Site planning algorithm for grinding large structural component by mobile robot

Abstract

The invention discloses a site planning method for grinding a large structural component through a movable grinding platform. The method comprises the steps of 1, establishing an optimization model; step 2, establishing a coding data solving format; 3, solving by adopting an optimization algorithm, and solving the proposed optimization model by using an improved non-dominated sorting genetic algorithm RL-NSGA-DP integrating reinforcement learning and double-population coevolution; step 4, selecting a site sequence planning scheme from a solving result to regulate and control the mobile grinding platform; the labor and time cost is effectively reduced, and the efficiency and the automation degree of the production process are improved.

Description

Technical Field

[0001] This invention belongs to the field of mobile robot grinding, and specifically relates to the planning of grinding stations for mobile robots in the grinding process of large, complex composite curved surface structures such as wind turbine blades. Background Technology

[0002] Wind turbine blades are a crucial structural component of wind turbine units, and their surface integrity and profile accuracy have a vital impact on the energy conversion efficiency, service life, and operational reliability of the turbine. Grinding and reprocessing the blade surface is an important method for maintaining profile quality. Wind turbine blades are large-sized, complex curved surface components made of fiberglass, and the grinding process generates a large amount of fine fibers and harmful dust, seriously affecting the health of workers. Applying robotic blade grinding offers advantages such as reducing operating costs and dust hazards, shortening blade delivery cycles, and ensuring the quality of curved surface processing. However, wind turbine blades are enormous, with offshore blades reaching hundreds of meters in length, necessitating mobile robots to move to multiple sites for blade surface grinding. Currently, the processing sites for mobile robots are determined manually, but the highly nonlinear distribution of the robot's operability within the task space, along with the complexity of the processed surface, makes determining the appropriate robot processing location very time-consuming and cannot guarantee optimal performance.

[0003] Existing technologies plan individual processing stations after determining the grinding trajectory. However, when dealing with large structural components, it is impossible to dynamically determine the number of processing stations with the highest efficiency based on the scale of the trajectory planning. Furthermore, relying on operator experience to determine the location of processing stations makes it difficult to quickly plan reasonable station locations after changes in the type and model of the structural component. Additionally, the station determination cycle is lengthy when the fan blade placement changes.

[0004] To address this issue, a highly efficient mobile robot processing station planning algorithm was designed to quickly plan the locations of a series of processing stations based on the grinding trajectory of the wind turbine blade surface. This solution will significantly improve the efficiency of grinding operations and ensure the quality of the processed surface. Summary of the Invention

[0005] To address the aforementioned shortcomings or deficiencies, this application proposes a site planning algorithm for mobile robots grinding large structural components. In the application scenario of grinding wind turbine blades, the algorithm automatically plans the location and trajectory allocation of the robot's processing sites, ensuring the grinding quality of the robot while improving production efficiency.

[0006] A site planning method for polishing large structural components using a mobile polishing platform, the mobile polishing platform including a mobile robot and an industrial robot, with the industrial robot mounted on the mobile robot; the mobile robot moves to the corresponding workstation according to the planned path, and the onboard industrial robot performs the polishing operation along the corresponding trajectory. Step 1: Establish an optimization model; Step 2: Establish the solution encoding data format; Step 3: Solve the proposed optimization model using an optimization algorithm. The improved non-dominated sorting genetic algorithm RL-NSGA-DP, which integrates reinforcement learning and dual-population co-evolution, is used. Step 4: Select a site sequence planning scheme from the solution results to control the mobile grinding platform.

[0007] The beneficial effects of this invention are: This algorithm first establishes an optimization model for the mobile robot's processing station and trajectory allocation. This model includes two optimization objectives: total completion time and robot operability, while also considering collision avoidance. Next, a novel two-layer meta-encoding mechanism is proposed. Based on this, an improved non-dominated sorting genetic algorithm RL-NSGA-DP, integrating reinforcement learning and dual-population co-evolution, is proposed to solve the optimization model. It can quickly and effectively select reasonable robot processing station locations and trajectory allocation schemes from the non-dominated solution set obtained by the algorithm according to actual needs, providing guidance for grinding fan blades. This algorithm effectively reduces labor and time costs, and improves the efficiency and automation of the production process. Attached Figure Description

[0008] Figure 1 A schematic diagram of the grinding trajectory on the surface of the wind turbine blade generated by the tangent plane method; Figure 2 A schematic diagram of the robot operation process for a single site; Figure 3 This is a schematic diagram of a two-layer meta-coding scheme; Figure 4 Flowchart of the improved non-dominated sorting genetic algorithm; Figure 5 This is a schematic diagram illustrating the search scope for the initial candidate robot sites; Figure 6 Here is a flowchart of the two-population collaborative search process; Figure 7 The workflow for the Q-learning-based adaptive environment selection mechanism; Figure 8 A schematic diagram of the site planning scheme with the shortest completion time; Figure 9 This is a schematic diagram of the site planning scheme with optimal robot operability. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only some, not all, of the embodiments of this disclosure, and are used merely to explain this disclosure and are not intended to limit it. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.

[0010] In the description of this disclosure, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," "outer," "both ends," "both sides," "top," and "bottom," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this disclosure and simplifying the description, and do not indicate or imply that the elements referred to must have a specific orientation or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this disclosure. Furthermore, the terms "first," "second," "upper-level," "lower-level," "primary," and "secondary," etc., are used for descriptive purposes only and can be simply used to more clearly distinguish different components, and should not be construed as indicating or implying relative importance.

[0011] In the description of this disclosure, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linkage" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, an integral connection, a mechanical connection, an electrical connection, a direct connection, or an indirect connection through an intermediate medium, or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this disclosure based on the specific circumstances.

[0012] like Figure 1 As shown, the grinding trajectory of the wind turbine blade surface is generated using the tangent plane method during the grinding process of large and complex composite curved surface structures such as wind turbine blades.

[0013] The mobile polishing platform comprises a mobile robot and an industrial robot, with the industrial robot mounted on the mobile robot. The mobile robot moves according to a planned path to its designated workstation, where the onboard industrial robot executes the corresponding trajectory. Clearly, the selection of the mobile robot's workstation locations determines the efficiency and quality of the processing task, necessitating the development of an effective optimization method to find the most suitable workstation locations and rationally allocate the polishing trajectory to these workstations.

[0014] This application proposes a planning algorithm for site planning of wind turbine blade surface grinding by assembling a grinding head using an AGV equipped with a KUKA KR180R3500 robot, which includes the following steps.

[0015] Step 1: Establish an optimization model.

[0016] An optimization model for the optimal robot station and trajectory allocation for mobile robot grinding of wind turbine blades was established for specific field applications. The optimization objective of this model is to simultaneously improve the efficiency and quality of the grinding operation, thus defining two objectives. Total polishing completion time and robot operability.

[0017] A well-designed robot station can significantly shorten the completion time of grinding and finishing processes by performing all surface trajectories with fewer stations. Objective function To minimize the processing time of the mobile robot, the operation flow of a single station robot is as follows: Figure 2 As shown, the completion time of the mobile robot at each processing station includes the travel time from the previous station to the current station, the grinding time, the time for the robot's end effector to jump between trajectories, and the transition time from the initial pose to the grinding start point and from the grinding end point back to the initial pose. The total completion time of the mobile robot is the sum of the travel, grinding, jumping, and transition times at all stations. Define the following formula (1): (1); in, Number of sites planned for the solution Indicates the first The completion time at each station is calculated from the time of movement to that station. Polishing time at this site Time between tracks and robot pose transition time composition.

[0018] As shown in formula (2), the objective function The aim is to maximize the robot's maneuverability by minimizing the deviation between the joint configuration and the intermediate value of joint motion. Since the range of motion of joints 4 and 6 of the industrial robot is large enough not to cause too much operational disadvantage, the angular deviations of joints 1, 2, 3, and 5 are chosen to represent the robot's maneuverability. As shown in formula (3), the relatively high exponent significantly increases the index when the joint approaches its physical limits. This index is obtained by inverse kinematics. In site planning, solutions with smaller deviations from the joint center value exhibit better motion maneuverability, thereby ensuring the processing quality of the robot.

[0019] (2); (3); in, Represents the robot's operability. This represents the number of trajectory points assigned to the k-th station. Indicates the first The deviation index of the joints at each station is determined by the robot at the trajectory points. The first Joint angle values Calculations show that and These represent the robot joints respectively. The median and maximum deviation angles,

[0020] Step 2: Establish the solution encoding data format.

[0021] The model established in Step 1 simultaneously optimizes the site and trajectory allocation for the mobile robot. For this model, a corresponding solution encoding data format is required.

[0022] A two-layer meta-coding data scheme is established to simultaneously characterize the solutions for site planning and trajectory allocation. The encoded information includes meta-variables, site vectors, and trajectory allocation components. Meta-variables serve as the basic building blocks of the solution; site vectors represent the location of each specific site; and the trajectory allocation component indicates the number of trajectory blocks assigned to that site and the specific start and end grinding point numbers within each trajectory block. The mobile robot moves sequentially to the corresponding sites according to the planned site sequence to complete the grinding task assigned to that site. Figure 3 As shown, the coded data can represent a solution containing a total of 18 trajectories and 9 candidate station locations, with the final station sequence being [1,4,5,7,9]. This coded data scheme effectively integrates processing station and trajectory allocation information into a single code, providing a computational basis for the optimization algorithm to solve the model.

[0023] Step 3: Solve the problem using an optimization algorithm.

[0024] The proposed optimization model is solved using an improved non-dominated sorting genetic algorithm RL-NSGA-DP, which integrates reinforcement learning and dual-population co-evolution. The algorithm flow is as follows: Figure 4 As shown, the process includes three main steps: initializing the population, dual-population collaborative search, and adaptive environment selection based on reinforcement learning.

[0025] The algorithm first determines the population size based on the set parameters. Initialize population The offspring population is then fed into a dual-population collaborative search mechanism, where different crossover and mutation operators are applied to generate offspring. The offspring and parent populations are then merged and fed into a reinforcement learning-based adaptive environment selection mechanism to select dominant individuals, which finally form the parent population. Continue inputting to the next iteration. Until the current iteration number. Reaching the maximum number of iterations Output the current population As the solution output by the RL-NSGA-DP algorithm.

[0026] First, during population initialization, while ensuring that all assigned blade surface grinding trajectories are reachable, the robot's singular positions and collision points with the workpiece need to be avoided. Therefore, the distribution of feasible robot processing station locations in Cartesian space is highly irregular and discontinuous. Furthermore, as the number of assigned trajectories increases, the difficulty of searching for feasible robot processing station locations continuously increases. Therefore, during initialization, each initial candidate robot processing station needs to be pre-assigned two consecutive trajectories, which are then randomly generated within the search range until both assigned trajectories are reachable.

[0027] like Figure 5 As shown, the search range depends on the robot's reachability and the boundary size parameters of the assigned trajectory. Assuming the robot's maximum reachable distance is... The boundary dimensions of the distribution trajectory covering the workpiece surface are So, a single site The search range is defined as The calculation formula (4) is as follows: (4); in To ensure a safe distance, the distance is set to half the width of the mobile robot, thus preventing collisions between the robot and the workpiece. Based on the above process, the positions of all robot stations and their corresponding assigned trajectories in the initial solution are determined within this safe area.

[0028] Second, a dual-population collaborative search strategy is used to promote the evolutionary search of the population. The specific process is as follows: Figure 6 As shown. First, the top 1 / 3 of solutions in terms of completion time and operability are selected, and their union forms the dominant population, while the remaining solutions form the subordinate population. The core idea of ​​the dual-population cooperative search is to use explicit and implicit crossover operators to accelerate the convergence speed of the dominant population, while each individual in the subordinate population undergoes simulated binary crossover with individuals in the dominant population to generate offspring, thereby maintaining population diversity and promoting global search capabilities.

[0029] Third, this improved algorithm also employs an adaptive environment selection mechanism based on reinforcement learning, the workflow of which is as follows: Figure 7 As shown, the proportion of previous solutions dominated by the current solution. As a standard for determining the evolutionary state of the population, Q-learning is used to dynamically switch between two selection strategies, thereby balancing convergence and diversity at different evolutionary stages. In Q-learning, the agent's actions are defined as two different selection mechanisms. Action 1 employs the NSGA-II environment selection mechanism, prioritizing solutions with better convergence through non-dominated sorting, while preserving some solutions in the critical layer to maintain diversity. Conversely, Action 2 focuses more on enhancing population diversity, iteratively removing the most similar suboptimal solutions based on a deletion criterion designed to consider both convergence and diversity, until the population size is reached. Deletion Criterion Design the following formula (5): (5); in This represents the most similar pair of solutions in the population. The solution is calculated using formula (6). The convergence of, and The solution is calculated using formula (7). Diversity of penalties.

[0030] (6); (7); Formula (6) uses the 2-norm value of the objective vector to evaluate the convergence of the solution, where This represents the normalized target vector. In formula (7) In addition to Distance in target space The most recent solution, Then it means except and distance The most recent solution. Therefore, the diversity penalty term. The molecule represents With corresponding solution The distance, and the denominator used for normalization represents the distance. The maximum distance to all other solutions. The smaller the value, the better the overall performance of the solution. The two solutions with the smallest distance have a high degree of similarity in population distribution, so deleting one of the solutions is reasonable and effective.

[0031] The reward function can evaluate the agent's action performance. In this algorithm, the reward function is designed based on the improvement rate of population fitness as shown in formula (8): (8); in, Represents the number of individuals in the population. and They represent individuals from the previous generation and the current generation of the population, respectively. .

[0032] This formula calculates the fitness improvement rate of the population from the previous generation to the current generation after the agent executes the corresponding environment selection strategy. It reflects the overall evolution of the population after taking the selected action. After multiple iterations, a Pareto front composed of non-dominated solutions can be obtained.

[0033] Step 4: Select a site sequence planning scheme for robot control.

[0034] Depending on whether the actual application scenario prioritizes completion time or robot operability, a site sequence planning scheme can be selected for actual deployment.

[0035] Figure 8 The site planning scheme for the mobile robot to complete the grinding process in the shortest time allows the mobile robot to simply move to site 1 and execute all grinding trajectories on the blade surface. Figure 9 This is the optimal site planning scheme for the operability of mobile robots, with a total of 6 processing stations that execute the corresponding grinding trajectories in sequence.

[0036] The present invention can dynamically and quickly generate site planning schemes based on actual on-site deployment, providing effective data guidance for on-site operation planning and accelerating product production and development.

[0037] This invention introduces a site planning algorithm for mobile robots grinding wind turbine blades. The feasibility of the algorithm was verified in a wind turbine blade grinding application scenario, improving actual production efficiency and automation. It is highly versatile and applicable to applications where mobile robots process large composite free-form surface structures, such as wind turbine blades, high-speed railway carriages, and aerospace components for grinding and painting. This solution offers high flexibility and can effectively address the uncertainty of the actual placement of wind turbine blades.

[0038] A general optimization model was established, simultaneously planning the locations of processing stations for mobile robots grinding wind turbine blades and allocating optimized trajectories. A novel two-layer meta-encoding mechanism was proposed, effectively integrating processing station and trajectory allocation information into a single code, providing a foundation for the optimization algorithm to solve the model. Based on the non-dominated sorting genetic algorithm, a dual-population co-evolution strategy and a reinforcement learning-based adaptive environment selection mechanism were introduced to improve the performance of the multi-objective optimization algorithm.

[0039] Appropriate extensions of this solution to other related or similar industries, such as grinding trajectory planning algorithms for large structural components and processing station planning algorithms for mobile robots grinding and spraying large structural components, can yield robot grinding trajectory and station planning solutions that combine efficiency, stability, and versatility.

[0040] Finally, it should be noted that the above description is merely an explanation of the present invention and is not intended to limit the invention. Although the present invention has been described in detail, those skilled in the art can still modify the technical solutions described above or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A site planning method for grinding large structural components using a mobile grinding platform, the mobile grinding platform comprising a mobile robot and an industrial robot, the industrial robot being mounted on the mobile robot; the mobile robot moves to the corresponding work station according to a planned path, and the mounted industrial robot performs the grinding operation along the corresponding trajectory, characterized in that: Step 1: Establish an optimization model; Step 2: Establish the solution encoding data format; Step 3: Solve the proposed optimization model using an optimization algorithm. The improved non-dominated sorting genetic algorithm RL-NSGA-DP, which integrates reinforcement learning and dual-population co-evolution, is used. Step 4: Select a site sequence planning scheme from the solution results to control the mobile grinding platform.

2. The site planning method for grinding large structural components using a mobile grinding platform according to claim 1, characterized in that, In step one, the optimization goal of the model is to simultaneously improve the efficiency and quality of the polishing operation. Two objectives are defined: total polishing completion time and robot operability.

3. The site planning method for grinding large structural components using a mobile grinding platform according to claim 2, characterized in that, objective function To minimize the completion time of the mobile robot's grinding process, the completion time of the mobile robot at each processing station includes the travel time from the previous station to the current station, the grinding time, the jump time of the robot's end effector between trajectories, and the transition time from the initial pose to the grinding start point and from the grinding end point back to the initial pose. The total completion time of the mobile robot is the sum of the travel, grinding, jump, and transition times at all stations. The definition is as follows: ， in, Number of sites planned for the solution Indicates the first The completion time at each station is calculated from the time of movement to that station. Polishing time at this site Time between tracks and robot pose transition time composition.

4. The site planning method for grinding large structural components using a mobile grinding platform according to claim 2, characterized in that, objective function The aim is to maximize robot maneuverability by minimizing the deviation between joint configuration and intermediate joint motion values. The angular deviation of the industrial robot joints is selected to reflect the robot's maneuverability. The objective function is... as follows, ， ， in, Represents the robot's operability. This represents the number of trajectory points assigned to the k-th station. Indicates the first The deviation index of the joints at each station is determined by the robot at the trajectory points. The first Joint angle values Calculations show that and These represent the robot joints respectively. The median and maximum deviation angles, 。 5. The site planning method for grinding large structural components using a mobile grinding platform according to claim 1, characterized in that, In step two, a two-layer meta-coding data scheme is established to simultaneously characterize the solutions for site planning and trajectory allocation, wherein the encoding information includes meta-variables, site vectors, and trajectory allocation components.

6. The site planning method for grinding large structural components using a mobile grinding platform according to claim 5, characterized in that, The meta-variables serve as the basic building blocks of the solution. The station vector represents the location of each specific station, and the trajectory allocation part represents the number of trajectory blocks allocated to that station and the specific start and end grinding point numbers within each trajectory block.

7. The site planning method for grinding large structural components using a mobile grinding platform according to claim 1, characterized in that, Step 3 includes step 1) initializing the population, step 2) dual-population collaborative search, and step 3) adaptive environment selection based on reinforcement learning.

8. The site planning method for grinding large structural components using a mobile grinding platform according to claim 7, characterized in that, Step 1) When initializing the population, each initial candidate robot processing station needs to be pre-assigned two consecutive trajectories and randomly generated within the search range until both assigned trajectories are reachable; The search range depends on the robot's accessibility and the boundary size parameters of the assigned trajectory.

9. The site planning method for grinding large structural components using a mobile grinding platform according to claim 8, characterized in that, Step 2) Use a dual-population collaborative search strategy to promote the evolutionary search of the population. First, select the top 1 / 3 of the solutions in terms of completion time and operability, and take their union to form the dominant population. The remaining solutions are the suboptimal population. Use explicit and implicit crossover operators to accelerate the convergence speed of the dominant population. At the same time, simulated binary crossover is performed between each individual in the suboptimal population and individuals in the dominant population to generate offspring, so as to maintain population diversity and promote global search capability.

10. The site planning method for grinding large structural components using a mobile grinding platform according to claim 9, characterized in that, Step 3) Employs an adaptive environment selection mechanism based on reinforcement learning to determine the proportion of previous solutions dominated by the current solution. As a standard to determine the evolutionary state of the population, Q-learning is used to dynamically switch between two selection strategies, thereby balancing convergence and diversity at different evolutionary stages.

Images