Multi-nozzle collaborative printing method

By constructing a three-dimensional spatial mesh model and optimizing the printing path using graph convolutional neural networks, the problems of overlapping printing areas and path conflicts in multi-head printing technology are solved, improving printing efficiency and stability and ensuring high-precision printing results.

CN121614097APending Publication Date: 2026-03-06YIXING HUALI TINPLATE PRINTING & CAN-MAKING CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202511800010.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-02
Publication Date
2026-03-06

AI Technical Summary

Technical Problem

In high-precision manufacturing fields such as flexible electronics, microstructure patterning, and biochip printing, multi-nozzle parallel printing technology suffers from problems such as overlapping printing areas, path conflicts, printhead waiting, and nozzle impacts. In particular, printing accuracy deviations and path interference occur frequently in irregular curved surfaces or areas with multiple interferences.

Method used

By constructing a three-dimensional spatial grid model, the printhead collaborative coverage parameters are obtained, a printhead task priority map is constructed, the minimum cross-entropy objective function is used for dynamic matching of printhead tasks, and a graph convolutional neural network is used to optimize the printing path and generate an optimized printing path scheme.

Benefits of technology

It significantly improves printhead utilization and overall printing efficiency, reduces printhead idle rate and waiting time, reduces path interference and printing errors, and improves the stability and finished product quality of multi-printhead printing systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121614097A_ABST
    Figure CN121614097A_ABST
Patent Text Reader

Abstract

The invention discloses a multi-nozzle collaborative printing method, and particularly relates to the technical field of printing path planning. According to the method, a three-dimensional space mesh model is constructed by collecting structural feature information of a target printing area; calculating collaborative coverage parameters of the sub-regions according to the space coverage capability and the region curvature characteristics of the nozzle; constructing a priority map based on nozzle performance and task suitability, and performing task matching by adopting a minimum cross entropy objective function to generate an initial path allocation scheme; constructing a jet printing path conflict graph, extracting conflict feature points, performing local path reconstruction by using a graph convolutional neural network, and generating an optimized jet printing path; and finally, the multiple nozzles are driven to cooperatively execute a jet printing task through time synchronization and instruction scheduling. According to the invention, the conflict rate and waiting time of the nozzles can be obviously reduced, the path planning efficiency and the jet printing precision are improved, and the method has good engineering applicability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of printing path planning technology, and more specifically to a multi-head collaborative printing method. Background Technology

[0002] Currently, in high-precision manufacturing fields such as flexible electronics, microstructure patterning, and biochip printing, multi-head parallel printing technology is gradually becoming mainstream due to its advantages of high efficiency and high resolution. However, in practical applications, the lack of efficient coordination of the working paths of multiple printheads often leads to problems such as overlapping printing areas, path conflicts, printhead waiting, and error superposition, which seriously affect the overall processing quality and efficiency.

[0003] Especially in micro-pattern printing tasks on irregular curved surfaces or in areas with multiple interferences, the area covered by each printhead is dynamically changing, and there are complex geometric overlaps or dead zones between printing paths. Most existing methods use preset sequences or simplified heuristic algorithms for printhead path allocation, lacking the ability to model and dynamically optimize printhead cooperation in real time. This makes them prone to printing accuracy deviations in local areas, or high-risk problems such as nozzle collisions and printing failures due to path interference during high-speed printhead switching. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-printer collaborative printing method to address the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a multi-head collaborative printing method, comprising:

[0006] Collect structural feature information of the target printing area and construct a three-dimensional spatial mesh model of the target printing area;

[0007] The three-dimensional spatial mesh model is divided into regions to obtain the cooperative coverage parameter Cn of each sub-region in the multi-nozzle reachability matrix, wherein the cooperative coverage parameter represents the degree of overlap of each nozzle on the sub-region;

[0008] Based on the collaborative coverage parameter Cn of each sub-region, a nozzle task priority map G is constructed to represent the task competition weight of each nozzle in different sub-regions.

[0009] Based on the priority map G, the printhead task is dynamically matched using the minimum cross-entropy objective function to obtain the initial printing path allocation scheme.

[0010] Based on the initial printing path allocation scheme, a printing path conflict map is constructed, and the path intersection conflict feature points F are extracted.

[0011] Based on the conflict feature points F, a graph convolutional neural network is used to locally reconstruct the path conflict and generate an optimized printing path scheme.

[0012] Based on the optimized path scheme P, each printhead is driven to perform collaborative printing operations according to the optimized path.

[0013] Preferably, the construction of the three-dimensional spatial mesh model of the target printing area includes:

[0014] Multi-angle surface data acquisition is performed on the target printing area to obtain a point cloud dataset containing depth information;

[0015] Noise filtering and curvature fitting are performed on the point cloud dataset to extract the boundary contour lines, local concave and convex regions, and curvature distribution features of the target printing area.

[0016] Based on the processed surface feature information, the octree partitioning algorithm is used to divide the printing area into multiple refined spatial units, forming a hierarchical three-dimensional spatial mesh model.

[0017] Preferably, the three-dimensional spatial mesh model is divided into regions to obtain the cooperative coverage parameter Cn of each sub-region in the multi-nozzle reachability matrix, including:

[0018] The three-dimensional spatial mesh model is clustered according to the geometric continuity and curvature similarity of each spatial unit to form multiple sub-region units with relatively uniform structure.

[0019] Based on the printing motion boundary and pose change range of each printhead, a printhead coverage area model is constructed, and ray projection judgment is performed on each sub-region unit to determine whether it is within the effective printing range of one or more printheads.

[0020] For a sub-region within the coverage area of ​​multiple printheads, calculate the printing coverage of each printhead and obtain the collaborative coverage parameter Cn of the sub-region using a linear superposition method.

[0021] Preferably, the step of constructing a nozzle task priority map G based on the collaborative coverage parameter Cn of each sub-region includes:

[0022] The collaborative coverage parameters of each sub-region are matched with the printing capability parameters of the corresponding printhead. Based on the printhead printing speed, printing resolution and jetting stability, a printhead performance evaluation vector is constructed to describe the printhead's operational adaptability in different sub-regions.

[0023] Based on the nozzle performance evaluation vector, weight allocation processing is performed on sub-regions with high collaborative coverage parameters. By calculating the comprehensive adaptation score of the nozzle in the sub-region, the initial value of the task competition weight of the nozzle for the sub-region is obtained.

[0024] The initial values ​​of the task competition weights of each nozzle in the same sub-region are normalized so that the sum of the competition weights of each nozzle in the sub-region is 1, thus forming a nozzle task competition weight matrix at the sub-region level.

[0025] The printhead task competition weight matrices of all sub-region levels are integrated according to spatial adjacency to construct a printhead task priority map, which is used to characterize the task priority distribution of each printhead in the global printing area.

[0026] Preferably, the step of dynamically matching printhead tasks using the minimum cross-entropy objective function based on the priority map G to obtain an initial printing path allocation scheme includes:

[0027] The task competition weights of each sub-region corresponding to each nozzle in the nozzle task priority map are used as the target probability distribution, and the predicted probability distribution of nozzle task matching is constructed as the model output.

[0028] A matching relationship matrix between nozzles and sub-regions is defined, and the optimization objective is to minimize the cross-entropy loss function. The matching weights are iteratively optimized to reduce the information difference between the predicted distribution and the target distribution.

[0029] In each iteration, matching constraints are introduced by combining the remaining operating capacity of the nozzles and the spatial continuity between sub-regions to limit the occurrence of unreasonable task allocation behavior.

[0030] After the cross-entropy loss converges to a set threshold or the maximum number of iterations is reached, the final matching result between the printhead and the sub-region is output, and an initial printing path allocation scheme is generated based on the matching relationship.

[0031] Preferably, based on the initial printing path allocation scheme, a printing path conflict map is constructed, and the path intersection conflict feature points F are extracted, including:

[0032] The path trajectory of each printhead in the initial inkjet path allocation scheme is discretized according to the time series to generate a path sequence dataset containing path point coordinates and timestamps.

[0033] All printhead path sequences are compared pairwise. Based on the overlap of spatial positions and the coincidence of time windows, it is determined whether there are potential conflicts in the paths. A print path conflict map is constructed with path pairs as edges.

[0034] In the conflict map, path intersection regions with high conflict intensity are identified, and path intersection conflict feature points are extracted as key nodes for local conflict optimization based on conflict frequency and spatial concentration indices.

[0035] Preferably, based on the conflict feature points F, a graph convolutional neural network is used to locally reconstruct the path conflict and generate an optimized printing path scheme, including:

[0036] The conflict feature points are constructed into a node set according to their spatial adjacency relationship, and an edge set is established based on the connection relationship between adjacent path points in the path sequence, thereby forming a conflict subgraph structure for learning.

[0037] The spatial coordinates, timestamps, and nozzle numbers of each node in the conflict subgraph are used as input feature vectors, and all input features are normalized to form the input feature matrix of the graph convolutional neural network.

[0038] A graph convolutional neural network with multiple convolutional kernels is used to extract features from the input feature matrix. The spatial correlation and temporal change patterns of conflict regions are learned through cross-node convolutional propagation, and the path adjustment vector of the output node is output.

[0039] The original path sequence is locally reconstructed based on the node path adjustment vector, and the path intersection area in the conflict subgraph is replaced with the alternative path segment optimized by the graph convolutional neural network, thereby generating an optimized inkjet printing path scheme.

[0040] Preferably, the step of driving each printhead to perform collaborative printing operations according to the optimized path scheme P includes:

[0041] The path sequence of each nozzle in the optimized path scheme is synchronized in time to build a unified execution time axis, ensuring that all nozzles have the ability to start and stop collaboratively at key nodes.

[0042] Based on the spatial displacement of each printhead path point and the printing rate, a printing control instruction set is generated, including printhead movement trajectory instructions, spray trigger instructions, and spray duration parameters.

[0043] Correcting execution deviations caused by printhead response time and mechanical inertia during the printing process;

[0044] According to the control instruction set, each printhead is driven to coordinately execute the printing task in the time axis scheduling order.

[0045] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0046] 1. This invention, by introducing multi-printerhead collaborative coverage parameters, task priority maps, and minimum cross-entropy optimization strategies, achieves dynamic matching and spatial coordination of multi-printerhead tasks, effectively solving problems such as duplicate printing paths, uneven resource allocation, and frequent path conflicts in existing technologies. Especially when processing complex graphics and large-scale printing tasks, it can significantly improve printer utilization and overall printing efficiency, and reduce printer idle rate and waiting time.

[0047] 2. This invention constructs a printing path conflict map and uses a graph convolutional neural network to locally reconstruct the conflict area, which can adaptively identify and optimize high-risk path intersections, greatly reducing path interference and printing errors. Combined with a collaborative scheduling and delay compensation mechanism based on a unified time axis, it ensures the synchronization and high-precision path tracking of each printhead during the execution process, further improving the stability and product quality of the multi-printhead printing system. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0049] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 embodiments of the present invention, not all embodiments. 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.

[0051] For examples, please refer to Figure 1 As shown in this embodiment, a multi-printer collaborative printing method includes:

[0052] Collect structural feature information of the target printing area and construct a three-dimensional spatial mesh model of the target printing area.

[0053] First, a high-precision 3D scanner or laser scanner is used to scan the target printing area from multiple angles. This scanner can capture point cloud data of the target surface, including the spatial location, shape, and depth information of the target area. After each scan, the device stores the measured spatial coordinate values ​​(i.e., point cloud) as a 3D dataset, which includes the X, Y, and Z coordinates of each point, as well as the associated surface reflection intensity information.

[0054] After obtaining point cloud data, noise filtering is performed first to remove invalid data points caused by environmental factors or equipment errors. Noise filtering typically uses statistical analysis methods, such as Gaussian filtering algorithms, to smooth the point cloud and eliminate irregular or isolated outlier data points.

[0055] Curvature fitting is performed on the processed point cloud data to extract the local curvature distribution of the target region. Curvature fitting is achieved by calculating the rate of change of neighboring points for each point, typically using the least squares fitting algorithm to obtain the curvature value of each point in the point cloud. Specifically, for each point in the point cloud dataset, the curvature change trend of its local neighborhood points is calculated to obtain the curvature value. Based on this, the boundary contour lines and local concave and convex regions of the target region are extracted. These features provide necessary geometric information for subsequent 3D spatial modeling. The boundary contour lines refer to the obvious boundaries outside the target region, while concave and convex regions are areas with significant surface undulations or large local curvature.

[0056] After curvature fitting, the obtained surface feature information is used for further spatial partitioning. To more accurately model the target printing area, an octree partitioning algorithm is used to divide the entire target area into multiple refined spatial units. In the specific implementation, the minimum bounding rectangle of the target area is first used as the overall reference frame, and spatial partitioning is performed layer by layer. The partition size of each layer is dynamically adjusted according to the local curvature and geometric features of the target area. For example, smaller spatial units are used in areas with large curvature changes, while larger spatial units can be used in areas with small curvature changes or flat areas.

[0057] Each spatial unit is recursively subdivided to ultimately form a hierarchical three-dimensional spatial mesh model. Each unit of this mesh model consists of a cube or cuboid, which contains the geometric feature information of the region, including spatial coordinates, curvature values, and obstacle locations.

[0058] The three-dimensional spatial mesh model is divided into regions to obtain the cooperative coverage parameter Cn of each sub-region in the multi-nozzle reachability matrix, wherein the cooperative coverage parameter represents the degree of overlap of each nozzle on the sub-region.

[0059] After constructing the 3D spatial mesh model, to achieve more efficient path planning and nozzle collaborative allocation, the entire printing area needs to be divided into several sub-regions with clearly defined characteristics. The specific method involves first extracting the geometric feature parameters of each spatial unit, including spatial coordinates, surface normal vectors, local curvature values, and unit volume. Then, a density-based clustering algorithm (such as the DBSCAN clustering algorithm) is used, employing the Euclidean distance and curvature change rate between spatial units as similarity metrics to perform spatial unit clustering. Curvature similarity is determined by setting a curvature difference threshold ΔK. When the curvature difference between two units is less than this threshold, and the spatial distance is less than a preset aggregation distance, they are considered adjacent units belonging to the same cluster. Ultimately, multiple sub-region units with relatively continuous structures and consistent geometric properties are formed, providing a regional basis for nozzle coverage modeling.

[0060] To determine the accessibility of each printhead to a sub-region during printing, a printhead coverage area model needs to be constructed. First, based on the printing parameters of each printhead, the maximum spray width, spray depth, and trajectory range of the printhead in each direction are obtained. Then, combining the printhead's installation position on the printing platform and the allowable angular variation range of the motion mechanism, the spatial coverage boundary of the printhead is constructed. This boundary is defined as the maximum coverage area of ​​the spatial region by the printhead under different postures. A geometric modeling method is used to define the coverage area of ​​each printhead as a set of three-dimensional cones or sectors, and spatially unfolded with the printhead nozzle as the vertex.

[0061] Based on this model, a ray projection determination of the spatial centroid coordinates is performed for each sub-region unit. Specifically, a spatial ray is emitted from the nozzle position to the center of the sub-region spatial unit. If the ray falls inside the printhead coverage area model and the angle between the ray and the printing direction is less than a set angle threshold, then the sub-region is determined to be within the effective printing range of the printhead.

[0062] For sub-regions determined to be covered by two or more printheads simultaneously, it is necessary to further evaluate the actual printing coverage of each printhead in that region. First, obtain the projected area of ​​the sub-region in the printhead coverage area model, and compare it with the theoretical maximum coverage area of ​​the printhead to obtain the printhead printing coverage rate of the sub-region, denoted as Pn, where n is the printhead number.

[0063] Then, the set of nozzle numbers with effective coverage is summed. The cooperative coverage parameter Cn is defined as: the sum of the coverage rates of all nozzles divided by the total number of nozzles, and then multiplied by the regional coverage consistency coefficient. This coefficient is calculated based on the difference in the spray direction angle of each nozzle to the sub-region; the smaller the angle, the higher the consistency.

[0064] Finally, the obtained collaborative coverage parameter Cn is normalized so that its value range is limited to between 0 and 1. The closer the value is to 1, the more consistent the coverage of the sub-region is with multiple nozzles. The closer the value is to 0, the weaker the coverage or the more effective it is with only a single nozzle. This parameter will serve as a key reference indicator for subsequent nozzle task priority allocation.

[0065] Based on the collaborative coverage parameter Cn of each sub-region, a nozzle task priority map G is constructed to represent the task competition weight of each nozzle in different sub-regions.

[0066] After obtaining the collaborative coverage parameters of the sub-regions, it is necessary to evaluate the printing adaptability of each printhead in different sub-regions by combining the specific physical performance characteristics of each printhead. First, a printhead performance evaluation vector is constructed for each printhead, which includes three dimensions: printing speed, printing resolution, and jetting stability.

[0067] Among them, printing speed represents the maximum effective spray length of the printhead per unit time, expressed in millimeters per second; printing resolution represents the minimum controllable spray spacing of the printhead, expressed in dots per inch; spray stability is modeled using historical error data and failure rate, and is defined as a stability coefficient between 0 and 1, with a larger value indicating more stable spraying.

[0068] After the nozzle performance evaluation vector is constructed, it is matched item by item with the cooperative coverage parameters of each sub-region. When a nozzle has high cooperative coverage for a sub-region and its performance parameters perform well within the spatial requirements of that sub-region, it is considered to have high adaptability. This matching process is implemented through a vector weighted scoring function, providing basic data for subsequent weight allocation.

[0069] For each sub-region, extract the corresponding printhead performance evaluation vector from the set of related printheads. Calculate the printhead's adaptation score for the sub-region using the following comprehensive adaptation function: multiply the printing speed, printing resolution, and jetting stability by their respective weighting coefficients (weights can be set to 0.4, 0.3, and 0.3), sum the weighted values, and then multiply by the printhead's collaborative coverage parameter for the sub-region to obtain the printhead's comprehensive adaptation score for that sub-region. This score is the initial value of the printhead's task competition weight in that sub-region, reflecting its operational priority in that region.

[0070] For example, for a certain printhead, if its printing speed in a certain sub-region is 150 mm / s, its printing resolution is 600 dots per inch, its spray stability is 0.95, and its cooperative coverage parameter is 0.85, then the adaptation score of the printhead in that region can be calculated according to the above weighting method.

[0071] To ensure that task weights are relatively distributed and comparable within each sub-region, the initial values ​​of the task competition weights for all nozzles in a given sub-region are normalized. Specifically, the initial competition weights of all nozzles are first summed, and then the initial weight of each nozzle is divided by this sum to obtain the normalized weight value.

[0072] After normalization, the sum of the competition weights of all nozzles in the sub-region is 1. This forms a task competition weight matrix with sub-regions as the unit and nozzle number as the dimension. Each element represents the competition priority of a certain nozzle in a certain sub-region, laying the data foundation for the subsequent construction of the global task map.

[0073] Since the actual printing path has spatial continuity, the adjacency relationship between each sub-region needs to be taken into account. Based on the constructed sub-region competition weight matrix, spatial adjacency constraints are introduced, and the continuity of nozzle task distribution between adjacent regions is determined by establishing a sub-region adjacency graph.

[0074] An integration strategy based on weighted average is adopted to smooth the competing weights of the same printhead in adjacent sub-regions, thereby reducing path jumps caused by frequent printhead switching. After integration, the competing weights of printheads in all sub-regions are connected in spatial order to form a printhead task priority graph G. This graph is a multi-dimensional array structure, where each dimension represents the task weight distribution trend of a printhead in the entire printing space. It is used to guide the priority selection of printhead tasks during path planning and reduce conflicts and waiting time between printheads.

[0075] Based on the priority map G, the printhead task is dynamically matched using the minimum cross-entropy objective function to obtain the initial printing path allocation scheme.

[0076] The nozzle task priority map provides the task competition weights for each nozzle across multiple sub-regions, reflecting the nozzle's priority for tasks within the current spatial layout. To achieve nozzle task allocation based on optimization objectives, this weight matrix needs to be normalized so that the sum of all weight values ​​is 1, thus forming the target probability distribution. This target distribution is denoted as the true label distribution and is used to supervise the learning of subsequent matching models.

[0077] Simultaneously, an output probability distribution based on the nozzle task prediction model is constructed as the prediction result of the matching probability between the nozzle and each sub-region. This prediction probability distribution is generated by a trainable neural network model, with the model input being a combination of the nozzle performance vector and the sub-region feature vector. The matching probability between each nozzle and each sub-region is calculated through forward propagation.

[0078] To clarify the task allocation status between nozzles and sub-regions, a matching relationship matrix is ​​constructed. Each row in this matrix represents a nozzle, each column represents a sub-region, and the matrix elements represent the matching probability between the corresponding nozzle and the sub-region.

[0079] Supervised optimization employs the cross-entropy loss function as the objective function. The cross-entropy loss function measures the information difference between two probability distributions; a smaller value indicates that the model's predicted distribution is closer to the target distribution. The model parameters are iteratively updated using the gradient descent algorithm, progressively optimizing the matching weights between the nozzle and the sub-region, thereby minimizing the cross-entropy loss.

[0080] To ensure the rationality and physical feasibility of task allocation, the following two types of matching constraints are introduced in each round of parameter updates:

[0081] First, there is a constraint on the remaining operating capacity of the nozzles. The maximum task load capacity of each nozzle is set as a constant. When the cumulative task weight of a nozzle exceeds this capacity, the probability of subsequent matching will be dynamically suppressed to near zero to prevent task allocation overload.

[0082] Second, spatial continuity constraints. Based on the geometric adjacency relationships between sub-regions, priority is given to preserving the matching continuity of nozzles between adjacent sub-regions to avoid abrupt changes in the movement path caused by task jumps. Continuity constraints are implemented by adding a penalty term for matching differences between spatially adjacent regions to the loss function, which helps control path coherence.

[0083] When the decrease in cross-entropy loss is less than a set threshold (e.g., 0.0001) or the number of iterations reaches the upper limit (e.g., 1000 rounds), the training is considered to have converged, and the optimization process is terminated. At this point, the maximum matching probability of the sub-region corresponding to each nozzle in the output matching matrix is ​​taken as the final allocation result.

[0084] Based on this result, an initial printing path allocation scheme is generated, which involves assigning tasks to corresponding printheads and, according to the topological order of the sub-regions in the spatial grid, initially constructing the printing path trajectory for each printhead. This path scheme will serve as the basic input data for subsequent path conflict detection and optimization reconstruction.

[0085] Based on the initial printing path allocation scheme, a printing path conflict map Cg is constructed, and the path intersection conflict feature points F are extracted from it.

[0086] After obtaining the initial print path allocation scheme, the path trajectories of multiple printheads need to be discretized to achieve collision detection. Specifically, according to a set time step Δt (e.g., 0.01 seconds), the continuous trajectory of each printhead in the printing task is sampled at time intervals, and the three-dimensional coordinate position of the printhead and the corresponding timestamp at each time point are recorded.

[0087] The sampled data constitutes a path sequence dataset. Each path sequence contains several path points, and each path point is represented by a quadruple, including the X, Y, and Z coordinates and the corresponding timestamp. This path sequence not only fully describes the nozzle's movement trajectory in space but also preserves temporal information, providing a basis for determining spatiotemporal overlap in collision detection.

[0088] After obtaining the path sequences of all nozzles, pairwise comparisons are performed. For any two nozzle paths, path points are sampled one by one and their spatial distance is calculated. If, within the same time window, the Euclidean distance between the path points of the two nozzles is less than the set spatial conflict threshold (e.g., 2 mm) and the time difference is less than the set synchronization threshold (e.g., 0.05 seconds), then the path pair is considered to have a potential conflict.

[0089] The spatial conflict threshold is defined as the minimum safe operating distance between two nozzles, and the time synchronization threshold is defined as the maximum time difference tolerance value for determining whether path intersection will cause actual interference.

[0090] Potentially conflicting path pairs are recorded as conflict edges, and nodes are the nozzle numbers or sub-path identifiers to which the path belongs, constructing an undirected printing path conflict graph. Each edge in the conflict graph represents the risk of spatiotemporal interference between two paths, used for subsequent conflict intensity analysis.

[0091] In the completed printing path conflict map, statistical analysis is performed on all conflict edges to calculate the frequency of conflict between each pair of paths, that is, the number of times a conflict occurs per unit time, and the spatial distribution density of conflict points is recorded.

[0092] A two-dimensional conflict intensity evaluation model is constructed, using conflict frequency as a horizontal indicator and spatial concentration as a vertical indicator. Spatial concentration is defined as the number of conflict points per unit volume. If multiple conflict points appear in a concentrated area, it indicates that the area is a high-risk intersection zone.

[0093] Based on the conflict intensity evaluation results, representative conflict points in path intersection areas with conflict density exceeding a preset threshold (e.g., more than 10 conflict points per cubic centimeter) are extracted as path intersection conflict feature points. This set of feature points is denoted as F and is used as key optimization nodes in subsequent path reconstruction to achieve precise adjustment and avoidance of path conflicts.

[0094] Based on the conflict feature points F, a graph convolutional neural network is used to locally reconstruct the path conflict and generate an optimized printing path scheme.

[0095] To achieve accurate modeling of path conflict areas, the extracted conflict feature points need to be transformed into a graph structure. First, all path intersection conflict feature points are selected, and each feature point is treated as a node in the graph. Next, spatial adjacency is determined based on the Euclidean distance of these feature points in three-dimensional space. If the spatial distance between two feature points is less than a set adjacency threshold (e.g., 2 mm), an undirected edge is established between the nodes, indicating that they are spatially adjacent.

[0096] Furthermore, based on the actual movement trajectory of the nozzles in the original path sequence, temporally continuous edge connections are established between adjacent discrete path points in the path, so that the graph structure includes both spatial adjacency relationships and temporally continuous relationships. The final conflict subgraph structure is defined by the set of nodes and the set of edges, and is used as input to the graph convolutional neural network for path conflict modeling and optimization learning.

[0097] To enable graph convolutional neural networks to effectively identify path conflict features, a descriptive feature vector needs to be constructed for each graph node. Each node's feature vector contains the following dimensions: three-dimensional spatial coordinates (X, Y, Z), the corresponding timestamp T, and the nozzle number N to which the path point belongs. The nozzle number is represented using one-hot encoding to ensure that the nozzle classification information is distinguishable in the feature space.

[0098] To prevent differences in the dimensions of different features from affecting the network's learning performance, all input features are normalized. The normalization method used is min-max normalization, mapping all values ​​to the range of 0 to 1. The processed feature vectors are organized in node order to form the input feature matrix, with a dimension equal to the number of nodes multiplied by the feature dimension, which serves as the input data for the graph convolutional neural network.

[0099] A graph convolutional neural network is used to extract features from the constructed conflict subgraph. This neural network contains multiple graph convolutional layers, with each layer propagating features through node adjacency relationships. During graph convolution, each node is weighted and aggregated based on the feature vectors of its neighbors, and then subjected to a non-linear transformation using an activation function (e.g., ReLU), thereby achieving joint modeling of spatial and temporal information.

[0100] The network is trained to learn the direction and displacement that each conflict point should be adjusted along the path. The final output is a path adjustment vector for each node, containing the proposed 3D coordinate offset, representing the expected position of that path point after conflict optimization. This output is used to locally replace the original path to alleviate path intersection conflicts.

[0101] After obtaining the path adjustment vectors of all conflict nodes, these vectors are mapped back to the corresponding path point positions in the original path sequence, and a path reconstruction operation is performed. Specifically, the non-conflicting path segments in the original path are retained, and only path segments containing conflict feature points are replaced. The adjusted coordinates output by the graph convolutional neural network are added to the original coordinates to form new path points, constituting alternative path segments in the conflict region.

[0102] To ensure path continuity, a smooth transition segment is introduced between the reconstructed path segments. B-spline interpolation is used to generate transition path points, ensuring the optimized path is both executable and has stable trajectory. Finally, the path sequences of all printheads are updated to the optimized version, forming an overall optimized printing path scheme for subsequent collaborative control and printing execution.

[0103] Based on the optimized path scheme P, each printhead is driven to perform collaborative printing operations according to the optimized path.

[0104] In the generated optimized printing path scheme, each printhead path is independent, with differences in spatial distribution and length. To ensure the synchronization of collaborative printing operations, the path sequences of all printheads need to be uniformly processed along a time axis. The specific method is as follows: First, extract the key control nodes in each printhead path, including the printing start point, path corner points, and end points, and record their corresponding path segment lengths. Based on the set printing rate of each printhead, calculate the theoretical motion time of each path segment, and construct a unified execution time axis based on the maximum path time.

[0105] Linear interpolation is used to divide the time axis into equal intervals, enabling all printheads to perform motion control on the same time axis. For printhead trajectories with shorter paths, interpolation pause or slowing strategies are used to extend the time, ensuring that all printheads simultaneously complete printing start-up, turning, and stopping at key nodes, thus possessing global collaborative execution capabilities.

[0106] Based on the time synchronization, a printing control instruction set is further generated. First, the spatial displacement distance of adjacent path points in each printhead path sequence is calculated, and the movement time of each path segment is calculated according to the target printing rate (in millimeters per second). At the same time, the trigger frequency and spray duration are determined based on the spatial distribution density of the path points and the target resolution.

[0107] The printhead control instruction set includes the following three types of information: first, printhead movement trajectory instructions, which record the printhead target position coordinates at each time step; second, spray trigger instructions, indicating whether to execute the spraying action at that time step; and third, spray duration parameters, specifying the duration of each spray in microseconds. This instruction set provides the printhead control unit with complete execution parameters, ensuring that the path is consistent with the printing task.

[0108] In actual printing, there is a certain response delay from the printhead receiving the control command to completing the printing process. Additionally, the printhead is affected by mechanical inertia when running at high speeds, leading to path tracking errors. To improve printing accuracy, a predictive delay compensation model is used for error correction.

[0109] The specific method is as follows: Based on the measured data of the nozzle type, a response time model is established. This model takes the nozzle acceleration, load mass, and control frequency as inputs to predict the delay time of the nozzle under different operating states. Combined with the velocity change rate of the path points, the possible offset position is estimated, and the injection trigger time in the control command is sent to the nozzle one correction cycle in advance (e.g., 0.01 seconds in advance).

[0110] In addition, for path corners or high acceleration sections, feedforward control logic is introduced. By pre-loading compensation position deviation values, the printhead can accurately complete the spraying operation when passing through critical path points, significantly improving the printing edge accuracy and overlap accuracy.

[0111] After the path synchronization, control command generation, and deviation compensation are completed, all nozzle control command sets are scheduled and executed according to a unified timeline. Specifically, at each time step, the scheduler sequentially reads the moving target position, spray status, and spray parameters of each nozzle and sends them to the nozzle control unit.

[0112] After receiving a control command, the nozzle is driven to perform precise positional movements and starts or shuts down according to the set spray duration. During execution, the scheduler continuously monitors the operating status of each nozzle and records execution feedback data, including position error, spray delay, and other information.

[0113] By synchronizing the entire process with time, we ensure that each printhead can achieve collaborative printing without conflict or waiting within the same printing area, thus achieving the printing quality and efficiency goals set after overall path optimization.

[0114] Example 2: To verify the effectiveness of the multi-printer collaborative printing path planning method described in this application, the following simulation experiment was conducted.

[0115] In this embodiment, a 3D printing platform simulation environment is constructed. The printing area is a 300 mm × 300 mm planar task surface, containing multiple curvature variation regions and obstacle interference areas. The printing task includes a complex graphic path with approximately 4500 line segments. The experiment uses four independent printheads with a nozzle diameter of 100 micrometers and a printing rate set to 120 mm per second. The following steps are completed using the method described in this application:

[0116] Collect the structural features of the target printing area and construct a three-dimensional spatial mesh model;

[0117] Segment the region and generate collaborative coverage parameters;

[0118] Construct a priority graph and use the minimum cross-entropy objective function to achieve dynamic task matching;

[0119] Preliminary route planning;

[0120] Path conflict graph modeling and graph convolutional neural network optimization;

[0121] Path time synchronization and collaborative execution control.

[0122] To verify the beneficial effects, the method of this invention was compared with traditional methods based on fixed-order allocation and simple A* path planning. The evaluation indicators included:

[0123] Task completion time (in seconds); total path length (in millimeters); number of path conflicts (number of times when the spatial distance between path points is less than 2 millimeters within the same time window); total nozzle waiting time (in seconds).

[0124] Table 1 Experimental Results

[0125] Method type Task completion time Total path length Number of path conflicts nozzle waiting time This application method 58.3 seconds 11,420 mm 3 times 1.2 seconds Fixed order + A* method 71.6 seconds 12,840 mm 21 times 8.4 seconds

[0126] As shown in Table 1, the method in this application achieves more reasonable task allocation and significantly reduces printhead waiting time by using collaborative coverage modeling and task priority matching. By introducing a graph convolutional neural network to locally reconstruct path conflict points, the number of path conflicts is reduced by approximately 85.7%. The optimized path, while maintaining printing integrity, has an overall shorter length of approximately 11%, improving path execution efficiency. The total task completion time is reduced by approximately 18.6%, demonstrating a significant efficiency advantage in multi-printhead scenarios.

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-jet coordinated printing method, characterized by: The method comprises the following steps: Collecting structural feature information of a target printing area, and constructing a three-dimensional space grid model of the target printing area; Segmenting the three-dimensional space grid model to obtain a collaborative coverage parameter Cn of each sub-region in a multi-nozzle accessibility matrix, wherein the collaborative coverage parameter represents the overlapping degree of coverage of each nozzle on the sub-region; According to the collaborative coverage parameter Cn of each sub-region, a nozzle task priority graph G is constructed to represent the task competition weight of each nozzle in different sub-regions; According to the priority graph G, a minimum cross-entropy objective function is used for dynamic matching of nozzle tasks to obtain an initial jet printing path allocation scheme; Based on the initial jet printing path allocation scheme, a jet printing path conflict graph is constructed, and a path intersection conflict feature point F is extracted therefrom; Based on the conflict feature point F, a graph convolutional neural network is used to locally reconstruct the path conflict to generate an optimized jet printing path scheme; Based on the optimized path scheme P, each nozzle is driven to perform collaborative jet printing operation according to the optimized path.

2. A multi-jet coordinated printing method according to claim 1, characterized in that: The method for constructing a three-dimensional space grid model of a target printing area comprises the following steps: Collecting multi-angle surface data of the target printing area to obtain a point cloud data set containing depth information; Noise filtering and curvature fitting processing are performed on the point cloud data set to extract the boundary contour line, local concave-convex region and curvature distribution feature of the target printing area; Based on the processed surface feature information, an octree partitioning algorithm is used to divide the printing area into multiple refined space units to form a three-dimensional space grid model with a hierarchical structure.

3. A multi-jet coordinated printing method according to claim 1, characterized in that: Segmenting the three-dimensional space grid model to obtain a collaborative coverage parameter Cn of each sub-region in a multi-nozzle accessibility matrix comprises the following steps: The three-dimensional space grid model is clustered according to the geometric continuity and curvature similarity of each space unit to form multiple sub-region units with relatively uniform structure; Based on the jet printing motion boundary and pose change range of each nozzle, a nozzle coverage area model is constructed, and ray projection judgment is performed on each sub-region unit to determine whether it is within the effective jet printing range of one or more nozzles; For the sub-regions within the coverage range of multiple nozzles, the jet printing coverage rate of each nozzle is calculated, and the linear superposition method is used to obtain the collaborative coverage parameter Cn of the sub-region.

4. The method of claim 1, wherein: The method for constructing a nozzle task priority graph G according to the collaborative coverage parameter Cn of each sub-region comprises the following steps: The collaborative coverage parameter of each sub-region is matched with the jet printing capability parameter of the corresponding nozzle, and a nozzle performance evaluation vector is constructed according to the jet printing speed, jet printing resolution and jet stability of the nozzle to describe the job adaptability of the nozzle in different sub-regions; Based on the nozzle performance evaluation vector, weight allocation processing is performed on the sub-region with high collaborative coverage parameter, and the comprehensive adaptation score of the nozzle in the sub-region is calculated to obtain the task competition weight initial value of the nozzle in the sub-region; The task competition weight initial values of each nozzle in the same sub-region are normalized so that the competition weight sum of each nozzle in the sub-region is 1, thereby forming a nozzle task competition weight matrix at the sub-region level; The nozzle task competition weight matrix of all sub-region levels is integrated according to the spatial adjacency relationship to construct a nozzle task priority map for representing the task priority distribution of each nozzle in the global printing area.

5. The method of claim 1, wherein: The initial printing path allocation scheme is obtained by adopting a minimum cross-entropy objective function to perform dynamic matching of the nozzle task according to the priority map G, including: The task competition weight of each nozzle corresponding sub-region in the nozzle task priority map is taken as a target probability distribution, and a prediction probability distribution of the nozzle task matching is constructed as a model output; A matching relationship matrix between the nozzles and the sub-regions is set, and an iterative optimization of the matching weight is performed by taking the minimization of the cross-entropy loss function as an optimization objective, so as to narrow the information difference between the prediction distribution and the target distribution; In each iteration, a matching constraint condition is introduced in combination with the residual job capacity of the nozzles and the spatial continuity between the sub-regions, so as to limit the occurrence of unreasonable task allocation behaviors; After the cross-entropy loss converges to a set threshold or the maximum number of iterations is reached, the final matching result of the nozzles and the sub-regions is output, and the initial printing path allocation scheme is generated according to the matching relationship.

6. A multi-jet coordinated printing method according to claim 1, wherein: Based on the initial printing path allocation scheme, a printing path conflict graph is constructed, and a path intersection conflict feature point F is extracted therefrom, including: The path trajectory of each nozzle in the initial printing path allocation scheme is discretized according to the time sequence to generate a path sequence dataset containing path point coordinates and time stamps; All nozzle path sequences are compared with each other, and it is determined whether there is a potential conflict between the paths according to the spatial position overlap degree and the time window coincidence, and a printing path conflict graph is constructed with the path pairs as edges; In the conflict graph, a path intersection area with high conflict intensity is identified, and a path intersection conflict feature point is extracted as a key node for local conflict optimization based on conflict frequency and spatial concentration indexes.

7. The method of claim 1, wherein: Based on the conflict feature point F, a graph convolutional neural network is adopted to locally reconstruct the path conflict to generate an optimized printing path scheme, including: The conflict feature points are constructed into a node set according to the spatial adjacency relationship, and an edge set is established according to the connection relationship between adjacent path points in the path sequence, so as to form a conflict subgraph structure for learning; The spatial coordinates, time stamps and nozzle numbers of each node in the conflict subgraph are taken as input feature vectors, and all input features are normalized to form an input feature matrix of the graph convolutional neural network; The input feature matrix is subjected to feature extraction by the graph convolutional neural network containing multiple convolution kernels, the spatial correlation and time variation mode of the conflict area are learned through cross-node convolution propagation, and a path adjustment vector of the node is output; Based on the node path adjustment vector, the original path sequence is locally reconstructed, the path intersection area in the conflict subgraph is replaced by a replacement path segment optimized by the graph convolutional neural network, and thus an optimized printing path scheme is generated.

8. The method of claim 1, wherein: The optimized path scheme P is used to drive each nozzle to perform collaborative printing operation according to the optimized path, including: The path sequence of each nozzle in the optimized path scheme is subjected to time synchronization processing to construct a unified execution time axis, so as to ensure that all nozzles have collaborative starting and stopping capabilities at key nodes; According to the spatial displacement of each jet path point and the jet printing rate, a jet printing control instruction set is generated, including jet moving track instructions, jet trigger instructions and jet duration parameters; An execution deviation caused by the response time of the jet and the mechanical inertia during the jet printing process is corrected; According to the control instruction set, each jet is driven to cooperatively execute the jet printing task in a time axis scheduling sequence.

Citation Information

Cited By

  • Double-nozzle collaborative and intelligent scheduling printing method and system, storage medium and program product

    CN121900712A

  • Multi-jet synchronous printing system based on a distributed architecture

    CN122331848A