CN102819219A - Intelligent movement control method for prolonging service life of slewing bearing - Google Patents

Abstract

The invention discloses an intelligent movement control method for prolonging service life of a slewing bearing. The intelligent movement control method is characterized by comprising the following steps of: inputting an expected track theta r into an impedance controller (1); measuring interaction force Fe of the tail end of a slewing bearing mechanical structure and an environment through a force sensor (7); inputting a compensation signal of a frictional resisting moment compensator (4) into the impedance controller (1); inputting a reference track theta r(t), theta r(t-1) and theta r(t-2) into a neural network compensator (2); and leading a drive torque signal T obtained by the impedance controller (1) into a torque motor (5), and driving the slewing bearing mechanical structure (6) by the torque motor. According to the intelligent movement control method for prolonging the service life of the slewing bearing, disclosed by the invention, accurate force and position control of the slewing bearing mechanical structure is achieved, therefore, power loss is reduced, the occurrence probability and the development speed of damage are effectively reduced, and the service life is prolonged.

Description

A kind of pivoting support smart motion control method in serviceable life that prolongs

Technical field

The present invention relates to a kind of pivoting support control method, relate in particular to a kind of pivoting support smart motion control method in serviceable life that prolongs.

Background technology

Pivoting support is the basic components that are widely used in need doing in the big machinery structures such as engineering machinery, aerogenerator, ocean platform, military hardware relative gyration; Its size is similar to bearing more again; So be referred to as turntable bearing again, but it has the characteristic that is different from plain bearing: require to bear simultaneously axial force, upsetting moment and radial force, low-speed heave-load; Adopt gear drive, working environment is extremely abominable.Pivoting support is equivalent to the movable joint of equipment, will cause complete machine to lose efficacy in case lose efficacy, even major accident; And because the pivoting support size is big, cost an arm and a leg, do not deposit spare part usually; And could keep in repair after need the thing of tens tons on top even hundreds of ton being promoted certain altitude, cause maintenance difficulty big, expense is high; Stop time is long, and loss is big.Therefore how research improves the pivoting support reliability, and it is significant effectively to increase the service life.

One of which; A lot of researchists work that conducts a research from the aspects such as contact model of circumference load distribution model, rolling body and raceway, expectation grasp pivoting support Life Calculation model, optimal design result; Improve because of fiduciary level is arranged; Thereby reach the purpose that increases the service life, be used for optimal design and set up the perfect Life Calculation model of pivoting support, remain modeling method and model improve aspect further theoretical and a large amount of experimental study.

Moreover some researchists are absorbed in Research on Fault Diagnosis Method, and expectation is in time found and keep in repair to increase the service life before important damage appears in pivoting support through setting up health monitoring systems, avoids the generation of major accident.And method for diagnosing faults is still waiting method research and experimental verification in the abominable low speed parts use of environment, and meaning can not get embodying on particular surroundings is used.

In recent years, there are some of the staff once to attempt adopting intelligent method to make up control system and realize that prolong serviceable life.In oil pumper, adopt the development of neural networks control system like the researchist of Jilin University, extract ability, make the life-span of oil pumper prolong 30% according to oil well actual load control oil pumper; The Hefei intelligence researchist of institute of the Chinese Academy of Sciences attempts artificial intelligence technology is introduced the self shifter control system, studies best schedule, improves the serviceable life of variator.But this respect research also is in the starting stage, and it is simpler to study a question, and still has many theories, technology and problem experimentally to need to solve.

Summary of the invention

The invention provides a kind of pivoting support smart motion control method in serviceable life that prolongs for addressing the above problem.Purpose is the accurate power and the position control of realization pivoting support physical construction, thereby reduces power attenuation, effectively reduces the generation probability of damage and slows down speed of development, increases the service life.

The present invention is for solving above technical matters, and the technical scheme that is adopted is:

A kind of pivoting support smart motion control method in serviceable life that prolongs comprises the steps: step 1:

With desired trajectory θ _rAnd

Input to impedance controller 1, wherein

Be θ _rDifferential, the expression angular velocity;

For

Differential, the expression angular acceleration;

Step 2:

Measure the interaction force F of pivoting support physical construction end and environment through force transducer 7 _e, utilize the position transducer 8 at turntable bearing gear wheel place to measure pivoting support actual displacement angle θ, with F _e, θ imports on-line identification device 3, and the on-line identification device contacts the inertial force M that produces in order to identification physical construction end with environment, ratio of damping P, stiffness coefficient R, satisfied relation is between these parameters:

In the formula:

is inertial force On-line Estimation value;

is ratio of damping On-line Estimation value, and

is stiffness coefficient On-line Estimation value;

Step 3:

Frictional resistance moment compensator 4 compensating signals are inputed to impedance controller 1, have frictional resistance moment because the variation of the outer load that receives of physical construction can make in the pivoting support between ball and raceway, so need the compensation moment of resistance, moment of resistance formula can be expressed as:

$T_f=\mathrm{\μ}\frac{d_m}{2}(\frac{4.4M}{d_m}+2.2\mathrm{Fr}+\mathrm{Fa})$

In the formula: M is the suffered upsetting moment of pivoting support; Fr is the pivoting support force in radial; Fa is the suffered axial force of pivoting support; d _mBe the rolling body diameter; μ is a pivoting support type selecting coefficient;

Step 4:

With reference locus θ _r(t), θ _r(t-1), θ _r(t-2) input to neural networks compensate device 2, wherein θ _r(t-1) be θ _r(t) previous moment value, wherein θ _r(t-2) be θ _r(t-1) previous moment value.Train function V to input to nerve network controller 2 error, wherein V satisfies following formula:

$V=\stackrel{\·\·}{E}+{\hat{M}}^{-1}(\hat{P}\stackrel{\·}{E}+\hat{R}E-F_e)$

In the formula: E is θ _r-θ;

Revise inner parameter through Neural Network Online; Export compensating signal

to impedance controller in each chronomere; The purpose of neural network compensation is that compensation is because the error that the modeling out of true causes; The interference of the outer bound pair of reduction system strengthens the control system robustness simultaneously;

Step 5:

The driving moment signal T that is obtained by impedance controller 1 imports torque motor 5, by supporting of torque motor driving rotational and physical construction 6.

Said step 1 middle impedance controller is by restraining the F gained by impedance Control, and the impedance Control rule is:

In the formula: K1 is the moment of inertia of pivoting support and physical construction, and K2 is a torque motor acting force radius, and c is the pivoting support ratio of damping, T _fBe pivoting support frictional resistance moment, F _eBe the interaction force of pivoting support physical construction end with environment;

U is expressed as in the formula:

$U=\stackrel{\·\·}{\mathrm{\θ}}+{\hat{M}}^{-1}(\hat{P}({\stackrel{\·}{\mathrm{\θ}}}_r-\stackrel{\·}{\mathrm{\θ}})+\hat{R}({\stackrel{\·}{\mathrm{\θ}}}_r-\mathrm{\θ})-F_e)$

In the formula:

is the θ differential, expression pivoting support actual speed.

The on-line identification device adopts young waiter in a wineshop or an inn's preconceived plan method through online fitting F in the said step 2 _e, θ calculates inertial force On-line Estimation value

Ratio of damping On-line Estimation value

Stiffness coefficient On-line Estimation value

Said step 4 neural networks compensate device adopts three layers of BP neural network; This network is made up of input layer, hidden layer, output layer three parts.

Said input layer is input as X _i=[θ _r(t) θ _r(t-1) θ _r(t-2)] ^T

Hidden layer is the corresponding with it W of each input weighting ¹ _Ij, and in each hidden layer node addition, wherein each hidden layer node has a non-linear activation function F (), is called the S type function, its amplitude is between-1 to 1:

$F(.)=\frac{1-\exp (-(.))}{1+\exp (-(.))}$

The hidden layer output valve is:

${\mathrm{\&phi;}}_j=\frac{1-\exp (-(\underset{i-1}{\overset{3}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="1"\; label="X\; i\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700021.png"\; state="\{\{state\}\}">X</figure-callout>}}_i{w^{}}_{}{{}^1}_{\mathrm{ij}}+{b^1}_j))}{1+\exp (-(\underset{i-1}{\overset{3}{\mathrm{\&Sigma;}}}{X}_i{{\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00001762008700011.png,HDA00001762008700021.png"\; state="\{\{state\}\}">w</figure-callout>}}^1}_{\mathrm{ij}}+{b^1}_j))}$

In the formula: W ¹ _IjBe the weighted value between input layer and hidden layer, b ¹ _jBe the bias of j neural unit in the hidden layer, output layer is input as φ _j, can obtain through linear output node like this:

${\mathrm{\&phi;}}_k=\underset{j-1}{\overset{S1}{\mathrm{\&Sigma;}}}({\mathrm{\&phi;}}_j{w^2}_{\mathrm{jk}}+{b^2}_{\text{k}})$

In the formula: W ² _JkBe the weighted value between hidden layer and output layer, b _k ²Be the bias of k neural unit in the output layer, choose the performance function of the quadratic form of training parameter V as training:

$J=\frac{1}{2}{V}^{T}V$

For making performance function J minimize the momentum improvement type that adopts backpropagation, can obtaining upgrading rule be:

${{\mathrm{\&Delta;b}}^1}_j\left(t\right)={\mathrm{\&lambda;\&phi;}}_k(1-{\mathrm{\&phi;}}_k)X_i\left(\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="2"\; label="V\; k\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700032.png"\; state="\{\{state\}\}">V</figure-callout>}}_k{w^{}}_{}{{}^2}_{\mathrm{jk}}\right)+\mathrm{\&gamma;\&Delta;}{b^1}_j(t-1)$

Δb ² _k(t)＝λV _k+γΔb ² _k(t-1)

${{\mathrm{\&Delta;w}}^1}_{\mathrm{ij}}\left(t\right)={\mathrm{\&lambda;\&phi;}}_k(1-{\mathrm{\&phi;}}_k)X_i\left(\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{V}_k{w^2}_{\mathrm{jk}}\right)+\mathrm{\&gamma;\&Delta;}{w^1}_{\mathrm{ij}}(t-1)$

Δw ² _jk(t)＝λV _kφ _k+γΔw ² _jk(t-1)

In the formula: λ is a turnover rate, and γ is the momentum term coefficient.

The invention has the beneficial effects as follows:

A kind of smart motion control method that prolongs pivoting support serviceable life of the present invention has realized the accurate power and the position control of pivoting support physical construction, thereby reduces power attenuation, the generation probability that effectively reduces damage with slow down speed of development, increase the service life.

Wherein the impedance controller of neural network can make the inside and outside environment of pivoting support oneself impression, independently improves controlled running environment, optimizes the displacement state, thereby improves adaptive capacity to environment and robustness, is superior to closed-loop controls such as traditional PID controller, impedance.

A kind of smart motion control method that prolongs pivoting support serviceable life of the present invention is than setting up health monitoring systems or grasping pivoting support Life Calculation model and realize having good practicality more easily.

Description of drawings

Fig. 1 is a pivoting support Machinery Control System general structure schematic diagram of the present invention.

Fig. 2 is an impedance controller cut-away view of the present invention.

Fig. 3 is a neural networks compensate device structural drawing of the present invention.

Fig. 4 is a pivoting support physical construction reduced graph of the present invention.

Fig. 5 is a pivoting support Machinery Control System techniqueflow chart of the present invention.

Fig. 6 is a pivoting support physical construction force tracking synoptic diagram of the present invention.

Fig. 7 is a pivoting support physical construction Position Tracking synoptic diagram of the present invention.

Fig. 8 is a pivoting support physical construction power tracking synoptic diagram of the present invention.

Among the figure: 1 is that impedance controller, 2 is that neural networks compensate device, 3 is that on-line identification device, 4 is that frictional resistance moment compensator, 5 is that torque motor, 6 is that pivoting support physical construction, 7 is that force transducer, 8 is a position transducer.

Embodiment

Below in conjunction with accompanying drawing the present invention is further specified.

Like Fig. 1 to 8, a kind of pivoting support smart motion control method in serviceable life that prolongs comprises the steps:

Step 1:

With desired trajectory θ _rAnd

Input to impedance controller 1, wherein

Be θ _rDifferential, the expression angular velocity;

For

Differential, the expression angular acceleration.

Step 2:

Measure the interaction force F of pivoting support physical construction end and environment through force transducer 7 _e, utilize the position transducer 8 at turntable bearing gear wheel place to measure pivoting support actual displacement angle θ, with F _e, θ imports on-line identification device 3, and identifier contacts the inertial force M that produces in order to identification physical construction end with environment, ratio of damping P, stiffness coefficient R, satisfied relation is between these parameters:

In the formula:

is inertial force On-line Estimation value;

is ratio of damping On-line Estimation value, and

is stiffness coefficient On-line Estimation value.

Step 3:

Frictional resistance moment compensator 4 compensating signals are inputed to impedance controller, have frictional resistance moment because the variation of the outer load that receives of physical construction can make in the pivoting support between ball and raceway, so need the compensation moment of resistance, moment of resistance formula can be expressed as:

$T_f=\mathrm{\&mu;}\frac{d_m}{2}(\frac{4.4M}{d_m}+2.2\mathrm{Fr}+\mathrm{Fa})$

In the formula: M is the suffered upsetting moment of pivoting support; Fr is the pivoting support force in radial; Fa is the suffered axial force of pivoting support; d _mBe the rolling body diameter; μ is a pivoting support type selecting coefficient.

Step 4:

With reference locus θ _r(t), θ _r(t-1), θ _r(t-2) input to neural networks compensate device 2, wherein θ _r(t-1) be θ _r(t) previous moment value, wherein θ _r(t-2) be θ _r(t-1) previous moment value.Train function V to input to nerve network controller 2 error, wherein V satisfies following formula:

$V=\stackrel{\&CenterDot;\&CenterDot;}{E}+{\hat{M}}^{-1}(\hat{P}\stackrel{\&CenterDot;}{E}+\hat{R}E-F_e)$

In the formula: E is θ _r-θ.

Revise inner parameter through Neural Network Online; Export compensating signal

to impedance controller in each chronomere; The purpose of neural network compensation is that compensation is because the error that the modeling out of true causes; The interference of the outer bound pair of reduction system strengthens the control system robustness simultaneously.

Step 5:

The driving moment signal T that is obtained by impedance controller 1 imports torque motor 5, by supporting of torque motor driving rotational and physical construction 6.

Wherein step 1 middle impedance controller is by shown in the accompanying drawing 2, and Fig. 2 is by impedance Control rule F gained, and the impedance Control rule is:

In the formula: K1 is the moment of inertia of pivoting support and physical construction, and K2 is a torque motor acting force radius, and c is the pivoting support ratio of damping, T _fBe pivoting support frictional resistance moment, F _eBe the interaction force of pivoting support physical construction end with environment.

U is expressed as in the formula:

$U=\stackrel{\&CenterDot;\&CenterDot;}{\mathrm{\&theta;}}+{\hat{M}}^{-1}(\hat{P}({\stackrel{\&CenterDot;}{\mathrm{\&theta;}}}_r-\stackrel{\&CenterDot;}{\mathrm{\&theta;}})+\hat{R}({\stackrel{\&CenterDot;}{\mathrm{\&theta;}}}_r-\mathrm{\&theta;})-F_e)$

In the formula:

is the θ differential, expression pivoting support actual speed.

Wherein the on-line identification device adopts young waiter in a wineshop or an inn's preconceived plan method through online fitting F in the step 2 _e, θ calculates inertial force On-line Estimation value

Ratio of damping On-line Estimation value

Stiffness coefficient On-line Estimation value

Wherein step 4 neural networks compensate device is by adopting three layers of BP neural network shown in the accompanying drawing 3.This network is made up of input layer, hidden layer, output layer three parts,

Input layer is input as X _i=[θ _r(t) θ _r(t-1) θ _r(t-2)] ^T

Hidden layer is the corresponding with it W of each input weighting ¹ _Ij, and in each hidden layer node addition, wherein each hidden layer node has a non-linear activation function F (), is called the S type function, its amplitude is between-1 to 1:

$F(.)=\frac{1-\exp (-(.))}{1+\exp (-(.))}$

The hidden layer output valve is:

${\mathrm{\&phi;}}_j=\frac{1-\exp (-(\underset{i-1}{\overset{3}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="1"\; label="X\; i\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700021.png"\; state="\{\{state\}\}">X</figure-callout>}}_i{w^{}}_{}{{}^1}_{\mathrm{ij}}+{b^1}_j))}{1+\exp (-(\underset{i-1}{\overset{3}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="1"\; label="X\; i\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700021.png"\; state="\{\{state\}\}">X</figure-callout>}}_i{w^{}}_{}{{}^1}_{\mathrm{ij}}+{b^1}_j))}$

In the formula: W ¹ _IjBe the weighted value between input layer and hidden layer, b ¹ _jBe the bias of j neural unit in the hidden layer, output layer is input as φ _j, can obtain through linear output node like this:

${\mathrm{\&phi;}}_k=\underset{j-1}{\overset{S1}{\mathrm{\&Sigma;}}}({\mathrm{\&phi;}}_{\mathrm{<figure-callout\; id="2"\; label="j\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700032.png"\; state="\{\{state\}\}">j</figure-callout>}}{w^{}}_{}{{}^2}_{\mathrm{jk}}+{b^2}_{\text{k}})$

In the formula: W ² _JkBe the weighted value between hidden layer and output layer, b _k ²Be the bias of k neural unit in the output layer, choose the performance function of the quadratic form of training parameter V as training:

$J=\frac{1}{2}{V}^{T}V$

For making performance function J minimize the momentum improvement type that adopts backpropagation, can obtaining upgrading rule be:

${{\mathrm{\&Delta;b}}^1}_j\left(t\right)={\mathrm{\&lambda;\&phi;}}_k(1-{\mathrm{\&phi;}}_k)X_i\left(\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="2"\; label="V\; k\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700032.png"\; state="\{\{state\}\}">V</figure-callout>}}_k{w^{}}_{}{{}^2}_{\mathrm{jk}}\right)+\mathrm{\&gamma;\&Delta;}{b^1}_j(t-1)$

Δb ² _k(t)＝λV _k+γΔb ² _k(t-1)

${{\mathrm{\&Delta;<figure-callout\; id="1"\; label="w"\; filenames="HDA00001762008700011.png,HDA00001762008700021.png"\; state="\{\{state\}\}">w</figure-callout>}}^1}_{\mathrm{ij}}\left(t\right)={\mathrm{\&lambda;\&phi;}}_k(1-{\mathrm{\&phi;}}_k)X_i\left(\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{<figure-callout\; id="2"\; label="V\; k\; w"\; filenames="HDA00001762008700011.png,HDA00001762008700032.png"\; state="\{\{state\}\}">V</figure-callout>}}_k{w^{}}_{}{{}^2}_{\mathrm{jk}}\right)+\mathrm{\&gamma;\&Delta;}{{\mathrm{<figure-callout\; id="1"\; label="w"\; filenames="HDA00001762008700011.png,HDA00001762008700021.png"\; state="\{\{state\}\}">w</figure-callout>}}^1}_{\mathrm{ij}}(t-1)$

Δw ² _jk(t)＝λV _kφ _k+γΔw ² _jk(t-1)

In the formula: λ is a turnover rate, and γ is the momentum term coefficient.

Be a kind of whole control scheme flow process that prolongs the smart motion control method in pivoting support serviceable life by shown in Figure 5; According to the further emulation of this scheme flow process by Fig. 6 pivoting support physical construction force tracking synoptic diagram; Fig. 7 pivoting support physical construction Position Tracking synoptic diagram; Fig. 8 pivoting support physical construction power tracking synoptic diagram can be explained and adopt this control method, has reached when making pivoting support power and position and has followed the tracks of, and can know that from the power tracking curve this method has reduced the pivoting support power attenuation; Can effectively reduce the generation probability of damage and slow down speed of development, increase the service life.

Embodiment recited above describes preferred implementation of the present invention; Be not that design of the present invention and scope are limited; Do not breaking away under the design concept prerequisite of the present invention; Common engineering technical personnel make technical scheme of the present invention in this area various modification and improvement all should fall into protection scope of the present invention, and the technology contents that the present invention asks for protection all is documented in claims.

Claims (5)

1. one kind prolongs the pivoting support smart motion control method in serviceable life, it is characterized in that comprising the steps:

Step 1:

With desired trajectory θ _rAnd

Input to impedance controller (1), wherein

Be θ _rDifferential, the expression angular velocity;

For

Differential, the expression angular acceleration;

Step 2:

Measure the interaction force F of pivoting support physical construction end and environment through force transducer (7) _e, utilize the position transducer (8) at turntable bearing gear wheel place to measure pivoting support actual displacement angle θ, with F _e, θ imports on-line identification device (3), and the on-line identification device contacts the inertial force M that produces in order to identification physical construction end with environment, ratio of damping P, stiffness coefficient R, satisfied relation is between these parameters:

In the formula: is inertial force On-line Estimation value; is ratio of damping On-line Estimation value, and

is stiffness coefficient On-line Estimation value;

Step 3:

Frictional resistance moment compensator (4) compensating signal is inputed to impedance controller (1); Because can making in the pivoting support, the variation of the outer load that receives of physical construction has frictional resistance moment between ball and raceway; So need the compensation moment of resistance, moment of resistance formula can be expressed as:

$T_f=\mathrm{\&mu;}\frac{d_m}{2}(\frac{4.4M}{d_m}+2.2\mathrm{Fr}+\mathrm{Fa})$

In the formula: M is the suffered upsetting moment of pivoting support; Fr is the pivoting support force in radial; Fa is the suffered axial force of pivoting support; d _mBe the rolling body diameter; μ is a pivoting support type selecting coefficient;

Step 4:

With reference locus θ _r(t), θ _r(t-1), θ _r(t-2) input to neural networks compensate device (2), wherein θ _r(t-1) be θ _r(t) previous moment value, wherein θ _r(t-2) be θ _r(t-1) previous moment value.Train function V to input to nerve network controller (2) error, wherein V satisfies following formula:

$V=\stackrel{\&CenterDot;\&CenterDot;}{E}+{\hat{M}}^{-1}(\hat{P}\stackrel{\&CenterDot;}{E}+\hat{R}E-F_e)$

In the formula: E is θ _r-θ;

Revise inner parameter through Neural Network Online; Export compensating signal

to impedance controller in each chronomere; The purpose of neural network compensation is that compensation is because the error that the modeling out of true causes; The interference of the outer bound pair of reduction system strengthens the control system robustness simultaneously;

Step 5:

The driving moment signal T that is obtained by impedance controller (1) imports torque motor (5), by supporting of torque motor driving rotational and physical construction (6).

2. the smart motion control method in prolongation pivoting support according to claim 1 serviceable life is characterized in that said step 1 middle impedance controller by by impedance Control rule F gained, and the impedance Control rule is:

In the formula: K1 is the moment of inertia of pivoting support and physical construction, and K2 is a torque motor acting force radius, and c is the pivoting support ratio of damping, T _fBe pivoting support frictional resistance moment, F _eBe the interaction force of pivoting support physical construction end with environment;

U is expressed as in the formula:

$U=\stackrel{\&CenterDot;\&CenterDot;}{\mathrm{\&theta;}}+{\hat{M}}^{-1}(\hat{P}({\stackrel{\&CenterDot;}{\mathrm{\&theta;}}}_r-\stackrel{\&CenterDot;}{\mathrm{\&theta;}})+\hat{R}({\stackrel{\&CenterDot;}{\mathrm{\&theta;}}}_r-\mathrm{\&theta;})-F_e)$

In the formula:

is the θ differential, expression pivoting support actual speed.

3. the smart motion control method in prolongation pivoting support according to claim 1 serviceable life is characterized in that in the said step 2 that the on-line identification device adopts young waiter in a wineshop or an inn's preconceived plan method through online fitting F _e, θ calculates inertial force On-line Estimation value

Ratio of damping On-line Estimation value

Stiffness coefficient On-line Estimation value

4. the smart motion control method in prolongation pivoting support according to claim 1 serviceable life is characterized in that said step 4 neural networks compensate device adopts three layers of BP neural network; This network is made up of input layer, hidden layer, output layer three parts.

5. the smart motion control method in prolongation pivoting support according to claim 4 serviceable life is characterized in that said input layer is input as X _i=[θ _r(t) θ _r(t-1) θ _r(t-2)] ^T

Hidden layer is the corresponding with it W of each input weighting ¹ _Ij, and in each hidden layer node addition, wherein each hidden layer node has a non-linear activation function F (), is called the S type function, its amplitude is between-1 to 1:

$F(.)=\frac{1-\exp (-(.))}{1+\exp (-(.))}$

The hidden layer output valve is:

${\mathrm{\&phi;}}_j=\frac{1-\exp (-(\underset{i-1}{\overset{3}{\mathrm{\&Sigma;}}}{X}_i{w^1}_{\mathrm{ij}}+{b^1}_j))}{1+\exp (-(\underset{i-1}{\overset{3}{\mathrm{\&Sigma;}}}{X}_i{w^1}_{\mathrm{ij}}+{b^1}_j))}$

In the formula: W ¹ _IjBe the weighted value between input layer and hidden layer, b ¹ _jBe the bias of j neural unit in the hidden layer, output layer is input as φ _j, can obtain through linear output node like this:

${\mathrm{\&phi;}}_k=\underset{j-1}{\overset{S1}{\mathrm{\&Sigma;}}}({\mathrm{\&phi;}}_j{w^2}_{\mathrm{jk}}+{b^2}_{\text{k}})$

In the formula: W ² _JkBe the weighted value between hidden layer and output layer, b _k ²Be the bias of k neural unit in the output layer, choose the performance function of the quadratic form of training parameter V as training:

$J=\frac{1}{2}{V}^{T}V$

For making performance function J minimize the momentum improvement type that adopts backpropagation, can obtaining upgrading rule be:

${{\mathrm{\&Delta;b}}^1}_j\left(t\right)={\mathrm{\&lambda;\&phi;}}_k(1-{\mathrm{\&phi;}}_k)X_i\left(\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{V}_k{w^2}_{\mathrm{jk}}\right)+\mathrm{\&gamma;\&Delta;}{b^1}_j(t-1)$

Δb ² _k(t)＝λV _k+γΔb ² _k(t-1)

${{\mathrm{\&Delta;w}}^1}_{\mathrm{ij}}\left(t\right)={\mathrm{\&lambda;\&phi;}}_k(1-{\mathrm{\&phi;}}_k)X_i\left(\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{V}_k{w^2}_{\mathrm{jk}}\right)+\mathrm{\&gamma;\&Delta;}{w^1}_{\mathrm{ij}}(t-1)$

Δw ² _jk(t)＝λV _kφ _k+γΔw ² _jk(t-1)

In the formula: λ is a turnover rate, and γ is the momentum term coefficient.

Images