CN104589349A - Combination automatic control method with single-joint manipulator under mixed suspension microgravity environments - Google Patents

Abstract

The invention provides a combination automatic control method with a single-joint manipulator under mixed suspension microgravity environments. The combination automatic control method comprises the following steps of 1, enabling a combination to be equivalent to an underwater robot, and establishing a kinematics equation and a dynamics equation; 2, approximating the dynamics equation of the combination by a radial basis function neural network, so as to obtain control force and control torque corresponding to the radial basis function neural network; 3, using a sliding mode control method, so as to obtain control force and control torque corresponding to sliding model control; 4, synthesizing the control force and control torque corresponding to the neural network and the control force and control torque obtained by the sliding model control method, and distributing thrust, so as to obtain a general vector which consists of thrust and joint torque of each propeller; approximating the thrust deviation of the corresponding thruster through the radial basis function neural network, so as to obtain the estimation value of the thrust deviation; 5, combining the results obtained in step 2, step 3 and step 4, obtaining the general vector consisting of the thrust and the joint torque of the corresponding propeller, and further obtaining the thrust and the joint torque of the corresponding propeller, so as to realize the automatic control.

Description

With the autonomous control method of assembly of simple joint mechanical arm under a kind of mix suspending microgravity environment

Technical field

The present invention relates to the movement control technology of underwater robot, be specially the autonomous control method of assembly with simple joint mechanical arm under a kind of mix suspending microgravity environment.

Background technology

Assembly is formed by testing after body and suspension target are docked, and comprising one can the joint of controlled rotation, because assembly working environment is in water, can be regarded as the underwater robot system being with simple joint mechanical arm.Underwater robot system have non-linear, time the feature such as change, close coupling, this proposes very large challenge to the design of its controller.

Large quantifier elimination has been carried out at present for the control technology of submarine navigation device.Multiple different control method comprises Linear Control, robust control, fuzzy control, and Self Adaptive Control etc. have been used to the motion control of submarine navigation device all.But above-mentioned often kind of control technology has its specific aim and limitation.

Traditional local linearization and PID control method need to have the parameter of hydrodynamic model to understand more accurately, and the controller designed only can ensure to have good performance near equalization point.Consider that the kinetic parameter of assembly can change along with the change of displacement state and configuration, controller must revise the control law of self in real time to ensure that whole control system can obtain satisfied performance all the time.

The developing into address this problem of Non-Linear Control Theory provides new thinking.Assembly under mix suspending microgravity environment is a typical multi input, multi output, non-linear, strongly coupled system, and being difficult to Obtaining Accurate due to its hydrodynamic force coefficient, the reasons such as change cannot obtain the mathematical models of system with assembly change of configuration for the kinematics of system and kinetic parameter.Therefore to meet high-precision control overflow, with regard to need controller can with the change of control object parameter the parameter of auto modification controller or structure, Self Adaptive Control and ANN Control be exactly in order to meet these type of needs and propose control strategy.

Adaptive controller can revise the characteristic of oneself in time with the change of the dynamic characteristic of suitable solution object, enables system keep optimum or suboptimum duty.In prior art, adopt the control two kinds of direct adaptive control rule being used for Autonomous Underwater Vehicle (AUV), the control effects of two kinds of control laws under measurement noises exists situation by simulation comparison.And for the AUV spatial movement dynamic system model containing uncertain item, propose a kind of self adaptation back stepping control device, the design of this controller does not need the hydrokinetic parameter predicting AUV, and simulation result shows that this control law performance is obviously better than traditional PID and controls.

Although above-mentioned adaptive controller can carry out identification to parameter uncertainty, the parameter uncertainty of kinetic model is compensated by real-time modifier controller parameter, but it is uncertain for the nonparametric such as external disturbance and Unmarried pregnancy, only rely on adaptive controller to be difficult to the stability of guarantee system, need to combine with other control strategy the robustness improving system.

Neutral net has the None-linear approximation mapping ability of height, adopts neutral net can realize accurately approaching unknown portions in submarine navigation device kinetics equation, thus by feedforward compensation, the high-precision motion realizing submarine navigation device controls.Disclose a neural network control device in prior art to realize controlling the six degree of freedom of submarine navigation device, wherein used two kinds of different neural network structures to approach the nonlinear kinetics of submarine navigation device, the kinetic model without the need to submarine navigation device can realize the good tracking to desired trajectory.But the method does not consider the saturated impact on controller performance of propeller thrust, and have ignored the approximate error of neutral net.

In order to meet the requirement of combination experiment body for controller stability, robustness and adaptive ability, consider that the kinetic model of assembly system cannot accurately obtain and propeller exists the conditions such as thrust constraint of saturation simultaneously, need the autonomous control method proposing assembly under a kind of new mix suspending microgravity environment.

Summary of the invention

For problems of the prior art, the invention provides a kind of control accuracy high, robustness is good, and the position control meeting six-freedom degree pose control and the joint of mechanical arm angle of testing body requires, with the autonomous control method of the assembly of simple joint mechanical arm under mix suspending microgravity environment.

The present invention is achieved through the following technical solutions:

With the autonomous control method of assembly of simple joint mechanical arm under a kind of mix suspending of the present invention microgravity environment, comprise,

Step one, is equivalent to underwater robot by assembly, sets up kinematical equation and kinetics equation;

Step 2, is approached by the kinetics equation of radial base neural net to assembly, obtains the control force and moment that radial base neural net is corresponding, is expressed as

Step 3, obtains sliding formwork by sliding-mode control and controls corresponding control force and moment, be expressed as τ _smc;

Step 4, by control corresponding for neutral net in step 2 and control moment, after the control obtained with sliding formwork control methods in step 3 and control moment synthesize, distributes the descriptor vector u obtaining each angle of rake thrust and joint moment of torsion composition by thrust _dit is as follows,

By radial base neural net, angle of rake thrust deflexion is approached, obtain thrust deflexion estimate

Step 5, the result obtained in combining step two to step 4, obtains the descriptor vector u of propeller propulsive force and joint moment of torsion composition, wherein, for the generalized inverse matrix of thrust allocation matrix B; Thus obtain propeller thrust and joint moment of torsion realize to assembly from main control.

Preferably, in step one,

The assembly kinematical equation set up is as follows,

$\stackrel{\·}{\mathrm{\ξ}}=\left[\begin{array}{c}{\stackrel{\·}{\mathrm{\η}}}_1\\ {\stackrel{\·}{\mathrm{\η}}}_2\\ \stackrel{\·}{q}\end{array}\right]=\left[\begin{array}{ccc}{J}_1\left(\mathrm{\η}\right)& O_{3\×3}& O_{3\×n}\\ O_{3\×3}& J_2\left(\mathrm{\η}\right)& O_{3\×n}\\ O_{n\×3}& O_{n\×3}& I_n\end{array}\right]\left[\begin{array}{c}{v}_1\\ v_2\\ \stackrel{\·}{q}\end{array}\right]=J^{\′}\left(\mathrm{\η}\right)\mathrm{\ζ};$

Wherein, represent position and the attitude vectors of in body-fixed coordinate system, testing body; represent linear velocity and the angular velocity vector of in the solid coordinate system of body, testing body; for body consolidates the transformation matrix of coordinate system and body-fixed coordinate system centerline velocities; J ₂(η) be the transformation matrix of angular speed in the solid coordinate system of body and body-fixed coordinate system; Q is the joint rotation angle of mechanical arm;

The assembly kinetics equation set up is as follows,

$M\left(q\right)\stackrel{\·}{\mathrm{\ζ}}+C(q,\mathrm{\ζ})\mathrm{\ζ}+D(q,\mathrm{\ζ})\mathrm{\ζ}+g(q,\mathrm{\η})=\mathrm{\τ};$

Wherein,

M (q) ∈ R ^{(6+n) × (6+n)}, be the inertial matrix caused by assembly and additional mass;

C (q, ζ) ∈ R ^{(6+n) × (6+n)}, the centripetal force caused by assembly and additional mass and coriolis force matrix;

D (q, ζ) ∈ R ^{(6+n) × (6+n)}, be hydrodynamic damping matrix;

G (q, η) ∈ R ⁽⁶⁺ⁿ⁾, be gravity and buoyancy vector;

τ ∈ R ⁽⁶⁺ⁿ⁾, be generalized force vector;

R represents real number, can represent real vector or real matrix according to subscript difference;

N represents the free degree of mechanical arm.

Further, in step 3,

Sliding formwork control item form in sliding-mode control is as follows:

τ _smc＝K _Ds′+K _Ssgn(s)；

Wherein,

$s^{\′}=\left[\begin{array}{c}{\tilde{v}}_1\\ {\tilde{v}}_2\\ \stackrel{\stackrel{\·}{~}}{q}\end{array}\right]+(\mathrm{\Λ}+K_D^{-1}{K}_P)\left[\begin{array}{c}R_I^B{\widetilde{\mathrm{\η}}}_1\\ \widetilde{\mathrm{\ϵ}}\\ \tilde{q}\end{array}\right]=\widetilde{\mathrm{\ζ}}+(\mathrm{\Λ}+K_D^{-1}{K}_P)\tilde{y},s=\widetilde{\mathrm{\ζ}}+\mathrm{\Λ}\tilde{y},$

S is the sliding-mode surface of design, Λ and K _pfor positive definite diagonal matrix, K _d, K _sfor symmetric positive definite matrix, sgn (s) is sign function; represent the deviation between desired value and actual value, desired value is exactly the desired value of experiment posture, speed and joint position, speed, and actual value is the experiment body movement state information that measuring system obtains, it is the experiment body attitude error represented by hypercomplex number;

According to desired control target ξ _d, ζ _dwith the assembly movement state information ξ that measurement obtains, ζ calculates s and s ', s and s ' is substituted into sliding formwork control item and namely obtains the control force and moment that sliding formwork controls correspondence.

Further again, following process is done to sign function sgn (s),

$\mathrm{sgn}\left(s\right)\≈\frac{s}{\left|s\right|+\mathrm{\δ}};$

Wherein, δ is an arithmetic number.

Further again, in step 2, pass through radial base neural net the nonlinear dynamical equation of assembly is approached,

ANN Control item is expressed as follows:

${\mathrm{\τ}}_{\mathrm{rbn}}=\hat{f}(\mathrm{\ξ},\mathrm{\ζ},{\mathrm{\ζ}}_r,{\stackrel{\·}{\mathrm{\ζ}}}_r)=\widehat{\mathrm{\ω}}\mathrm{\γ}(x,c);$

Wherein, the input vector x of neutral net is by four generalized state vectors of assembly composition, ζ _r=ζ+s, output vector is control and the control moment of assembly, and ω, for exporting weight matrix, will be determined by adaptive updates rule.

Further again, in step 4, utilize radial base neural net propeller thrust deviation delta u is estimated, estimate for introducing in controller as feedback quantity to compensate thrust deflexion Δ u.

Further again, in step 2 and step 4, according to the change of assembly motion state and configuration, the output weight matrix of online updating neutral net;

The neutral net being used for approaching system dynamics exports weight adaptive updates and restrains as follows,

${\stackrel{\stackrel{\·}{^}}{\mathrm{\ω}}}^T={\mathrm{\Γ}}_{\mathrm{\ω}}\mathrm{\γ}(x,c)s^T;$

The neutral net being used for approaching thrust deflexion exports weight adaptive updates and restrains as follows,

${{\stackrel{\stackrel{\·}{^}}{\mathrm{\ω}}}_u}^T={\mathrm{\Γ}}_u{\mathrm{\γ}}_u(x,c)s^{T}B;$

Wherein, matrix B is thrust allocation matrix, Γ _ωand Γ _ufor positive definite diagonal matrix, represent the adaptive gain of weight more new law.

Further again, in step 4, allocation matrix when assembly thrust is distributed is as follows,

$B=\left[\begin{array}{cc}L& 0\\ 0& I\end{array}\right]$

Wherein, L is the mapping matrix carrying out thrust distribution; I is the joint torque distribution matrix of mechanical arm, is unit battle array.

Compared with prior art, the present invention has following useful technique effect:

The present invention is approached by radial base neural net and the nonlinear kinetics of compensating group zoarium and thrust deflexion value, realizes the estimation to controlling force and moment; Utilize sliding-mode control simultaneously, by the setting of sliding formwork control item, be used for compensating the approximate error of neutral net and improving robustness and the response speed of system, make to control more accurately with stable; Finally utilize thrust to distribute and realize controlling the correspondence of each thruster; To realize under mix suspending microgravity environment with the assembly of simple joint mechanical arm from main control, the six-freedom degree pose meeting experiment body controls and the position control requirement at joint of mechanical arm angle, and practicality is good, and exploitativeness is high.

Further, the adaptive updates rule of weight is exported by network, complete on-line study according to the motion state of assembly or the auxiliary nervous network of output weight matrix of thrust design value online updating neutral net, ensure that the accuracy that neutral net is approached mission nonlinear dynamics and thrust deflexion.

Accompanying drawing explanation

Fig. 1 is the radial base neural net structural representation for approaching kinetics equation described in example of the present invention.

Fig. 2 is the propeller layout of assembly described in example of the present invention.

Fig. 3 is the adaptive neural network sliding formwork control logic block diagram of control method autonomous described in example of the present invention.

Detailed description of the invention

Below in conjunction with specific embodiment, the present invention is described in further detail, and the explanation of the invention is not limited.

The present invention is based on ANN Control, sliding formwork controls and Adaptive Control Theory, with the autonomous control method of assembly of simple joint mechanical arm under proposition mix suspending microgravity environment, realize assembly under mix suspending microgravity experiment environment from main control, the six-freedom degree pose of namely testing body controls and the position control of joint of mechanical arm.Wherein neutral net is used for approaching nonlinear kinetics and the thrust deflexion value of assembly, adaptive updates rule completes on-line study by the auxiliary nervous network of output weight matrix of online updating neutral net, and sliding formwork controls robustness and the response speed of approximate error and the raising system being then mainly used to compensate neutral net.Concrete autonomous control method is as follows.

Step one: set up assembly kinematics and kinetics equation.

Because assembly can be equivalent to underwater robot, therefore set up kinematical equation and kinetics equation is as follows,

The kinematical equation of assembly is as follows:

$\stackrel{\·}{\mathrm{\ξ}}=\left[\begin{array}{c}{\stackrel{\·}{\mathrm{\η}}}_1\\ {\stackrel{\·}{\mathrm{\η}}}_2\\ \stackrel{\·}{q}\end{array}\right]=\left[\begin{array}{ccc}{J}_1\left(\mathrm{\η}\right)& O_{3\×3}& O_{3\×n}\\ O_{3\×3}& J_2\left(\mathrm{\η}\right)& O_{3\×n}\\ O_{n\×3}& O_{n\×3}& I_n\end{array}\right]\left[\begin{array}{c}{v}_1\\ v_2\\ \stackrel{\·}{q}\end{array}\right]=J^{\′}\left(\mathrm{\η}\right)\mathrm{\ζ}---\left(1\right);$

Wherein, represent position and the attitude vectors of in body-fixed coordinate system, testing body; represent linear velocity and the angular velocity vector of in the solid coordinate system of body, testing body; for body consolidates the transformation matrix of coordinate system and body-fixed coordinate system centerline velocities; J ₂(η) be the transformation matrix of angular speed in the solid coordinate system of body and body-fixed coordinate system; Q is the joint rotation angle of mechanical arm.

The kinetics equation of assembly is as follows:

$M\left(q\right)\stackrel{\·}{\mathrm{\ζ}}+C(q,\mathrm{\ζ})\mathrm{\ζ}+D(q,\mathrm{\ζ})\mathrm{\ζ}+g(q,\mathrm{\η})=\mathrm{\τ}---\left(2\right)$

Wherein,

M (q) ∈ R ^{(6+n) × (6+n)}---assembly and the inertial matrix caused by additional mass;

C (q, ζ) ∈ R ^{(6+n) × (6+n)}---assembly and the centripetal force caused by additional mass and coriolis force matrix;

D (q, ζ) ∈ R ^{(6+n) × (6+n)}---hydrodynamic damping matrix;

G (q, η) ∈ R ⁽⁶⁺ⁿ⁾---gravity and buoyancy vector;

τ ∈ R ⁽⁶⁺ⁿ⁾---generalized force vector;

R represents real number, can represent real vector or real matrix according to subscript difference;

N represents the free degree of mechanical arm, and the mechanical arm due to the assembly that the present invention is directed to only has a joint, therefore n=1.

The each symbol implication representing assembly motion state in subsequent step is identical with the implication of respective symbol in (2) with the equation of motion (1), is no longer described in subsequent step.

Step 2: the Design and implementation of ANN Control item.

Radial base neural net (RBN) is a kind of three layers of feedforward network with single hidden layer, has proved that RBN can approach a given nonlinear function with arbitrary accuracy in theory.RBN is nonlinear by the mapping being input to output, and hidden layer space is linear to the mapping of output region, and due to the action function in RBN be Gaussian bases, its value is nonzero value in limited range in the input space, thus RBN network is the neutral net of partial approximation, so adopt RBN greatly can accelerate pace of learning and avoid local minimum problem, be applicable to the needs controlled in real time.

The representation that radial base neural net is general is as follows:

${\hat{f}}_i\left(x\right)=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\widehat{\mathrm{\ω}}}_{\mathrm{ij}}{\mathrm{\γ}}_j\left(x{,c}_j\right)i=1,...,n$

Wherein, x=[x ₁, x ₂..., x _n] ^tfor the input vector of network, the Gaussian bases of a hidden layer jth node, c _jand σ _jrepresent center vector and the base width parameter of a hidden layer jth node respectively, be the output weight parameter of network, N is the nodes of network hidden layer, and n is the dimension of network output vector.

Radial base neural net as shown in Figure 1 will be adopted in the present invention to the nonlinear kinetics of assembly $M\left(q\right){\stackrel{\·}{\mathrm{\ζ}}}_r+C(q,\mathrm{\ζ}){\mathrm{\ζ}}_r+D(q,\mathrm{\ζ}){\mathrm{\ζ}}_r+g\left(\mathrm{\ζ}\right)$ Approach.

ANN Control item in controller can be expressed as follows:

${\mathrm{\τ}}_{\mathrm{rbn}}=\hat{f}(\mathrm{\ξ},\mathrm{\ζ},{\mathrm{\ζ}}_r,{\stackrel{\·}{\mathrm{\ζ}}}_r)=\widehat{\mathrm{\ω}}\mathrm{\γ}(x,c)---\left(3\right)$

Wherein, the input vector x of neutral net is by four generalized state vectors of assembly system composition, the expression of s refers to step 3, and output vector is control and the control moment of assembly, and exporting weight matrix ω will be determined by adaptive updates rule, and more the specific implementation form of new law is see step 4.

Step 3: the Design and implementation of sliding formwork control item.

Consider the computational efficiency of controller, the node of radial base neural net hidden layer should not arrange too much, therefore inevitably there is certain error to the compensation of assembly nonlinear kinetics in neutral net, the present invention compensates the approximate error of neutral net by sliding-mode control, improves the response speed of controller and the robustness of system simultaneously.

Sliding formwork control item form is as follows:

τ _smc＝K _Ds′+K _Ssgn(s) (4)

Wherein,

$s^{\′}=\left[\begin{array}{c}{\tilde{v}}_1\\ {\tilde{v}}_2\\ \stackrel{\stackrel{\·}{~}}{q}\end{array}\right]+(\mathrm{\Λ}+K_D^{-1}{K}_P)\left[\begin{array}{c}R_I^B{\widetilde{\mathrm{\η}}}_1\\ \widetilde{\mathrm{\ϵ}}\\ \tilde{q}\end{array}\right]=\widetilde{\mathrm{\ζ}}+(\mathrm{\Λ}+K_D^{-1}{K}_P)\tilde{y},s=\widetilde{\mathrm{\ζ}}+\mathrm{\Λ}\tilde{y},$

S is the sliding-mode surface of design, Λ and K _pfor positive definite diagonal matrix, K _d, K _sfor symmetric positive definite matrix; represent the deviation between desired value and actual value, expected value of the present invention is exactly the desired value of experiment posture, speed and joint position, speed, and actual value is the experiment body movement state information that measuring system obtains, it is the experiment body attitude error represented by hypercomplex number.

By according to desired control target ξ _d, ζ _dwith the assembly movement state information ξ that measuring system provides, ζ calculates s and s ', s and s ' is substituted into the sliding formwork control item shown in formula (4) and can obtain the control force and moment that sliding formwork controls correspondence.In actual use, in order to eliminate the flutter that sign function sgn (s) causes, generally following process is done:

$\mathrm{sgn}\left(s\right)\≈\frac{s}{\left|s\right|+\mathrm{\δ}};$

Wherein, δ is a little arithmetic number, carries out choosing from Row sum-equal matrix according to control effects.

Step 4, controller thrust is distributed.

The assembly position of experiment and gesture stability finally realize by controlling 6 angle of rake thrusts, so need the control of experiment body and control moment to carry out thrust distribution.And it is relevant with angle of rake layout with control moment to test control suffered by body.In the present invention, assembly adopts three groups of totally 6 screw propellers, often organizes and arranges along body coordinate system three axial symmetry respectively, and propeller produces thrust T ₁, T ₂, T ₃, T ₄, T ₅, T ₆direction as shown in Figure 2.

Experiment body control and control moment vector τ _bcan be expressed as follows:

τ _b＝LT

Wherein, T=[T ₁, T ₂, T ₃, T ₄, T ₅, T ₆] ^tthe vector be made up of each angle of rake thrust; L is the mapping matrix carrying out thrust distribution, and consider the propeller layout of experiment body, L is expressed as follows:

$L=\left[\begin{array}{cccccc}1& 1& 0& 0& 0& 0\\ 0& 0& 1& 1& 0& 0\\ 0& 0& 0& 0& 1& 1\\ 0& 0& l_3& -l_4& 0& 0\\ 0& 0& 0& 0& l_5& -l_6\\ l_1& -l_2& 0& 0& 0& 0\end{array}\right]$

Wherein, l _irepresent the vertical range between the solid coordinate system reference axis of i-th propeller and body parallel with it.

The joint torque distribution matrix considering mechanical arm is unit battle array, and the final form of assembly thrust allocation matrix is as follows:

$B=\left[\begin{array}{cc}L& 0\\ 0& I\end{array}\right]$

Wherein, after the control obtain sliding formwork control item in ANN Control item in step 2 and step 3 and control moment synthesize, with the generalized inverse matrix of thrust allocation matrix B be multiplied by the descriptor vector u that final control and control moment vector can obtain thrust needed for each propeller and joint moment of torsion composition _d.

Consider that the propeller of assembly exists thrust saturated phenomenon, the thrust of the therefore actual output of propeller with Controller gain variations thrust u _dbetween there is deviation delta u, this deviation will have influence on the control performance of controller, so also utilize radial base neural net in the present invention propeller thrust deviation delta u is estimated, estimate introduced in controller by as feedback quantity to compensate thrust deflexion Δ u.Wherein, the input vector of neutral net is Controller gain variations thrust u _d, output vector is the thrust deflexion Δ u of assembly, exports weight matrix ω _uto be determined by adaptive updates rule.

In order to realize the on-line study of neutral net, needing that weight matrix is exported to it and carrying out adaptive updates, namely according to the change of assembly motion state and configuration, the output weight matrix of online updating neutral net,

The neutral net being used for approaching system dynamics exports weight adaptive updates and restrains as follows:

${\stackrel{\stackrel{\·}{^}}{\mathrm{\ω}}}^T={\mathrm{\Γ}}_{\mathrm{\ω}}\mathrm{\γ}(x,c)s^T---\left(5\right)$

The neutral net being used for approaching thrust deflexion exports weight adaptive updates and restrains as follows:

${{\stackrel{\stackrel{\·}{^}}{\mathrm{\ω}}}_u}^T={\mathrm{\Γ}}_u{\mathrm{\γ}}_u(x,c)s^{T}B---\left(6\right)$

Wherein, matrix B is thrust allocation matrix, and concrete form is see step 5; Γ _ωand Γ _ufor positive definite diagonal matrix, represent the adaptive gain of weight more new law.By the hidden layer function gamma (x in step 2, the s that obtains substitutes into formula (5) and formula (6) ξ) and in step 3, the rate of change that two neutral nets export weight matrix can be obtained, can be obtained the updated value of weight matrix by integration according to the rate of change of weight matrix, new weight matrix will be utilized in step 2 and step 4.

Step 5: Controller gain variations result

The result of combining step two to step 4, the final way of realization that can obtain controller is as follows:

In formula (7), the implication of each symbol and implementation method have been described in detail in step 2 to step 4.

The present invention as shown in Figure 3, realizes it and is made up of three parts from the method for main control.One is ANN Control item, utilizes radial base neural net to approach and the nonlinear kinetics of compensating group zoarium and thrust deflexion value, represents in block diagrams with circle dotted box portion.Two is sliding formwork control items, is mainly used to compensate the approximate error of neutral net and improve robustness and the response speed of system, represents in block diagrams by dotted-line ellipse frame part.Three is adaptive updates rules that network exports weight, and it act as and completes on-line study according to the motion state of assembly or the auxiliary nervous network of output weight matrix of thrust design value online updating neutral net, represents in block diagrams by square solid box part.Executing agency due to assembly is six screw propellers and joint motor, so controller also will carry out thrust distribution, in Fig. 3 for the generalized inverse matrix of thrust allocation matrix B, for control and control moment are mapped as each angle of rake thrust and joint control moment.All the other symbols in block diagram elaborate in other parts of invention.Can according to the assembly movement state information comprising experiment body position attitude, speed and joint position, the desired control target of desired value of speed and measuring system and obtain, calculate required propeller thrust and joint moment of torsion, and it can be used as control instruction to issue assembly to complete motion control to assembly, and be less than 3 centimetres by its control accuracy of case verification satisfied experiment body position by mistake, attitude error is less than 3 degree, and pose error is less than 1 degree.To realize under mix suspending microgravity environment with the assembly of simple joint mechanical arm from main control, the six-freedom degree pose meeting experiment body controls and the position control requirement at joint of mechanical arm angle.

In the following example, the control effects adopting control method of the present invention to obtain is as follows respectively.Example 1, experiment body position attitude remains unchanged, and controls joint and rotates.

η ₀＝η _d＝[0 0 0 0 0 0] ^Tm,rad

q ₀＝0 rad $q_d=\frac{\mathrm{\π}}{2}\mathrm{rad}$

Propeller thrust saturation value is set to 30N, and joint moment of torsion saturation value is set to 30Nm.

Wherein, η ₀and η _dbe respectively initial position attitude and the desired locations attitude of experiment body.Q ₀and q _dbe respectively initial value and the desired value of assembly joint rotation angle.

Control law optimum configurations is as follows:

Λ＝0.5diag{0.5 0.5 0.5 0.5 0.5 0.5 0.5}

K _D＝50diag{1 1 1 1 1 1 0.5}

K _S＝50diag{1 1 1 1 1 1 0.5}

K _P＝50diag{1 1 1 1 1 1 0.5}

Γ _ω=0.1I σ _w=30c _w∈ [-4 4] number of network node=401

Γ _u=0.1I σ _u=1000c _u∈ [-400 400] number of network node=201

Experiment body position attitude remains unchanged, and control joint and rotate, be i.e. position retentive control, its control accuracy is shown in 1:

Table 1 combination experiment body position retentive control precision

Example 2: joint position remains unchanged, controls body motion.

η ₀＝[0 0 0 0 0 0] ^Tm,rad

η _d＝[1 1 1 0 0 0] ^Tm,rad

q ₀＝q _d＝0 rad

Propeller thrust saturation value is set to 30N, and joint moment of torsion saturation value is set to 30Nm.

Wherein, η ₀and η _dbe respectively initial position attitude and the desired locations attitude of experiment body.Q ₀and q _dbe respectively initial value and the desired value of assembly joint rotation angle.

Control law optimum configurations is as follows:

Λ＝0.5diag{0.5 0.5 0.5 0.5 0.5 0.5 0.3}

K _D＝50diag{1 1 1 1 1 1 0.3}

K _S＝50diag{1 1 1 1 1 1 0.3}

K _P＝50diag{1 1 1 1 1 1 0.3}

Γ _ω=0.1I σ _w=30c _w∈ [-4 4] number of network node=401

Γ _u=0.1I σ _u=1000c _u∈ [-400 400] number of network node=201

Assembly joint position remains unchanged, and Control release posture reaches setting value, i.e. assembly position control, and its control accuracy is in table 2:

Table 2 combination experiment body position control precision

Can be found out by table 1 and table 2, for described two scenes, the controller of the present invention's design is adopted to control assembly position, attitude and joint position, its steady-state error is all very little, illustrate thus controller given by the present invention at control object Unknown Parameters and systematic parameter change with assembly configuration and change still have good control effects.The stability of the control law that simultaneous verification is proposed by the invention.

Claims (8)

1. under mix suspending microgravity environment with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, comprise,

Step one, is equivalent to underwater robot by assembly, sets up kinematical equation and kinetics equation;

Step 2, is approached by the kinetics equation of radial base neural net to assembly, obtains the control force and moment that radial base neural net is corresponding, is expressed as

Step 3, obtains sliding formwork by sliding-mode control and controls corresponding control force and moment, be expressed as τ _smc;

Step 4, by control corresponding for neutral net in step 2 and control moment, after the control obtained with sliding formwork control methods in step 3 and control moment synthesize, distributes the descriptor vector u obtaining each angle of rake thrust and joint moment of torsion composition by thrust _dit is as follows,

By radial base neural net, angle of rake thrust deflexion is approached, obtain thrust deflexion estimate

Step 5, the result obtained in combining step two to step 4, obtains the descriptor vector u of propeller propulsive force and joint moment of torsion composition, wherein, for the generalized inverse matrix of thrust allocation matrix B; Thus obtain propeller thrust and joint moment of torsion realize to assembly from main control.

2. under a kind of mix suspending microgravity environment according to claim 1 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, in step one,

The assembly kinematical equation set up is as follows,

$\stackrel{.}{\mathrm{\ξ}}=\left[\begin{array}{c}{\stackrel{.}{\mathrm{\η}}}_1\\ {\stackrel{.}{\mathrm{\η}}}_2\\ \stackrel{.}{q}\end{array}\right]\left[\begin{array}{ccc}{J}_1\left(\mathrm{\η}\right)& O_{3\×3}& O_{3\×n}\\ O_{3\×3}& J_2\left(\mathrm{\η}\right)& O_{3\×n}\\ O_{n\×3}& O_{n\×3}& I_n\end{array}\right]\left[\begin{array}{c}{v}_1\\ v_2\\ q\end{array}\right]=J^{\′}\left(\mathrm{\η}\right)\mathrm{\ζ};$

Wherein, represent position and the attitude vectors of in body-fixed coordinate system, testing body; represent linear velocity and the angular velocity vector of in the solid coordinate system of body, testing body; for body consolidates the transformation matrix of coordinate system and body-fixed coordinate system centerline velocities; J ₂(η) be the transformation matrix of angular speed in the solid coordinate system of body and body-fixed coordinate system; Q is the joint rotation angle of mechanical arm;

The assembly kinetics equation set up is as follows,

$M\left(q\right)\stackrel{.}{\mathrm{\ζ}}+C(q,\mathrm{\ζ})\mathrm{\ζ}+D(q,\mathrm{\ζ})\mathrm{\ζ}+g(q,\mathrm{\η})=\mathrm{\τ};$

Wherein,

M (q) ∈ R ^{(6+n) × (6+n)}, be the inertial matrix caused by assembly and additional mass;

C (q, ζ) ∈ R ^{(6+n) × (6+n)}, the centripetal force caused by assembly and additional mass and coriolis force matrix;

D (q, ζ) ∈ R ^{(6+n) × (6+n)}, be hydrodynamic damping matrix;

G (q, η) ∈ R ⁽⁶⁺ⁿ⁾, be gravity and buoyancy vector;

τ ∈ R ⁽⁶⁺ⁿ⁾, be generalized force vector;

R represents real number, can represent real vector or real matrix according to subscript difference;

N represents the free degree of mechanical arm.

3. under a kind of mix suspending microgravity environment according to claim 2 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, in step 3,

Sliding formwork control item form in sliding-mode control is as follows:

τ _smc＝K _Ds′+K _Ssgn(s)；

Wherein,

$s^{\′}=\left[\begin{array}{c}{\tilde{v}}_1\\ {\tilde{v}}_2\\ \stackrel{\stackrel{.}{~}}{q}\end{array}\right]+(\mathrm{\Λ}+K_D^{-1}{K}_P)\left[\begin{array}{c}{R_I^B\widetilde{\mathrm{\η}}}_1\\ \widetilde{\mathrm{\ϵ}}\\ \tilde{q}\end{array}\right]=\widetilde{\mathrm{\ζ}}+(\mathrm{\Λ}+K_D^{-1}{K}_P)\tilde{y},s=\widetilde{\mathrm{\ζ}}+\mathrm{\Λ}\tilde{y},$

S is the sliding-mode surface of design, Λ and K _pfor positive definite diagonal matrix, K _d, K _sfor symmetric positive definite matrix, sgn (s) is sign function; represent the deviation between desired value and actual value, desired value is exactly the desired value of experiment posture, speed and joint position, speed, and actual value is the experiment body movement state information that measuring system obtains, it is the experiment body attitude error represented by hypercomplex number;

According to desired control target ξ _d, ζ _dwith the assembly movement state information ξ that measurement obtains, ζ calculates s and s ', s and s ' is substituted into sliding formwork control item and namely obtains the control force and moment that sliding formwork controls correspondence.

4. under a kind of mix suspending microgravity environment according to claim 3 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, following process is done to sign function sgn (s),

$\mathrm{sgn}\left(s\right)\≈\frac{s}{\left|s\right|+\mathrm{\δ}};$

Wherein, δ is an arithmetic number.

5. under a kind of mix suspending microgravity environment according to claim 3 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, in step 2, pass through radial base neural net the nonlinear dynamical equation of assembly is approached,

ANN Control item is expressed as follows:

${\mathrm{\τ}}_{\mathrm{rbn}}=\hat{f}(\mathrm{\ζ},\mathrm{\ζ},{\mathrm{\ζ}}_r,{\stackrel{.}{\mathrm{\ζ}}}_r)=\widehat{\mathrm{\ω}}\mathrm{\γ}(x,c);$

Wherein, the input vector x of neutral net is by four generalized state vectors of assembly composition, ζ _r=ζ+s, output vector is control and the control moment of assembly, and ω, for exporting weight matrix, will be determined by adaptive updates rule.

6. under a kind of mix suspending microgravity environment according to claim 5 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, in step 4, utilize radial base neural net propeller thrust deviation delta u is estimated, estimate for introducing in controller as feedback quantity to compensate thrust deflexion Δ u.

7. under a kind of mix suspending microgravity environment according to claim 6 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, in step 2 and step 4, according to the change of assembly motion state and configuration, the output weight matrix of online updating neutral net;

The neutral net being used for approaching system dynamics exports weight adaptive updates and restrains as follows,

${\stackrel{\stackrel{.}{^}}{\mathrm{\ω}}}^T={\mathrm{\Γ}}_{\mathrm{\ω}}\mathrm{\γ}(x,c)s^T;$

The neutral net being used for approaching thrust deflexion exports weight adaptive updates and restrains as follows,

${{\stackrel{\stackrel{.}{^}}{\mathrm{\ω}}}_u}^T={\mathrm{\Γ}}_u{\mathrm{\γ}}_u(x,c)s^{T}B;$

Wherein, matrix B is thrust allocation matrix, Γ _ωand Γ _ufor positive definite diagonal matrix, represent the adaptive gain of weight more new law.

8. under a kind of mix suspending microgravity environment according to claim 7 with the autonomous control method of assembly of simple joint mechanical arm, it is characterized in that, in step 4, assembly thrust distribute time allocation matrix as follows,

$B=\left[\begin{array}{cc}L& 0\\ 0& I\end{array}\right]$

Wherein, L is the mapping matrix carrying out thrust distribution; I is the joint torque distribution matrix of mechanical arm, is unit battle array.