CN102042893A - Soft-measuring method for tension of band steel between rollers of continuous annealing unit - Google Patents

Abstract

The invention discloses a soft-measuring method for the tension of band steel between rollers of a continuous annealing unit, comprising the following steps of: (1) acquiring process data; (2) preprocessing the process data; (3) constructing a soft-measuring model for combining a cascade observer with a neural network compensator; and (4) correcting the model. The moment current of a frequency converter corresponding to each roller between two tension measuring instruments, the speed and the boundary tension of each roller are used as the input quantity, the tension between every two rollers is calculated by the recursion in a cascade manner by adopting a frequency domain observer arithmetic, thus the indirect calculation and measuring on the tension of the band steel at the position where no tension measuring instrument is installed can be better realized, a favorable means is provided for the stability monitoring of the operation condition of the continuous annealing unit, high-precision calculation on the tension of the band steel at the position where no tension measuring instrument is installed is realized relatively accurately through compensating the nonlinear process characteristic and the unmodeled dynamics by applying a main model and neural network method, and a basis can be provided for the optimization on the control process parameter of tension in the continuous annealing unit.

Description

The flexible measurement method of strip tension between a kind of continuous annealing unit roller

Technical field

The present invention relates to a kind of tension detecting method, is a kind of Applied Computer Techniques specifically, comes the flexible measurement method of alternative hardware with software, particularly a kind of flexible measurement method that is used for strip tension between any roller of continuous annealing unit.

Background technology

Strip tension detection technique in the band steel continuous annealing process, move back the reliable and stable operation of unit for connecting, prevent to be with the steel broken belt, and improve strip speed and belt steel product quality, give full play to the productive capacity of equipment, it all is crucial satisfying various new technologies and modern production control.

Now in the band steel production run, the continuous annealing unit is for keeping the continous-stable operation, the control of employing tension closed loop, as stove district Tension Control, adopt tension link, speed ring tandem control mode to realize, survey Zhang Yi (TM5, TM opens instrument english abbreviation 5 for numbering for surveying, and is as follows) tension force and realize for as shown in Figure 1 No. 5 by 14 rollers of HF furnace roller (numbering HF1R-HF14R) synchronous variable-speed.And, do special consideration when needing to idler roller and furnace roller design for guaranteeing balancing tension, each roller torque equilibrium everywhere, see Fig. 2 a, Fig. 2 b.As can be seen from the figure, different with idler roller, the load balance of stove district Tension Control and each roller thereof is not based on direct Tension Control, promptly each roller torque current signal is not carried out feedback calculation, and the self-regulation of mainly depending on the DROOP (control of hanging down) of each roller, promptly using roller is that the mode of roller speed adjusted in concert is controlled tension force, and DROOP (control of hanging down) regulative mode allows each roller to have roller speed elementary errors, it is improper directly to be provided with at roller, be not complementary with the band steel degree of expanding with heat and contract with cold or exist under the situation such as obvious disturbance, can guarantee to survey an instrument position tension stability though just exist, middle do not have to survey open instrument place tension force and a unusual situation occurs, influence the unit ordinary production even cause pernicious fault such as broken belt.

And the detection method of strip tension is to carry out online detection by open an installation survey Zhang Yi in limited survey at present.All there is tension force in the every place of band steel in the continuous annealing production line, is unpractical but on production line survey Zhang Yi is installed everywhere because of being subjected to factor affecting such as technology, equipment, cost.Therefore the strip tension in the most of zone of continuous annealing production line does not detect, and fault majorities such as broken belt occur in the blind area that these strip tensions detect.Therefore, the method that adopts survey Zhang Yi to carry out the tension force detection can not satisfy the demand of production far away.

At document Ku Chin Lin, Ming-Ching Tsai; Kung-Yi Chen; Web tension control of a start-up process using observer techniques with friction and inertia compensation[J] .Industrial Electronics Society, 2001.IECON ' 01.The 27th Annual Conference of the IEEE.2001, (11.1:529-534. the observer technology that utilization has friction and inertia compensation realizes the start-up course Tension Control) and document Ruiz C., J., Sbarbaro H., D.Estimating the strip tension in multi-drivesystems[J] .Industrial Electronics, Control and Instrumentation, 1997.IECON 97.23rdInternational Conference.1997,11.3:1432-1437. set forth a kind of tension force flexible measurement method in (estimation of multi-drive system strip tension), it all is the flexible measurement method that is proposed for a roller (as uncoiler) at the control purpose, and be purpose, and do not relate to as yet at the tension force flexible measurement method that connects the tension control system that moves back this special multiple roll coupled synchronization speed governing with the operating states of the units monitoring, diagnosing.

In Jap.P. JP61023066, the tension monitoring device that a kind of multiple roll drives is disclosed.This patent is introduced the electromagnetic induction equation of equilibrium of voltage when building object model, but observation during tension force owing to state equation become can see canonical form, only investigate the may observe part, do not use this information.The model that this patent provides is introduced state equation with boundary tension as state variable, and observation model is output as two border tension force and intermediate tensions.To be with the steel bomb property coefficient directly as model parameter in this patent model, and they are different everywhere under the actual conditions, and the motion of band steel and factor such as relevant with temperature, make this parameter (roller characteristic of comparing that often changes, band steel characteristic is more unstable), and be difficult to estimate, thereby introduce model error.In addition, be model parameter directly in the model with roller inertia, do not consider the influence to whole inertia such as motor and band steel, do not consider the influence of bearing friction in the roller equalising torque formula yet.In addition, the tension fluctuation influence that causes owing to the existence of roller off-centre is bigger, does not add to handle direct application and then be difficult to draw set state is judged the result who is of practical significance.

Summary of the invention

The technical problem to be solved in the present invention is: the measuring method that strip tension between a kind of accurate more continuous annealing unit roller is provided, when this measuring method has been avoided the modeling of employing observation model, owing to will be with the steel bomb property coefficient directly as model parameter, and the model error of generation and the strip tension measuring error brought owing to the influence of ignoring bearing friction power.

In order to solve the problems of the technologies described above, the present invention has adopted following technical scheme:

The flexible measurement method of strip tension between a kind of continuous annealing unit roller may further comprise the steps:

The first step, the process data collection

Set up mechanism and roller equalising torque as can be known by tension force, tension force and each roller torque current between roller, roller speed, boundary tension are directly related, so at first selected following system measurable variable is as auxiliary variable: roller speed (v ₁～v _N), torque current (I ₁～I _N), finite point tension force F _I, F _O

So, the variation of the auxiliary variable that measures then can reflect tension variation between roller.So soft measuring principle of the present invention is by collection v, I, F draws tension force between each roller indirectly.

In addition, gather necessary procedure parameter, as moment constant, the whole inertia of roller motor etc., and, gather the specification data of producing the band steel, as waiting until roller footpath R by scene annealing unit roller data for adapting to calculating needs at different size _i, the roller inertia J _i, reduction gear ratio Value is consulted the frequency converter data and is obtained moment coefficient ξ φ _i, power factor η _i, torque current full scale value I _t, check in each section strip steel length l by the PLC program _i, obtain each roller bearing friction T by the trial run data _0iWith the wide w of band steel, each parameter value of thick h;

Second step, the process data pre-service

The process data pre-service comprises data conversion and correction, and filtering roller bearing off-centre causes high fdrequency component;

The 3rd step, the foundation of the soft measurement master cast of strip tension

Basic thought: with mechanism model is that equalising torque is the recurrence relation that strip tension has been set up on the basis, and the compensation that wherein relates to comprises: moment of inertia is got the compensation of band steel inertia, bearing friction compensation, the output torque calculating of the moment of inertia value of roller and band steel integral body in the distribution compensation that the deadweight of balancing band steel, temperature influence tension force, the roller equalising torque;

Set up the strip tension soft-sensing model at the strip tension object, the input of model, output variable are:

Model input variable: each roller roller speed v (v ₁～v _N), each roller electric current I (I ₁～I _N), upstream tension force F _I

Model output variable: tension force between roller

F wherein _I, F _OFor instrument tension force, F are opened in the survey that can survey _IBe boundary condition, F _OBe used for the verification model,

By following system of equations, calculate tension force between each roller again, obtain i roller exit output tension force with tandem mode recursion

${\hat{F}}_i={\hat{F}}_{i^{-1}}^{\′}-(T_{\mathrm{ei}}-v_i\frac{J_{i}s}{R_i})\×\frac{1}{(T_{c}s+1)R_i}---\left(18\right)$

${\hat{F}}_i^{\′}={\hat{F}}_i+\mathrm{\Δ}{F}_i---\left(19\right)$

${\hat{F}}_0^{\′}=F_I---\left(20\right)$

T _ei＝(ξφ _i×I _i-T _0i)×d _i (21)

$d_i=\left\{\begin{array}{cc}{G}_r^i{\mathrm{\&eta;}}^i& \mathrm{\&xi;}{\mathrm{\&phi;}}_i\&times;I_i>T_{0i}\\ G_r^i/{\mathrm{\&eta;}}^i& \mathrm{\&xi;}{\mathrm{\&phi;}}_i\&times;I_i<T_{0i}\end{array}\right.---\left(22\right)$

J _i＝J _iO+J _iS＝J _iM+J _iR+J _iS (23)

i＝1，2，...，N

Wherein, I _i(i=1,2 ..., but N) be the dynamometry square electric current of i roller, ξ φ _iBe frequency converter moment constant, T _EiBe i motor output torque, T _0iBe i roller no-load torque, T _EiBe the output torque of i roller, v _iBe the corresponding motor speed of i roller, R _iBe i roller radius,

Be upstream boundary tension force,

Be i roller exit output tension force,

Be i+1 roller porch tension force, d _iBe the output torque coefficient of i roller, J _iBe the inertia of i roller motor, roller, band steel integral body,

Be the reduction gear ratio of i roller reducing gear, η ⁱBe the output efficiency of i roller frequency converter, observer parameter T _cBig more, error convergence is fast more, but delay is big more, and T _cToo little error convergence is slow excessively, T _cThe influence compromise of the delay estimated to error convergence speed and to tension force according to this parameter of selection select;

The 4th step, the foundation of neural networks compensate model

For the compensating non-linear process characteristic reaches modeling dynamically to the influence of model accuracy, described measuring method adds the neural networks compensate model, and the input of model, output variable are:

Model input variable: each roller torque current I ₁～I _N, roller speed v ₁～v _N

Model output variable: correction error It is added to the soft measurement master cast output of tension force between roller

Obtain

Output as soft-sensing model

The 5th step, model tuning

Model tuning detects the on-the-spot big power condition changing that occurs by reasoning by cases, adopts this moment the process data of the new accumulation of long-term alignment technique utilization to upgrade neural network model, and to obtain higher model accuracy, step is as follows:

STEP1: adopt the reasoning by cases method to calculate the process variable of collection in real time and the similarity of historical data

Whether the size by this similarity surpasses threshold value Sim _ThJudging whether to occur big power condition changing, is then to change more new model of STEP2;

$\mathrm{Sim}(I\left(t\right),{\stackrel{\&OverBar;}{I}}_{\mathrm{old}})=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&epsiv;}}_j(1-\frac{|I_j\left(t\right)-{\stackrel{\&OverBar;}{I}}_{\mathrm{old},j}|}{\max (I_j\left(t\right),{\stackrel{\&OverBar;}{I}}_{\mathrm{old},j})})---\left(24\right)$

Wherein, I (t) is the torque current vector of gathering in real time,

Be the average of torque current historical data, ε _jBe weighting coefficient, satisfy Select according to experience.

STEP2: use the process data of accumulation again to upgrade neural network model;

Neural network input: newly adopt each roller torque current I of data ₁～I _N, roller speed v ₁～v _N

Neural network output: the evaluated error of master cast

Flexible measurement method according to strip tension between continuous annealing unit roller of the present invention, it is characterized in that: the pretreated data conversion of described process data comprises current values conversion and number of speed value transform, the practical measurement of current value is generally represented with number percent, with electric current measured value and full scale value I _tMultiply each other, obtain actual current value I, the general employing rpm expression of speed measured value unit is converted into standard unit's metre per second (m/s), so conveniently data is specifically calculated.

Flexible measurement method according to strip tension between continuous annealing unit roller of the present invention, it is characterized in that: the pretreated adjustment of data of described process data adopts the recurrence average filtering algorithm, by the spectrum analysis of actual measurement tension curve as can be seen, roller bearing off-centre can cause the tension force cyclic fluctuation, under the little situation of the amplitude of this fluctuation, think that itself and broken belt have nothing to do, the purpose of considering soft-sensing model is the early warning broken belt, at first with the speed that detects, electric current, the periodic component that tension signal filtering off-centre causes, and use the data behind the filtering eccentricity component to carry out following tension state observation

$Y\left(n\right)=\frac{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}X(n+i)}{N}$

Wherein, X (m) is a unfiltered data, and establishing data volume is M, m=1, and 2 ..., M, n=1,2 ..., M-N+1, on average counting is N=π R/v/T _s, R is that roller footpath, v are roller speed, sampling period T _s=0.01s.

Flexible measurement method according to strip tension between continuous annealing unit roller of the present invention is characterized in that: following method is adopted in described Tension Distribution compensation:

Use Δ F _iWhen the compensating band steel transmits between roller, the influence of the Tension Distribution that deadweight of band steel and temperature produce

When i is top roll:

$\mathrm{\&Delta;}{F}_i=F_i^g-m_{i}g---\left(25\right)$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i={\hat{F}}_i+F_i^g-m_{i}g---\left(26\right)$

When i is lower roll:

$\mathrm{\&Delta;}{F}_i=F_i^g+m_{i}g---\left(27\right)$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i={\hat{F}}_i+F_i^g+m_{i}g---\left(28\right)$

Wherein,

Be temperature compensation, m _iG is band steel deadweight compensation;

m _i＝ρ×l _i×h×w (29)

$F_i^g={\hat{F}}_i(E_i-E_{i+1})/E_i---\left(30\right)$

$E_i=f\left(T_{\mathrm{fi}}\right)=(208570-{0.20966T}_{\mathrm{fi}}^2)\&times;{10}^6\mathrm{Pa}\left({N/m}^2\right)---\left(31\right)$

T wherein _fBe furnace temperature everywhere, ρ is a gas density, l _iBe roll gap, h is a belt steel thickness, and w is a strip width.

Flexible measurement method according to strip tension between continuous annealing unit roller of the present invention is characterized in that: following method is adopted in described band steel inertia compensation:

Inertia comprises three parts in the model: roller inertia, motor inertia, band steel inertia, and whole inertia calculates according to following formula:

J _i＝J _iO+J _iS＝J _iM+I _iR+J _iS (32)

J _IS-Strip GD ², J _IO-machine GD ², J _IM-motor GD ², J _IR-roll GD ², J _i-total moment of inertia, wherein

ρ-band steel density, A-band steel area, l _i-band steel length, R _i-roller footpath,

-reduction gear ratio, A=w * h.

Flexible measurement method according to strip tension between continuous annealing unit roller of the present invention is characterized in that: following method is adopted in described bearing friction compensation:

The rapid-result nonlinear relationship of bearing friction and roller, and speed is big more, bearing friction is big more, and the relation between no-load current and the rotating speed is by formula (15) expression, do not have the data fitting of gathering in real time when carrying out no load test under the band steel situation by the scene and obtains, as HF 1R T _0iThe unloaded moment that the The Representation Equation friction causes shown in the use formula (16).

T _0i＝f(v _i)(i＝1，2，...，N) (33)

$T_{0i}=\left\{\begin{array}{cc}0.3+\frac{1}{110}{v}_i& 0\&le;v_i<44\\ 0.7+\frac{3}{440}{v}_i& 44\&le;v_i<88\\ 1+\frac{1}{176}{v}_i& 88\&le;v_i<264\\ 2+\frac{1}{880}{v}_i& 264\&le;v_i<440\\ 2.2+\frac{3}{4400}{v}_i& 440\&le;v_i\&le;880\end{array}\right.---\left(34\right)$

Flexible measurement method according to strip tension between continuous annealing unit roller of the present invention is characterized in that: described output torque is calculated the following method that adopts:

Divide three kinds of situations to calculate output torque, the i.e. pure motoring condition of furnace roller motor (ξ φ * I＞T ₀), generating state and accurate motoring condition (ξ φ * I＜T ₀).

$d_i=\left\{\begin{array}{cc}{G}_r^i{\mathrm{\&eta;}}^i& \mathrm{\&xi;}{\mathrm{\&phi;}}_i\&times;I_i>T_0\\ G_r^i/{\mathrm{\&eta;}}^i& \mathrm{\&xi;}{\mathrm{\&phi;}}_i\&times;I_i<T_0\end{array}\right.---\left(35\right)$

Wherein Be the reduction gear ratio of i roller, η ⁱOutput efficiency for i roller frequency converter.

Beneficial effect of the present invention is: move back strip tension flexible measurement method between furnace roller by the company that the present invention describes, can realize preferably not installing surveying and open the strip tension indirect calculation measurement of instrument place, provide good means for connecting the STABILITY MONITORING of moving back operating states of the units, effectively avoided because the broken belt accident that local tension force causes greatly unusually.By using master cast and neural net method compensating non-linear process characteristic and modeling is not dynamic, realize comparatively accurately not installing and survey the high precision computation of opening instrument place strip tension, thereby when having avoided the modeling of employing observation model, owing to will be with the steel bomb property coefficient directly as model parameter, and the model error that produces and because the strip tension measuring error that the influence of ignoring bearing friction power brings, this moves back unit furnace internal tension CONTROL PROCESS Parameter Optimization for company good foundation is provided.

Description of drawings

Fig. 1 is TM5 Tension Control mode figure;

Fig. 2 is load balance figure;

Fig. 3 moves back process tension force object diagram for connecting;

Fig. 4 is soft-sensing model overall frame structure figure;

Fig. 5 is tandem tension state observer master cast figure;

Fig. 6 is single roller tension state observer figure;

Fig. 7 is the neural networks compensate structural drawing;

Fig. 8 is a tension force results estimated proof diagram;

Fig. 9 is for there being tension model estimated result proof diagram under the uncompensated two kinds of situations of compensation;

Figure 10 is the probability density function and the autocorrelation function synoptic diagram of tension model residual error;

The comparison chart of TM6 model value and measured value when Figure 11 is stove district broken belt.

Embodiment

Below for a more detailed description to the present invention in conjunction with the accompanying drawings with embodiment.These embodiment only are the descriptions to best mode for carrying out the invention, scope of the present invention are not had any restriction.

Embodiment 1

Earlier symbol in each accompanying drawing is done a unified explanation:

Symbol description is as follows among Fig. 1:

TM5 is No. 5 tensile measurer MRH-datum velocity ATR-tension link controllers

ACR1～ACR14-electric current loop 1～electric current loop 14

ASR1～ASR14-speed ring controller 1～speed ring controller 14

The roller speed measured value of HF1Rv_Sp～No. 1 roller of HF14Rv_Sp-HF stove～No. 14 roller roller speed of HF stove measured value

The roller speed actual measurement setting value of HF1Rv～No. 1 roller of HF14Rv-HF stove～No. 14 rollers of HF stove

F _1R2R～F _14R15RTension force between the roller of No. 15 rollers of-No. 1 roller No. 2 roller～No. 14 roller;

Symbol description is as follows among Fig. 2 a:

ACR1～ACR5-electric current loop device 1～electric current loop 5

ASR1～ASR5-speed ring controller 1～speed ring controller 5

LOAD BALANCE-load balancing algorithm

The roller speed measured value of roller speed measured value～No. 5 rollers of No. 4 idler rollers of 4BR1Rv_Sp～No. 1 roller of 4BR5Rv_Sp-4 idler roller

The roller speed actual measurement setting value of roller speed setting value～No. 5 rollers of No. 4 idler rollers of 4BR1Rv～No. 1 roller of 4BR5Rv-4 idler roller

F _1R2R～F _14R15RTension force between the roller of No. 5 rollers of tension force～No. 4 roller between the roller of No. 2 rollers of-No. 1 roller

Symbol description is as follows among Fig. 2 b:

SP-ref-speed setting value SP-speed actual value

Iref-current setting value ASR GAIN-gain calculating

ASR-speed ring controller FF-ASR-speed ring feedforward compensation

The DROOP-frequency converter controlled variable that hangs down

Symbol description is as follows among Fig. 3:

F _I-inlet tension force F _O-outlet tension force

I ₁～I _N-No. 1 roller torque current～N roller torque current

v ₁～v _N-No. 1 roller roller speed～N roller roller speed v _O-downstream rollers speed

R ₁～R _N-No. 1 roller～N roller S ₁～S _N-No. 1 roller downstream band steel～N roller downstream band steel

F ₁～F _N-roller 1～roller N exit tension force F ' ₁～F ' _N-1-roller 2～roller N porch tension force

Symbol description is as follows among Fig. 4:

F _IEach roller torque current I of-inlet tension force measured value I- ₁～I _NEach roller torque current v of v- ₁～v _N

F _O-outlet tension force measured value

-outlet tension force estimated value

-outlet tension force master cast calculated value

Tension force estimated value between-each roller E-Model Calculation error

-Model Calculation error estimate

Symbol description is as follows among Fig. 5:

F _I-inlet tension force F _O-outlet tension force

I ₁～I _N-No. 1 roller torque current～N roller torque current

v ₁～v _N-No. 1 roller roller speed～N roller roller speed v _O-downstream rollers speed

R ₁～R _N-No. 1 roller～N roller S ₁～S _N-No. 1 roller downstream band steel～N roller downstream band steel

F ₁～F _N-roller 1～roller N exit tension force

-roller 1～roller N exit tension force is estimated

F ' ₁～F ' _N-1-roller 2～roller N porch tension force

-roller 2～roller N porch tension force is estimated

Symbol description is as follows among Fig. 6:

I _iBut the dynamometry square electric current ξ φ of-i roller _i-frequency converter moment constant T _Ei-i motor output torque

T _0i-i roller no-load torque T _EiThe output torque v of-i roller _iThe corresponding motor speed R of-i roller _i-i roller radius

d _iThe output torque coefficient J of-i roller _iThe inertia of-i roller motor, roller, band steel integral body

-upstream boundary tension force

-Di i roller exit output tension force T _c-observer parameter

-Di i+1 roller porch tension force Δ F _iThe Tension Distribution compensation of-band steel

Symbol description is as follows among Fig. 7:

I ₁～I _N-No. 1 roller torque current～N roller torque current v ₁～v _N-No. 1 roller roller speed～N roller roller speed

The error compensation value of tension force between-roller

Present embodiment is finished on continuous annealing production line, opens on value F (19 s') the production line and implements flexible measurement method having real-time measuring roll speed v, torque current I, limited survey.

The data acquisition of speed, electric current, tension force and pre-service:

Use connects the speed (measurement range: 0～6.25V), electric current (measurement range: 0～4V) signal, the analog quantity (measurement range :-5～5V) tension signals of a survey instrument output of moving back the output of unit data acquisition system (DAS) collection AO plate.Take all factors into consideration the time constant of tension link, speed ring, electric current loop, the sampling period is decided to be 10ms, and promptly every s surveys 100 times, and data are carried out (length of window N=[π R/v/T after standardization and the recurrence average Filtering Processing _s] adjust with the velocity variations self-adaptation, during as v=600/60m/s, R=0.4m, N=12, wherein [] is rounding operation), obtain pretreated roller speed v (v ₁～v _N), torque current I (I ₁～I _N), survey to open the tension force F of instrument place, deposit in the database, the data that adopt when modeling and Model Calculation are all according to handling with upper type.

Set up soft-sensing model then:

From database, pick out the sample data that can cover operating mode substantially, and soft-sensing model is set up in segmentation.As follows with the soft-sensing model explanation modeling procedure that TM10～TM11 section is set up:

The first step: selected modeling section TM10～TM11, read procedure parameter, i.e. model parameter (moment constant, inertia etc.);

A, roller footpath R _i=0.4m (i=1,2 ..., 22)

B, moment coefficient ξ φ _i=5.693N.m (i=1,2 ..., 22)

C, power factor η _i=0.95 (i=1,2 ..., 22)

D, roller inertia J _IO=14.13kgm ²(i=1,2 ..., 22)

E, each section strip steel length $l_i=\left\{\begin{array}{cc}18m& (i=\mathrm{1,2},...\mathrm{19,21,22})\\ 3.6m& (i=20)\end{array}\right.$

F, each roller bearing friction T _0i(i=1,2 ..., 22)

G, reduction gear ratio $G_r^i=1(i=\mathrm{1,2},...,22)$

H, band steel specification: w=0.823m, h=0.193mm

i、N＝22

Second step: the foundation of tension force soft-sensing model master cast;

With model parameter substitution formula (18)～(23), and take all factors into consideration model speed of convergence and the crucial observer parameter T of steady-state error setting _c=0.2, the computing formula that obtains master cast is as follows:

${\hat{F}}_i={\hat{F}}_{i^{-1}}^{\&prime;}-(T_{\mathrm{ei}}-v_i\frac{J_{i}s}{0.4})\&times;\frac{1}{(0.2s+1)0.4}---\left(36\right)$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i---\left(37\right)$

${\hat{F}}_0^{\&prime;}={\mathrm{<figure-callout\; id="1"\; label="F"\; filenames="H2009101975953E0000022.png,H2009101975953E0000032.png"\; state="\{\{state\}\}">F</figure-callout>}}_1,$ F _I＝TM ₁₀ (38)

T _ei＝(5.693×I _i-T _0i)×d _i (39)

$d_i=\left\{\begin{array}{cc}1\&times;0.95& 5.693\&times;I_i\&GreaterEqual;1.7\\ 1/0.95& 5.693\&times;I_i<1.7\end{array}\right.---\left(40\right)$

J _i＝J _iO+J _iS＝14.13+J _iS (41)

I=1,2 ..., thereby 22,

Master cast estimated value for TM11.

Each compensation tache calculation specifications is as follows in the master cast:

A, Tension Distribution compensation

Use Δ F _iWhen the compensating band steel transmits between roller, the influence of the Tension Distribution that deadweight of band steel and temperature produce.

For the band steel of specification w=0.823m, h=0.193mm, consider length l between known roller _i, strip quality is calculated as follows between roller, in addition, and T everywhere in the 2OA stove district _Fi≈ 280, then

$m_i=\mathrm{\&rho;}\&times;l_i\&times;h\&times;w=\left\{\begin{array}{cc}7.8\&times;{10}^3\&times;18\&times;0.823\&times;0.000193=22.3& i=\mathrm{1,2},..,\mathrm{19,21,22}\\ 7.8\&times;{10}^3\&times;3.6\&times;0.823\&times;0.000193=4.5& i=20\end{array}\right.$

Calculate the influence of deadweight of band steel and temperature in the following manner to Tension Distribution:

When roller i (i=2,4 ... 16,18,21) when being top roll:

$\mathrm{\&Delta;}{F}_i=F_i^g-m_{i}g=-22.3\&times;9.8=-218.5N$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i={\hat{F}}_i-218.5$

When roller i (i=1,3 ... 17,19,20,22) when being lower roll:

$\mathrm{\&Delta;}{F}_i=F_i^g+m_{i}g=\left\{\begin{array}{cc}0+22.3\&times;9.8=218.5& i=\mathrm{1,3},...,\mathrm{17,19,22}\\ 0+4.5\&times;9.8=44.1& i=20\end{array}\right.$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i={\hat{F}}_i+F_i^g+m_{i}g=\left\{\begin{array}{cc}{\hat{F}}_i+218.5& i=\mathrm{1,3},...,\mathrm{17,19,22}\\ {\hat{F}}_i+44.1& i=20\end{array}\right.$

B, the compensation of band steel inertia

Inertia comprises three parts in the model: roller inertia, motor inertia, band steel inertia, and whole inertia calculates according to following formula:

$J_{\mathrm{iS}}=\frac{\mathrm{\&rho;}\&times;A\&times;l_i\&times;{\mathrm{<figure-callout\; id="2"\; label="R\; i"\; filenames="H2009101975953E0000011.png,H2009101975953E0000022.png"\; state="\{\{state\}\}">R</figure-callout>}}_i^{}}{4\&times;{\mathrm{<figure-callout\; id="2"\; label="G\; r"\; filenames="H2009101975953E0000011.png,H2009101975953E0000022.png"\; state="\{\{state\}\}">G</figure-callout>}}_r^{}}=\left\{\begin{array}{cc}\frac{\mathrm{\&rho;}\&times;A\&times;l_i\&times;{\mathrm{<figure-callout\; id="2"\; label="R\; i"\; filenames="H2009101975953E0000011.png,H2009101975953E0000022.png"\; state="\{\{state\}\}">R</figure-callout>}}_i^{}}{4\&times;{\mathrm{<figure-callout\; id="2"\; label="G\; r"\; filenames="H2009101975953E0000011.png,H2009101975953E0000022.png"\; state="\{\{state\}\}">G</figure-callout>}}_r^{}}=0.9& i=\mathrm{1,2},...,\mathrm{19,21,22}\\ \frac{\mathrm{\&rho;}\&times;A\&times;l_i\&times;{\mathrm{<figure-callout\; id="2"\; label="R\; i"\; filenames="H2009101975953E0000011.png,H2009101975953E0000022.png"\; state="\{\{state\}\}">R</figure-callout>}}_i^{}}{4\&times;{\mathrm{<figure-callout\; id="2"\; label="G\; r"\; filenames="H2009101975953E0000011.png,H2009101975953E0000022.png"\; state="\{\{state\}\}">G</figure-callout>}}_r^{}}=0.18& i=20\end{array}\right.$

Thereby, $J_i={\mathrm{<figure-callout\; id="0"\; label="J\; i"\; filenames="H2009101975953E0000023.png,H2009101975953E0000041.png"\; state="\{\{state\}\}">J</figure-callout>}}_i+J_{\mathrm{iS}}=\left\{\begin{array}{cc}14.13+0.9=15.03& i=\mathrm{1,2},...,\mathrm{19,21,22}\\ 14.13+0.18=14.31& i=\mathrm{1,2},...,\mathrm{19,21,22}\end{array}\right..$

C, bearing friction compensation

The rapid-result nonlinear relationship of bearing friction and roller, and speed is big more, and bearing friction is big more, and the pass between no-load current and the rotating speed is a nonlinear function, to each roller in the section between the TM10TM11, do not have the data fitting of gathering in real time when carrying out no load test under the band steel situation by the scene and obtain piecewise function.

$T_{0i}=\left\{\begin{array}{cc}0.2+\frac{1}{110}{v}_i& 0\&le;v_i<44\\ 0.4+\frac{2}{440}{v}_i& 44\&le;v_i<88\\ 0.3+\frac{1}{176}{v}_i& 88\&le;v_i<264\\ 1.5+\frac{1}{880}{v}_i& 264\&le;v_i<440\\ 1.8+\frac{2}{4400}{v}_i& 440\&le;v_i\&le;880\end{array}\right.---\left(42\right)$

D, output torque are calculated

D when calculating the motor output torque _iCalculation of parameter is as follows:

$d_i=\left\{\begin{array}{cc}1\&times;0.95& 5.693\&times;I_i\&GreaterEqual;T_{0\mathrm{<figure-callout\; id="1"\; label="i"\; filenames="H2009101975953E0000022.png,H2009101975953E0000032.png"\; state="\{\{state\}\}">i</figure-callout>}}\\ 1/0.95& 5.693\&times;I_i<T_{0i}\end{array}\right.---\left(43\right)$

The 3rd step: the foundation of neural network model:

For the compensating non-linear process characteristic reaches modeling dynamically to the influence of model accuracy, add the neural networks compensate link, see Fig. 7, with each roller torque current I ₁～I _N, roller speed v ₁～v _NBe the input of neural network model, correction error

Be output, training data is the process data I of actual measurement ₁～I _N, v ₁～v _NAnd the master cast error of calculation of boundary tension

Input number of nodes is 2N=44, and the number of hidden nodes elects 80 as, and the output node number is 1, adopts the statistical standard method, and training precision is got ε=0.001, and maximum iteration time 500 obtains the weighting parameter and the threshold parameter of neural network through training.

The 4th step: the final output of calculating tension force soft-sensing model between roller:

Use newly sampled this model tuning error of the above neural network model of setting up

Will

Be added to the soft measurement master cast output of tension force between roller Promptly obtain

As the output of soft-sensing model, see Fig. 4.

${\hat{F}}_o={\hat{F}}_{\mathrm{om}}+\hat{e}---\left(44\right)$

Model testing

With flexible measurement method of the present invention, be used for estimating in real time tension value between roller, and will push away survey open instrument place tension force estimated value and compare with the measured value of surveying Zhang Yi, divide following three kinds of situation verification models:

When a, set speed adjustment, the results are shown in Figure 8 a;

B, (strengthen) when there is disturbance the centre, the results are shown in Figure 8 b as certain roller load;

(being the mode of opening of transferring in the on-the-spot stove) the results are shown in Figure 8 c when c, the adjusting of tension force slope.

And for this paper backoff algorithm validity is described, The model calculation and probability of error density histogram under compensation, the uncompensated two kinds of situations are relatively arranged, see Fig. 9, from Fig. 9 a, Fig. 9 b more as can be seen, after adding the various compensation (Tension Distribution, the compensation of band steel inertia etc.) and neural networks compensate in the master cast, not only estimation does not have partially and amplitude of variation reduces.

The precision of adding up 6 actual measurements and soft measurement comparing result sees Table 1

Table 1 model accuracy

Numbering	1	2	3	4	5	6
							Precision	96.03％	94.94％	95.29％	94.47％	95.45％	95.31％

Above-mentioned precision is defined as: because of obtaining to survey a measured value of instrument place tension force, use the downstream to survey and open the tension force F of instrument place _OVerification model, model error k is defined as follows, and promptly gets the relative error average absolute:

$k=\frac{\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}\frac{|g_i-h_i|}{\left|g_i\right|}}{n}---\left(45\right)$

G is measured data F in the formula _O, h is a soft-sensing model result of calculation

N data number.Precision is defined as 1-k.

Interpretation of result

As can be seen, to record the result very approaching with surveying Zhang Yi for the tension value that calculates with flexible measurement method of the present invention from comparison chart 8 and table 1 as a result.Error between concrete measured value more as can be known and the Model Calculation value is within ± 5%.

Next analyze the probability distribution of the degree of correlation and error, analyze the TM11 measured value and compare the b that sees Fig. 8 with tension force soft-sensing model calculated value result, the two coefficient R=0.9967, the degree of correlation is higher, thus low frequency trend unanimity as a result as can be seen.The tension model error of calculation is asked probability density function, and result such as Fig. 9 near normal distribution, illustrate that model error is mainly noise and causes.

Model uses

By the residual error tracing trouble of definition, see Fig. 5,

F wherein _OOpen instrument place actual measurement tension value for surveying,

Be soft-sensing model calculated value mentioned above.

Use the soft-sensing model of setting up between TM5～TM6 to calculate the estimated value of TM6 to the data of gathering under certain broken belt situation

With TM6 measured value F _OComparison as shown in figure 10.

With variance D (r (t)) the early warning broken belt of residual error, i.e. D (r (t))＞D _Lmax, threshold value D is set by on-site experience _Lmax=10, during t=100, broken belt has taken place in D (r (t))=16.6976.

The present invention is based on equalising torque, accurately detecting under the tension force prerequisite, reject the roller bearing Influence from Eccentric, ignore under wearing and tearing and the prerequisite of vibratory impulse the influence of tension force, main consider the tension fluctuation that causes because of broken belt and abnormal disturbances, creatively consider band steel inertia, under the Tension Distribution prerequisite of band steel temperature influence, obtain tension force between any roller by setting up frequency domain tension state observer, realize not installing surveying and open the strip tension indirect calculation measurement of instrument place, good means are provided for connecting the STABILITY MONITORING of moving back operating states of the units, by using master cast and neural net method compensating non-linear process characteristic and modeling is not dynamic, realize comparatively exactly not installing and survey the high precision computation of opening instrument place strip tension, from checking of the present invention and effect, model accuracy is higher, can satisfy demand of practical production, and it is unusual to can be used for the tension force that detects under the unit failure condition, and production is had actual directive significance.

Claims (7)

1. the flexible measurement method of strip tension between a continuous annealing unit roller may further comprise the steps:

The first step, the process data collection

Gather variable roller speed (v ₁～v _N), torque current (I ₁～I _N), finite point tension force F _I, F _OValue, and obtain roller footpath R by scene annealing unit roller data _i, the roller inertia J _i, reduction gear ratio

Value is consulted the frequency converter data and is obtained moment coefficient ξ φ _i, power factor η _i, torque current full scale value I _t, check in each section strip steel length l by the PLC program _i, obtain each roller bearing friction T by the trial run data _0iWith the wide w of band steel, each parameter value of thick h;

Second step, the process data pre-service

The process data pre-service comprises the data conversion and the adjustment of data, the high fdrequency component that filtering roller bearing off-centre causes;

The 3rd step, the foundation of the soft measurement master cast of strip tension

With mechanism model is that equalising torque is the recurrence relation that strip tension has been set up on the basis, and the compensation that wherein relates to comprises: moment of inertia is got the compensation of band steel inertia, bearing friction compensation, the output torque calculating of the moment of inertia value of roller and band steel integral body in the distribution compensation that the deadweight of balancing band steel, temperature influence tension force, the roller equalising torque;

Set up the strip tension soft-sensing model at the strip tension object, the input of model, output variable are:

Model input variable: each roller roller speed v (v ₁～v _N), each roller electric current I (I ₁～I _N), upstream tension force F _I

Model output variable: tension force between roller

F wherein _I, F _OFor instrument tension force, F are opened in the survey that can survey _IBe boundary condition, F _OBe used for the verification model,

By following system of equations, calculate tension force between each roller again, obtain i roller exit output tension force with tandem mode recursion

${\hat{F}}_i={\hat{F}}_{i^{-1}}^{\&prime;}-(T_{\mathrm{ei}}-v_i\frac{J_{i}s}{R_i})\&times;\frac{1}{(T_{c}s+1)R_i}---\left(1\right)$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i---\left(2\right)$

${\hat{F}}_0^{\&prime;}=F_I---\left(3\right)$

T _ei＝(ξφ _i×I _i-T _0i)×d _i (4)

$d_i=\left\{\begin{array}{cc}{G}_r^i{\mathrm{\&eta;}}^i& \mathrm{\&xi;}{\mathrm{\&phi;}}_i\&times;I_i>T_{0i}\\ G_r^i/{\mathrm{\&eta;}}^i& \mathrm{\&xi;}{\mathrm{\&phi;}}_i\&times;I_i<T_{0i}\end{array}\right.---\left(5\right)$

J _i＝J _iO+J _iS＝J _iM+J _iR+J _iS (6)

i＝1，2，...，N

Wherein, I _i(i=1,2 ..., but N) be the dynamometry square electric current of i roller, ξ φ _iBe frequency converter moment constant, T _EiBe i motor output torque, T _0iBe i roller no-load torque, T _EiBe the output torque of i roller, v _iBe the corresponding motor speed of i roller, R _iBe i roller radius, Be upstream boundary tension force,

Be i roller exit output tension force, Be i+1 roller porch tension force, d _iBe the output torque coefficient of i roller, J _iBe the inertia of i roller motor, roller, band steel integral body,

Be the reduction gear ratio of i roller reducing gear, η ⁱBe the output efficiency of i roller frequency converter, observer parameter T _c=0.2, T _cBig more, error convergence is fast more, but delay is big more, and T _cToo little error convergence is slow excessively, T _cThe influence compromise of the delay estimated to error convergence speed and to tension force according to this parameter of selection select;

The 4th step, the foundation of neural networks compensate model

For the compensating non-linear process characteristic reaches modeling dynamically to the influence of model accuracy, described measuring method adds the neural networks compensate model, and the input of model, output variable are:

Model input variable: each roller torque current I ₁～I _N, roller speed v ₁～v _N

Model output variable: correction error

It is added to the soft measurement master cast output of tension force between roller Obtain

Output as soft-sensing model

The 5th step, model tuning

Model tuning is that the similarity by reasoning by cases judges on-the-spot when big power condition changing occurring, adopt long-term alignment technique, utilize the process data of new accumulation to upgrade neural network model, to reduce unit operation a period of time, obtain more high precision because the model parameter that factors such as unit equipment wearing and tearing cause changes the harmful effect that model accuracy is caused.

2. the flexible measurement method of strip tension between continuous annealing unit roller according to claim 1, it is characterized in that: the pretreated data conversion of described process data comprises current values conversion and number of speed value transform, the practical measurement of current value is generally represented with percentage, with electric current measured value and full scale value I _tMultiply each other, obtain actual current value I, speed measured value unit is general to adopt rpm, is converted into standard unit's metre per second (m/s).

3. the flexible measurement method of strip tension between continuous annealing unit roller according to claim 1 is characterized in that: the pretreated adjustment of data of described process data adopts the recurrence average filtering algorithm:

$Y\left(n\right)=\frac{\underset{i=0}{\overset{N-1}{\mathrm{\&Sigma;}}}X(n+i)}{N}$

Wherein, X (m) is a unfiltered data, and establishing data volume is M, m=1, and 2 ..., M, n=1,2 ..., M-N+1, on average counting is N=π R/v/T _s, R is that roller footpath, v are roller speed, sampling period T _s=0.01s.

4. the flexible measurement method of strip tension between continuous annealing unit roller according to claim 1 is characterized in that: following method is adopted in described Tension Distribution compensation:

Use Δ F _iWhen the compensating band steel transmits between roller, the influence of the Tension Distribution that deadweight of band steel and temperature produce

When i is top roll:

$\mathrm{\&Delta;}{F}_i=F_i^g-m_{i}g---\left(7\right)$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i={\hat{F}}_i+F_i^g-m_{i}g---\left(8\right)$

When i is lower roll:

$\mathrm{\&Delta;}{F}_i=F_i^g+m_{i}g---\left(9\right)$

${\hat{F}}_i^{\&prime;}={\hat{F}}_i+\mathrm{\&Delta;}{F}_i={\hat{F}}_i+F_i^g+m_{i}g---\left(10\right)$

Wherein,

Be temperature compensation, m _iG is band steel deadweight compensation;

m _i＝ρ×l×h×w (11)

$F_i^g={\hat{F}}_i(E_i-E_{i+1})/E_i---\left(12\right)$

$E_i=f\left(T_{\mathrm{fi}}\right)=(208570-{0.20966T}_{\mathrm{fi}}^2)\&times;{10}^6\mathrm{Pa}\left({N/m}^2\right)---\left(13\right)$

T wherein _FiBe furnace temperature everywhere, ρ is a gas density, and l is a roll gap, and h is a belt steel thickness, and w is a strip width.

5. the flexible measurement method of strip tension between continuous annealing unit roller according to claim 1 is characterized in that: following method is adopted in described band steel inertia compensation:

Inertia comprises three parts in the model: roller inertia, motor inertia, band steel inertia, and whole inertia calculates according to following formula:

J _i＝J _iO+J _iS＝J _iM+J _iR+J _iS (14)

J _IS-Strip GD ², J _IO-machine GD ², J _IM-motor GD ², J _IR-roll GD ², J _i-total moment of inertia, wherein

ρ-band steel density, A-band steel area, l _i-band steel length, R _i-roller footpath,

Reduction gear ratio, A=w * h.

6. the flexible measurement method of strip tension between continuous annealing unit roller according to claim 1 is characterized in that: following method is adopted in described bearing friction compensation:

The rapid-result nonlinear relationship of bearing friction and roller, and speed is big more, bearing friction is big more, and the relation between no-load current and the rotating speed is by formula (15) expression, do not have the data fitting of gathering in real time when carrying out no load test under the band steel situation by the scene and obtains, as HF 1R T _0iThe unloaded moment that the The Representation Equation friction causes shown in the use formula (16).

T _0i＝f(v _i)(i＝1，2，...，N) (15)

$T_{0i}=\left\{\begin{array}{cc}0.3+\frac{1}{110}{v}_i& 0\&le;v_i<44\\ 0.7+\frac{3}{440}{v}_i& 44\&le;v_i<88\\ 1+\frac{1}{176}{v}_i& 88\&le;v_i<264\\ 2+\frac{1}{880}{v}_i& 264\&le;v_i<440\\ 2.2+\frac{3}{4400}{v}_i& 440\&le;v_i\&le;880\end{array}\right.---\left(16\right)$

7. the flexible measurement method of strip tension between continuous annealing unit roller according to claim 1 is characterized in that: described output torque is calculated the following method that adopts:

Divide three kinds of situations to calculate output torque, the i.e. pure motoring condition of furnace roller motor (ξ φ * I＞T ₀), generating state and accurate motoring condition (ξ φ * I＜T ₀).

$d_i=\left\{\begin{array}{cc}{G}_r^i{\mathrm{\&eta;}}^i& \mathrm{\&xi;}\mathrm{\&phi;}\&times;I>T_0\\ G_r^i/{\mathrm{\&eta;}}^i& \mathrm{\&xi;}\mathrm{\&phi;}\&times;I<T_0\end{array}\right.---\left(17\right)$

Wherein

Be the reduction gear ratio of i roller, η ⁱOutput efficiency for i roller frequency converter.

Images