Adaptive path planning method and system for grinding and polishing robots based on RBF neural network

Abstract

The present invention discloses a method and system for adaptive path planning of a grinding and polishing robot based on RBF neural network, wherein the method comprises the steps of: establishing and training a radial basis function neural network model; collecting and preprocessing grinding and polishing data in real time; inputting the preprocessed grinding and polishing data into the trained radial basis function neural network model to generate the initial grinding and polishing path of the grinding and polishing robot; optimizing the initial grinding and polishing path using a particle swarm algorithm to obtain the optimal grinding and polishing path of the grinding and polishing robot; performing the grinding and polishing task according to the optimal grinding and polishing path, and feeding back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model. Its significant effect is that it can generate and optimize the grinding and polishing path of the robot in real time, and can adapt to workpieces of different shapes and materials, realizing efficient and automatic path planning under various grinding and polishing conditions, thereby improving grinding and polishing efficiency and processing quality.

Description

Technical Field

[0001] The invention relates to the technical field of robot control, and in particular to an adaptive path planning method and system for a grinding and polishing robot based on RBF neural network. Background Art

[0002] Grinding and polishing robots are commonly used for automated grinding and polishing processes, especially in manufacturing industries that require precision machining, such as automotive, aviation, and mold manufacturing. These robots need to operate according to preset paths to ensure that the surface of the workpiece reaches the required accuracy and finish. Traditional grinding and polishing robots mainly rely on the following methods for path planning: Pre-programmed paths: The robot's motion path is defined by programming. This method works well when dealing with standardized and highly repetitive tasks, but lacks flexibility and is difficult to adapt to workpieces with complex shapes or varying sizes; Manual settings: The operating technician manually sets the grinding and polishing path based on experience. This relies on the operator's skills and experience, is inefficient, and the results are easily affected by human factors.

[0003] As the manufacturing industry develops towards the goal of high efficiency and high quality, higher requirements are placed on robot technology, especially in terms of automation and intelligence. The main problems encountered in grinding and polishing operations include: adaptability to complex geometric shapes: traditional methods have difficulty in processing irregular or complex-shaped workpieces, which are very common in the aviation and automotive industries; real-time adjustment capabilities: the unevenness of material removal during grinding and polishing requires the robot to be able to adjust the path in real time to ensure processing quality.

[0004] Although some existing technologies attempt to solve these problems through more complex control algorithms or sensor technologies, most of them still face the following limitations: Slow response speed: It takes time to analyze data and adjust paths, which makes it difficult to meet the needs of high-efficiency production; High technical complexity and cost: Advanced control systems and sensors not only increase the complexity of the robot system, but also greatly increase the cost. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide an adaptive path planning method and system for a grinding and polishing robot based on a radial basis function (RBF) neural network. By integrating the RBF neural network and the particle swarm optimization (PSO) algorithm, the robot's grinding and polishing path can be generated and optimized in real time to adapt to workpieces of different shapes and materials, thereby improving the grinding and polishing efficiency and processing quality.

[0006] The technical solution adopted by the present invention is as follows:

[0007] One of the purposes of the present invention is to propose an adaptive path planning method for a grinding and polishing robot based on an RBF neural network, the key of which is that it comprises the following steps:

[0008] Step 1: Establish a radial basis function neural network model and input the historical grinding and polishing data of the grinding and polishing robot for training;

[0009] Step 2: real-time acquisition of grinding and polishing data of the grinding and polishing robot during the grinding and polishing process and data preprocessing;

[0010] Step 3, inputting the preprocessed grinding and polishing data into the trained radial basis function neural network model to generate the initial grinding and polishing path of the grinding and polishing robot;

[0011] Step 4: Using a particle swarm algorithm to optimize the initial grinding and polishing path to obtain an optimal grinding and polishing path for the grinding and polishing robot;

[0012] Step 5: The grinding and polishing robot performs the grinding and polishing task according to the obtained optimal grinding and polishing path, and feeds back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model to adjust the grinding and polishing path in real time.

[0013] Furthermore, the radial basis function neural network model in step 1 includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input the collected grinding and polishing data; the hidden layer is used to calculate the input grinding and polishing data using the radial basis function to obtain the path adjustment parameters; the output layer is used to generate an optimized grinding and polishing path according to the path adjustment parameters output by the hidden layer.

[0014] Furthermore, the formula for obtaining the path adjustment parameter by calculating the input polishing data in the hidden layer is:

[0015]

[0016] Among them, h i (x) is the path adjustment parameter, x is the input grinding and polishing data, c i is the center of the i-th hidden layer node, σ i is the width parameter of the i-th hidden layer node.

[0017] Furthermore, the output layer generates the optimized grinding and polishing path according to the path adjustment parameters as follows:

[0018]

[0019] Among them, y(x) is the optimized grinding and polishing path, w i is the weight of the i-th hidden layer node, h i (x) is the path adjustment parameter.

[0020] Furthermore, the grinding and polishing data include the geometric shape and position of the workpiece to be ground and polished, and the posture, speed, acceleration and torque of the robot.

[0021] Furthermore, the data preprocessing includes coordinate conversion processing, normalization processing and posture calculation processing.

[0022] Furthermore, the specific process of optimizing the initial grinding and polishing path using the particle swarm algorithm in step 4 is as follows:

[0023] Step 4.1, initialize the particle swarm;

[0024] Step 4.2, for each particle, generate a path based on the parameters of the current radial basis function neural network model and calculate the fitness function value;

[0025] Step 4.3, update the velocity and position of the particle according to the fitness function value;

[0026] Step 4.4: Return to step 4.2 until the adaptability function value converges and the optimal grinding and polishing path is obtained.

[0027] Furthermore, the speed and position of the particle are updated as follows:

[0028] v i (t+1)=ωv i (t)+c1r1( pi -x i (t))+c2r2(gx i (t))

[0029] x i (t+1)=x i (t)+v i (t+1)

[0030] Among them, v i is the particle velocity, x i is the particle position, ω is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers, p i is the best historical position of the particle, g is the global best position, and t is the number of iterations.

[0031] Furthermore, the mathematical expression of the adaptability function is:

[0032] F(x)=α·L(x)+β·S(x)+γ·O(x)

[0033] Among them, L(x) is the path length, S(x) is the path smoothness, O(x) is the obstacle avoidance, and α, β, and γ are weight coefficients.

[0034] The second object of the present invention is to propose an adaptive path planning system for a grinding and polishing robot based on an RBF neural network, for use in the method described in one of the objects, comprising:

[0035] RBF neural network building module, used to build radial basis function neural network model and use the historical grinding and polishing data of the grinding and polishing robot for training;

[0036] The data acquisition and preprocessing module is used to collect the grinding and polishing data of the grinding and polishing robot in real time and perform data preprocessing;

[0037] An initial grinding and polishing path generation module is used to generate an initial grinding and polishing path of the grinding and polishing robot based on the preprocessed grinding and polishing data using a trained radial basis function neural network model;

[0038] A PSO optimization module is used to optimize the initial grinding and polishing path by using a particle swarm algorithm to obtain an optimal grinding and polishing path for the grinding and polishing robot;

[0039] The execution and feedback module is used to control the grinding and polishing robot to perform the grinding and polishing task according to the obtained optimal grinding and polishing path, and to feed back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model to adjust the grinding and polishing path in real time.

[0040] The remarkable effects of the present invention are:

[0041] 1. The present invention introduces a radial basis function (RBF) neural network. In robot path planning, the RBF neural network can quickly generate or adjust the path based on real-time collected data, thereby improving the flexibility and efficiency of grinding and polishing operations.

[0042] 2. By integrating RBF neural network and particle swarm optimization (PSO) algorithm, the present invention can generate and optimize the robot's grinding and polishing path in real time, and can adapt to workpieces of different shapes and materials, realizing efficient and automated path planning under various grinding and polishing conditions, thereby improving grinding and polishing efficiency and processing quality.

[0043] 3. The present invention has adaptive capabilities: through the RBF neural network driven by real-time data, the system can adapt to different processing conditions and changing task requirements.

[0044] 4. The present invention can be optimized in real time: combined with the PSO algorithm, the system can continuously optimize the path during operation, thereby improving operating efficiency and processing quality.

[0045] 5. The present invention realizes closed-loop feedback: the design of the execution and feedback modules enables the entire system to have the ability to learn and self-optimize, and its performance can be continuously improved as the usage time increases. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a method flow chart of the method of the present invention;

[0047] Figure 2It is a structural principle block diagram of the system described in the present invention. DETAILED DESCRIPTION

[0048] The specific implementation manner and working principle of the present invention are further described in detail below with reference to the accompanying drawings.

[0049] Embodiment 1:

[0050] like Figure 1 As shown, this embodiment proposes an adaptive path planning method for a grinding and polishing robot based on an RBF neural network, and the specific steps are as follows:

[0051] Step 1: Establish a radial basis function neural network model and input the historical grinding and polishing data of the grinding and polishing robot for training;

[0052] The radial basis function neural network model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input the collected grinding and polishing data; the hidden layer is used to calculate the input grinding and polishing data using the radial basis function to obtain the path adjustment parameters; the output layer is used to generate the optimized grinding and polishing path according to the path adjustment parameters output by the hidden layer, specifically:

[0053] The formula for the hidden layer to calculate the input polishing data to obtain the path adjustment parameter is:

[0054]

[0055] Among them, h i (x) is the path adjustment parameter, x is the input grinding and polishing data, c i is the center of the i-th hidden layer node, σ i is the width parameter of the i-th hidden layer node.

[0056] The output layer generates the optimized polishing path according to the path adjustment parameters as follows:

[0057]

[0058] Among them, y(x) is the optimized grinding and polishing path, w i is the weight of the i-th hidden layer node, h i (x) is the path adjustment parameter.

[0059] In this example, the historical grinding and polishing data include the geometric shape and position of the workpiece to be ground and polished, and the posture, speed, acceleration, and torque of the robot.

[0060] Step 2: real-time acquisition of grinding and polishing data of the grinding and polishing robot during the grinding and polishing process and data preprocessing;

[0061] In this example, an optical 3D motion capture system is used to collect the grinding and polishing data of the grinding and polishing robot during the grinding and polishing process. The specific implementation method is as follows:

[0062] Motion capture cameras: Installed around the work area to form a field of view covering the entire grinding and polishing workpiece and robot.

[0063] Marking points: reflective or luminous marking points are attached to key parts of the grinding and polishing robot and workpieces for tracking by the motion capture system.

[0064] Data preprocessing unit: responsible for processing the three-dimensional position information captured by the camera.

[0065] In this example, the data types included in the grinding and polishing data are consistent with the above-mentioned historical grinding and polishing data, specifically including the geometric shape and position of the workpiece to be ground and polished, and the posture, speed, acceleration, and torque of the robot.

[0066] The specific steps of the above optical 3D motion capture system for collecting grinding and polishing data are as follows:

[0067] First, data collection is performed;

[0068] Position data: The motion capture camera captures the 3D position of the marker in real time, and calculates the accurate 3D coordinates of the marker by fusing the perspectives of multiple cameras.

[0069] Speed ​​data, acceleration data, etc. can be calculated by software.

[0070] Then, data preprocessing is performed;

[0071] Coordinate conversion: convert the coordinate system of the motion capture system into the coordinate system of the robot control system to ensure data consistency;

[0072] Normalize the data.

[0073] Finally, perform pose calculation;

[0074] Inverse kinematics calculation: Based on the three-dimensional position data, the robot inverse kinematics algorithm is used to calculate the position information of each joint of the robot;

[0075] Posture estimation: Calculate the robot's posture (represented by rotation matrix or quaternion) using the relative position changes of the markers.

[0076] Step 3, inputting the preprocessed grinding and polishing data into the trained radial basis function neural network model to generate the initial grinding and polishing path of the grinding and polishing robot;

[0077] Step 4: Using a particle swarm algorithm to optimize the initial grinding and polishing path to obtain an optimal grinding and polishing path for the grinding and polishing robot;

[0078] The specific process of obtaining the optimal grinding and polishing path of the grinding and polishing robot is as follows:

[0079] Step 4.1, initialize the particle swarm: each particle represents a candidate path solution, that is, a set of RBF neural network parameters, including center (c), width (σ) and weight (w);

[0080] Initialization velocity: assign an initial velocity to each particle;

[0081] Let the number of iterations t = 0;

[0082] Step 4.2, for each particle, generate a path based on the parameters of the current radial basis function neural network model and calculate the fitness function value;

[0083] Step 4.3: Update the particle's velocity and position according to the fitness function value:

[0084] v i (t+1)=ωv i (t)+c1r1( pi -x i (t))+c2r2(9-x i (t))

[0085] x i (t+1)=x i (t)+v i (t+1)

[0086] Among them, v i is the particle velocity, x i is the particle position, ω is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers, p i is the best historical position of the particle, g is the global best position, and t is the current iteration number;

[0087] Step 4.4: Return to step 4.2 until the adaptability function value converges and the optimal grinding and polishing path is obtained.

[0088] In this example, the adaptability function is used to evaluate the quality of the path, with path length and processing quality as the optimization objectives. The adaptability function takes into account factors such as path length, smoothness, and obstacle avoidance. The mathematical expression of the adaptability function is:

[0089] F(x)=α·L(x)+β·S(x)+γ·O(x)

[0090] in:

[0091] L(x): path length, the shorter the better;

[0092] S(x): path smoothness, the smaller the curvature, the better;

[0093] O(x): obstacle avoidance, the farther away from obstacles the better;

[0094] α, β, γ: weight coefficients, adjusted according to actual needs.

[0095] Further:

[0096] The calculation formula of path length L(x) is:

[0097]

[0098] The calculation formula of path smoothness S(x) is:

[0099]

[0100] The calculation formula of obstacle avoidance O(x) is:

[0101]

[0102] Where N is the number of particles in the particle group, d i is the distance from the waypoint to the nearest obstacle.

[0103] Step 5: The grinding and polishing robot performs the grinding and polishing task according to the obtained optimal grinding and polishing path, and feeds back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model to adjust the grinding and polishing path in real time.

[0104] Embodiment 2:

[0105] like Figure 2 As shown, this embodiment proposes an adaptive path planning system for a grinding and polishing robot based on an RBF neural network, which is used to implement the method described in Example 1, including: an RBF neural network construction module, a data acquisition and preprocessing module, an initial grinding and polishing path generation module, a PSO optimization module, and an execution and feedback module; wherein:

[0106] RBF neural network building module, used to build radial basis function neural network model and use the historical grinding and polishing data of the grinding and polishing robot for training;

[0107] The data acquisition and preprocessing module is used to collect the grinding and polishing data of the grinding and polishing robot in real time and perform data preprocessing;

[0108] An initial grinding and polishing path generation module is used to generate an initial grinding and polishing path of the grinding and polishing robot based on the preprocessed grinding and polishing data using a trained radial basis function neural network model;

[0109] A PSO optimization module is used to optimize the initial grinding and polishing path by using a particle swarm algorithm to obtain an optimal grinding and polishing path for the grinding and polishing robot;

[0110] The execution and feedback module is used to control the grinding and polishing robot to perform the grinding and polishing task according to the obtained optimal grinding and polishing path, and to feed back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model to adjust the grinding and polishing path in real time.

[0111] From this we can see that:

[0112] The RBF neural network building module is trained through historical grinding and polishing data to learn the grinding and polishing paths of various materials and shapes. The network weights and deviations are optimized using regression analysis methods to ensure that the output path data has high accuracy.

[0113] The data acquisition and preprocessing module:

[0114] Function: Responsible for collecting all necessary input data, including the geometry and position of the workpiece and the current posture information of the robot;

[0115] Data source: The input data can come from a variety of sensors, such as visual systems, force sensors, position sensors, etc. This embodiment uses an optical 3D motion capture system;

[0116] Before the optical 3D motion capture system collects data, its motion capture cameras are installed around the working area to form a field of view covering the entire grinding and polishing workpiece and the robot; at the same time, reflective or luminous marking points are attached to the key parts of the grinding and polishing robot and the workpiece for tracking by the motion capture system;

[0117] When the optical three-dimensional motion capture system collects data, the motion capture camera captures the three-dimensional position of the marker point in real time;

[0118] Preprocess the collected data: calculate the accurate three-dimensional coordinates of the marker points through multi-camera perspective fusion to form position data, and then calculate the velocity data and acceleration data; convert the coordinate system of the motion capture system to the coordinate system of the robot control system to ensure data consistency, and normalize the data; then use the robot inverse kinematics algorithm to calculate the position information of each joint of the robot based on the three-dimensional position data; finally, use the relative position change of the marker points to calculate the robot's posture (represented by rotation matrix or quaternion).

[0119] Purpose: To ensure that the RBF neural network module receives enough information to generate an effective preliminary path.

[0120] The initial grinding and polishing path generation module:

[0121] Function: Use the pre-processed grinding and polishing data and the pre-trained RBF neural network to generate the initial grinding and polishing path.

[0122] Advantages: RBF neural network can handle nonlinear problems due to its excellent function approximation ability and is suitable for complex path planning tasks.

[0123] Implementation method: The network model is trained with historical data and can quickly adjust and optimize the path based on real-time data.

[0124] The PSO optimization module:

[0125] Function: Optimize the generated initial grinding and polishing path, define the adaptability function, take the path length and processing quality as the optimization goals, and obtain the optimal grinding and polishing path.

[0126] Core mechanism: Use particle swarm optimization algorithm to find the global optimal solution, continuously update the particle position and velocity, and explore potential better paths by simulating the behavior of bird flocks.

[0127] Optimization goals include minimizing the grinding and polishing path length, optimizing energy efficiency, and improving the machined surface quality.

[0128] The execution and feedback module:

[0129] Function: The robot performs the grinding and polishing tasks according to the optimized optimal grinding and polishing path and collects execution data in the process.

[0130] Feedback application: The collected data is used to retrain and optimize the RBF neural network to achieve self-learning and continuous improvement.

[0131] In summary, the embodiment of the present invention introduces a radial basis function (RBF) neural network, which can quickly generate or adjust the path based on real-time collected data in the robot path planning, thereby improving the flexibility and efficiency of the grinding and polishing operation; at the same time, by integrating the RBF neural network and the particle swarm optimization (PSO) algorithm, it can generate and optimize the robot's grinding and polishing path in real time, and can adapt to workpieces of different shapes and materials, thereby realizing efficient and automated path planning under various grinding and polishing conditions, thereby improving the grinding and polishing efficiency and processing quality; in addition, through the RBF neural network driven by real-time data, the system can adapt to different processing conditions and changing task requirements; the design of the execution and feedback module enables the entire system to have the ability to learn and self-optimize, and its performance can be continuously improved with the increase of usage time.

[0132] The technical solution provided by the present invention is described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, the present invention can also be improved and modified in a number of ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. An adaptive path planning method for a grinding and polishing robot based on RBF neural network, characterized in that: The steps include: Step 1: Establish a radial basis function neural network model and input the historical grinding and polishing data of the grinding and polishing robot for training; Step 2: real-time acquisition of the grinding and polishing data of the grinding and polishing robot during the grinding and polishing process and data preprocessing; Step 3, inputting the preprocessed grinding and polishing data into the trained radial basis function neural network model to generate the initial grinding and polishing path of the grinding and polishing robot; Step 4: Using a particle swarm algorithm to optimize the initial grinding and polishing path to obtain the optimal grinding and polishing path of the grinding and polishing robot; Step 5: The grinding and polishing robot performs the grinding and polishing task according to the obtained optimal grinding and polishing path, and feeds back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model to adjust the grinding and polishing path in real time; The radial basis function neural network model in step 1 includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input the collected polishing data; the hidden layer is used to calculate the input polishing data using the radial basis function to obtain the path adjustment parameters; the output layer is used to generate an optimized polishing path according to the path adjustment parameters output by the hidden layer; The grinding and polishing data include the geometric shape and position of the workpiece to be ground and polished, as well as the posture, speed, acceleration and torque of the robot; The specific process of optimizing the initial grinding and polishing path using the particle swarm algorithm in step 4 is as follows: Step 4.1, initialize the particle swarm; Step 4.2, for each particle, generate a path based on the parameters of the current radial basis function neural network model and calculate the fitness function value; Step 4.3, update the velocity and position of the particle according to the fitness function value; Step 4.4: Return to step 4.2 until the adaptability function value converges and the optimal grinding and polishing path is obtained.

2. The adaptive path planning method for a grinding and polishing robot based on RBF neural network according to claim 1 is characterized in that: The formula for the hidden layer to calculate the input polishing data to obtain the path adjustment parameter is: Among them, h i (x) is the path adjustment parameter, x is the input grinding and polishing data, c i is the center of the i-th hidden layer node, σ i is the width parameter of the i-th hidden layer node.

3. The adaptive path planning method for a grinding and polishing robot based on RBF neural network according to claim 1 is characterized in that: The output layer generates the optimized polishing path according to the path adjustment parameters as follows: Among them, y(x) is the optimized grinding and polishing path, w i is the weight of the i-th hidden layer node, h i (x) is the path adjustment parameter.

4. The adaptive path planning method for a grinding and polishing robot based on RBF neural network according to claim 1 is characterized in that: The data preprocessing includes coordinate conversion processing, normalization processing and posture calculation processing.

5. The adaptive path planning method for a grinding and polishing robot based on RBF neural network according to claim 1 is characterized in that: The particle's velocity and position are updated as follows: v i (t+1)=ωv i (t)+c1r1(p i -x i (t))+c2r2(g-x i (t)) x i (t+1)=x i (t)+v i (t+1) Among them, v i is the particle velocity, x i is the particle position, ω is the inertia weight, c1 and c2 are acceleration constants, r1 and r2 are random numbers, p i is the best historical position of the particle, g is the global best position, and t is the number of iterations.

6. The adaptive path planning method for a grinding and polishing robot based on RBF neural network according to claim 1 is characterized in that: The mathematical expression of the adaptability function is: F(x)=α·L(x)+β·S(x)+γ·O(x) Among them, L(x) is the path length, S(x) is the path smoothness, O(x) is the obstacle avoidance, and α, β, and γ are weight coefficients.

7. An adaptive path planning system for a grinding and polishing robot based on RBF neural network, used to implement the method described in any one of claims 1 to 6, characterized in that: include: RBF neural network building module, used to build radial basis function neural network model and use the historical grinding and polishing data of the grinding and polishing robot for training; The data acquisition and preprocessing module is used to collect the grinding and polishing data of the grinding and polishing robot in real time and perform data preprocessing; An initial grinding and polishing path generation module is used to generate an initial grinding and polishing path of the grinding and polishing robot based on the preprocessed grinding and polishing data using a trained radial basis function neural network model; A PSO optimization module is used to optimize the initial grinding and polishing path by using a particle swarm algorithm to obtain an optimal grinding and polishing path for the grinding and polishing robot; The execution and feedback module is used to control the grinding and polishing robot to perform the grinding and polishing task according to the obtained optimal grinding and polishing path, and to feed back the grinding and polishing data collected during the execution of the grinding and polishing task to the radial basis function neural network model to adjust the grinding and polishing path in real time.

Images