Strong impact load operation robot tail end vibration suppression method

Abstract

The invention discloses a strong impact load operation robot tail end vibration suppression method, which comprises the following steps: taking transmission mechanism data and tail end vibration measured by a combined force-visual sensor as input in an impact operation process of a robot, preprocessing, and then adopting a model driving and data driving method to establish a prediction model of the robot tail end vibration. Model predictive control (MPC) is adopted, closed-loop control is conducted on tail end vibration of the robot in each control period, active disturbance rejection control (ADRC) is introduced to restrain nonlinear discontinuous vibration generated in the impact process, and active vibration restraining in the impact process is achieved. And detecting the vibration of the tail end of the robot after suppression, and proportionally compensating suppression errors in the next corresponding impact process by using a self-learning method. And after multiple times of compensation tends to be stable, a compensation curve is recorded and called in the subsequent impact process. According to the method, the nonlinear and discontinuous vibration problems of the tail end of the impact load operation robot can be effectively solved, and the operation stability of the impact robot is improved.

Description

Technical Field

[0001] This invention belongs to the field of automation control and robotics technology, and specifically relates to a method for suppressing end-effector vibration of a robot subjected to strong impact loads. Background Technology

[0002] To address the issue of instability in front-of-furnace robots operating under strong impact loads, the nonlinear and discontinuous vibrations generated by the robot's impact end colliding with the furnace wall during operation affect the accuracy of opening and closing the furnace holes, and in severe cases, can even damage them. Corresponding reverse compensation for the end-of-furnace vibration is an effective measure to solve this problem. However, in existing compensation methods, the vibration models are mostly purely mathematical models with low calculation accuracy, and reverse compensation can actually negatively impact the system's stability.

[0003] Lin Xianzong et al. applied for an invention patent entitled "Predictive Control Method for Torque Rate Control and Vibration Suppression," proposing an MPC-based method to control the torque and torque rate in robot motion, effectively suppressing robot vibration and improving the accuracy and stability of industrial robot operation. Gong Chunyang et al. were granted an invention patent entitled "A Bidirectional Buck-Boost Converter Control Method Based on Reduced-Order Active Disturbance Rejection Strategy," proposing a bidirectional buck-boost converter control method based on an ADRC+MPC dual closed-loop structure. The method uses ADRC to obtain the current loop reference value and MPC to control the PWM duty cycle voltage adjustment, thereby improving the dynamic response capability of the bidirectional converter. Zhang Zhenbin applied for an invention patent entitled "Model Predictive Control Method Based on Neural Network," proposing a model predictive control method based on a neural network. This method predicts the weight coefficients of a permanent magnet synchronous motor through a neural network, taking into account multiple indicators such as electromagnetic torque, stator current, and system losses, to automatically adjust motor performance under different operating conditions, ensuring control accuracy and efficiency in both dynamic and steady-state conditions.

[0004] Therefore, in the vibration suppression and compensation of the end effector of a robot subjected to strong impact loads, a neural network is used to establish a vibration prediction model for the robot's end effector under the current working conditions, based on historical data from the robot's impact operation experiments. The MPC method is used to perform reverse compensation for the vibration of the robot's end effector in three degrees of freedom (forward / backward, left / right, and up / down) within each control cycle. Simultaneously, ADRC is introduced to suppress the nonlinear and discontinuous vibration disturbance caused by the collision between the robot's end effector and the furnace wall during the impact, achieving active closed-loop vibration suppression control. The compensated end effector vibration is detected, and a self-learning method is used to compensate for system errors, applying the difference in the next corresponding impact process. This allows for more accurate compensation of the end effector vibration at the corresponding moment, effectively improving the nonlinear and discontinuous vibration problem of the robot's end effector under impact loads, making the impact operation of the furnace-front robot more stable. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method for suppressing end-effector vibration of robots subjected to strong impact loads. This method improves upon existing methods for suppressing nonlinear and discontinuous end-effector vibrations during operation of robots subjected to impact loads. By combining neural networks, model predictive control, and a self-learning system, the method predicts and compensates for end-effector vibrations under different operating conditions, thereby reducing end-effector vibrations during impacts.

[0006] To address the above problems, embodiments of the present invention provide the following solutions:

[0007] A method for suppressing end-effector vibration in robots subjected to strong impact loads is mainly applied to robots performing strong impact opening operations in front of metallurgical furnaces, categorized into four operating modes based on different working conditions: industrial silicon, calcium carbide, ferrosilicon, and silicomanganese furnaces. The method establishes a predictive model for the robot's end-effector vibration under the current working condition and uses MPC+ADRC to suppress the vibration. Relevant data from the robot's transmission mechanism are collected during impact operation experiments, and the corresponding end-effector vibration measured by a force-vision sensor is used as input to establish a predictive model for the robot's end-effector vibration. Within a set control cycle, reverse compensation is performed on the vibration of the three degrees of freedom of the robot's end-effector. The compensated end-effector vibration is then detected, and a self-learning method is used to compensate for errors in the predictive model and control system, incorporating the difference into the next corresponding impact process. After compensation, the compensation curve for the suppressed impact vibration is recorded. The method is characterized by the following steps:

[0008] Step S1: Establish an impact vibration simulation and analysis model to obtain the vibration characteristics of various structures of the robot during the impact process. In the experiment, relevant data of the transmission mechanism of the robot working under strong impact loads are collected, and the corresponding end-effector vibration is measured by force-vision sensors as input. After preprocessing, a neural network is used to establish a predictive model of the robot's end-effector vibration under the current working condition, combined with model-driven and data-driven approaches.

[0009] Step S2: Set the time period for vibration suppression, and perform reverse compensation for the robot's end effector vibration with Δt as one control cycle. The compensation process is divided into movement and impact processes, as follows: Figure 2 As shown, different vibration compensation strategies are employed in the two stages.

[0010] Step S3: Within each control cycle, reverse compensation is performed on the vibrations of the robot's end effector in the forward / backward, left / right, and up / down directions during operation. In the experiment, the compensated end effector vibration is detected, and a self-learning method is used to compensate for errors in the prediction model and control system. The difference is then compensated for in the next corresponding impact process.

[0011] Step S4: After vibration compensation is completed, record the current compensation curve for suppressing impact vibration, and call it in subsequent actual production operations.

[0012] Preferably, the method for suppressing end-effector vibration of a robot subjected to strong impact loads is characterized in that step S1 is specifically implemented as follows:

[0013] Step S1.1: Establish an impact vibration simulation and analysis model. By analyzing the impact vibration dynamics simulation and the corresponding mode shape cloud diagrams of the multibody structure under the robot application scenario, and establishing the impact vibration energy flow model and the impact stress transmission-coupling-diffusion mechanism model of the robotic arm, the vibration characteristics of various structures of the robot during the impact process are obtained.

[0014] Step S1.2: Based on the vibration characteristics, a force-visual fusion sensing method is used to measure the nonlinear discontinuous vibration of the robot's end effector during operation. During the robot's movement phase, the end effector vibration is primarily measured through visual positioning; during the impact phase, visual measurements are obstructed, so a force-visual sensor fusion method is used to measure the end effector vibration. The formula for the combined measurement of the two sensors is:

[0015] Vib = α × Vib F +(1-α)×Vib V (1)

[0016] Where Vib represents the final vibration result, Vib F Vib measures vibrations using force sensors. V The vibration is measured by a visual sensor, and α is the proportion of force measurement, which can be adjusted according to the site conditions.

[0017] Step S1.3: Combine the measured vibrations with the mathematical model from Step S1.1 and relevant data from existing sensors measuring the robot during operation, using this as the training set for neural network training. Perform data preprocessing and dimensionality reduction on the data used, and then employ the neural network to establish vibration prediction models for the robot's end effector under different impact scenarios.

[0018] Preferably, step S1.3 is specifically implemented as follows:

[0019] Step S1.3.1: Collect and organize the data required to establish the end-effector vibration prediction model for the robot subjected to strong impact loads. The data is first preprocessed. Since subsequent control operates on a control cycle of Δt, data alignment is required for sensors with different sampling periods. For sampling periods faster than Δt, an averaging method is used, and the correction formula is as follows:

[0020]

[0021] Where n is the total number of data collected by the sensor within Δt, Data i For the collected data, Data new This is the converted data.

[0022] For sensors with sampling periods slower than Δt, data augmentation is used for data alignment. After data alignment, missing values ​​are filled and outliers are corrected. Then, principal component analysis is used to reduce the dimensionality of the high-dimensional input data, reducing the data dimensionality while retaining important information and accelerating model building time.

[0023] Step S1.3.2: Shift the prediction results used for training the input model by one data point to predict the robot's end effector vibration in advance, reducing the execution delay of the control system. The shifting method is as follows: Figure 3 As shown.

[0024] Step S1.3.3: Divide the processed data into four categories according to the four operating methods of industrial silicon, calcium carbide, ferrosilicon and ferromanganese furnace, and divide them into training set, test set and validation set. Use XGBoost for supervised learning to establish robot end-effector vibration prediction models.

[0025] Preferably, the method for suppressing end-effector vibration of a robot subjected to strong impact loads is characterized in that the moving process in step S2 refers to the process of the impact robot accelerating towards the impact surface, during which vibration in three directions is mainly controlled by coarse adjustment. The impact process involves the robot's centering and eye-opening, and repeated drilling to enlarge the hole, requiring precise control and adjustment of vibration in the forward and backward directions.

[0026] Preferably, step S3 is specifically implemented as follows:

[0027] Step S3.1: Based on the output of the robot end-effector vibration prediction model, the MPC method is used to perform reverse compensation for the vibration in three directions of the robot end during operation, and ADRC is used to suppress the nonlinear and discontinuous vibration disturbance caused by the collision between the robot end and the furnace wall during the impact process.

[0028] Step S3.2: Design corresponding control strategies based on the different drive methods of the robot operating under strong impact loads, and compensate for the vibration offset in the robot's movement, impact, pitch, and rotation actuators to suppress vibration. The drive sources are mainly hydraulic and electric motor drives. The compensation method involves a secondary server calculating the vibration to be suppressed and transmitting it to the PLC control unit. The compensation value is then used as an additional input to the program variables of the PLC responsible for controlling the corresponding degrees of freedom.

[0029] Step S3.3: After vibration compensation, the robot's end effector vibration is detected again. Self-learning is used to compensate for the errors in the prediction model and the execution errors of the control system. Based on the detected vibration of the robot's end effector after vibration compensation, the error of vibration compensation for the three degrees of freedom is calculated, and the compensation curve at the corresponding moment of the next impact is corrected. The correction formula is as follows:

[0030]

[0031] Where, N x_new N y_new and N z_new N represents the vibration compensation value after self-learning optimization for the three degrees of freedom; x_old N y_old and N z_old N represents the vibration compensation amount for the three degrees of freedom before optimization. x_dev N y_dev and N z_dev This is the error compensation for vibration in three degrees of freedom.

[0032] Preferably, the self-learning method described in step S3.3 adjusts the vibration suppression curve in the experiment. To prevent excessive fluctuations in the transmission system caused by directly compensating for all differences, which would affect the stable operation of the robot, a compensation coefficient n is set in the formula. A progressive compensation strategy is adopted, with n set at 15% in the coarse adjustment stage and 25% in the fine adjustment stage. This ratio can be adjusted according to actual production. In actual operation, environmental factors can affect the accuracy of visual vibration measurement. Therefore, the compensation coefficient in formula (3) is reduced to 10% to reduce the impact of measurement errors on the control system.

[0033] Preferably, the method for suppressing end-effector vibration of a robot subjected to strong impact loads is characterized in that, in step S4, the compensation curve for recording the current impact vibration is performed by performing multiple reverse compensations proportionally on the end-effector vibration of the impact robot. When the envelope of the vibration in the three directions is less than Δm, the vibration compensation curve is recorded in the operation scenario data of the corresponding model and can be directly called for vibration suppression during subsequent production operations.

[0034] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0035] In this embodiment of the invention, a neural network is used to establish a predictive model for the robot's end effector vibration under the current working condition, based on historical data from the robot's impact operation experiments. MPC+ADRC is used to control the robot's end effector, performing reverse compensation for vibration deviations in the robot's three degrees of freedom (forward / backward, left / right, and up / down) to actively suppress nonlinear and discontinuous end effector vibrations. The compensated end effector vibration is detected, and a self-learning method is used to compensate for errors in the predictive model and control system, incorporating the difference into the next corresponding impact process. This invention effectively improves the nonlinear and discontinuous vibration problem of the robot's end effector under impact load operations, enhancing the stability of robot production operations. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating a method for suppressing end-effector vibration of a robot subjected to strong impact loads according to the present invention.

[0038] Figure 2 This diagram illustrates the different vibration compensation strategies employed by the robot during two phases of its operation. In the diagram, the horizontal axis represents the robot's operation time, t0 to t1 represents the robot's movement process, and t1 to t2 represents the robot's impact process.

[0039] Figure 3 This is a schematic diagram of the input data target value offset method for the vibration prediction model of an impact robot. In the diagram, the fourth row is the normal corresponding vibration target value, and the fifth row is the vibration target value after forward offset.

[0040] Figure 4 These are vibration suppression schemes designed for impact robots based on their specific operating conditions. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. In order to make the objectives, technical solutions and advantages of the present invention clearer, the preferred embodiments described below are only examples to further describe the present invention in detail. Other obvious variations can be conceived by those skilled in the art. The basic principles of the present invention defined in the following description can be applied to other implementation schemes, modifications, equivalent schemes and other counting schemes that do not depart from the spirit and scope of the present invention. The specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] This invention provides a method for suppressing end-effector vibration in robots subjected to strong impact loads. For example... Figure 1 The diagram shown is a flowchart of the method in an embodiment of the present invention. The method combines neural networks, model predictive control, and a self-learning system to predict the end-effector vibration under different working conditions and to perform reverse compensation for the vibration of the robot's end-effector in three degrees of freedom: forward / backward, left / right, and up / down.

[0043] In this embodiment, the control period Δt is 2ms, the force measurement ratio α is 70%, and the envelope circle Δm is 3cm. Regarding the power source, the robot's forward and backward movement, pitch, and rotation are controlled by hydraulic drive, while the robot's chisel impact is controlled by electric motor drive. In this embodiment, the method specifically includes the following steps:

[0044] Step S1: Establish and analyze the impact vibration simulation model. Analyze the impact vibration dynamics simulation and mode shape cloud maps of the corresponding multi-body structure under robot application scenarios. Establish the impact vibration energy flow model and the impact stress transmission-coupling-diffusion mechanism model of the robotic arm to obtain the vibration characteristics of various structures of the robot during impact. Based on the vibration characteristics, use a force-vision fusion perception method to measure the nonlinear discontinuous vibration of the robot's end effector during operation. During the robot's movement phase, the end effector vibration is mainly measured through visual positioning; during the impact phase, visual measurement is obstructed, so a force-vision sensor joint measurement of the end effector vibration is used. The formula for the joint measurement of the two sensors is:

[0045] Vib = α × Vib F +(1-α)×Vib V (1)

[0046] Where Vib represents the final vibration result, Vib F Vib measures vibrations using force sensors. V The vibration is measured by the visual sensor, and α is the proportion of force measurement, which is taken as 60% to 75% in this embodiment.

[0047] In the experiment, relevant data of the transmission mechanism of the robot under strong impact loads were collected during operation. The corresponding end-effector vibrations were measured using a force-vision sensor and used as the training set input data for network training, followed by preprocessing. Since subsequent control operates on a 2ms control cycle, data alignment is required for sensors with different sampling periods. For sampling periods faster than 2ms, an averaging method is used, and the correction formula is as follows:

[0048]

[0049] Where n is the total number of data collected by the sensor within 2ms, and Data i For the collected data, Data new This is the converted data.

[0050] For sensors with sampling periods slower than 2ms, data augmentation is used for data alignment. After data alignment, missing values ​​are filled and outliers are corrected. Then, principal component analysis is used to reduce the dimensionality of the high-dimensional input data. The prediction results used to train the input model are then shifted one data point earlier to predict the robot's end effector vibration, reducing the execution delay of the control system. The shifting method is as follows: Figure 3 As shown. The processed data is divided into four categories according to four operating methods: industrial silicon, calcium carbide, ferrosilicon, and ferromanganese furnaces. Training, testing, and validation sets are also defined. Supervised learning using XGBoost is employed to establish robot end-effector vibration prediction models for each category.

[0051] Step S2: Set the time period for vibration suppression, and perform reverse compensation for the robot's end effector vibration every 2ms as a control cycle. The compensation process is divided into movement and impact processes, as follows: Figure 2 As shown, different vibration compensation strategies are adopted in the two stages. During the movement process, hydraulic drive control is used in three directions, and vibration is mainly coarsely adjusted, with vibration compensation performed once every 5 to 10 control cycles. During the impact process, the main design involves motor-driven front and rear impact unit control and hydraulically driven rotation and pitch mechanism control. Vibration requires precise control and adjustment, and vibration compensation is required in each control cycle.

[0052] Step S3: Based on the results of the robot end-effector vibration prediction model, nonlinear model predictive control is used to perform reverse compensation for the vibration of the robot end-effector in the front-back, left-right, and up-down directions during operation. ADRC is introduced to suppress the nonlinear and discontinuous vibration disturbance caused by the collision between the robot end-effector and the furnace wall during the impact process.

[0053] After vibration compensation, the robot's end effector vibration is detected again. Self-learning is used to compensate for the errors in the prediction model and the execution errors of the control system. Based on the detected vibration of the robot's end effector after vibration compensation, the errors of vibration compensation for the three degrees of freedom are calculated, and the compensation curve at the time corresponding to the next impact is corrected. The correction formula is as follows:

[0054]

[0055] Where, N x_new N y_new and N z_new N represents the vibration compensation value after self-learning optimization for the three degrees of freedom; x_old N y_old and N z_old N represents the vibration compensation amount for the three degrees of freedom before optimization. x_dev N y_dev and N z_dev This is the error compensation for vibration in three degrees of freedom.

[0056] The self-learning method described above is used to adjust the vibration suppression curve in the experiment. To prevent excessive fluctuations in the transmission system caused by directly compensating for all differences, which would affect the stable operation of the robot, a compensation coefficient n is set in the formula. In the coarse adjustment control stage, n is taken as 15%, and in the fine adjustment control stage, n is taken as 25%. This ratio can be adjusted according to actual production. In actual operation, environmental factors will affect the accuracy of visual vibration measurement. Therefore, the compensation coefficient in formula (2) is reduced to 10% to reduce the impact of measurement error on the control system.

[0057] Depending on the driving source, corresponding control methods are adopted to compensate for vibration offsets in the robot's actuators responsible for movement, impact, pitch, and rotation, thereby suppressing vibration. The compensation method involves a secondary server calculating the vibration to be suppressed and transmitting it to the PLC control unit via a communication protocol. The compensation value is then used as additional input to the program variables of the PLC responsible for controlling the corresponding degrees of freedom.

[0058] Step S4: Perform multiple reverse compensations on the vibration at the end of the impact robot proportionally. When the envelope of the vibration in the three directions is less than 3cm, record the current compensation curve for suppressing impact vibration into the data of the corresponding machine model's operating scenario, and call it for control in the subsequent actual production operation.

[0059] In a preferred embodiment of the present invention, the acquisition of historical robot data during the impact operation is achieved through software that monitors production data. At each moment, the difference between the given and actual mill speed is calculated, the average value of the data for each time period is calculated, and the impact speed drop is corrected; this is implemented using PLC programming.

[0060] This invention improves the nonlinear and discontinuous vibration problem of the end effector of existing impact-loaded robots. By combining neural networks, model predictive control and a self-learning system, it predicts the end effector vibration under different working conditions and performs reverse compensation for the vibration deviation of the robot's end effector in the three degrees of freedom of front-back, left-right and up-down. This can minimize the end effector vibration during impact and improve the stability of the robot during the impact process.

[0061] For those skilled in the art, modifications can still be made to the technical solutions described in the foregoing embodiments, or some of the technical features can be replaced. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the protection scope of this invention.

Claims

1. A method for suppressing end-effector vibration of a robot subjected to strong impact loads, primarily applied to robots used for opening holes in front of metallurgical furnaces under strong impact loads. Its characteristics are: Includes the following steps: Step S1: Establish an impact vibration simulation and analysis model to obtain the vibration characteristics of various structures of the robot during the impact process. In the experiment, relevant data of the transmission mechanism of the robot working under strong impact loads are collected, and the corresponding end-effector vibration is measured by force-vision sensors as input. After preprocessing, a neural network is used to establish a predictive model of the robot's end-effector vibration under the current working condition, combined with model-driven and data-driven approaches. Step S2: Set the time period for vibration suppression, and perform reverse compensation for the robot's end effector vibration with Δt as one control cycle. The compensation process is divided into movement and impact processes, as shown in Figure 2, with different vibration compensation strategies applied in the two stages. Step S3: Within each control cycle, reverse compensation is performed on the vibrations of the robot's end effector in the forward / backward, left / right, and up / down directions during operation. In the experiment, the compensated end effector vibrations are detected, and a self-learning method is used to compensate for errors in the prediction model and control system, incorporating the difference into the next corresponding impact process. Step S4: After vibration compensation is completed, record the current compensation curve for suppressing impact vibration, and call it in subsequent actual production operations.

2. The method for suppressing end-effector vibration of a robot subjected to strong impact loads according to claim 1, characterized in that, Step S1 is specifically implemented as follows: Step S1.1: Establish an impact vibration simulation and analysis model. By analyzing the impact vibration dynamics simulation and the corresponding mode shape cloud diagrams of the multibody structure under the robot application scenario, and establishing the impact vibration energy flow model and the impact stress transmission-coupling-diffusion mechanism model of the robotic arm, the vibration characteristics of various structures of the robot during the impact process are obtained. Step S1.2: Based on the vibration characteristics, a force-visual fusion sensing method is used to measure the nonlinear discontinuous vibration of the robot's end effector during operation. During the robot's movement phase, the end effector vibration is primarily measured through visual positioning; during the impact phase, visual measurements are obstructed, so a force-visual sensor fusion method is used to measure the end effector vibration. The formula for the combined measurement of the two sensors is: Vib=α×Vib F +(1-α)×Vib V (1) Where Vib represents the final vibration result, Vib F Vib measures vibrations using force sensors. V The vibration is measured by a visual sensor, and α is the proportion of force measurement, which can be adjusted according to the site conditions. Step S1.3: Combine the measured vibrations with the mathematical model from Step S1.1 and relevant data from existing sensors measuring the robot during operation, using this as the training set for neural network training. Perform data preprocessing and dimensionality reduction on the data used, and then employ the neural network to establish vibration prediction models for the robot's end effector under different impact scenarios.

3. The method for suppressing end-effector vibration of a robot subjected to strong impact loads according to claim 1, characterized in that, Step S1.3 is specifically implemented as follows: Step S1.3.1: Collect and organize the data required to establish the end-effector vibration prediction model for the robot subjected to strong impact loads. The data is first preprocessed. Since subsequent control uses Δt as a control cycle, data alignment is required for sensors with different sampling periods. For sampling periods faster than Δt, an averaging method is used, and the correction formula is as follows: Where n is the total number of data collected by the sensor within Δt, Data i For the collected data, Data new This is the converted data. For sensors with sampling periods slower than Δt, data augmentation is used for data alignment. After data alignment, missing values ​​are filled and outliers are corrected. Then, principal component analysis is used to reduce the dimensionality of the high-dimensional input data, reducing the data dimensionality while retaining important information and accelerating model building time. Step S1.3.2: Shift the prediction results of the input model used for training by one data point to predict the robot end effector vibration in advance, thereby reducing the execution delay of the control system. Step S1.3.3: Divide the processed data into four categories according to the four operating methods of industrial silicon, calcium carbide, ferrosilicon and ferromanganese furnace, and divide them into training set, test set and validation set. Use XGBoost for supervised learning to establish robot end-effector vibration prediction models.

4. The method for suppressing end-effector vibration of a robot subjected to strong impact loads according to claim 1, characterized in that, The moving process described in step S2 refers to the process of the impact robot accelerating towards the impact surface. During this process, the vibration in the three directions is mainly coarse. The impact process involves the robot's centering and eye-opening, and repeated drilling to enlarge the hole. Precise control and adjustment of vibration in the front and rear directions are required.

5. The method for suppressing end-effector vibration of a robot subjected to strong impact loads according to claim 1, characterized in that, Step S3 is specifically implemented as follows: Step S3.1: Based on the output of the robot end-effector vibration prediction model, the MPC method is used to perform reverse compensation for the vibration in three directions of the robot end during operation, and ADRC is used to suppress the nonlinear and discontinuous vibration disturbance caused by the collision between the robot end and the furnace wall during the impact process. Step S3.2: Design corresponding control strategies based on the different drive methods of the robot operating under strong impact loads, and compensate the vibration offset to the actuators of the robot's movement, impact, pitch, and rotation to suppress the vibration. The compensation method calculates the vibration to be suppressed through a secondary server and transmits it to the PLC control unit. The compensation value is then used as an additional input to compensate the program variables of the PLC responsible for controlling the corresponding degrees of freedom. Step S3.3: After vibration compensation, the robot's end effector vibration is detected again. Self-learning is used to compensate for the errors in the prediction model and the execution errors of the control system. Based on the detected vibration of the robot's end effector after vibration compensation, the error of vibration compensation for the three degrees of freedom is calculated, and the compensation curve at the corresponding moment of the next impact is corrected. The correction formula is as follows: Where, N x_new N y_new and N z_new N represents the vibration compensation value after self-learning optimization for the three degrees of freedom; x_old N y_old and N z_old N represents the vibration compensation amount for the three degrees of freedom before optimization. x_dev N y_dev and N z_dev This is the error compensation for vibration in three degrees of freedom.

6. The method for suppressing end-effector vibration of a robot subjected to strong impact loads according to claim 5, characterized in that, The self-learning method was used to adjust the vibration suppression curve in the experiment. To prevent excessive fluctuations in the transmission system caused by directly compensating for all differences, which would affect the stable operation of the robot, a compensation coefficient n was set in the formula. A progressive compensation strategy was adopted, with n set at 15% in the coarse adjustment stage and 25% in the fine adjustment stage. This ratio can be adjusted according to actual production. In actual operation, environmental factors can affect the accuracy of visual vibration measurement. Therefore, the compensation coefficient in formula (2) was reduced to 10% to reduce the impact of measurement errors on the control system.

7. The method for suppressing end-effector vibration of a robot subjected to strong impact loads according to claim 1, characterized in that, The step S4, which records the current compensation curve for suppressing impact vibration, involves performing multiple reverse compensations on the vibration of the impact robot's end effector in a proportional manner. When the envelope of the vibration in the three directions is less than Δm, the vibration compensation curve is recorded in the corresponding machine model's operating scenario data and can be directly called up for vibration suppression during subsequent production operations.

Images