Nondestructive distributed car cleaning system for complex interior structure train carriages with frozen coal

Abstract

This invention belongs to the field of machine learning technology, specifically relating to a non-destructive, decentralized cleaning system for unloading frozen coal from train carriages with complex internal structures. The system includes a data acquisition module, a data analysis module, and a frozen coal cleaning module. This invention acquires point cloud data of the carriage interior through 3D laser scanning and introduces two indicators—surface anomaly and structural inconsistency—to analyze point cloud features. These two indicators are combined to correct the distance between data points, and the DBSCAN algorithm is improved for clustering, achieving precise separation of the frozen coal area from the carriage structure. This effectively avoids damage to the carriage during cleaning operations, ensuring non-destructive, efficient, and automated cleaning.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology. More specifically, this invention relates to a non-destructive, decentralized cleaning system for unloading frozen coal from train carriages with complex internal structures. Background Technology

[0002] In cold regions, coal transported by train is prone to freezing and adhering to the inner walls of the carriages, forming a hard layer of frozen coal, which poses a challenge to unloading the coal. To achieve automated and non-destructive unloading operations, 3D laser scanning equipment is typically used to acquire point cloud data of the carriage interior. By processing and analyzing the point cloud, a 3D model of the frozen coal and the carriage is constructed, thereby guiding the automated cleaning system to subsequently break up and clean the frozen coal.

[0003] Density-Based Spatial Clustering of Applications with Noise (DBSCAN), a traditional machine learning algorithm, can identify clusters of arbitrary shapes by evaluating the local density of point cloud data. Therefore, DBSCAN can be used to distinguish between frozen coal point cloud clusters and wagon body point cloud clusters. However, in actual operating scenarios, frozen coal is often not simply piled up inside the wagon body, but rather tightly frozen and adhered to the side walls, floor, and other complex structures. In this case, the point cloud of frozen coal and the point cloud of the wagon body structure are spatially continuous, and there are no low-density gaps between them that can be identified by the DBSCAN algorithm. Therefore, the point cloud clusters corresponding to frozen coal and the point cloud clusters corresponding to the wagon body structure may be incorrectly merged into a single cluster. This incorrect segmentation result may prevent the system from accurately defining the actual boundaries of the frozen coal, potentially causing damage to the wagon body when the automatic cleaning system removes frozen coal, affecting the wagon's service life, and posing certain safety hazards. Summary of the Invention

[0004] To address the technical problem that the DBSCAN algorithm might erroneously merge the point cloud clusters corresponding to frozen coal and the point cloud clusters corresponding to the carriage structure into a single cluster, this invention provides a non-destructive, decentralized cleaning system for unloading frozen coal from train carriages with complex internal structures. The system includes the following modules:

[0005] The data acquisition module acquires point cloud data inside the carriage, using any data point in the point cloud data as the target data point. The data analysis module acquires neighborhood data points of the target data point, performs principal component analysis on the target data point and its neighborhood data points to obtain the corresponding eigenvalues ​​and eigenvectors, and obtains the surface anisotropy of the target data point based on the corresponding eigenvalues. Based on the cosine similarity between the unit normal vector of the target data point and the unit normal vectors of other data points, and the difference between the surface anisotropy of the data points, the structural inconsistency between the target data point and other data points is obtained. Based on the structural inconsistency between data points and the Euclidean distance, a corrected distance is obtained. The corrected distance is used to replace the Euclidean distance to perform DBSCAN clustering on the point cloud data points to obtain each cluster. The mean of the surface anisotropy of the data points in each cluster is calculated, and the frozen coal area is obtained based on the mean of the surface anisotropy. The frozen coal cleaning module controls the cleaning and recycling equipment to clean the frozen coal.

[0006] This invention addresses the problem in existing technologies that incorrectly group frozen coal and the car body wall into the same cluster when they are tightly adhered and lack low-density gaps, leading to inaccurate identification of frozen coal boundaries. It introduces surface anomaly and structural inconsistency, and creates a corrected distance based on these two indicators to replace the traditional Euclidean distance. This considers not only the spatial location of data points but also the differences in the structural properties of the surfaces on which the points are located. Even if data points corresponding to frozen coal and those corresponding to the car body structure are spatially adjacent, differences in surface roughness and normal vectors will create a distance between them, allowing for successful separation during clustering. This solves the problem of distinguishing adhered frozen coal from the car body structure, achieving precise segmentation of frozen coal areas. This provides a reliable data foundation for subsequent non-destructive cleaning operations and avoids damage to the car body during frozen coal removal, thus preventing impact on the car's service life.

[0007] Preferably, the surface anisotropy satisfies the following relationship: In the formula, For the first Surface anisotropy of each data point , and The first Three feature values ​​for each data point It is a minimum value function.

[0008] Preferably, the surface anisotropy satisfies the following relationship: In the formula, For the first Surface anisotropy of each data point , and The first Three feature values ​​for each data point It is a minimum value function. It is a function for maximizing the value.

[0009] This invention assesses the local surface complexity of data points by measuring surface anisotropy. By using the minimum eigenvalue to reflect the dispersion of the point cloud along the normal vector direction, it can stably quantify the surface complexity of the corresponding region of the point cloud, providing effective data support for distinguishing between the structure of the train carriage and frozen coal.

[0010] Preferably, acquiring the neighborhood data points of the target data point includes: using the nearest neighbor data points to the target data point. The data points are the neighboring data points of the target data point, among which It is 30.

[0011] Preferably, the structural inconsistency satisfies the following relationship: In the formula, For the first The data point and the first The structure is inconsistent between the data points. For the first The unit normal vector of each data point For the first The unit normal vector of each data point For the first Surface anisotropy of each data point For the first Surface anisotropy of each data point This is the cosine similarity function.

[0012] This invention obtains structural inconsistencies by combining the difference in normal vectors between two data points with the difference in surface anisotropy. By combining the differences in data points in two dimensions—geometric direction and surface texture—it can effectively distinguish the boundary between frozen coal and the car body wall, further enhancing the accuracy of frozen coal area identification.

[0013] Preferably, the corrected distance satisfies the following relationship: In the formula, For the first The data point and the first Corrected distance between data points For the first The data point and the first Euclidean distance between data points For the first The data point and the first The structure is inconsistent between the data points.

[0014] Preferably, the DBSCAN clustering of point cloud data points includes: setting the neighborhood radius and the minimum number of neighbors of the core point, and performing DBSCAN clustering on the point cloud data points by replacing Euclidean distance with a corrected distance, wherein the neighborhood radius is 0.2 and the minimum number of neighbors of the core point is 20.

[0015] This invention improves the DBSCAN algorithm by modifying the distance, making it easier for data points with similar corresponding structures to be clustered into the same category, and making it easier for data points with different corresponding structures to be clustered into different categories. By combining the spatial distance and structural distance between data points, the clustering algorithm can better separate data points corresponding to frozen coal from data points corresponding to the structure of the wagon.

[0016] Preferably, the step of obtaining the frozen coal region based on the mean of surface anomaly includes: using the median of the mean of surface anomaly corresponding to each cluster as a threshold, and taking the region corresponding to the cluster with the mean of surface anomaly greater than the threshold as the frozen coal region.

[0017] Preferably, the step of obtaining the frozen coal region based on the mean of surface anomaly includes: using the average of the mean surface anomaly values ​​corresponding to each cluster as a threshold, and taking the region corresponding to the cluster with the mean surface anomaly value greater than the threshold as the frozen coal region.

[0018] This invention obtains frozen coal areas by calculating the mean surface anisotropy of all clusters, enabling the segmentation criteria to be dynamically adjusted according to the overall distribution of the current data, thereby improving the adaptability of the cleaning system to different working conditions and the reliability of the identification results.

[0019] Preferably, the method of controlling the cleaning and recycling equipment to clean frozen coal includes: after identifying the frozen coal area, controlling the cleaning and recycling equipment to move to the frozen coal area, and starting the crusher to crush and clean the frozen coal area.

[0020] The beneficial effects of this invention are as follows: Addressing the technical problem that traditional point cloud clustering algorithms struggle to distinguish frozen coal from the car wall due to their tight adhesion, this invention introduces two indices—surface anomaly and structural inconsistency—to analyze the local geometric and textural features of the point cloud. By combining these two indices, a modified distance is constructed to replace the traditional Euclidean distance, thereby improving the DBSCAN clustering algorithm. This effectively amplifies the structural differences between frozen coal and the car wall, allowing for accurate separation in the feature space even when frozen coal and the car wall are spatially tightly connected. This enables accurate identification of frozen coal areas, ensuring the non-destructive and efficient operation of subsequent automated cleaning. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure, as described in this invention.

[0022] Figure 2 This is a schematic diagram illustrating the system block diagram of the data analysis module in this invention;

[0023] Figure 3 This is a diagram showing the clustering results in this invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0026] This invention provides a non-destructive, decentralized cleaning system for unloading frozen coal from train carriages with complex internal structures. For example... Figure 1 As shown, the non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure includes a data acquisition module 100, a data analysis module 200, and a frozen coal cleaning module 300, which are described in detail below.

[0027] The data acquisition module 100 is used to acquire point cloud data inside the carriage.

[0028] Specifically, a set of point cloud data inside the carriage is obtained by a three-dimensional laser scanning device deployed on the frozen coal cleaning and recycling equipment. Any point cloud data point in the set is used as the target data point for subsequent calculations. The point cloud data acquisition frequency is 30Hz, and the implementers can determine the acquisition frequency of the point cloud data according to the actual situation.

[0029] The data analysis module 200 is used to analyze whether the point cloud data corresponds to frozen coal or the internal structure of the car body.

[0030] Specifically, such as Figure 2 As shown, the data analysis module includes a feature extraction submodule 201, a distance metric submodule 202, and a density clustering submodule 203, which are described in detail below:

[0031] The feature extraction submodule 201 is used to obtain the surface anisotropy of data points and the structural inconsistency between data points, providing data support for the distance measurement submodule.

[0032] It should be noted that different locations inside the car body have different surface characteristics. Some areas are relatively flat, while others are relatively complex. Frozen coal areas usually correspond to relatively complex surfaces. Therefore, frozen coal areas inside the car body can be found by observing the surface conditions of different areas. The distribution of data points corresponding to areas with different surface conditions is also different. Since the distribution of data points can be analyzed by principal component analysis, this invention performs principal component analysis on the neighboring data points of the data points to obtain eigenvalues, and obtains the surface anisotropy of the data points based on the eigenvalues.

[0033] Specifically, the closest to the target data point The data points are the neighboring data points of the target data point. Principal component analysis is performed on the target data point and its neighboring data points using the coordinates of the data points in the three dimensions of X, Y, and Z to obtain the eigenvalues ​​and eigenvectors corresponding to the target data point. The surface anisotropy of the target data point is obtained based on the eigenvalues ​​of the target data point.

[0034] For example, It is 30.

[0035] It should be further explained that the number of eigenvalues ​​obtained by principal component analysis is the same as the number of dimensions of the input data. Since point cloud data points have values ​​in three dimensions, X, Y, and Z, principal component analysis of the neighborhood data points of the target data point will yield three eigenvalues ​​and three eigenvectors, with each eigenvalue corresponding to one eigenvector.

[0036] In one embodiment, the surface anisotropy satisfies the following relationship:

[0037]

[0038] In the formula, For the first Surface anisotropy of each data point , and The first Three feature values ​​for each data point It is a minimum value function.

[0039] Since principal component analysis yields eigenvalues ​​that represent the variance of the data along the eigenvectors corresponding to those eigenvalues, therefore , and It can measure the first The degree of dispersion of the neighborhood data points of a data point in the three eigenvector directions. Representing the The degree of dispersion of neighborhood data points of the data point along the normal vector direction of the plane containing the data point; the larger this value, the more dispersed the neighboring data points of the data point. The more likely the plane containing the first data point is uneven, the better. The greater the surface anomaly of a data point, the smaller the value indicates that the... The more likely the plane containing the nth data point is to be flat, the better. The smaller the surface anomaly of each data point.

[0040] In another embodiment, the surface anisotropy satisfies the following relationship:

[0041]

[0042] In the formula, For the first Surface anisotropy of each data point , and The first Three feature values ​​for each data point It is a minimum value function. It is a function for maximizing the value.

[0043] because and ,and and Compared to being closer, using and The ratio between the minimum and maximum eigenvalues ​​is used as the surface anomaly of the data points. This makes the surface anomaly more sensitive to the complex conditions of the surface corresponding to the data points. Therefore, in this embodiment, the ratio of the minimum eigenvalue to the maximum eigenvalue is used as the surface anomaly.

[0044] Furthermore, the eigenvector corresponding to the smallest eigenvalue among the three eigenvalues ​​of the data point is taken as the normal vector of the data point, and the unit vector of the normal vector of the target data point is taken as the unit normal vector. Based on the cosine similarity between the unit normal vector of the target data point and the unit normal vectors of other data points, as well as the difference between the surface anisotropy of the data points, the structural inconsistency between the target data point and other data points is obtained.

[0045] It should be noted that, since the surface of the frozen coal area is uneven, while the area outside the frozen coal is usually flat, there will be a certain difference between the normal vector of the data point corresponding to the surface of the frozen coal area and the normal vector of the data point on the surface of the area outside the frozen coal. Therefore, this invention obtains the structural inconsistency between data points based on the surface anisotropy of the data points and the normal vector corresponding to the data points.

[0046] Structural inconsistency satisfies the following relation:

[0047]

[0048] In the formula, For the first The data point and the first The structure is inconsistent between the data points. For the first The unit normal vector of each data point For the first The unit normal vector of each data point For the first Surface anisotropy of each data point For the first Surface anisotropy of each data point This is the cosine similarity function.

[0049] in, The smaller the value, the better. The unit normal vector of the nth data point and the nth data point The greater the difference between the unit normal vectors of the nth data points, the greater the difference between them. The data point and the first The more likely a data point is to belong to a different area within the carriage, the more likely the 1st data point is to belong to a different area within the carriage. The data point and the first The greater the structural inconsistency between data points, the greater the structural inconsistency. The larger the value, the more likely it is to be the first. The unit normal vector of the nth data point and the nth data point The smaller the difference between the unit normal vectors of the nth data points, the better. The data point and the first The more likely a data point is to belong to a region with the same structure within the carriage, the more likely the data point is to belong to that region. The data point and the first The smaller the structural inconsistency between data points.

[0050] The larger the number, the higher the number of... The data point and the first The more likely the smoothness of the carriage interior surface corresponding to each data point is to be different, the more likely the 1st data point will be. The data point and the first The greater the structural inconsistency between data points, the greater the structural inconsistency. The smaller, the more... The data point and the first The more likely the smoothness of the inner surface of the carriage is to be the same for each data point, the more likely the first data point is to be the smoothest. The data point and the first The smaller the structural inconsistency between data points.

[0051] Thus, the surface anomaly of the data points and the structural inconsistency between the data points were obtained.

[0052] The distance measurement submodule 202 is used to obtain the corrected distance between data points.

[0053] It should be noted that the surface corresponding to the point cloud data points of frozen coal and the surface corresponding to the data points of the carriage structure have different structures, but their point cloud data point densities may be similar. Therefore, directly clustering the point cloud data points using the DBSCAN algorithm may not be able to separate the region corresponding to frozen coal from the structural region inside the carriage. In order to avoid the situation where the point cloud data points corresponding to frozen coal and the data points corresponding to the carriage structure cannot be separated, which would affect the cleaning of frozen coal inside the carriage, this invention obtains the corrected distance between data points based on the structural inconsistency between data points.

[0054] Specifically, obtain the Euclidean distance between any two data points, and obtain the corrected distance between the two data points based on the Euclidean distance between the two data points and the structural inconsistency between the two data points.

[0055] The corrected distance satisfies the following relationship:

[0056]

[0057] In the formula, For the first The data point and the first Corrected distance between data points For the first The data point and the first Euclidean distance between data points For the first The data point and the first The structure is inconsistent between the data points.

[0058] in, Representing the The data point and the first The distance between data points The smaller, the first The data point and the first The more likely two data points are to belong to the same area within the carriage, the smaller the corrected distance between them should be to avoid clustering them into different categories. The larger, the more The data point and the first The more likely two data points belong to different areas within the carriage, the greater the corrected distance between them should be, so that the two data points are clustered into different categories.

[0059] Representing the The data point and the first The difference between each data point on the corresponding surface structure The larger the number, the more likely it is to be the first. The data point and the first The more likely a data point is to belong to either the frozen coal area or the carriage structure area, the more likely it is to be the first data point. The data point and the first The more data points need to be divided into different categories, the more... The data point and the first The larger the correction distance between data points, the greater the correction distance; The smaller the number, the more likely it is to be the first The data point and the first The more likely the data points belong to the same frozen coal area or the same carriage structure area, the more likely the first data point is to belong to the same frozen coal area or the same carriage structure area. The data point and the first The more data points need to be classified into the same category, the more... The data point and the first The smaller the correction distance between data points.

[0060] The density clustering submodule 203 is used to perform data point clustering.

[0061] Specifically, the neighborhood radius and the minimum number of neighbors for the core point are set. The point cloud data points are clustered using DBSCAN by replacing the Euclidean distance with a modified distance to obtain each cluster. The mean of the surface anisotropy of the data points in each cluster is calculated, and the frozen coal area is obtained based on the mean of the surface anisotropy.

[0062] For example, the neighborhood radius is 0.2 and the minimum number of neighbors for the core point is 20.

[0063] In one embodiment, the average value of the surface anomaly corresponding to each cluster is used as a threshold, and the region corresponding to the cluster whose average surface anomaly is greater than the threshold is regarded as the frozen coal region.

[0064] In another embodiment, the median of the mean surface anomaly of each cluster is used as a threshold, and the regions corresponding to clusters with a mean surface anomaly greater than the threshold are designated as frozen coal regions.

[0065] Figure 3 This diagram illustrates the clustering results in this invention. The clustering method of this invention can better separate frozen coal that is closely connected to the carriage structure, providing better input data for the frozen coal cleaning module.

[0066] The frozen coal cleaning module 300 is used to control the cleaning and recycling equipment to clean frozen coal.

[0067] Specifically, after identifying the frozen coal area, the cleaning and recycling equipment is moved to the frozen coal area, and the crusher is started to crush and clean the frozen coal area.

Claims

1. A non-destructive, decentralized cleaning system for unloading frozen coal from train carriages with complex internal structures, characterized in that: include: The data acquisition module acquires point cloud data inside the carriage, using any data point in the point cloud data as the target data point; The data analysis module acquires neighborhood data points of the target data point, performs principal component analysis on the target data point and its neighborhood data points to obtain the eigenvalues ​​and eigenvectors corresponding to the target data point, and obtains the surface anisotropy of the target data point based on the eigenvalues, satisfying the following relationship: ； In the formula, For the first Surface anisotropy of each data point , and The first Three feature values ​​for each data point It is a minimum value function. It is a function for maximizing the value; Representing the The degree of dispersion of neighborhood data points of the data point along the normal vector direction of the plane containing the data point; the larger this value, the more dispersed the neighboring data points of the data point. The more likely the plane containing the first data point is uneven, the better. The greater the surface anomaly of a data point, the smaller the value indicates that the... The more likely the plane containing the nth data point is to be flat, the better. The smaller the surface anomaly of each data point; Based on the cosine similarity between the unit normal vector of the target data point and the unit normal vectors of other data points, and the difference between the surface heterogeneity of the data points, the structural inconsistency between the target data point and other data points is obtained, satisfying the following relationship: ； In the formula, For the first The data point and the first The structure is inconsistent between the data points. For the first The unit normal vector of each data point For the first The unit normal vector of each data point For the first Surface anisotropy of each data point The cosine similarity function; The larger the number, the higher the number of... The data point and the first The more likely the smoothness of the carriage interior surface corresponding to each data point is to be different, the more likely the 1st data point will be. The data point and the first The greater the structural inconsistency between data points, the greater the structural inconsistency. The smaller, the more... The data point and the first The more likely the smoothness of the inner surface of the carriage is to be the same for each data point, the more likely the first data point is to be the smoothest. The data point and the first The smaller the structural inconsistency between data points; The corrected distance is obtained based on the structural inconsistencies between data points and the Euclidean distance, specifically as follows: ； In the formula, For the first The data point and the first Corrected distance between data points For the first The data point and the first Euclidean distance between data points; By replacing Euclidean distance with corrected distance, DBSCAN clustering is performed on the point cloud data points to obtain each cluster; the mean of surface anomaly of data points in each cluster is calculated, and the frozen coal area is obtained based on the mean of surface anomaly. The frozen coal cleaning module controls the cleaning and recycling equipment to clean frozen coal.

2. The non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure as described in claim 1, characterized in that: The surface anisotropy satisfies the following relationship: ； In the formula, For the first Surface anisotropy of each data point , and The first Three feature values ​​for each data point It is a minimum value function.

3. The non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure as described in claim 1, characterized in that: The acquisition of neighborhood data points of the target data point includes: the nearest neighbor data points to the target data point. The data points are the neighboring data points of the target data point, among which It is 30.

4. The non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure as described in claim 1, characterized in that: The DBSCAN clustering of point cloud data points includes: setting the neighborhood radius and the minimum number of neighbors for the core point, and performing DBSCAN clustering on the point cloud data points by replacing Euclidean distance with a corrected distance, wherein the neighborhood radius is 0.2 and the minimum number of neighbors for the core point is 20.

5. The non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure as described in claim 1, characterized in that: The step of obtaining frozen coal regions based on the mean of surface anomaly includes: using the median of the mean surface anomaly of each cluster as a threshold, and taking the regions corresponding to clusters with a mean surface anomaly greater than the threshold as frozen coal regions.

6. The non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure as described in claim 1, characterized in that: The step of obtaining frozen coal regions based on the mean of surface anomaly includes: using the average of the mean surface anomaly values ​​corresponding to each cluster as a threshold, and taking the regions corresponding to clusters with mean surface anomaly values ​​greater than the threshold as frozen coal regions.

7. The non-destructive decentralized cleaning system for unloading frozen coal from a train carriage with a complex internal structure as described in claim 1, characterized in that: The process of controlling the cleaning and recycling equipment to clean frozen coal includes: after identifying the frozen coal area, controlling the cleaning and recycling equipment to move to the frozen coal area, starting the crusher to crush and clean the frozen coal area.

Images