Wall surface crack repairing method based on mechanical arm control optimization

Through robotic arm control optimization technology, combined with deep reinforcement learning and PID control, efficient and uniform repair of building wall cracks is achieved, the problems of low efficiency and safety hazards in the existing technology are solved, and the repair effect and safety are improved.

CN119021503APending Publication Date: 2024-11-26GREATER BAY AREA INST FOR INNOVATION HUNAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411196663.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art has problems such as low efficiency, inaccurate control of force, and waste of materials in the repair of cracks on building walls, resulting in poor repair results and safety hazards in manual repair.

Method used

Using a method based on robotic arm control optimization, the robotic arm control is optimized through deep reinforcement learning algorithms and force and motion combined controllers, combined with PID controllers and laser scanners to achieve uniform extrusion and adhesion of materials at different speeds and directions. , ensure the efficiency and effect of crack filling, and surface polishing is performed through quick tool changer.

Benefits of technology

It realizes efficient and even repair of wall cracks, improves repair effect and safety, ensures sufficient filling of materials and smooth surfaces, and reduces the risk of manual repair.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119021503A_ABST
    Figure CN119021503A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of robot control, and particularly relates to a wall crack repairing method based on mechanical arm control optimization, and the method comprises the steps: carrying out the self-calibration of an extruder, carrying out the mechanical arm force and motion combined control optimization algorithm, controlling the material printing force of an extrusion head, and the like. And the speed and the material extrusion rate are adjusted according to the size and the shape of the crack in the repairing process through sensors such as a PID controller and a laser scanner, and the crack is filled with the material, so that the optimal repairing effect is achieved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of robot control, and in particular relates to a wall crack repair method based on robot arm control optimization. Background Art

[0002] In recent years, with the construction of infrastructure and residential buildings, the problem of wall cracks has gradually emerged as a key challenge. It not only concerns the stability and appearance of the building, but also has a profound impact on people's safety and quality of life. Surface cracks are usually non-structural, and will accelerate the deterioration of the building structure due to the intrusion of moisture or chemicals, shortening its service life, and may even induce safety hazards such as fractures and collapses, posing a direct threat to people's safety. At present, most of the repairs for building wall cracks are done manually, but there are problems such as low efficiency of manual repair, inaccurate force control, and waste of materials, which leads to poor results in building crack repair. Summary of the invention

[0003] The purpose of the present invention is to provide a wall crack repair method based on robot arm control optimization to solve the problems mentioned in the background technology.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is as follows:

[0005] 1. A wall crack repair method based on robot arm control optimization, characterized in that it includes the following steps:

[0006] Step 1: Initial extrusion calibration:

[0007] Ensure the consistency and accuracy of material extrusion at different speeds. Determine the relationship between robot arm speed and material extrusion volume by printing material strips at different speeds and measuring their cross-sectional areas.

[0008] Step 2: Robotic arm control optimization:

[0009] Use deep reinforcement learning algorithms and force and motion joint controllers to optimize robot arm control so that the robot moves along the optimized path, and adjust the torque and movement speed in vibrating or unstable environments to ensure the efficiency and effectiveness of crack filling;

[0010] Step 3: Material filling pressure control:

[0011] By real-time monitoring and adjusting the extrusion force, the material can be ensured to have uniform adhesion and interlayer bonding strength in different printing layers and in different printing directions, avoiding uneven extrusion or material accumulation;

[0012] Step 4: Repair verification:

[0013] After the filling process is completed, a laser scanner is used to perform a post-fill scan to verify the repair effect by calculating the difference between the pre-filled contour and the post-filled contour;

[0014] Step 5: Surface treatment:

[0015] After the repair is completed, use the quick tool changer to replace the end effector tool with sandpaper material, polish the repair area flat, and ensure a smooth surface.

[0016] Compared with the prior art, the present invention has at least the following advantages:

[0017] This application achieves the best repair effect by autonomously calibrating the extruder, optimizing the joint control algorithm of the robot arm force and motion, controlling the printing force of the extrusion head material, etc., and adjusting the speed and material extrusion rate to fill the cracks according to the size and shape of the cracks through sensors such as PID controllers and laser scanners during the repair process. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The present invention is further illustrated by means of the non-limiting examples given in the accompanying drawings.

[0019] Figure 1 Schematic diagram of the steps of optimizing the robot arm control of the present invention.

[0020] Figure 2 Schematic diagram of the steps for controlling the filling pressure of the material of the present invention.

[0021] Figure 3 It is a schematic diagram of the overall process of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0023] like Figure 1-3 As shown, a wall crack repair method based on robot arm control optimization includes the following steps:

[0024] Step 1: Initial extrusion calibration:

[0025] Ensure the consistency and accuracy of material extrusion at different speeds. Determine the relationship between robot arm speed and material extrusion volume by printing material strips at different speeds and measuring their cross-sectional areas.

[0026] Step 2: Robotic arm control optimization:

[0027] Use deep reinforcement learning algorithms and force and motion joint controllers to optimize robot arm control so that the robot moves along the optimized path, and adjust the torque and movement speed in vibrating or unstable environments to ensure the efficiency and effectiveness of crack filling;

[0028] Step 3: Material filling pressure control:

[0029] By real-time monitoring and adjusting the extrusion force, the material can be ensured to have uniform adhesion and interlayer bonding strength in different printing layers and in different printing directions, avoiding uneven extrusion or material accumulation;

[0030] Step 4: Repair verification:

[0031] After the filling process is completed, a laser scanner is used to perform a post-fill scan to verify the repair effect by calculating the difference between the pre-filled contour and the post-filled contour;

[0032] Step 5: Surface treatment:

[0033] After the repair is completed, use the quick tool changer to replace the end effector tool with sandpaper material, polish the repair area flat, and ensure a smooth surface.

[0034] In step 1, in order to meet the needs of building material repair for different buildings, it is necessary to determine the extrusion rate at different speeds and extrude the repair material at the correct speed and amount to achieve the best repair effect. The specific steps are as follows:

[0035] S1: Print material strips and identify edges, extrude materials at different speeds, and measure their cross-sectional areas based on a laser scanner. The scanner provides 2D measurement data, where the x-axis represents the width of the scan line and the y-axis represents the distance from the scanner to the material. The edge of the printed material is identified by the maximum rate of change of the laser depth z to determine the corresponding relationship between the robot's movement speed and the amount of extruded material, satisfying the following formula;

[0036] x1=x[i1]

[0037] x2=x[i2]

[0038] Where x1 and x2 are the maximum x values ​​of the laser depth z change at index i1 and i2.

[0039] S2: Calculate the cross-sectional area. Use the identified edges x1 and x2 to calculate the cross-sectional area A by integration or summation, satisfying the following formula:

[0040]

[0041] Where Δx is the change between values ​​of x and A is the cross-sectional area.

[0042] S3: Find the center position and calculate the center position of the cross section, which can be achieved by averaging the identified edge index. According to the measurement results, the relationship between the robot speed and the material extrusion amount is established to ensure that the robot speed can be adjusted according to the size of the crack during the crack repair process to obtain the best filling effect. Satisfy the following formula:.

[0043]

[0044] (c x ,c y )=(x[i avg ],z[i avg ])

[0045] Among them, i avg is the average value of the index, c x and c y are the center positions of the crack on the x-axis and y-axis at the laser depth z, respectively.

[0046] In step 2, the robot moves to the crack and uses a laser scanner to refine the crack coordinates. After obtaining the precise crack position and shape, the crack repair work can be started. The robot moves along the predetermined path of the crack and controls the extruder to inject the material into the crack through a multi-joint robotic arm. To optimize the control accuracy and flexibility of the multi-joint robotic arm, two deep reinforcement learning algorithms (DRL), Soft Actor-Critic (SAC) algorithm and Proximal Policy Optimization (PPO) algorithm, are used. The hyperparameters of these two algorithms are optimized through Tree-structured Parzen Estimator (TPE). At the same time, a joint motion and force controller is used to realize the path control of motion and force to facilitate the repair operation of the robotic arm.

[0047] Among them, the PPO algorithm is a DRL algorithm known for its stability and sample efficiency. It aims to optimize the strategy while ensuring that the update is not too drastic. Its objective function can be expressed as:

[0048]

[0049] Among them, L CLIP (θ) is the objective function of clipping, represents the expected value on a batch of samples, r t (θ) is the probability ratio of the current strategy to the old strategy, is the advantage function estimate at time t, and ∈ is the clipping parameter.

[0050] The SAC algorithm is a model-free deep reinforcement learning algorithm based on maximum entropy, which is suitable for continuous action space. Its objective function can be expressed as:

[0051]

[0052] Among them, θ is the policy parameter, γ is the trajectory sampled in the policy, r(s t ,a t ) is the immediate reward, α is the temperature parameter, H(π(·|s t )) is the entropy of the strategy.

[0053] TPE is a tree-based Bayesian optimization algorithm used to solve the global optimization problem of black box functions. Its innovative approach includes creating a binary tree model to map the distribution of various hyperparameters, thereby effectively optimizing the robot control with efficiency and accuracy. The algorithm optimizes by iteratively selecting hyperparameters to satisfy the following formula:

[0054]

[0055] Here, θ represents the optimized hyperparameters and P(·) represents the conditional probability, i.e., the probability that the objective function is better or worse than the current best.

[0056] The Lagrangian method is a method used to analyze dynamic systems, especially physical systems in classical mechanics, focusing on the energy changes of the system. Its equation satisfies the following formula:

[0057]

[0058] Where L is the Lagrangian, q i are generalized coordinates, is the generalized velocity, which is the derivative of the generalized coordinates.

[0059] ①TPE optimization algorithm includes the following steps:

[0060] (1) Determine the hyperparameter set. A basic hyperparameter set is determined by randomly sampling a series of hyperparameters, which can produce favorable results. Specifically, the following parameters are included:

[0061] A. Joint angle (θ)

[0062] B. End effector position (EE pos )

[0063] C. Target location pos )

[0064] D.Action (A t): corresponds to the individual motions of each joint of the robot arm, provided by the agent to update the state of the environment.

[0065] (2) Through the reward function (R t ) refines the accuracy and responsiveness of the model and systematically improves model performance through iterative learning. Satisfies the following equation:

[0066]

[0067] Among them, (x EE ,y EE ,z EE ) represents the current position of the end effector, (x Goal ,y Goal ,z Goal ) represents the coordinates of the target position. t It is a negative value of distance, which means the closer you are to the target crack, the higher the reward.

[0068] The model trained with TPE and default hyperparameters is used to optimize the control of the robot arm. During the entire repair process, the robot's movement speed is dynamically adjusted according to the cross-sectional area of ​​the crack. If the cross-sectional area of ​​the crack is large, the robot will slow down to ensure that the material is fully filled; on the contrary, if the area is small, the speed will be increased accordingly. This speed adjustment mechanism ensures that the crack is filled efficiently and evenly, thereby improving the durability and durability of the repaired structure.

[0069] ②The steps of the joint control algorithm of the robot arm force and motion are as follows:

[0070] (1) Dynamic analysis of the robot arm. Using the Lagrangian method, the Lagrangian vector L of the system is defined as:

[0071]

[0072] Among them, KE is the total kinetic energy and PE is the total potential energy.

[0073] For the robotic arm, its kinetic energy satisfies the following formula:

[0074]

[0075] Among them, m i is the quality of the ith connection, ν i is the linear velocity, ω i is the angular velocity, I i is the inertia matrix of the ith link relative to the inertial reference frame.

[0076] For the robotic arm, its potential energy satisfies the following formula:

[0077]

[0078] Where g is the acceleration due to gravity, h is the i is the height of the centroid of the ith link.

[0079] Finally, the Lagrangian L of the system can be defined as:

[0080]

[0081] (2) Use the Jacobian matrix to express the linear velocity and angular velocity as joint velocity, satisfying the following formula:

[0082]

[0083] in, is the joint velocity v i is the linear velocity, ω i is the angular velocity.

[0084] (3) Calculate the joint torque τ required to achieve the desired motion, and obtain the following equation using the Lagrange equation:

[0085]

[0086] Among them, a kj represents the dynamic acceleration on which the kth and jth depend, is the angular acceleration of the jth joint, It represents the torque change caused by the change of joint velocity. represents the torque effect of potential energy on joint k.

[0087] (4) Controller design. The combined controller aims to meet the free space control objective by integrating motion control and force control into a single comprehensive design. Trajectory tracking is achieved through a position error dynamic model that satisfies the following equation:

[0088]

[0089] Where e is the position error, x is the coordinate of the system initialization reference frame, and x ref is the trajectory position vector, is the speed error, is based on the difference between the position error and the velocity error. C is a 3×3 gain matrix that is used to describe the dynamic behavior of the error e over time and to control the corresponding input.

[0090] (5) Motion control. The motion control of the robot arm is achieved by optimizing the output parameters of the controller. The actual position and speed of the end effector are measured in real time and fed back to the control algorithm for error calculation and torque update to satisfy the following equation:

[0091]

[0092] Among them, J T is the Jacobian matrix transpose, which transforms the moment from task space to joint space, K p is the proportional gain, which is proportional to the position error and represents the torque in the direction of the desired trajectory, Represents the integral gain, which is proportional to the speed error and is used to provide resistance to reduce system oscillation.

[0093] In step three, the repair material is delivered to the end effector through a delivery pump, which is equipped with a print head capable of extruding the material;

[0094] At the beginning of the repair, a small amount of material will be pre-extruded to ensure the uniformity of material flow and remove possible impurities in the nozzle. This pre-extrusion operation helps to reduce the unevenness of material output;

[0095] When filling the cracks with repair materials, the extrusion speed of the material is monitored and adjusted in real time. By combining Bayesian optimization with transfer learning and using an optimized PID controller, the dynamic changes of the filling process can be adapted, and the deviations between fillings can be monitored and corrected in real time to ensure that the material can fill the cracks evenly and continuously.

[0096] The PID controller is used to control the extrusion speed, maintain the driving force at the required constant reference value, and can make appropriate adjustments to the basic process while handling external disturbances.

[0097] Bayesian Optimization (BO) is a data-driven iterative optimization method used to optimize black box functions that can only be sampled point by point. This method is used to adjust the printing process parameters and update the rules and obtain new sample points through the Gaussian Process (GP). The following formula is satisfied:

[0098] μ t (x) = k t (x) T (K t +ρ 2 I) -1 y t

[0099]

[0100] Among them, μ t (x) is the posterior mean at point x, which is used to describe the expected value of the probability distribution of parameter updates, k t (x) kernel function vector, whose elements are k(x,x i ), x iis the previously observed point, K t is the kernel matrix, whose element K t (i,j)=k(x i ,x j ), ρ 2 is the noise variance, I is the unit matrix, y t is the result vector collected so far. is the posterior variance, a key quantity used to predict uncertainty.

[0101] Transfer learning (TL) is the use of a transfer learning strategy to initialize the Bayesian optimization process. This involves using a dataset collected from a previous task to guide the optimization process of a new task, thereby accelerating convergence and improving tuning efficiency.

[0102] A black box objective function is one in which only the relationship between input and output is known, but the internal working principle or calculation process of the function is unclear.

[0103] Expected Improvement (EI) is a commonly used acquisition function in Bayesian optimization, which is used to guide the search process in the parameter space to find the minimum value of the objective function.

[0104] The specific material output pressure control method is as follows:

[0105] ① Optimize the structure of the PID controller. Use the improved PID controller to adjust the extrusion force. By modifying the derivative part, it contains two derivative terms. The fast differential is used to calculate the machine motion system with a smaller time constant, and the slow differential is used for the extrusion system with a larger time constant. The purpose is to calculate the numerical derivative of the force measurement at different times to meet the following formula:

[0106]

[0107] Where u(t) is the controller output, e(t) is the difference between the set point and the actual extrusion force, and K p , K i , K d , K dd They are proportional, integral, fast derivative and slow derivative gains respectively.

[0108] ②Perform continuous Bayesian optimization. In order to fill the required shape and amount of cracks, it is necessary to execute the Bayesian optimization program during filling, and perform a large number of Bayesian iterations in one filling. The specific working principle is as follows:

[0109] a. Execute the command for the required crack filling, and the filling head moves along the predefined trajectory;

[0110] b. Initialize basic control by setting the desired force reference and random initial controller;

[0111] c. The controller records 10s of printing data, and the Bayesian optimization program evaluates the filling results and immediately deploys new controller parameters. This is repeated until the filling is completed or the user interrupts.

[0112] The error between the measured extrusion force and the expected reference force is minimized by calculating the root mean square error (RMSE) of the force, given the reference force F r and continuity The measurement sequence satisfies the following formula:

[0113]

[0114] ③ Evaluate the controller performance. The controller performance is evaluated by calculating the RMSE of the data recorded when the controller is deployed, with the goal of minimizing RMSE. Given a set of controller gains x = {K p ,K i ,K d ,K dd}, and the forces recorded when the controller is deployed The defined Bayesian black box objective function satisfies the following formula:

[0115]

[0116] The expected improvement (EI) function is selected as the information collection strategy. In each iteration, the point that is most likely to bring improvement is selected, satisfying the following formula:

[0117]

[0118] in, is the currently known best objective function value, f(x) is the predicted value at point x, and ξ is a small positive number (exploration noise) used to control exploration, which is used to find new points with a certain probability when the predicted value is good.

[0119] ④ Use transfer learning to accelerate parameter tuning. Expand the controller gain x = {K p ,K i ,K d ,K dd ,F r}, using the normalized root mean square error as the optimization target, this method can effectively improve the efficiency and effect of Bayesian optimization, thereby better realizing the control of extrusion force and satisfying the following formula:

[0120]

[0121] In step 4, after the filling process is completed, a laser scanner is used to perform a post-fill scan. The absolute normalized difference between the pre-filled contour and the crack contour after filling is calculated to quantify the accuracy of the repair. The heuristic algorithm used in the initial scan and calibration process is applied to the scan data after filling. The unfilled points will have a certain depth, while the filled points will have a certain height. The effectiveness of the repair is evaluated by comparing the pre-filled and post-filled profiles. The following formula is satisfied:

[0122]

[0123] Among them, ∈ fill represents the filling error, A pre is the cross-sectional area of ​​the crack before filling, A post Represents the cross-sectional area of ​​the crack that is filled at exactly the same coordinates. Lower values ​​mean a more effective repair and a flatter surface.

[0124] In step 5, after the wall cracks are filled, the surface needs to be polished to maintain the aesthetics;

[0125] After the wall filling material is stable, use the quick-change tool to replace the print head with a sandpaper grinding tool to process the repaired wall to remove uneven parts, keep the repaired part smooth with the surrounding wall, and reduce the visibility of the repair marks; use the FORGE framework to achieve adaptive grinding and repair of the wall under the maximum allowable force condition to avoid damage to the wall due to excessive force. The framework uses reinforcement learning. The algorithm can learn and optimize the grinding strategy in a simulated environment, and simulate various uncertainties and changes in the real world through dynamic randomization to ensure the robustness and adaptability of the strategy. The specific steps include:

[0126] ① Set the force threshold. During the grinding process, excessive force can cause damage to the wall and cause cracks to form again. In order to develop a safe and reliable surface treatment, we first use the force threshold to set F th :π(a|o,F th ), if the contact force sensed by the robot Exceeding the force threshold F th , the strategy will be punished by adding an additional penalty term to the reward function to satisfy the following formula:

[0127]

[0128] in, is the penalty caused by excessive contact force, β is the weight parameter used to adjust the penalty intensity, is the magnitude of the contact force at time t, F th is the force threshold set. This penalty term is subtracted from the total reward to ensure that the policy takes force control into account during the learning process.

[0129] ② Use dynamic randomization technology to simulate the diversity and uncertainty of real walls to enhance the robustness of the strategy. The specific steps include:

[0130] (1) Controller randomization. The controller has a great influence on pressure sensing, which uses impedance control, where the applied force satisfies the following equation:

[0131]

[0132] in, is the target position at time t, a t is the relative position of the strategy output, p fixed is the position of the wall polishing, λ is the action scale parameter, which is used to limit the distance between the target position and the current position of the end effector, and the clip function ensures that the target position does not exceed the operating range of the robot arm. targ is the force exerted by the target on the end effector, k p and k d are the proportional and differential gains, respectively, used to calculate the required force to reach the target position, is the current position of the end effector, is the current speed of the end effector.

[0133] (2) Expected return of dynamic randomization. A parameterized strategy is used to make the robot follow the corresponding command to maximize the expected return. Combined with the force threshold, the strategy can adjust its actions to limit the contact force, satisfying the following equation:

[0134]

[0135] Among them, J(π θ ) means that under the condition of dynamic randomization, the strategy π θ The expected return of p(τ,|π θ ,Ψ) represents the strategy π θ The probability distribution of the trajectory τ under dynamic randomization is considered, r t is the immediate reward at time t, and γ is the discount factor used to calculate the current value of future rewards.

[0136] ③Completion termination. During polishing, the laser scanner detects the wall condition in real time. When the processed wall surface is the same as the surrounding wall surface, the polishing operation is terminated. Through a trained success predictor, the time when the wall polishing is completed can be predicted in real time. Set a termination reward function Used to punish incorrect successful predictions, satisfying the following formula:

[0137]

[0138] in, Represents the termination action at time t, which is used to predict whether the polishing is completed. Its output is between 0 and 1. It is compared with the confidence threshold p to determine whether the completion of the strategy is credible. When it is greater than p, the task is completed. is the termination reward at time t, used to train the policy to correctly predict task completion, and a t is the action output, y t is the label of true success, success is 1 and failure is 0.

[0139] Furthermore, in step 2, the robot arm control optimization is specifically as follows:

[0140] 1. Refine the coordinates: The robot moves to the crack and uses a laser scanner to refine the crack coordinates;

[0141] 2. Pre-extrusion of a small amount of material: ensure uniformity of material flow and remove possible impurities in the nozzle;

[0142] 3. Hyperparameter optimization, specifically:

[0143] (1) Determine the hyperparameter set: Determine a basic hyperparameter set by randomly sampling a series of hyperparameters that can produce favorable results;

[0144] (2) Refine the accuracy and responsiveness of the model: Systematically improve model performance through reward functions and iterative learning;

[0145] (3) Optimizing the control of the robot arm: Using the model trained with TPE and default hyperparameters, the robot’s movement speed is dynamically adjusted according to the cross-sectional area of ​​the crack during the entire repair process.

[0146] Furthermore, in step three, the material filling pressure control is specifically as follows:

[0147] 1. The system first controls the movement of the nozzle through the motion controller and controls the extrusion of the material according to the printing trajectory.

[0148] 2. The sensor measures the force of material extrusion in real time and feeds the data back to the extrusion controller;

[0149] 3. The Bayesian optimization algorithm uses this data to continuously adjust the gain parameters of the extrusion controller to ensure the stability and accuracy of the printing process;

[0150] 4. This closed-loop system can gradually improve printing quality, reduce errors, and achieve optimized printing effects.

[0151] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A wall crack repair method based on robot arm control optimization, characterized in that: The following steps are involved: Step 1: Initial extrusion calibration: Ensure the consistency and accuracy of material extrusion at different speeds. Determine the relationship between robot arm speed and material extrusion volume by printing material strips at different speeds and measuring their cross-sectional areas. Step 2: Robotic arm control optimization: Use deep reinforcement learning algorithms and force and motion joint controllers to optimize robot arm control so that the robot moves along the optimized path, and adjust the torque and movement speed in vibrating or unstable environments to ensure the efficiency and effectiveness of crack filling; Step 3: Material filling pressure control: By real-time monitoring and adjusting the extrusion force, the material can be ensured to have uniform adhesion and interlayer bonding strength in different printing layers and in different printing directions, avoiding uneven extrusion or material accumulation; Step 4: Repair verification: After the filling process is completed, a laser scanner is used to perform a post-fill scan to verify the repair effect by calculating the difference between the pre-filled contour and the post-filled contour; Step 5: Surface treatment: After the repair is completed, use the quick tool changer to replace the end effector tool with sandpaper material, polish the repair area flat, and ensure a smooth surface.

2. According to claim 1, a wall crack repair method based on robot arm control optimization is characterized in that: In step 1, in order to meet the needs of repairing building materials used in different buildings, it is necessary to determine the extrusion rate at different speeds, extrude the repair material at the correct speed and amount, so as to achieve the best repair effect. The specific steps are as follows: S1: Print material strips and identify edges; S2: Calculate the cross-sectional area; S3: Find the center location.

3. The wall crack repair method based on robot arm control optimization according to claim 2 is characterized in that: In the step 2, the robot moves to the crack and uses a laser scanner to refine the crack coordinates. After obtaining the precise crack position and shape, the crack repair work can be started. The robot moves along the predetermined path of the crack, and controls the extruder to inject the material into the crack through the multi-joint robotic arm. In order to optimize the accuracy and flexibility of the control of the multi-joint robotic arm, two deep reinforcement learning algorithms, the Soft Actor-Critic algorithm and the proximal strategy optimization algorithm, are used to optimize the hyperparameters of the two algorithms through the Tree-structured Parzen Estimator. At the same time, a joint controller of motion and force is used to realize the path control of motion and force, so as to facilitate the repair operation of the robotic arm.

4. The wall crack repair method based on robot arm control optimization according to claim 3 is characterized in that: In the step 3, the repair material is delivered to the end effector through a delivery pump, and the end effector is equipped with a print head capable of extruding the material; At the beginning of the repair, a small amount of material will be pre-extruded to ensure the uniformity of material flow and remove possible impurities in the nozzle. This pre-extrusion operation helps to reduce the unevenness of material output; When filling the cracks with repair materials, the extrusion speed of the material is monitored and adjusted in real time. By combining Bayesian optimization with transfer learning and using an optimized PID controller, the dynamic changes of the filling process can be adapted, and the deviations between fillings can be monitored and corrected in real time to ensure that the material can fill the cracks evenly and continuously.

5. The wall crack repair method based on robot arm control optimization according to claim 4 is characterized in that: In step 4, after the filling process is completed, a laser scanner is used to perform a post-filling scan. The absolute normalized difference between the pre-filled contour and the crack contour after filling is calculated to quantify the accuracy of the repair. The heuristic algorithm used in the initial scanning and calibration process is applied to the scanned data after filling. The unfilled points will have a certain depth, and the filled points will have a certain height. The effectiveness of the repair is evaluated by comparing the pre-filled and post-filled profiles.

6. The wall crack repair method based on robot arm control optimization according to claim 5 is characterized in that: In step 5, after the wall cracks are filled, the surface needs to be polished to maintain the aesthetics; After the wall filling material is stable, use the quick-change tool to replace the print head with a sandpaper grinding tool to process the repaired wall to remove uneven parts, so that the repaired part and the surrounding wall are smooth and the visibility of the repair marks is reduced; The FORGE framework is used to achieve adaptive grinding and repair of walls under the maximum allowable force conditions to avoid damage to the wall due to excessive force. The framework uses reinforcement learning, which can learn and optimize the grinding strategy in a simulated environment. It simulates various uncertainties and changes in the real world through dynamic randomization to ensure the robustness and adaptability of the strategy.