Machine Learning-Based Dynamic Precision Optimization Method for Manipulators

Abstract

The present invention relates to the field of robots, and discloses a method for optimizing the dynamic accuracy of a manipulator based on machine learning, including image data acquisition, data processing, error judgment, error analysis, error evaluation, and error adjustment feedback. By processing and feature extraction of the collected motion state data and attitude information data, analyzing the error distribution state of the extracted data, screening out the errors generated by data acquisition, and further analyzing the error data after excluding the influence of the errors generated by data acquisition, calculating a comprehensive error evaluation coefficient to adjust the accuracy of the dynamic error of the manipulator, and executing an accuracy adjustment instruction to trigger an accuracy adjustment execution operation, which is beneficial to making the dynamic accuracy optimization control of the manipulator more intelligent, and dynamically adjusting the error of the manipulator in real time according to the production task, so as to always maintain the best working state of the manipulator.

Description

Technical Field

[0001] The present invention relates to the technical field of robots, and more particularly to a method for optimizing the dynamic accuracy of a manipulator based on machine learning. Background Art

[0002] With the development of the manufacturing industry, the requirements for the accuracy and performance of manipulators are getting higher and higher. Manipulators need to accurately grasp and place objects on the production line to ensure product quality and production efficiency. However, the dynamic characteristics of manipulators, such as speed, acceleration, and inertia, etc., will affect their accuracy and precision. Machine learning algorithms can automatically learn patterns and rules from a large amount of data, so as to realize the prediction and control of complex systems. By collecting and analyzing the motion data, sensor data, and actual production data of the manipulator, the factors affecting accuracy are found, and then machine learning algorithms are used to establish a prediction model or optimize the controller;

[0003] However, the above process still has the following disadvantages:

[0004] Firstly, the existing dynamic accuracy optimization of manipulators is to collect the dynamic data of the manipulator and use machine learning algorithms to perform model prediction on a large amount of real-time collected data. However, the large amount of real-time collected data may lead to data collection errors during the collection process, and the errors caused by data collection cannot be excluded, which will cause inaccurate prediction results, thus making it impossible to optimize the dynamic accuracy of the manipulator;

[0005] Secondly, the existing manipulators lack intelligence and adaptability in terms of dynamic accuracy optimization, and cannot automatically adjust and optimize parameters according to the real-time changes of production tasks to maintain the best working state, resulting in the need to improve the accuracy optimization of the manipulator. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for optimizing the dynamic accuracy of a manipulator based on machine learning to solve the problems existing in the above background art.

[0007] The present invention provides the following technical solutions: A method for optimizing the dynamic accuracy of a manipulator based on machine learning, including:

[0008] S1: Real-time capture the motion state image of the manipulator through a vision sensor, use image processing technology to extract the motion state data and pose information data in the motion state image of the manipulator, and transmit the extracted motion state data and pose information data to S2;

[0009] S2: Process the motion state data and attitude information data through a machine learning model, extract the relevant parameters affecting the dynamic precision error of the manipulator, then analyze and calculate the position error, torque error, speed error, and motion time error through the motion state data analysis, and analyze and calculate the pitch angle error and yaw angle error through the attitude information data analysis, and transmit the extracted motion error data and attitude error data to S3;

[0010] S3: Conduct a detailed analysis of the extracted motion error data and attitude error data, analyze the motion error distribution state index and attitude error distribution state index, and further analyze the motion error peak and attitude error peak, so as to judge whether the error generation of the manipulator is related to data acquisition;

[0011] S4: Conduct further error analysis on the motion error data and attitude error data of the manipulator, analyze the motion error deviation value through the motion error data analysis, analyze the attitude error through the attitude error data analysis, and simultaneously monitor the influence of temperature on the manipulator error through the temperature influence coefficient, and transmit the analysis results to S5;

[0012] S5: Calculate the comprehensive error evaluation coefficient, adjust the precision of the dynamic error of the manipulator through the comprehensive error evaluation coefficient, and automatically send a precision adjustment instruction to S6;

[0013] S6: Trigger the precision adjustment execution operation by executing the precision adjustment instruction, automatically adjust the error of the manipulator until the adjustment result is maintained within the normal precision range and stop executing the adjustment, and transmit a precision optimization completion feedback instruction to S6.

[0014] Preferably, the acquisition method of the motion state data and attitude information data of the manipulator in S1 is as follows:

[0015] Install a vision sensor on the end effector of the manipulator, adjust the position and angle of the sensor, and perform parameter settings. Start the vision sensor to capture the motion state image of the end effector of the manipulator in real time; preprocess the collected motion state image through image processing technology, including noise removal, image enhancement, and grayscale conversion, and then use an edge detection algorithm to extract the motion state data and attitude information data in the preprocessed motion state image.

[0016] Preferably, the processing process of the motion state data and attitude information data by the machine learning model in S2 includes data cleaning, data standardization, and data integration. The machine learning model extracts the relevant parameters affecting the dynamic precision error of the manipulator, including position parameters, attitude parameters, temperature, speed, and motion time, and then calculates the errors generated by the currently collected position parameters, attitude parameters, temperature, speed, and motion time;

[0017] The position error is calculated based on the difference between the actual position coordinates (x 实际 , y 实际 , z 实际 ) of the end effector of the currently acquired manipulator and the set position coordinates (x 预设 , y 预设 , z 预设 ) to obtain the position error

[0018] The torque error is calculated based on the difference between the actual torque value q 实际 of the end effector of the currently acquired manipulator and the set torque value q 预设 to obtain the torque error Q = |q 预设 - q 实际 |;

[0019] The speed error is calculated based on the difference between the actual speed v 实际 of the end effector of the currently acquired manipulator and the preset speed v 预设 to obtain the speed error V = |v 预设 - v 实际 |;

[0020] The motion time error is calculated based on the difference between the actual motion time t 实际 of the end effector of the currently acquired manipulator and the preset motion time t 预设 to obtain the motion time error Δt = |t 预设 - t 实际 |;

[0021] The pitch angle error is calculated based on the difference between the actual pitch angle r 实际 of the end effector of the currently acquired manipulator and the preset pitch angle r 预设 to obtain the pitch angle error Δr = |r 预设 - r 实际 |;

[0022] The yaw angle error is calculated based on the error between the actual yaw angle Δγ 实际 of the end effector of the currently acquired manipulator and the preset yaw angle Δγ 预设 to obtain the yaw angle error Δγ = γ 预设 - γ 实际 |.

[0023] Preferably, in S3, the motion error distribution state index is calculated through the position error X, the torque error Q, the speed error V, and the motion time error Δt as a1, a2, a3, and a4 are the weight coefficients of position error, torque error, velocity error, and motion time error respectively; the attitude error distribution state index is calculated through the pitch angle error Δr and the yaw angle error Δγ as b1 and b2 respectively represent the weight coefficients of the pitch angle error and the yaw angle error;

[0024] The calculation formula for the peak value of the motion error calculated through multiple groups of motion error data collected in real time is E 1i represents the motion error distribution state index analyzed for the i-th data collection, represents the average value of the motion error distribution state index analyzed for n data collections, σ E1 represents the standard deviation of the motion error distribution state index analyzed for n data collections, and n represents collecting n data;

[0025] The calculation formula for the peak value of the attitude error calculated through multiple groups of attitude error data collected in real time is E 2i represents the attitude error distribution state index analyzed for the i-th data collection, represents the average value of the attitude error distribution state index analyzed for n data collections, represents the standard deviation of the attitude error distribution state index analyzed for n data collections, and n represents collecting n data;

[0026] By comparing the peak value of the motion error m1 with the first threshold M1 and the peak value of the attitude error m2 with the second threshold M2, when the peak value of the motion error m1 is less than or equal to the first threshold M1 and the peak value of the attitude error m2 is less than or equal to the second threshold M2, it is determined that the error generation of the manipulator has nothing to do with data collection, and there is no need to re-collect the data, and the motion error data and the attitude error data are transmitted to S4. When the peak value of the motion error m1 is greater than the first threshold M1 or when the peak value of the attitude error m2 is greater than the second threshold M2, it is determined that the error generation of the manipulator is related to data collection, and the data needs to be re-collected.

[0027] Preferably, S4 further analyzes the motion error data and the attitude error data imported after excluding the errors caused by data collection of the motion error and the attitude error of the manipulator, and analyzes the change trends of the current motion error and the attitude error of the manipulator;

[0028] The analysis method of the motion error deviation value is as follows:

[0029] Step S411: Calculate the position error, torque error, velocity error, and motion time error at the acquisition time points 1, 2, 3..., k,..., J;

[0030] Step S412: Comprehensively analyze the position error, torque error, velocity error, and motion time error calculated through J acquisition time points, and calculate the deviation value of the motion error. X k represents the position error at the k-th acquisition time point, Q k represents the torque error at the k-th acquisition time point, V k represents the velocity error at the k-th acquisition time point, Δt k represents the motion time error at the k-th acquisition time point;

[0031] The attitude error is calculated in real time through the pitch angle error Δr at the k-th acquisition time point and the yaw angle error Δγ at the k-th acquisition time point. The attitude error is Δr1, Δr2,..., Δr J represents the pitch angle errors at the 1st, 2nd,..., J-th acquisition time points, Δγ1, Δγ2,..., Δγ J represents the yaw angle errors at the 1st, 2nd,..., J-th acquisition time points.

[0032] Preferably, in S5, the comprehensive error evaluation coefficient is calculated by comprehensively analyzing the deviation value of the motion error, the attitude error, and the temperature influence coefficient, the error during the movement of the manipulator is evaluated and monitored in real time, and the real-time change of the error is visually displayed in the form of a graph;

[0033] The calculation formula of the comprehensive error evaluation coefficient is s = ln(1 + C) × F T , C represents the attitude error, F represents the deviation value of the motion error, T represents the temperature influence coefficient; compare the comprehensive error evaluation coefficient with the preset error threshold to determine whether the manipulator needs to be adjusted for accuracy. If the comprehensive error evaluation coefficient s is less than the preset error threshold S, the error of the manipulator is within the normal accuracy range, and it is determined that the manipulator does not need to be adjusted for accuracy. Then return to S1 to collect and analyze the motion state data and attitude information data of the manipulator again. If the comprehensive error evaluation coefficient s is equal to or greater than the preset error threshold S, the error of the manipulator exceeds the normal accuracy range, and an accuracy adjustment instruction is immediately generated automatically and transmitted to S6.

[0034] Preferably, when S6 receives the accuracy adjustment instruction, it immediately triggers the adjustment operation of the manipulator accuracy and monitors the adjustment result in real time. When it is monitored that the manipulator error is adjusted to within the normal accuracy range, the adjustment is immediately stopped, and the accuracy optimization completion is fed back to S6, and the transmission of the accuracy adjustment instruction is stopped. At the same time, a dynamic accuracy optimization report is automatically generated according to the manipulator error adjustment process and saved in the database.

[0035] The technical effects and advantages of the present invention:

[0036] The present invention extracts motion state data and attitude information data from the captured motion images of the manipulator, processes the data and extracts features from the motion state data and attitude information data through a machine learning model, thereby extracting relevant parameters affecting the dynamic precision error of the manipulator. Through the motion error distribution state index and the attitude error distribution state index, the motion error peak and the attitude error peak are further analyzed, so as to judge whether the error generation of the manipulator is related to data acquisition. By screening out the errors generated in the data acquisition process, it is beneficial to exclude the errors caused by data acquisition, making the analysis result of the error more accurate. By analyzing the motion error deviation value and the attitude error, and monitoring the influence of temperature on the manipulator error through the temperature influence coefficient, the dynamic error of the manipulator is adjusted in precision by calculating the comprehensive error evaluation coefficient, and the precision adjustment instruction is executed to trigger the precision adjustment execution operation, which is beneficial to making the intelligent optimization control of the dynamic precision of the manipulator more intelligent, and dynamically adjusting the error of the manipulator in real time according to the production task, so as to always maintain the best working state of the manipulator and further improve the dynamic precision optimization ability of the manipulator. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a method step diagram of the present invention.

[0038] Figure 2 It is a system structure block diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. In addition, the forms of the structures described in the following embodiments are merely examples, and the method for optimizing the dynamic precision of the manipulator based on machine learning involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0040] The present invention provides a method for optimizing the dynamic precision of a manipulator based on machine learning, including:

[0041] S1: The motion state image of the manipulator is captured in real time through a vision sensor, and the motion state data and attitude information data in the motion state image of the manipulator are extracted by using image processing technology, and the extracted motion state data and attitude information data are transmitted to S2.

[0042] In this embodiment, the acquisition methods of the motion state data and attitude information data of the manipulator in S1 are:

[0043] By installing a vision sensor on the end effector of the manipulator, adjusting the position and angle of the sensor, and performing parameter settings, the vision sensor is activated to capture the motion state image of the end effector of the manipulator in real time; the collected motion state image is preprocessed through image processing technology, including noise removal, image enhancement, and grayscale conversion, and then an edge detection algorithm is used to extract the motion state data and pose information data in the preprocessed motion state image.

[0044] S2: The motion state data and pose information data are processed and feature extracted through a machine learning model. The relevant parameters affecting the dynamic precision error of the manipulator are extracted. Then, through the analysis of the motion state data, the position error, torque error, speed error, and motion time error are calculated. Through the analysis of the pose information data, the pitch angle error and yaw angle error are calculated, and the extracted motion error data and pose error data are transmitted to S3.

[0045] In this embodiment, the process of processing the motion state data and pose information data by the machine learning model in S2 includes data cleaning, data standardization, and data integration. The relevant parameters affecting the dynamic precision error of the manipulator are extracted by the machine learning model, including position parameters, pose parameters, temperature, speed, and motion time. Then, the errors generated by the currently collected position parameters, pose parameters, temperature, speed, and motion time are calculated;

[0046] The position error is calculated based on the difference between the actual position coordinates (x 实际 , y 实际 , z 实际 ) of the end effector of the currently collected manipulator and the set position coordinates (x 预设 , y 预设 , z 预设 ) to obtain the position error

[0047] The torque error is calculated based on the difference between the actual torque value q 实际 of the end effector of the currently collected manipulator and the set torque value q 预设 to obtain the torque error Q = |q 预设 - q 实际 |;

[0048] The speed error is calculated based on the difference between the actual speed v 实际 of the end effector of the currently collected manipulator and the preset speed v 预设 to obtain the speed error V = |v 预设 - v 实际 |;

[0049] The motion time error is calculated based on the actual motion time t of the end effector of the current acquisition manipulator 实际 and the preset motion time t 预设 The difference between them is calculated to obtain the motion time error Δt = |t 预设 - t 实际 |;

[0050] The pitch angle error is calculated based on the actual pitch angle r of the end effector of the current acquisition manipulator 实际 and the preset pitch angle r 预设 The difference between them is calculated to obtain the pitch angle error Δr = |r 预设 - r 实际 |;

[0051] The yaw angle error is calculated based on the actual yaw angle Δγ of the end effector of the current acquisition manipulator 实际 and the preset yaw angle Δγ 预设 The error between them is calculated to obtain the yaw angle error Δγ = γ 预设 - γ 实际 |.

[0052] S3: By analyzing the extracted motion error data and attitude error data in detail, analyzing the motion error distribution state index and the attitude error distribution state index, and further analyzing the motion error peak and the attitude error peak, so as to judge whether the error generation of the manipulator is related to data acquisition.

[0053] In this embodiment, S3 calculates the motion error distribution state index through the position error X, the torque error Q, the velocity error V, and the motion time error Δt as a1, a2, a3, a4 are the weight coefficients of the position error, the torque error, the velocity error, and the motion time error respectively; the attitude error distribution state index is calculated through the pitch angle error Δr and the yaw angle error Δγ as b1 and b2 respectively represent the weight coefficients of the pitch angle error and the yaw angle error;

[0054] The formula for calculating the motion error peak through multiple groups of motion error data collected in real time is E 1i represents the motion error distribution state index analyzed in the i-th data acquisition, represents the average value of the motion error distribution state index analyzed in n data acquisitions, σ E1 represents the standard deviation of the motion error distribution state index analyzed in n data acquisitions, and n represents collecting n data;

[0055] The formula for calculating the attitude error peak through multiple groups of attitude error data collected in real time is E2i Denotes the attitude error distribution state index analyzed for the i-th data acquisition, Denotes the average value of the attitude error distribution state indices analyzed for n data acquisitions, Denotes the standard deviation of the attitude error distribution state indices analyzed for n data acquisitions, where n represents the number of data acquisitions;

[0056] By comparing the peak motion error m1 with the first threshold M1 and the peak attitude error m2 with the second threshold M2, when the peak motion error m1 is less than or equal to the first threshold M1 and the peak attitude error m2 is less than or equal to the second threshold M2, it is determined that the error generation of the manipulator has nothing to do with data acquisition, there is no need to re-acquire the data, and the motion error data and attitude error data are transmitted to S4. When the peak motion error m1 is greater than the first threshold M1 or when the peak attitude error m2 is greater than the second threshold M2, it is determined that the error generation of the manipulator is related to data acquisition, and the data needs to be re-acquired.

[0057] Specifically, the calculation formula for the standard deviation of the motion error distribution state index analyzed for n data acquisitions is The calculation formula for the standard deviation of the attitude error distribution state index analyzed for n data acquisitions is

[0058] Judgment method of the error distribution state:

[0059] When the peak motion error m1 is less than or equal to the first threshold M1, it indicates that the distribution state of the motion error is flat and not an error caused by the data acquisition process. When the peak motion error m1 is greater than the first threshold M1, it indicates that the motion error is caused by improper data acquisition. When the peak attitude error m2 is less than or equal to the second threshold M2, it indicates that the distribution state of the attitude error is flat and not an error caused by the data acquisition process. When the peak attitude error m2 is greater than the second threshold M2, it indicates that the attitude error is caused by improper data acquisition.

[0060] S4: Through further error analysis of the motion error data and attitude error data of the manipulator, the motion error deviation value is analyzed from the motion error data, the attitude error is analyzed from the attitude error data, and at the same time, the influence of temperature on the manipulator error is monitored through the temperature influence coefficient, and the analysis results are transmitted to S5.

[0061] In this embodiment, S4 performs further error analysis on the motion error data and attitude error data imported after excluding the errors caused by data acquisition of the motion error and attitude error of the manipulator, and analyzes the change trend of the current motion error and attitude error of the manipulator;

[0062] The analysis method of the motion error deviation value is:

[0063] Step S411: Calculate the position error, torque error, velocity error, and motion time error at the 1st, 2nd, 3rd, …, kth, …, Jth acquisition time points;

[0064] Step S412: Through comprehensive analysis of the position error, torque error, velocity error, and motion time error calculated respectively at the J acquisition time points, calculate the motion error deviation value X k represents the position error at the kth acquisition time point, Q k represents the torque error at the kth acquisition time point, V k represents the velocity error at the kth acquisition time point, Δt k represents the motion time error at the kth acquisition time point;

[0065] The attitude error is calculated in real time through the pitch angle error Δr at the kth acquisition time point and the yaw angle error Δγ at the kth acquisition time point, and the attitude error is Δr1, Δr2, …, Δr J represents the pitch angle errors at the 1st, 2nd, …, Jth acquisition time points, Δγ1, Δγ2, …, Δγ J represents the yaw angle errors at the 1st, 2nd, …, Jth acquisition time points.

[0066] Specifically, the temperature influence coefficient is obtained by respectively collecting the temperatures of the end effector of the manipulator at the J acquisition time points, and then calculating the temperature influence coefficient formula as T k represents the temperature of the end effector of the manipulator at the kth acquisition time point, represents the average value of the temperatures of the end effector of the manipulator at the J acquisition time points.

[0067] S5: Calculate the comprehensive error evaluation coefficient, adjust the accuracy of the dynamic error of the manipulator through the comprehensive error evaluation coefficient, and automatically send an accuracy adjustment instruction to S6.

[0068] In this embodiment, S5 calculates the comprehensive error evaluation coefficient through the analyzed motion error deviation value, attitude error, and temperature influence coefficient, evaluates and monitors the error during the motion of the manipulator in real time, and visually displays the real-time change of the error in the form of a graph;

[0069] The calculation formula of the comprehensive error evaluation coefficient is s = ln(1 + C) × F T, C represents the attitude error, F represents the deviation value of the motion error, and T represents the temperature influence coefficient; compare the comprehensive error evaluation coefficient with the preset error threshold to determine whether the manipulator needs to be adjusted for accuracy. If the comprehensive error evaluation coefficient s is less than the preset error threshold S, the error of the manipulator is within the normal accuracy range, and it is determined that the manipulator does not need to be adjusted for accuracy. Return to S1 to collect and analyze the motion state data and attitude information data of the manipulator again. If the comprehensive error evaluation coefficient s is equal to or greater than the preset error threshold S, the error of the manipulator exceeds the normal accuracy range, and an accuracy adjustment instruction is immediately generated automatically and transmitted to S6.

[0070] S6: By executing the accuracy adjustment instruction, trigger the accuracy adjustment execution operation, and automatically adjust the error of the manipulator until the adjustment result is maintained within the normal accuracy range and stop executing the adjustment, and transmit an accuracy optimization completion feedback instruction to S6.

[0071] In this embodiment, when S6 receives the accuracy adjustment instruction, it immediately triggers the adjustment operation of the manipulator's accuracy and monitors the adjustment result in real time. When it is detected that the manipulator error is adjusted to within the normal accuracy range, the adjustment is immediately stopped, and the accuracy optimization is completed and fed back to S6, and the transmission of the accuracy adjustment instruction is stopped. At the same time, a dynamic accuracy optimization report is automatically generated according to the manipulator error adjustment process and saved to the database.

[0072] As Figure 2 shown, the implementation system corresponding to the dynamic accuracy optimization method of the manipulator based on machine learning provided in this embodiment includes an image data acquisition module, a data processing module, an error judgment module, an error analysis module, an error evaluation module, and an error adjustment feedback module. The image data acquisition module is connected to the data processing module, the data processing module is connected to the error judgment module, the error judgment module is connected to the error analysis module, the error analysis module is connected to the error evaluation module, and the error evaluation module is connected to the error adjustment feedback module.

[0073] The image data acquisition module captures the motion state image of the manipulator in real time through a vision sensor, and uses image processing technology to extract the motion state data and attitude information data in the motion state image of the manipulator;

[0074] The data processing module performs data processing and feature extraction on the motion state data and attitude information data through a machine learning model, extracts the relevant parameters that affect the dynamic accuracy error of the manipulator, and then analyzes and calculates the position error, torque error, speed error, and motion time error through the motion state data analysis, and calculates the pitch angle error and yaw angle error through the attitude information data analysis;

[0075] The error judgment module analyzes the extracted motion error data and attitude error data in detail, analyzes the motion error distribution state index and the attitude error distribution state index, and further analyzes the motion error peak value and the attitude error peak value, so as to judge whether the error generation of the manipulator is related to data acquisition;

[0076] The error analysis module further analyzes the motion error data and attitude error data of the manipulator, analyzes the motion error deviation value through the motion error data analysis, analyzes the attitude error through the attitude error data analysis, and monitors the influence of temperature on the manipulator error through the temperature influence coefficient at the same time;

[0077] The error evaluation module calculates the comprehensive error evaluation coefficient, adjusts the accuracy of the dynamic error of the manipulator through the comprehensive error evaluation coefficient, and automatically issues an accuracy adjustment instruction;

[0078] The error adjustment feedback module triggers the accuracy adjustment execution operation by executing the accuracy adjustment instruction, automatically adjusts the error of the manipulator until the adjustment result is maintained within the normal accuracy range and stops executing the adjustment.

[0079] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0080] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claimed rights.

Claims

1. A method for optimizing the dynamic accuracy of a manipulator based on machine learning, characterized in that: include: S1: Capture the motion state image of the manipulator in real time through the visual sensor, extract the motion state data and posture information data from the motion state image of the manipulator using image processing technology, and transmit the extracted motion state data and posture information data to S2; S2: Process the motion state data and attitude information data and extract features through the machine learning model, extract the relevant parameters that affect the dynamic accuracy error of the manipulator, and then analyze and calculate the position error, torque error, speed error and motion time error through the motion state data. Analyze and calculate the pitch angle error and yaw angle error through the attitude information data, and transmit the extracted motion error data and attitude error data to S3; S3: By analyzing the extracted motion error data and posture error data in detail, analyzing the motion error distribution state index and the posture error distribution state index, and further analyzing the motion error peak value and the posture error peak value, it is determined whether the error generation of the manipulator is related to data acquisition; S4: further error analysis is performed on the motion error data and attitude error data of the manipulator, the motion error deviation value is analyzed through the motion error data, the attitude error is analyzed through the attitude error data, and the influence of temperature on the manipulator error is monitored through the temperature influence coefficient, and the analysis result is transmitted to S5; S5: Calculate the comprehensive error evaluation coefficient, adjust the accuracy of the dynamic error of the manipulator through the comprehensive error evaluation coefficient, and automatically send out the accuracy adjustment command to S6; S6: By executing the precision adjustment instruction, the precision adjustment execution operation is triggered, and the error of the robot is automatically adjusted until the adjustment result is kept within the normal precision range, and the adjustment is stopped, and the precision optimization completion feedback instruction is transmitted to S6.

2. The method for optimizing the dynamic accuracy of a manipulator based on machine learning according to claim 1, characterized in that: The data collection method of the manipulator motion state data and posture information of S1 is: By installing the visual sensor on the end effector of the manipulator, adjusting the position and angle of the sensor, and setting the parameters, the visual sensor is started to capture the motion state image of the end effector of the manipulator in real time; the collected motion state image is preprocessed by image processing technology, including noise removal, image enhancement and grayscale conversion, and then the edge detection algorithm is used to extract the motion state data and posture information data in the preprocessed motion state image.

3. The method for optimizing the dynamic accuracy of a manipulator based on machine learning according to claim 1, characterized in that: The S2 processes the motion state data and posture information data through a machine learning model, including data cleaning, data standardization, and data integration. The machine learning model extracts the relevant parameters that affect the dynamic accuracy error of the manipulator, including position parameters, posture parameters, temperature, speed, and motion time, and then calculates the errors caused by the currently collected position parameters, posture parameters, temperature, speed, and motion time; The position error is based on the actual position coordinates (x 实际 ,y 实际 , z 实际 ) and the set position coordinates (x 预设 ,y 预设 , z 预设 ) is calculated to obtain the position error The torque error is based on the actual torque value q of the current acquisition manipulator end effector 实际 With the set torque value q 预设 The difference between them is calculated to obtain the torque error Q = |q 预设 -q 实际 |; The speed error is based on the actual speed v of the current acquisition manipulator end effector 实际 With the preset speed v 预设 The difference between them is calculated to obtain the speed error V = |v 预设 -v 实际 |; The motion time error is based on the actual motion time t of the end effector of the current acquisition manipulator. 实际 With the preset movement time t 预设 The difference between them is calculated to obtain the motion time error Δt=|t 预设 -t 实际 |; The pitch angle error is based on the actual pitch angle r of the current acquisition manipulator end effector. 实际 With the preset pitch angle r 预设 The difference between them is calculated to obtain the pitch angle error Δr = |r 预设 -r 实际 |; The yaw angle error is based on the actual yaw angle Δγ of the current acquisition manipulator end effector. 实际 With the preset yaw angle Δγ 预设 The error between is calculated to obtain the yaw angle error Δγ=|γ 预设 -γ 实际 |.

4. The method for optimizing the dynamic accuracy of a manipulator based on machine learning according to claim 1, characterized in that: The S3 calculates the motion error distribution state index through the position error X, the torque error Q, the speed error V and the motion time error Δt: a1, a2, a3, a4 are the weight coefficients of position error, torque error, velocity error and motion time error respectively; the attitude error distribution state index is calculated by the pitch angle error Δr and the yaw angle error Δγ: b1, b2 represent the weight coefficients of pitch angle error and yaw angle error respectively; The motion error peak is calculated by using multiple sets of motion error data collected in real time. E 1i represents the motion error distribution state index analyzed by the i-th data acquisition, It represents the average value of the motion error distribution state index analyzed by n data acquisitions, It represents the standard deviation of the motion error distribution state index analyzed by n data collections, and n represents the collection of n data; The peak value of the posture error is calculated by using multiple sets of posture error data collected in real time: E 2i represents the attitude error distribution state index analyzed by the i-th data collection, It represents the average value of the attitude error distribution state index analyzed by n data collections, It represents the standard deviation of the attitude error distribution state index analyzed by n data collections, where n represents the collection of n data; By comparing the motion error peak m1 with the first threshold M1, and comparing the posture error peak m2 with the second threshold M2, when the motion error peak m1 is less than or equal to the first threshold M1 and the posture error peak m2 is less than or equal to the second threshold M2, it is determined that the error generation of the manipulator is not related to data acquisition, and there is no need to re-collect the data, and the motion error data and the posture error data are transmitted to S4; when the motion error peak m1 is greater than the first threshold M1 or when the posture error peak m2 is greater than the second threshold M2, it is determined that the error generation of the manipulator is related to data acquisition, and the data needs to be re-collected.

5. The method for optimizing the dynamic accuracy of a manipulator based on machine learning according to claim 1, characterized in that: S4 further analyzes the change trend of the current motion error and posture error of the manipulator by performing error analysis on the motion error data and posture error data imported after eliminating the errors caused by data acquisition; The motion error deviation value is analyzed as follows: Step S411: Calculate the position error, torque error, velocity error and motion time error of 1, 2, 3, ..., k, ..., J acquisition time points; Step S412: Comprehensively analyze the position error, torque error, speed error and motion time error calculated at J acquisition time points, and calculate the motion error deviation value X k represents the position error at the kth acquisition time point, Q k represents the torque error at the kth acquisition time point, V k represents the velocity error at the kth acquisition time point, Δt k represents the motion time error at the kth acquisition time point; The attitude error is calculated in real time by the pitch angle error Δr at the kth acquisition time point and the yaw angle error Δγ at the kth acquisition time point. The attitude error is: Δr1, Δr2, …, Δr J represents the pitch angle error at 1, 2, ..., J acquisition time points, Δγ1, Δγ2, ..., Δγ J Represents the yaw angle error at 1, 2, ..., J acquisition time points.

6. The method for optimizing the dynamic accuracy of a manipulator based on machine learning according to claim 1, characterized in that: The S5 calculates the comprehensive error evaluation coefficient by analyzing the motion error deviation value, posture error and temperature influence coefficient, evaluates and monitors the error of the manipulator in real time, and visualizes the real-time change of the error in a graphical manner; The calculation formula of the comprehensive error evaluation coefficient is s=ln(1+C)×F T , C represents the attitude error, F represents the motion error deviation value, and T represents the temperature influence coefficient; The comprehensive error evaluation coefficient is compared with the preset error threshold to determine whether the robot needs to perform precision adjustment. If the comprehensive error evaluation coefficient s is less than the preset error threshold S, the error of the robot is within the normal precision range, and it is determined that the robot does not need to perform precision adjustment. The system will return to S1 to re-collect and analyze the motion state data and posture information data of the robot. If the comprehensive error evaluation coefficient s is equal to or greater than the preset error threshold S, the error of the robot exceeds the normal precision range, and a precision adjustment instruction is automatically generated and transmitted to S6.

7. The method for optimizing the dynamic accuracy of a manipulator based on machine learning according to claim 1, characterized in that: When receiving the precision adjustment instruction, S6 immediately triggers the adjustment operation of the manipulator precision and monitors the adjustment result in real time. When it is monitored that the manipulator error is adjusted to within the normal precision range, the adjustment is stopped immediately, and the precision optimization completion is fed back to S6, and the transmission of the precision adjustment instruction is stopped. At the same time, a dynamic precision optimization report is automatically generated according to the manipulator error adjustment process and saved in the database.

Images