A training method and system for a robotic arm control model

Abstract

The present invention discloses a training method and system for a robot arm control model, which relates to the field of robot arm control technology. The robot arm to be simulated is modeled by computer simulation software to generate a visual and dynamically responsive model, which is then divided into a plurality of joint units, and feature data of each joint unit is collected. Simulation is performed under preset loading conditions, and the position state of the end of the robot arm is monitored. A recurrent neural network technology is used for training, and a terminal spatial position deviation value Pdb constructed by feature extraction is used to determine a target qualified area of ​​the terminal position, and then an optimization training analysis instruction is issued, and then the aggregation degree of the terminal position is further analyzed, and the terminal position aggregation index Pdj is calculated, and the index is combined with a mean value of the spatial position deviation to fit a training evaluation index Zpg. Finally, a comparison analysis is performed with a preset evaluation threshold K to comprehensively determine the training effect of the robot arm control model, and take corresponding operations.

Description

Technical Field

[0001] The present invention relates to the technical field of robot arm control, and in particular to a training method and system for a robot arm control model. Background Art

[0002] As an important part of modern industrial automation, robotic arms are widely used in manufacturing, assembly, welding and handling. With the continuous advancement of technology, the flexibility and intelligence of robotic arms have been improved. However, the control accuracy of robotic arms is still a key factor in achieving efficient operation. Accurate control not only affects the motion trajectory and operational stability of the robotic arms, but is also directly related to the quality and production efficiency of the final product. Therefore, improving the control accuracy of robotic arms has become one of the current research hotspots, especially in complex working environments. How to ensure that robotic arms maintain high accuracy when performing precision tasks is an urgent problem to be solved in the industry.

[0003] In order to improve the precision control of the robotic arm, especially in the precision control of the spatial position of the robotic arm end, traditional methods often rely on empirical parameter adjustment and simplified models, which are difficult to meet the increasingly complex operation requirements. The robotic arm control model training method and system proposed in this paper, by constructing a three-dimensional model and combining recurrent neural network technology, analyzes the deviation state between the actual spatial position of the robotic arm end and the target spatial position during multiple simulations of the control model, and provides a systematic training and evaluation mechanism. The implementation of this process solves the problem of precision control of the spatial position of the robotic arm end, can improve the control accuracy of the robotic arm in practical applications, and ensure its reliability and stability in changing environments. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a training method and system for a robotic arm control model, which solves the problems in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a training system for a robot arm control model, including a three-dimensional model building module, a parameter acquisition module, a model training analysis module and a training evaluation module;

[0006] The three-dimensional model building module is used to build the mechanical arm to be simulated using computer simulation software, and integrate it with the three-dimensional model of the mechanical arm through the controller to generate a visual and dynamically responsive three-dimensional model of the mechanical arm, and is also used to divide the three-dimensional model of the mechanical arm into joints to generate a plurality of joint units;

[0007] The parameter acquisition module is used to collect characteristic data information of several joint units, obtain the characteristic data information, and execute a simulation program on the three-dimensional model of the robotic arm through pre-set loading condition data, and monitor the position state of the end of the robotic arm after several simulations to obtain relevant end position data information;

[0008] The model training and analysis module obtains the robot control prediction model based on the recurrent neural network technology training, extracts and analyzes the features of the relevant terminal position data information to construct several groups of terminal spatial position deviation values ​​Pdb, and determines the target qualified area of ​​the robot terminal position in the three-dimensional model of the robot according to the numerical values ​​of the spatial position deviation values ​​Pdb of several groups of terminals, and issues an optimization training and analysis instruction. After receiving the optimization training and analysis instruction, the relevant terminal position data information obtained after several simulations is analyzed for the degree of aggregation, and the terminal position aggregation index Pdj is obtained, which is compared with the spatial position deviation mean. The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model;

[0009] The training evaluation module is used to pre-set an evaluation threshold K, and compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model and take corresponding operations.

[0010] Preferably, the three-dimensional model building module includes a simulation modeling unit and a joint division unit;

[0011] The simulation modeling unit is used to construct the robot arm to be simulated in proportion according to the task requirements of assembly on the production line using the extended toolbox Simulink in the computer simulation software MATLAB, and integrate the controller with the three-dimensional model of the robot arm to control the robot arm, so as to generate a visualized and dynamically responsive three-dimensional model of the robot arm, and then check the integrity and connection reliability of the components of the three-dimensional model of the robot arm, and the specific inspection content includes whether there are missing components and unconnected components in the three-dimensional model of the robot arm;

[0012] The joint division unit is used to identify the overall structure of the robotic arm of the three-dimensional model of the robotic arm. According to the identified joint motion types, kinematic characteristics and connection methods of adjacent links of each joint, the three-dimensional model of the robotic arm is divided into units, each unit corresponding to a joint, to generate a number of independent joint units.

[0013] Preferably, the parameter acquisition module includes a parameter acquisition unit and a simulation monitoring unit;

[0014] The parameter acquisition unit is used to collect characteristic data information of a plurality of joint units, and execute a simulation program on the three-dimensional model of the robotic arm through pre-set loading condition data, wherein the characteristic data information includes the joint numbers, joint lengths and initial joint angles of the plurality of joint units, and the loading condition data includes the rotation angles of each joint, the adjacent joint connection distances, the adjacent joint displacement distances and the adjacent joint torsion angles;

[0015] The simulation monitoring unit monitors the position state of the end of the manipulator in real time during several simulations according to the preset loading condition data to obtain relevant end position data information, wherein the relevant end position data information includes the end target space position and the actual spatial position of the end .

[0016] Preferably, the model training analysis module includes a model training unit, a preliminary analysis unit, an optimization training analysis unit and a training comprehensive evaluation unit;

[0017] The model training unit is used to construct an original control model using recurrent neural network technology, and train and test the original control model with its characteristic data information and related terminal position data information, in combination with a predetermined target spatial position, and use the trained original control model as a terminal control state recognition model to obtain spatial positioning information within the terminal control state recognition model, and use the obtained spatial positioning information to train and test the terminal control state recognition model through a homogeneous transformation matrix algorithm, and then use the trained terminal control state recognition model as a robotic arm control prediction model in combination with training optimization analysis instructions.

[0018] Preferably, the preliminary analysis unit is used to calculate the terminal target spatial position according to the relevant terminal position data information. and the actual spatial position of the end Analysis is performed to construct the spatial position deviation value Pdb of several groups of terminals, with the spatial position deviation value after the i-th simulation For example, it is obtained in the following way:

[0019] ;

[0020] In the formula, It is expressed as the spatial position deviation value after the i-th simulation, It is represented as the X-axis coordinate value of the actual spatial position of the terminal after the i-th simulation. It is represented as the Y-axis coordinate value of the actual spatial position after the i-th simulation. It is expressed as the Z-axis coordinate value of the actual spatial position of the end after the i-th simulation. It is expressed as the X-axis coordinate value of the end target space position. It is expressed as the Y-axis coordinate value of the end target space position. Expressed as the Z-axis coordinate value of the end target space position.

[0021] Preferably, the spatial position deviation values ​​Pdb of several groups of terminals are compared with a preset deviation threshold K. When the spatial position deviation value Pdb does not exceed the deviation threshold K, the corresponding actual spatial position of the terminal is All of them are marked as target qualified areas. When the number of position points in the target qualified area exceeds 95% of the total number of position points, an optimization training analysis instruction is issued.

[0022] According to the spatial position deviation value obtained after the i-th simulation The spatial position deviation values ​​Pdb of several groups of terminals are obtained respectively, and the mean value of the spatial position deviation is calculated based on the statistical mean value algorithm. .

[0023] Preferably, the optimization training analysis unit is used to perform a clustering degree analysis on the relevant terminal position data information obtained after several simulations after receiving the optimization training analysis instruction, and obtain the terminal position clustering index Pdj, and the terminal position clustering index Pdj is obtained by the following formula;

[0024] ;

[0025] Reflects the distribution characteristics of the terminal spatial position, indicating the concentrated area and distribution form of the terminal spatial position points. In the formula, It is represented as the spatial position deviation value after the i-th simulation, where i=1, 2, 3, ..., n, and n represents the number of simulations.

[0026] Preferably, the training comprehensive evaluation unit is used to obtain the terminal position aggregation index Pdj and the spatial position deviation mean The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model. The training evaluation index Zpg is obtained by the following formula:

[0027] ;

[0028] In the formula, and Respectively represented as the terminal position aggregation index Pdj and the mean of spatial position deviation The weight value of , A is represented as the correction constant.

[0029] Preferably, the training evaluation module is used to pre-set an evaluation threshold K, and compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the manipulator control model. The specific contents are as follows:

[0030] If the training evaluation index Zpg ≥ the evaluation threshold K, it means that the training effect of the current robot arm control model meets the robot arm control accuracy requirement, and the first training feedback result is obtained, indicating that the loaded condition data when the current trained control model executes the simulation program is correct, and the loaded condition data is recorded, and the data is imported into the robot arm control platform to perform precision assembly of the production line;

[0031] If the training evaluation index Zpg is less than the evaluation threshold K, it means that the training effect of the current robotic arm control model does not meet the robotic arm control accuracy requirements, and the second training feedback result is obtained, indicating that there are errors in the loaded condition data when the currently trained control model executes the simulation program, including the rotation angles of each joint, the connection distances between adjacent joints, the displacement distances between adjacent joints and the torsion angles of adjacent joints. The loading condition data in the parameter acquisition module continues to be adjusted, and the training and analysis process of the original control model by the model training and analysis module is executed until the first training feedback result is obtained.

[0032] A training method for a robotic arm control model comprises the following steps:

[0033] Step 1: Use computer simulation software to construct the robotic arm to be simulated, and integrate it with the robotic arm three-dimensional model through a controller to generate a visual and dynamically responsive three-dimensional model of the robotic arm, and also divide the three-dimensional model of the robotic arm into joints to generate a number of joint units;

[0034] Step 2: Collect the characteristic data information of several joint units, obtain the characteristic data information, and execute the simulation program on the three-dimensional model of the robotic arm through the pre-set loading condition data, and monitor the end position state of the robotic arm after several simulations to obtain relevant end position data information;

[0035] Step 3: The robot control prediction model is obtained by training based on the recurrent neural network technology, and the relevant terminal position data information is extracted and analyzed to construct several groups of spatial position deviation values ​​Pdb of the terminal. According to the numerical values ​​of the spatial position deviation values ​​Pdb of several groups of terminals, the target qualified area of ​​the terminal position of the robot in the three-dimensional model of the robot is determined, and an optimization training analysis instruction is issued. After receiving the optimization training analysis instruction, the relevant terminal position data information obtained after several simulations is analyzed for the degree of aggregation, and the terminal position aggregation index Pdj is obtained, which is compared with the mean value of the spatial position deviation. The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model;

[0036] Step 4: pre-set the evaluation threshold K, compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model and take corresponding operations.

[0037] The present invention provides a training method and system for a robotic arm control model, which has the following beneficial effects:

[0038] (1) Through modular construction, parameter collection and training analysis based on recurrent neural networks, the accuracy of the end position of the robot arm can be dynamically adjusted and optimized. First, the three-dimensional model construction module uses a visual dynamic response model generated by computer simulation software to provide a real and reliable basis for the control and simulation of the robot arm, so that the movement of the robot arm can be intuitively observed and debugged in a virtual environment. Secondly, the parameter acquisition module effectively collects and monitors the characteristic data information of each joint unit, as well as the simulation results after setting loading conditions, ensuring the accuracy of the data. It not only provides a rich source of information for subsequent model training, but also lays the foundation for optimizing control strategies. The model training and analysis module uses recurrent neural network technology for deep learning, which can extract and analyze the end position data information. By constructing the end spatial position deviation value Pdb, the target qualified area of ​​the robot arm is identified, and the optimization training and analysis instructions are issued. , which improves the responsiveness and adaptability of the control model. Finally, the training evaluation module realizes the quantitative evaluation of the training effect by setting the evaluation threshold K. It can provide timely feedback on the accuracy of the model during the training process to ensure its effectiveness and stability on the production line. When the training evaluation index Zpg reaches the preset standard, the system will promptly record the successful loading conditions and import them into actual production. Otherwise, it will quickly adjust the parameters to correct the errors. This closed-loop feedback mechanism optimizes the training process. In general, by constructing a three-dimensional model and combining recurrent neural network technology, the deviation state between the actual spatial position of the end of the robot and the target spatial position of the control model during multiple simulations is analyzed, providing a systematic training and evaluation mechanism. The realization of this process solves the problem of precision control of the spatial position of the end of the robot, which can improve the control accuracy of the robot in practical applications and ensure its reliability and stability in a changing environment.

[0039] (2) By adopting the recurrent neural network technology and the homogeneous transformation matrix algorithm, not only the accurate training and testing of the original control model is achieved, but also the control accuracy of the end of the robot arm is improved. The preliminary analysis unit constructs the end spatial position deviation value Pdb, which makes the end target spatial position and the actual spatial position of the end The difference between them can be quantified and compared with the preset deviation threshold K to ensure the qualification of the terminal position. When the actual position points in the qualified area exceed 95%, the optimization training analysis instruction is automatically issued to adjust the training strategy in time. This efficient feedback mechanism not only enhances the model's adaptability to dynamic environmental changes, but also provides data support for subsequent refined operations. The mean value of spatial position deviation calculated by statistical methods , providing a scientific basis for the evaluation of training effects.

[0040] (3) By presetting the evaluation threshold K and effectively comparing the training evaluation index Zpg, the training effect of the robot arm control model and the reliability of its actual application are improved. When the training evaluation index Zpg reaches or exceeds the threshold K, it indicates that the training of the control model has met the accuracy requirements, so that it can smoothly enter the precision assembly stage of the production line, ensuring the improvement of production efficiency and quality. This instant feedback mechanism not only ensures the transparency and traceability of the training process, but also provides a practical basis for optimizing the loading condition data, which helps to continuously improve the control strategy. On the contrary, if the training evaluation index Zpg is less than the threshold K, it can quickly identify potential problems, adjust the loading conditions in time, and ensure the accuracy of each parameter, thereby avoiding errors and risks in actual operation. This closed-loop feedback design makes the training of the robot arm control model approach perfection in continuous adjustment and optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A training system block diagram of a robot arm control model of the present invention;

[0042] Figure 2 The present invention is a flow chart of a training method for a robot arm control model. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Example 1

[0045] See also Figure 1 , the present invention provides a training system for a robot arm control model, including a three-dimensional model building module, a parameter acquisition module, a model training analysis module and a training evaluation module;

[0046] The three-dimensional model building module is used to build the mechanical arm to be simulated using computer simulation software, and integrate it with the three-dimensional model of the mechanical arm through the controller to generate a visual and dynamically responsive three-dimensional model of the mechanical arm, and is also used to divide the three-dimensional model of the mechanical arm into joints to generate a plurality of joint units;

[0047] The parameter acquisition module is used to collect characteristic data information of several joint units, obtain the characteristic data information, and execute a simulation program on the three-dimensional model of the robotic arm through pre-set loading condition data, and monitor the position state of the end of the robotic arm after several simulations to obtain relevant end position data information;

[0048] The model training and analysis module obtains the robot control prediction model based on the recurrent neural network technology training, extracts and analyzes the features of the relevant terminal position data information to construct several groups of terminal spatial position deviation values ​​Pdb, and determines the target qualified area of ​​the robot terminal position in the three-dimensional model of the robot according to the numerical values ​​of the spatial position deviation values ​​Pdb of several groups of terminals, and issues an optimization training and analysis instruction. After receiving the optimization training and analysis instruction, the relevant terminal position data information obtained after several simulations is analyzed for the degree of aggregation, and the terminal position aggregation index Pdj is obtained, which is compared with the spatial position deviation mean. The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model;

[0049] The training evaluation module is used to pre-set an evaluation threshold K, and compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model and take corresponding operations.

[0050] In this embodiment, the dynamic response model created by computer simulation software ensures the visualization and practicality of the robot arm in a variety of working environments, providing an intuitive basis for subsequent operation and control. Secondly, the parameter acquisition module systematically collects the characteristic data information of the joint unit and simulates it under the set loading conditions, ensuring the reliability and accuracy of the end position state, providing rich data support for subsequent training. The model training and analysis module uses recurrent neural network technology to extract features and analyze deviations of relevant end position data, effectively constructing the target qualified area of ​​the end position, thereby optimizing the training process. This data-driven method not only improves the adaptability of the model, but also provides a more accurate decision-making basis for real-time control. In addition, the aggregation degree analysis and the calculation of the end position aggregation index Pdj enable the system to deeply understand the distribution characteristics of each position point, thereby further improving the model's processing capabilities for complex tasks. Finally, the training evaluation module systematically judges the training effect by setting the evaluation threshold K, making the training adjustment of the robot arm control model more efficient.

[0051] Example 2

[0052] Please refer to Figure 1 ,Specifically: the three-dimensional model building module includes a simulation modeling unit and a joint division unit;

[0053] The simulation modeling unit is used to construct the robot arm to be simulated in proportion according to the task requirements of assembly on the production line using the extended toolbox Simulink in the computer simulation software MATLAB, and integrate the controller with the three-dimensional model of the robot arm to control the robot arm, so as to generate a visualized and dynamically responsive three-dimensional model of the robot arm, and then check the integrity and connection reliability of the components of the three-dimensional model of the robot arm, and the specific inspection content includes whether there are missing components and unconnected components in the three-dimensional model of the robot arm;

[0054] The joint division unit is used to identify the overall structure of the robotic arm of the three-dimensional model of the robotic arm. According to the identified joint motion types, kinematic characteristics and connection methods of adjacent links of each joint, the three-dimensional model of the robotic arm is divided into units, each unit corresponding to a joint, to generate a number of independent joint units.

[0055] In this embodiment, the accuracy and efficiency of the robot design and control are improved through the effective combination of the simulation modeling unit and the joint division unit. First, the simulation modeling unit uses MATLAB and its Simulink toolbox to construct the robot in proportion according to the specific task requirements of the production line assembly, thereby ensuring the true reproduction and functionality of the model. This modeling method based on actual needs not only enhances the performance of the robot in dynamic response, but also provides a visualization platform for subsequent control, which is convenient for engineers to monitor and adjust operating parameters in real time. By systematically checking the integrity of the model components and the reliability of the connections, the stability of the robot during operation is ensured. Secondly, the joint division unit identifies and analyzes the overall structure of the robot, so that the motion type and kinematic characteristics of each joint can be clarified, thereby achieving accurate unit division in computer simulation software.

[0056] Example 3

[0057] Please refer to Figure 1 ,Specifically: the parameter acquisition module includes a parameter acquisition unit and a simulation monitoring unit;

[0058] The parameter acquisition unit is used to collect characteristic data information of a plurality of joint units, and execute a simulation program on the three-dimensional model of the robotic arm through pre-set loading condition data, wherein the characteristic data information includes the joint numbers, joint lengths and initial joint angles of the plurality of joint units, and the loading condition data includes the rotation angles of each joint, the adjacent joint connection distances, the adjacent joint displacement distances and the adjacent joint torsion angles;

[0059] The simulation monitoring unit monitors the position state of the end of the manipulator in real time during several simulations according to the preset loading condition data to obtain relevant end position data information, wherein the relevant end position data information includes the end target space position and the actual spatial position of the end .

[0060] In this embodiment, the accuracy and reliability of the robot arm operation are improved through the organic combination of the parameter acquisition unit and the simulation monitoring unit. First, the parameter acquisition unit systematically collects the characteristic data information of each joint unit, such as the joint number, joint length and initial angle, to ensure the understanding of the overall structure of the robot arm and provide a basis for subsequent simulation. At the same time, the preset loading condition data ensures that the real operating environment can be reflected in actual work. This data collection not only improves the credibility of the simulation model, but also provides an effective basis for model optimization. Secondly, the simulation monitoring unit monitors the end position status of the robot arm in real time, including the end target space position. and the actual spatial position of the end , ensuring the acquisition of position data during multiple simulations, and providing a data basis for the subsequent simulation process of the robotic arm three-dimensional model.

[0061] Example 4

[0062] Please refer to Figure 1 ,Specifically: the model training and analysis module includes a model training unit, a preliminary analysis unit, an optimization training and analysis unit, and a training comprehensive evaluation unit;

[0063] The model training unit is used to construct an original control model using recurrent neural network technology, and train and test the original control model with its characteristic data information and related terminal position data information, in combination with a predetermined target spatial position, and use the trained original control model as a terminal control state recognition model to obtain spatial positioning information within the terminal control state recognition model, and use the obtained spatial positioning information to train and test the terminal control state recognition model through a homogeneous transformation matrix algorithm, and then use the trained terminal control state recognition model as a robotic arm control prediction model in combination with training optimization analysis instructions.

[0064] It should be noted that the homogeneous transformation matrix algorithm is a mathematical tool used to describe and calculate the position and posture transformation of objects in space. It can uniformly represent the motion posture of translation or rotation as matrix operations, thereby simplifying the spatial positioning calculation in robot arm control and simulation.

[0065] Specifically, the preliminary analysis unit is used to calculate the terminal target spatial position according to the relevant terminal position data information. and the actual spatial position of the end Analysis is performed to construct the spatial position deviation value Pdb of several groups of terminals, with the spatial position deviation value after the i-th simulation For example, it is obtained in the following way:

[0066] ;

[0067] In the formula, It is expressed as the spatial position deviation value after the i-th simulation, It is represented as the X-axis coordinate value of the actual spatial position of the terminal after the i-th simulation. It is represented as the Y-axis coordinate value of the actual spatial position after the i-th simulation. It is expressed as the Z-axis coordinate value of the actual spatial position of the end after the i-th simulation. It is expressed as the X-axis coordinate value of the end target space position. It is expressed as the Y-axis coordinate value of the end target space position. Expressed as the Z-axis coordinate value of the end target space position.

[0068] Specifically, the spatial position deviation values ​​Pdb of several groups of terminals are compared with the preset deviation threshold K. When the spatial position deviation value Pdb does not exceed the deviation threshold K, the corresponding actual spatial position of the terminal is All of them are marked as target qualified areas. When the number of position points in the target qualified area exceeds 95% of the total number of position points, an optimization training analysis instruction is issued.

[0069] According to the spatial position deviation value obtained after the i-th simulation The spatial position deviation values ​​Pdb of several groups of terminals are obtained respectively, and the mean value of the spatial position deviation is calculated based on the statistical mean value algorithm. .

[0070] In this embodiment, by combining recurrent neural network and systematic deviation analysis, the training effect and application accuracy of the robot arm control model are improved. First, the model training unit uses recurrent neural network technology to construct the original control model, which can effectively process complex data features and terminal position data, so that the model has a high degree of adaptability and learning ability. This intelligent training method, combined with the homogeneous transformation matrix algorithm, conducts in-depth analysis of spatial positioning information, so that the terminal control state recognition model can accurately map the dynamic behavior of the robot arm. The preliminary analysis unit calculates the terminal spatial position deviation value Pdb, which provides an important basis for subsequent performance optimization. By comparing the actual position with the target position, potential control deviations can be identified in time. When the terminal position deviation value Pdb does not exceed the preset deviation threshold K, the system automatically marks the position that meets the standard as the target qualified area. This process helps to ensure the quality and availability of training data. In addition, if the number of position points in the target qualified area exceeds 95%, the system issues an optimization training analysis instruction to promote further model optimization analysis. This adaptive feedback mechanism not only speeds up the training process, but also improves the accuracy and reliability of the model. At the same time, the statistical mean algorithm calculation makes the entire evaluation process more scientific and reasonable.

[0071] Example 5

[0072] Please refer to Figure 1 Specifically: the optimization training analysis unit is used to perform a clustering degree analysis on the relevant terminal position data information obtained after several simulations after receiving the optimization training analysis instruction, and obtain the terminal position clustering index Pdj, and the terminal position clustering index Pdj is obtained by the following formula;

[0073] ;

[0074] It reflects the distribution characteristics of the terminal spatial position, indicating the concentrated area and distribution form of the terminal spatial position points. The smaller the value, the more concentrated it is, and the larger the value, the more dispersed it is. In the formula, It is represented as the spatial position deviation value after the i-th simulation, where i=1, 2, 3, ..., n, and n represents the number of simulations.

[0075] Specifically, the training comprehensive evaluation unit is used to obtain the terminal position aggregation index Pdj and the spatial position deviation mean The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model. The training evaluation index Zpg is obtained by the following formula:

[0076] ;

[0077] In the formula, and Respectively represented as the terminal position aggregation index Pdj and the mean of spatial position deviation The weight value of , A is represented as the correction constant.

[0078] In this embodiment, the accuracy and reliability of the robot arm control model are improved by analyzing the terminal position aggregation index Pdj. The terminal position aggregation index Pdj reflects the distribution characteristics of the terminal spatial position, so that the model can clearly identify the concentrated area and dispersion of the terminal position points, and then help optimize the control strategy through quantitative evaluation of the degree of aggregation. When the terminal position aggregation index Pdj is small, indicating that the position points are highly aggregated, targeted adjustments can be made to ensure that the robot arm has higher stability in complex environments. At the same time, the training comprehensive evaluation unit calculates the terminal position aggregation index Pdj and the mean value of the spatial position deviation. This systematic evaluation mechanism not only provides a scientific basis for subsequent model training optimization, but also accelerates the improvement of control accuracy, ensuring the efficiency and reliability of the robotic arm when performing precision tasks.

[0079] Example 6

[0080] Please refer to Figure 1 Specifically: the training evaluation module is used to pre-set the evaluation threshold K, and compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model. The specific contents are as follows:

[0081] If the training evaluation index Zpg ≥ the evaluation threshold K, it means that the training effect of the current robot arm control model meets the robot arm control accuracy requirement, and the first training feedback result is obtained, indicating that the loaded condition data when the current trained control model executes the simulation program is correct, and the loaded condition data is recorded, and the data is imported into the robot arm control platform to perform precision assembly of the production line;

[0082] If the training evaluation index Zpg is less than the evaluation threshold K, it means that the training effect of the current robotic arm control model does not meet the robotic arm control accuracy requirements, and the second training feedback result is obtained, indicating that there are errors in the loaded condition data when the currently trained control model executes the simulation program, including the rotation angles of each joint, the connection distances between adjacent joints, the displacement distances between adjacent joints and the torsion angles of adjacent joints. The loading condition data in the parameter acquisition module continues to be adjusted, and the training and analysis process of the original control model by the model training and analysis module is executed until the first training feedback result is obtained.

[0083] In this embodiment, by presetting the evaluation threshold K and comparing the training evaluation index Zpg, the training effect and reliability of the actual application of the robot arm control model are improved. When the training evaluation index Zpg reaches or exceeds the threshold K, it indicates that the control model meets the accuracy requirements, so that it can smoothly enter the precision assembly stage of the production line, ensuring the improvement of production efficiency and quality. This instant feedback mechanism not only ensures the transparency and traceability of the training process, but also provides a practical basis for optimizing the loading condition data, which is helpful to continuously improve the control strategy. On the contrary, if the training evaluation index Zpg does not meet the standard, the system can quickly identify potential problems, adjust the loading conditions in time, and ensure the accuracy of each parameter, thereby avoiding errors and risks in actual operations. This closed-loop feedback design makes the training effect of the robot arm control model perfect in continuous adjustment and optimization.

[0084] Example 7

[0085] Please refer to Figure 1 and Figure 2 ,Specifically: A training method for a robotic arm control model includes the following steps;

[0086] Step 1: Use computer simulation software to construct the robotic arm to be simulated, and integrate it with the robotic arm three-dimensional model through a controller to generate a visual and dynamically responsive three-dimensional model of the robotic arm, and also divide the three-dimensional model of the robotic arm into joints to generate a number of joint units;

[0087] Step 2: Collect the characteristic data information of several joint units, obtain the characteristic data information, and execute the simulation program on the three-dimensional model of the robotic arm through the pre-set loading condition data, monitor the position state of the end of the robotic arm after several simulations, so as to obtain the relevant end position data information;

[0088] Step 3: The robot control prediction model is obtained by training based on the recurrent neural network technology, and the relevant terminal position data information is extracted and analyzed to construct several groups of spatial position deviation values ​​Pdb of the terminal. According to the numerical values ​​of the spatial position deviation values ​​Pdb of several groups of terminals, the target qualified area of ​​the terminal position of the robot in the three-dimensional model of the robot is determined, and an optimization training analysis instruction is issued. After receiving the optimization training analysis instruction, the relevant terminal position data information obtained after several simulations is analyzed for the degree of aggregation, and the terminal position aggregation index Pdj is obtained, which is compared with the mean value of the spatial position deviation. The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model;

[0089] Step 4: pre-set the evaluation threshold K, compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model and take corresponding operations.

[0090] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A training system for a robot arm control model, characterized in that: It includes 3D model building module, parameter acquisition module, model training analysis module and training evaluation module; The three-dimensional model building module is used to build the mechanical arm to be simulated using computer simulation software, and integrate it with the three-dimensional model of the mechanical arm through the controller to generate a visual and dynamically responsive three-dimensional model of the mechanical arm, and is also used to divide the three-dimensional model of the mechanical arm into joints to generate a plurality of joint units; The parameter acquisition module is used to collect characteristic data information of several joint units, obtain the characteristic data information, and execute a simulation program on the three-dimensional model of the robotic arm through pre-set loading condition data, and monitor the position state of the end of the robotic arm after several simulations to obtain relevant end position data information; The model training and analysis module obtains the robot control prediction model based on the recurrent neural network technology training, extracts and analyzes the features of the relevant terminal position data information to construct several groups of terminal spatial position deviation values ​​Pdb, and determines the target qualified area of ​​the robot terminal position in the three-dimensional model of the robot according to the numerical values ​​of the spatial position deviation values ​​Pdb of several groups of terminals, and issues an optimization training and analysis instruction. After receiving the optimization training and analysis instruction, the relevant terminal position data information obtained after several simulations is analyzed for the degree of aggregation, and the terminal position aggregation index Pdj is obtained, which is compared with the spatial position deviation mean. The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model; The model training analysis module includes a model training unit, a preliminary analysis unit, an optimization training analysis unit and a training comprehensive evaluation unit; The optimization training analysis unit is used to perform a clustering degree analysis on the relevant terminal position data information obtained after several simulations after receiving the optimization training analysis instruction, and obtain the terminal position clustering index Pdj, wherein the terminal position clustering index Pdj is obtained by the following formula; ； Reflects the distribution characteristics of the terminal spatial position, indicating the concentrated area and distribution form of the terminal spatial position points. In the formula, It is represented as the spatial position deviation value after the i-th simulation, where i=1, 2, 3, ..., n, and n represents the number of simulations; The training comprehensive evaluation unit is used to obtain the terminal position aggregation index Pdj and the spatial position deviation mean The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model. The training evaluation index Zpg is obtained by the following formula: ； In the formula, and Respectively represented as the terminal position aggregation index Pdj and the mean of spatial position deviation The weight value of , A is represented as the correction constant; The training evaluation module is used to pre-set an evaluation threshold K, and compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model and take corresponding operations; The training evaluation module is used to pre-set the evaluation threshold K, and compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model. The specific contents are as follows: If the training evaluation index Zpg ≥ the evaluation threshold K, it means that the training effect of the current robot arm control model meets the robot arm control accuracy requirement, and the first training feedback result is obtained, indicating that the loaded condition data when the current trained control model executes the simulation program is correct, and the loaded condition data is recorded, and the data is imported into the robot arm control platform to perform precision assembly of the production line; If the training evaluation index Zpg is less than the evaluation threshold K, it means that the training effect of the current robotic arm control model does not meet the robotic arm control accuracy requirements, and the second training feedback result is obtained, indicating that there are errors in the loaded condition data when the currently trained control model executes the simulation program, including the rotation angles of each joint, the connection distances between adjacent joints, the displacement distances between adjacent joints and the torsion angles of adjacent joints. The loading condition data in the parameter acquisition module continues to be adjusted, and the training and analysis process of the original control model by the model training and analysis module is executed until the first training feedback result is obtained.

2. A training system for a robot arm control model according to claim 1, characterized in that: The three-dimensional model building module includes a simulation modeling unit and a joint division unit; The simulation modeling unit is used to construct the robot arm to be simulated in proportion according to the task requirements of assembly on the production line using the extended toolbox Simulink in the computer simulation software MATLAB, and integrate the controller with the three-dimensional model of the robot arm to control the robot arm, so as to generate a visualized and dynamically responsive three-dimensional model of the robot arm, and then check the integrity and connection reliability of the components of the three-dimensional model of the robot arm, and the specific inspection content includes whether there are missing components and unconnected components in the three-dimensional model of the robot arm; The joint division unit is used to identify the overall structure of the robotic arm of the three-dimensional model of the robotic arm. According to the identified joint motion types, kinematic characteristics and connection methods of adjacent links of each joint, the three-dimensional model of the robotic arm is divided into units, each unit corresponding to a joint, to generate a number of independent joint units.

3. A training system for a robot arm control model according to claim 2, characterized in that: The parameter acquisition module includes a parameter acquisition unit and a simulation monitoring unit; The parameter acquisition unit is used to collect characteristic data information of a plurality of joint units, and execute a simulation program on the three-dimensional model of the robotic arm through pre-set loading condition data, wherein the characteristic data information includes the joint numbers, joint lengths and initial joint angles of the plurality of joint units, and the loading condition data includes the rotation angles of each joint, the adjacent joint connection distances, the adjacent joint displacement distances and the adjacent joint torsion angles; The simulation monitoring unit monitors the position state of the end of the manipulator in real time during several simulations according to the preset loading condition data to obtain relevant end position data information, wherein the relevant end position data information includes the end target space position and the actual spatial position of the end .

4. A training system for a robot arm control model according to claim 3, characterized in that: The model training analysis module includes a model training unit, a preliminary analysis unit, an optimization training analysis unit and a training comprehensive evaluation unit; The model training unit is used to construct an original control model using recurrent neural network technology, and train and test the original control model with its characteristic data information and related terminal position data information, in combination with a predetermined target spatial position, and use the trained original control model as a terminal control state recognition model to obtain spatial positioning information within the terminal control state recognition model, and use the obtained spatial positioning information to train and test the terminal control state recognition model through a homogeneous transformation matrix algorithm, and then use the trained terminal control state recognition model as a robotic arm control prediction model in combination with training optimization analysis instructions.

5. A training system for a robot arm control model according to claim 4, characterized in that: The preliminary analysis unit is used to analyze the terminal target spatial position according to the relevant terminal position data information. and the actual spatial position of the end Analysis is performed to construct the spatial position deviation value Pdb of several groups of terminals, with the spatial position deviation value after the i-th simulation For example, it is obtained in the following way: ； In the formula, It is expressed as the spatial position deviation value after the i-th simulation, It is represented as the X-axis coordinate value of the actual spatial position of the terminal after the i-th simulation. It is represented as the Y-axis coordinate value of the actual spatial position after the i-th simulation. It is expressed as the Z-axis coordinate value of the actual spatial position of the end after the i-th simulation. It is expressed as the X-axis coordinate value of the end target space position. It is expressed as the Y-axis coordinate value of the end target space position. Expressed as the Z-axis coordinate value of the end target space position.

6. A training system for a robot arm control model according to claim 5, characterized in that: The spatial position deviation values ​​Pdb of several groups of terminals are compared with the preset deviation threshold K. When the spatial position deviation value Pdb does not exceed the deviation threshold K, the corresponding actual spatial position of the terminal is All of them are marked as target qualified areas. When the number of position points in the target qualified area exceeds 95% of the total number of position points, an optimization training analysis instruction is issued. According to the spatial position deviation value obtained after the i-th simulation The spatial position deviation values ​​Pdb of several groups of terminals are obtained respectively, and the mean value of the spatial position deviation is calculated based on the statistical mean value algorithm. .

7. A training method for a robot arm control model, used to implement a training system for a robot arm control model as described in any one of claims 1 to 6, characterized in that: The steps include: Step 1: Use computer simulation software to construct the robotic arm to be simulated, and integrate it with the robotic arm three-dimensional model through a controller to generate a visual and dynamically responsive three-dimensional model of the robotic arm, and also divide the three-dimensional model of the robotic arm into joints to generate a number of joint units; Step 2: Collect the characteristic data information of several joint units, obtain the characteristic data information, and execute the simulation program on the three-dimensional model of the robotic arm through the pre-set loading condition data, monitor the position state of the end of the robotic arm after several simulations, so as to obtain the relevant end position data information; Step 3: The robot control prediction model is obtained by training based on the recurrent neural network technology, and the relevant terminal position data information is extracted and analyzed to construct several groups of spatial position deviation values ​​Pdb of the terminal. According to the numerical values ​​of the spatial position deviation values ​​Pdb of several groups of terminals, the target qualified area of ​​the terminal position of the robot in the three-dimensional model of the robot is determined, and an optimization training analysis instruction is issued. After receiving the optimization training analysis instruction, the relevant terminal position data information obtained after several simulations is analyzed for the degree of aggregation, and the terminal position aggregation index Pdj is obtained, which is compared with the mean value of the spatial position deviation. The training evaluation index Zpg is obtained by fitting and combining with the robot arm control prediction model; Step 4: pre-set the evaluation threshold K, compare and analyze the training evaluation index Zpg with the evaluation threshold K to comprehensively determine the training effect of the robot arm control model and take corresponding operations.

Images