CN108445764A - The Active Compliance Control strategy of Stewart platforms - Google Patents
The Active Compliance Control strategy of Stewart platforms Download PDFInfo
- Publication number
- CN108445764A CN108445764A CN201810244210.3A CN201810244210A CN108445764A CN 108445764 A CN108445764 A CN 108445764A CN 201810244210 A CN201810244210 A CN 201810244210A CN 108445764 A CN108445764 A CN 108445764A
- Authority
- CN
- China
- Prior art keywords
- load
- moving platform
- test specimen
- movement locus
- stewart platforms
- 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.)
- Granted
Links
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
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of Active Compliance Control strategies of Stewart platforms, belong to the control field of Stewart platforms.Using six-dimension force sensor as acquisition elements, pass through the Active Compliance Control strategy of the Stewart platforms of neural fusion.In conjunction with trajectory planning and neural network, the relational model between platform movement locus and output loads can be obtained in the case of Platform Designing unknown parameters.After it is expected the load input model applied to test specimen again, you can obtain the movement locus of platform when test specimen being made to reach anticipated load, and then realize Stewart platforms using load as the control method of parameter by Bit andits control.Compared with prior art, the present invention avoid conventional needle in the control of Stewart platforms complicated decoupling and coupled problem, and prior information of the design of control strategy without equipment, serious forgiveness is high, is not interfered by factors such as rigging error, measurement errors.
Description
Technical field
It is the present invention relates to the control field of Stewart platforms, more particularly to a kind of using six-dimension force sensor as acquisition member
Part, the Active Compliance Control strategy by the Stewart platforms of neural fusion.The present invention can be flight simulator, automatic
The technologies scope such as load, positioning of Stewart platforms involved by the fields such as control, parallel machine, material properties test, provides
A kind of Active Compliance Control mode of simplicity.
Background technology
Moving platform and fixed platform are attached by parallel institution by more than two independent kinematic chains, form one
Kind carries out the close loop mechanism of drive control with parallel way.The appearance of parallel institution can date back the 1930s.Wherein
Six degree of freedom platform is the parallel institution of domestic and foreign scholars' most study as a branch in parallel robot mechanism.Most
The early article in relation to 6-dof parallel mechanism is delivered by Stewart.D in nineteen sixty-five, is originally designed for Tire testing, afterwards again
This kind of mechanical structure is placed in plane simulation device and is used as flight simulator, thus this motion is also known as Stewart
Mechanism.The mechanism is made of moving platform and silent flatform, four companies being made up of a revolute and a universal joint therebetween
The connection that linkage carries out.Parallel institution has the characteristics that large carrying capacity, fine motion precision are high, exercise load is small, and with connect
Mechanism compares, the rigidity with bigger, stable structure.On position solves, serial mechanism normal solution is easy, but anti-solution is very tired
Difficulty, and the difficult anti-solution of parallel institution normal solution is very easy to.Therefore, at present for the theoretical research of 6-dof parallel mechanism,
The fields such as theory of mechanisms, kinematics, dynamics and control strategy study are focused primarily upon, " washing such as input motion signal is related to
Go out ", pose forward and reverse solution, the kinematics of mechanism and Singularity etc..
Robot control is a key areas in robotics research.When the TRAJECTORY CONTROL of single position is inevitable
When ground generates some undesirable environment contact forces, it is desirable to force detection sensor is introduced, to detect that manipulator connects with environment
The state of touching for information about, by the processing of response control, realizes the adaptable control of machine human and environment, this is for improving machine
Device human nature energy, expands its application range and has great importance the adaptability for enhancing robot.Core lance in robot research
One of shield is：Robot exists when specific contact environment is operated to that can generate arbitrary active force high request flexible and robot
Contradiction when free space operates between the high request of position servo stiffness and mechanical structure rigidity.Robot can be to contact
This ability of environment compliance is referred to as compliance.To solve this contradiction, a large amount of research has been carried out both at home and abroad, it is referred to as soft
Sequence system is studied.Compliance has two kinds of active compliance and passive compliance.External force is generated by auxiliary compliant mechanism suitable
From mode be known as passive compliance, and by power acquisition elements, according to feedback information, controlled by certain control strategy
Mode be known as active compliance, i.e. so-called power control.Typical 6-dof parallel mechanism power control is mainly base at present
In position, control carries out, by select the elements such as wrist force sensor to be used as detection instrument or be introduced directly into the modes such as force signal into
Row closed loop feedback control.
To sum up, although the power prosecutor face about Stewart platforms has more people and studies and propose various different control strategies,
But most of is to use sensor being distributed on each driving cylinder of Stewart platforms, by being adopted to single cylinder output loads
Collection with control couple and then obtains the total output loads of platform；Or the Stewart platforms for hydraulic-driven, pass through control
The parameters such as flow, the flow velocity of each hydraulic cylinder carry out the control of output loads, but under the conditions of practical application, and it is flat that electricity drives Stewart
The parameters such as motor speed, the actual output torque of platform are difficult to measure, and can not be controlled, therefore this method is not suitable for electricity drive
Stewart platforms.It using Stewart platforms as loading device, is acquired by six-dimension force sensor, is applied to material property and surveys
In terms of examination and treats test block and apply and specify the control means of load load there has been no researchs.
Invention content
It is that one kind is with load the purpose of the present invention is to provide a kind of Active Compliance Control strategy of Stewart platforms
The control method of parameter, solving existing Stewart platforms is kept away with the present situation of speed, displacement parameter in order to control
Exempted from conventional needle in the control of Stewart platforms complicated decoupling and coupled problem, and the design of control strategy is not necessarily to equipment
Prior information, serious forgiveness is high, is not interfered by factors such as rigging error, measurement errors.The present invention is with six-dimension force sensor
As acquisition elements, pass through the Active Compliance Control strategy of the Stewart platforms of neural fusion.In conjunction with trajectory planning and god
Through network, the relational model between platform movement locus and output loads can be obtained in the case of Platform Designing unknown parameters.
After it is expected the load input model applied to test specimen again, you can obtain the movement rail of platform when test specimen being made to reach anticipated load
Mark, and then realize Stewart platforms using load as the control method of parameter by Bit andits control.
The above-mentioned purpose of the present invention is achieved through the following technical solutions：
The loading device, by the fixed platform 1 of Stewart platforms, driving unit 2, moving platform 3, six-dimension force sensor
4, upper fixture 5, lower fixture 7, the composition of test specimen 6.
The Active Compliance Control strategy of Stewart platforms controls Stewart platforms using load as parameter, application
Object is the Stewart platforms using six-dimension force sensor as power acquisition module, is included the following steps：
Step 1：Trajectory planning
Due to being loading unit with the moving platform 3 of Stewart platforms, moving platform 3 is controlled by displacement, therefore
Demand goes out to apply the test specimen 6 of load the movement locus of corresponding moving platform 3 during specific load；First with fixing end,
Descend at fixture 7 is initial coordinate system R, tache motorice during calculation testing piece pure bending, i.e. movement locus at upper fixture 5；By
Deflection curve equation can obtain the A locus of points, and set the A locus of points as C1；It on this basis, can since AB sections of connector is rigid section
The B points in coordinate system R, the i.e. track of 1 center of moving platform of Stewart platforms are asked, and sets the central point rail of the moving platform 3
Mark is C2；And in the practical control of Stewart platforms, be with platform local Coordinate System R ' be reference, to moving platform center, that is, B
Point carries out TRAJECTORY CONTROL；It therefore need to be further according to the spatial relationship of coordinate system R and R ', by coordinate system transformation, by C2It is converted to
The middle moving platform centers coordinate system R ', i.e. the movement locus of B points, and the B locus of points are set as C3；Track C3As make test specimen 6
Inputted when Combined Loading the actual motion track of Stewart platforms；
Step 2：Error analysis
Since power acquisition elements are using six-dimension force sensor 4, when 4 pose of such six-dimension force sensor changes, due to
, by gravity, can be had an impact to six-dimension force sensor 4 leads to collection result there are errors for it；Therefore it needs first in idle condition
Under, make moving platform 3 along required track C3Movement, and it is loaded to record 4 institute of six-dimension force sensor in the process；This load is
The error that six-dimension force sensor 4 is generated due to gravity influence during the motion；
Step 3：Model training
After test specimen 6 to be loaded is connect with the moving platform 3 of Stewart platforms, by movement locus C3It is flat to input Stewart
Platform, and it is loaded by 6 institute of test specimen in acquisition 3 motion process of moving platform of six-dimension force sensor 4；Gained load is cut into step 2
The error amount of middle gained, as real load suffered by test specimen；With six component (F of real loadxFyFzMx My Mz) it is line number,
Using the load data number n acquired as columns, the six-dimensional force matrix of a 6*n is formed；Call the neural network in matlab
Six-dimensional force matrix is set as input quantity by tool box, and the movement locus of moving platform 3 is set as output quantity, and transformational relation between the two is
Hidden layer, the output loads for establishing Stewart platforms move relationship between track, form what a 6 inputs-hidden layer -6 exported
Model；Wherein, algorithm selects Feed-forward backprop i.e. feedforward BP neural network, the algorithm advantage to be to have real
The function of what incumbent complex nonlinear mapping, and network is generated with certain popularization, abstract ability.Training method is selected
TRAINSCG, i.e. conjugate gradient method, the method take less, and convergence is fast, and can be automatically stopped when not restraining.In the nerve of hidden layer
In terms of first number, by empirical equation
K=(I+O)/2
Wherein K is neuron number, and I is input layer variable number, and O is output layer variable number；The neuron of hidden layer can be obtained
Number is 6；Remaining parameter selection is given tacit consent to, and finally establishes neural network model；
Step 4：Model prediction
The quasi- load applied to test specimen is inputted into trained model, you can obtain corresponding moving platform under the loading environment
3 movement locus；When being moved due to Stewart platforms there are between 2 each driving cylinder of driving unit coupling and cooperative motion, move
The path that platform 3 reaches a certain particular pose is uncertain.Stewart platforms output to be made specifies load and avoids movement rail
The uncertainty of mark damages material for test performance, need to carry out differential to the movement locus of moving platform 3, it is ensured that motion path
Accuracy；Therefore target output loads is are subdivided into m parts, six components of the formation one with load by the method that uses
(FxFyFzMx My Mz) it is line number, using m as the loading matrix of the 6*m of columns；By the 6*m Input matrixes of gained trained mould
In type, you can obtain the movement locus of moving platform 3 under the specified load.
By calculating the movement locus of test specimen to be loaded in loading procedure, is converted according to coordinate, test specimen to be made can be obtained
Obtain the movement locus of Stewart platforms when expected load；It enables moving platform 3 be moved in the unloaded state along required track, leads to
Six-dimension force sensor 4 is crossed, force acting on transducer state in motion process is acquired.This is influence of the gravity to six-dimension force sensor 4；It enables
Moving platform 3 moves in the state of clamping test specimen 6 along required track, by six-dimension force sensor 4, acquires test specimen in motion process
Stress；It is real load suffered by test specimen 6 that gained load, which is cut influence of the gravity to six-dimension force sensor 4,；By test specimen 6
Suffered real load is trained with corresponding platform track by neural network, and the output loads of Stewart platforms are established
Relationship between track is moved, an output model of 6 inputs-hidden layer -6 is obtained,；Desired load is inputted into trained mould
Type obtains the platform movement locus under the loading environment, and then can realize the Active Compliance Control of Stewart platforms.
By the above-mentioned moving platform movement locus for finding out Stewart platforms by test specimen movement locus to be loaded, further according to dynamic
Platform movement locus and test specimen it is loaded between relationship establish neural network model, and desired load inputted trained
Model, which is instead released, makes platform output it is expected the method for the movement locus of platform when load, it can be achieved that Stewart platforms are with load
The control model of parameter.
The beneficial effects of the present invention are：
(a) avoid conventional needle in the control of Stewart platforms complicated decoupling and coupled problem, utilize multilayer nerve
The activation primitive that the hidden layer neuron of network uses, effectively solves the problems, such as such nonlinear Control.
(b) traditional self-adaptation control method needs the prior information of model to design control program.Due to neural network
Approximation capability, controller do not need specific model information.
(c) under the MPP framework of neural network, certain nodes damage of network has no effect on entire god
Overall performance through network effectively increases the serious forgiveness of control system.
(d) it is not necessarily to consider the influence that rigging error, measurement error of mechanism etc. are brought.
Description of the drawings
Attached drawing described herein is used to provide further understanding of the present invention, and is constituted part of this application, this hair
Bright illustrative example and its explanation is not constituted improper limitations of the present invention for explaining the present invention.
Fig. 1 is the Stewart platform and loading device of the present invention；
Fig. 2 is the geometrical analysis method of the pure bending load of the present invention；
Fig. 3 is the structure chart of the neural network model of the present invention；
Fig. 4 is the neural metwork training fitting result of the present invention；
Fig. 5 is the neural network whole sample regression coefficient analysis chart of the present invention.
Specific implementation mode
The detailed content and its specific implementation mode further illustrated the present invention below in conjunction with the accompanying drawings.
Shown in Fig. 1 to Fig. 5, the Active Compliance Control strategy of Stewart platforms of the invention, using load as parameter pair
Stewart platforms are controlled, and application is the Stewart platforms using six-dimension force sensor as power acquisition module.The present invention
In conjunction with trajectory planning and neural network, platform movement locus and output can be obtained in the case of Platform Designing unknown parameters
Relational model between load.After it is expected the load input model applied to test specimen again, you can acquisition makes test specimen reach expected load
The movement locus of platform when lotus, and then realize Stewart platforms using load as the control method of parameter by Bit andits control.
Include the following steps：
Step 1：Trajectory planning
Stewart platforms are made of fixed platform 1, driving unit 2, moving platform 3.Due to the moving platform 3 of Stewart platforms
It is controlled by displacement for loading unit and the platform, therefore demand goes out and applies specific load mistake to the test specimen 6 of load
The movement locus of corresponding moving platform 3 in journey；First with fixing end, that is, it is initial coordinate system R, calculation testing piece to descend at fixture 6
Tache motorice during pure bending, i.e. movement locus at upper fixture 5；The A locus of points can be obtained by deflection curve equation, and set the A point rails
Mark is C1；On this basis, since AB sections of connector is rigid section, the B points in coordinate system R can be sought, i.e., Stewart platforms is dynamic
The track of 1 center of platform, and the central point locus of the moving platform 3 is set as C2；And in the practical control of Stewart platforms
In, be with platform local Coordinate System R ' it is reference, TRAJECTORY CONTROL is carried out to moving platform center, that is, B points；It therefore need to be further according to coordinate
It is R and R ' spatial relationship, by coordinate system transformation, by C2Be converted to the middle moving platform centers coordinate system R ', i.e. B points
Movement locus, and the B locus of points are set as C3；Track C3Input Stewart platforms when test specimen 6 as being made to carry out Combined Loading
Actual motion track；
Step 2：Error analysis
Since power acquisition elements are using six-dimension force sensor 4, when 4 pose of such six-dimension force sensor changes, due to
, by gravity, can be had an impact to six-dimension force sensor 4 leads to collection result there are errors for it；Therefore it needs first in idle condition
Under, make moving platform 3 along required track C3Movement, and it is loaded to record 4 institute of six-dimension force sensor in the process；This load is
The error that six-dimension force sensor 4 is generated due to gravity influence during the motion；
Step 3：Model training
After test specimen 6 to be loaded is connect with the moving platform 3 of Stewart platforms, by movement locus C3It is flat to input Stewart
Platform, and it is loaded by 6 institute of test specimen in acquisition 3 motion process of moving platform of six-dimension force sensor 4；Gained load is cut into step 2
The error amount of middle gained, as real load suffered by test specimen 6；With six component (F of real loadxFyFzMx My Mz) it is row
Number forms the six-dimensional force matrix of a 6*n using the load data number n acquired as columns；Call the nerve net in matlab
Six-dimensional force matrix is set as input quantity by network tool box, and the movement locus of moving platform 3 is set as output quantity, transformational relation between the two
For hidden layer, the output loads for establishing Stewart platforms move relationship between track, form a 6 inputs-hidden layer -6 and export
Model；Wherein, algorithm selects Feed-forward backprop i.e. feedforward BP neural network, algorithm advantage to be have
Realize the function of any complex nonlinear mapping, and generate network there is certain popularization, abstract ability.Training method is selected
TRAINSCG, i.e. conjugate gradient method, the method take less, and convergence is fast, and can be automatically stopped when not restraining.In the nerve of hidden layer
In terms of first number, by empirical equation
K=(I+O)/2
Wherein K is neuron number, and I is input layer variable number, and O is output layer variable number；The neuron of hidden layer can be obtained
Number is 6；Remaining parameter selection is given tacit consent to, and finally establishes neural network model；
Step 4：Model prediction
The quasi- load applied to test specimen is inputted into trained model, you can obtain corresponding moving platform under the loading environment
3 movement locus；When being moved due to Stewart platforms there are between 2 each driving cylinder of driving unit coupling and cooperative motion, move
The path that platform 3 reaches a certain particular pose is uncertain.Stewart platforms output to be made specifies load and avoids movement rail
The uncertainty of mark damages material for test performance, need to carry out differential to the movement locus of moving platform 3, it is ensured that motion path
Accuracy；Therefore target output loads is are subdivided into m parts, six components of the formation one with load by the method that uses
(FxFyFzMxMyMz) it is line number, using m as the loading matrix of the 6*m of columns；By the 6*m Input matrixes of gained trained model
In, you can obtain the movement locus of moving platform 3 under the specified load.
By calculating the movement locus of test specimen to be loaded in loading procedure, is converted according to coordinate, test specimen to be made can be obtained
Obtain the movement locus of Stewart platforms when expected load；It enables moving platform 3 be moved in the unloaded state along required track, leads to
Six-dimension force sensor 4 is crossed, force acting on transducer state in motion process is acquired.This is influence of the gravity to six-dimension force sensor 4；It enables
Moving platform 3 moves in the state of clamping test specimen 6 along required track, by six-dimension force sensor 4, acquires test specimen in motion process
Stress；It is real load suffered by test specimen 6 that gained load, which is cut influence of the gravity to six-dimension force sensor 4,；By test specimen 6
Suffered real load is trained with corresponding platform track by neural network, and the output loads of Stewart platforms are established
Relationship between track is moved, an output model of 6 inputs-hidden layer -6 is obtained,；Desired load is inputted into trained mould
Type obtains the platform movement locus under the loading environment, and then can realize the Active Compliance Control of Stewart platforms.
Stewart in test specimen loading procedure can get according to coordinate transform by the track in calculation testing piece loading procedure
The movement locus of platform；In the unloaded state, so that platform is moved along track, obtain the error that sensor generates under the influence of gravity
Value；Under test specimen aid state, platform is made to be moved along track, it is loaded to obtain test specimen institute；Test specimen institute loaded is subtracted into zero load
The sensor error value of gained, as test specimen suffered real load in loading procedure under state；By real load suffered by test specimen
It is trained by neural network with corresponding platform track, obtains the model of a 6 inputs-hidden layer -6 output, as
Platform movement locus and test specimen it is loaded between relational model；By the input of desired load, trained model, acquisition should add
Platform track under the conditions of load.This movement locus is inputted into Stewart platforms, you can so that platform is applied test specimen and specify load.
This method provides one kind using load as parameter to the Stewart platforms using six-dimension force sensor for power acquisition elements
Control strategy.
Embodiment：
Shown in Fig. 1 to Fig. 5, test specimen is applied and specifies load load.Here by taking pure bending loads as an example.
Step 1：Trajectory planning.Since using the moving platform of Stewart platforms as loading unit, demand goes out test specimen curved
The movement locus of moving platform during song.First with fixing end, that is, it is initial coordinate system R, calculation testing piece pure bending to descend at fixture
Tache motorice in the process, i.e. movement locus at upper fixture.By pure bending deflection equationA points can be obtained in coordinate system R
Track C1For
Y=0 (2)
U=0 (4)
V=α (5)
Z=0 (6)
In formula, fBFor sag, α is bending corner, MeFor bending moment；E and I be respectively test specimen elasticity modulus and
Cross sectional moment of inertia；X, Y, Z are respectively in a coordinate system along the displacement in x, y, z direction；U, V, W be respectively in a coordinate system around x, y,
The corner of z-axis.
On this basis, since AB sections of connector is rigid section, the B points in coordinate system R can be sought, i.e. Stewart platforms
The track of moving platform center, and the locus of points is set as C2, it is
Y=0 (8)
U=0 (10)
V=α (11)
W=0 (12)
And in the practical control of Stewart platforms, it is to be to moving platform center for reference with platform local Coordinate System R '
B points carry out TRAJECTORY CONTROL.Therefore it need to pass through coordinate system transformation further according to the spatial relationship of coordinate system R and R '
By C2The middle moving platform centers coordinate system R ', the i.e. movement locus of B points are converted to, and sets the locus of points as C3.It should
Track is the actual motion track of platform input when test specimen being made to carry out Combined Loading.Wherein θ is R and R ' angle of coordinate system.
[X′Y′Z′]TRespectively along the displacement of x ', y ', the directions z ' in coordinate system R '.
Step 2：Error analysis.Since power acquisition elements are using six-dimension force sensor 4, such six-dimension force sensor 4
When pose changes, since it is by gravity, can be had an impact to six-dimension force sensor 4 leads to collection result there are errors；Cause
This need to make moving platform 3 along required track C first under idle condition3Movement, and record 4 institute of six-dimension force sensor in the process
It is loaded；This load is the error that six-dimension force sensor 4 is generated due to gravity influence during the motion.
Step 3：Model training.After test specimen 6 to be loaded is connect with the moving platform 3 of Stewart platforms, by movement locus
C3Stewart platforms are inputted, and loaded by 6 institute of test specimen in acquisition 3 motion process of moving platform of six-dimension force sensor 4；By institute
Obtain the error amount that load cuts gained in step 2, as real load suffered by test specimen 6；With six components of real load
(FxFyFzMxMyMz) it is line number, using the load data number n acquired as columns, the six-dimensional force matrix of a 6*n is formed, is passed through
Matlab establishes the relationship between platform movement locus and output loads.
It is involved Matlab operating processes below：
(a) nntool orders are inputted in command window, into neural network module.Loading matrix is inputted into [Inputs] window
Mouthful, by movement locus Input matrix [Targets] window.It clicks [New Network], establishes three layers of nerve of one hidden layer of band
Network.Wherein, algorithm selects Feed-forward backprop i.e. feedforward BP neural network, the algorithm advantage to be to have real
The function of what incumbent complex nonlinear mapping, and network is generated with certain popularization, abstract ability.Training method is selected
TRAINSCG, i.e. conjugate gradient method, the method take less, and convergence is fast, and can be automatically stopped when not restraining.In the nerve of hidden layer
In terms of first number, by empirical equation
K=(I+O)/2
Wherein K is neuron number, and I is input layer variable number, and O is output layer variable number.It can thus be concluded that the nerve of hidden layer
First number is 6.In terms of other parameters, Layer 1Transfer function select TANSIG, 2 Transfer of Layer
Function selects LOGSIG.After all parameter inputs, point Create buttons are created.
(b) Train buttons are clicked and reaches Training info tabss.It sets loading matrix to input quantity, moves square
Battle array is set as output quantity.Parameter is carried out in Train Parameters tabss to fill in.Wherein, iterations [epochs] select
1000 are selected, or is voluntarily adjusted according to convergence result.Target error [goal] can be according to goal* sample sizes<0.5 voluntarily adjusts.
When result restrains, i.e., training error is not more than standard error, train successfully.
Step 4：The quasi- load applied to test specimen is inputted into trained model, you can obtain corresponding under the loading environment
3 movement locus of moving platform；When being moved due to Stewart platforms there are between 2 each driving cylinder of driving unit coupling and cooperative motion,
Therefore it is uncertain that moving platform 3, which reaches the path of a certain particular pose,.Stewart platforms output to be made is specified load and is avoided
The uncertainty of movement locus damages material for test performance, need to carry out differential to the movement locus of moving platform 3, it is ensured that fortune
The accuracy in dynamic path；Therefore target output loads is are subdivided into m parts, six with load of formation one by the method that uses
Component (FxFyFzMxMyMz) it is line number, using m as the loading matrix of the 6*m of columns；The 6*m Input matrixes of gained is trained
In model, you can obtain the movement locus of moving platform 3 under the specified load.
Model prediction.The quasi- load applied to test specimen is inputted into trained model, you can it is right under the loading environment to obtain
3 movement locus of moving platform answered；When being moved due to Stewart platforms there are between 2 each driving cylinder of driving unit coupling with cooperate with
Movement, therefore it is uncertain that moving platform 3, which reaches the path of a certain particular pose,.Load is specified in Stewart platforms output to be made
And the uncertainty of movement locus is avoided to damage material for test performance, differential need to be carried out to the movement locus of moving platform 3,
Ensure the accuracy of motion path；When applying pure bending load, F is enabledx、Fy、Fz、My、MzIt is always zero, by MxM parts are divided into,
Form the loading matrix of a 6*m.By loading matrix input, trained model is predicted, can obtain one group of moving platform movement
Track.This movement locus is inputted into Stewart platforms, you can so that Stewart platforms is applied test specimen and specify load.
The foregoing is merely the preferred embodiments of the present invention, are not intended to restrict the invention, for the technology of this field
For personnel, the invention may be variously modified and varied.All any modification, equivalent substitution, improvement and etc. made for the present invention,
It should all be included in the protection scope of the present invention.
Claims (2)
1. a kind of Active Compliance Control strategy of Stewart platforms, it is characterised in that：It is parameter to Stewart platforms using load
It is controlled, application is the Stewart platforms using six-dimension force sensor as power acquisition module, is included the following steps：
Step 1：Trajectory planning
Stewart platforms are by fixed platform（1）, driving unit（2）, moving platform（3）It constitutes, due to the dynamic flat of Stewart platforms
Platform（3）For loading unit, moving platform（3）It is controlled by displacement, therefore demand goes out the test specimen to load（6）Apply special
Determine moving platform corresponding in loading（3）Movement locus；First with fixing end, that is, descend fixture（7）Place is initial coordinate
It is R, tache motorice during calculation testing piece pure bending, i.e. upper fixture（5）The movement locus at place；A point rails can be obtained by deflection curve equation
Mark, and the A locus of points are set as C1；On this basis, since AB sections of connector is rigid section, the B points in coordinate system R can be sought, i.e.,
The moving platform of Stewart platforms（1）The track of center, and set the moving platform（3）Central point track be C2；And
Be with platform local Coordinate System R' it is reference in the practical control of Stewart platforms, track is carried out to moving platform center, that is, B points
Control；It therefore need to be further according to the spatial relationship between coordinate system R and R', by coordinate system transformation, by C2It is converted in coordinate system R'
The movement locus of moving platform center, i.e. B point, and the B locus of points are set as C3；Track C3As make test specimen（6）It carries out compound
The actual motion track of Stewart platforms is inputted when load；
Step 2：Error analysis
Since power acquisition elements are using six-dimension force sensor（4）, six-dimension force sensor（4）Pose change when, due to its by
Gravity, can be to six-dimension force sensor（4）Having an impact leads to collection result there are errors；Therefore it needs first in idle condition
Under, make moving platform（3）Along required track C3Movement, and record six-dimension force sensor in the process（4）Institute is loaded；This load
As six-dimension force sensor（4）The error generated during the motion due to gravity influence；
Step 3：Model training
By test specimen to be loaded（6）With the moving platform of Stewart platforms（3）After connection, by movement locus C3It is flat to input Stewart
Platform, and pass through six-dimension force sensor（4）Acquire moving platform（3）Test specimen in motion process（6）Institute is loaded；Gained load is cut
The error amount of gained, as real load suffered by test specimen in step 2；With six component (F of real loadxFyFzMx My Mz) be
Line number forms the six-dimensional force matrix of a 6*n using the load data number n acquired as columns；Call the nerve in matlab
Six-dimensional force matrix is set as input quantity by network tool case, and 3 movement locus of moving platform is set as output quantity, transformational relation between the two
For hidden layer, the output loads for establishing Stewart platforms move relationship between track, form a 6 inputs-hidden layer -6 and export
Model；Wherein, algorithm selects Feed-forward backprop i.e. feedforward BP neural network, training method to select
TRAINSCG, i.e. conjugate gradient method；In terms of the neuron number of hidden layer, by formula
K=(I+O)/2
Wherein K is neuron number, and I is input layer variable number, and O is output layer variable number；The neuron number that hidden layer can be obtained is
6；Remaining parameter selection is given tacit consent to, and finally establishes neural network model；
Step 4：Model prediction
The quasi- load applied to test specimen is inputted into trained model, you can obtain corresponding moving platform under the loading environment（3）
Movement locus；Since there are driving units when Stewart platforms move（2）Coupling between each driving cylinder and cooperative motion, therefore
Moving platform（3）The path for reaching a certain particular pose is uncertain；Stewart platforms output to be made specifies load and avoids transporting
The uncertainty of dynamic rail mark damages material for test performance, need to be to moving platform（3）Movement locus carry out differential, it is ensured that fortune
The accuracy in dynamic path；Therefore target output loads is are subdivided into m parts, six with load of formation one by the method that uses
Component (FxFyFzMx My Mz) it is line number, using m as the loading matrix of the 6*m of columns；The 6*m Input matrixes of gained have been trained
Model in, you can moving platform under the specified load（3）Movement locus.
2. the Active Compliance Control strategy of Stewart platforms according to claim 1, it is characterised in that：It is waited for by calculating
The test specimen of load（6）Movement locus in loading procedure, is converted according to coordinate, can get test specimen（6）It is moved in loading procedure flat
Platform（3）Movement locus；Enable moving platform（3）It is moved in the unloaded state along required track, passes through six-dimension force sensor（4）Acquisition
Test specimen in motion process（6）Stress；This is gravity to six-dimension force sensor（4）Influence；Enable moving platform（3）It is being clamped
It is moved along required track in the state of test specimen, passes through six-dimension force sensor（4）, acquire test specimen stress in motion process；It will
Gained load cuts gravity to six-dimension force sensor（4）Influence be real load suffered by test specimen；It will really be carried suffered by test specimen
Lotus and corresponding moving platform（3）Movement locus be trained by neural network, obtain 6 inputs-hidden layer -6 it is defeated
Go out model, it is established that movement locus and test specimen relational model loaded；Desired load is inputted into trained model, is obtained
The platform movement locus under the loading environment is obtained, and then can realize the Active Compliance Control of Stewart platforms.
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 true CN108445764A (en) | 2018-08-24 |
| CN108445764B 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) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109760063A (en) * | 2019-03-15 | 2019-05-17 | 京东方科技集团股份有限公司 | Control method, device, equipment and the storage medium of parallel robot |
| CN109940605A (en) * | 2019-01-29 | 2019-06-28 | 中国工程物理研究院激光聚变研究中心 | Merge the submissive assembly system and method for 3D vision and contact force analysis |
| CN110030234A (en) * | 2019-04-30 | 2019-07-19 | 北京航空航天大学 | A kind of interaction safety control strategy based on hydraulic submissive 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 |
| CN114831847A (en) * | 2022-03-29 | 2022-08-02 | 中国农业大学 | Neck rehabilitation training robot with four branched chain parallel mechanisms and force control method thereof |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060200253A1 (en) * | 1999-02-01 | 2006-09-07 | Hoffberg Steven M | Internet appliance system and method |
| 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 |
-
2018
- 2018-03-23 CN CN201810244210.3A patent/CN108445764B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060200253A1 (en) * | 1999-02-01 | 2006-09-07 | Hoffberg Steven M | Internet appliance system and method |
| 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 |
|---|
| CHUL-GOO KANG: "Closed-form force sensing of a 6-axis force transducer based on the Stewart platform", 《SENSORS AND ACTUATORS》 * |
| SONGLIN ZHOU: "Method of designing a six-axis force sensor for stiffness decoupling based on Stewart platform", 《MEASUREMENT》 * |
| 王志军 等: "机器人运动中六维力传感器的重力补偿研究", 《机械设计与制造》 * |
| 许金凯 等: "驱动冗余重型并联机构的动力学性能", 《吉林大学学报》 * |
| 钱承 等: "Stewart六自由度并联平台动力学模型振动分析", 《湖南大学学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109940605A (en) * | 2019-01-29 | 2019-06-28 | 中国工程物理研究院激光聚变研究中心 | Merge the submissive assembly system and method for 3D vision and contact force analysis |
| CN109760063A (en) * | 2019-03-15 | 2019-05-17 | 京东方科技集团股份有限公司 | Control method, device, equipment and the storage medium of parallel robot |
| 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 |
| CN110030234A (en) * | 2019-04-30 | 2019-07-19 | 北京航空航天大学 | A kind of interaction safety control strategy based on hydraulic submissive driver |
| 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 |
| CN114831847A (en) * | 2022-03-29 | 2022-08-02 | 中国农业大学 | Neck rehabilitation training robot with four branched chain parallel mechanisms and force control method thereof |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108445764B (en) | 2021-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108445764A (en) | The Active Compliance Control strategy of Stewart platforms | |
| Belchior et al. | Off-line compensation of the tool path deviations on robotic machining: Application to incremental sheet forming | |
| Khosla et al. | Parameter identification of robot dynamics | |
| CN105690388B (en) | A kind of tendon driving manipulator tendon tension restriction impedance adjustment and device | |
| CN110000787A (en) | A kind of control method of super redundant mechanical arm | |
| Zhang et al. | Kinematic calibration of cable-driven parallel robots considering the pulley kinematics | |
| Wang et al. | Dynamic performance analysis of parallel manipulators based on two-inertia-system | |
| Korayem et al. | Development of ICASBOT: a cable-suspended robot’s with Six DOF | |
| Fung et al. | Dynamic analysis and system identification of an LCD glass-handling robot driven by a PMSM | |
| CN108638058A (en) | A kind of posture decision dynamic programming method | |
| Khalilpour et al. | Tip-trajectory tracking control of a deployable cable-driven robot via output redefinition | |
| Alamdar et al. | Investigation of a hybrid kinematic calibration method for the “sina” surgical robot | |
| Chong et al. | Evaluation of dynamic isotropy and coupling acceleration capacity for a parallel manipulator with mixed DoFs | |
| Najafi et al. | Hybrid simulation with multiple actuators: A state-of-the-art review | |
| Yu et al. | Microinstrument contact force sensing based on cable tension using BLSTM–MLP network | |
| Yu | A new pose accuracy compensation method for parallel manipulators based on hybrid artificial neural network | |
| Zhang et al. | A newton-raphson and bp neural network hybrid algorithm for forward kinematics of parallel manipulator | |
| Sun et al. | Force/motion transmissibility analysis and parameters optimization of hybrid mechanisms with prescribed workspace | |
| Rahmani et al. | Application of a novel elimination algorithm with developed continuation method for nonlinear forward kinematics solution of modular hybrid manipulators | |
| Irani et al. | Tip tracking control of a rigid-flexible manipulator based on deflection estimation using neural networks: Application to needle insertion | |
| Ding et al. | Quantitative analysis of decoupling and spatial isotropy of a generalised rotation-symmetric 6-DOF Stewart platform | |
| Zaccaria et al. | Design and experimental equilibrium stability assessment of a RFRFR continuum parallel robot | |
| Zhang et al. | Lightweight long-reach 5-dof robot arm for farm application | |
| Yu et al. | Research on stiffness control of a redundant cable-driven parallel mechanism | |
| Xu et al. | Force analysis of the redundantly actuated parallel mechanism 2RPR+ P considering different control methodologies |
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 |