CN104880947B - A kind of multichannel arrowband control algolithm of machinery active vibration isolation - Google Patents

Abstract

The present invention relates to the multichannel arrowband control algolithm of a kind of machinery active vibration isolation, reference signal is picked up with vibrating sensor near vibration source, time domain line spectrum reference signal is extracted with the narrow band filter of the most ready-portioned each frequency range, obtain the filtered reference signal of secondary channel inversion model for Self Adaptive Control by multichannel 2 rank filtering operation, guarantee stability simultaneously, convergence rate that can be faster than FxLMS algorithm；The phase difference compensation that series connection one can adjust with frequency fluctuation online adaptive containing only 2 rank parameters, parameter after narrow band filter, original reference signals before making the line spectrum reference signal after extraction and extracting is without phase contrast, thus the controller optimal solution fluctuation suppressing line spectral frequencies fluctuation to cause, improve the control algolithm robustness to frequency fluctuation.Fast convergence rate of the present invention, good to frequency fluctuation robustness, control effective under mechanical vibration amplitude, frequency fluctuation operating mode.

Description

A kind of multichannel arrowband control algolithm of machinery active vibration isolation

Technical field

The present invention relates to the multichannel arrowband control algolithm of a kind of machinery active vibration isolation.

Background technology

Mechanical vibration contain abundant and prominent line spectrum composition, and these line spectrum energy are concentrated, amplitude is far above surrounding frequencies Broadband noise, and each line spectrum amplitude, frequency all can fluctuate along with machine operation operating mode.

The line spectrum pair Ship Radiated-Noise of ship machinery has a major impact.Plant equipment active Vibration Isolation presses down as one The effective means of line spectrum processed and be increasingly subject to pay attention to, control effect to realize more preferable line spectrum, machinery active vibration isolation active Control algolithm needs to break through following difficult point: (1) has many line spectrum vibration needs to be controlled, and uses traditional wideband adaptive to calculate Method can only effectively control the strongest 1～2 frequency vibration；(2) plant equipment is supported by multiple vibration isolators, and structure is the most relatively Complexity, secondary channel matrix exgenvalue dispersion is big, causes convergence stability and speed to be difficult to take into account；(3) the shaking of plant equipment Width and frequency are all likely to occur fluctuation, and control algolithm should have very fast convergence rate and have preferable robustness to frequency fluctuation, Good line spectrum could be obtained and control effect.

The narrowband self-adaption algorithm announced at present can obtain good for many line spectrum vibration (noise) controls of difficult point (1) Effect.Narrowband self-adaption algorithm is divided into Time-Domain algorithm, frequency domain algorithm and time-domain and frequency-domain combination algorithm, but current narrow-long slot parts is equal Difficult point (2), (3) cannot be solved simultaneously.

" He Lin, Li Yan, Yang Jun. magnetic suspension-air bag master passive hybrid vibration isolation device is theoretical and tests [J]. acoustic journal, 2013,38 (2): 241-249. " and " climb. the Active Vibration Control technical research [D] of reciprocating machine peculiar to vessel, Harbin work Cheng great Xue Master's thesis, 2013. ", have studied time-domain narrowband algorithm, by narrow band filter extract time domain reference signal and Line spectrum in time domain error signal, the filtering of secondary channel convolution is reduced to ten multistage, and each controller independent parallel controls many Root line spectrum, compared with the algorithm of broadband, amount of calculation decreases, and control effect and convergence rate to many line spectrums increase, but And unresolved secondary channel matrix exgenvalue dispersion is big and be difficult to take into account convergence stability and the problem of speed；And arrowband band The phase-frequency characteristic of bandpass filter is the most precipitous, and therefore in frequency fluctuation operating mode, the fluctuation of this algorithmic controller optimal solution is violent, robust Property is poor.

Also have a kind of narrow-long slot parts that frequency mismatch is had robustness, reference signal Risk factor arrange frequency because of Son, uses class LMS algorithm that this factor is carried out self-adaptative adjustment, makes reference signal frequency move closer to actual frequency, but this is frequently The self-adaptative adjustment of the rate factor needs to use the error signal of whole multi-channel control system, secondary channel model and controller system Number, operand is big and has strong coupling with control algolithm；If simulation result shows that actual frequency, for static state, adjust reference Signal frequency also needs iteration thousands of times close to this frequency, does not provides the tracking effect for frequency fluctuation.

Patent CN102629469A discloses a kind of time-frequency domain mixed self-adapting Active noise control using algorithm, this algorithm according to The time delay feature of noise transmission, is divided into two pieces by whole control wave filter, and one piece completes filtering to ensure control in time domain The generation link of signal is without time delay, and another block uses frequency domain filtering, and two blocking filters more new capital completes on frequency domain.But this is special The time domain controller of profit is not directed to how to extract time domain line spectrum from time domain reference signal and error signal, is not directed to frequency fluctuation Operating mode robustness.

From the above it can be seen that above-mentioned arrowband control algolithm is mainly for amplitude and the operating mode of frequency stable, but in plant equipment Under line spectrum amplitude and frequency fluctuation operating mode, in addition it is also necessary to research convergence rate, real-time, the more preferable algorithm of robustness, to guarantee this Many line spectrums vibration control effect under operating mode.

Summary of the invention

It is an object of the invention to the active isolation for the vibration of mechanical many line spectrums, it is desirable to provide a kind of at secondary channel matrix In the case of eigenvalue dispersion is big, still can take into account stability and convergence rate, and can start building at mechanical vibration amplitude, frequency wave Under condition, there is the multichannel arrowband control algolithm of the mechanical active vibration isolation of good robustness and control effect.

The implementation of the object of the invention is, the multichannel arrowband control algolithm of a kind of machinery active vibration isolation, concrete steps As follows:

1) near vibration source, reference signal v is picked up with vibrating sensor^oN (), by full range original reference signals v^oN () inputs Transfer function isHilbert transformer, it is thus achieved that signalBy v^oN () input M point postpones Device, it is thus achieved that signal x^o(n)；Owing to Hilbert transform has all-pass characteristic and full frequency band phase shift-pi/2 characteristic, therefore in formulaFor x^oN () signal after full frequency band phase shift-pi/2, subscript Q represents phase shift-pi/2；

2) according to the most ready-portioned frequency range, with the narrow band filter B of each frequency range^mZ () is to full range reference signal x^o (n) and full range orthogonal reference signalIt is filtered, it is thus achieved that be in the line spectrum signal u of each frequency range^m(n) andBy u^m(n) WithInput a phase difference compensation containing only 2 rank parameters, it is thus achieved that line spectrum reference signal x of each frequency range^mN () and line spectrum are just Hand over reference signal2 rank parameters of this phase difference compensation can adjust with frequency fluctuation online adaptive, to guarantee x^m(n) WithIt is respectively relative to the signal x before extracting^o(n) andAlmost without phase contrast；Subscript m represents that frequency range is numbered；

3) setting control system hardware and have L actuator and L error pick-up, each actuator is right in control algolithm Ying Yilu adaptive controller, in each frequency range, obtains adaptive controller output signal according to equation belowL= 1 ..., L:

$y_l^m\left(n\right)=\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{w}_{\mathrm{li}}^m\left(n\right)x^m(n-i),l=1,...,L---\left(8\right),$

In formulaI=0 ..., N-1 is the time-domain coefficients of the l road adaptive controller of m frequency range, with self adaptation Controller transfer functionCorresponding；

In formula, N is the exponent number of adaptive controller；

4) by each frequency range control signalL=1 ..., obtain control signal after L superpositionL=1 ..., L； Control signalL=1 ..., L is through secondary channel $S_{\mathrm{lk}}\left(z\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}$ Arrive kth=1 ..., L error vibrations sensing At device, and kth=1 ..., the desired signal at L error vibrations sensorK=1 ..., after L acts on jointly, To error signalK=1 ..., L；

S in formula_klZ () represents that l road control signal is via the secondary channel between l road actuator to kth road error signal Transfer function；

5) with the narrow band filter B of each frequency range^mZ () is to error signalK=1 ..., L is filtered, and extracts Go out the line spectrum error signal of each frequency rangeK=1 ..., L；

6) in each frequency range, B is used^mZ () is to line spectrum reference signal x^m(n) and line spectrum orthogonal reference signalCarry out once again Narrow-band filtering, obtains signalWithThen secondary channel is carried out in each frequency range by multichannel 2 rank filtering operation inverse Model filtering, obtains the filtered reference signal of secondary channel inversion model of each frequency range $r_{\mathrm{lk}}^m\left(n\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array};$

7) in each frequency range, according to the equation below time-domain coefficients to each road adaptive controller $w_{\mathrm{li}}^m\left(n\right)$ $\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}$ Enter Row sum-equal matrix:

$w_{\mathrm{li}}^m(n+1)=w_{\mathrm{li}}^m\left(n\right)-\widetilde{\mathrm{\μ}}\underset{k=1}{\overset{L}{\mathrm{\Σ}}}{e}_k^m\left(n\right)r_{\mathrm{lk}}^m(n-i),\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}---\left(9\right),$

WhereinFor restraining step-length,Each frequency rangeCan be independently arranged；

Repeat step 1)～7), make the object function of each frequency rangeIt is gradually reduced, thus realizes many line spectrums Active Vibration Control.

The present invention improves on the basis of the disclosed frequency domain Newton algorithms such as Elliott: extract with narrow band filter Line spectrum in time domain reference signal, and series connection one can be with frequency fluctuation containing only 2 rank parameters, parameter after narrow band filter The phase difference compensation that online adaptive adjusts, the original reference signals before making the line spectrum reference signal after extraction and extracting is without phase Potential difference, thus the controller optimal solution fluctuation suppressing line spectral frequencies fluctuation to cause, significantly improve control algolithm to frequency fluctuation Robustness；Time domain line spectrum reference signal is carried out secondary channel inversion model filtering, as a example by L actuator and L sensor, logical Cross L × L 2 rank filtering operations, it is thus achieved that the filtered reference signal of secondary channel inversion model, be used for carrying out Self Adaptive Control, While guaranteeing stability, more faster convergence rate than FxLMS algorithm (especially dispersion can be obtained in secondary channel matrix exgenvalue In the case of degree is big)；This algorithm the convergence speed is fast, good to frequency fluctuation robustness, can start building at mechanical vibration amplitude, frequency wave Obtain under condition and well control effect.

The invention has the beneficial effects as follows:

1, time domain line spectrum reference signal is carried out secondary channel inversion model filtering, by filtered for secondary channel inversion model ginseng Examine signal for Self Adaptive Control, while guaranteeing stability, can obtain more faster convergence rate than FxLMS algorithm, to width The vibration of value and frequency fluctuation has and preferably controls effect；

2, the filtering of above-mentioned secondary channel inversion model only need to process time-domain signal, it is not necessary to carries out time-frequency conversion, and to time domain Signal carries out pointwise process, and real-time is the best；

3, the secondary channel inversion model filtering of above-mentioned each frequency range, only needs L × L 2 rank filtering to realize, and operand is very Little；

4, above-mentioned secondary channel inversion model is L × L rank complex matrixs, uses off-line method identification and inverts, being not take up Line operand；Another frequency range divides and after Model Distinguish, can quickly check whether control algolithm meets the essential condition of convergence；

5, after narrow band filter, series connection one can be adjusted with frequency fluctuation online adaptive containing only 2 rank parameters, parameter Whole phase difference compensation, the original reference signals before making the line spectrum reference signal after extraction and extracting is without phase contrast, thus presses down Controller optimal solution fluctuation during line spectral frequencies processed fluctuation, can significantly improve the control algolithm robustness to frequency fluctuation；

6, the phase difference compensation of above-mentioned each frequency range is containing only 2 rank parameters, and operand is little；This phase difference compensation only needs to process Time-domain signal, it is not necessary to carry out time-frequency conversion, real-time is the best；

7, the parameter adaptive adjustment link of this phase difference compensation, need not call control system secondary channel characteristic and Controller coefficient, do not affected by isolation mounting physical characteristic, coupling weak, therefore fast convergence rate, can be fast along with frequency fluctuation The frequency response of the whole phase difference compensation of velocity modulation；

8, this phase difference compensation compensates only for narrow band filter frequency response at score spectral frequency, does not interferes with resistance Band attenuating；

9, this phase difference compensation will not be to control system amount of imports external signal, thus without introducing additional noise；

10, each frequency range only needs a phase difference compensation, for compensating phase contrast when extracting reference signal；Without Multichannel error signal to carry out phase difference compensation, and therefore shared in multi-channel system operand is the least；

Even if the present invention is in the case of secondary channel frequency response matrix eigenvalue dispersion is big, guaranteeing the same of stability Time, still can obtain very fast convergence rate；Controller optimal solution fluctuation when line spectral frequencies can be suppressed to fluctuate, significantly improves control The algorithm robustness to frequency fluctuation, can obtain under mechanical vibration amplitude, frequency fluctuation operating mode and well control effect.

Accompanying drawing explanation

Fig. 1 is the theory diagram of inventive algorithm,

Fig. 2 a is in inventive algorithm, the theory diagram of phase difference compensation,

Fig. 2 b is in inventive algorithm, the theory diagram of secondary channel inversion model wave filter,

Fig. 3 is the frequency domain equivalent schematic diagram of inventive algorithm,

Fig. 4 is the present invention a kind of configuration figure of applied environment when being embodied as,

Fig. 5 a is the inventive algorithm G in each frequency range_LeftEigenvalue real partActual measurement result figure,

Fig. 5 b is at the G of each frequency range for the FxLMS algorithm of contrast_LeftEigenvalue real partActual measurement Result figure,

Fig. 6 a, b are respectively diesel engine speed 1200r/min, and inventive algorithm controls the vibration of 109.5Hz, 139Hz line spectrum Error signal power convergence curve,

Error when Fig. 7 a, b are respectively diesel engine speed 1100r/min, 1200r/min, before and after inventive algorithm convergence Power spectrum density figure,

When Fig. 8 a, b, c respectively control 120Hz, 180Hz, 195Hz vibration, inventive algorithm and the mistake of FxLMS algorithm Difference signal power convergence curve comparison,

Fig. 9 a, b are respectively when controlling sweep velocity 0.1Hz/s, the vibration of sweep velocity 0.4Hz/s, inventive algorithm (have/ Without phase difference compensation) error signal power convergence curve contrast.

Detailed description of the invention

The present invention picks up reference signal with vibrating sensor near vibration source, and full frequency band is divided into before controlling a fixed width Narrowband slice, uses offline mode to take its mid frequency identification secondary channel model L × L rank plural number square formation in each frequency rangeSo Rear off-line is inverted, and obtains the secondary channel inversion model of each frequency rangeL × L represents L actuator and L sensor.

Described division frequency range and take mid frequency in each frequency range and come identification secondary channel model, the reality in each frequency range During frequency of vibration off-center frequency, the actual frequency response of secondary channel can change thus and there is error between model, but logical Cross classifying rationally frequency range, can make that error is less, make control algolithm remain to meet the essential condition of convergence.Whether inspection frequency range divides Rationally, the method for secondary channel model the most satisfied convergence essential condition is as follows: at the low-limit frequency of certain frequency range (left hand edge) Identification secondary channel model(right hand edge) identification secondary channel model at the highest frequency of this frequency rangeAt mid frequency Place's identification secondary channel modelCalculateWithAnd calculateEigenvalueWithEigenvalueIf each eigenvalue real part is satisfied byWithThen illustrated divisions frequency range is reasonable, secondary channel modelWithError is little, by center frequency The secondary channel model of rateIt is rational for control algolithm, when actual line spectral frequencies is positioned at the left hand edge of each frequency range, the right During Yuan He center, all can meet the essential condition of control algolithm convergence.Above-mentioned identification and the method for inspection, automatic operation able to programme, The rapidest.When machinery active vibration isolation target for control 20～200Hz line spectrums vibration, and frequency range divide sufficiently fine (about 5～ Time 10Hz), the convergence at each frequency range left hand edge, right hand edge and center generally can represent the convergence of this frequency range other frequencies interior.

The present invention uses the Hilbert transform of all-pass characteristicAnd M point delayer is former to full range Beginning reference signal v^oN () is filtered, it is thus achieved that full range reference signal x^o(n) and full range orthogonal reference signalWherein Relative to x^o(n) phase shift-pi/2.

With reference to Fig. 1, Fig. 2 a, with the narrow band filter B of the most ready-portioned each frequency range^mZ () is to x^o(n) andEnter Row filtering, it is thus achieved that be in the line spectrum signal u of each frequency range^m(n) andThen by u^m(n) andInput one containing only 2 rank ginsengs The phase difference compensation of number.

The phase difference compensation of described each frequency range, as shown in Figure 2 a, its 2 rank parameter is respectively its signal flow block diagramWithThis 2 rank parameter can be adjusted by the online adaptive link relatively independent with control system, then multiple Make in control system, to signal u^m(n) andIt is filtered processing, narrow band filter under real-Time Compensation frequency fluctuation Phase contrast；Thus obtain and be respectively relative to extract front x^o(n) andLine spectrum reference signal x almost without phase contrast^m(n) and Line spectrum orthogonal reference signal

The phase difference compensation of described each frequency range compensates the step of narrow band filter phase contrast:

1) B is used^mZ () is to line spectrum signal u^m(n) andIt is filtered, obtains line spectrum signalWith $u_{\mathrm{QB}}^m\left(n\right)=B^m\left(z\right)u_Q^m\left(n\right);$

2) parameter is usedWithRightWithIt is filtered obtaining line spectrum signalAs follows:

$u_{\mathrm{BC}}^m\left(n\right)=c_R^m\left(n\right)u_B^m\left(n\right)-c_I^m\left(n\right)u_{\mathrm{QB}}^m\left(n\right)---\left(10\right)$

3) calculateWith u^mError between (n)

$e_u^m\left(n\right)=u^m\left(n\right)-u_{\mathrm{BC}}^m\left(n\right)---\left(11\right)$

4) parameter is adjusted according to following algorithm online adaptiveWith

$\begin{array}{c}{c}_R^m(n+1)=c_R^m\left(n\right)+\widetilde{\mathrm{\ϵ}}{u}_B^m\left(n\right)e_u^m\left(n\right)\\ c_I^m(n+1)=c_I^m\left(n\right)-\widetilde{\mathrm{\ϵ}}{u}_{\mathrm{QB}}^m\left(n\right)e_u^m\left(n\right)\end{array}---\left(12\right)$

It is whereinConvergence step-length,Each frequency rangeCan be independently arranged.

5) whenWithAdjusted convergence by online adaptive, have $e_u^m\left(n\right)\→0,$ I.e. $u_{\mathrm{BC}}^m\left(n\right)\→u^m\left(n\right)$

6) willWithCopy in control system, to u^m(n) andIt is filtered obtaining line spectrum signal x^m(n) andAs follows:

$\begin{array}{c}{x}^m\left(n\right)=c_R^m\left(n\right)u^m\left(n\right)-\mathrm{\ϵ}{c}_I^m\left(n\right)u_Q^m\left(n\right)\\ x_Q^m\left(n\right)=c_R^m\left(n\right)u_Q^m\left(n\right)+\mathrm{\ϵ}{c}_I^m\left(n\right)u^m\left(n\right)\end{array}---\left(13\right)$

7) step 1 is repeated)-6), make x^m(n) andIt is respectively relative to the signal x before extracting^o(n) andAlmost without Phase contrast；If frequency fluctuation, then formula (12) can adjust rapidly parameterWithMake phase difference compensation tracking frequency Fluctuate, remain phase difference compensation effect.

Knowable to formula (12), 2 rank parameter adaptives of phase difference compensation adjust without the secondary using control system Passage $S_{\mathrm{lk}}\left(z\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}$ Characteristic, controller coefficient $w_{\mathrm{lk}}^m\left(n\right),\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}$ With line spectrum error signalK=1 ..., L.Handled signal is time-domain signal.

Formula (12) need not call control system secondary channel characteristic and controller coefficient, not by isolation mounting physical characteristic shadow Ring, coupling weak, therefore fast convergence rate, operand are little.Additionally from step 1)-6) and Fig. 2 a knowable to, phase difference compensation is not Can be to control system amount of imports external signal, thus without introducing additional noise.

With reference to Fig. 1, Fig. 2 b, in each frequency range, use B^mZ () is to line spectrum reference signal x^m(n) and line spectrum orthogonal reference signal Carry out a narrow-band filtering again, obtain signalWithThen input secondary channel inversion model wave filter, it is not necessary to time-frequency Conversion, can realize the filtering of secondary channel inversion model at each frequency range L × L 2 rank filtering operations, obtain secondary time-domain signal The filtered reference signal of passage inversion model $S_{\mathrm{lk}}^m\left(n\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}.$

Step 6 in claim 1) described in each frequency range secondary channel inversion model filtering, control start before Offline mode is used to take its mid frequency identification secondary channel model in each frequency rangeFor L × L rank plural number square formation, its kth Row l row complex elementThe most corresponding l road controls signal to the secondary channel frequency response model value of kth road error signal；Then Off-line is inverted to obtain the secondary channel inversion model of each frequency rangeFor L × L rank plural number square formation.

Step 6 in claim 1) described in each frequency range secondary channel inversion model filtering, its signal flow block diagram As shown in Figure 2 b.If secondary channel inversion modelThe conjugation of l row kth row complex element isCount the most as follows Calculate each frequency range filtered reference signal of secondary channel inversion model $r_{\mathrm{lk}}^m\left(n\right),$ $\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}:$

$r_{\mathrm{lk}}^m\left(n\right)={\hat{p}}_{\mathrm{lk}}^m{x}_B^m\left(n\right)+{\hat{q}}_{\mathrm{lk}}^m{x}_{\mathrm{QB}}^m\left(n\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}---\left(14\right)$

Knowable to formula (14), in each frequency range, only need to carry out L × L 2 rank filtering and can realize the filter of secondary channel inversion model Ripple, handled signalWithFor time-domain signal, time-frequency conversion need not be carried out.

In each frequency range, respectively obtain the output signal of adaptive controllerL=1 ..., L；By each frequency range control signalL=1 ..., obtain control signal after L superpositionL=1 ..., L.This control is believed Number pass through secondary channel $S_{\mathrm{lk}}\left(z\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}$ Arrive kth=1 ..., at L error vibrations sensor, with desired signalK=1 ..., after L acts on jointly, obtain error signalK=1 ..., L.

With the narrow band filter B of each frequency range^mZ () is to error signalK=1 ..., L is filtered, and extracts The line spectrum error signal of each frequency rangeK=1 ..., L.

In each frequency range, according to formula $w_{\mathrm{li}}^m(n+1)=w_{\mathrm{li}}^m\left(n\right)-\widetilde{\mathrm{\μ}}\underset{k=1}{\overset{L}{\mathrm{\Σ}}}{e}_k^m\left(n\right)r_{\mathrm{lk}}^m(n-i),\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}$ Update self-adaptive controlled Device time-domain coefficients processed $w_{\mathrm{lk}}^m\left(n\right),\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}.$

WhereinFor restraining step-length,Each frequency rangeCan be independently arranged.

With reference to Fig. 1, in figure by 2 actuator, 2 error pick-ups, 2 narrowband slice control algolithm as a example by, have narrow Band frequency range 1 controls subsystem and narrowband slice 2 controls subsystem, and two control subsystem structures are identical；It practice, actuator, Error pick-up, narrowband slice quantity can be not limited to 2.

In Fig. 1, the subscript o of each variable represents that full frequency band, subscript 1 represent that narrowband slice 1, subscript 2 represent narrowband slice 2, By that analogy ...；For universality, follow-up explanation m represents.The Double-number subscript 11,12,21,22 of each variable is compiled with actuator Number l and error pick-up numbering k are relevant；For universality, represent with lk or kl in follow-up explanation.Under the single digital of each variable Mark 1,2 is relevant with actuator numbering l and error pick-up numbering k；For universality, represent with l or k in follow-up explanation.Become Amount adds inferior Q and represents that ratio is without Q identical variable phase shift-pi/2；Variable adds inferior B and represents ratio without the identical variable of B again Have passed through a narrow-band bandpass filtering.

V in Fig. 1^oN () is full range original reference signals, obtain full range reference signal x after being input to M point delayer^o(n), Full range orthogonal reference signal is obtained after input Hilbert changer (being realized by 2 × M+1 rank FIR filter)Relative to x^o(n) phase shift-pi/2.In narrowband slice m controls subsystem, B^mZ () is that narrowband slice m is corresponding Narrow band filter, u^mN () is for using B^mZ () is to x^oThe line spectrum signal extracted after the filtering of (n) narrow-band bandpass,For with B^m Z () is rightThe line spectrum signal extracted after narrow-band bandpass filtering,Relative to u^mN () still keeps phase shift-pi/2；By u^m (n) andInput narrow-band bandpass filtered phase difference adaptive compensator, each frequency range all has an independent above-mentioned compensation Device, this compensator is as shown in Figure 2 a.After compensator, line spectrum reference signal x of output^m(n) andIt is respectively relative to x^o (n) andWithout phase contrast.Use B^mZ () is to x^m(n) andObtain after carrying out a narrow-band filtering againWith WillWithIn input secondary channel inversion model wave filter, each frequency range all have one independent, as shown in Figure 2 b Secondary channel inversion model wave filter.

In Fig. 1, P_kZ () is primary vibration source to the primary channel transfer function between the error signal of kth road, W_l ^m(for arrowband frequency Section m controls the l road controller transfer function in subsystem, and each road controller of narrowband slice m is all with x^mN () is as input letter Number, y_l ^mN () is and W_l ^mZ control output signal that () is corresponding, y_lN () is that the l road after each narrowband subsystem parallel connection superposition controls Output signal.S_klZ () is that l road control signal is via the secondary channel between l road actuator to kth road error signal.d_k ^o (n) and e_k ^oN () is respectively kth road desired signal and error signal, e_k ^mN () is for using B^mZ () is to e_k ^oN () carries out narrow-band bandpass filter The line spectrum error signal that ripple extracts.

Part in dotted line frame in Fig. 2 a, for the narrow-band bandpass filtered phase difference adaptive compensator of designed, designed of the present invention (narrowband slice m's) structure chart.The input signal of this compensator is u^m(n) andCompensator parameter isWithWith this Two parameters are multiplied with input signal, are added, and obtaining output signal is x^m(n) andCompensator parameterWithCan be online Self-adaptative adjustment, the structure of adjustment link as shown in Figure 2 a, whereinWithFor with B^mZ () is to u^m(n) andAgain The signal obtained after carrying out a narrow-band filtering, uses parameterWithWithWithIt is multiplied, is added, obtain signalWith $e_i^m\left(n\right)=u^m\left(n\right)-u_{\mathrm{BC}}^m\left(n\right)$ Self-adaptative adjustment parameter is carried out as errorWithParameter can be madeWith Compensate narrow band filter B^mThe phase contrast of (z), so that x^m(n) andIt is respectively relative to the signal x before extracting^o(n) WithAlmost without phase contrast.

Part in dotted line frame in Fig. 2 b, for secondary channel inversion model wave filter (the narrowband slice m of designed, designed of the present invention ) structure chart.The input signal of this inversion model wave filter isWithFilter parameterFor narrowband slice m time Level channel pattern inverse matrixIn the value of real part of l row kth row complex element,For corresponding imaginary values；r_lk ^mN () is warp Crossing the filtered reference signal of secondary channel inversion model of narrowband slice m, filtering computing formula isIt is identical with the structured flowchart in Fig. 2 b.

The algorithm that the present invention is explained in detail below still can take in the case of secondary channel frequency response matrix eigenvalue dispersion is big Obtain the principle of very fast convergence rate；And phase difference compensation improves control algolithm robustness principle.

The time domain FB(flow block) (Fig. 1 and Fig. 2 a, b) of inventive algorithm is equivalent to the frequency domain theory diagram of Fig. 3, the most here Illustrate in conjunction with Fig. 3.

Owing to each frequency component of linear system is mutually orthogonal, the most only analysis single-frequency situation can not be lost, if line spectrum letter Number frequency is ω^m。

Below from the principle of inventive control algorithm, what is said or talked about beneficial effects of the present invention:

The parameter adaptive adjustment principle of phase difference compensation:

In frequencies omega^mPlace, if x^oN the frequency spectrum of () is plural number X^o, B^mZ the frequency response of () is plural numberPhase contrast is mended The frequency response repaying device is plural number $C^m=c_R^m+j{c}_I^m,$ Line spectrum signal u^m(n)、x^mN () is in frequencies omega^mPlace Frequency spectrum be respectively plural number U^m、X^m, as it is shown on figure 3, wherein U^m=B^mX^o, then the parameter of phase difference compensation Self-adaptative adjustment step is equivalent at frequency domain:

1) B is used^mTo U^mIt is filtered, obtains

2) C is used^mRightIt is filteredAs follows:

$U_{\mathrm{BC}}^m=C^m{U}_B^m---\left(15\right)$

3) calculateWith U^mBetween error

$E_u^m=U^m-I_{\mathrm{BC}}^m---\left(16\right)$

4) C is adjusted according to following algorithm online adaptive^m:

$C^m=C^m+\widetilde{\mathrm{\ϵ}}{\left(U_B^m\right)}^{*}{E}_u^m---\left(17\right)$

Wherein subscript^*Represent conjugation.

5) C is worked as^mAdjusted convergence by online adaptive, haveI.e.That is $C^m{U}_B^m\→{\left(B^m\right)}^{-1}{U}_B^m,$ This Time C^m→(B^m)^-1。

6) by C^mCopy in control system, to U^mIt is filtered obtaining X^mAs follows:

X^m=C^mU^m=(B^m)^-1B^mX^o=X^o (18)

From formula (18), C^mCan compensate for narrow band filter frequency response B^mPhase contrast, make the X after extraction^mRelative to X before extraction^oAlmost without phase contrast.

7) step 1 is repeated)-6)；If frequencies omega^mFluctuation, then formula (17) can adjust rapidly C^m, make phase difference compensation follow the tracks of Frequency fluctuation, remain phase difference compensation effect.

Phase difference compensation raising control algolithm robustness principle:

In frequencies omega^mPlace, if Multi Channel Controller frequency response is complex vector locatedWherein recover ElementRepresent the controller frequency response of l road；If secondary channel frequency response is plural number square formation S^m, wherein row k l row complex element (S^m)_kl Represent that l road controls signal to the secondary channel frequency response of kth road error signal.

If signal x^o(n)、x^mN () is in frequencies omega^mThe frequency spectrum at place is plural number X^oAnd X^m；If multi channel signals is vectorial $e^o\left(n\right)={[e_1^o\left(n\right),...,e_k^o\left(n\right),...,e_L^o\left(n\right)]}^T,d^o\left(n\right)={[d_1^o\left(n\right),...,d_k^o\left(n\right),...,d_L^o\left(n\right)]}^T,$ $e^m\left(n\right)={[e_1^m\left(n\right),...,e_k^m\left(n\right),...,e_L^m\left(n\right)]}^T,d^m\left(n\right)={[d_1^m\left(n\right),...,d_k^m\left(n\right),...,d_L^m\left(n\right)]}^T,$ $y^m\left(n\right)={[y_1^m\left(n\right),...,y_l^m\left(n\right),...,y_L^m\left(n\right)]}^T$ At ω^mThe frequency spectrum at place is complex vector located $E^o={[E_1^o,...,E_k^o,...,E_L^o]}^T,$ $D^o={[D_1^o,...,D_k^o,...,D_L^o]}^T,E^m={[E_1^m,...,E_k^m,...,E_L^m]}^T,D^m={[D_1^m,...,D_k^m,...,D_L^m]}^T$ With As it is shown on figure 3, there is X^m=B^mX^o, D^m=B^mD^o, E^m=B^mE^o.Subscript T represents transposition.

Above-mentioned frequency spectrum and frequency response have a following mutual relation:

Y^m=W^mX^m (19)

E^m=B^m(D^o+S^mY^m)=(D^m+B^mS^mW^mX^m) (20)

From formula (20), containing the control algolithm of narrow-band bandpass filtering link, optimal controller coefficient is

$\begin{array}{c}{W}_{\mathrm{opt}}^m=-{\left(B^m\right)}^{-1}{\left(S^m\right)}^{-1}{\left(X^m\right)}^{-1}{D}^m\\ =-{\left(B^m\right)}^{-1}{\left(S^m\right)}^{-1}{\left(X^o\right)}^{-1}{D}^o\end{array}---\left(21\right)$

In engineering, rotation, excited frequency ω of reciprocating machine^mMay fluctuate in certain limit.Owing to narrow-band bandpass is filtered The passband phase-frequency characteristic of ripple device is the most precipitous, and frequency fluctuation can make B^mLarger fluctuation occurs, consequently, it is possible to cause in formula (21) 'sLarger fluctuation occurs, affects Self Adaptive Control effect.

After have employed the phase difference compensation of the present invention, after this compensator string is associated in narrow band filter, such as Fig. 3 Shown in, then there is X^m=C^mU^m=C^mB^mX^o.From above, compensator parameter can with frequency fluctuation self-adaptative adjustment, it is ensured that its Frequency response C^m→(B^m)^-1, then have X^m→X^o, substitute into formula (21) and understand, now haveCan make controller Excellent solutionNot by B^mImpact, suppresses frequencies omega^mCaused by narrow-band bandpass filtering link during fluctuationFluctuation, thus improve line Spectral frequency ω^mThe robustness actively controlled during fluctuation.

The algorithm of the present invention still can obtain very fast receipts in the case of secondary channel frequency response matrix eigenvalue dispersion is big Hold back the principle of speed:

From above, the adaptive controller time-domain coefficients more new formula of inventive algorithm is:

$w_{\mathrm{li}}^m(n+1)=w_{\mathrm{li}}^m\left(n\right)-\widetilde{\mathrm{\μ}}\underset{k=1}{\overset{L}{\mathrm{\Σ}}}{e}_k^m\left(n\right)r_{\mathrm{lk}}^m(n-i),\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}$

$r_{\mathrm{lk}}^m\left(n\right)={\hat{p}}_{\mathrm{lk}}^m{x}_B^m\left(n\right)+{\hat{q}}_{\mathrm{lk}}^m{x}_{\mathrm{QB}}^m\left(n\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}$

WhereinFor secondary channel inversion modelThe conjugation of l row kth row complex element, i.e.

If secondary channel inversion model filtered reference signal matrixAt frequency Rate ω^mThe frequency spectrum at place is complex matrixAnd use B^mZ () is to x^mN () carries out an arrowband filter again The signal obtained after rippleIts frequency spectrum isHaveAs shown in Figure 3.

The most above-mentioned adaptive controller time-domain coefficients more new formula is equivalent at frequency domain:

${W_l}^m={W_l}^m-\widetilde{\mathrm{\μ}}\underset{k=1}{\overset{L}{\mathrm{\Σ}}}{\left(R_{\mathrm{lk}}^m\right)}^{*}{E}_k^m,l=1,...,L---\left(22\right)$

$R_{\mathrm{lk}}^m={\left[{\left({\hat{S}}^m\right)}^{-1}\right]}_{\mathrm{lk}}^{*}{X}_B^m,\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}---\left(23\right)$

It is written as vector sum matrix form, is had:

$W^m=W^m-\widetilde{\mathrm{\μ}}{\left(R^m\right)}^{*}{E}^m---\left(24\right)$

$R^m={\left[{\left({\hat{S}}^m\right)}^{-1}\right]}^{*}{X}_B^m---\left(25\right)$

So that

$W^m=W^m-\widetilde{\mathrm{\μ}}{\left({\hat{S}}^m\right)}^{-1}{\left(X_B^m\right)}^{*}{E}^m---\left(26\right)$

Formula (21) is deformed intoSubstitution formula (26), simultaneous formula (20) then has simultaneously:

$W^m(n+1)-W_{\mathrm{opt}}^m=[I-\widetilde{\mathrm{\μ}}{\left({\hat{S}}^m\right)}^{-1}{S}^m][W^m\left(n\right)-W_{\mathrm{opt}}^m]---\left(27\right)$

Understand the algorithm of the present invention, convergence and matrixRelevant, if modeling error is little, then matrix L eigenvalue all approximate 1, and restrain step-lengthShould choose according to eigenvalue of maximum inverse, therefore be easier to choose suitable step-lengthMake the multi-way contral algorithm of the present invention on the premise of ensureing stability, Fast Convergent.

By contrast, FxLMS algorithm has:

$W^m(n+1)-W_{\mathrm{opt}}^m=[I-\widetilde{\mathrm{\μ}}{\left({\hat{S}}^m\right)}^H{S}^m][W^m\left(n\right)-W_{\mathrm{opt}}^m]--\left(28\right)$

Wherein subscript H represents conjugate transpose.

Understand FxLMS algorithm, convergence and matrixRelevant, for baroque machinery active vibration isolation manifold Road system, matrixL eigenvalue in, the magnitude of eigenvalue of maximum and minimal eigenvalue generally differs greatly, feature Value dispersion is even up to 10²～10⁵, for ensureing stability, restrain step-lengthShould choose according to eigenvalue of maximum inverse, therefore FxLMS algorithm the convergence speed is slower.

The process that realizes and the measured result of the present invention be described in detail in detail by specific embodiment and comparative example below, and with FxLMS algorithm Contrast.

Embodiment 1,

Use a 200kW diesel generating set master peculiar to vessel passive hybrid vibration isolation device as implementation environment.Such as Fig. 4 institute Showing, in acceleration transducer i.e. figure installed by upper strata under-chassis, 1 as reference sensor；Pacify near 6 hybrid isolators of pedestal Fill 6 acceleration transducers (1#～6# measuring point) as error pick-up, wherein 2#, 4# and 6# measuring point i.e. 5-1,5-2 and 5-3, And 1#, 3# and 5# measuring point is arranged on and surveys point-symmetric unit another side with 2#, 4# and 6#.This unit 2 weighs about 3.6 tons；By 6 magnetic Suspension actuator-passive hybrid isolator of air bag master 3 supports.Wideband and many line spectrum vibrations are excited, at the beginning of being during unit operating Level vibration source；The magnetic suspension actuator 4 being arranged in air bag vibration isolator is as secondary vibration source.

Control system uses the C6678 of TI company as core processor, dominant frequency 1GHz, is passed through by reference sensor signal Preposition amplification, anti-aliasing filter, AD conversion (sample rate 1000Hz) are digital signal, as reference signal v^o(n)；Run this Bright control algolithm, produces control signalL=1 ..., 6, through DA change (turnover rate 1000Hz) into analogue signal, Magnetic suspension actuator is driven, the Expected Response caused at error pick-up with primary vibration source by power amplifierK= 1 ..., obtain error signal after 6 common effectsK=1 ..., 6；With the narrow band filter B of each frequency range^mZ () is right Error signalK=1 ..., 6 are filtered, and extract the line spectrum error signal of each frequency rangeK=1 ..., 6；? Each frequency range withFor object function, according to inventive algorithm, update adaptive controller time-domain coefficients $w_{\mathrm{li}}^m\left(n\right),$ $\begin{array}{c}l=1,...,6\\ i=0,...,N-1\end{array}.$

From control signalL=1 ..., 6 arrive error signalK=1 ..., the secondary channel transfer function of 6 It is $S_{\mathrm{kl}}\left(z\right),\begin{array}{c}l=1,...,6\\ k=1,...,6\end{array}.$ Additionally, due to air bag vibration isolator passive vibration isolation respond well (> 30dB), therefore needed for secondary vibration source Control power much smaller than the exciting force of primary vibration source, the negligible secondary vibration source feedback coupling influence to reference signal.

Algorithm design parameter is provided that

According to the Frequency Response of device, 17.5～202.5Hz frequency ranges being pressed 5Hz bandwidth partition frequency range, mid frequency is respectively For 20Hz, 25Hz ... 200Hz, each narrowband slice takes mid frequency and carrys out identification secondary channel modelIt is that 6 × 6 rank are multiple Number square formation.

The narrow band filter B of corresponding each frequency range^mZ () is 8 rank elliptic filters, key property is: passband width 5Hz, intermediate zone width 2.5Hz, stopband attenuation is 40dB, and passband ripple is 2dB.

The Hilbert transformer of full frequency band is realized by rank, (2M+1)=21 FIR filter, and corresponding delay device is M=10 Individual point.

The exponent number N=15 of the adaptive controller of each frequency range, therefore its time-domain coefficients is $w_{\mathrm{li}}^m\left(n\right),\begin{array}{c}l=1,...,6\\ i=0...,14\end{array}.$

The secondary channel inversion model filtering of each frequency range is realized by 6 × 62 rank filtering.

Measured result is as follows:

The each frequency range of identificationCalculateWithAnd calculate eigenvalueWithThrough inspection, each eigenvalue real part is satisfied byWithFrom Fig. 5 a, frequency range divides rationally, and secondary channel model meets inventive control algorithm convergence Essential condition, and each eigenvalue real part of inventive algorithm relatively approximates 1；

For FxLMS algorithm, calculateWithAnd calculate eigenvalueWithFrom Fig. 5 b, dividing according to this frequency range, secondary channel model also can meet necessity of FxLMS algorithmic statement Condition (eigenvalue real part is all higher than 0), but the magnitude difference of each eigenvalue real part of FxLMS algorithm is the biggest.

Fig. 6 a, b are diesel engine speed 1200r/min, and inventive algorithm controls the mistake of 109.5Hz and 139Hz line spectrum vibration Difference signal power convergence curve, in figure, solid line is convergence curve, and the time of dotted line alignment is the control algolithm opening time.From figure Visible, control algolithm is opened at about 8s, the existing larger fluctuation of the vibrational line spectral power before control, presents amplitude unstable state, And inventive algorithm still can quickly, stable, effectively control.

Error signal merit when Fig. 7 a, b are diesel engine speed 1100r/min and 1200r/min, before and after control algolithm convergence Rate spectrum density, before wherein dotted line is for controlling, after solid line is control algolithm convergence.Understanding under each operating mode, control algolithm is 20 ～the most prominent 10～11 line spectrums of 200Hz Frequency Band Selection energy are controlled, the control effect of main line spectrum be up to 10～ 35dB/ root, the line spectrum after control is as little as submerged in broadband noise.

Embodiment 2,

With the vibrator 6 in Fig. 4 as primary vibration source, with vibrator voltage as reference signal, remaining is arranged all with real Execute example 1 identical.

Vibrator output 120Hz, 180Hz, 195Hz vibration.

Run inventive algorithm, the most also run arrowband FxLMS algorithm, to contrast convergence rate.

Fig. 8 a, b, c give inventive algorithm, convergence step size mu₁FxLMS algorithm, convergence step size mu₂=1.5 μ₁'s FxLMS algorithm controls 120Hz, the error power convergence curve contrast of 180Hz, 195Hz vibration.Understand FxLMS algorithm (μ₁) receive Hold back speed and be much slower than inventive algorithm, if strengthening convergence step-length to μ₂, then FxLMS algorithm dissipated the most rapidly in the starting stage.And Inventive algorithm, in the case of guaranteeing to control process stabilization convergence, convergence rate can after being more than FxLMS algorithm, and convergence Reach excellent control effect.

Embodiment 3,

With the vibrator 6 in Fig. 4 as primary vibration source, with vibrator voltage as reference signal, remaining is arranged all with real Execute example 1 identical.

Vibrator output is vibrated from 107.5Hz to 112.5Hz frequency sweep, and frequency is suddenlyd change to 107.5Hz's from 112.5Hz Vibration.

Running inventive algorithm, apply/do not apply phase difference compensation in inventive algorithm, contrast is with/without phase contrast The active under vibrator output 107.5Hz to 112.5Hz swept-frequency signal and frequency discontinuity operating mode of the inventive algorithm of compensator Control effect.As shown in Fig. 9 a, b.

Sweep velocity in Fig. 9 a, b is respectively 0.1Hz/s and 0.4Hz/s, the most corresponding frequency of each error power spike in figure Rate is suddenlyd change.Understand the algorithm without phase compensator from Fig. 9 a and can obtain the line spectrum of about 15～19dB when sweep velocity 0.1Hz/s Control effect, and be only capable of obtaining the control effect of about 5～8dB when Fig. 9 b sweep velocity 0.4Hz/s；And have phase compensator Inventive algorithm is when above-mentioned sweep velocity, and the line spectrum that can obtain about 35～38dB and 25～29dB respectively controls effect.

This explanation phase contrast adaptive compensator significantly improves the inventive algorithm control robustness when frequency fluctuation.

In Fig. 9 a, b, the frequency discontinuity of each error power spike correspondence 112.5Hz to 107.5Hz, it is known that with/without phase place The inventive algorithm of compensator is respectively provided with preferable robustness to frequency discontinuity, after line spectral frequencies suddenlys change, all can restrain rapidly, Unstable divergence problem does not occurs.

Claims (7)

1. the multichannel arrowband control algolithm of a mechanical active vibration isolation, it is characterised in that: specifically comprise the following steps that

1) near vibration source, reference signal v is picked up with vibrating sensor^oN (), by full range original reference signals v^o(n) input transfer Function isHilbert transformer, it is thus achieved that signalBy v^oN () input M point delayer, obtains Obtain signal x^o(n)；Owing to Hilbert transform has all-pass characteristic and full frequency band phase shift-pi/2 characteristic, therefore in formulaFor x^o N () signal after full frequency band phase shift-pi/2, subscript Q represents phase shift-pi/2；

2) according to the most ready-portioned frequency range, with the narrow band filter B of each frequency range^mZ () is to full range reference signal x^o(n) and Full range orthogonal reference signalIt is filtered, it is thus achieved that be in the line spectrum signal u of each frequency range^m(n) andBy u^m(n) andInput a phase difference compensation containing only 2 rank parameters, it is thus achieved that line spectrum reference signal x of each frequency range^mN () and line spectrum are orthogonal Reference signal2 rank parameters of this phase difference compensationWithCan adjust with frequency fluctuation online adaptive, with Guarantee x^m(n) andIt is respectively relative to the signal x before extracting^o(n) andAlmost without phase contrast；Subscript m represents that frequency range is compiled Number；

3) setting control system hardware and have L actuator and L error pick-up, each actuator is correspondence one in control algolithm Road adaptive controller, in each frequency range, obtains adaptive controller output signal according to equation belowL=1 ..., L:

$y_l^m\left(n\right)=\underset{i=0}{\overset{N-1}{\Σ}}{w}_{li}^m\left(n\right)x^m(n-i),l=1,...,L---\left(8\right),$

In formulaI=0 ..., N-1 is the time-domain coefficients of the l road adaptive controller of m frequency range, with Self Adaptive Control Device transfer functionCorresponding；

In formula, N is the exponent number of adaptive controller；

4) by each frequency range control signalL=1 ..., obtain control signal after L superpositionL=1 ..., L；Control letter NumberL=1 ..., L is through secondary channel S_kl(z),Arrive kth=1 ..., at L error vibrations sensor, With kth=1 ..., the desired signal at L error vibrations sensorK=1 ..., after L acts on jointly, obtain error SignalK=1 ..., L；

S in formula_klZ () represents that l road control signal shifts via the secondary channel between l road actuator to kth road error signal Function；

5) with the narrow band filter B of each frequency range^mZ () is to error signalK=1 ..., L is filtered, and extracts each The line spectrum error signal of frequency rangeK=1 ..., L；

6) in each frequency range, B is used^mZ () is to line spectrum reference signal x^m(n) and line spectrum orthogonal reference signalCarry out an arrowband again Filtering, obtains signalWithThen input secondary channel inversion model wave filter, is filtered by multichannel 2 rank in each frequency range Ripple computing carries out secondary channel inversion model filtering, obtains the filtered reference signal of secondary channel inversion model of each frequency range

7) in each frequency range, according to the equation below time-domain coefficients to each road adaptive controller Adjust Whole:

$w_{li}^m(n+1)=w_{li}^m\left(n\right)-\widetilde{\mathrm{\μ}}\underset{k=1}{\overset{L}{\Σ}}{e}_k^m\left(n\right)r_{lk}^m(n-i),\begin{array}{c}l=1,...,L\\ i=0,...,N-1\end{array}---\left(9\right),$

WhereinFor restraining step-length,Each frequency rangeCan be independently arranged；

Repeat step 1)～7), make the object function of each frequency rangeIt is gradually reduced, thus realizes the vibration of many line spectrums Actively control.

The multichannel arrowband control algolithm of a kind of machinery active vibration isolation the most according to claim 1, it is characterised in that: control The front narrowband slice that full frequency band is divided into a fixed width, uses offline mode to take its mid frequency identification secondary channel in each frequency range Model L × L rank plural number square formationThen off-line is inverted, and obtains the secondary channel inversion model of each frequency rangeL × L represents L work Dynamic device and L sensor.

The multichannel arrowband control algolithm of a kind of machinery active vibration isolation the most according to claim 1 and 2, it is characterised in that: Inspection ready-portioned frequency range in advance is the most reasonable, the method for secondary channel model the most satisfied convergence essential condition is as follows: at certain Identification secondary channel model at the low-limit frequency of frequency range left hand edgeAt the highest frequency of this frequency range right hand edge, identification secondary is led to Road modelIdentification secondary channel model at mid frequencyCalculateWithAnd calculateEigenvalueWithEigenvalueIf each eigenvalue real part is satisfied byWithThen illustrated divisions frequency range is reasonable, secondary channel modelWith Error is little, by the secondary channel model of mid frequencyIt is rational for control algolithm, when actual line spectral frequencies is positioned at respectively When the left hand edge of frequency range, right hand edge and center, all can meet the essential condition of control algolithm convergence.

The multichannel arrowband control algolithm of a kind of machinery active vibration isolation the most according to claim 1, it is characterised in that: described The phase difference compensation of each frequency range, its 2 rank parameter is respectivelyWithThis 2 rank parameter can be able to be passed through with frequency fluctuation The online adaptive link relatively independent with control system is adjusted, and is then copied in control system, to signal u^m(n) andIt is filtered processing, the phase contrast of narrow band filter under real-Time Compensation frequency fluctuation；Thus obtain and be respectively relative to Signal x before extraction^o(n) andLine spectrum reference signal x almost without phase contrast^m(n) and line spectrum orthogonal reference signal

5. according to the multichannel arrowband control algolithm of a kind of machinery active vibration isolation described in claim 1 or 4, it is characterised in that: The phase difference compensation of described each frequency range compensates the step of narrow band filter phase contrast:

1) B is used^mZ () is to line spectrum signal u^m(n) andIt is filtered, obtains line spectrum signalWith

2) parameter is usedWithRightWithIt is filtered obtaining line spectrum signalAs follows:

$u_{BC}^m\left(n\right)=c_R^m\left(n\right)u_B^m\left(n\right)-c_I^m\left(n\right)u_{QB}^m\left(n\right)---\left(10\right)$

3) calculateWith u^mError between (n)

$e_u^m\left(n\right)=u^m\left(n\right)-u_{BC}^m\left(n\right)---\left(11\right)$

4) parameter is adjusted according to following algorithm online adaptiveWith

$\begin{array}{c}{c}_R^m(n+1)=c_R^m\left(n\right)+\widetilde{\mathrm{\ϵ}}{u}_B^m\left(n\right)e_u^m\left(n\right)\\ c_I^m(n+1)=c_I^m\left(n\right)-\widetilde{\mathrm{\ϵ}}{u}_{QB}^m\left(n\right)e_u^m\left(n\right)\end{array}---\left(12\right)$

It is whereinConvergence step-length,Each frequency rangeCan be independently arranged；

5) whenWithAdjusted convergence by online adaptive, haveI.e.

6) willWithCopy in control system, as follows to u^m(n) andIt is filtered obtaining line spectrum signal x^m (n) and

$\begin{array}{c}{x}^m\left(n\right)=c_R^m\left(n\right)u^m\left(n\right)-c_I^m\left(n\right)u_Q^m\left(n\right)\\ x_Q^m\left(n\right)=c_R^m\left(n\right)u_Q^m\left(n\right)+c_I^m\left(n\right)u^m\left(n\right)\end{array}---\left(13\right)$

7) step 1 is repeated)-6), make x^m(n) andIt is respectively relative to the signal x before extracting^o(n) andAlmost without phase place Difference；If frequency fluctuation, then formula (12) can adjust rapidly parameterWithMake phase difference compensation tracking frequency fluctuate, Remain phase difference compensation effect；

Knowable to formula (12), 2 rank parameter adaptives of phase difference compensation adjust without the secondary channel using control system S_kl(z),Characteristic, controller coefficient With line spectrum error signalK=1 ..., L.Institute Process signal is time-domain signal.

The multichannel arrowband control algolithm of a kind of machinery active vibration isolation the most according to claim 1, it is characterised in that: in power Profit requires the step 6 of 1) described in each frequency range secondary channel inversion model filtering, control start before used offline mode Its mid frequency identification secondary channel model is taken in each frequency rangeFor L × L rank plural number square formation, its row k l row complex elementThe most corresponding l road controls signal to the secondary channel frequency response model value of kth road error signal；Then off-line is inverted The secondary channel inversion model of each frequency rangeFor L × L rank plural number square formation.

The multichannel arrowband control algolithm of a kind of machinery active vibration isolation the most according to claim 1, it is characterised in that: in power Profit requires the step 6 of 1) described in each frequency range secondary channel inversion model filtering, if the secondary channel inversion model of each frequency rangeThe conjugation of l row kth row complex element isThen it is calculated as follows the secondary channel inversion model filter of each frequency range Reference signal after ripple

$r_{lk}^m\left(n\right)={\hat{p}}_{lk}^m{x}_B^m\left(n\right)+{\hat{q}}_{lk}^m{x}_{QB}^m\left(n\right),\begin{array}{c}l=1,...,L\\ k=1,...,L\end{array}---\left(14\right)$

Knowable to formula (14), in each frequency range, only need to carry out L × L 2 rank filtering and can realize the filtering of secondary channel inversion model, Handled signal is time-domain signal.