A vibration suppression control method for cable parallel robots based on mechanism model and data driving

Abstract

The application discloses a kind of mechanism model and data driving-based cable parallel robot vibration suppression control method, comprising: the dynamics modeling of cable-driven parallel robot, the inherent frequency of each point in workspace is solved;Randomly select point in workspace, move to selected position, use reaction force device to excite, vibration sensor records pose change;According to vibration data, the actual inherent frequency of system is identified, and the spatial position-inherent frequency residual data set is used as input to train the deep learning model describing the relationship between position coordinates and inherent frequency residual;With the design goal of minimizing inherent frequency vibration signal, the controller is designed for vibration suppression control, and the inherent frequency of each position is obtained by summing the predicted value of the dynamics model and the residual given by the deep learning model. Through the method, different cable parallel robots can be quickly suppressed, and the working accuracy and stability of the end of the cable parallel robot can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of cable parallel robot technology, and in particular to a vibration suppression control method for cable parallel robots based on a mechanism model and data-driven approach. Background Technology

[0002] Cable-driven parallel robots consist of multiple cables and a suspended rigid mobile platform. Due to the lightweight, low-cost, and easily replaceable and reconfigurable nature of the cables, they have great application potential in scenarios requiring high-speed operation or large workspaces. However, compared to traditional rigid-link robots, cable-driven parallel robots use lightweight cables to connect the base platform and the mobile platform. Because of the flexibility and unidirectional force limitations of the cable structure, they are susceptible to external disturbances, leading to undesirable vibrations that affect operational accuracy and stability.

[0003] Existing research on vibration suppression control of cable parallel robots often uses model-based controllers, which require high accuracy in the dynamic modeling of cable parallel robots. Ignoring too much of the factors such as hysteresis, creep, and stiffness of the rope can easily lead to inaccurate system models, resulting in poor vibration suppression effects. On the other hand, overly complex modeling processes greatly increase the computational difficulty and lack feasibility in practical operation. Summary of the Invention

[0004] To overcome the problems existing in the prior art, the purpose of this invention is to provide a vibration suppression control method for cable-driven parallel robots based on mechanistic models and data-driven approaches. The method uses a reaction force device to excite vibration for modal analysis, obtains the natural frequency residuals to train a deep learning model, compensates for the dynamic modeling results, and designs a sliding mode controller or other state-space controller with the goal of minimizing the vibration signal of a certain order natural frequency. This avoids overly complex dynamic modeling and simultaneously achieves rapid vibration suppression for cable-driven robots, improving the operational accuracy and stability of cable-driven robots.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A vibration suppression control method for a cable-parallel robot based on a mechanism model and data-driven approach includes the following steps: Treating the cable as a massless spring with constant stiffness, and neglecting secondary factors, the robot's dynamics are modeled to solve for the natural frequencies of the robot at various points within its workspace; selecting a certain number of points within the workspace, moving the end effector platform of the cable-parallel robot until its center of mass coincides with the selected location, exciting it using a reaction force device, and recording the pose changes of the cable-parallel robot during vibration using an inertial sensing unit; identifying the actual natural frequency of the cable-parallel robot's vibration at that location based on the data recorded by the inertial sensing unit after excitation, and then determining the natural frequency of the selected natural frequency. A dataset is generated by combining the position coordinates of a point with the natural frequency residuals of the cable-parallel robot at that point. This spatial position-natural frequency residual dataset is used as input to train a deep learning model that describes the relationship between the workspace position coordinates and the corresponding natural frequency residuals, thereby correcting the natural frequencies obtained from the dynamic model. Based on the dynamic model of the cable-parallel robot, a state-space controller is designed with the goal of minimizing the vibration signal of a certain order natural frequency. This controller drives a reaction force device to output a damping force to control the vibration of the cable-parallel robot. The natural frequency at each position is obtained by summing the predicted value from the dynamic model and the residual given by the deep learning model.

[0007] Alternatively, the rope can be treated as a massless spring with constant stiffness, and secondary factors can be ignored. The Newton-Euler method or the Lagrange method can be used to model the dynamics of the cable-parallel robot. The dynamic model is as follows:

[0008]

[0009] Where M is the mass matrix of the cable-parallel robot, C is the damping matrix of the cable-parallel robot, K is the stiffness matrix of the cable-parallel robot, and F is the time-varying external load. For nodal acceleration vectors, q is the nodal velocity vector, and q is the nodal displacement vector;

[0010] Neglecting damping effects and external forces, assuming free vibration is a harmonic response, we obtain the following equation:

[0011] (KM -1 -ω 2 I)φ=0

[0012] Where is KM -1 The generalized stiffness matrix, ω represents the natural frequency of the cable-parallel robot, φ represents the vibration mode of the cable-parallel robot, and I is the identity matrix; it can calculate the dynamic model prediction of the natural frequency of all points in the workspace of the cable-parallel robot.

[0013] Optionally, the secondary factors include the rope's own weight, hysteresis effect, creep effect, and pulley size.

[0014] Optionally, the reaction force device may be an unbalanced rotating inertia actuator, a jet nozzle, or a rotor-type aerodynamic reaction force device;

[0015] As needed, a reaction force device is used to induce vibration of the moving platform of the cable-parallel robot, and an inertial sensing unit is used to record the pose changes of the moving platform in different directions.

[0016] Optionally, the step of identifying the actual natural frequency of the cable-parallel robot's vibration at that location based on the data recorded by the inertial sensing unit after excitation, and generating a dataset from the position coordinates of the selected point and the natural frequency residual of the cable-parallel robot at that point, includes:

[0017] The data collected by the inertial sensing unit is preprocessed, including filtering, noise reduction, and normalization.

[0018] Fourier transform is used to convert the collected time series data into frequency domain data, and modal analysis is used to obtain the actual natural frequency of the cable-parallel robot.

[0019] The residual Δω of the natural frequency of the cable-parallel robot at that location is obtained by using the actual natural frequency ω of the cable-parallel robot and the predicted natural frequency ω' calculated based on the dynamic model, thus generating a spatial position-natural frequency residual dataset.

[0020] Optionally, the modal analysis technique employs peak picking, curve fitting, or an autoregressive model.

[0021] Optionally, the spatial location-intrinsic frequency residual dataset is used as input to train a deep learning model that describes the relationship between workspace location coordinates and the corresponding intrinsic frequency residuals, resulting in...

[0022]

[0023] Where q is the nodal displacement vector. These are the parameters for the deep learning model.

[0024] The residual natural frequency of the cable-parallel robot at each point in the workspace is obtained from the deep learning model, and the natural frequency obtained from the dynamic model is corrected.

[0025] Optionally, the deep learning model adopts a multilayer perceptron model, specifically including an input layer, several hidden layers, and an output layer; the input layer contains three neurons representing the spatial coordinates of the selected point, and the output layer contains one neuron representing the intrinsic frequency residual; the ReLU function is used as the activation function, the MSE function as the loss function, and Adam as the optimizer.

[0026] Optionally, based on the dynamic model of the cable parallel robot, a state space controller is designed with the goal of minimizing the vibration signal of a certain order natural frequency, and the reaction force device is driven to output the vibration damping force to perform vibration damping control on the cable parallel robot.

[0027] The end effector trajectory of the cable parallel robot is divided into several segments. The average natural frequency of all path points in each segment is taken as the natural frequency of that segment. With the goal of minimizing the vibration signal of a certain order natural frequency, a controller is designed based on the characteristics of the cable parallel robot. The controller adopts pole placement, linear quadratic regulator, linear quadratic Gaussian regulator, sliding mode controller, model predictive control, adaptive control or H∞ robust control, etc.

[0028] Compared with the prior art, the present invention has the following advantages:

[0029] This invention presents a vibration suppression control method for cable-parallel robots based on mechanistic models and data-driven approaches. Through deep learning, the method compensates for and corrects the deviation between the dynamic model and the actual situation, avoiding the complex dynamic modeling and analysis process. At the same time, it achieves rapid vibration suppression at the end of the cable-parallel robot, improving the positioning accuracy and stability of the cable-parallel robot.

[0030] The present invention provides a vibration suppression control method for cable parallel robots based on mechanistic models and data-driven approaches. This method uses the same set of reaction force devices to achieve both excitation and suppression functions, thereby improving the integration of cable parallel robots and reducing the requirements for their load performance. Attached Figure Description

[0031] Figure 1 This is a flowchart of a vibration suppression control method for a cable parallel robot based on a mechanism model and data-driven approach, provided in an embodiment of this application. Detailed Implementation

[0032] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting the present invention.

[0033] Specifically, Figure 1 This is a flowchart illustrating a vibration suppression control method for a cable parallel robot based on a mechanism model and data-driven approach, as provided in an embodiment of this application.

[0034] like Figure 1 As shown, the vibration suppression control method for cable parallel robots based on mechanism models and data-driven approaches includes the following steps:

[0035] In step S101, the rope is regarded as a massless spring with constant stiffness. Ignoring secondary factors such as pulley size, the Newton-Euler method or the Lagrange method is used to perform dynamic modeling of the cable parallel robot, and the dynamic model of all points in the workspace is solved to predict the natural frequency.

[0036] In actual implementation,

[0037] The dynamic model of the cable-driven parallel robot can be written in the following form:

[0038]

[0039] Where M is the mass matrix of the cable-parallel robot, C is the damping matrix of the cable-parallel robot, K is the stiffness matrix of the cable-parallel robot, and F is the time-varying external load. For nodal acceleration vectors, q is the nodal velocity vector, and q is the nodal displacement vector;

[0040] Neglecting damping effects and external forces, and assuming that free vibration is a harmonic response, the following equation can be obtained:

[0041] (KM -1 -ω 2 I)φ=0

[0042] Where is KM -1 The generalized stiffness matrix, ω is the natural frequency of the cable-parallel robot, φ is the vibration mode of the cable-parallel robot, and I is the identity matrix.

[0043] The above expression can be rewritten in the following form:

[0044] ω'=f(q,KM -1 )

[0045] In other words, the natural frequency ω' predicted by the dynamic model is a function of the spatial position coordinates and the generalized stiffness matrix. Since the effects of the rope's own weight, hysteresis, creep, and pulley size on the cable exit point are not considered in the dynamic modeling process, there is a residual Δω between the natural frequency ω' predicted by the dynamic model and the actual natural frequency ω.

[0046] In step S102, a certain number of points are selected in the workspace, the end effector platform of the cable parallel robot is moved until the center of mass coincides with the selected position, the reaction force device is used to excite the vibration, and the pose change of the cable parallel robot during the vibration process is recorded by the inertial sensing unit.

[0047] In actual implementation, considering the influence of the workspace boundary on the cable stiffness, the natural frequency of the cable-parallel robot end effector changes more rapidly near the workspace boundary. Therefore, more excitation points should be taken near the workspace boundary. Let G(p) be the probability that the point corresponding to the displacement vector p is taken, and p0 be the displacement vector corresponding to the geometric center of the workspace. G(p) is required to follow a Cauchy distribution relative to 1 / ||p-p0||.

[0048] In step S103, the actual natural frequency of the cable-parallel robot's vibration at that location is identified based on the data recorded by the inertial sensing unit after excitation. A dataset is generated from the position coordinates of the selected point and the residual of the cable-parallel robot's natural frequency at that point.

[0049] In actual execution, it is necessary to first preprocess the inertial sensing unit data collected under a certain excitation condition, including filtering, noise reduction, normalization and other operations, to obtain the data of displacement x, velocity v, acceleration a or angular displacement θ, angular velocity ω and angular acceleration α changing with time.

[0050] From the Fourier transform formula:

[0051]

[0052] Fourier transform is used to convert the collected displacement x, velocity v, acceleration a or angular displacement θ, angular velocity ω and angular acceleration α time series data into frequency domain data. Modal analysis techniques such as peak picking, curve fitting, and autoregressive model are used to extract the actual natural frequency ω of the cable parallel robot. The selected point position p is used as a sample, and the corresponding natural frequency residual Δω is used as a label, and added to the dataset in the form of (p, Δω).

[0053] In step S104, the spatial location-natural frequency residual dataset is used as input to train a deep learning model that describes the relationship between workspace location coordinates and corresponding natural frequency residuals, and the natural frequencies obtained from the dynamic model are corrected.

[0054] Optionally, a multilayer perceptron model can be used for learning, specifically comprising an input layer, several hidden layers, and an output layer. The input layer contains three neurons, representing the spatial coordinates of the selected point, and the output layer contains one neuron, representing the intrinsic frequency residual. The number of hidden layers and the number of neurons in each layer can be set according to the actual working scenario, and no specific restrictions are imposed here.

[0055] The ReLU function is used as the activation function, the MSE function as the loss function, and Adam as the optimizer. During execution, the initial learning rate is set by technicians based on the actual work scenario. The spatial location-intrinsic frequency residual dataset is divided into training and test sets proportionally, such as 80% for training and 20% for testing. The model is trained iteratively several times, and its performance is evaluated on the validation set after each training round. Based on the results on the test set, the model structure and hyperparameters are adjusted until satisfactory prediction accuracy is achieved.

[0056] In step S105, based on the dynamic model of the cable parallel robot, a state space controller such as a sliding mode controller is designed with the goal of minimizing the vibration signal of a certain order natural frequency, and the reaction force device is driven to output the vibration damping force to perform vibration damping control on the cable parallel robot.

[0057] In actual implementation, the dynamic equations of the cable-parallel robot need to be converted into a state-space representation, as follows:

[0058]

[0059] Where x(t) is the state vector, Let u(t) be the first derivative of the state vector with respect to time, y(t) be the controlled variable, y(t) be the output variable, A be the system matrix, B be the input matrix, C be the output matrix, and D be the pass-through matrix.

[0060] The controller is designed based on the state-space equation. It should be noted that the embodiment uses sliding mode control as an example for illustration. In actual operation, various controllers such as pole placement, linear quadratic regulator, linear quadratic Gaussian regulator, model predictive control, adaptive control, and H∞ robust control can be used, and no restrictions are imposed here.

[0061] Specifically, the design should satisfy the following form for the sliding surface S:

[0062] S = c1x1 + x2

[0063] Where x1 and x2 are the corresponding components of the state vector, and c1 is the sliding surface design parameter, determined according to the dynamic equations of the actual system. A reaching law is designed as needed, and the stability of the cable-parallel robot is determined using Lyapunov functions, thereby obtaining the specific parameter range of the designed controller.

[0064] It is worth noting that although the technical solutions and embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to the specific embodiments described above. The embodiments described above are merely illustrative. Those skilled in the art can make many other forms based on the inspiration of the present invention without departing from the spirit and scope of the claims, and these all fall within the scope of protection of the present invention.

Claims

1. A vibration suppression control method for cable parallel robots based on mechanistic models and data-driven approaches, characterized in that, Includes the following steps: Treating the rope as a massless spring with constant stiffness and ignoring secondary factors, a dynamic model of the cable-parallel robot is constructed to solve for the natural frequencies of the robot at various points within its workspace. A certain number of points are selected within the workspace, and the end effector platform of the cable-parallel robot is moved until its center of mass coincides with the selected location. A reaction force device is used to excite the robot, and an inertial sensor unit records the pose changes during the vibration process. Based on the data recorded by the inertial sensor unit after excitation, the actual natural frequency of the cable-parallel robot at that location is identified. A dataset is generated from the position coordinates of the selected points and the natural frequency residuals of the cable-parallel robot at those points. This spatial position-natural frequency residual dataset is used as input to train a deep learning model describing the relationship between the workspace position coordinates and the corresponding natural frequency residuals, correcting the natural frequencies obtained from the dynamic model. Based on the dynamic model of the cable-parallel robot, a state-space controller is designed with the goal of minimizing the vibration signal of a certain order of natural frequency. This controller drives the reaction force device to output a damping force to control the vibration of the cable-parallel robot. The natural frequency at each location is obtained by summing the predicted value from the dynamic model and the residual given by the deep learning model. The secondary factors include the rope's own weight, hysteresis effect, creep effect, and pulley size.

2. The method according to claim 1, characterized in that, Treating the rope as a massless spring with constant stiffness and neglecting secondary factors, the dynamics of the cable-parallel robot are modeled using the Newton-Euler method or the Lagrange method. The dynamic model takes the form of: Where M is the mass matrix of the cable-parallel robot, C is the damping matrix of the cable-parallel robot, K is the stiffness matrix of the cable-parallel robot, and F is the time-varying external load. For nodal acceleration vectors, q is the nodal velocity vector, and q is the nodal displacement vector; Neglecting damping effects and external forces, assuming free vibration is a harmonic response, we obtain the following equation: (KM -1 -ω 2 I)φ=0 KM -1 Let ω be the generalized stiffness matrix, φ be the natural frequency of the cable-parallel robot, φ be the vibration mode of the cable-parallel robot, and I be the identity matrix; it can calculate the natural frequency of the dynamic model of all points in the workspace of the cable-parallel robot.

3. The method according to claim 1, characterized in that, The reaction force device is an unbalanced rotating inertia actuator, a jet nozzle, or a rotor-type aerodynamic reaction force device. As needed, a reaction force device is used to induce vibration of the moving platform of the cable-parallel robot, and an inertial sensing unit is used to record the pose changes of the moving platform in different directions.

4. The method according to claim 1, characterized in that, The actual natural frequency of the cable-parallel robot's vibration at that location is identified based on the data recorded by the inertial sensing unit after excitation. A dataset is generated from the position coordinates of the selected point and the residual of the cable-parallel robot's natural frequency at that point, including: The data collected by the inertial sensing unit is preprocessed, including filtering, noise reduction, and normalization. Fourier transform is used to convert the collected time series data into frequency domain data, and modal analysis is used to obtain the actual natural frequency of the cable-parallel robot. The residual Δω of the natural frequency of the cable-parallel robot at that location is obtained by using the actual natural frequency ω of the cable-parallel robot and the predicted natural frequency ω' calculated based on the dynamic model, thus generating a spatial position-natural frequency residual dataset.

5. The method according to claim 4, characterized in that, The modal analysis technique employs peak picking, curve fitting, or autoregressive models.

6. The method according to claim 1, characterized in that, Using the spatial location-intrinsic frequency residual dataset as input, a deep learning model describing the relationship between workspace location coordinates and their corresponding intrinsic frequency residuals is trained, resulting in... Where q is the nodal displacement vector. These are parameters for the deep learning model. The residual natural frequency of the cable-parallel robot at each point in the workspace is obtained from the deep learning model, and the natural frequency obtained from the dynamic model is corrected.

7. The method according to claim 1, characterized in that, The deep learning model employs a multilayer perceptron model, specifically comprising an input layer, several hidden layers, and an output layer. The input layer contains three neurons representing the spatial coordinates of the selected point, and the output layer contains one neuron representing the intrinsic frequency residual. The ReLU function is used as the activation function, the MSE function as the loss function, and Adam as the optimizer.

8. The method according to claim 1, characterized in that, Based on the dynamic model of the cable parallel robot, a state space controller is designed with the goal of minimizing the vibration signal of a certain order natural frequency. The controller drives the reaction force device to output the vibration damping force to control the vibration of the cable parallel robot. The end effector trajectory of the cable parallel robot is divided into several segments. The average natural frequency of all path points in each segment is taken as the natural frequency of that segment. With the goal of minimizing the vibration signal of a certain order natural frequency, a controller is designed based on the characteristics of the cable parallel robot. The controller adopts pole placement, linear quadratic regulator, linear quadratic Gaussian regulator, sliding mode controller, model predictive control, adaptive control or H∞ robust control.

Images