Specific embodiment
As shown in Figure 1, permanent-magnet synchronous motor with five degrees of freedom without bearing 11 is by two degrees of freedom bearing-free permanent magnet synchronous motor 111
It is formed with Three Degree Of Freedom AC-DC hybrid magnetic bearing 112, two degrees of freedom bearing-free permanent magnet synchronous motor 111 controls rotor radial
xa, yaDisplacement, Three Degree Of Freedom AC-DC hybrid magnetic bearing 112 control rotor radial xb, ybAnd axial direction zbDisplacement.
As shown in FIG. 2 and 3, it is hanged by bearing-free permanent magnet synchronous motor 11, two degrees of freedom bearing-free permanent magnet synchronous motor
Buoyancy winding current control module 4, Three Degree Of Freedom AC-DC hybrid magnetic bearing current control module 5, displacement sensor 12, photoelectricity
Encoder 13 and angle calculation module 14 form composite controlled object 1.
Suspending power subsystem line neural network is sequentially connected in series between composite controlled object 1 against module 2 and additional control
Device module 3, suspending power subsystem line neural network connect the input of composite controlled object 1, suspending power against the output of module 2
The output and the input of composite controlled object 1 of system line neural network against module 2 are to constant current
The output of composite controlled object 1 is real-time displacement xa(t), ya(t), xb(t), yb(t), zb(t), i.e., four real-time radial displacement xa
(t), ya(t), xb(t), yb(t) and a real-time axial displacement zb(t)。
Additional controller module 3 is by the first sliding mode controller 31 (SMC31), the second sliding mode controller 32 (SMC32), third
Sliding mode controller 33 (SMC33), the 4th sliding mode controller 34 (SMC34) and the 5th sliding mode controller 35 (SMC35) this five
Sliding mode controller composition.The input of additional controller module 3 is given displacementEach cunning therein
The input of mould controller is a given displacementWith corresponding real-time displacement xa(t), ya(t), xb(t),
yb(t), zb(t) difference eax(t), eay(t), ebx(t), eby(t), ebz(t), that output is corresponding Bit andits control amount v1, v2,
v3, v4, v5.Namely: the input of the first sliding mode controller 31 is a given displacementWith real-time displacement xa(t) difference eax
(t), that output is Bit andits control amount v1;The input of second sliding mode controller 32 is given displacementWith real-time displacement ya(t) difference
Value eay(t), that output is Bit andits control amount v2;The input of third sliding mode controller 33 is given displacementWith real-time displacement xb
(t) difference ebx(t), that output is Bit andits control amount v3;The input of 4th sliding mode controller 34 is given displacementWith it is real-time
It is displaced yb(t) difference eby(t), that output is Bit andits control amount v4;The input of 5th sliding mode controller 35 is given displacementWith
Real-time displacement zb(t) difference ebz(t), that output is Bit andits control amount v5。
Suspending power subsystem line neural network against module 2 by nerve network system 22, on-line learning algorithm module 21 with
And 10 integrator S-1Composition, the input of the output Connection Neural Network system 22 of on-line learning algorithm module 21, five sliding formworks
The integrated device S of the output of controller-1The input of Connection Neural Network system 22.The input of on-line learning algorithm module 21 is five
Five Bit andits control amount v of sliding mode controller output1, v2, v3, v4, v5Double integral Respectively with corresponding real-time displacement xa(t), ya(t), xb(t), yb(t), zb(t) error e1(t), e2(t), e3
(t), e4(t), e5(t), the output of on-line learning algorithm module 21 is weight matrix W adjusted0(t+1), weight matrix W0
It (t+1) is one of nerve network system 22 input.Specifically:
Nerve network system 22 shares 16 inputs, the displacement control that each sliding mode controller 31,32,33,34,35 exports
Amount v processed1, v2, v3, v4, v5All input nerve network system 22, the displacement control that each sliding mode controller 31,32,33,34,35 exports
Amount v processed1, v2, v3, v4, v5Through an integrator S-1The multiple integral obtained afterwards all inputs nerve network system 22, each cunning
The Bit andits control amount v that mould controller 31,32,33,34,35 exports1, v2, v3, v4, v5Two integrator S through concatenating-1Afterwards
The double integral arrivedNerve network system 22 is all inputted, along with on-line study is calculated
The weight matrix W of the input nerve network system 22 of method module 210(t+1), therefore totally ten six inputs.On-line learning algorithm module
21 altogether there are five input, wherein first input be the first sliding mode controller 31 output Bit andits control amount v1Through concatenating
One, second integrator S-1The double integral obtained afterwardsWith real-time displacement xa(t) error e1(t), second input be
The Bit andits control amount v of second sliding mode controller 32 output2Third, the 4th integrator S through concatenating-1The double integral obtained afterwardsWith real-time displacement ya(t) error e2(t), third input is the Bit andits control amount v that third sliding mode controller 33 exports3
The the 5th, the 6th integrator S through concatenating-1The double integral obtained afterwardsWith real-time displacement xb(t) error e3(t), the 4th
A input is the Bit andits control amount v of the 4th sliding mode controller 34 output4The the 7th, the 8th integrator S through concatenating-1It obtains afterwards
Double integralWith real-time displacement yb(t) error e4(t), the 5th input is the displacement of the 5th sliding mode controller 35 output
Control amount v5The the 9th, the tenth integrator S through concatenating-1The double integral obtained afterwardsWith real-time displacement zb(t) error e5
(t)。
As shown in figure 3, composite controlled object 1 is shaftless come real-time detection two degrees of freedom using 5 eddy current displacement sensors
The radial displacement x of bearing permanent magnet synchronous electric motor 111a(t), ya(t), the radial displacement x of Three Degree Of Freedom AC-DC hybrid magnetic bearing 112b
(t), yb(t) and axial displacement zb(t)。
Two degrees of freedom bearing-free permanent magnet synchronous motor levitation force winding current control module 4 is adjusted by pi regulator 41, PI
Device 42, IPARK converter 43, SVPWM44, voltage source inverter 45, current sensor 46, CLARK converter 47 and PARK
Converter 48 forms;Current sensor 46 detects the levitation force winding electric current of two degrees of freedom bearing-free permanent magnet synchronous motor 111
iBa, iBb, iBc, the input terminal of the output end connection CLARK converter 47 of current sensor 46, through the generation of CLARK converter 47 α-
Electric current i under β coordinate systemBα, iBβ, the input terminal of the output end connection PARK converter 48 of CLARK converter 47, angle calculation
Module 14 obtains angle, θ, calculation formula according to the rotational speed omega that photoelectric encoder 13 measures are as follows: θ=ω t, PARK converter 48 according to
Electric current i under d-q coordinate system is generated according to the counted θ of angle calculation module 14Bd, iBq, this electric current is that two degrees of freedom bearing-free permanent magnet is same
The feedback current for walking motor levitation force winding is exported against module 2 to constant current with suspending power subsystem line neural network
After obtain difference, difference obtains the given voltage signal under d-q coordinate system after modulating again through pi regulator 41,42The output end of pi regulator 41,42 is connected with the input terminal of IPARK converter 43, the foundation again of IPARK converter 43
The counted θ of angle calculation module 15 generates the voltage under alpha-beta coordinate systemVoltageIt is generated through SVPWM44
The switching signal S of voltage source inverter 45A(A=1,2,3,4,5,6), switching signal S of the voltage source inverter 45 according to offerA
(A=1,2,3,4,5,6) 111 levitation force winding of two degrees of freedom bearing-free permanent magnet synchronous motor is controlled.
Three Degree Of Freedom AC-DC hybrid magnetic bearing current control module 5 by pi regulator 51, pi regulator 52, SVPWM53,
Voltage source inverter 54, current sensor 55 and CLARK converter 56 form.Current sensor 55 detects that Three Degree Of Freedom is handed over
The radial displacement of direct current hybrid magnetic bearing 112 controls electric power ia, ib, ic, the output end connection CLARK transformation of current sensor 55
The input terminal of device 56 generates radial displacement under alpha-beta coordinate system through CLARK converter 56 and controls electric current ix, iy, with suspending power subsystem
Line neural network of uniting is exported against module 2 to constant currentAfter obtain difference, difference is adjusted through pi regulator 51,52 again
The given voltage signal under alpha-beta coordinate system is obtained after systemVoltageVoltage source inverter 54 is generated through SVPWM53
Switching signal SH(H=1,2,3,4,5,6), switching signal S of the voltage source inverter 54 according to offerH(H=1,2,3,4,5,
6) 112 radial displacement electric current of Three Degree Of Freedom AC-DC hybrid magnetic bearing is controlled;Three Degree Of Freedom AC-DC hybrid magnetic bearing
112 axial displacement controls electric current izThe given current signal exported by suspending power subsystem line neural network against module 2Through
Power amplifier 57 obtains.
For composite controlled object 1, five degree of freedom is established to the working principle of five free bearing-free permanent magnet synchronous motors 11
Bearing-free permanent magnet synchronous motor suspending power subsystem mathematical model carries out 11 rotor of permanent-magnet synchronous motor with five degrees of freedom without bearing
Mechanical analysis considers the coupling influence between the gyroscopic effect and each freedom degree of magnetic suspension bearing system, establishes the equation of motion,
And it choosesMake
For the state variable of composite controlled object 1, U=[u1,u2,u3,u4,u5]T=[iBd *,iBq *,ix *,iy *,iz *]TAs composite quilt
Control the input variable of object 1, Y=[y1,y2,y3,y4,y5]T=[xa(t),ya(t),xb(t),yb(t),zb(t)]TAs compound
The output variable of controlled device 1 establishes the state equation of composite controlled object 1, derivation is carried out to output variable Y, until each
The aobvious U containing input variable of a component, obtains opposite order α=(α of composite controlled object 11,α2,α2,α4,α5)=(2,2,2,2,
2) reversibility Analysis, is carried out to composite controlled object 1 and knows that composite controlled object 1 is reversible.
Using random current signal [iBd *,iBq *,ix *,iy *,iz *] motivated, obtain the output of composite controlled object 1
[xa(t),ya(t),xb(t),yb(t),zb(t)], displacement x is acquired using five point value derivative algorithmsa(t), ya(t), xb(t), yb
(t), zb(t) single order, second dervative constitute the input sample collection of neural networkWith desired output sample
Collect [iBd *,iBq *,ix *,iy *,iz *], then data are normalized.
The present invention uses structure for 15 × 32 × 5 BP neural network, and the excitation function of hidden layer neuron is chosen for70% in 5000 groups of samples that sampling is obtained is used as training sample, and remaining 30% is used as test specimens
This.Network is trained using LM learning algorithm, after the training of 1200 steps, error precision reaches 0.001, obtains trained
Nerve network system 22 saves its structure and parameter, with 10 integrators and instruction known to the opposite order of composite controlled object 1
The nerve network system 22 perfected can construct the offline Neural Network Inverse System of composite controlled object 1.
Trained neural network input/output relation can be expressed asWherein u is output vector, and z is
Input variable, the connection weight matrix of input layer to hidden layer are V0, the connection weight matrix of hidden layer to output layer is W0=
[w1,w2,w3,w4,w5]T∈R32×5, in formula, w1, w2, w3, w4, w5Indicate the matrix of 1 row 32 column;T is transposition;R32×5Table
Show any one 32 row, 5 column matrix;wq=[w1q,w2q,…,w11q,w32q], w1q,w2q,…,w11q,w32qFor connection weight, q=
1,2,3,4,5;σ () is general hidden layer excitation function.
Initial time initializes suspending power subsystem line neural network against module 2, and off-line training is obtained
The connection weight matrix W of nerve network system 220And V0Initial weight as on-line study neural network.Based on basic function
Thought, only to the W being affected to neural network approximation properties0It is adjusted.T moment, according to the defeated of each sliding mode controller
The error e of each output valve of integrated value Yu composite controlled object 1 of signal outi(t), i=1,2,3,4,5, wherein
Calculate the connection weight matrix w for obtaining t momentij(t) correction amount wij(t):
In formula, Δ wijIt (t) is connection weight matrix wij(t) correction amount;eiIt (t) is each sliding mode controller output signal
The error of differential value and 1 output valve of composite controlled object;For error ei(t) to connection weight wij(t) local derviation;μj>0
For adjustable parameter;I=1,2,3,4,5;J=1,2 ..., 32.
Set error threshold { ε1,ε2,ε3,ε4,ε5, wherein εiFor lesser constant, i=1,2,3,4,5.When | ei(t)|
< εiWhen, connection weight wij(t) it does not adjust, still there is W0 (t+1)=W0 (t), when | ei(t) | > εiWhen, obtain the company at t+1 moment
Meet weight matrix wij(t+1).Its calculation method is obtained by following formula:
In formula, Δ wijIt (t) is connection weight wij(t) correction amount;eiIt (t) is the differential of each sliding mode controller output signal
The error of value and 1 output valve of composite controlled object;For error ei(t) to connection weight matrix wij(t) local derviation;μj>0
For adjustable parameter;I=1,2,3,4,5;J=1,2 ..., 32, to obtain the connection weight square for updating the t+1 moment adjusted
Battle array W0(t+1)。
The parameter of on-line tuning nerve network system 22, until ei(t)=0, i=1,2,3,4,5.Suspending power subsystem exists
Line nerve network reverse module 2 is connected with compound controlled system 1 may make up the second order puppet of 5 single-input single-outputs as shown in Figure 4
Linear displacement subsystem.
Additional controller module 3 is to make system closed-loop control to the sliding mode controller of pseudo-linear system construction.It is compound controlled
Object 1 obtains 5 pseudo-linear systems, respectively second order radial displacement x after decouplinga, ya, xb, ybSubsystem and second order axial displacement
zbSubsystem.
In order to eliminate the intrinsic buffeting problem of Sliding mode variable structure control, the present invention on the basis of conventional exponentially approaching rule,
It is proposed a kind of novel exponentially approaching rule, expression are as follows:Wherein, s is sliding-mode surface, and C is to be
System and has state variableL >=0, ε > 0, k > 0 are system design parameters.
The steady-state error and rapidity of consideration system, k here are the nonlinear function of Error Absolute Value, and Fig. 5 is to become ginseng
Number k flow chart, if e is the system given value for inputting SMC and the error of real-time detection value, i.e. e is eax(t), eay(t), ebx(t),
eby(t), ebz(t), znFor given fiducial value, there is z1< z2< ... < zn, mnFor the selective value of k after comparison, there is m0< m1< ...
mn.Will | e | with z1Compare, if | e |≤z1, select k=m0, second step is otherwise executed, second step is incited somebody to action | e | with z2Compare, if | e |≤
z2, select k=m1, third step is otherwise executed, and so on, compared k and obtains optimal value mn, value mnI.e. optimal k value.
First sliding mode controller 31 is for second order radial displacement xaSubsystem design, take system state equation expression formula
Are as follows:r1For system state variables and have For state variable r1Derivative and be denoted asChoosing
Take the sliding-mode surface of system are as follows: s1=c1r1+r2, solvec1For sliding-mode surface coefficient,For sliding-mode surface s1Lead
Number, the novel Reaching Law that the first sliding mode controller 31 uses may be expressed as:Then the first sliding formwork control
The output v of device 311It is obtained by following calculation formula:Wherein, l1>=0, ε1> 0, k1>0
It is system design parameters.Construct Lyapunov function:According to Lyapunov Theory of Stability it is found that sliding mode
Accessibility condition are as follows:By can be calculated:It can
Know radial displacement xaSubsystem can reach sliding-mode surface by free position in finite time.
Similarly, the second sliding mode controller 32 is for second order radial displacement yaSubsystem design, take system mode side
Journey expression formula are as follows:r3For system state variables and have For state variable r3Derivative and be denoted asThe sliding-mode surface of selecting system are as follows: s2=c2r3+r4, solvec2For sliding-mode surface coefficient,For sliding formwork
Face s2Derivative, the second sliding mode controller 32 use novel Reaching Law may be expressed as:Then second
The output v of sliding mode controller 322It is obtained by following calculation formula:Wherein, l2>=0,
ε2> 0, k2> 0 is system design parameters.Construct Lyapunov function:According to Lyapunov Theory of Stability it is found that
The accessibility condition of sliding mode are as follows:By can be calculated:Know radial displacement yaSubsystem can be by appointing in finite time
Meaning state reaches sliding-mode surface.
Similarly, third sliding mode controller 33 is for second order radial displacement xbSubsystem design, take system mode side
Journey expression formula are as follows:r5For system state variables and have For state variable r5Derivative and be denoted asThe sliding-mode surface of selecting system are as follows: s3=c3r5+r6, solvec3For sliding-mode surface coefficient,For sliding-mode surface
s3Derivative, third sliding mode controller 33 use novel Reaching Law may be expressed as:Then third is sliding
The output v of mould controller 333It is obtained by following calculation formula:Wherein, l3>=0, ε3
> 0, k3> 0 is system design parameters.Construct Lyapunov function:It is according to Lyapunov Theory of Stability it is found that sliding
The accessibility condition of dynamic model state are as follows:By can be calculated:Know radial displacement xbSubsystem can be by any in finite time
State reaches sliding-mode surface.
Similarly, the 4th sliding mode controller 34 is for second order radial displacement ybSubsystem design, take system mode side
Journey expression formula are as follows:r7For system state variables and have For state variable r7Derivative and be denoted asThe sliding-mode surface of selecting system are as follows: s4=c4r7+r8, solvec4For sliding-mode surface coefficient,For sliding formwork
Face s4Derivative, the 4th sliding mode controller 34 use novel Reaching Law may be expressed as:Then the 4th is sliding
The output v of mould controller 344It is obtained by following calculation formula:Wherein, l4>=0, ε4>
0, k4> 0 is system design parameters.Construct Lyapunov function:According to Lyapunov Theory of Stability it is found that sliding
The accessibility condition of mode are as follows:By can be calculated:
Know radial displacement ybSubsystem can reach sliding-mode surface by free position in finite time.
Similarly, the 5th sliding mode controller 35 is for second order radial displacement zbSubsystem design, take system mode side
Journey expression formula are as follows:r9For system state variables and have For state variable r9Derivative and be denoted asThe sliding-mode surface of selecting system are as follows: s5=c5r9+r10, solvec5For sliding-mode surface coefficient,For sliding formwork
Face s5Derivative, the 5th sliding mode controller 35 use novel Reaching Law may be expressed as:Then the 5th
The output v of sliding mode controller 355It is obtained by following calculation formula:Wherein, r9To be
System and has state variablel5>=0, ε5> 0, k5> 0 is system design parameters.Construct Lyapunov function:According to Lyapunov Theory of Stability it is found that the accessibility condition of sliding mode are as follows:It can by calculating
:Know axial displacement zbSubsystem can in finite time
Sliding-mode surface is reached by free position.