Invention content
Goal of the invention:
The purpose of the present invention is by quadruple accuracy compensation, increasing substantially the positioning accuracy of large-scale heavy duty mechanical arm, gram
Take the contradiction between large-scale heavy duty mechanical arm and high accuracy positioning.Quadruple precision compensation method through the invention can make big
The positioning accuracy of type heavy-duty machinery arm improves two orders of magnitude or more, reaches within 0.5mm, while payload can be overcome inclined
Terminal angle error caused by setting.
Technical solution:
A kind of quadruple precision compensation method for large-scale heavy duty mechanical arm, the quadruple precision compensation method specific steps are such as
Under:
Step 1:Realize return difference compensating control method;
Step 2:On the basis of return difference is compensated and controlled, model compensation calibration is carried out, determines new kinematics model;
Step 3:Model compensation is controlled and is added in Mechanical arm control method, in model compensation control → return difference compensation control
On the basis of system, gravity compensation calibration is carried out;
Step 4:Gravity compensation is controlled and is added in Mechanical arm control method, in gravity compensation control → model compensation control
On the basis of system → return difference compensation control, inclined square compensation calibration is carried out;
Step 5:Inclined square is compensated control to be added in Mechanical arm control method, forms inclined square compensation control → gravity compensation
The quadruple accuracy compensation control method of control → model compensation control → return difference compensation control, makes manipulator motion precision substantially carry
It is high.
Return difference compensation control comprises the concrete steps that:
Reading position sensor reading, judge the error of joint output position and target location whether reach allowable range it
It is interior;If not reaching allowable range, to target location direction with very slow speed kinematic error × proportionality coefficient away from
From;Then position sensor reading is read again, judges whether the error of joint output position and target location reaches permission model
Within enclosing;As do not reached allowable range, it is necessary to repeat above-mentioned compensation campaign process, until the mistake of joint output position
Until difference reaches allowable range;If reaching allowable range, indicate that return difference compensation controls successfully, exits the compensation control of this return difference
Cyclic process.
Step 2 comprises the concrete steps that:
Processing, installation error indicated with error matrix, processing, installation error place may to three translation, three
Error is generated on the total six direction of rotation, uses dx respectivelyi、dyi、dzi、αi、βi、γiIt indicates, wherein i indicates the sequence of error matrix
Number;The form that error matrix is expressed as:
Ideally, each error component is zero, then it is quadravalence unit matrix, i.e. this error matrix that error matrix, which is degenerated,
It disappears in the kinematics model of mechanical arm;Conversely, mechanical arm ideal movements model any two transformation matrix it
Between be inserted into an error matrix, indicate the two movement transmission links between there may be processing, installation error;
Gravitational settling indicates in the kinematics model of mechanical arm, and the translation transformation matrix of characterization connecting rod will be made to become as follows
Form:
Wherein, i indicates that the serial number of translation transformation matrix, I indicate three rank unit matrixs;PiFor three dimensional vectors, the company of expression
The translational movement of bar in the ideal situation;LiIt is a column vector, expression and the relevant length of connecting rod of gravitational settling herein, i-th
Gravitational settling at connecting rod is to this connecting rod and this connecting rod before until all connecting rods of mechanical arm pedestal are all related, so LiIt is
I dimensions;
In containing balanced new kinematics model, by the translation transformation matrix of all connecting rods with vertical settlement effects
All replace with TRANgiForm;It is as follows to obtain form that is more accurate and meeting actual new kinematics model:
T=...i-1Ti·TRANgi·Tei·iTi+1·TRANgi+1·… (3)
Wherein,i-1TiIndicate that the transformation matrix of i-th of joint, T indicate that mechanical arm pedestal becomes to the totality of end flange
Change matrix;
Model compensation calibration is carried out, after establishing the form of precise motion model, mechanical arm tail end flange center is obtained and exists
The expression-form of position coordinates P under base coordinate system, is denoted as:
P=F (α, β, γ, dx, dy, dz, p);Wherein, α, β, γ, dx, dy, dz, p indicate three kinds of rotation errors, three respectively
The column vector that kind translation error, gravitational settling coefficient are constituted;If above-mentioned error is all sufficiently small, following total differential form is just obtained
Error difference equation:
Wherein, Δ P is the difference of absolute fix and theoretical position, and δ is the column vector that all margins of error of mechanical arm are constituted,
L δ are the matrix representation forms of Δ P;In above formula, δ is that model compensation calibration needs obtained result;L is each freedom of mechanical arm
The function of position is spent, each margin of error being directed to when calculating L is all denoted as 0;The theoretical position of mechanical arm is each by mechanical arm
Degree of freedom position is calculated by ideal movements model;So if manipulator motion is measured end to a certain position
The absolute fix at end flanges center, re-records down the position of each degree of freedom of mechanical arm at this time, and it is mechanical arm just to obtain Δ P=L δ
The equation of the margin of error;Measure enough position P1,…,Pm, just obtain the big matrix equation about δ:
Δ Q=LL δ (5)
Wherein,
The value of m should at least make the line number of matrix L L be more than columns;
Under normal conditions, LL is not sequency spectrum;It needs to find out linear relevant Column vector groups in LL at this time, by it
In a row remove, and remove error component corresponding with this row in δ simultaneously;This operation is constantly repeated, until LL becomes row completely
Until order matrix, this sequency spectrum matrix is denoted as LL', error column vector corresponding with LL' is denoted as δ ';Finally obtained square
Battle array equation be:
Δ Q=LL' δ ' (6)
The error component removed is mathematically relevant with remaining error component, that is, the error component removed does not have
Independent error effects, does not demarcate it;The above-mentioned matrix equation about the margin of error solves following least squqre approximation
Solution:
δ '=(LL'TLL')-1LL'TΔQ (7)
In the case of mechanical arm not loading, the target of laser tracker is placed in mechanical arm tail end flange center position
Ball;It allows manipulator motion to be combined to a series of joint position, the position coordinate value x, y, z of target ball is recorded with laser tracker,
And corresponding joint position combination is recorded, need the practical joint position for recording joint output position sensor feedback;Profit
The concrete numerical value of the various margins of error in compound movement model is fitted with the method described above;Selected joint position group
The working space of mechanical arm will fully be covered by closing, while in order to reduce workload, not choose the position outside working space;Because
Model compensation calibration is not the final demarcating steps of quadruple accuracy compensation method, so model compensation allows have remainder error;In order to
Further decreasing workload should be regarded as managing to acceptable range for the unconspicuous error link of some error effects
Think, the output position in certain joints is made to become unrelated with the calibration of model error amount in this way, dramatically reduces model and mend
Repay the workload of calibration.
Model compensation control comprises the concrete steps that:
After model compensation calibration, obtains the more accurate of mechanical arm and meet actual compound movement model, be denoted as:
T=Kinaccu(q) (8)
Wherein q indicates the vector of each joint position composition of mechanical arm;It is applied in model compensation control, needs to solve
Go out the Inverse Kinematics Solution of the model:
In general, after model error factor is added in Mechanical transmission test, kinematics model can become very complicated and be difficult to obtain
The Inverse Kinematics Solution of parsing needs the numerical solution for using Inverse Kinematics Solution at this time;
There is the inverse solution of Mechanical transmission test, with regard to carrying out model compensation control;When an end target for specifying mechanical arm
When pose, end object pose is converted to mechanical arm pedestal to the transformation matrix T of end flange, then pass through Inverse Kinematics Solution
Solve joint output position q;Q is the joint of mechanical arm target output position with model compensation, while being also return difference compensation
It is input in return difference compensating control method by the input of control, just continues return difference compensation after model compensation.
Gravity compensation calibration comprises the concrete steps that:
X, y, z and payload w four dimensions to the effective working space of mechanical arm carry out mesh generation;It is four-dimensional
Each node (x, y, z, w) of grid indicates that mechanical arm moves to (x, y, z) in working space when payload is w
Point;Each node of four-dimensional grid is demarcated, is measured between mechanical arm tail end actual measurement pose and theory target pose
Difference establishes gravity compensation database, completes gravity compensation calibration process;Wherein, for payload w for mechanical arm tail end
Flange generates great torque, and in gravity compensation calibration, it is 0 to make the inclined square that payload w is generated, and is not 0 by inclined square
Situation, which is separated, gives inclined square compensation solution;
It is which type of terminal angle (x, y, z) point is moved to for mechanical arm, in gravity compensation calibration, mechanical arm
Calibration point is moved to some common special terminal angle, obtains the nominal data under this special posture;Gravity compensation number
According to the nominal data contained only in library under special posture, because mesh generation can not also be carried out to terminal angle, otherwise database
In data volume will be increased to the degree that cannot achieve;The case where task posture of mechanical arm is other postures utilizes an algorithm to solve
Certainly, the error of mechanical arm tail end tool and involved in this algorithm, needs first to demarcate end-of-arm tooling.
End-of-arm tooling calibration comprises the concrete steps that:
End-of-arm tooling all be can be regarded as in calibration for end-of-arm tooling, the part after mechanical arm tail end flange, including for reality
The work that existing quadruple accuracy compensation method and the six-dimension force sensor and mechanical arm that install additional must install additional to realize job task
Tool;The model of tool indicates that is, the displacement in tri- directions X, Y, Z is plus the corner around Y, around X with X, Y, Z, B, C;B, the name of C
Adopted value is 0, and X, Y, Z also have respective nominal value;
First, allow manipulator motion to a specific position, the command value of object attitude angle is 0;Measure mechanical arm end
The position of ending tool and posture, obtain at this time mechanical arm base coordinate system to the homogeneous transformation T of tool coordinates systemXYZABC;Then will
The end flange rotary joint of mechanical arm rotates specific angle z, measures position and the posture of mechanical arm end-of-arm tooling again, then
It is secondary to obtain a TXYZABC;Enough times are measured, a series of z-T are obtainedXYZABC, there is following relational expression:
TXYZABC(z)=TBase→FlangeTRot(z)Ttool,BCTtool,XYZ (10)
Wherein, TBase→FlangeIndicate mechanical arm pedestal to the transformation matrix of end flange, TRot(z) rotational angle z is indicated
The rotational transformation matrix of generation, Ttool,BCAnd Ttool,XYZIndicate true rotation transformation and the translation transformation square of mechanical arm tail end tool
Battle array;
The posture part in relational expression (10) is only investigated, following relationship is obtained:
RABC(z)=RBase→FlangeRot(z)Rtool,BC (11)
RABC(z)、RBase→Flange、Rot(z)、Rtool,BCT is indicated respectivelyXYZABC(z)、TBase→Flange、TRot(z)、Ttool,BC
Spin matrix part;Lateral attitude angle B, C that tool is found out from posture relational expression (11), are asked with the method for intelligent search
Solution;The phase of the absolute value of the difference for the attitude angle that the target function of intelligent search is characterized by above-mentioned posture relationship "=" both sides
It hopes, smaller selected tool tilt attitude angle B, C of explanation of target function value is better;For selected one group of specific tool
Lateral attitude angle B, C, are handled as follows:First R is found out in z=0Base→Flange, then utilize posture relationship when z ≠ 0
Formula and RBase→FlangeFind out target function value;
In TXYZABCRelational expression in, specify following symbol:
Then obtain the relational expression as follows about position:
Further transformation obtains:
According to the different values of the rotational angle z of end flange rotary joint, a series of relationship shaped like (13) is obtained
Formula combines them:
X, Y, Z are acquired using least square method;To obtained whole feature X, Y in mechanical arm tail end tool model,
Z, B, C complete end-of-arm tooling calibration.
For one group of specific value of terminal position theoretical value, by the theoretical value of three attitude angles A, B, C of terminal angle
It is taken as A0, 0,0, wherein A0For the specific angles an A value;Have at this time:
Wherein,For the measured value of mechanical arm tail end pose,For the theory target of mechanical arm tail end pose
Value,For the gravity compensation matrix of mechanical arm;
By gravity compensation matrixPose form, 3 position compensation amounts and 3 attitude angle compensation rates are converted into, this
6 numerical value are exactly the calibration result of this calibration point;All nodes in four-dimensional mesh generation are measured, and calculate the mark of each node
Calmly as a result, ultimately forming gravity compensation database, gravity compensation calibration process is completed.
Gravity compensation control comprises the concrete steps that:
What gravity compensation to be done is exactly the database established when being demarcated by inquiring, and finds corresponding Tcomp, pass through relationship
Formula:
TcompTreal=Tthe (16)
Calculate Tthe, by TtheIt is transmitted to mechanical arm as the pose after gravity compensation, after so that mechanical arm is moved by this pose
Just reach Treal;
It is A for there was only attitude angle in database0, 0,0 when compensation rate, and face is posture when gravity compensation controls
The problem of the case where angle is arbitrary value A, B, C, needs to use the end-of-arm tooling carried out before and demarcates to solve;If it is known that
Tthe,ABC, pass through following formula:
It is calculatedWherein, Ttool,XYZ,theIndicate ideal end-of-arm tooling transformation, TZtIndicate mechanical arm wrist
Center is to the transformation at end flange center, TRot(A-A0, B, C) and it indicates from attitude angle A0, 0,0 to attitude angle A, B, C rotation become
Change matrix;Then by following formula, byCalculate to obtain Tcomp,ABC:
Wherein, Ttool,XYZAnd Ttool,BCRespectively indicate end-of-arm tooling actual end-of-arm tooling translation transformation obtained by calibrating and
End-of-arm tooling rotation transformation;
All it is and T for the compensation rate in gravity compensation databasetheIt is corresponding, and currently known is Treal, Wo Menwu
Method direct basis TrealUsing the problem of obtaining required compensation rate with Multi-dimension interpolation method of tabling look-up in gravity compensation database, have
Gravity compensation relational expression:
Tcomp,ABCTreal,ABC=Tthe,ABC (19)
I.e. as known Treal,ABCWhen, as long as providing Tcomp,ABC, just obtain Tthe,ABC;
It then passes through cycle and tables look-up approximatioss solve can not direct basis TrealIt is obtained with Multi-dimension interpolation method using tabling look-up
Obtain required compensation rate:First assume Tthe,ABCEqual to Treal,ABC, then T is obtained in gravity compensation database lookupcomp,ABC;According to
Formula (19) updates Tthe,ABC, then by new Tthe,ABCTable look-up update Tcomp,ABC;T is updated further according to formula (19)the,ABC, constantly repeat
This process, until reaching required precision.
Inclined square compensation calibration comprises the concrete steps that:
First, the inclined square range that the payload of mechanical arm may generate in two directions is determined;Within the scope of inclined square
Mesh generation is carried out to the inclined square in both direction, generates a two-dimensional calibration grid;In the grid must include point (0,
0) benchmark that, (0,0) is compensated as inclined square, i.e., inclined square compensation rate is all 0 when not loading inclined square to mechanical arm;To two-dimensional grid
In each node, it is real-time using six-dimension force sensor according to inclined square numerical value to the inclined square of mechanical arm tail end tool loads two dimension
Inclined square value is measured, ensures that load is accurate;
The influence for excluding load gravity and mechanical arm dead weight is deformed is had to when square compensation is demarcated partially, so, first
Pose on the basis of the most frequently used pose of one mechanical arm of definition, all inclined squares compensation calibration are carried out in this pose;Then right
Control machinery arm is needed once to be moved using benchmark pose as target, at this after load is accurate in each two dimension inclined square
Application in order is needed to demarcate all accuracy compensation control methods of completion in movement, to eliminate its other than inclined square
The influence of his error source, i.e.,:
Gravity compensation control → model compensation control → return difference compensation control;
After the completion of movement, the pose of mechanical arm tail end tool at this time is measured, thus calculates end-of-arm tooling around both direction
Deflection angleThe square combination partially of all two dimensions is measured, corresponding deflection angle is calculated, forms inclined square compensation database;By
In the benchmark for compensating (0,0) as inclined square, so for the square combination partially of all two dimensions, it is real caused by inclined square
Border deflection angle is:
Inclined square compensation control comprises the concrete steps that:
First, the numerical value of inclined square is read using six-dimension force sensor, is then tabled look-up and is utilized in inclined square compensation database
Two-dimensional interpolation obtains deflection angle β, γ, carries out inclined square compensation control using following formula later:
Tout=TinTtool,XYZ,the -1TRot(-β,-γ)Ttool,XYZ,the (20)
Wherein, TinIndicate the object pose before inclined square compensation, TRotThe rotation transformation square that (- β ,-γ) expression-β ,-γ are generated
Battle array, ToutIndicate the pose of inclined square compensation output.
Advantage and effect:
Quadruple accuracy compensation method is started with from four kinds of different main error sources of large-scale heavy duty mechanical arm, is moved to it continuously
Four serial accuracy compensations are carried out, kinematic error can be made to decline 2 orders of magnitude, absolute precision is made to reach and repeatable accuracy phase
Same magnitude, reaches the absolute fix precision of 0.5mm.Using quadruple accuracy compensation method, need end flange in mechanical arm with
It installs six-dimension force sensor between tool additional, and the output end (end of joint transmission chain) in each joint of mechanical arm is needed to add
Holding position sensor.
Quadruple accuracy compensation method makes the exhausted of the mechanical arm when being applied to large-scale heavy duty mechanical arm-proton therapeutic bed
40mm is more than in the case of only carrying out zero position calibration to positioning accuracy, is improved to using small after quadruple accuracy compensation method
In 0.2mm, more than 2 orders of magnitude of absolute fix precision improvement.
Specific implementation mode
Quadruple accuracy compensation method is started with from four kinds of different main error sources of large-scale heavy duty mechanical arm, is moved to it continuously
Four serial accuracy compensations are carried out, kinematic error can be made to decline 2 orders of magnitude, absolute precision is made to reach and repeatable accuracy phase
Same magnitude, reaches the absolute fix precision of 0.5mm.Using quadruple accuracy compensation method, need end flange in mechanical arm with
It installs six-dimension force sensor between tool additional, and the output end (end of joint transmission chain) in each joint of mechanical arm is needed to add
Holding position sensor.
A kind of quadruple precision compensation method for large-scale heavy duty mechanical arm, the quadruple precision compensation method specific steps are such as
Under:
Step 1:Realize return difference compensating control method;
Step 2:On the basis of return difference is compensated and controlled, model compensation calibration is carried out, determines new kinematics model;
Step 3:Model compensation is controlled and is added in Mechanical arm control method, in model compensation control → return difference compensation control
On the basis of system, gravity compensation calibration is carried out;
Step 4:Gravity compensation is controlled and is added in Mechanical arm control method, in gravity compensation control → model compensation control
On the basis of system → return difference compensation control, inclined square compensation calibration is carried out;
Step 5:Inclined square is compensated control to be added in Mechanical arm control method, forms inclined square compensation control → gravity compensation
The quadruple accuracy compensation control method of control → model compensation control → return difference compensation control, makes manipulator motion precision substantially carry
It is high.
Wherein, Mechanical arm control method is the prior art.
Quadruple accuracy compensation is respectively return difference compensation, model compensation, gravity compensation, the compensation of inclined square.Quadruple accuracy compensation method
Essence be that one virtual target pose is gone out by compensation calculation by the object pose of mechanical arm, mechanical arm presses uncompensated mode
(except return difference compensation) moves to virtual target pose, and virtual condition just has arrived at object pose.Quadruple accuracy compensation method
Mechanical arm control in compensation flow be:
Inclined square compensation → gravity compensation → model compensation → return difference compensation
Quadruple accuracy compensation method before application, needs to demarcate each compensation process, and calibration is needed according to following
Sequence carries out:
Return difference compensates calibration → model compensation calibration → gravity compensation calibration → square compensation calibration partially
The calibration in a stage is often carried out, it is necessary to which the backoff algorithm in corresponding stage is applied to the control of mechanical arm
In, this is to continue with the premise for carrying out follow-up phase calibration.
Return difference compensates:
The target compensation of return difference compensation is that the backhaul of joint of mechanical arm transmission link is poor.Its compensation effect is to improve mechanical arm
Joint position output accuracy makes the error of joint output position and target location within the scope of precision allowable error.Return difference is mended
It is one most special in four-stage to repay, it is actually a kind of control method, and other stages are actually backoff algorithm.
Return difference compensation need not demarcate, but directly apply its control method in mechanical arm control system.
Return difference compensating control method:
After each joint output end loading position sensor of mechanical arm, the joint control mode of mechanical arm becomes such as Fig. 1
Shown in double circle structure.Due to the presence of joint transmission return difference, after joint of robot motor completes positioning, joint carry-out bit
Set random a small amount of there are still one, this random quantity will generate appreciable error through armed lever amplification in robot end.So
After joint motor completes positioning, need to enter return difference compensation process.
Return difference compensation needs reading position sensor reading first, judges that the error of joint output position and target location is
It is no to reach within allowable range;As do not reached allowable range, it is necessary to (have with very slow speed to the direction of target location
Body speed is measured by testing, and speed is too low to cause compensation efficiency low, and speed is excessively high to will produce inertia mistake at transmission back clearance
Rush and make return difference compensation that can not succeed) distance of kinematic error × proportionality coefficient;Then position sensor reading is read again, is sentenced
Whether the error of disconnected joint output position and target location reaches within allowable range;As do not reached allowable range, it is necessary to
Repeat above-mentioned compensation campaign process, until the error of joint output position reaches allowable range.Return difference compensation control
Method flow is as shown in Figure 2.
Model compensation:
Model compensation is processing, installation error to be considered, caused by gravity on the basis of robot ideal movements model
These factors are added to kinematics model by sedimentation and deformation, are obtained more accurate and are met actual new kinematics model.
Processing, installation error are indicated with error matrix.It may be to three translations, three in the presence of the place of processing, installation error
Error is generated on a total six direction of rotation, is used respectively
dxi、dyi、dzi、αi、βi、γi
It indicates, wherein i indicates the serial number of error matrix.The form that then error matrix is expressed as:
Ideally, each error component is zero, then it is quadravalence unit matrix, i.e. this error matrix that error matrix, which is degenerated,
It disappears in the kinematics model of mechanical arm.Conversely, can also be converted in any two of the ideal movements model of mechanical arm
It is inserted into an error matrix between matrix, indicates that there may be processing, installation errors between the two movement transmission links.
Gravitational settling error is to cause itself link tilt due to the dead weight of large-scale heavy duty mechanical arm, caused by connecting rod
End changes the relative position of connecting rod initiating terminal.Since mechanical arm is complicated multi-body system, gravitational settling is extremely complex,
So being simplified shown as gravitational settling coefficient here.Gravitational settling indicates in the kinematics model of mechanical arm, will make table
The translation transformation matrix of sign connecting rod becomes following form:
Wherein, i indicates that the serial number of translation transformation matrix, I indicate three rank unit matrixs;PiFor three dimensional vectors, the company of expression
The translational movement of bar in the ideal situation;LiIt is a column vector, expression and the relevant length of connecting rod of gravitational settling herein, i-th
Gravitational settling at connecting rod is to this connecting rod and this connecting rod before until all connecting rods of mechanical arm pedestal are all related, so LiIt is
I dimensions;In some cases connecting rod is after the rotation of rotary joint, it can be all applied to by being not the dead weight of this connecting rod
Gravitational settling is generated on connecting rod before, such case can be equivalent to LiIt is multiplied by the dimension of respective link before middle correspondence
The trigonometric function of one joint rotation angles;
piIt is the gravitational settling coefficient matrix of 3 × i, this matrix is a sparse matrix under normal conditions, specific in matrix
Which element is 0, which element is the same gravitational settling coefficient, needs to be built according to the concrete structure and coordinate system of mechanical arm
Cube formula is made a concrete analysis of, and there is no universal methods.
Ideally, gravitational settling coefficient matrix piFor null matrix, TRANgiIt is perfect translational transformation matrix to degenerate.And
In containing balanced new kinematics model, the translation transformation matrix of all connecting rods with vertical settlement effects should all be replaced
For TRANgiForm.
Finally, it is as follows to obtain form that is more accurate and meeting actual new kinematics model:
T=...i-1Ti·TRANgi·Tei·iTi+1·TRANgi+1·… (3)
Wherein,i-1TiIndicate that the transformation matrix of i-th of joint, T indicate that mechanical arm pedestal becomes to the totality of end flange
Change matrix.
Model compensation calibration process:
After the form for establishing precise motion model, mechanical arm tail end flange center can be obtained under base coordinate system
Position coordinates P expression-form, can be denoted as:
P=F (α, β, γ, dx, dy, dz, p);Wherein, α, β, γ, dx, dy, dz, p indicate three kinds of rotation errors, three respectively
The column vector that kind translation error, gravitational settling coefficient are constituted.If above-mentioned error is all sufficiently small, so that it may to obtain following total differential
The error difference equation of form:
Wherein, Δ P is the difference of absolute fix and theoretical position, and δ is the column vector that all margins of error of mechanical arm are constituted,
L δ are the matrix representation forms of Δ P.
In above formula, δ is that model compensation calibration needs obtained result;L is the function of each degree of freedom position of mechanical arm, is being counted
It calculates each margin of error being directed to when L and is all denoted as 0;The theoretical position of mechanical arm can be led to by each degree of freedom position of mechanical arm
Ideal movements model is crossed to be calculated.So if by manipulator motion to a certain position, the reality at end flange center is measured
Location is set, and the position of each degree of freedom of mechanical arm at this time is re-recorded down, so that it may to obtain Δ P=L δ as the side of the mechanical arm margin of error
Journey.Measure enough position P1,…,Pm, so that it may to obtain the big matrix equation about δ:
Δ Q=LL δ (5)
Wherein,
The value of m should at least make the line number of matrix L L be more than columns.
Under normal conditions, LL is not sequency spectrum.It needs to find out linear relevant Column vector groups in LL at this time, by it
In a row remove, and remove error component corresponding with this row in δ simultaneously.This operation is constantly repeated, until LL becomes row completely
Until order matrix, this sequency spectrum matrix is denoted as LL', error column vector corresponding with LL' is denoted as δ '.Finally obtained square
Battle array equation be:
Δ Q=LL' δ ' (6)
The error component removed is mathematically relevant with remaining error component, that is, the error component removed does not have
Independent error effects can not demarcate it.The above-mentioned matrix equation about the margin of error can solve following minimum
Two multiply approximate solution:
δ '=(LL'TLL')-1LL'TΔQ (7)
In the case of mechanical arm not loading, the target of laser tracker is placed in mechanical arm tail end flange center position
Ball.It allows manipulator motion to be combined to a series of joint position, the position coordinate value x, y, z of target ball is recorded with laser tracker,
And record corresponding joint position combination (needing to record the practical joint position of joint output position sensor feedback).This
Sample can fit the concrete numerical value of the various margins of error in compound movement model using the method described above.Selected
Joint position combines the working space that fully cover mechanical arm, while in order to reduce workload, not choose outside working space
Position.Because model compensation calibration is not the final demarcating steps of quadruple accuracy compensation method, model compensation allows to have surplus
Remaining error.The unconspicuous error link of some error effects is answered in order to further decrease workload to acceptable range
It is regarded as preferably, the output position in certain joints being made to become unrelated with the calibration of model error amount in this way, Ke Yiji
Reduce to big degree the workload of model compensation calibration.
Model compensation control method:
After model compensation calibration, the more accurate of mechanical arm can be obtained and meet actual compound movement model,
It is denoted as:
T=Kinaccu(q) (8)
Wherein q indicates the vector of each joint position composition of mechanical arm.It is applied in model compensation control, needs to solve
Go out the Inverse Kinematics Solution of the model:
In general, after model error factor is added in Mechanical transmission test, kinematics model can become very complicated and be difficult to obtain
The Inverse Kinematics Solution of parsing needs the numerical solution for using Inverse Kinematics Solution at this time.
There is the inverse solution of Mechanical transmission test, so that it may to carry out model compensation control.When an end for specifying mechanical arm
When object pose, end object pose can be converted to mechanical arm pedestal to the transformation matrix T of end flange, then pass through fortune
Dynamic learn against solution solves joint output position q.Q is the joint of mechanical arm target output position with model compensation, is also simultaneously
The input of return difference compensation control, it is input in return difference compensating control method, so that it may to continue back after model compensation
Difference compensation.
Gravity compensation:
After model compensation calibration, a part of remainder error is still suffered from.On the other hand, large-scale heavy duty mechanical arm is effective
Load can also introduce very important position error, and there is no compensated the error that payload introduces before this.So connecing
Get off and needs to continue with gravity compensation.X, y, z and payload w of the gravity compensation to the effective working space of mechanical arm
Four dimensions carry out mesh generation.Each node (x, y, z, w) expression mechanical arm of four-dimensional grid is w lucks in payload
Move (x, y, z) point in working space.Each node of four-dimensional grid is demarcated, mechanical arm tail end actual measurement position is measured
Difference between appearance and theory target pose, so that it may to establish gravity compensation database.It is mechanical whenever giving again after establishing database
When the task of one (x, y, z, w) form of arm, so that it may to inquire the data in database, and Multi-dimension interpolation method is combined, to this
Task carries out effective gravity compensation.
Here there are two problems:First, great torque generated to mechanical arm tail end flange without reference to payload w, two
Be without reference to mechanical arm it is which type of terminal angle (x, y, z) point is moved to.For first problem, payload w
Load deflection torque, referred to as inclined square are called to the torque that end flange generates;In gravity compensation calibration, payload w is made to produce
Raw inclined square is 0, and the case where not being 0 by inclined square separates and gives the compensation of inclined square and solve.For Second Problem, in gravity
In compensation calibration, we allow mechanical arm to move to calibration with some common special terminal angle (such as 3 attitude angles are all for 0)
Point can obtain the nominal data under this special posture in this way;The mark under special posture is contained only in gravity compensation database
Fixed number evidence, because we can not also carry out mesh generation to terminal angle, otherwise will be increased to can not for the data volume in database
The degree of realization;The case where task posture of mechanical arm is other postures utilizes an algorithm to solve, and involved in this algorithm
The error of mechanical arm tail end tool, so needing first to demarcate end-of-arm tooling.
End-of-arm tooling is demarcated:
In accurately compensating for strategy, need accurately to demarcate the end-of-arm tooling of mechanical arm.Mechanical arm tail end method
End-of-arm tooling all be can be regarded as in part after orchid, include the six-dimension force sensor installed additional to realize quadruple accuracy compensation method, with
And the tool that mechanical arm must install additional to realize job task.The model of tool indicates with X, Y, Z, B, C, i.e. X, Y, Z
The displacement in three directions is plus the corner around Y, around X.B, the nominal value of C is 0, and X, Y, Z also have respective nominal value.
Allow manipulator motion to a specific position first, the command value of object attitude angle is 0.Measure mechanical arm tail end
The position of tool and posture, can obtain at this time mechanical arm base coordinate system to the homogeneous transformation T of tool coordinates systemXYZABC.Then
The end flange rotary joint of mechanical arm is rotated into specific angle z, measures position and the posture of mechanical arm end-of-arm tooling again,
A T can be obtained againXYZABC.Enough times are measured, a series of z-T can be obtainedXYZABC, there is following relational expression:
TXYZABC(z)=TBase→FlangeTRot(z)Ttool,BCTtool,XYZ (10)
Wherein, TBase→FlangeIndicate mechanical arm pedestal to the transformation matrix of end flange, TRot(z) rotational angle z is indicated
The rotational transformation matrix of generation, Ttool,BCAnd Ttool,XYZIndicate true rotation transformation and the translation transformation square of mechanical arm tail end tool
Battle array.
The posture part in relational expression above is only investigated, following relationship can be obtained:
RABC(z)=RBase→FlangeRot(z)Rtool,BC (11)
Our target is that lateral attitude angle B, C of tool are found out from above-mentioned posture relationship, can use intelligent search
Method solves.The absolute value of the difference for the attitude angle that the target function of intelligent search is characterized by above-mentioned posture relationship "=" both sides
Expectation, selected tool tilt attitude angle B, C of the smaller explanation of target function value is better.It is specific for selected one group
Tool tilt attitude angle B, C, we can be handled as follows:First R is found out in z=0Base→Flange, then utilize z ≠ 0
When posture relational expression and RBase→FlangeFind out target function value.
In TXYZABCRelational expression in, specify following symbol:
The relational expression about position as follows can then be obtained:
Further transformation can obtain:
According to the different values of the rotational angle z of end flange rotary joint, can obtain a series of shaped like (13)
Relational expression combines them:
It can be in the hope of X, Y, Z using least square method.
Whole feature X, Y, Z, B, C in mechanical arm tail end tool model have thus been obtained, end-of-arm tooling mark is completed
It is fixed.
Gravity compensation calibration process:
For each node in the mesh generation of payload w, mechanical arm tail end tool is added according to load requirement
It carries.The six-dimension force sensor installed additional using mechanical arm tail end flange weighs to load, to ensure the accuracy of load, simultaneously also
Inclined square is not generated after utilizing six-dimension force sensor to ensure load.After the completion of load, control machinery arm is transported with particular end posture
It moves in effective working space in each nodes of locations of mesh generation, i.e. mesh generation nodes of locations and particular end posture
For the object pose of mechanical arm.After the completion of movement, the attained pose of mechanical arm is measured, it is bound to wrong between object pose
Difference, this error are exactly the compensation rate of gravity compensation.
Under normal conditions, for one group of specific value of terminal position theoretical value, we are by three postures of terminal angle
The theoretical value of angle A, B, C are taken as A0, 0,0, wherein A0For the specific angles an A value.Have at this time:
Wherein,For the measured value of mechanical arm tail end pose,For the theory target of mechanical arm tail end pose
Value,For the gravity compensation matrix of mechanical arm.
By gravity compensation matrixIt is converted into pose form (3 position compensation amounts, 3 attitude angle compensation rates), this
6 numerical value are exactly the calibration result of this calibration point.All nodes in four-dimensional mesh generation are measured, and calculate the mark of each node
Calmly as a result, ultimately forming gravity compensation database, gravity compensation calibration process is completed.
Gravity compensation control method:
Gravity compensation is embodied in control method, is exactly by calculating, changing the object pose of mechanical arm.To mechanical arm
Object pose is actually the pose T that mechanical arm needs actual motion to arrivereal.If without gravity compensation directly by TrealLife
Order is transmitted to mechanical arm, and mechanical arm can not finally reach this pose because of gravitational settling.What gravity compensation to be done is exactly to pass through
The database established when inquiry calibration, finds corresponding Tcomp, pass through relational expression:
TcompTreal=Tthe (16)
Calculate Tthe, by TtheIt is transmitted to mechanical arm as the pose after gravity compensation, after so that mechanical arm is moved by this pose
Just reach Treal。
For problem one, needs to use the end-of-arm tooling carried out before and demarcate to solve.If it is known that Tthe,ABC, Ke Yitong
Cross following formula:
It is calculatedWherein, Ttool,XYZ,theIndicate ideal end-of-arm tooling transformation, TZtIndicate mechanical arm wrist
Center is to the transformation at end flange center, TRot(A-A0, B, C) and it indicates from attitude angle A0, 0,0 to attitude angle A, B, C rotation become
Change matrix.Following formula is may then pass through, byCalculate to obtain Tcomp,ABC:
Wherein, Ttool,XYZAnd Ttool,BCRespectively indicate end-of-arm tooling actual end-of-arm tooling translation transformation obtained by calibrating and
End-of-arm tooling rotation transformation.
For problem two, there is gravity compensation relational expression:
Tcomp,ABCTreal,ABC=Tthe,ABC (19)
I.e. as known Treal,ABCWhen, as long as providing Tcomp,ABC, so that it may to obtain Tthe,ABC.Then it can pass through such as Fig. 3 institutes
The gravity compensation control flow that shows solves the problems, such as two.
Inclined square compensation:
Inclined square refers to the payload loaded on mechanical arm tail end tool, since load center of gravity is not or not end flange center
And the loading moment relative to end flange center generated.The main effects of inclined square is to make six-dimension force sensor generate deflection to become
Shape, to generate very important influence to mechanical arm tail end positioning accuracy.Inclined square compensation is on the amendment hand of this influence
Section.
The load of mechanical arm, there may be the inclined square of both direction, is passed around six-dimensional force respectively relative to end flange center
The x-axis direction of sensor, the y-axis direction around six-dimension force sensor.Correspondingly, the anamorphic effect that inclined square generates be exactly end-of-arm tooling around
The deflection in the two directions.
Inclined square compensates calibration process:
The inclined square range that the payload of mechanical arm may generate in two directions is determined first.It is right within the scope of inclined square
Inclined square in both direction carries out mesh generation, generates a two-dimensional calibration grid.Must include point (0,0) in the grid,
The benchmark that (0,0) is compensated as inclined square, i.e., inclined square compensation rate is all 0 when not loading inclined square to mechanical arm.To in two-dimensional grid
Each node surveyed in real time using six-dimension force sensor according to inclined square numerical value to the inclined square of mechanical arm tail end tool loads two dimension
Inclined square value is measured, ensures that load is accurate.
The influence for excluding load gravity and mechanical arm dead weight is deformed is had to when square compensation is demarcated partially.To reach this mesh
, pose on the basis of the most frequently used pose of a mechanical arm is defined first, and all inclined squares compensation calibration is carried out in this pose.
Then control machinery arm is needed once to be moved using benchmark pose as target after load is accurate each inclined square of two dimension,
Need application in order to demarcate all accuracy compensation control methods of completion in this movement, with eliminate in addition to inclined square it
The influence of other outer error sources, i.e.,:
Gravity compensation control → model compensation control → return difference compensation control;
After the completion of movement, the pose of mechanical arm tail end tool at this time is measured, thus calculates end-of-arm tooling around both direction
Deflection angleThe square combination partially of all two dimensions is measured, corresponding deflection angle is calculated, forms inclined square compensation database.By
In the benchmark for compensating (0,0) as inclined square, so for the square combination partially of all two dimensions, it is real caused by inclined square
Border deflection angle is:
Inclined square compensating control method:
Inclined square compensation is embodied in control method, is exactly by calculating, changing the object pose of mechanical arm so that mechanical arm
The deflection that inclined square generates can be just offset according to the pose movement after compensation, just as not loading inclined square also without using inclined square
Compensation control is the same.
The numerical value of inclined square is read first with six-dimension force sensor, then tables look-up in inclined square compensation database and utilizes two
Dimension interpolation obtains deflection angle β, γ, carries out inclined square compensation control using following formula later:
Tout=TinTtool,XYZ,the -1TRot(-β,-γ)Ttool,XYZ,the (20)
Wherein, TinIndicate the object pose before inclined square compensation, TRotThe rotation transformation square that (- β ,-γ) expression-β ,-γ are generated
Battle array, ToutIndicate the pose of inclined square compensation output.
Control main-process stream:
If before no any accuracy compensation of progress, the object pose of mechanical arm is denoted as Tar, then quadruple accuracy compensation
The control main-process stream of method is as shown in Figure 4.As seen in Figure 4, the control of quadruple accuracy compensation method carried out inclined square compensation before this, then carried out
Gravity compensation is model compensation later, is finally return difference compensation.
Quadruple accuracy compensation method is when being applied to large-scale heavy duty mechanical arm-proton therapeutic bed, existing proton therapeutic
Bed weight is at 2.5 tons or so, and mechanical arm nominal load is 150kg, and arm exhibition is 3 meters, and quadruple accuracy compensation method makes the mechanical arm
Absolute fix precision is more than 40mm in the case of only carrying out zero position calibration, is improved to using after quadruple accuracy compensation method
It is better than 0.2mm, more than 2 orders of magnitude of absolute fix precision improvement.