CN108445764B - Active compliance control strategy of Stewart platform - Google Patents
Active compliance control strategy of Stewart platform Download PDFInfo
- Publication number
- CN108445764B CN108445764B CN201810244210.3A CN201810244210A CN108445764B CN 108445764 B CN108445764 B CN 108445764B CN 201810244210 A CN201810244210 A CN 201810244210A CN 108445764 B CN108445764 B CN 108445764B
- Authority
- CN
- China
- Prior art keywords
- platform
- load
- test piece
- stewart
- track
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The invention relates to an active compliance control strategy of a Stewart platform, and belongs to the field of control of Stewart platforms. And the active compliance control strategy of the Stewart platform is realized by taking a six-dimensional force sensor as an acquisition element and through a neural network. By combining the trajectory planning and the neural network, a relation model between the motion trajectory of the platform and the output load can be obtained under the condition that the design parameters of the platform are unknown. And inputting the expected load applied to the test piece into the model to obtain the motion track of the platform when the test piece reaches the expected load, and further realizing the control method of the Stewart platform by taking the load as a parameter through displacement control. Compared with the prior art, the method avoids the problems of complex decoupling and coupling in the traditional control aiming at the Stewart platform, and the design of the control strategy does not need prior information of equipment, so that the fault tolerance rate is high, and the interference of factors such as assembly errors, measurement errors and the like is avoided.
Description
Technical Field
The invention relates to the field of control of a Stewart platform, in particular to an active compliance control strategy of the Stewart platform, which is realized by taking a six-dimensional force sensor as an acquisition element and through a neural network. The invention can provide a simple and convenient active flexible control mode for the technical categories of loading, positioning and the like of the Stewart platform in the fields of flight simulators, automatic control, parallel machine tools, material performance test and the like.
Background
The parallel mechanism is a closed loop mechanism which is formed by connecting the movable platform and the fixed platform through more than two independent kinematic chains and is driven and controlled in a parallel mode. The advent of parallel mechanisms has been dating back to the 30 s of the 20 th century. The six-degree-of-freedom platform is used as a branch in a parallel robot mechanism and is the parallel mechanism which is most researched by scholars at home and abroad. The earliest article on parallel six-degree-of-freedom mechanisms was published by stewart.d in 1965, and was originally designed for tire testing, and such mechanical structures were then placed in an aircraft simulator as a flight simulator, and hence the kinematic mechanism was also referred to as the Stewart mechanism. The mechanism consists of a movable platform and a static platform which are connected through a four-bar linkage consisting of a revolute pair and a universal joint. The parallel mechanism has the characteristics of large bearing capacity, high micro-motion precision and small motion load, and has higher rigidity and stable structure compared with a serial mechanism. In the position solution, the serial mechanism is easy to solve positively, but difficult to solve negatively, and the parallel mechanism is difficult to solve positively and easy to solve negatively. Therefore, at present, theoretical research on parallel six-degree-of-freedom mechanisms mainly focuses on the fields of mechanistic research, kinematics, dynamics, control strategy research and the like, and relates to the aspects such as washout of input motion signals, pose positive and inverse solution, kinematics of mechanisms, singular configurations and the like.
Robot control is an important area in robotics research. When the track control of a single position inevitably generates some undesirable environmental contact forces, a force detection sensor is required to be introduced to detect the relevant information of the contact state of the manipulator and the environment, and the control of the robot which is adaptive to the environment is realized through the processing of response control, which has important significance for improving the performance of the robot, enhancing the adaptability of the robot and expanding the application range of the robot. One of the core contradictions in robotic research is: the high requirement of the robot on flexibility capable of generating any acting force when the robot operates in a specific contact environment and the high requirement of the robot on position servo rigidity and mechanical structure rigidity when the robot operates in a free space are contradictory. This ability of the robot to be compliant to the contact environment is referred to as compliance. To solve this conflict, a great deal of research is conducted at home and abroad, called compliance control research. The flexibility comprises active flexibility and passive flexibility. The mode of generating compliance to external forces by means of the auxiliary compliance mechanism is called passive compliance, and the mode of performing control by a certain control strategy according to feedback information by means of the force acquisition element is called active compliance, namely, force control. At present, the typical parallel six-degree-of-freedom mechanism force control is mainly carried out based on position control, and closed-loop feedback control is carried out by selecting elements such as a wrist force sensor as a detection tool or directly introducing a force signal and the like.
In conclusion, although many people have studied and proposed various control strategies in the aspect of force control of the Stewart platform, most of the control strategies adopt that sensors are distributed on all driving cylinders of the Stewart platform, and the total output load of the platform is obtained by coupling acquisition and control of single-cylinder output load; or aiming at the hydraulically driven Stewart platform, the output load is controlled by controlling parameters such as flow rate, flow speed and the like of each hydraulic cylinder, but under the condition of practical application, the parameters such as the motor rotating speed, the actual output torque and the like of the electrically driven Stewart platform are difficult to measure and cannot be controlled, so that the method is not suitable for the electrically driven Stewart platform. A control means for taking a Stewart platform as a loading device, collecting by a six-dimensional force sensor, applying the Stewart platform to the material performance test and applying specified load loading to a test piece to be tested has not been researched yet.
Disclosure of Invention
The invention aims to provide an active compliance control strategy of a Stewart platform, which is a control method taking load as a parameter, solves the current situation that most of the existing Stewart platform takes speed and displacement as control parameters, avoids the problems of complex decoupling and coupling in the traditional control aiming at the Stewart platform, and has the advantages of no need of prior information of equipment in the design of the control strategy, high fault tolerance rate and no interference of factors such as assembly errors, measurement errors and the like. The invention relates to an active compliance control strategy of a Stewart platform, which is realized by taking a six-dimensional force sensor as an acquisition element through a neural network. By combining the trajectory planning and the neural network, a relation model between the motion trajectory of the platform and the output load can be obtained under the condition that the design parameters of the platform are unknown. And inputting the expected load applied to the test piece into the model to obtain the motion track of the platform when the test piece reaches the expected load, and further realizing the control method of the Stewart platform by taking the load as a parameter through displacement control.
The above object of the present invention is achieved by the following technical solutions:
the loading device comprises a fixed platform 1 of a Stewart platform, a driving unit 2, a movable platform 3, a six-dimensional force sensor 4, an upper clamp 5, a lower clamp 7 and a test piece 6.
The active compliance control strategy of the Stewart platform controls the Stewart platform by taking a load as a parameter, and an application object is the Stewart platform which takes a six-dimensional force sensor as a force acquisition module, and the method comprises the following steps of:
step 1: trajectory planning
The moving platform 3 of the Stewart platform is used as a loading unit, and the moving platform 3 is controlled by displacement, so that the motion track of the moving platform 3 corresponding to the loaded test piece 6 in the process of applying a specific load is required to be obtained; firstly, taking a fixed end, namely the position of a lower clamp 7, as an initial coordinate system R, and calculating a motion track of a motion end, namely the position of an upper clamp 5, in the pure bending process of a test piece; the locus of the point A can be obtained by a flexible line equation and is set as C1(ii) a On the basis, as the AB section of the connecting piece is a rigid section, the track of the B point in the coordinate system R, namely the track of the central point of the moving platform 3 of the Stewart platform can be obtained, and the track of the central point of the moving platform 3 is set as C2(ii) a In the actual control of the Stewart platform, the center of the moving platform, namely the point B, is subjected to track control by taking the coordinate system R' of the platform as reference; therefore, C needs to be transformed by the coordinate system according to the spatial relationship between the coordinate systems R and R2Conversion into the central position of the movable platform in the coordinate system R', namely the point BAnd the locus of the point B is set as C3(ii) a The locus C3Namely inputting the actual motion track of the Stewart platform when the test piece 6 is subjected to composite loading;
step 2: error analysis
Because the force acquisition element adopts the six-dimensional force sensor 4, when the pose of the six-dimensional force sensor 4 changes, the six-dimensional force sensor 4 is influenced by gravity, so that an acquisition result has errors; therefore, the moving platform 3 is required to move along the required track C under the no-load condition3Moving and recording the load borne by the six-dimensional force sensor 4 in the process; the load is an error generated by the six-dimensional force sensor 4 due to the influence of gravity in the motion process;
and step 3: model training
Connecting the test piece 6 to be loaded with the movable platform 3 of the Stewart platform, and then connecting the motion trail C3Inputting the load to a Stewart platform, and acquiring the load of a test piece 6 in the motion process of the movable platform 3 through a six-dimensional force sensor 4; subtracting the error value obtained in the step 2 from the obtained load to obtain the real load borne by the test piece; with six components (F) of the true loadx Fy Fz Mx My Mz) Taking the number n of the collected load data as the number of columns to form a six-dimensional force matrix of 6 x n; calling a neural network tool box in matlab, setting a six-dimensional moment array as an input quantity, setting the motion track of the movable platform 3 as an output quantity, and establishing a relation between the output load of the Stewart platform and the motion track thereof to form a model of 6 input-hidden layer-6 output, wherein the conversion relation between the six-dimensional moment array and the motion track of the movable platform is a hidden layer; the algorithm selects a Feed-forward back prop (feedforward BP) neural network, has the function of realizing any complex nonlinear mapping, and has certain popularization and generalization capability in a generated network. The training mode adopts the method of TRAINSCG, namely a conjugate gradient method, which has the advantages of less time consumption, fast convergence and automatic stop when the convergence is not realized. In the aspect of the number of the neurons of the hidden layer, the method is based on an empirical formula
K＝(I+O)/2
Wherein K is the number of neurons, I is the number of input layer variables, and O is the number of output layer variables; the number of the neurons of the obtained hidden layer is 6; selecting default for other parameters, and finally establishing a neural network model;
and 4, step 4: model prediction
Inputting the load to be applied to the test piece into the trained model, and obtaining the corresponding motion track of the movable platform 3 under the loading condition; because the Stewart platform moves, the coupling and the cooperative motion of the driving cylinders of the driving unit 2 exist, and the path of the movable platform 3 reaching a certain specific attitude is uncertain. In order to enable the Stewart platform to output the specified load and avoid the uncertainty of the motion track from damaging the material performance of the test piece, the motion track of the movable platform 3 needs to be differentiated to ensure the accuracy of the motion path; the method adopted is that the target output load is subdivided into m parts to form six components (F) of the loadxFy Fz Mx My Mz) A 6 x m load matrix with m as columns and rows; and inputting the obtained 6 × m matrix into the trained model to obtain the motion track of the movable platform 3 under the specified load.
Calculating the motion track of a test piece to be loaded in the loading process, and obtaining the motion track of the Stewart platform when the test piece is expected to be loaded according to coordinate conversion; the movable platform 3 is driven to move along the required track in the no-load state, and the stress state of the sensor in the movement process is collected through the six-dimensional force sensor 4. This is the effect of gravity on the six-dimensional force sensor 4; the movable platform 3 is driven to move along the required track in the state of clamping the test piece 6, and the stress state of the test piece in the movement process is collected through the six-dimensional force sensor 4; the influence of the gravity on the six-dimensional force sensor 4 is reduced from the obtained load, and the real load borne by the test piece 6 is obtained; training the real load borne by the test piece 6 and the corresponding platform track through a neural network, establishing the relationship between the output load of the Stewart platform and the motion track of the Stewart platform, and obtaining a 6-input-hidden-layer-6 output model; and inputting the expected load into the trained model to obtain the motion trail of the platform under the loading condition, thereby realizing the active compliance control of the Stewart platform.
By the method, the moving track of the moving platform of the Stewart platform is obtained according to the moving track of the test piece to be loaded, a neural network model is established according to the relation between the moving track of the moving platform and the load borne by the test piece, the expected load is input into the trained model, and the moving track of the platform is reversely deduced when the platform outputs the expected load, so that the control mode of the Stewart platform taking the load as a parameter can be realized.
The invention has the beneficial effects that:
(a) the problems of complex decoupling and coupling in the traditional control aiming at the Stewart platform are solved, and the nonlinear control problem is effectively solved by utilizing the activation function adopted by the hidden layer neuron of the multilayer neural network.
(b) Conventional adaptive control methods require a priori information of the model to design the control scheme. Due to the approximation capability of the neural network, the controller does not need specific model information.
(c) Under the large-scale parallel processing architecture of the neural network, the damage of certain nodes of the network does not affect the overall performance of the whole neural network, and the fault tolerance rate of the control system is effectively improved.
(d) The influence brought by the assembly error, the measurement error and the like of the mechanism does not need to be considered.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention.
FIG. 1 is a Stewart platform and loading device of the present invention;
FIG. 2 is a geometric resolution method of pure bending loading according to the present invention;
FIG. 3 is a block diagram of a neural network model of the present invention;
FIG. 4 is a fitting result of neural network training of the present invention;
FIG. 5 is a graph of regression coefficient analysis of all samples of the neural network of the present invention.
Detailed Description
The details of the present invention and its embodiments are further described below with reference to the accompanying drawings.
Referring to fig. 1 to 5, according to the active compliance control strategy of the Stewart platform, the Stewart platform is controlled by taking a load as a parameter, and an application object is the Stewart platform which takes a six-dimensional force sensor as a force acquisition module. The invention combines the track planning and the neural network, and can obtain a relation model between the motion track of the platform and the output load under the condition that the design parameters of the platform are unknown. And inputting the expected load applied to the test piece into the model to obtain the motion track of the platform when the test piece reaches the expected load, and further realizing the control method of the Stewart platform by taking the load as a parameter through displacement control.
The method comprises the following steps:
step 1: trajectory planning
The Stewart platform is composed of a fixed platform 1, a driving unit 2 and a movable platform 3. Because the moving platform 3 of the Stewart platform is used as a loading unit and is controlled by displacement, the motion trail of the moving platform 3 corresponding to the loaded test piece 6 in the process of applying a specific load is required to be solved; firstly, taking a fixed end, namely the position of a lower clamp 6, as an initial coordinate system R, and calculating a motion track of a motion end, namely the position of an upper clamp 5, in the pure bending process of a test piece; the locus of the point A can be obtained by a flexible line equation and is set as C1(ii) a On the basis, as the AB section of the connecting piece is a rigid section, the track of the B point in the coordinate system R, namely the track of the central point of the moving platform 3 of the Stewart platform can be obtained, and the track of the central point of the moving platform 3 is set as C2(ii) a In the actual control of the Stewart platform, the center of the moving platform, namely the point B, is subjected to track control by taking the coordinate system R' of the platform as reference; therefore, C needs to be transformed by the coordinate system according to the spatial relationship between the coordinate systems R and R2Converting into the central position of the movable platform in the coordinate system R', namely the motion track of the point B, and setting the track of the point B as C3(ii) a The locus C3Namely inputting the actual motion track of the Stewart platform when the test piece 6 is subjected to composite loading;
step 2: error analysis
Because the force acquisition element adopts the six-dimensional force sensor 4, when the pose of the six-dimensional force sensor 4 changes, the six-dimensional force sensor can be aligned to the six-dimensional force under the action of gravityThe force sensor 4 generates influence to cause errors in the acquisition result; therefore, the moving platform 3 is required to move along the required track C under the no-load condition3Moving and recording the load borne by the six-dimensional force sensor 4 in the process; the load is an error generated by the six-dimensional force sensor 4 due to the influence of gravity in the motion process;
and step 3: model training
Connecting the test piece 6 to be loaded with the movable platform 3 of the Stewart platform, and then connecting the motion trail C3Inputting the load to a Stewart platform, and acquiring the load of a test piece 6 in the motion process of the movable platform 3 through a six-dimensional force sensor 4; subtracting the error value obtained in the step 2 from the obtained load to obtain the real load borne by the test piece 6; with six components (F) of the true loadx Fy Fz Mx My Mz) Taking the number n of the collected load data as the number of columns to form a six-dimensional force matrix of 6 x n; calling a neural network tool box in matlab, setting a six-dimensional moment array as an input quantity, setting the motion track of the movable platform 3 as an output quantity, and establishing a relation between the output load of the Stewart platform and the motion track thereof to form a model of 6 input-hidden layer-6 output, wherein the conversion relation between the six-dimensional moment array and the motion track of the movable platform is a hidden layer; the algorithm selects a Feed-forward back prop (feedforward BP) neural network, has the function of realizing any complex nonlinear mapping, and has certain popularization and generalization capability in a generated network. The training mode adopts the method of TRAINSCG, namely a conjugate gradient method, which has the advantages of less time consumption, fast convergence and automatic stop when the convergence is not realized. In the aspect of the number of the neurons of the hidden layer, the method is based on an empirical formula
K＝(I+O)/2
Wherein K is the number of neurons, I is the number of input layer variables, and O is the number of output layer variables; the number of the neurons of the obtained hidden layer is 6; selecting default for other parameters, and finally establishing a neural network model;
and 4, step 4: model prediction
Inputting the load to be applied to the test piece into the trained model, and obtaining the corresponding motion track of the movable platform 3 under the loading condition; due to the presence of drive cylinders of the drive unit 2 between the Stewart platform when it is in motionCoupled and coordinated motion, the path of the moving platform 3 to a particular pose is uncertain. In order to enable the Stewart platform to output the specified load and avoid the uncertainty of the motion track from damaging the material performance of the test piece, the motion track of the movable platform 3 needs to be differentiated to ensure the accuracy of the motion path; the method adopted is that the target output load is subdivided into m parts to form six components (F) of the loadxFy Fz Mx My Mz) A 6 x m load matrix with m as columns and rows; and inputting the obtained 6 × m matrix into the trained model to obtain the motion track of the movable platform 3 under the specified load.
Calculating the motion track of a test piece to be loaded in the loading process, and obtaining the motion track of the Stewart platform when the test piece is expected to be loaded according to coordinate conversion; the movable platform 3 is driven to move along the required track in the no-load state, and the stress state of the sensor in the movement process is collected through the six-dimensional force sensor 4. This is the effect of gravity on the six-dimensional force sensor 4; the movable platform 3 is driven to move along the required track in the state of clamping the test piece 6, and the stress state of the test piece in the movement process is collected through the six-dimensional force sensor 4; the influence of the gravity on the six-dimensional force sensor 4 is reduced from the obtained load, and the real load borne by the test piece 6 is obtained; training the real load borne by the test piece 6 and the corresponding platform track through a neural network, establishing the relationship between the output load of the Stewart platform and the motion track of the Stewart platform, and obtaining a 6-input-hidden-layer-6 output model; and inputting the expected load into the trained model to obtain the motion trail of the platform under the loading condition, thereby realizing the active compliance control of the Stewart platform.
By calculating the track in the test piece loading process and according to coordinate transformation, the motion track of the Stewart platform in the test piece loading process can be obtained; in the no-load state, the platform moves along the track to obtain an error value generated by the sensor under the influence of gravity; in the test piece clamping state, the platform moves along the track to obtain the load borne by the test piece; subtracting the error value of the sensor obtained in the no-load state from the load borne by the test piece to obtain the real load borne by the test piece in the loading process; training the real load borne by the test piece and the corresponding platform track through a neural network to obtain a model of 6 input-hidden layer-6 output, namely a relation model between the platform motion track and the load borne by the test piece; and inputting the expected load into the trained model to obtain the platform track under the loading condition. And inputting the motion track into a Stewart platform, so that the platform can apply specified load to the test piece.
The method provides a control strategy taking load as a parameter for a Stewart platform taking a six-dimensional force sensor as a force acquisition element.
Example (b):
referring to fig. 1 to 5, a specified load is applied to the test piece. Here, a pure bending load is taken as an example.
Step 1: and planning a track. Because the moving platform of the Stewart platform is used as a loading unit, the motion track of the moving platform in the bending process of the test piece needs to be calculated. Firstly, a fixed end, namely a lower clamp position, is taken as an initial coordinate system R, and a motion track of a motion end, namely an upper clamp position, in the pure bending process of the test piece is calculated. From pure bending deflection equation
The track C of the point A in the coordinate system R can be obtained1Is composed of
Y＝0 (2)
U＝0 (4)
V＝α (5)
Z＝0 (6)
In the formula (f)BFor bending deflection, alpha is the bending angle, MeIs a bending moment; e and I are the modulus of elasticity and the cross section of the test piece, respectivelyMoment of inertia; x, Y, Z are displacements in the x, y, z directions in the coordinate system, respectively; u, V, W are the angles of rotation in the coordinate system about the x, y, and z axes, respectively.
On the basis, as the AB section of the connecting piece is a rigid section, the track of a point B in a coordinate system R, namely the central point of a moving platform of the Stewart platform can be obtained, and the track of the point is set as C2Is a
Y＝0 (8)
U＝0 (10)
V＝α (11)
W＝0 (12)
In the actual control of the Stewart platform, the center of the moving platform, namely the point B, is subjected to track control by taking the coordinate system R' of the platform as a reference. Therefore, it is necessary to transform the coordinate system according to the spatial relationship between the coordinate systems R and R
C is to be2Converting into the central position of the movable platform in the coordinate system R', namely the motion track of the point B, and setting the track of the point as C3. The track is the actual motion track input by the platform when the test piece is subjected to composite loading. Wherein θ is the angle between the coordinate systems R and R'. [ X ' Y ' Z ']TRespectively, displacements in x ', y', z 'directions in the coordinate system R'.
Step 2: and (5) error analysis. Because the force acquisition element adopts the six-dimensional force sensor 4, when the pose of the six-dimensional force sensor 4 changes, the six-dimensional force sensor 4 is influenced by gravityCausing errors in the acquisition results; therefore, the moving platform 3 is required to move along the required track C under the no-load condition3Moving and recording the load borne by the six-dimensional force sensor 4 in the process; the load is an error generated by the six-dimensional force sensor 4 due to the influence of gravity during the movement process.
And step 3: and (5) training a model. Connecting the test piece 6 to be loaded with the movable platform 3 of the Stewart platform, and then connecting the motion trail C3Inputting the load to a Stewart platform, and acquiring the load of a test piece 6 in the motion process of the movable platform 3 through a six-dimensional force sensor 4; subtracting the error value obtained in the step 2 from the obtained load to obtain the real load borne by the test piece 6; with six components (F) of the true loadx FyFz Mx My Mz) And (4) forming a six-dimensional force matrix of 6 x n by taking the number n of the collected load data as the number of rows and the number of columns, and establishing the relation between the motion track of the platform and the output load through Matlab.
The Matlab operational flow involved is as follows:
(a) and inputting an nntool command in a command window and entering a neural network module. The load matrix is input into an [ Inputs ] window, and the motion trajectory matrix is input into a [ Targets ] window. And clicking [ New Network ], and establishing a three-layer neural Network with a hidden layer. The algorithm selects a Feed-forward back prop (feedforward BP) neural network, has the function of realizing any complex nonlinear mapping, and has certain popularization and generalization capability in a generated network. The training mode adopts the method of TRAINSCG, namely a conjugate gradient method, which has the advantages of less time consumption, fast convergence and automatic stop when the convergence is not realized. In the aspect of the number of the neurons of the hidden layer, the method is based on an empirical formula
K＝(I+O)/2
Wherein K is the number of neurons, I is the number of input layer variables, and O is the number of output layer variables. This gives a number of neurons in the hidden layer of 6. In other parameters, TANSIL is used for the Layer 1Transfer function, and LOGSIG is used for the Layer 2Transfer function. After all parameters are entered, the point Create button is created.
(b) Click the Train button to get to the tracking info tab. The load matrix is set as an input quantity and the motion matrix is set as an output quantity. Parameter filling is done in the Train Parameters tab. Wherein, the iteration times [ epochs ] is selected to be 1000, or self-adjusted according to the convergence result. The target error [ gold ] can be adjusted by itself according to the number of samples per gold < 0.5. When the result converges, i.e. the training error is not greater than the standard error, the training is successful.
And 4, step 4: inputting the load to be applied to the test piece into the trained model, and obtaining the corresponding motion track of the movable platform 3 under the loading condition; because the Stewart platform moves, the coupling and the cooperative motion of the driving cylinders of the driving unit 2 exist, and the path of the movable platform 3 reaching a certain specific attitude is uncertain. In order to enable the Stewart platform to output the specified load and avoid the uncertainty of the motion track from damaging the material performance of the test piece, the motion track of the movable platform 3 needs to be differentiated to ensure the accuracy of the motion path; the method adopted is that the target output load is subdivided into m parts to form six components (F) of the loadx Fy Fz Mx My Mz) A 6 x m load matrix with m as columns and rows; and inputting the obtained 6 × m matrix into the trained model to obtain the motion track of the movable platform 3 under the specified load.
And (5) model prediction. Inputting the load to be applied to the test piece into the trained model, and obtaining the corresponding motion track of the movable platform 3 under the loading condition; because the Stewart platform moves, the coupling and the cooperative motion of the driving cylinders of the driving unit 2 exist, and the path of the movable platform 3 reaching a certain specific attitude is uncertain. In order to enable the Stewart platform to output the specified load and avoid the uncertainty of the motion track from damaging the material performance of the test piece, the motion track of the movable platform 3 needs to be differentiated to ensure the accuracy of the motion path; when applying pure bending load, let Fx、Fy、Fz、My、MzAlways is zero, will MxThe load is divided into m parts to form a load matrix of 6 x m. And inputting the load matrix into the trained model for prediction to obtain a group of motion tracks of the movable platform. Inputting the motion track into a Stewart platform, so that the Stewart platform can apply specified load to the test piece。
The above description is only a preferred example of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like of the present invention shall be included in the protection scope of the present invention.
Claims (2)
1. An active compliance control strategy for a Stewart platform, comprising: the Stewart platform is controlled by taking the load as a parameter, and an application object is the Stewart platform which takes a six-dimensional force sensor as a force acquisition module, and the method comprises the following steps:
step 1: trajectory planning
The Stewart platform is composed of a fixed platform (1), a driving unit (2) and a movable platform (3), and the movable platform (3) of the Stewart platform is used as a loading unit, and the movable platform (3) is controlled through displacement, so that the motion track of the movable platform (3) corresponding to the loaded test piece (6) in the process of applying a specific load is required to be obtained; firstly, taking a fixed end, namely the position of a lower clamp (7), as an initial coordinate system R, and calculating a motion track of a motion end, namely the position of an upper clamp (5), in the pure bending process of a test piece; the locus of the point A can be obtained by a flexible line equation and is set as C1(ii) a On the basis, as the AB section of the connecting piece is a rigid section, the track of the B point in the coordinate system R, namely the central point of the moving platform (3) of the Stewart platform, can be obtained, and the track of the central point of the moving platform (3) is set as C2(ii) a In the actual control of the Stewart platform, the center of the moving platform, namely the point B, is subjected to track control by taking the coordinate system R' of the platform as reference; therefore, C needs to be transformed by the coordinate system according to the spatial relationship between the coordinate systems R and R2Converting into the central position of the movable platform in the coordinate system R', namely the motion track of the point B, and setting the track of the point B as C3(ii) a The locus C3Namely, the actual motion track of the Stewart platform is input when the test piece (6) is subjected to composite loading;
step 2: error analysis
Because the force acquisition element adopts a six-dimensional force sensor (4), the six-dimensional force sensor(4) When the pose changes, due to the gravity action, the pose influences the six-dimensional force sensor (4), so that errors exist in the acquisition result; therefore, the movable platform (3) is required to move along the required track C under the no-load condition3Moving and recording the load borne by the six-dimensional force sensor (4) in the process; the load is an error generated by the six-dimensional force sensor (4) due to the influence of gravity in the motion process;
and step 3: model training
Connecting a test piece (6) to be loaded with a movable platform (3) of a Stewart platform, and then connecting a motion track C3Inputting the load to a Stewart platform, and acquiring the load borne by a test piece (6) in the moving process of the movable platform (3) through a six-dimensional force sensor (4); subtracting the error value obtained in the step 2 from the obtained load to obtain the real load borne by the test piece; with six components (F) of the true loadxFyFzMx My Mz) Taking the number n of the collected load data as the number of columns to form a six-dimensional force matrix of 6 x n; calling a neural network tool box in matlab, setting a six-dimensional moment array as an input quantity, setting a motion track of a movable platform (3) as an output quantity, and establishing a relation between an output load of a Stewart platform and the motion track of the Stewart platform to form a model of 6 input-hidden layer-6 output, wherein a conversion relation between the six-dimensional moment array and the motion track of the movable platform is a hidden layer; wherein the algorithm selects Feed-forward back prop, namely a feedforward BP neural network, and the training mode selects TRAINSCG, namely a conjugate gradient method; in the aspect of the number of the neurons of the hidden layer, the formula
K=(I+O)/2
Wherein K is the number of neurons, I is the number of input layer variables, and O is the number of output layer variables; the number of the neurons of the obtained hidden layer is 6; selecting default for other parameters, and finally establishing a neural network model;
and 4, step 4: model prediction
Inputting the load to be applied to the test piece into the trained model, and obtaining the corresponding motion track of the movable platform (3) under the loading condition; because the coupling and the cooperative motion of the driving cylinders of the driving unit (2) exist when the Stewart platform moves, the path of the movable platform (3) reaching a certain specific attitude is notDetermining; in order to enable the Stewart platform to output the specified load and avoid the uncertainty of the motion track from damaging the material performance of the test piece, the motion track of the movable platform (3) needs to be differentiated to ensure the accuracy of the motion path; the method adopted is that the target output load is subdivided into m parts to form six components (F) of the loadxFyFzMx My Mz) A 6 x m load matrix with m as columns and rows; and inputting the obtained 6-m matrix into a trained model to obtain the motion trail of the movable platform (3) under the specified load.
2. The active compliance control strategy of Stewart platform of claim 1, wherein: the motion track of the movable platform (3) in the loading process of the test piece (6) to be loaded can be obtained by calculating the motion track of the test piece (6) to be loaded in the loading process and converting the coordinates; the movable platform (3) is made to move along the required track in an idle state, and the stress state of the test piece (6) in the movement process is collected through the six-dimensional force sensor (4); this is the effect of gravity on the six-dimensional force sensor (4); the movable platform (3) is enabled to move along the track under the state of clamping the test piece, and the stress state of the test piece in the movement process is collected through the six-dimensional force sensor (4); the influence of the load on the six-dimensional force sensor (4) is reduced, and the real load borne by the test piece is obtained; training the real load borne by the test piece and the motion trail of the corresponding movable platform (3) through a neural network to obtain a 6 input-hidden layer-6 output model, and establishing a relation model of the motion trail and the load borne by the test piece; and inputting the expected load into the trained model to obtain the motion trail of the platform under the loading condition, thereby realizing the active compliance control of the Stewart platform.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810244210.3A CN108445764B (en) | 2018-03-23 | 2018-03-23 | Active compliance control strategy of Stewart platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810244210.3A CN108445764B (en) | 2018-03-23 | 2018-03-23 | Active compliance control strategy of Stewart platform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108445764A CN108445764A (en) | 2018-08-24 |
| CN108445764B true CN108445764B (en) | 2021-01-12 |
Family
ID=63196755
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810244210.3A Active CN108445764B (en) | 2018-03-23 | 2018-03-23 | Active compliance control strategy of Stewart platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108445764B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109940605B (en) * | 2019-01-29 | 2021-04-20 | 中国工程物理研究院激光聚变研究中心 | Compliant assembly system and method integrating three-dimensional vision and contact force analysis |
| CN109760063B (en) * | 2019-03-15 | 2021-02-05 | 京东方科技集团股份有限公司 | Parallel robot control method, device, equipment and storage medium |
| CN110030234B (en) * | 2019-04-30 | 2023-01-13 | 北京航空航天大学 | Interactive safety control strategy based on hydraulic compliant driver |
| CN113967918A (en) * | 2021-11-19 | 2022-01-25 | 伯朗特机器人股份有限公司 | Parallel six-axis robot position and posture correction method based on BP neural network |
| CN114831847B (en) * | 2022-03-29 | 2023-06-09 | 中国农业大学 | Four-branched-chain parallel mechanism neck rehabilitation training robot and force control method thereof |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101451933A (en) * | 2008-12-18 | 2009-06-10 | 北京林业大学 | Six-dimension loading material mechanical test machine |
| CN105589558A (en) * | 2015-11-25 | 2016-05-18 | 南京航空航天大学 | Virtual reality helmet burden alleviation servo system and method based on surface electromyogram signal |
| CN106774362A (en) * | 2016-11-25 | 2017-05-31 | 集美大学 | The tank test control method and system of a kind of flexible six-degree-of-freedom wire saws |
| CN107121930A (en) * | 2017-05-26 | 2017-09-01 | 山东科技大学 | Six Degree-of-Freedom Parallel Platform motion control method based on dummy model |
| CN109094817A (en) * | 2018-08-29 | 2018-12-28 | 哈尔滨工业大学 | The adaptive undercarriage of carrier-based helicopter warship simulation system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7904187B2 (en) * | 1999-02-01 | 2011-03-08 | Hoffberg Steven M | Internet appliance system and method |
-
2018
- 2018-03-23 CN CN201810244210.3A patent/CN108445764B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101451933A (en) * | 2008-12-18 | 2009-06-10 | 北京林业大学 | Six-dimension loading material mechanical test machine |
| CN105589558A (en) * | 2015-11-25 | 2016-05-18 | 南京航空航天大学 | Virtual reality helmet burden alleviation servo system and method based on surface electromyogram signal |
| CN106774362A (en) * | 2016-11-25 | 2017-05-31 | 集美大学 | The tank test control method and system of a kind of flexible six-degree-of-freedom wire saws |
| CN107121930A (en) * | 2017-05-26 | 2017-09-01 | 山东科技大学 | Six Degree-of-Freedom Parallel Platform motion control method based on dummy model |
| CN109094817A (en) * | 2018-08-29 | 2018-12-28 | 哈尔滨工业大学 | The adaptive undercarriage of carrier-based helicopter warship simulation system |
Non-Patent Citations (5)
| Title |
|---|
| Closed-form force sensing of a 6-axis force transducer based on the Stewart platform;Chul-Goo Kang;《Sensors and Actuators》;20001211;第31-37页 * |
| Method of designing a six-axis force sensor for stiffness decoupling based on Stewart platform;Songlin Zhou;《Measurement》;20190822;第1-10页 * |
| Stewart六自由度并联平台动力学模型振动分析;钱承 等;《湖南大学学报》;20160225;第36-42页 * |
| 机器人运动中六维力传感器的重力补偿研究;王志军 等;《机械设计与制造》;20180708;第252-255页 * |
| 驱动冗余重型并联机构的动力学性能;许金凯 等;《吉林大学学报》;20170715;第1138-1143页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108445764A (en) | 2018-08-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108445764B (en) | Active compliance control strategy of Stewart platform | |
| CN109382828B (en) | Robot shaft hole assembling system and method based on teaching learning | |
| CN103495977B (en) | 6R-type industrial robot load identification method | |
| Wilson et al. | Relative end-effector control using cartesian position based visual servoing | |
| CN110977992B (en) | Non-kinematic model trajectory tracking method for mechanical arm and mechanical arm system | |
| CN113601509B (en) | Multi-degree-of-freedom mechanical arm flexible control method and system | |
| CN110161850A (en) | A kind of identification of industrial robot variable element rigidity and modeling method | |
| CN109623827B (en) | High-performance redundant manipulator repetitive motion planning method and device | |
| CN111639749A (en) | Industrial robot friction force identification method based on deep learning | |
| CN110253574A (en) | A kind of detection of multitask mechanical arm pose and error compensating method | |
| CN113500599A (en) | Track tracking method and system of rope-driven mechanical arm based on neural network | |
| CN114942593A (en) | Mechanical arm self-adaptive sliding mode control method based on disturbance observer compensation | |
| CN113927603B (en) | Mechanical arm dragging control method and device, computer equipment and storage medium | |
| Parvaresh et al. | Dynamics and path tracking of continuum robotic arms using data-driven identification tools | |
| CN116214516A (en) | Real-time grabbing and track tracking method and system for mobile redundant mechanical arm | |
| Luo et al. | Segmented hybrid motion-force control for a hyper-redundant space manipulator | |
| JP2629291B2 (en) | Manipulator learning control method | |
| CN112847351B (en) | Parallel robot trajectory tracking control method based on exponential approach rate | |
| CN113134839B (en) | Robot precision flexible assembly method based on vision and force position image learning | |
| Wei et al. | Sensorimotor coordination and sensor fusion by neural networks | |
| Pajak et al. | Planning of a point to point collision-free trajectory for mobile manipulators | |
| Motion control method of multi degree of freedom industrial robot for intelligent manufacturing | ||
| CN113043269A (en) | Robot contact force observation system based on robot model | |
| CN117584137B (en) | Random-working-condition-oriented hydraulic mechanical arm load model online correction method and system | |
| Yang et al. | Machine Leaning-Based Method for Kinematics Parameters Identification of Twin-Pivot Cable-Driven Continuum Robots |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |