is provided for automatic material taking and discharging operation of a travelling crane.

Description

Technical Field

[0001] This invention belongs to the field of automatic control technology, and specifically relates to a material unloading method. Background Technology

[0002] The unloading environment for metallurgical raw materials is characterized by high temperature, high dust, and high noise, making it unsuitable for long-term manual operation. Existing remote control equipment frequently suffers from errors in visual judgment of grab bucket height and position, as well as poor adaptability to material transport operations.

[0003] Chinese patent CN112669376A discloses a method and system for calculating the height of a material pile stack. This method acquires two-dimensional point cloud data and fits it to a three-dimensional spatial point cloud. The average value of the z-axis of the three-dimensional point cloud set is taken as the height of the grabbing platform. However, the material contour analysis is time-consuming and the accuracy of the material height information is low, which can easily lead to errors in the grab bucket height judgment.

[0004] Chinese patent CN114348678A discloses a control method, device, and electronic equipment for an automated material handling vehicle. This method controls the vehicle's automatic operation solely based on the coordinates of the picking and unloading points. However, in actual operation, the height of the material pile traversed by the trolley and the auxiliary trolley varies. To avoid interference between the hopper and the pile, the hopper must be raised to a higher position, which is time-consuming and energy-intensive, hindering efficient material transport. Summary of the Invention

[0005] The purpose of this invention is to provide a material unloading method that can reliably achieve fully automated operation.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for unloading materials,

[0008] Step A. Construct a stockpile model;

[0009] Step a1. Scan the material pile to obtain a 3D laser point cloud dataset on the surface of the material pile. Establish a coordinate system with the length of the material yard as the x-axis, the width as the y-axis, and the height as the z-axis. Convert the spatial coordinates of each point in the point cloud dataset into (x, y, z).

[0010] Step a2. Select three non-collinear points from the point cloud dataset as the interior point sample set;

[0011] Step a3. Perform plane fitting on the interior point sample set obtained in step a2 to obtain a planar model;

[0012] The fitting expression is ax + by + cz + d = 0. Substitute the coordinates of the interior point sample set into the solution to obtain a, b, c, and d. Output the normal vector n = (a, b, c,) of the plane model and the feature point P(x0, y0, z0). The feature point is a point located on the plane model, which can be the center point, corner point, or point in the interior point sample set of the plane model.

[0013] Step a4. Calculate the distance from the points in the point cloud dataset to the planar model obtained in step a3, and determine that the points with a distance less than a preset threshold are interior points, and the rest are exterior points;

[0014] Step a5. Check the interior point ratio threshold;

[0015] Based on the results of step a4, calculate the ratio of the number of interior points c1 to the total number of points c0 participating in the fitting. If c1 / c0 < k1, where k1 is an empirical value and k1∈(0,1), the plane model is determined to be invalid and it is necessary to return to step a2 to refit the plane model; otherwise, proceed to step a6.

[0016] Step a6. Verify the consistency of the spatial distribution of interior points;

[0017] Project the interior points obtained in step a4 onto the planar model, calculate the minimum bounding rectangle area S1 and the bounding box area S2 of all interior points projected onto the planar model. If S1 / S2≥k2 and the aspect ratio of the minimum bounding rectangle is less than or equal to k3, where k2∈[0.5,1) and k3∈(0,0.2], the planar model is deemed valid. Extract the normal vector and feature point coordinates of the planar model to form the material pile model and proceed to step B. Otherwise, the planar model is deemed invalid and it is necessary to return to step a2 to refit the planar model.

[0018] Step B. Automatic unloading by the overhead crane;

[0019] The work area is divided into square grid-like areas along the width and length of the material yard.

[0020] b1. Mark the cell field where the throwing point of the throwing machine is located as the material picking area. Combined with the material pile model obtained in step A, select several cell fields as the unloading area.

[0021] b2. Based on the stockpile model obtained in step A, determine the material picking coordinates and unloading coordinates of the crane, the material picking height and unloading height of the grab bucket, and drive the crane and grab bucket to move to achieve automatic unloading.

[0022] Compared with the prior art, the present invention has the following technical effects: the planar model is verified twice, which significantly improves the reliability of the material pile surface fitting, and thus enables the construction of a model with high conformity to the real surface of the material pile, providing a reliable guarantee for the automatic material picking and unloading operation of the crane. Attached Figure Description

[0023] The following is a brief explanation of the contents depicted in the accompanying drawings and the markings within them:

[0024] Figure 1 This is a schematic diagram of the present invention. Detailed Implementation

[0025] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and through the description of the examples.

[0026] A method for unloading materials includes the following steps:

[0027] Step A. Construct a stockpile model.

[0028] Step a1. Use a lidar to scan the material pile and obtain a 3D lidar point cloud dataset of the pile surface. In this embodiment, the material yard is rectangular in shape. The lidar is mounted on a crane, and the crane is driven to move along the length of the material yard to scan the entire material yard. For the analysis and processing of the point cloud data, a coordinate system is established with the length of the material yard as the x-axis, the width as the y-axis, and the height as the z-axis. The spatial coordinates of each point in the point cloud dataset are converted into (x, y, z).

[0029] Step a2. Select three non-collinear points from the point cloud dataset as the interior point sample set.

[0030] In the preferred scheme, to avoid selecting noisy points into the inner point sample set and improve the efficiency of material pile model construction, the point cloud dataset obtained in step a1 can be analyzed first to identify continuously distributed point clouds. Local continuously distributed point cloud sets can be established for these continuously distributed point clouds respectively. Then, three non-collinear points can be selected from each local continuously distributed point cloud set as the inner point sample set of that region. That is, a planar model can be constructed for each continuously distributed point cloud region. In this way, the planar model constructed based on the inner point sample set fits the real surface of the material pile better.

[0031] Step a3. Perform plane fitting on the interior point sample set obtained in step a2 to obtain a planar model. In this embodiment, the least squares method is used to fit the planar model, and the fitting expression is ax + by + cz + d = 0. Substitute the coordinates of the points in the interior point sample set into the expression to solve for a, b, c, and d, and output the normal vector n = (a, b, c) and feature points P(x0, y0, z0) of the planar model. The feature points are points located on the planar model. The feature points can be obtained from the fitting expression, or they can be a point in the interior point sample set, or they can be calculated from the interior point sample set.

[0032] Step a4. Calculate the distance from each point in the point cloud dataset to the planar model obtained in step a3, also known as the Euclidean distance. Points with an Euclidean distance less than a preset threshold are considered interior points, and the rest are considered exterior points. Interior points are considered valid points, and exterior points are considered noise points. These interior points constitute the fitting range of the planar model; that is, the boundary of the fitting region of the planar model is determined by the interior points. Points can be extracted from the point cloud dataset with a radius of R centered on the feature points of the planar model to test interior points, or all points in the point cloud dataset can be used to test interior points.

[0033] Step a5. Check the interior point ratio threshold.

[0034] Based on the results of step a4, calculate the ratio of the number of interior points c1 to the total number of points c0 participating in the fitting. If c1 / c0 < k1, where k1 is an empirical value and k1∈(0,1), the plane model is determined to be invalid and it is necessary to return to step a2 to refit the plane model; otherwise, proceed to step a6.

[0035] This step verifies the fit between the planar model and the actual surface of the stockpile in the height direction. A higher c1 / c0 ratio indicates a higher proportion of points whose distance from the planar model meets the threshold, signifying a better fit between the planar model and the actual stockpile surface. A low c1 / c0 ratio indicates a poor fit between the planar model and the actual stockpile surface, necessitating refitting.

[0036] Step a6. Verify the consistency of the spatial distribution of interior points.

[0037] Project the interior points obtained in step a4 onto the planar model, calculate the minimum bounding rectangle area S1 and the bounding box area S2 of all interior points projected onto the planar model. If S1 / S2≥k2 and the aspect ratio of the minimum bounding rectangle is less than or equal to k3, where k2∈[0.5,1) and k3∈(0,20], the planar model is deemed valid. Extract the normal vector and feature point coordinates of the planar model to form the material pile model and proceed to step B. Otherwise, the planar model is deemed invalid and it is necessary to return to step a2 to refit the planar model.

[0038] If S1 / S2 meets the threshold, it indicates that the surface of the material pile composed of the interior points is relatively flat, and there are no obvious protrusions or depressions in the area where the planar model is located. The aspect ratio of the smallest bounding rectangle meets the threshold, indicating that the material pile in the area where the planar model is located is not a long and thin strip, thus avoiding the identification of long and thin strip edges such as material pile baffles as the material pile plane.

[0039] This step is used to verify whether the interior points are continuously distributed in space, that is, whether the interior points of the planar model are adjacent points, and whether the interior points can represent a relatively continuous plane. If the interior points are adjacently distributed on the xy plane, it means that the planar model fits these interior points well. If the interior points are scattered and far apart on the xy plane, it means that the planar model fits points that are far apart into a plane, and the planar model is far from the actual surface of the material pile. Therefore, to simplify the calculation, the preferred solution for this step is to project the planar model onto the xy plane to form the projection analysis plane of the planar model, project the interior points obtained in step a4 onto the projection analysis plane of the planar model, and calculate the minimum bounding rectangle area S1 and the bounding box area S2 of all the interior points projected onto the projection analysis plane.

[0040] After verifying the planar model twice, errors in the construction of the stockpile model can be effectively avoided, thus providing a reliable guarantee for the automated operation of the crane.

[0041] Step B. Automatic unloading by the overhead crane.

[0042] This embodiment divides the work area into square grid-like regions along the width and length of the material yard. The edges of each cell are parallel to the length or width of the material yard, and the size of each cell can be adjusted according to the actual layout of the material yard. Dividing the material yard into several cell regions facilitates accurate positioning and updating of the material's location, type, and pile outline height, and provides clear coordinate instructions for subsequent crane unloading operations. Furthermore, based on this, in step A, the point cloud data within each cell region can be analyzed to construct a pile model. This allows for simultaneous planar simulation of each cell region, improving data analysis efficiency.

[0043] b1. Mark the cell field containing the throwing point of the throwing machine as the material picking area 11. Based on the stockpile model obtained in step A, select several cell fields as unloading areas 12. In actual production, raw materials are transported to the material yard by conveyor belt and then thrown to the designated location by the throwing machine. Therefore, the material picking area 11 is usually relatively fixed. However, different batches of raw materials need to be transferred to different areas of the material yard, and raw materials of the same batch or type also need to have their unloading areas determined according to the stockpile model. Therefore, the unloading area 12 needs to be adjusted in real time. The material picking area 11 and the unloading area 12 can be selected manually or automatically by the host computer according to preset rules.

[0044] b2. Based on the stockpile model obtained in step A, determine the crane's pick-up and unloading coordinates, and the grab bucket's pick-up and unloading heights. Then, drive the crane and grab bucket to move and achieve automatic unloading. (See attached diagram) Figure 1As shown, by selecting the material picking area 11 and the material unloading area 12, the material picking coordinates and the material unloading coordinates of the crane can be obtained. Combined with the material pile model obtained in step A, the material pile height of the material picking area 11 and the material unloading area 12 can be obtained, and then the material picking and unloading height of the crane hopper can be calculated.

[0045] Based on this, it can be understood that the stockpile model is used to provide accurate height information of various locations in the stockyard for crane operations. Therefore, in the preferred scheme, in order to facilitate the guidance of crane operations, in step a3, the normal vector of the plane model can be preset to be parallel to the z-axis, so that the fitted plane model is an xy plane parallel to the ground. In this way, the height of various locations in the stockpile can be quickly obtained based on the stockpile model.

[0046] In the preferred embodiment, to make full use of the material storage space, in step b2, when the unloading area 12 includes multiple cell fields, the current cell field is filled to the preset height before the next cell field is filled. This can make full use of the material storage area, avoid frequent switching of the crane operation coordinates, and make the crane operation instructions clearer and more specific.

[0047] As attached Figure 1 As shown, the unloading area 12 has multiple cell fields in both the length and width directions of the material yard, i.e., multiple cell fields in the x-axis and y-axis directions respectively. To improve efficiency, the cell fields are filled sequentially in the width direction of the material yard from far to near, and in the length direction of the material yard from near to far. That is, the cell fields arranged horizontally at the top, furthest from the material picking area 11, are filled first, and these horizontal cell fields are filled sequentially from left to right. After these three horizontal cell fields are filled to the specified height, the cell fields located below are then filled. This avoids the cell fields arranged adjacent to the material picking area 11 from being filled to the highest point of the material, which would affect the passage of the hopper and thus avoid potential safety hazards.

[0048] In this embodiment, to further improve the reliability of the overhead crane unloading operation, a grab bucket unloading algorithm model is applied to control the grab bucket operation. Specifically, in step b2, the planar model data of the corresponding area in the stockpile model obtained in step A is input into the grab bucket unloading algorithm model. The grab bucket unloading algorithm model outputs the working height and angle of the overhead crane grab bucket, and the overhead crane performs the unloading operation according to the working height output by the grab bucket unloading algorithm model.

[0049] The preferred approach employs a lightweight convolutional neural network to construct the grab bucket unloading algorithm model. The model training data includes the stockpile model data obtained in step A, material type labels, grab bucket model, the grab bucket's descent height and working angle when manual operation is successful, and the grab bucket's descent height and working angle when manual operation fails. Inputting failed operation cases into the model helps identify the "boundary conditions" and abnormal patterns that lead to adverse results, making it more robust to complex and changing production environments. Simultaneously, noise ±h is added to the grab bucket's descent height, where h is a real number ∈ [0, 10cm], and features such as material moisture content and grab bucket wear degree are added to improve the model's generalization ability, thereby ensuring stable and reliable grab bucket operation.

[0050] In a further optimized solution, to avoid the possibility of deformation of the material surface due to contact with the grab bucket, a real-time feedback adjustment component is added. Specifically, a laser rangefinder sensor is installed on the outer surface of the grab bucket side to measure the real-time distance between the grab bucket and the material surface. When the grab bucket descends to the specified position output by the grab bucket unloading algorithm model, if the measured distance between the material on the side of the grab bucket and the laser rangefinder sensor is less than a preset value, it is determined that material collapse has occurred, and the grab bucket is driven to continue to move downwards by 1-3cm before performing the material grabbing operation. That is, the descent height of the grab bucket is finely adjusted to ensure the amount of material picked up in a single operation.

Claims

1. A method for unloading materials, comprising the following steps: Step A. Construct a stockpile model; Step a1. Scan the material pile to obtain a 3D laser point cloud dataset on the surface of the material pile. Establish a coordinate system with the length of the material yard as the x-axis, the width as the y-axis, and the height as the z-axis. Convert the spatial coordinates of each point in the point cloud dataset into (x, y, z). Step a2. Select three non-collinear points from the point cloud dataset as the interior point sample set; Step a3. Perform plane fitting on the interior point sample set obtained in step a2 to obtain a planar model; Step a4. Calculate the distance from the points in the point cloud dataset to the planar model obtained in step a3, and determine that the points with a distance less than a preset threshold are interior points, and the rest are exterior points; Step a5. Check the interior point ratio threshold; Based on the results of step a4, calculate the ratio of the number of interior points c1 to the total number of points c0 participating in the fitting. If c1 / c0 < k1, where k1 is an empirical value and k1∈(0,1), the plane model is determined to be invalid and it is necessary to return to step a2 to refit the plane model; otherwise, proceed to step a6. Step a6. Verify the consistency of the spatial distribution of interior points; Project the interior points obtained in step a4 onto the planar model, calculate the minimum bounding rectangle area S1 and the bounding box area S2 of all interior points projected onto the planar model. If S1 / S2≥k2 and the aspect ratio of the minimum bounding rectangle is less than or equal to k3, where k2∈[0.5,1) and k3∈(0,0.2], the planar model is deemed valid. Extract the normal vector and feature point coordinates of the planar model to form the material pile model and proceed to step B. Otherwise, the planar model is deemed invalid and it is necessary to return to step a2 to refit the planar model. Step B. Automatic unloading by the overhead crane; The work area is divided into square grid-like areas along the width and length of the material yard. b1. Mark the cell field where the throwing point of the throwing machine is located as the material taking area (11). Combine the material pile model obtained in step A, select several cell fields as the unloading area (12). b2. Based on the stockpile model obtained in step A, determine the material picking coordinates and unloading coordinates of the crane, the material picking height and unloading height of the grab bucket, and drive the crane and grab bucket to move to achieve automatic unloading.

2. The unloading method according to claim 1, characterized in that: In step a2, the point cloud dataset obtained in step a1 is analyzed, and several locally continuous point cloud sets are extracted. From each locally continuous point cloud set, three non-collinear points are selected as the interior point sample set of that region.

3. The unloading method according to claim 1 or 2, characterized in that: The work area is divided into several square grid areas along the width and length of the material yard. In step A, the point cloud data in each cell is analyzed to construct the material pile model.

4. The unloading method according to claim 1, characterized in that: In step a3, the normal vector of the preset planar model is parallel to the z-axis, so that the fitted planar model is an xy plane parallel to the ground.

5. The unloading method according to claim 1, characterized in that: In step a6, the planar model is projected onto the xy plane to form the projection analysis plane of the planar model. The interior points obtained in step a4 are projected onto the projection analysis plane of the planar model. The minimum bounding rectangle area S1 and the bounding box area S2 of all interior points projected onto the projection analysis plane are calculated.

6. The unloading method according to claim 1, characterized in that: In step b2, when the unloading area (12) includes multiple cell fields, each cell field is filled in the order from far to near in the width direction of the material yard and from near to far in the length direction of the material yard; after the current cell field is filled to the preset height, the next cell field is filled.

7. The unloading method according to claim 1, characterized in that: In step b2, the planar model data of the corresponding area in the stockpile model obtained in step A is input into the grab bucket unloading algorithm model. The grab bucket unloading algorithm model outputs the working height and angle of the crane grab bucket. The crane completes the unloading operation according to the output results of the grab bucket unloading algorithm model. A lightweight convolutional neural network was used to construct a grab bucket unloading algorithm model. The model training data included the stockpile model data obtained in step A, material type labels, grab bucket model, grab bucket descent height and working angle when manual operation was successful, and grab bucket descent height and working angle when manual operation failed. Meanwhile, noise ±h is added to the descent height of the grab bucket, where h is a real number and h∈[0,10cm], and the humidity of the characteristic material and the wear degree of the grab bucket are also added.

8. The unloading method according to claim 7, characterized in that: A laser rangefinder sensor is installed on the outer surface of the side of the grab bucket to measure the real-time distance between the grab bucket and the material surface. When the grab bucket descends to the specified position output by the grab bucket unloading algorithm model, if the distance between the material on the side of the grab bucket and the laser rangefinder sensor is less than the preset value, it is determined that material collapse has occurred, and the grab bucket is driven to continue to move downward by 1-3cm before the material grabbing operation is carried out.

Images