An integrated automatic powder cleaning system and method for additive manufacturing molding cylinders
By improving the Euclidean clustering algorithm and adaptive powder cleaning path planning, the problems of powder identification misjudgment and uneven cleaning in the powder cleaning of the additive manufacturing molding cylinder were solved, and efficient and safe automated powder cleaning effect was achieved.
Patent Information
- Application Number
- CN202611285200.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-24
- Publication Date
- 2026-11-17
AI Technical Summary
Existing additive manufacturing molding cylinder powder cleaning technology relies on manual operation or automated equipment with fixed parameters, which suffers from problems such as powder identification misjudgment, uneven cleaning, equipment blockage and low efficiency. It cannot effectively distinguish the subtle differences between powder and cylinder structure, resulting in a working environment that is harmful to human health and is inefficient.
An improved Euclidean clustering algorithm is adopted to adjust the clustering radius based on the difference in the angle between the local normal vectors of the point cloud, generate a powder adhesion distribution map, dynamically plan the powder cleaning path, and combine an adaptive powder cleaning execution mechanism and a multi-mode cleaning strategy to achieve accurate identification and efficient cleaning of powder clusters.
It improves the accuracy of powder cluster boundary recognition to the millimeter level, optimizes the powder cleaning path planning, improves powder cleaning efficiency, reduces powder residue and equipment blockage, and improves the safety of the working environment.
Smart Images

Figure CN123104069A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of automated control technology in additive manufacturing, and in particular to an integrated automatic powder cleaning system and method for additive manufacturing molding cylinders. Background Technology
[0002] Current additive manufacturing molding cylinder powder cleaning mainly relies on manual operation or fixed-parameter automated equipment. Manual powder cleaning involves manually removing residual powder using tools such as scrapers and vacuum cleaners. The high dust concentration in the working environment poses a risk of lung disease to operators with prolonged exposure, and the cleaning process can easily scratch the inner wall of the molding cylinder or damage undemolded parts. Fixed-parameter automated equipment uses laser scanning to acquire point cloud data of the molding cylinder, then directly segments the point cloud using a preset threshold and generates a straight reciprocating path. The point cloud segmentation uses a fixed-radius Euclidean clustering algorithm, which cannot identify subtle geometric differences between the powder and the cylinder structure. It often misclassifies powder clusters and cylinder wall protrusions as the same object, or incorrectly merges dispersed powder into a single cluster. The path generation does not consider differences in powder distribution density. In areas with excessive powder accumulation, the powder cleaning actuator may experience pipe blockage due to excessive suction volume in a single operation, leading to repeated empty runs in areas with small amounts of powder, reducing operational efficiency.
[0003] Traditional point cloud processing lacks a quantitative model of powder spatial distribution, relying solely on 3D coordinate mapping to generate paths, thus failing to distinguish between powder adhesion thickness and cylinder structural deformation. The powder cleaning mechanism traverses the entire forming cylinder at a uniform speed, wasting time in sparse powder areas and leaving residual powder in powder-agglomerated areas due to insufficient suction. Right-angled areas on the inner wall of the forming cylinder are often ignored by the algorithm due to missing point cloud data, forming cleaning blind spots. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose an integrated automatic powder cleaning system and method for additive manufacturing molding cylinders.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: an integrated automatic powder cleaning method for additive manufacturing molding cylinders, comprising: Obtain the original point cloud data set of the additive manufacturing molding cylinder to be cleaned, the original point cloud data set including the point cloud of the inner bottom surface of the molding cylinder, the point cloud of the inner side wall, and the point cloud of the residual powder bed surface; The original point cloud data set is subjected to point cloud segmentation processing based on an improved Euclidean clustering algorithm to obtain a segmented powder cluster point cloud set and a segmented cylinder structure point cloud set. The improved Euclidean clustering algorithm adjusts the clustering radius based on the difference in the angle between the local normal vectors of the point cloud. The powder adhesion distribution map of the molding cylinder is generated based on the segmented powder cluster point cloud set and the segmented cylinder structure point cloud set. The powder adhesion distribution map is used to mark the position and volume of powder clusters in each region inside the molding cylinder. Based on the powder adhesion distribution map, a powder cleaning execution path sequence is generated, and the powder cleaning execution path sequence is transmitted to the integrated powder cleaning execution mechanism.
[0006] As a further aspect of the present invention, the workflow of the improved Euclidean clustering algorithm includes: Traverse each data point in the original point cloud data set and calculate the neighborhood point set of each data point. The neighborhood point set consists of all neighboring data points whose spatial straight-line distance from the currently calculated data point is less than the initial fixed radius. For each data point, the local plane of the current calculated data point is fitted using all neighboring data points in the neighborhood point set, and the normal vector of the local plane is calculated as the local normal vector of the current calculated data point. Calculate the cosine of the angle between the local normal vector of the current data point and the local normal vector of the corresponding neighboring data point, and determine the normal matching score between the current data point and the corresponding neighboring data point based on the cosine of the angle. When the normal matching score is greater than the preset matching threshold, the corresponding neighboring data point is marked as the normal consistent neighborhood point of the currently calculated data point; The number of normal-consistent neighborhood points for each data point is counted, and the adaptive clustering radius corresponding to the current data point is determined based on the number of normal-consistent neighborhood points. The adaptive clustering radius is inversely proportional to the number of normal-consistent neighborhood points, and the value range of the adaptive clustering radius is limited to between the minimum radius value and the maximum radius value. Using each data point as the starting point of the search, and using the adaptive clustering radius as the search radius, a region growing clustering operation is performed. Data points with continuous connectivity of neighboring points with consistent normal directions are grouped into the same point cloud cluster, and the segmented powder cluster point cloud set and the segmented cylinder structure point cloud set are output.
[0007] As a further aspect of the present invention, point cloud segmentation processing based on an improved Euclidean clustering algorithm is performed on the original point cloud data set to obtain a segmented powder cluster point cloud set and a segmented cylinder structure point cloud set, further comprising: Before performing point cloud segmentation processing on the original point cloud data set, voxel downsampling filtering processing is performed on the original point cloud data set to generate a downsampled point cloud data set. The voxel downsampling filtering processing includes dividing the spatial range of the original point cloud data set into three-dimensional voxel grids of equal size. In each three-dimensional voxel grid, only a single data point corresponding to the centroid coordinates of all data points in the current three-dimensional voxel grid is retained. Statistical outlier removal processing is performed on the downsampled point cloud dataset. For each data point in the downsampled point cloud dataset, the average distance between the current calculated data point and a preset number of neighboring data points is calculated. When the average distance deviates from the global average distance by more than a preset standard deviation multiple, the current calculated data point is marked as an outlier and removed from the downsampled point cloud dataset to generate a clean point cloud dataset. The clean point cloud dataset is input into the improved Euclidean clustering algorithm to perform point cloud segmentation processing.
[0008] As a further aspect of the present invention, generating a powder adhesion distribution map of the molding cylinder based on the segmented powder cluster point cloud set and the segmented cylinder structure point cloud set includes: The inner bottom surface and inner sidewall of the forming cylinder are discretized into a two-dimensional grid pattern according to a preset grid size. Each grid cell in the two-dimensional grid pattern corresponds to a fixed spatial region inside the forming cylinder. Traverse all data points in the segmented cylinder block structure point cloud set, project each data point onto the corresponding grid cell of the two-dimensional grid map, and mark the currently projected grid cell as the cylinder block structure occupying cell; Traverse each powder cluster point cloud in the segmented powder cluster point cloud set, calculate the three-dimensional bounding box of all data points contained in the powder cluster point cloud, and obtain the spatial occupancy range of the powder cluster point cloud inside the molding cylinder. The set of grid cells covered by the current powder cluster point cloud is determined based on the spatial occupancy range. Each grid cell in the set of grid cells is marked as a powder cluster occupying cell. At the same time, the total number of data points in the current powder cluster point cloud is calculated as the powder cluster volume index corresponding to the current set of grid cells. The grid cells in the two-dimensional grid map that are simultaneously marked as cylinder structure occupying units and powder cluster occupying units are selected to form powder attachment area markers. The powder attachment area markers and the powder cluster volume index corresponding to each powder attachment area marker are stored together as the powder attachment distribution map.
[0009] As a further aspect of the present invention, a powder removal execution path sequence is generated based on the powder adhesion distribution map, including: Extract all grid cells with powder adhesion region markers from the powder adhesion distribution map to form a powder adhesion grid set; The grid cells in the powder adhesion grid set are sorted from largest to smallest according to their corresponding powder cluster volume index to generate a priority cleaning grid sequence; The grid cell with the largest powder agglomeration volume index is selected from the priority cleaning grid sequence as the current cleaning target cell, and the center point of the current cleaning target cell is used as the moving target point of the powder cleaning actuator. A path planning algorithm is used to calculate the shortest movement path between the current position of the cleaning actuator and the moving target point, and the shortest movement path is recorded as a path segment in the cleaning execution path sequence; After the cleaning actuator reaches the moving target point, the dwell time at the moving target point is determined according to the powder cluster volume index corresponding to the current cleaning target unit. The dwell time is proportional to the powder cluster volume index. The moving target point is marked as cleared and removed from the priority cleaning grid sequence. The operations of retrieving, calculating, recording, determining the dwell time and marking the cleared state are repeated until the priority cleaning grid sequence is empty, thus obtaining a complete powder cleaning execution path sequence.
[0010] As a further aspect of the present invention, the path planning algorithm includes: A two-dimensional grid map is constructed inside the forming cylinder. Each grid cell in the two-dimensional grid map corresponds one-to-one with each grid cell in the two-dimensional grid map. Grid cells with cylinder structure occupying cells are marked as obstacle grids, and grid cells without cylinder structure occupying cells are marked as free grids. The current grid cell of the powder cleaning actuator is obtained as the starting grid, and the grid cell where the moving target point is located is obtained as the target grid; Starting from the initial grid, a heuristic search method is used to explore adjacent grids. For each explored grid cell, the estimated cost of reaching the target grid from the initial grid through the explored grid cells is calculated. The estimated cost is obtained by adding the actual movement cost from the initial grid to the explored grid cell to the estimated Manhattan distance cost from the explored grid cell to the target grid cell. Prioritize expanding the raster cell with the lowest estimated cost, and record the predecessor raster cell for each expanded raster cell, until the target raster is expanded; Starting from the target grid, the system backtracks to the starting grid sequentially based on the predecessor grid cells of each grid cell. The center points of all the grid cells traversed during the backtracking are connected in the backtracking order to obtain the shortest moving path.
[0011] As a further aspect of the present invention, the number of normally consistent neighborhood points for each data point is counted, and the adaptive clustering radius corresponding to the currently calculated data point is determined based on the number of normally consistent neighborhood points, including: Obtain the number of normal-consistent neighbor points of the data point; Calculate the absolute value of the difference between the number of normal-consistent neighbor points and the preset threshold for the total number of neighbor points; The absolute value of the difference is input into a preset radius mapping function to calculate the preliminary cluster radius. The radius mapping function is configured such that the output value is inversely proportional to the input value. The initial clustering radius is compared with a preset minimum radius threshold. If the initial clustering radius is less than the minimum radius threshold, the minimum radius threshold is determined as the adaptive clustering radius of the data points. If the initial clustering radius is not less than the minimum radius threshold, then the initial clustering radius is compared with the preset maximum radius threshold. If the initial clustering radius is greater than the maximum radius threshold, then the maximum radius threshold is determined as the adaptive clustering radius of the data points. If the initial clustering radius is not greater than the maximum radius threshold, then the initial clustering radius is directly determined as the adaptive clustering radius of the data points.
[0012] As a further aspect of the present invention, the method further includes a powder removal strategy classification process performed after generating the powder adhesion distribution map: Based on the powder cluster volume index corresponding to each powder adhesion region marker in the powder adhesion distribution map, the powder adhesion region markers are divided into heavily adhered regions, moderately adhered regions, and lightly adhered regions. The heavily adhered regions correspond to regions where the powder cluster volume index is greater than the first volume threshold, the moderately adhered regions correspond to regions where the powder cluster volume index is between the second volume threshold and the first volume threshold, and the lightly adhered regions correspond to regions where the powder cluster volume index is less than the second volume threshold. A high-pressure pulse purging and dust removal mode is configured for the heavily adhered area. The high-pressure pulse purging and dust removal mode includes the on-time of the high-pressure air source, the pulse frequency, and the pulse duty cycle, wherein the pulse frequency and the pulse duty cycle are proportional to the volume index of the powder clusters in the heavily adhered area. A continuous airflow purging and dust removal mode is configured for the moderately attached region. The continuous airflow purging and dust removal mode includes the flow rate of the continuous airflow and the purging duration, wherein the flow rate of the continuous airflow is proportional to the powder cluster volume index of the moderately attached region. A rotary brush cleaning mode is configured for the lightly attached area. The rotary brush cleaning mode includes the rotation speed of the brush motor and the number of brush reciprocations, wherein the number of brush reciprocations is fixed as a single reciprocation. The high-pressure pulse purging cleaning mode, the continuous airflow purging cleaning mode, and the rotary brush cleaning mode are sequentially bound to the corresponding moving target points according to the access order of the moving target points in the cleaning execution path sequence, thereby generating a cleaning execution path sequence with cleaning mode markings.
[0013] As a further aspect of the present invention, the method further includes a real-time verification operation of the powder removal effect performed during the execution of the powder removal execution path sequence: During the execution of the toner cleaning execution path sequence with the attached toner cleaning mode marker, when the toner cleaning execution mechanism moves to any moving target point and completes the toner cleaning mode operation corresponding to the arbitrary moving target point, the on-site cloud collection command is triggered. Real-time point cloud data of the area where the current moving target point is located is collected by a scanning probe installed at the end of the powder cleaning actuator, and the point cloud data after local powder cleaning is obtained. The point cloud data after local powder removal is input into the improved Euclidean clustering algorithm to perform point cloud segmentation processing, thereby obtaining a set of point clouds of residual powder clusters after powder removal; Calculate the total number of all data points in the point cloud set of residual powder clusters after the powder cleaning, and use it as the residual powder amount index; The residual powder quantity index is compared with the preset residual quantity qualified threshold. If the residual powder quantity index is less than the residual quantity qualified threshold, the current moving target point is marked as a qualified powder cleaning point and the next path segment in the powder cleaning execution path sequence is continued. If the residual powder quantity index is greater than or equal to the residual quantity qualification threshold, the current moving target point is added back to the head of the priority cleaning grid sequence, and the remaining powder cleaning execution path sequence is regenerated.
[0014] As a further aspect of the present invention, the method further includes a residual powder collection operation performed after the powder cleaning execution path sequence is completed: Obtain the opening status signal of the bottom powder discharge port of the molding cylinder, and confirm that the bottom powder discharge port is in the open position based on the opening status signal; The integrated powder cleaning actuator is moved to the center position of the inner bottom surface of the forming cylinder, and the integrated powder cleaning actuator is driven to move from the center position to the bottom powder discharge port along a spiral scanning path. The pitch of the spiral scanning path is determined by a fixed step length parameter. As the integrated powder cleaning actuator moves along the spiral scanning path, the negative pressure adsorption device is continuously activated, and the adsorption port of the negative pressure adsorption device is attached to the inner bottom surface of the forming cylinder. When the integrated powder cleaning actuator reaches the edge of the bottom powder discharge port, it shuts down the negative pressure adsorption device and generates a signal indicating that the residual powder collection is complete.
[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: An improved Euclidean clustering algorithm adjusts the clustering radius based on the difference in the angle between local normal vectors of the point cloud, calculating the rate of change of the angle between the normal vectors of adjacent points in real time during the point cloud segmentation stage. When the difference in the angle between the normal vectors of a local region exceeds a preset threshold, the clustering radius is automatically reduced to accurately separate the thin layer of powder adhering to the cylinder wall from the point cloud of the sidewall structure; in flat areas of the powder bed, the clustering radius is increased to quickly merge scattered powder point clouds into complete clusters. This dynamic adjustment mechanism eliminates the over-segmentation and under-segmentation phenomena of fixed-radius algorithms in complex geometric regions, improving the accuracy of powder cluster boundary recognition to the millimeter level.
[0016] The powder adhesion distribution map establishes a three-dimensional spatial topological relationship of powder distribution within the molding cylinder by quantifying and marking the position coordinates and volume parameters of powder clusters in each region. The powder cleaning execution path sequence is generated based on the cluster volume data in the map, prioritizing efficient coverage paths for large-volume powder areas and switching to a jogging cleaning mode in areas with trace amounts of powder. The path nodes integrate powder volume weight coefficients, driving the actuator to adaptively adjust the moving speed and negative pressure power of the powder suction head, employing a low-speed, high-flow mode in densely powdered areas and switching to a high-speed, low-power mode in sparsely powdered areas. Attached Figure Description
[0017] Figure 1 This is a flowchart of an integrated automatic powder cleaning method for additive manufacturing molding cylinders according to the present invention; Figure 2 A flowchart for the improved Euclidean clustering algorithm; Figure 3 A flowchart for generating a powder adhesion distribution map. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0020] See Figure 1 This invention provides an integrated automatic powder cleaning method for additive manufacturing molding cylinders, the specific method including: The original point cloud data set of the additive manufacturing molding cylinder to be cleaned is acquired using a 3D scanning device. This set completely includes the 3D spatial coordinate information of the cylinder's internal bottom surface, internal sidewalls, and the surface of the residual powder bed attached to them. Point cloud segmentation processing based on an improved Euclidean clustering algorithm is performed on the original point cloud data set. This algorithm dynamically adjusts the clustering search radius based on the difference in the angle between the local normal vectors of the point clouds, thereby effectively distinguishing the powder clusters attached to the cylinder surface from the cylinder structure itself, outputting a segmented powder cluster point cloud set and a segmented cylinder structure point cloud set. A powder adhesion distribution map is generated based on these two segmented point cloud sets. This map marks the presence of powder clusters and their estimated volume in various spatial regions inside the molding cylinder in a grid-like form. Based on this powder adhesion distribution map, a series of cleaning execution locations and their access order are automatically planned, forming a cleaning execution path sequence. This sequence is transmitted to the integrated cleaning execution mechanism to drive it to complete the automated cleaning operation.
[0021] In one embodiment of the present invention, see [reference] Figure 2The algorithm iterates through each data point in the original point cloud dataset. For the current computational data point, it finds all neighboring data points within its spatial neighborhood that are less than an initial fixed radius, forming a neighborhood point set. Using all points in this neighborhood point set, a plane fitting algorithm is used to calculate the local microplane containing the current computational data point, and then the normal vector of this plane is obtained as the local normal vector of the current computational data point. For each neighboring data point in the neighborhood point set, the cosine of the angle between its local normal vector and the local normal vector of the current computational data point is calculated. This cosine value is defined as the normal matching degree score between the two points. When this score is greater than a preset matching degree threshold, the neighboring data point is considered to have the same surface orientation as the current computational data point and is marked as a normal-consistent neighboring point. The number of all normal-consistent neighbors of the current computational data point is counted. The adaptive clustering radius corresponding to the current point is determined based on this quantity. The determination process includes: obtaining the number of normally consistent neighboring points of the data point; calculating the absolute value of the difference between this number and a preset threshold for the total number of neighboring points; inputting this absolute value into a preset radius mapping function to calculate the preliminary clustering radius, which is configured such that the output value is inversely proportional to the input value; comparing the obtained preliminary clustering radius with a preset minimum radius threshold; if the preliminary clustering radius is less than the minimum radius threshold, the minimum radius threshold is determined as the adaptive clustering radius of the data point; if the preliminary clustering radius is not less than the minimum radius threshold, it is further compared with a preset maximum radius threshold; if the preliminary clustering radius is greater than the maximum radius threshold, the maximum radius threshold is determined as the adaptive clustering radius of the data point; if the preliminary clustering radius is not greater than the maximum radius threshold, the preliminary clustering radius is directly determined as the adaptive clustering radius of the data point. This mechanism allows for the use of a larger search radius in regions where the point cloud normal changes gently to accelerate clustering, and a smaller search radius in regions where the normal changes drastically to improve segmentation accuracy. Using each data point as a seed point, a region growth clustering operation is performed using its corresponding adaptive clustering radius as the search radius. Points with consistent normals and spatial connectivity are grouped into the same point cloud cluster, thereby outputting a set of segmented powder cluster point clouds and a set of segmented cylinder structure point clouds.
[0022] In its implementation, the improved Euclidean clustering algorithm begins. The algorithm iterates through each data point in the original point cloud dataset. For the current data point, it searches its spatial neighborhood for all neighboring data points whose linear distance to the current data point is less than the initial fixed radius. These neighboring data points form a neighborhood set. For the current data point, the algorithm fits a local plane using all neighboring data points in the neighborhood set and calculates the normal vector of this local plane. This normal vector is then assigned to the current data point as its local normal vector. Next, the algorithm calculates the cosine of the angle between the local normal vector of the current data point and the local normal vectors of each neighboring data point in the neighborhood set. This cosine is defined as the normal matching score. In practice, when the normal matching score exceeds a preset matching threshold, the algorithm marks the corresponding neighboring data point as a normal-consistent neighbor of the current data point.
[0023] After counting the number of normally consistent neighbors for each data point, the algorithm determines the adaptive clustering radius corresponding to the current data point based on the number of normally consistent neighbors. In some embodiments, the process of determining the adaptive clustering radius includes obtaining the number of normally consistent neighbors for each data point and calculating the absolute value of the difference between the number of normally consistent neighbors and a preset threshold for the total number of neighbors. In a specific implementation, the absolute value of the difference can be input into a preset radius mapping function to calculate the preliminary clustering radius. The radius mapping function is configured such that the output value is inversely proportional to the input value. An optional radius mapping function expression is:
[0024] in: Represents the initial cluster radius. Represents the fundamental radius constant. Represents the scaling constant. The number of normally-consistent neighborhood points of a data point. This represents a preset threshold for the total number of neighboring points. It can be understood that this function ensures that the greater the difference between the number of normally aligned neighboring points and the threshold for the total number of neighboring points, the larger the initial cluster radius. The smaller.
[0025] After obtaining the initial cluster radius Then, the algorithm performs threshold comparisons to finally determine the adaptive clustering radius. The initial clustering radius is then... With the preset minimum radius threshold Comparison, if the initial cluster radius Less than the minimum radius threshold The algorithm then sets the minimum radius threshold. The adaptive clustering radius for the data points is determined. If the initial clustering radius... Not less than the minimum radius threshold The algorithm then sets the initial cluster radius. Compared with the preset maximum radius threshold Comparison. In some embodiments, if the initial cluster radius... Greater than the maximum radius threshold The algorithm then sets the maximum radius threshold. The adaptive clustering radius for the data points is determined. If the initial clustering radius... Not greater than the maximum radius threshold The algorithm then sets the initial cluster radius. The adaptive clustering radius is directly determined as the value of the data points. This process can be understood as strictly limiting the range of values for the adaptive clustering radius to a minimum radius threshold. and maximum radius threshold between.
[0026] In the specific implementation, after calculating the adaptive clustering radius for each data point, the algorithm uses each data point as the starting point and the adaptive clustering radius corresponding to the data point as the search radius to perform a region growing clustering operation. Specifically, the region growing clustering operation starts with any unvisited data point and adds it to the seed point queue of the current cluster. The algorithm retrieves a seed point from the seed point queue and searches for all neighboring data points whose spatial distance to the seed point is less than its adaptive clustering radius and which have already been marked as seed points with consistent normal direction. Optionally, if these found neighboring data points have not yet been classified into any point cloud cluster, they are classified into the current point cloud cluster and added to the seed point queue. This process is repeated until the seed point queue is empty, thus grouping a region of continuously connected data points with consistent normal direction neighbors into the same point cloud cluster. The algorithm traverses all data points and finally outputs a set of segmented powder cluster point clouds and a set of segmented cylinder structure point clouds.
[0027] In one embodiment of the present invention, before applying the improved Euclidean clustering algorithm to the original point cloud dataset, voxel downsampling filtering is performed to generate a downsampled point cloud dataset. The voxel downsampling filtering process involves uniformly dividing the three-dimensional space occupied by the original point cloud dataset into a series of fixed-size three-dimensional voxel grids. For each voxel grid, the average of the three-dimensional coordinates of all original data points within it is calculated, i.e., the centroid coordinates. Only the data point corresponding to these centroid coordinates is retained to replace all original points within that voxel grid. This step significantly reduces the total data volume while maintaining the overall shape characteristics of the point cloud. Statistical outlier removal processing is then performed on the downsampled point cloud dataset to remove noise points. For each data point in the downsampled point cloud dataset, the average distance to its K nearest neighbors is calculated, where K is a preset number. The global mean and standard deviation of the average distances corresponding to all points in the entire point cloud dataset are calculated. If the average distance of a point exceeds the global mean plus N times the standard deviation, where N is a preset standard deviation multiple, this point is identified as an outlier and removed from the point cloud dataset. After the above filtering and denoising processes, a clean point cloud dataset is generated. This clean point cloud dataset will be used as input to the improved Euclidean clustering algorithm to perform subsequent point cloud segmentation processing.
[0028] In specific implementations, voxel downsampling filtering is performed on the original point cloud dataset. This process involves dividing the spatial extent of the original point cloud dataset into equally sized three-dimensional voxel grids. The size of each three-dimensional voxel grid is predefined, for example, setting the side length of the three-dimensional voxel grid to 2.0 mm. In specific implementations, the algorithm traverses each divided three-dimensional voxel grid. For each three-dimensional voxel grid, the average of the three-dimensional spatial coordinates of all original data points in the current three-dimensional voxel grid is calculated. The spatial point corresponding to this average value is the barycentric coordinate. In specific implementations, voxel downsampling filtering removes all original data points from the current three-dimensional voxel grid, retaining only the single data point determined by the barycentric coordinate. In some embodiments, the original point cloud dataset may contain millions of data points. After voxel downsampling filtering, the number of data points in the generated downsampled point cloud dataset is significantly reduced, for example, to one-fifth of the original number, while essentially maintaining the geometric contour features of the molding cylinder and powder clusters.
[0029] In practical implementation, after generating the downsampled point cloud dataset, statistical outlier removal is performed. For each data point in the downsampled point cloud dataset, the average distance between the current data point and a preset number of its neighboring data points is calculated. In some embodiments, the preset number K can be set to 50, meaning that for each data point, the algorithm needs to find its 50 closest spatially adjacent data points in the downsampled point cloud dataset. The average distance is then calculated. The formula can be expressed as:
[0030] in: This represents the coordinate vector of the i-th data point in the downsampled point cloud dataset. represent The coordinate vector of the j-th nearest neighbor data point, Represents the preset quantity. This indicates the calculation of the Euclidean distance between two points. In practice, it is necessary to calculate the Euclidean distance for each data point in the downsampled point cloud dataset. Calculate its corresponding .
[0031] In practice, all data points are calculated. Then, the algorithm calculates the global mean of these average distances. and global standard deviation This is understandable; the global mean... This represents the overall density level of the data point distribution in the downsampled point cloud dataset, while the global standard deviation... This measures the dispersion of the density distribution. Statistical outlier removal processes remove outliers from each data point. It is compared to a threshold defined by a global statistic, which is the global mean. Add the preset standard deviation multiple Compared with global standard deviation The product. Optional, preset standard deviation multiple. It can be set to 2.5. When data points... of Greater than the threshold At that time, determine the data points Outlier points are identified. In practice, the algorithm removes all data points marked as outliers from the downsampled point cloud dataset. This process filters out isolated noise points far from the main point cloud caused by scanning noise or environmental dust. After statistical outlier removal, the resulting point cloud dataset is called the clean point cloud dataset. The clean point cloud dataset serves as input for subsequent point cloud segmentation processing and is fed into the improved Euclidean clustering algorithm. It can be understood that the combined effect of voxel downsampling filtering and statistical outlier removal reduces the total data volume and improves data quality while preserving the key geometric features of the original point cloud dataset.
[0032] In one embodiment of the present invention, see [reference] Figure 3 The internal bottom and sidewalls of the molding cylinder are unfolded and discretized, and divided into a regular two-dimensional grid array according to a preset grid size, forming a two-dimensional grid map. Each grid cell in the map corresponds to a fixed-size spatial region inside the molding cylinder. All data points in the segmented cylinder structure point cloud set are traversed, and each data point is projected onto the corresponding grid cell in the two-dimensional grid map according to its three-dimensional coordinates. This grid cell is marked as a cylinder structure occupying unit; this step is used to characterize the position of the cylinder surface. Each powder cluster point cloud in the segmented powder cluster point cloud set is traversed, and the minimum and maximum coordinate values of all data points within the cluster in three-dimensional space are calculated to obtain its three-dimensional axial bounding box. This bounding box defines the spatial occupancy range of the current powder cluster inside the molding cylinder. Based on this spatial occupancy range, all grid cells covered by this range are determined on the two-dimensional grid map, forming a grid cell set. Each grid cell in this set is marked as a powder cluster occupying unit. The total number of data points contained in the current powder cluster point cloud is counted and used as the volume estimation index for the powder cluster, i.e., the powder cluster volume index. This index is then associated with its corresponding set of grid cells. In the two-dimensional grid map, grid cells that are simultaneously marked as cylinder structure occupancy cells and powder cluster occupancy cells are selected. These cells represent the actual powder attachment area and are defined as powder attachment area markers. Integrating and storing all powder attachment area markers and their respective associated powder cluster volume indices constitutes the final powder attachment distribution map.
[0033] In specific implementation, the process of generating a powder adhesion distribution map begins by discretizing the inner bottom surface and inner sidewalls of the molding cylinder into a two-dimensional mesh map according to a preset grid size. In some embodiments, the preset grid size can be set to a square grid with a side length of 5 mm, and each grid cell in the two-dimensional mesh map corresponds to a fixed spatial region of 5 mm by 5 mm inside the molding cylinder. In specific implementation, the algorithm traverses all data points in the segmented cylinder structure point cloud set, projects the three-dimensional coordinates of each data point onto the corresponding two-dimensional plane coordinates according to the geometric model of the molding cylinder, and maps them to a specific grid cell in the two-dimensional mesh map, marking the currently projected grid cell as the cylinder structure occupying unit. It can be understood that this process establishes the occupancy mapping of the inner wall surface of the molding cylinder in the two-dimensional mesh map.
[0034] In practice, the process then iterates through each powder cluster point cloud in the segmented powder cluster point cloud set. For each powder cluster point cloud, the maximum and minimum values of all data points it contains are calculated along the X, Y, and Z coordinate axes. An optional formula for calculating the spatial occupancy range is as follows:
[0035] in: This represents the calculated three-dimensional axial bounding box. , , These represent the minimum X, Y, and Z coordinates of all data points in the current powder cluster point cloud, respectively. , , These represent the corresponding maximum coordinate values. This bounding box defines the spatial occupancy of the current powder cluster point cloud within the forming cylinder. Based on the spatial occupancy... The algorithm determines all grid cells within the X and Y coordinate coverage area of a two-dimensional grid map; these grid cells constitute a grid cell set. In some embodiments, for a spatial occupancy area... For powder clusters of [(10,20,5), (25,35,12)] (unit: mm), at a grid size of 5 mm, the grid cells they cover span from column 2 to column 5 in the X direction and from row 4 to row 7 in the Y direction, and the set of grid cells thus determined contains multiple grid cells.
[0036] In its implementation, the algorithm marks each grid cell in the grid cell set as a powder cluster occupying unit. The total number of data points in the current powder cluster point cloud is calculated, and this number is used as the powder cluster volume index corresponding to the current grid cell set. The powder cluster volume index can be understood as a quantitative indicator of the relative volume of powder clusters. The algorithm filters out grid cells in the two-dimensional grid map that are simultaneously marked as cylinder structure occupying units and powder cluster occupying units; these units constitute powder attachment region markers. All powder attachment region marker information and the powder cluster volume index corresponding to each marker are stored together to form the final powder attachment distribution map. Optionally, some data in the powder attachment distribution map can be stored in tabular form, as shown in Table 1, which records the association information between some grid cells and powder clusters.
[0037] Table 1: Powder Adhesion Distribution Map Data Table ; In one embodiment of the present invention, all grid cells marked as powder-attached areas are extracted from the powder adhesion distribution map to form a powder adhesion grid set. Based on the powder cluster volume index associated with each grid cell, the grid cells in this set are sorted from largest to smallest to generate a priority cleaning grid sequence, ensuring that areas with large powder amounts are processed first. The grid cell with the largest powder cluster volume index is selected from this priority cleaning grid sequence as the current cleaning target cell, and the coordinates of the center point corresponding to this cell within the molding cylinder are set as the target position that the powder cleaning actuator needs to move to, i.e., the target point. A two-dimensional grid map corresponding to the two-dimensional grid map is constructed, where each grid cell corresponds one-to-one with a grid cell. In the grid map, grid cells marked as cylinder structure occupants in the powder adhesion distribution map are marked as obstacle grids, representing areas that the powder cleaning actuator cannot traverse; the remaining grids are marked as free grids. The grid cell where the powder cleaning actuator is currently located is obtained as the starting grid, and the grid cell where the target point is located is obtained as the target grid. Starting from the initial grid, a heuristic search algorithm is used for path exploration. For each explored grid cell, the actual cost of moving from the initial grid to that grid is calculated, and the Manhattan distance from that grid to the target grid is added as the estimated total cost. The algorithm prioritizes expanding the grid cells with the lowest estimated total cost and records the predecessor grids of each expanded grid cell until the target grid cell is expanded. Starting from the target grid cell, the algorithm backtracks to the initial grid cell based on the recorded predecessor grid information, connecting the center points of all grid cells along the backtracking path in sequence to obtain the shortest movement path from the current position to the target point. This path is recorded as a path segment in the powder cleaning execution path sequence. When the powder cleaning actuator moves along this path segment to the target point, a dwell time is calculated based on the powder cluster volume index of the grid cell corresponding to the target point. This dwell time is proportional to the volume index, and the cleaning action is performed accordingly. This target point is marked as cleaned and removed from the priority cleaning grid sequence. Repeat the above operations of retrieving the next target unit from the sequence, planning the path, recording the path, determining the dwell time, and marking the cleaned status until the priority cleaned grid sequence becomes empty, resulting in a complete cleanup execution path sequence containing multiple path segments and corresponding dwell times.
[0038] In practical implementation, the process begins with generating a powder removal execution path sequence based on the powder adhesion distribution map. The algorithm extracts all grid cells with powder adhesion region markers from the powder adhesion distribution map, forming a powder adhesion grid set. In some embodiments, the powder adhesion grid set may contain dozens of grid cells, each associated with a powder cluster volume index. The algorithm sorts the grid cells in the powder adhesion grid set from largest to smallest according to their corresponding powder cluster volume index, generating an ordered list as a priority cleaning grid sequence, where the grid cell with the largest powder cluster volume index is located at the beginning of the sequence. See Table 2 for an example of the first few data points of the priority cleaning grid sequence.
[0039] Table 2: Priority Cleanup Grid Sequence List ; In practice, a path planning algorithm is initiated to calculate the movement path. The algorithm first constructs a two-dimensional grid map of the inside of the forming cylinder, with each grid cell corresponding one-to-one with each grid cell in the powder adhesion distribution map. The algorithm marks grid cells occupied by cylinder structure units as obstacle grids and grid cells not occupied by cylinder structure units as free grids. The current grid cell of the powder cleaning actuator is obtained as the starting grid, and the grid cell with the largest powder agglomeration volume index is selected from the priority cleaning grid sequence as the current cleaning target unit. The center point of the current cleaning target unit is used as the movement target point, and the grid cell containing the movement target point is obtained as the target grid. Starting from the starting grid, a heuristic search method is used to explore adjacent grid cells. For each explored grid cell, the path planning algorithm calculates an estimated cost. ,calculate The formula is:
[0040] in: This represents the actual cost of moving from the starting grid cell to the currently explored grid cell. This represents the estimated cost of the Manhattan distance from the currently explored raster cell to the target raster. The algorithm prioritizes expansion. The algorithm identifies the grid cell with the smallest value and records the predecessor grid cell for each expanded grid cell. This process continues until the target grid cell is expanded. Starting from the target grid cell, the algorithm backtracks to the starting grid cell based on the predecessor grid cell information recorded for each grid cell. It connects the center point coordinates of all grid cells traversed during the backtracking process in the backtracking order to obtain the shortest path from the starting position to the current target point. In practice, this shortest path is recorded as a segment in the powder cleaning execution path sequence.
[0041] After the cleaning actuator reaches the target point along the calculated shortest path, the system determines the dwell time at the target point based on the powder cluster volume index corresponding to the current cleaning target unit. In some embodiments, the dwell time... With powder cluster volume index Proportional, for example through a linear relationship Confirmed, among which and The system uses preset coefficients. It marks the grid cell corresponding to the current moving target point as cleaned and removes it from the priority cleaning grid sequence. Optionally, the algorithm then checks if the priority cleaning grid sequence is empty. If not, it takes the new grid cell with the largest powder cluster volume index from the head of the updated priority cleaning grid sequence as the next current cleaning target cell, and repeats the operations of acquiring the target grid, calculating the shortest movement path, recording the path segment, determining the dwell time, and marking it as cleaned. In practice, this process is repeated until the priority cleaning grid sequence becomes an empty set, resulting in a complete powder cleaning execution path sequence containing multiple ordered moving target points, their corresponding path segments, and dwell times.
[0042] In one embodiment of the present invention, after generating a powder adhesion distribution map, the adhesion areas are classified according to the powder cluster volume index corresponding to each powder adhesion area in the map. Areas with a powder cluster volume index greater than a first volume threshold are classified as heavily adhered areas, areas with an index between a second volume threshold and the first volume threshold are classified as moderately adhered areas, and areas with an index less than the second volume threshold are classified as lightly adhered areas. Differentiated cleaning modes are configured for different levels of areas. A high-pressure pulse purging cleaning mode is configured for heavily adhered areas, with parameters including the on-time of the high-pressure air source, pulse frequency, and pulse duty cycle, and the pulse frequency and duty cycle being proportional to the powder cluster volume index of the area. A continuous airflow purging cleaning mode is configured for moderately adhered areas, with parameters including the flow rate of the continuous airflow and the purging duration, and the flow rate being proportional to the powder cluster volume index. A rotary brush cleaning mode is configured for lightly adhered areas, with parameters including the rotation speed of the brush motor and a fixed number of single reciprocating brush strokes. These powder removal mode parameters are bound to the corresponding target points according to the access order of each moving target point in the powder removal execution path sequence, generating a powder removal execution path sequence with powder removal mode markers. During the execution of this sequence, when the powder removal execution mechanism completes the bound powder removal mode operation at any moving target point, a point cloud acquisition command is triggered. Real-time point cloud data of the area surrounding that point is acquired through a scanning probe installed at the end of the execution mechanism, obtaining local point cloud data after powder removal. This local point cloud data is input into an improved Euclidean clustering algorithm for processing, obtaining a set of residual powder cluster point clouds in the area after powder removal. The total number of all data points in this set is calculated as the residual powder amount index. This index is compared with a preset residual amount qualification threshold. If it is less than the threshold, the current point is marked as a qualified powder removal point, and the next path segment in the path sequence continues to be executed. If it is greater than or equal to the threshold, the grid cell corresponding to the current point is reinserted into the head of the priority cleaning grid sequence, and the remaining powder removal execution path sequence is re-planned and generated based on this for further cleaning. After the entire powder removal execution path sequence is completed, the residual powder collection operation is performed. The system acquires the opening status signal of the powder discharge port at the bottom of the molding cylinder, confirming that it is open. The integrated powder cleaning actuator is moved to the center of the bottom surface inside the molding cylinder and driven to move along a spiral path from the center towards the bottom powder discharge port. The pitch of the spiral path is determined by a fixed step length parameter. During the movement, the negative pressure adsorption device is continuously activated, with its adsorption port adhering to the bottom surface of the cylinder to collect the scattered powder swept to the bottom. When the actuator moves to the edge of the powder discharge port, the negative pressure adsorption device is turned off, and a signal indicating that residual powder collection is complete is generated.
[0043] In specific implementation, after generating the powder adhesion distribution map, a powder removal strategy classification process is performed. This classification process divides regions based on the powder cluster volume index corresponding to each powder adhesion region marker in the powder adhesion distribution map. In some embodiments, a first volume threshold is set to 1500, and a second volume threshold is set to 500. Powder adhesion regions with a powder cluster volume index greater than the first volume threshold of 1500 are classified as heavily adhered regions; powder adhesion regions with a powder cluster volume index between the second volume threshold of 500 and the first volume threshold of 1500 are classified as moderately adhered regions; and powder adhesion regions with a powder cluster volume index less than the second volume threshold of 500 are classified as lightly adhered regions. In specific implementation, a high-pressure pulse purging powder removal mode is configured for heavily adhered regions. The high-pressure pulse purging powder removal mode includes the on-time of the high-pressure air source, the pulse frequency, and the pulse duty cycle, wherein the pulse frequency and pulse duty cycle are proportional to the powder cluster volume index of the heavily adhered region. An optional formula for calculating the pulse frequency is:
[0044] in: This represents the calculated pulse frequency. Represents the frequency proportionality coefficient. This represents the powder agglomeration volume index corresponding to the currently heavily adhered area. A continuous airflow purging cleaning mode is configured for the moderately adhered area. This mode includes the flow rate and duration of the continuous airflow, where the flow rate is proportional to the powder agglomeration volume index of the moderately adhered area. A rotary brush cleaning mode is configured for the lightly adhered area. This mode includes the rotational speed of the brush motor and the number of brush reciprocations, where the number of reciprocations is fixed as a single reciprocation. In practice, the high-pressure pulse purging cleaning mode, the continuous airflow purging cleaning mode, and the rotary brush cleaning mode are sequentially bound to the corresponding moving target points according to the access order of the moving target points in the cleaning execution path sequence, generating a cleaning execution path sequence with cleaning mode markers.
[0045] In practical implementation, real-time verification of the powder removal effect is performed during the execution of the powder removal execution path sequence with attached powder removal mode markings. During the execution path sequence, when the powder removal actuator moves to any target point and completes the corresponding powder removal mode operation, the system automatically triggers a point cloud acquisition command. Real-time point cloud data of the area where the current target point is located is acquired using a scanning probe installed at the end of the powder removal actuator, obtaining local post-powder removal point cloud data. This local post-powder removal point cloud data reflects the surface state of the local area after the powder removal operation. In practical implementation, the local post-powder removal point cloud data is input into an improved Euclidean clustering algorithm for point cloud segmentation processing, resulting in a set of residual powder cluster point clouds after powder removal. The total number of all data points in the set of residual powder cluster point clouds after powder removal is calculated, and this total number is used as the residual powder quantity index. The residual powder quantity index is compared with a preset residual quantity qualification threshold, which can be set to 50. In some embodiments, if the residual powder index is less than the residual acceptable threshold of 50, the system marks the current moving target point as a qualified powder cleaning point and controls the powder cleaning actuator to continue executing the next path segment in the powder cleaning execution path sequence. If the residual powder index is greater than or equal to the residual acceptable threshold of 50, the system re-adds the current moving target point to the head of the priority cleaning grid sequence and re-plans and generates the remaining powder cleaning execution path sequence based on the updated priority cleaning grid sequence. Optionally, the re-addition operation means that the position will be cleaned again in subsequent powder cleaning cycles.
[0046] In specific implementation, residual powder collection is performed after the powder cleaning execution path sequence is completed. The system acquires the opening status signal of the bottom powder discharge port of the molding cylinder, and confirms that the bottom powder discharge port is in the open position based on the opening status signal. It can be understood that the opening of the bottom powder discharge port is a prerequisite for powder collection. The integrated powder cleaning actuator is moved to the center position of the inner bottom surface of the molding cylinder, and driven to move along the spiral scanning path from the center position towards the bottom powder discharge port. In some embodiments, the pitch of the spiral scanning path is determined by a fixed step length parameter, for example, the fixed step length parameter is set to 20 mm, which means that the radius increases by 20 mm for each rotation of the spiral path. In specific implementation, during the movement of the integrated powder cleaning actuator along the spiral scanning path, the negative pressure adsorption device is continuously activated, and the adsorption port of the negative pressure adsorption device moves in contact with the inner bottom surface of the molding cylinder. When the integrated powder cleaning actuator moves to the edge of the bottom powder discharge port, the system closes the negative pressure adsorption device and generates a residual powder collection completion signal.
[0047] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.
Claims
1. An additive manufacturing method of forming a cylinder-integrated automatic powder cleaning method, characterized by, The method includes: Obtain the original point cloud data set of the additive manufacturing molding cylinder to be cleaned, the original point cloud data set including the point cloud of the inner bottom surface of the molding cylinder, the point cloud of the inner side wall, and the point cloud of the residual powder bed surface; The original point cloud data set is subjected to point cloud segmentation processing based on an improved Euclidean clustering algorithm to obtain a segmented powder cluster point cloud set and a segmented cylinder structure point cloud set. The improved Euclidean clustering algorithm adjusts the clustering radius based on the difference in the angle between the local normal vectors of the point cloud. The powder adhesion distribution map of the molding cylinder is generated based on the segmented powder cluster point cloud set and the segmented cylinder structure point cloud set. The powder adhesion distribution map is used to mark the position and volume of powder clusters in each region inside the molding cylinder. Based on the powder adhesion distribution map, a powder cleaning execution path sequence is generated, and the powder cleaning execution path sequence is transmitted to the integrated powder cleaning execution mechanism.
2. The method of claim 1, wherein the method further comprises: The workflow of the improved Euclidean clustering algorithm includes: Traverse each data point in the original point cloud data set and calculate the neighborhood point set of each data point. The neighborhood point set consists of all neighboring data points whose spatial straight-line distance from the currently calculated data point is less than the initial fixed radius. For each data point, the local plane of the current calculated data point is fitted using all neighboring data points in the neighborhood point set, and the normal vector of the local plane is calculated as the local normal vector of the current calculated data point. Calculate the cosine of the angle between the local normal vector of the current data point and the local normal vector of the corresponding neighboring data point, and determine the normal matching score between the current data point and the corresponding neighboring data point based on the cosine of the angle. When the normal matching score is greater than the preset matching threshold, the corresponding neighboring data point is marked as the normal consistent neighborhood point of the currently calculated data point; The number of normal-consistent neighborhood points for each data point is counted, and the adaptive clustering radius corresponding to the current data point is determined based on the number of normal-consistent neighborhood points. The adaptive clustering radius is inversely proportional to the number of normal-consistent neighborhood points, and the value range of the adaptive clustering radius is limited to between the minimum radius value and the maximum radius value. Using each data point as the starting point of the search, and using the adaptive clustering radius as the search radius, a region growing clustering operation is performed. Data points with continuous connectivity of neighboring points with consistent normal directions are grouped into the same point cloud cluster, and the segmented powder cluster point cloud set and the segmented cylinder structure point cloud set are output.
3. The method of claim 1, wherein the method further comprises: Perform point cloud segmentation processing based on an improved Euclidean clustering algorithm on the original point cloud dataset to obtain a segmented powder cluster point cloud set and a segmented cylinder structure point cloud set, and further include: Before performing point cloud segmentation processing on the original point cloud data set, voxel downsampling filtering processing is performed on the original point cloud data set to generate a downsampled point cloud data set. The voxel downsampling filtering processing includes dividing the spatial range of the original point cloud data set into three-dimensional voxel grids of equal size. In each three-dimensional voxel grid, only a single data point corresponding to the centroid coordinates of all data points in the current three-dimensional voxel grid is retained. Statistical outlier removal processing is performed on the downsampled point cloud dataset. For each data point in the downsampled point cloud dataset, the average distance between the current calculated data point and a preset number of neighboring data points is calculated. When the average distance deviates from the global average distance by more than a preset standard deviation multiple, the current calculated data point is marked as an outlier and removed from the downsampled point cloud dataset to generate a clean point cloud dataset. The clean point cloud dataset is input into the improved Euclidean clustering algorithm to perform point cloud segmentation processing.
4. The method of claim 1, wherein the method further comprises: The powder adhesion distribution map of the molding cylinder is generated based on the segmented powder cluster point cloud set and the segmented cylinder structure point cloud set, including: The inner bottom surface and inner sidewall of the forming cylinder are discretized into a two-dimensional grid pattern according to a preset grid size. Each grid cell in the two-dimensional grid pattern corresponds to a fixed spatial region inside the forming cylinder. Traverse all data points in the segmented cylinder block structure point cloud set, project each data point onto the corresponding grid cell of the two-dimensional grid map, and mark the currently projected grid cell as the cylinder block structure occupying cell; Traverse each powder cluster point cloud in the segmented powder cluster point cloud set, calculate the three-dimensional bounding box of all data points contained in the powder cluster point cloud, and obtain the spatial occupancy range of the powder cluster point cloud inside the molding cylinder. The set of grid cells covered by the current powder cluster point cloud is determined based on the spatial occupancy range. Each grid cell in the set of grid cells is marked as a powder cluster occupying cell. At the same time, the total number of data points in the current powder cluster point cloud is calculated as the powder cluster volume index corresponding to the current set of grid cells. The grid cells in the two-dimensional grid map that are simultaneously marked as cylinder structure occupying units and powder cluster occupying units are selected to form powder attachment area markers. The powder attachment area markers and the powder cluster volume index corresponding to each powder attachment area marker are stored together as the powder attachment distribution map.
5. The method of claim 4, wherein the method further comprises: Based on the powder adhesion distribution map, a powder cleaning execution path sequence is generated, including: Extract all grid cells with powder adhesion region markers from the powder adhesion distribution map to form a powder adhesion grid set; The grid cells in the powder adhesion grid set are sorted from largest to smallest according to their corresponding powder cluster volume index to generate a priority cleaning grid sequence; The grid cell with the largest powder agglomeration volume index is selected from the priority cleaning grid sequence as the current cleaning target cell, and the center point of the current cleaning target cell is used as the moving target point of the powder cleaning actuator. A path planning algorithm is used to calculate the shortest movement path between the current position of the cleaning actuator and the moving target point, and the shortest movement path is recorded as a path segment in the cleaning execution path sequence; After the cleaning actuator reaches the moving target point, the dwell time at the moving target point is determined according to the powder cluster volume index corresponding to the current cleaning target unit. The dwell time is proportional to the powder cluster volume index. The moving target point is marked as cleared and removed from the priority cleaning grid sequence. The operations of retrieving, calculating, recording, determining the dwell time and marking the cleared state are repeated until the priority cleaning grid sequence is empty, thus obtaining a complete powder cleaning execution path sequence.
6. The method of claim 5, wherein the method further comprises: The path planning algorithm includes: A two-dimensional grid map is constructed inside the forming cylinder. Each grid cell in the two-dimensional grid map corresponds one-to-one with each grid cell in the two-dimensional grid map. Grid cells with cylinder structure occupying cells are marked as obstacle grids, and grid cells without cylinder structure occupying cells are marked as free grids. The current grid cell of the powder cleaning actuator is obtained as the starting grid, and the grid cell where the moving target point is located is obtained as the target grid; Starting from the initial grid, a heuristic search method is used to explore adjacent grids. For each explored grid cell, the estimated cost of reaching the target grid from the initial grid through the explored grid cells is calculated. The estimated cost is obtained by adding the actual movement cost from the initial grid to the explored grid cell to the estimated Manhattan distance cost from the explored grid cell to the target grid cell. Prioritize expanding the raster cell with the lowest estimated cost, and record the predecessor raster cell for each expanded raster cell, until the target raster is expanded; Starting from the target grid, the system backtracks to the starting grid sequentially based on the predecessor grid cells of each grid cell. The center points of all the grid cells traversed during the backtracking are connected in the backtracking order to obtain the shortest moving path.
7. The method of claim 2, wherein the method further comprises: The number of normally consistent neighborhood points for each data point is counted, and the adaptive clustering radius corresponding to the current data point is determined based on the number of normally consistent neighborhood points, including: Obtain the number of normal-consistent neighbor points of the data point; Calculate the absolute value of the difference between the number of normal-consistent neighbor points and the preset threshold for the total number of neighbor points; The absolute value of the difference is input into a preset radius mapping function to calculate the preliminary cluster radius. The radius mapping function is configured such that the output value is inversely proportional to the input value. The initial clustering radius is compared with a preset minimum radius threshold. If the initial clustering radius is less than the minimum radius threshold, the minimum radius threshold is determined as the adaptive clustering radius of the data points. If the initial clustering radius is not less than the minimum radius threshold, then the initial clustering radius is compared with the preset maximum radius threshold. If the initial clustering radius is greater than the maximum radius threshold, then the maximum radius threshold is determined as the adaptive clustering radius of the data points. If the initial clustering radius is not greater than the maximum radius threshold, then the initial clustering radius is directly determined as the adaptive clustering radius of the data points.
8. The integrated automatic powder cleaning method for additive manufacturing molding cylinders according to claim 4, characterized in that, The method further includes a powder removal strategy grading process performed after the powder adhesion distribution map is generated: Based on the powder cluster volume index corresponding to each powder adhesion region marker in the powder adhesion distribution map, the powder adhesion region markers are divided into heavily adhered regions, moderately adhered regions, and lightly adhered regions. The heavily adhered regions correspond to regions where the powder cluster volume index is greater than the first volume threshold, the moderately adhered regions correspond to regions where the powder cluster volume index is between the second volume threshold and the first volume threshold, and the lightly adhered regions correspond to regions where the powder cluster volume index is less than the second volume threshold. A high-pressure pulse purging and dust removal mode is configured for the heavily adhered area. The high-pressure pulse purging and dust removal mode includes the on-time of the high-pressure air source, the pulse frequency, and the pulse duty cycle, wherein the pulse frequency and the pulse duty cycle are proportional to the volume index of the powder clusters in the heavily adhered area. A continuous airflow purging and dust removal mode is configured for the moderately attached region. The continuous airflow purging and dust removal mode includes the flow rate of the continuous airflow and the purging duration, wherein the flow rate of the continuous airflow is proportional to the powder cluster volume index of the moderately attached region. A rotary brush cleaning mode is configured for the lightly attached area. The rotary brush cleaning mode includes the rotation speed of the brush motor and the number of brush reciprocations, wherein the number of brush reciprocations is fixed as a single reciprocation. The high-pressure pulse purging cleaning mode, the continuous airflow purging cleaning mode, and the rotary brush cleaning mode are sequentially bound to the corresponding moving target points according to the access order of the moving target points in the cleaning execution path sequence, thereby generating a cleaning execution path sequence with cleaning mode markings.
9. The integrated automatic powder cleaning method for additive manufacturing molding cylinders according to claim 8, characterized in that, The method also includes a real-time verification operation of the cleanup effect performed during the execution of the cleanup execution path sequence: During the execution of the toner cleaning execution path sequence with the attached toner cleaning mode marker, when the toner cleaning execution mechanism moves to any moving target point and completes the toner cleaning mode operation corresponding to the arbitrary moving target point, the on-site cloud collection command is triggered. Real-time point cloud data of the area where the current moving target point is located is collected by a scanning probe installed at the end of the powder cleaning actuator, and the point cloud data after local powder cleaning is obtained. The point cloud data after local powder removal is input into the improved Euclidean clustering algorithm to perform point cloud segmentation processing, thereby obtaining a set of point clouds of residual powder clusters after powder removal; Calculate the total number of all data points in the point cloud set of residual powder clusters after the powder cleaning, and use it as the residual powder amount index; The residual powder quantity index is compared with the preset residual quantity qualified threshold. If the residual powder quantity index is less than the residual quantity qualified threshold, the current moving target point is marked as a qualified powder cleaning point and the next path segment in the powder cleaning execution path sequence is continued. If the residual powder quantity index is greater than or equal to the residual quantity qualification threshold, the current moving target point is added back to the head of the priority cleaning grid sequence, and the remaining powder cleaning execution path sequence is regenerated.
10. The integrated automatic powder cleaning method for additive manufacturing molding cylinders according to claim 1, characterized in that, The method further includes a residual powder collection operation performed after the powder cleaning execution path sequence has been completed: Obtain the opening status signal of the bottom powder discharge port of the molding cylinder, and confirm that the bottom powder discharge port is in the open position based on the opening status signal; The integrated powder cleaning actuator is moved to the center position of the inner bottom surface of the forming cylinder, and the integrated powder cleaning actuator is driven to move from the center position to the bottom powder discharge port along a spiral scanning path. The pitch of the spiral scanning path is determined by a fixed step length parameter. As the integrated powder cleaning actuator moves along the spiral scanning path, the negative pressure adsorption device is continuously activated, and the adsorption port of the negative pressure adsorption device is attached to the inner bottom surface of the forming cylinder. When the integrated powder cleaning actuator reaches the edge of the bottom powder discharge port, it shuts down the negative pressure adsorption device and generates a signal indicating that the residual powder collection is complete.
