A method and system for trajectory planning of an automated glue-applying robot for irregular curved surfaces

Abstract

This invention belongs to the field of industrial robots, specifically relating to a method and system for trajectory planning of an automatic glue-applying robot on irregular curved surfaces. The method includes: scanning the workpiece to be processed using a 3D laser scanner to obtain a point cloud 3D network model; interpolating the point cloud 3D network model to obtain a smooth 3D network model; using the smooth 3D network model, employing a deep learning-optimized A* algorithm to plan the shortest path trajectory to obtain an initial optimal path trajectory; optimizing the initial optimal path trajectory to obtain a final optimal path trajectory; and having the control center guide the robot to perform the glue-applying operation based on the final optimal path trajectory. This invention enables precise and efficient glue application, significantly improving the robot's glue-applying efficiency.

Description

Technical Field

[0001] This invention belongs to the field of industrial robots, specifically relating to a trajectory planning method and system for an automatic glue-applying robot on irregular curved surfaces. Background Technology

[0002] In the manufacturing process of many products, adhesive application is a common and crucial step, and its quality directly affects the overall quality of the final product. With increasing product requirements and the growing variety and complexity of products, the precision and consistency of adhesive application have become increasingly important. Traditional adhesive application methods, relying on manual experience or simple trajectory planning, are insufficient to meet the adhesive application needs of increasingly complex irregular curved surfaces.

[0003] With the development of automation technology, robotic adhesive application technology has gradually gained attention. However, existing robotic adhesive application technologies mainly rely on fixed trajectories and simple surface processing, which cannot efficiently handle complex irregular surfaces. Traditional methods are prone to uneven adhesive application and insufficient precision when dealing with such complex surfaces, leading to poor product quality. In addition, the current technology has a low degree of automation, making it difficult to achieve real-time adjustments and optimizations, which affects production efficiency and product consistency.

[0004] This invention aims to solve the aforementioned problems by proposing an innovative trajectory planning method. By introducing deep learning and employing heuristic functions to dynamically adjust based on complex surface features, the efficiency and accuracy of path planning are improved. This method can efficiently and accurately handle complex irregular surfaces, achieving high-quality and consistent adhesive application, thereby enhancing production efficiency and product quality. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a trajectory planning method and system for an automated adhesive application robot on irregular curved surfaces. The method includes:

[0006] S1: Use a 3D laser scanner to scan the workpiece to be processed to obtain a 3D point cloud network model;

[0007] S2: Interpolate the point cloud 3D network model to obtain a smooth 3D network model;

[0008] S3: Based on the smooth surface of the 3D network model, the shortest path trajectory is planned using the A* algorithm optimized by deep learning to obtain the initial optimal path trajectory;

[0009] S4: Optimize the initial optimal path trajectory to obtain the final optimal path trajectory;

[0010] S5: The control center guides the robot to perform the glue application operation based on the final optimal path trajectory.

[0011] Preferably, the process of obtaining the point cloud 3D network model includes: using a 3D laser scanner to scan the surface to be processed from multiple perspectives to obtain key coordinate points on the workpiece surface and form point cloud data from multiple perspectives; preprocessing all point cloud data to remove noise and redundant points; aligning and registering the preprocessed point cloud data from multiple perspectives to generate a complete point cloud 3D mesh model.

[0012] Preferably, the interpolation process for the point cloud 3D network model includes: interpolating the 3D model using an adaptive cubic B-spline curve to generate a smooth motion trajectory curve, thereby obtaining a smooth 3D network model; the interpolation formula for the adaptive cubic B-spline curve is:

[0013]

[0014] Where, parameter u represents the average weight of the control points; Q(u) represents the control point position vector matrix under parameter u; Q j M represents the position vector of the j-th control point; j,h (u) represents the h-th order B-spline basis function corresponding to the j-th control point, and m represents the number of control points;

[0015]

[0016] Among them, u j u represents the weight parameter of the j-th control point; j+h-1 u represents the weight parameter of the (j+h-1)th control point. j+h M represents the weight parameter of the (j+h)th control point. j,h-1 (u) represents the h-1 degree B-spline basis function corresponding to the j-th control point, M j+1,h-1 (u) represents the h-1th order B-spline basis function corresponding to the (j+1)th control point, where h represents the order of the B-spline curve.

[0017] Preferably, the process of shortest path trajectory planning using the A* algorithm optimized by deep learning includes:

[0018] S31: The trained DNN model is used to process the surface features of the current node, the target node, and the 3D network model to obtain the predictive heuristic estimate of the current node.

[0019] S32: Calculate the total cost estimate from the current node to the target node based on the predictive heuristic estimate of the current node;

[0020] S33: Reduce the dimensionality of the 3D network model to a 2D raster model;

[0021] S34: Based on the total cost estimate, the A* algorithm is used to process the two-dimensional grid model to obtain the initial optimal path trajectory.

[0022] Furthermore, the formula for calculating the total cost estimate of the target node is as follows:

[0023] H(v)=p(v)+h DNN (v)

[0024] Where H(v) represents the estimated total cost from the starting node through the current node v to the target node, p(v) represents the actual cost from the starting node to the current node v, and h DNN (v) represents the heuristic estimate of the current node v as output by the DNN model.

[0025] Furthermore, the process of processing the two-dimensional raster model using the A* algorithm includes:

[0026] S341: Add the initial nodes in the 2D raster model to the open set;

[0027] S342: Select the node v with the smallest total cost estimate H(v) in the open set and use it as the node to be processed; remove node v from the open set and add it to the closed set that has been processed;

[0028] S343: If the adjacent node is in the close set, ignore it; if the adjacent node is not in the open set, add it to the open set and set the current node v as the parent node of the adjacent node.

[0029] S344: If the neighboring node is already in the open set, check whether the path to the neighboring node through the current node v is better than the previously found path. If so, update the parent node of the neighboring node to v and recalculate the total cost estimate H(v) of the neighboring node.

[0030] S345: Repeat steps S342-S344. When the target node is added to the open set, it means that the optimal path has been found, or when the open set is empty, it means that there is no feasible path.

[0031] S346: Starting from the target node, backtrack along the parent node chain to the starting point to generate the initial optimal path trajectory.

[0032] Preferably, the process of optimizing the initial optimal path trajectory includes: using a particle swarm optimization algorithm to optimize the initial optimal path trajectory with the goal of minimizing the time, so as to obtain the final optimal path trajectory.

[0033] An automatic glue-applying robot trajectory planning system for irregular curved surfaces includes: a surface recognition module, a data processing module, a trajectory planning module, and a control execution module;

[0034] The irregular curved surface recognition module is used to acquire the three-dimensional point cloud data of the glued workpiece;

[0035] The data processing module is used to process 3D point cloud data and generate a 3D mesh model;

[0036] The trajectory planning module is used to generate an initial optimal adhesive application trajectory and optimize the initial optimal adhesive application trajectory to generate the final optimal path trajectory.

[0037] The control execution module is used to execute robot glue application control commands based on the final optimal path trajectory.

[0038] The beneficial effects of this invention are as follows: This invention uses a 3D laser scanner to accurately reconstruct the 3D model of irregular curved surfaces, efficiently acquiring 3D point cloud data of the workpiece to be processed; by introducing an adaptive cubic B-spline curve interpolation method, it can generate highly smooth motion trajectories on complex curved surfaces; the use of a deep learning-optimized A* algorithm combined with a particle swarm optimization algorithm further improves the smoothness and efficiency of the adhesive application trajectory; through a real-time feedback mechanism in the control center, it achieves precise control and automatic adjustment of the adhesive application task, improving the smoothness and efficiency of adhesive application, reducing uneven adhesive application, and ensuring the accuracy and efficient implementation of adhesive application. This invention not only improves production efficiency but also significantly enhances product consistency and quality, making it particularly suitable for industrial scenarios with high requirements for adhesive application precision, such as aerospace and automotive manufacturing. Attached Figure Description

[0039] Figure 1 This is a flowchart of the trajectory planning method for an automatic glue-applying robot on irregular curved surfaces in this invention;

[0040] Figure 2 This is a flowchart of the shortest path trajectory planning using the A* algorithm optimized by deep learning in this invention. Detailed Implementation

[0041] 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, and 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.

[0042] This invention proposes a trajectory planning method and system for an automated glue-applying robot on irregular curved surfaces. The method involves using a 3D laser scanner to scan the workpiece to acquire point cloud data of its irregular curved surface; processing the point cloud data to reconstruct a 3D mesh model of the workpiece; interpolating the 3D mesh model using an adaptive cubic B-spline curve to generate a smooth motion trajectory; employing a deep learning-optimized A* algorithm for shortest path trajectory planning; and further optimizing the trajectory generated by the A* algorithm based on a particle swarm optimization algorithm, using minimum time as the optimization metric, thereby improving glue-applying efficiency and path smoothness. The optimized trajectory is transmitted to the control center to generate glue-applying instructions, achieving efficient and precise automated glue-applying operation.

[0043] like Figure 1 As shown, the method specifically includes:

[0044] S1: Use a 3D laser scanner to scan the workpiece to be processed to obtain a 3D point cloud network model.

[0045] A 3D laser scanner mounted on the front end of the workpiece stage scans the workpiece surface from multiple perspectives to obtain key coordinate points, forming multi-view point cloud data. All point cloud data undergoes preprocessing, specifically: noise is removed using a machine learning-based noise filtering method; redundant points are removed using a redundant point detection algorithm to simplify the point cloud data; the processed multi-view point cloud data is then registered and surface reconstructed to generate a complete 3D point cloud mesh model. In some preferred embodiments, an adaptive algorithm based on workpiece features is used for noise filtering; data redundancy is reduced through feature point extraction and redundant point removal; and the alignment process is optimized based on multi-view registration technology and the Iterative Closest Point (ICP) algorithm.

[0046] S2: Interpolate the point cloud 3D network model to obtain a smooth 3D network model.

[0047] In some preferred embodiments, adaptive cubic B-spline curves are used to interpolate the three-dimensional model to generate smooth motion trajectory curves, thereby obtaining a smooth three-dimensional network model.

[0048] The interpolation formula for adaptive cubic B-spline curves is as follows:

[0049]

[0050] Where, parameter u represents the average weight of the control points; Q(u) represents the control point position vector matrix under parameter u; Q j M represents the position vector of the j-th control point; j,h (u) represents the h-th order B-spline basis function corresponding to the j-th control point, and m represents the number of control points;

[0051]

[0052] Among them, u j u represents the weight parameter of the j-th control point; j+h-1 u represents the weight parameter of the (j+h-1)th control point. j+h This represents the weight parameter of the (j+h)th control point; the weight parameter ranges from 0 to 1, where 0 represents the start of the curve and 1 represents the end of the curve; M j,h-1 (u) represents the h-1 degree B-spline basis function corresponding to the j-th control point, M j+1,h-1 (u) represents the (h-1)th order B-spline basis function corresponding to the (j+1)th control point, where h represents the order of the B-spline curve. h is used to adjust the step size of the point on the spline curve as it moves from one control point to the next. Here, h represents the step size from u... j to u j+h The step size is the increment in the parameterization process.

[0053] This interpolation method dynamically adjusts the interpolation density of the B-spline curve through an adaptive algorithm to adapt to complex irregular surface features, ensuring that the generated trajectory has high smoothness and continuity on the workpiece surface.

[0054] S3: Based on the smooth 3D network model, the shortest path trajectory is planned using the A* algorithm optimized by deep learning to obtain the initial optimal path trajectory.

[0055] like Figure 2 As shown, the process of shortest path trajectory planning using the A* algorithm optimized by deep learning includes:

[0056] S31: The trained DNN model is used to process the surface features of the current node, the target node, and the 3D network model to obtain the predictive heuristic estimate of the current node.

[0057] This invention introduces a deep learning-based heuristic function h. DNN (v) uses point cloud data and the corresponding optimal path as a dataset to train a deep neural network (DNN): the input features include the current node (x) v ,y v ,z v ), target node coordinates (x g ,y g ,z g The dataset contains the surface features C(v) (curvature, normal vector) and the actual cost D(v) from the start point to the end point. Each sample in the dataset can be represented as:

[0058] S={(x v ,y v ,zv ),(x g ,y g ,z g ),C(v),D(v)}

[0059] The input layer of the DNN model receives feature data from the nodes and the environment, including the three-dimensional coordinates (x, y, y) of the current node and the target node. v ,y v ,z v ) and (x g ,y g ,z g The surface feature is C(v). The hidden layers employ a multi-layer fully connected structure, with ReLU used as the activation function in each layer. The output layer is a single neuron used to predict the heuristic estimate h. DNN (v), the model architecture is represented as:

[0060] h DNN (v)=DNN((x v ,y v ,z v ),(x g ,y g ,z g ),C(v))

[0061] The regression task is used, and the training objective is to minimize the heuristic cost h of the prediction. DNN The mean squared error (MSE) between the true cost D(v) and the actual cost D(v). The loss function is expressed as:

[0062]

[0063] Where N is the number of training samples, D(v i The value represents the true path cost in the sample. Cross-validation is used to evaluate the model's generalization ability, ensuring that the DNN model can accurately predict the heuristic cost in different surface and path environments.

[0064] The surface features of the current node, the target node, and the 3D network model are input into a trained DNN model for processing to obtain a predictive heuristic estimate of the current node.

[0065] S32: Calculate the total cost estimate from the current node to the target node based on the predictive heuristic estimate of the current node.

[0066] The A* algorithm of this invention uses a trained DNN model as a heuristic function to plan the shortest path trajectory and calculate the total cost estimate from the current node v to the target node g. The trajectory evaluation formula is as follows:

[0067] H(v)=p(v)+h DNN(v)

[0068] Where H(v) represents the estimated total cost from the starting node through the current node v to the target node, p(v) represents the actual cost from the starting node to the current node v, and h DNN (v) represents the heuristic estimate of the current node v as output by the DNN model.

[0069] S33: Reduce the dimensionality of the 3D network model to a 2D raster model.

[0070] During path planning, the 3D model is reduced to a 2D grid model, and each grid point G(i,j,k) is set to represent a possible path point on the surface; the gridded point set G will serve as the search space for the A* algorithm.

[0071] S34: Based on the total cost estimate, the A* algorithm is used to process the two-dimensional grid model to obtain the initial optimal path trajectory.

[0072] The process of processing a two-dimensional raster model using the A* algorithm includes:

[0073] S341: Add the initial nodes in the 2D grid model to the open set, which is the set of nodes to be processed.

[0074] S342: Select the node v with the smallest total cost estimate H(v) in the open set and use it as the node to be processed; remove node v from the open set and add it to the closed set that has been processed.

[0075] The adjacent nodes of node v are processed as follows:

[0076] S343: If the neighboring node is in the close set, ignore it; if the neighboring node is not in the open set, add it to the open set and set the current node v as the parent node of the neighboring node.

[0077] S344: If the neighboring node is already in the open set, check whether the path to the neighboring node through the current node v is better than the previously found path. If so, update the parent node of the neighboring node to v and recalculate the total cost estimate H(v) of the neighboring node.

[0078] To check if the path from the current node v to the adjacent node is better than the previously found path, specifically: evaluate the value of H(v) while considering all possible successor nodes when expanding the node set. For each successor node v, calculate the H(v) value for reaching v from the starting point through the current node u. If the newly calculated H(v) value is less than the lowest known H(v) value (i.e., the value previously calculated or stored in the open set), then the current path is considered better. This indicates that the total cost of reaching v through the current node u plus the estimated cost from v to the target is lower than the previously known path.

[0079] The open list (open set) stores all nodes to be expanded, sorted by their H(v) values. Node selection is based on minimizing H(v) values ​​to ensure that the optimal path most likely to reach the goal is considered first.

[0080] S345: Repeat steps S342-S344. When the target node is added to the open set, it means that the optimal path has been found, or when the open set is empty, it means that there is no feasible path.

[0081] If no feasible path is found, modify the conditions (such as adjusting the positions of the start or end points, or using different path planning algorithm parameters) and search for the optimal path trajectory again.

[0082] S346: Starting from the target node, backtrack along the parent node chain to the starting point to generate the initial optimal path trajectory.

[0083] S4: Optimize the initial optimal path trajectory to obtain the final optimal path trajectory.

[0084] In the current path planning algorithm, the primary optimization objective is to minimize the overall task completion time. Therefore, the cost function G(z) is defined as the sum of the times required for all steps on the path; its expression is:

[0085]

[0086] Where p represents the number of optimization objectives, and here p = 1, representing time. The time z for each step... k It is estimated as the minimum time required to complete this step, while β k It is a time-related weighting factor used to adjust the proportion of time cost in the total cost function.

[0087] In some preferred embodiments, a particle swarm optimization algorithm is used to optimize the initial optimal path trajectory with the goal of minimizing the time, so as to obtain the final optimal path trajectory.

[0088] The objective function is expressed as:

[0089]

[0090] Among them, t k This represents the time required for the Kth step. This optimization goal enables the system to complete the adhesive application process as quickly as possible while ensuring quality.

[0091] The basic steps are as follows:

[0092] (1) Randomly generate the initial position and velocity of each particle in the particle swarm (each particle represents a combination of coating parameters, such as velocity and acceleration);

[0093] (2) Calculate the value of the objective function based on the position of the particle, which represents the trajectory optimization quality of the particle;

[0094] (3) Compare the current fitness value (objective function value) of each particle with its historical best solution and update the individual best position of the particle;

[0095] (4) Compare the individual optimal solutions of all particles and find the best solution among them as the current global optimal solution;

[0096] (5) Update the velocity and position of each particle according to the following formula:

[0097] w t+1 =w t +φ1η1(P best (t)-x t )+φ2η2(G best (t)-x t )

[0098] x t+1 =x t +w t+1

[0099] Among them, w t φ represents the particle's velocity at time t; φ1 and φ2 are learning factors, adjusting the step size of the particle as it flies towards its individual optimal position and the global optimal position, respectively; η1 and η2 are random numbers to maintain population diversity; P best (t) represents the particle's current optimal individual position; G best (t) represents the current global optimal position of the particle swarm; x t This indicates the position of the particle at time t.

[0100] Repeat the above steps until the predetermined number of iterations is reached or the convergence condition is met, to obtain the final optimal path trajectory.

[0101] S5: The control center guides the robot to perform the glue application operation based on the final optimal path trajectory.

[0102] The generated optimal path trajectory is transmitted to the control center. The control center guides the robot to perform the glue application operation based on the final optimal path trajectory. This includes: the control center receiving the glue application trajectory (optimal path trajectory) from the sensor and issuing control commands to control the robot to perform a glue application task on the workpiece according to the glue application trajectory at a specific glue application speed and acceleration. At the same time, the control center has a safety protection system: it can perform emergency stop and fault alarm.

[0103] This invention also proposes a trajectory planning system for an automatic glue-applying robot with irregular curved surfaces. This system can execute the above-mentioned trajectory planning method for an automatic glue-applying robot with irregular curved surfaces, and includes: a surface recognition module, a data processing module, a trajectory planning module, and a control execution module.

[0104] The irregular curved surface recognition module is used to acquire the three-dimensional point cloud data of the glued workpiece;

[0105] The data processing module is used to process 3D point cloud data and generate a 3D mesh model;

[0106] The trajectory planning module is used to generate an initial optimal adhesive application trajectory and optimize the initial optimal adhesive application trajectory to generate the final optimal path trajectory.

[0107] If no feasible path is found, the system provides feedback to the user or control system explaining why a path could not be found. This feedback can be a simple error message or more detailed diagnostic information explaining why a path could not be found (such as obstacles blocking the way, or the start or end point being unreachable). If the environment allows, the system modifies the conditions (such as adjusting the positions of the start or end points, or using different path planning algorithm parameters) and searches for the optimal path trajectory again.

[0108] The control execution module is used to execute robot glue application control commands based on the final optimal path trajectory.

[0109] In summary, this invention constructs a 3D model using point cloud data generated by a 3D scanning device, employs a combination of deep learning-optimized A* algorithm and particle swarm optimization algorithm for trajectory planning, and controls the robot's glue application task through a control center. This enables precise and efficient glue application, greatly improving the robot's glue application efficiency.

[0110] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An irregular curved surface automatic gluing robot trajectory planning method, characterized in that, Comprise: S1: using a three-dimensional laser scanner to scan the workpiece to be processed, and obtaining a point cloud three-dimensional network model; The process of obtaining the point cloud three-dimensional network model comprises: using a three-dimensional laser scanner to scan the surface of the workpiece to be processed from multiple angles, obtaining key coordinate points of the surface of the workpiece to be processed, and forming multi-angle point cloud data; preprocessing the multi-angle point cloud data to remove noise and redundant points; aligning and registering the preprocessed multi-angle point cloud data to generate a complete point cloud three-dimensional network model; S2: interpolating the point cloud three-dimensional network model to obtain a smooth curved three-dimensional network model; the process of interpolating the point cloud three-dimensional network model comprises: using an adaptive cubic B-spline curve to interpolate the point cloud three-dimensional network model to generate a smooth motion trajectory curve, and then obtaining a smooth curved point cloud three-dimensional network model; the interpolation formula of the adaptive cubic B-spline curve is: ； wherein the parameters represent the average weight of the control points; the parameters represent the control point position vector matrix under the parameters the position vector of the th control point; the position vector of the th control point corresponds to the th spline basis function, the number of control points; ； wherein, denotes a weight parameter of the control point; denotes a weight parameter of the control point, denotes a weight parameter of the control point, denotes a weight parameter of the control point, denotes the spline basis function, denotes the spline basis function, denotes the spline basis function, denotes the order of the spline curve; S3: according to the smooth curved three-dimensional network model, using a deep learning optimized A* algorithm to plan a shortest path trajectory, and obtaining an initial optimal path trajectory; the process of using the deep learning optimized A* algorithm to plan the shortest path trajectory comprises: S31: using a trained DNN model to process the current node, the target node and the surface features of the smooth curved three-dimensional network model, and obtaining a predicted heuristic estimate value of the current node; S32: calculating the total cost estimate from the current node to the target node according to the predicted heuristic estimate value of the current node; the formula for calculating the total cost estimate from the current node to the target node is: ； wherein, denotes the total cost estimate from the start node through the current node to the target node, denotes the actual cost from the start node to the current node , denotes the predicted heuristic estimate value of the current node output by the DNN model; S33: reducing the smooth curved three-dimensional network model to a two-dimensional grid model; S34: according to the total cost estimate, using an A* algorithm to process the two-dimensional grid model to obtain an initial optimal path trajectory; S4: optimizing the initial optimal path trajectory to obtain a final optimal path trajectory; S5: the control center guides the robot to perform the glue application operation according to the final optimal path trajectory.

2. The irregular curved surface automatic gluing robot trajectory planning method according to claim 1, characterized in that, The process of optimizing the initial optimal path trajectory comprises: using a particle swarm optimization algorithm to optimize the initial optimal path trajectory with minimum time as the optimization target to obtain the final optimal path trajectory.

3. A system for performing the method of any one of claims 1-2, wherein the system is a system for planning a trajectory of a robot for applying glue to an irregular curved surface. Comprise: An irregular curved surface recognition module, a data processing module, a trajectory planning module, and a control execution module; The irregular curved surface recognition module is used to obtain three-dimensional point cloud data of the workpiece to be processed; The data processing module is used to process the three-dimensional point cloud data and generate a point cloud three-dimensional grid model; The trajectory planning module is used to generate an initial optimal path trajectory, and optimize the initial optimal path trajectory to generate a final optimal path trajectory; The control execution module is used to execute a robot glue application control instruction according to the final optimal path trajectory.

Images