An adaptive control method for collective doffing spindle vibration

Abstract

The present application relates to the field of industrial control technology, specifically to a self-adaptive control method for collective doffing spindle vibration, the present application firstly establishes an electromechanical combined dynamics model, optimizes the generation of non-standard flexible motion trajectory which can suppress vibration from the source, at the same time, by constructing a disturbance model library and prospectively analyzing PLC instructions, a predictive compensation signal is generated; the predictive compensation signal is used to realize double feedforward control: by analyzing its energy characteristics to dynamically adjust the servo feedforward gain of the manipulator, time-varying stiffness control is realized; then the spindle seat actuator is driven to actively damp to dissipate disturbance energy; on this basis, the system further predicts the residual vibration after compensation, and calculates the anti-waveform signal based on the inverse dynamics model to accurately offset; finally, the PLC time sequence phase-locked algorithm is used for online correction of each model, which comprehensively uses predictive control, feedforward compensation and adaptive correction, and realizes high-precision and whole-process suppression of spindle vibration.

Description

Technical Field

[0001] This invention relates to the field of industrial control technology, specifically to an adaptive control method for the vibration of a collective doffing spindle. Background Technology

[0002] In many high-speed, high-precision automated production equipment, the operating efficiency and positioning stability of their electromechanical servo systems are generally plagued by residual vibrations generated when the end effector interacts with the load. For example, the high acceleration and deceleration motion of the servo-driven actuator, the flexible deformation of its own structural links, and the physical coupling between the task point and the controlled object can all generate complex broadband vibrations, which can seriously affect the system's production cycle and the quality of the final product.

[0003] Traditional control algorithms (such as PID) are typically based on a simplified, linear, time-invariant system mathematical model, with control parameters tuned offline during equipment commissioning. However, the actual dynamic characteristics of the controlled object are affected by factors such as mechanical wear, dynamic changes in workload, and lubrication conditions, exhibiting significant nonlinearity and time-varying characteristics, leading to model parameter drift. Existing technologies generally lack a closed-loop mechanism for online identification of model parameters and real-time adaptive tuning of the control law, resulting in poor control robustness and a significant decrease in performance over time and under varying operating conditions. These traditional methods mainly rely on sensor-based hysteresis feedback compensation, representing a typical "error-driven" approach.

[0004] Therefore, an adaptive control method for the vibration of collective doffing spindles is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive control method for vibration of collective doffing spindles. By combining forward-looking prediction, multi-level active compensation, and online adaptive correction, it achieves high-precision, full-process suppression of vibration, thereby significantly improving the stability and reliability of high-speed doffing operations. To achieve the above objective, this invention provides the following technical solution: An adaptive control method for vibration of a collective doffing spindle includes: A combined electromechanical dynamics model is established, consisting of a doffing robot, a servo drive system, and a spindle-bearing-yarn tube system. Based on the model, reverse optimization calculations are performed to generate non-standard flexible motion trajectories. A disturbance model library containing residual mechanical shock and motor starting torque pulsation is created. The PLC instruction queue is synchronized and parsed in real time to capture upcoming operations and time points. The corresponding model is called from the disturbance model library to generate a predictive compensation signal. The predictive compensation signal is analyzed using an online time-series prediction model. Based on a meta-learning model, the characteristics of the system state and the predictive compensation signal are analyzed in real time to generate optimal control parameters. The predictive energy injection curve is calculated and generated to dynamically adjust the feedforward gain of the servo drive system when executing the non-standard flexible motion trajectory. The predictive compensation signal is used to drive the actuator below the spindle to perform active damping control; the residual vibration waveform after feedforward gain is predicted, and the inverse waveform drive signal to cancel the residual vibration waveform is calculated based on the system inverse dynamics model; the inverse waveform drive signal is injected into the actuator's drive amplifier before the predicted vibration occurs.

[0006] Preferably, the construction of the electromechanical joint dynamics model includes: performing finite element modeling of the mechanical structure of the doffing robot to obtain stiffness and modal characteristics; establishing a dynamic response model for the servo drive system based on the motor torque-speed characteristics and driver response delay; performing multibody dynamics modeling of the spindle-bearing-yarn tube system to extract natural frequencies and damping characteristics; integrating it into a joint simulation platform to construct a virtual prototype system that can reflect the dynamic coupling relationship between each subsystem; performing closed-loop dynamic simulation to obtain simulation results; based on the simulation results, with minimizing the vibration energy of the spindle as the optimization objective, using an optimization algorithm to iteratively optimize the motion trajectory parameters in the simulation environment until an optimal non-standard flexible motion trajectory is calculated; and compiling the non-standard flexible motion trajectory into a control program and downloading it to the programmable logic controller of the doffing robot.

[0007] Preferably, the establishment of the disturbance model library includes: collecting instantaneous data of residual mechanical impact and motor starting torque pulsation under different working conditions through offline calibration experiments, and performing frequency domain analysis on the instantaneous data to extract the characteristic frequency, amplitude and phase information of the main harmonic components constituting the disturbance signal, forming a structured disturbance characteristic parameter set; Based on the set of disturbance characteristic parameters, a parameterized disturbance model is established for each independent disturbance source, including but not limited to motor starting and robotic gripper closure. The parameterized disturbance model calls upon the corresponding set of feature parameters according to the upcoming operation command, reconstructs and predicts the disturbance waveform through Fourier series. Based on a meta-learning model, the waveform and energy characteristics of the predictive compensation signal are received in real time and combined with the current operating state to form a decision state vector. The meta-learning model performs real-time analysis on the decision state vector. Based on the analysis results, optimal control parameters are generated, and composite drive instructions are synthesized and executed. The parameterized disturbance model is modularly encapsulated, a globally unique identifier is assigned, and an index table is created and stored. The index table is used to establish a one-to-one correspondence between the unique identifier and the programmable logic controller instruction code that triggers the corresponding disturbance, and to construct a query path. The disturbance characteristic parameter set, parameterized disturbance model, and index table are stored in the disturbance model library.

[0008] Preferably, the generation of the predictive compensation signal includes: setting up a forward-looking instruction parsing module, which synchronously reads the queue of unexecuted instructions stored in the PLC in real time, identifies the target operation instruction to be executed in the queue, and predicts the future execution time based on the position of the target operation instruction in the queue and the scan cycle of the PLC; using the instruction code of the target operation instruction as an index, queries an index table to locate and call the parameterized disturbance model corresponding to the target operation instruction; generating a predicted disturbance waveform in real time based on the disturbance characteristic parameter set inside the parameterized disturbance model; and calculating an inverse waveform with the same amplitude but opposite phase as the predicted disturbance waveform based on the predicted disturbance waveform, and outputting the inverse waveform as a predictive compensation signal.

[0009] Preferably, the dynamic adjustment of the feedforward gain of the servo system when executing the non-standard flexible motion trajectory includes: setting up an energy observation module, analyzing the predictive compensation signal based on a preset system energy transfer model, calculating and generating a predicted energy injection curve describing the timing, amplitude, and power of disturbance energy injection within a future time window, and when the doffing robot's servo system tracks and executes the non-standard flexible motion trajectory, the feedforward controller synchronously receives the predicted energy injection curve, increases the feedforward gain before the predicted energy injection peak is about to arrive, and decreases the gain during the smooth period of energy injection.

[0010] Preferably, using the predictive compensation signal to drive the actuator below the spindle seat to perform active damping control includes: sending the original waveform of the predictive compensation signal as a driving command to the actuator below the spindle seat; the controller of the actuator executing an active damping control algorithm, which applies a damping force proportional to and opposite in direction to the vibration velocity based on the received command waveform and the real-time detected spindle vibration velocity.

[0011] Preferably, calculating the inverse waveform drive signal includes: using an online timing prediction model to predict the residual vibration waveform that will still exist in the future time window after the feedforward gain is dynamically adjusted; inverting the predicted residual vibration waveform to generate a target cancellation trajectory with the same amplitude but opposite phase as the residual vibration waveform; substituting the target cancellation trajectory as the motion output into the system inverse dynamics model, which is a pre-established model describing the relationship between the actuator input signal and the spindle vibration output; and calculating the inverse waveform drive signal by inverse solution.

[0012] Preferably, the inverse waveform drive signal is injected into the actuator's drive amplifier before the predicted vibration occurs; the correction of the model using a PLC timing lock-in algorithm includes: using a clock synchronization mechanism to inject the signal into the actuator's drive amplifier before the precise moment when the residual vibration occurs.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By establishing an electromechanical joint dynamic model, a non-standard flexible motion trajectory is generated. This trajectory starts from the source of motion planning, actively avoids system resonance, fundamentally reduces the internal vibration excitation, and provides a highly stable motion benchmark for the system. Compared with traditional trajectory planning, it can more effectively ensure the stability and reliability of the high-speed doffing process from the root.

[0014] 2. This invention can accurately predict future disturbances by constructing a disturbance model library and proactively analyzing the PLC instruction queue. This mechanism actively generates compensation signals before the disturbance occurs, completely overcoming the response delay of traditional feedback control. This "pre-knowledge and pre-control" mode improves the real-time performance and accuracy of control and enhances the system's anti-disturbance capability under high-speed operation.

[0015] 3. A multi-level collaborative control strategy is proposed. By utilizing the predicted signal, active damping energy dissipation and dynamic servo gain adjustment are simultaneously achieved. After this dual feedforward control, the system performs secondary prediction of residual vibration and generates an inverse waveform for final cancellation. This strategy organically combines multiple control methods to suppress vibration layer by layer, making the system's adaptability far exceed that of traditional single control methods. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of an adaptive control method for collective doffing spindle vibration according to the present invention. Figure 2 This is a flowchart illustrating an adaptive control method for collective doffing spindle vibration according to the present invention. Figure 3 This is a schematic diagram illustrating the process of constructing the electromechanical combined dynamics model of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figures 1 to 3 This invention provides an adaptive control method for the vibration of a collective doffing spindle, the technical solution of which is as follows: An adaptive control method for vibration of collective doffing spindles, referring to Figure 1 Step flowchart and Figure 2 The flowchart is as follows: A combined electromechanical dynamics model is established, consisting of a doffing robot, a servo drive system, and a spindle-bearing-yarn tube system. Based on the model, reverse optimization calculations are performed to generate non-standard flexible motion trajectories. A disturbance model library containing residual mechanical shock and motor starting torque pulsation is created. The PLC instruction queue is synchronized and parsed in real time to capture upcoming operations and time points. The corresponding model is called from the disturbance model library to generate a predictive compensation signal. The predictive compensation signal is analyzed using an online time-series prediction model. Based on a meta-learning model, the characteristics of the system state and the predictive compensation signal are analyzed in real time to generate optimal control parameters. The predictive energy injection curve is calculated and generated to dynamically adjust the feedforward gain of the servo drive system when executing the non-standard flexible motion trajectory. The predictive compensation signal is used to drive the actuator below the spindle to perform active damping control; the residual vibration waveform after feedforward gain is predicted, and the inverse waveform drive signal to cancel the residual vibration waveform is calculated based on the system inverse dynamics model; the inverse waveform drive signal is injected into the actuator's drive amplifier before the predicted vibration occurs.

[0019] Example 1: This embodiment is set as a large spinning mill with multiple high-speed ring spinning production lines, each equipped with 200 spindles, and using a centralized doffing robot for automatic doffing. Due to high-speed operation and frequent doffing operations, spindle vibration is a serious problem, affecting yarn quality and equipment life. The adaptive control method of this application is adopted to reduce spindle vibration caused by doffing operations during the spinning process.

[0020] First, an electromechanical joint dynamics model is established and a non-standard flexible motion trajectory is generated, referring to... Figure 3This is a schematic diagram of the process for constructing the electromechanical co-dynamic model of the present invention. The construction of the electromechanical co-dynamic model includes: performing finite element modeling of the mechanical structure of the doffing robot to obtain stiffness and modal characteristics; specifically, using ANSYS Workbench software, the doffing robot, including the robotic arm, gripping mechanism, transmission components, etc., is modeled in three dimensions; the model contains 100,000 tetrahedral elements, and the material properties are set as steel, i.e., Young's modulus 200 GPa, Poisson's ratio 0.3, and density 7850 kg / m³; modal analysis is performed to obtain the first six natural frequencies and mode shapes; for example, the first natural frequency is 15 Hz, mainly manifested as bending vibration of the robotic arm; the second natural frequency is 25 Hz, manifested as torsional vibration of the gripping mechanism; at the same time, static stiffness analysis is performed to determine that the equivalent stiffness of the end of the robotic arm is 10^6 N / m.

[0021] Based on the motor torque-speed characteristics and driver response delay, a dynamic response model is established for the servo drive system. Specifically, MATLAB / Simulink is used to build the servo drive system model for the doffing robot. The servo motor is a permanent magnet synchronous motor with a rated power of 5kW, a rated torque of 20Nm, and a rated speed of 3000rpm. The driver adopts a three-loop control structure, including a current loop, a speed loop, and a position loop, and its response delay is determined experimentally. The response time of the current loop is 0.5ms, the response time of the speed loop is 5ms, and the response time of the position loop is 10ms. The motor torque-speed characteristics are obtained by fitting measured data. For example, in the range of 0-3000rpm, the torque decreases linearly with the speed, and the slope of the decrease is 0.005Nm / rpm.

[0022] Multibody dynamics modeling was performed on the spindle-bearing-yarn tube system to extract its natural frequencies and damping characteristics. Specifically, ADAMS software was used to model a single spindle-bearing-yarn tube system. The spindle model was 500 mm long and 10 mm in diameter, made of high-carbon steel. The bearing was a rolling bearing model with a stiffness coefficient of 10^8 N / m and a damping coefficient of 500 Ns / m. The yarn tube was a hollow cylinder model with a length of 200 mm, a diameter of 40 mm, and a mass of 50 g. Modal analysis was used to extract the first two natural frequencies of the spindle-bearing-yarn tube system. For example, the first natural frequency was 80 Hz, mainly representing the coupled bending vibration of the spindle and yarn tube; the second natural frequency was 150 Hz, representing the higher-order bending vibration of the spindle. The damping ratio was 0.02 based on experimental modal analysis.

[0023] By identifying and quantifying the vibration transmission influence matrix between each spindle through offline system identification, pre-coordinated feedforward control and post-coordinated feedback control are executed. Specifically, the action of a single spindle within a future window is predicted, the vibration transmission influence matrix is ​​queried, and a compensation strategy is calculated. Vibration sensors are deployed on the spindles to detect residual vibration information. When the residual vibration information is too large, the active damping strategy of adjacent spindles is adjusted to help offset the residual vibration information. The simulation results are obtained by integrating the simulation into a co-simulation platform to construct a virtual prototype system that reflects the dynamic coupling relationship between each subsystem. Closed-loop dynamic simulation is performed to obtain simulation results. Specifically, the finite element model, servo drive model, and multibody dynamics model established above are imported into the MSC.AdamsMechatronics co-simulation platform. In this platform, the robot arm is set to complete a doffing cycle within 0.5 seconds, including actions such as grabbing the yarn tube, lifting, moving, putting down the empty tube, and resetting. 100 closed-loop dynamic simulations are performed, each simulation simulating a complete doffing process, and the vibration displacement and acceleration data of the spindle during the doffing process are recorded.

[0024] Based on the simulation results, with minimizing the spindle vibration energy as the optimization objective, an optimization algorithm is used to iteratively optimize the motion trajectory parameters in the simulation environment until an optimal non-standard flexible motion trajectory is calculated. Specifically, the spindle vibration acceleration data obtained from the simulation is integrated and squared to calculate the spindle vibration energy for each simulation cycle. A genetic algorithm is used as the optimization algorithm, with a population size of 50 and 200 iterations. The optimization variables include the velocity curve of the robotic arm and the key node parameters of the acceleration curve. For example, the traditional trapezoidal velocity curve parameters, where the acceleration time is 0.1s, the constant speed time is 0.3s, and the deceleration time is 0.1s, are modified to the five key control points of an S-shaped curve, which are the velocity values ​​corresponding to time points 0s, 0.1s, 0.25s, 0.4s, and 0.5s. After 200 iterations, an optimal set of motion trajectory parameters is found, which reduces the spindle vibration energy by 30% compared to the traditional trapezoidal trajectory.

[0025] The non-standard flexible motion trajectory is compiled into a control program and downloaded to the programmable logic controller (PLC) of the doffing robot. Specifically, the optimized S-shaped flexible motion trajectory parameters, such as the position, velocity, and acceleration command values ​​of each axis at each 1ms time step, are used to generate G-code or a PLC-recognizable instruction format. Then, the control program is downloaded to the PLC via the EtherCAT bus to control each servo axis of the doffing robot. By performing detailed modeling and co-simulation of subsystems such as machinery, servo motors, and spindles, the true dynamic characteristics of the equipment can be accurately reproduced. Based on this high-fidelity model, the optimized motion trajectory can proactively avoid the system's inherent resonance modes at their source. This optimization method based on virtual prototypes ensures the scientific rigor and optimality of motion planning, fundamentally guaranteeing the system's low-vibration operation.

[0026] Furthermore, the establishment of the disturbance model library includes: Offline calibration experiments were conducted to collect instantaneous data on residual mechanical impact and motor starting torque pulsation under different operating conditions. Frequency domain analysis was performed on the instantaneous data to extract the characteristic frequencies, amplitudes, and phase information of the main harmonic components constituting the disturbance signal, forming a structured set of disturbance characteristic parameters. Specifically, the following offline calibration experiments were performed on a testing platform in a spinning mill: The residual mechanical impact simulation included operations such as the contact between the robotic arm's gripper and the yarn tube, gripper closure, the empty tube falling into the spinning machine, and robotic arm resetting. Near each operation point, accelerometers were used to collect triaxial acceleration data from the robotic arm body and the spindle holder, with a sampling frequency set to 10kHz. For example, at the instant the gripper closed, a 20ms-long, peak-valued impact acceleration of 10g was measured on the robotic arm body along the X-axis. These time-domain data were analyzed using Fast Fourier Transform (FFT) to extract the main harmonic components. For instance, the gripper closure impact had a peak at 100Hz with an amplitude of 0.5g / Hz and a phase of 30 degrees; and another peak at 250Hz with an amplitude of 0.2g / Hz and a phase of -60 degrees.

[0027] Motor starting torque pulsation is measured by testing the servo motor at different starting accelerations, such as 500 rad / s², 1000 rad / s², and 1500 rad / s², and under different loads, such as no-load and 20% of rated load. A high-precision torque sensor (e.g., HBMT40B) is used to collect the motor output torque data, with the sampling frequency set to 5 kHz. For example, at a starting acceleration of 1000 rad / s², a torque pulsation lasting 50 ms with a peak value of 5 Nm is measured at the moment of motor start-up. FFT analysis is performed on the torque data, and for example, a main harmonic component exists at 50 Hz with an amplitude of 0.8 Nm / Hz and a phase of 15 degrees. The extracted characteristic frequencies, amplitudes, and phases, along with the corresponding operating parameters, including the robot's movement speed, load mass, and motor starting acceleration, are stored as a structured perturbation characteristic parameter set in JSON format.

[0028] Based on the disturbance characteristic parameter set, a parameterized disturbance model is established for each independent disturbance source, including but not limited to motor starting and robotic gripper closure. The parameterized disturbance model calls the corresponding characteristic parameter set according to the upcoming operation command, and reconstructs and predicts the disturbance waveform through Fourier series. Specifically, for the disturbance source "robotic gripper closure," a parameterized model is established. When the PLC command recognizes the gripper closure command, the model uses the current robotic speed and the weight of the gripped object as operating parameters, calls the corresponding harmonic components from the disturbance characteristic parameter set (e.g., 100Hz, 0.5g / Hz, 30°; 250Hz, 0.2g / Hz, -60°), normalizes the amplitude, frequency, and phase, and then uses Fourier series, for example... ,in It is the amplitude. It's frequency. It represents the phase, N is the number of cosine waves, and t is the time. The function is a cosine function; the predicted acceleration impact waveform is reconstructed, with a duration of 50ms.

[0029] The parameterized disturbance model is modularly encapsulated, assigned a globally unique identifier, and an index table is created and stored. This index table establishes a one-to-one correspondence between the unique identifier and the programmable logic controller (PLC) instruction code that triggers the corresponding disturbance, constructing a query path. The disturbance characteristic parameter set, parameterized disturbance model, and index table are stored in a disturbance model library. Specifically, each parameterized disturbance model is encapsulated as an independent C++ class or Python module and assigned a globally unique identifier, such as "MEC_IMPACT_CLAW_CLOSE_001". Then, an SQL database is created as the index table, containing the "PLC instruction code", such as MOVA, B, R100 – robot axis A moves to position B, R100 represents gripper closure, and the corresponding "disturbance model ID", such as "MEC_IMPACT_CLAW_CLOSE_001". This data is stored together on the factory's central server as a disturbance model library.

[0030] Through offline calibration experiments and frequency domain analysis, accurate quantification and feature extraction of various disturbance sources were achieved. This data-driven modeling approach enables the disturbance model library to truly reflect the impact and pulsation characteristics in the physical world. This provides an accurate and reliable data foundation for subsequent forward-looking compensation, greatly improving the fidelity of the predicted signal and the accuracy of the compensation effect.

[0031] Furthermore, the generation of predictive compensation signals includes: setting up a forward-looking instruction parsing module, which synchronously reads the queue of unexecuted instructions stored in the PLC in real time, identifies the target operation instruction to be executed in the queue, and predicts the future execution time based on the position of the target operation instruction in the queue and the scan cycle of the PLC; specifically, a communication module based on the OPCUA protocol is developed and connected to the PLC in real time; this module reads the instruction buffer of the PLC at a period of 1ms to obtain the queue of instructions to be executed in the next 500ms; for example, if there is an instruction "MOVA, B, R100" (gripper closure instruction) in the instruction queue, which is the 10th instruction in the queue, and the scan cycle of the PLC is 1ms, then it is predicted that the instruction will be executed in the next 10ms; Using the instruction code of the target operation instruction as an index, the system queries the index table to locate and call the parameterized perturbation model corresponding to the target operation instruction. Specifically, after the forward-looking instruction parsing module identifies the instruction "MOVA, B, R100", it uses "R100" as an index to query the SQL index table mentioned above and locates the corresponding perturbation model ID "MEC_IMPACT_CLAW_CLOSE_001". Based on the set of disturbance characteristic parameters inside the parameterized disturbance model, a predicted disturbance waveform is generated in real time. Specifically, the “MEC_IMPACT_CLAW_CLOSE_001” model is called. This model selects the best matching parameter combination from the set of disturbance characteristic parameters based on the current real-time operating status of the robot arm, including speed, position, etc. These data are also obtained in real time through OPCUA, and a predicted acceleration impact waveform lasting 50ms is generated in real time. Based on the predicted disturbance waveform, a reverse waveform with the same amplitude but opposite phase to the predicted disturbance waveform is calculated, and the reverse waveform is output as a predictive compensation signal. Specifically, the generated predicted acceleration impact waveform is negativeed point by point to obtain a reverse waveform with the same amplitude but opposite phase to the predicted waveform. This reverse waveform, for example, is an acceleration waveform lasting 50ms with a peak value of -10g, which is used as the predictive compensation signal.

[0032] By reading the PLC's instruction queue that has not yet been executed in real time, the system can "predict" future disturbances. This mechanism enables the generation and output of compensation signals to precede the actual occurrence of the disturbance, completely eliminating the response delay of traditional control methods. This proactive control strategy is the key to achieving high-precision real-time vibration suppression and greatly improves the dynamic response speed of the system.

[0033] Furthermore, the dynamic adjustment of the feedforward gain of the servo system when executing the non-standard flexible motion trajectory includes: setting up an energy observation module to analyze the predictive compensation signal based on a preset system energy transfer model, and calculating and generating a predictive energy injection curve describing the timing, amplitude, and power of disturbance energy injection within a future time window; specifically, the energy observation module receives the aforementioned predictive compensation signal (reverse acceleration waveform) and combines it with a preset system energy transfer model (e.g., describing the conversion relationship from acceleration signal to spindle vibration energy through a transfer function); this model considers parameters such as the mass, stiffness, and damping of the manipulator and spindle system; the method for establishing this model is: firstly, extracting from the previous multibody dynamics simulation results... The equivalent mass, stiffness, and damping ratio parameters of the spindle under the main vibration modes are obtained. Then, a standard second-order system model is constructed using these parameters to describe the dynamic response relationship between the input acceleration and the output vibration displacement and velocity. Finally, the energy observation module calculates the future vibration velocity and displacement based on the model, and combines the known equivalent mass and stiffness to calculate the corresponding sum of kinetic and potential energy in real time, thereby generating a predicted energy injection curve. For example, it is predicted that the gripper closing operation will occur in the next 10 ms, and inject 2 joules of vibration energy in the next 50 ms, with its power peaking at 40 watts at 30 ms. This information is integrated into a predicted energy injection curve, describing the dynamic changes of disturbance energy in the next 200 ms.

[0034] A control primitive strategy library is constructed, which encapsulates independent vibration control units. Specifically, the vibration control unit includes: a dynamic stiffness adjustment unit corresponding to dynamically adjusting the feedforward gain; an active damping unit corresponding to executing active damping control; and a waveform cancellation unit corresponding to calculating the inverse waveform drive signal. This decouples the fixed control flow into a modular control toolbox, greatly improving the flexibility and scalability of the control strategy. Multiple control methods can be freely combined according to different disturbance characteristics to cope with working conditions that are far more complex than a single flow.

[0035] When the doffing robot's servo system tracks and executes the non-standard flexible motion trajectory, the feedforward controller synchronously receives the predicted energy injection curve. It increases the feedforward gain before the predicted energy injection peak arrives and decreases the gain during periods of flat energy injection. Specifically, the feedforward control module in the doffing robot's servo controller receives the predicted energy injection curve in real time. When the energy observation module predicts an energy injection peak within the next 5ms (i.e., between 5ms and 10ms, assuming the prediction is for an event occurring at 10ms), the feedforward controller increases the position and velocity feedforward gains from the usual 0.8 to 1.2. During periods of flat energy injection, such as between 15ms and 25ms, the feedforward gain returns to 0.8. This dynamic adjustment ensures that the servo system can more accurately track the flexible motion trajectory and counteract the effects of predicted disturbances.

[0036] Energy analysis is performed on predicted disturbances, and the servo feedforward gain is dynamically adjusted accordingly, giving the robotic arm the characteristic of "time-varying stiffness". The stiffness is enhanced before the impact to resist the disturbance, and the stiffness is reduced during smooth movement to ensure compliance. This intelligent gain scheduling strategy achieves the optimal balance between disturbance resistance and smoothness of the system.

[0037] Furthermore, using the predictive compensation signal to drive the actuator below the spindle seat to perform active damping control includes: The original waveform of the predictive compensation signal is used as a drive command and sent to the actuator below the spindle. Specifically, the predictive compensation signal generated above, that is, the inverse waveform of the predicted disturbance waveform, such as an acceleration waveform lasting 50ms and peak value of -10g, is sent in real time via EtherCAT bus to the piezoelectric actuator installed below each spindle. Each actuator has an output force of 100N, a stroke of 10µm, and a response frequency of up to 2kHz. The actuator's controller executes an active damping control algorithm. This algorithm applies a damping force proportional to and opposite to the vibration velocity, based on the received command waveform and the real-time detected spindle vibration velocity. Specifically, each piezoelectric actuator is equipped with an independent digital controller. This controller receives a predictive compensation signal at a sampling frequency of 5 kHz and simultaneously acquires real-time spindle vibration velocity data via a laser vibration meter integrated on the spindle base, such as a Polytec PSV-500. The controller internally executes an LQR active damping control algorithm. According to the algorithm, the controller calculates a force proportional to (with a proportionality coefficient of -0.5 Ns / m) and opposite to the real-time detected spindle vibration velocity, and drives the piezoelectric actuator to apply this force, thereby dissipating vibration energy and suppressing spindle vibration. Using predictive compensation signals to drive actuators to perform active damping control is equivalent to installing an intelligent, predictable "virtual damper" for the spindle system. This method directly acts on the dissipation of vibration energy, which can quickly suppress oscillations and shorten the decay time. Compared with position compensation alone, active damping can absorb and eliminate vibration energy more efficiently.

[0038] Further, calculating the inverse waveform drive signal includes: An online time-series prediction model is employed to predict the residual vibration waveform that will still exist within a future time window after the dual feedforward compensation effects of dynamic feedforward gain adjustment and active damping control. Specifically, the input of this online time-series prediction model (e.g., a Kalman filter-based predictor) synchronously receives and comprehensively analyzes the following three types of information: first, the robot motion command data after feedforward gain adjustment; second, the original predictive compensation signal used to drive the actuator to perform active damping; and third, the current vibration state of the spindle as fed back in real time by sensors. Based on the above complete input, the model uses 1 The model uses a time step of ms to predict the actual residual vibration waveform on the spindle within the next 20ms. For example, the model predicts that between 60ms and 80ms after the gripper closes, the spindle will still have a residual vibration with a peak value of 0.5g and a frequency of 80Hz. The predicted residual vibration waveform is then inverted to generate a target cancellation trajectory with the same amplitude but opposite phase as the residual vibration waveform. Specifically, the predicted residual vibration waveform is negativeed point by point, for example, the peak value of 0.5g is inverted to -0.5g to obtain a trajectory with the same amplitude but opposite phase as the residual vibration.

[0039] The target cancellation trajectory is taken as the motion output and substituted into the system inverse dynamics model. The system inverse dynamics model is a pre-established model that describes the relationship between the actuator input signal and the spindle vibration output. The inverse waveform drive signal is calculated by inverse solution. Specifically, the pre-established system inverse dynamics model is obtained through system identification method and describes the transfer function between the voltage input signal of the piezoelectric actuator and the spindle vibration displacement output. This model takes into account the nonlinearity and hysteresis characteristics of the piezoelectric actuator. The target cancellation trajectory, such as the desired displacement of the spindle to -10µm at a certain moment, is taken as the desired output of the inverse dynamics model. The voltage signal required to drive the piezoelectric actuator to generate the cancellation trajectory is calculated by inverse solution. For example, a voltage waveform lasting 20ms with a peak value of 100V is the inverse waveform drive signal.

[0040] A secondary prediction and cancellation mechanism for residual vibration was set up, enabling refined vibration suppression. The small vibrations that still exist after the initial compensation were predicted by an online model, and the accurate inverse waveform signal was calculated based on the inverse dynamics model. This secondary compensation measure greatly improved the final accuracy of vibration control and achieved ultimate suppression of vibration.

[0041] Furthermore, before the predicted vibration occurs, the inverse waveform drive signal is injected into the actuator's drive amplifier; the model is corrected using a PLC timing lock-in algorithm, including: Using a clock synchronization mechanism, the signal is injected into the actuator's drive amplifier precisely before the occurrence of residual vibration. Specifically, the control system achieves high-precision time synchronization between the PLC, servo driver, and actuator controller via the NTP protocol, with synchronization accuracy reaching the microsecond level. When the predicted residual vibration is expected to occur 5ms later, the calculated inverse waveform drive signal, such as a voltage waveform with a peak value of 100V, is injected into the piezoelectric actuator's drive amplifier 3ms in advance (as a safety margin) via the EtherCAT bus. The drive amplifier amplifies the voltage signal and drives the actuator to generate the required counteracting force.

[0042] A PLC-based phase-locked loop (PLL) algorithm is employed to periodically correct and update the online timing prediction model and the disturbance model library by comparing the phase difference and amplitude difference between the actual vibration and the predicted results. Specifically, the PLC PLL algorithm is executed every 100 doffing cycles. This algorithm uses a laser vibration meter to collect the actual vibration waveform of the spindle during the doffing process in real time. Then, the actual vibration waveform is compared with the vibration waveform predicted by the online timing prediction model and the disturbance model library. The phase difference (e.g., 10 degrees) and amplitude difference (e.g., 20%) are calculated. If the phase difference exceeds 5 degrees or the amplitude difference exceeds 10%, a correction mechanism is triggered. This correction mechanism utilizes this error information and employs an adaptive filtering algorithm, such as the LMS algorithm, to perform online correction of the parameters of the online timing prediction model, such as the covariance matrix of the Kalman filter. The adjustment mechanism also includes an automatic update module for the disturbance model library. The update logic of this module is as follows: when the average prediction amplitude error of a specific disturbance source is detected to be consistently higher than a preset threshold, the system will fine-tune the amplitude parameters of each harmonic component corresponding to that disturbance source in the disturbance model library by a preset ratio based on the magnitude and direction of the error. Similarly, the continuous average phase error will also be used to fine-tune the corresponding phase parameters. This periodic, small-amplitude automatic adjustment allows the disturbance model to gradually track and adapt to the actual changes in the system's characteristics. If the prediction error of a specific disturbance source, such as "motor starting torque pulsation," remains large, the system will record this situation and periodically prompt maintenance personnel to re-perform offline calibration experiments to update the corresponding disturbance characteristic parameter set in the disturbance model library.

[0043] A PLC timing-locked loop algorithm is used to periodically correct and update the online timing prediction model and the disturbance model library by comparing the phase difference and amplitude difference between the actual vibration and the predicted results, and to generate maintenance instructions for equipment components. Specifically, a historical prediction error database is established, corresponding one-to-one with each disturbance source in the disturbance model library, to store the amplitude difference and phase difference data associated with the disturbance source calculated by the PLC timing-locked loop algorithm in a time series manner. A trend analysis module is set up to perform statistical learning and time series analysis on the data in the historical prediction error database to extract the trend characteristics of the prediction error corresponding to a specific disturbance source over time. The trend characteristics are compared with a preset health baseline model in real time to calculate the deviation. When the deviation exceeds a preset fault threshold, a health status assessment report and predictive maintenance instructions for the equipment components associated with the disturbance source are generated. Clock synchronization ensures the precise timing of the compensation signal injection, while the PLC timing phase-locked algorithm gives the entire control system the ability to learn and adapt. The algorithm continuously corrects the model by comparing the actual and predicted results, ensuring that the system can adapt to the characteristic changes caused by wear and other factors during long-term operation and maintain the long-term stability of control performance.

[0044] This embodiment introduces an adaptive control method for spindle vibration in a spinning mill. First, through electromechanical co-simulation, a non-standard flexible motion trajectory is optimized to suppress vibration at its source. Then, by establishing a disturbance model library and proactively analyzing PLC instructions, the method predicts impending mechanical shocks and torque pulsations in real time, generating a predictive compensation signal. This signal is used to drive a composite active control system: on the one hand, it dynamically adjusts the servo feedforward gain to actively resist shocks; on the other hand, it drives the actuators under the spindle to perform active damping and cancel out the inverse waveform of residual vibration. Finally, the system uses a phase-locked loop algorithm to compare the actual and predicted vibration errors, periodically correcting the model to achieve adaptive optimization of control performance.

[0045] Example 2: In a smart spinning workshop, there are three different types of spinning machines, each using spindles of different sizes and weights. At the same time, there are two doffing robots, M1 and M2, of the same type but with different workloads, working together to handle the doffing task of the spinning machines. Robot M1 usually handles spindles of type A and B, while robot M2 handles spindles of type C.

[0046] For each type of spindle-bearing-yarn tube system (i.e., type A, type B, type C), a multibody dynamics model is established to extract their natural frequencies and damping characteristics. For example, the natural frequency of spindle A is 80Hz, that of type B is 75Hz, and that of type C is 90Hz. At the same time, finite element modeling is performed on two doffing robots, M1 and M2, to consider the stiffness and modal differences that may be caused by their actual operating wear and load conditions. Furthermore, based on multiple sets of electromechanical joint dynamic models, customized non-standard flexible motion trajectories are generated for each "manipulator-spindle model" combination (e.g., M1-A, M1-B, M2-C) through reverse optimization calculations; for example, the S-shaped trajectory parameters used by M1 when processing type A spindles are different from those used when processing type B spindles. The disturbance model library will be refined based on "manipulator ID" and "spindle model"; offline calibration experiments will be conducted separately for two manipulators, M1 and M2, and 2*3=6 combinations of three spindle models, A, B, and C; for example, the disturbance characteristic parameter set of "manipulator M1 - gripper closed - model A spindle" is different from the parameter set of "manipulator M2 - gripper closed - model C spindle"; the index table of the disturbance model library will add "manipulator ID" and "spindle model" as query fields to achieve more refined disturbance source identification.

[0047] In addition to reading PLC instructions, the forward-looking instruction parsing module also obtains the ID of the robot currently performing the doffing operation and the model of the spindle it is processing in real time; when generating predictive disturbance waveforms and predictive compensation signals, the system calls the disturbance model corresponding to the robot ID and spindle model for calculation; when adjusting the gain, the feedforward controller also dynamically adjusts the parameters of the energy transfer model based on the characteristics of the current robot and spindle model. Furthermore, the actuators under each spindle will be calculated using the corresponding system inverse dynamics model based on their spindle model when calculating the inverse waveform drive signal. For example, the inverse dynamics model of the A-type spindle actuator has different model parameters than that of the B-type spindle actuator. When the PLC timing lock-in algorithm corrects the online timing prediction model and disturbance model library, it will perform error analysis and parameter updates for different combinations of manipulators and spindle models to ensure targeted correction. For example, if the prediction error of the M1-B type spindle combination is consistently large, the system will only update the model parameters related to that combination and will not affect the model of the M2-C type spindle combination. This embodiment demonstrates that the present invention can adapt to more complex production environments and effectively solves the problems of using spindles of different sizes and weights for various types of spinning machines, thus achieving more reliable vibration adaptive control.

[0048] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive control method for the vibration of a collective doffing spindle, characterized in that, Includes the following steps: An electromechanical joint dynamic model is established, consisting of a doffing robot, a servo drive system, and a spindle-bearing-yarn tube. Based on this model, a reverse optimization calculation is performed to generate a non-standard flexible motion trajectory. This is then integrated into a joint simulation platform to construct a virtual prototype system that reflects the dynamic coupling relationship between the subsystems. Closed-loop dynamic simulation is performed to obtain simulation results. Based on these results, with the goal of minimizing the spindle vibration energy, an optimization algorithm is used to iteratively optimize the motion trajectory parameters in the simulation environment until an optimal non-standard flexible motion trajectory is calculated. When the doffing robot's servo system tracks and executes the non-standard flexible motion trajectory, the feedforward controller synchronously receives the predicted energy injection curve, increases the feedforward gain before the predicted energy injection peak is about to arrive, and decreases the gain during the flat period of energy injection. A disturbance model library containing residual mechanical shock and motor starting torque pulsation is created. The PLC instruction queue is synchronized and parsed in real time to capture upcoming operations and time points. The corresponding model is called from the disturbance model library to generate a predictive compensation signal. The predictive compensation signal is analyzed using an online time-series prediction model. Based on a meta-learning model, the characteristics of the system state and the predictive compensation signal are analyzed in real time to generate optimal control parameters. The predictive energy injection curve is calculated and generated to dynamically adjust the feedforward gain of the servo drive system when executing the non-standard flexible motion trajectory. The predictive compensation signal is used to drive the actuator below the spindle to perform active damping control; the residual vibration waveform after feedforward gain is predicted, and the inverse waveform drive signal to cancel the residual vibration waveform is calculated based on the system inverse dynamics model; the inverse waveform drive signal is injected into the actuator's drive amplifier before the predicted vibration occurs.

2. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, The construction of the electromechanical combined dynamics model includes: Finite element modeling was performed on the mechanical structure of the doffing robot to obtain its stiffness and modal characteristics. A dynamic response model of the servo drive system was established based on the motor torque-speed characteristics and driver response delay. Multibody dynamics modeling was performed on the spindle-bearing-yarn tube system to extract its natural frequencies and damping characteristics. These were integrated into a co-simulation platform to construct a virtual prototype system that reflects the dynamic coupling relationship between the subsystems. Closed-loop dynamic simulation was performed to obtain simulation results. Based on these results, minimizing the spindle vibration energy was used as the optimization objective. An optimization algorithm was employed to iteratively optimize the motion trajectory parameters in the simulation environment until an optimal non-standard flexible motion trajectory was calculated. This non-standard flexible motion trajectory was then compiled into a control program and downloaded to the programmable logic controller (PLC) of the doffing robot.

3. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, The establishment of the disturbance model library includes: Through offline calibration experiments, instantaneous data of residual mechanical impact and motor starting torque pulsation were collected under different working conditions. Frequency domain analysis was performed on the instantaneous data to extract the characteristic frequency, amplitude and phase information of the main harmonic components constituting the disturbance signal, forming a structured set of disturbance characteristic parameters. Based on the disturbance characteristic parameter set, a parameterized disturbance model is established for each independent disturbance source, including but not limited to motor starting and robotic gripper closure. The parameterized disturbance model calls the corresponding feature parameter set according to the upcoming operation instruction, reconstructs and predicts the disturbance waveform using Fourier series. The parameterized disturbance model is modularly encapsulated, assigned a globally unique identifier, and an index table is created and stored. This index table is used to establish a one-to-one correspondence between the unique identifier and the programmable logic controller instruction code that triggers the corresponding disturbance, constructing a query path. The disturbance characteristic parameter set, parameterized disturbance model, and index table are stored in a disturbance model library.

4. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, The generation of the predictive compensation signal includes: A forward-looking instruction parsing module is set up. This module synchronously reads the queue of unexecuted instructions stored in the PLC in real time, identifies the target operation instruction that is about to be executed in the queue, and predicts the future execution time based on the position of the target operation instruction in the queue and the scan cycle of the PLC. Based on a meta-learning model, the waveform and energy characteristics of the predictive compensation signal are received in real time and combined with the current operating state to form a decision state vector. The meta-learning model performs real-time parsing of the decision state vector. Based on the analysis results, optimal control parameters are generated, and composite drive instructions are synthesized and executed. Using the instruction code of the target operation instruction as an index, the index table is searched to locate and call the parameterized disturbance model corresponding to the target operation instruction. Based on the disturbance characteristic parameter set inside the parameterized disturbance model, a predicted disturbance waveform is generated in real time. Based on the predicted disturbance waveform, an inverse waveform with the same amplitude but opposite phase to the predicted disturbance waveform is calculated, and the inverse waveform is output as a predictive compensation signal.

5. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, The dynamic adjustment of the feedforward gain of the servo system when executing the non-standard flexible motion trajectory includes: An energy observation module is set up to analyze the predictive compensation signal based on a preset system energy transfer model, calculate and generate a predicted energy injection curve describing the time, amplitude and power of disturbance energy injection within a future time window. When the doffing robot's servo system tracks and executes the non-standard flexible motion trajectory, the feedforward controller synchronously receives the predicted energy injection curve, increases the feedforward gain before the predicted energy injection peak arrives, and decreases the gain during the smooth period of energy injection.

6. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, Using the predictive compensation signal to drive the actuator below the spindle to perform active damping control includes: The original waveform of the predictive compensation signal is sent as a driving command to the actuator below the spindle. The controller of the actuator executes an active damping control algorithm, which applies a damping force that is proportional to the vibration velocity and opposite in direction, based on the received command waveform and the real-time detected spindle vibration velocity.

7. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, The calculation of the inverse waveform drive signal includes: using an online timing prediction model to predict the residual vibration waveform that will still exist in the future time window after the feedforward gain is dynamically adjusted; inverting the predicted residual vibration waveform to generate a target cancellation trajectory with the same amplitude but opposite phase as the residual vibration waveform; substituting the target cancellation trajectory as the motion output into the system inverse dynamics model, which is a pre-established model describing the relationship between the actuator input signal and the spindle vibration output; and calculating the inverse waveform drive signal by inverse solution.

8. The adaptive control method for collective doffing spindle vibration according to claim 1, characterized in that, Injecting the inverse waveform drive signal into the actuator's drive amplifier before the predicted vibration occurs includes: using a clock synchronization mechanism to inject the signal into the actuator's drive amplifier before the point at which the residual vibration occurs.

Images