CN100358739C - Automobile semi-active pendant damping control method based on wavelet decomposition - Google Patents

Abstract

The present invention relates to an automobile semi-active suspension damping control method based on wavelet dividing. The present invention uses an acceleration sensor for extracting vibration signals of the automobile body of a running automobile in an on-line mode, the vibration signals are divided into a plurality of frequency bands from high to low by a wavelet transform method, an energy statistical method is used for determining the main frequency bands of the current vibration signals, suspension damping coefficients are adjusted through a controllable damping vibroshock according to damping coefficients and response curves of the vibration of axles and the automobile body, and thereby, the automobile can obtain good smoothness and control stability. The method uses the wavelet transform for the multiresolution analysis of vibration signals of the automobile body, the resolution of a high-frequency signal part is low, and the resolution of a low-frequency signal part is high, which conforms to the characteristic of vibration signals of the automobile. Because the algorithm only measures the acceleration of the spring loading weight, the road surface does not need predicting, the low frequency vibration and the high frequency vibration of the suspension are compromised, the present invention can be conveniently applied to the semi-active suspension of the automobile to realize the real-time control.

Description

A kind of method of controlling based on the auto magnetorheological semi-active suspension damping of wavelet decomposition

Technical field

The invention belongs to vehicle vibration damping control technology field, be specifically related to a kind of method of controlling based on the auto magnetorheological semi-active suspension damping of wavelet decomposition.

Background technology

Since the D.A.Crosby of the U.S. in 1973 and D.C.Karnopp propose the notion of semi-active suspension, the various countries scholar has carried out the semi-active suspension control method research based on the controllable damping shock absorber successively, what wherein find application on actual vehicle is skyhook damping control, its imagination is arranged on shock absorber between spring carried mass and the inertial coordinate-ceiling absolute motion speed of direct control spring carried mass.The skyhook damping control algorithm is simple, studies show that to need only the choose reasonable parameter, can be good at suppressing the vertical shake of spring carried mass, improves travelling comfort.But skyhook damping need be measured the absolute velocitye of spring carried mass and the relative velocity of spring carried mass and nonspring carried mass simultaneously, and obtaining of spring carried mass absolute velocitye is quite difficult, skyhook damping also requires the time lag of actuating unit less, otherwise cause the misoperation of " switch " control easily, and it is a cost to sacrifice landing property of tire, is unacceptable under some operating mode.

Control theories such as optimal control, fuzzy control, adaptive control, prediction control also are applied to automobile half active damping controls, but these methods usually are to be based upon on the basis of various hypothesis, with actual condition bigger difference are arranged; And calculation of complex restrains slowly mostly, and real-time is poor, therefore mainly rests on laboratory simulations, is difficult to be applied to real vehicle control.

Summary of the invention

Goal of the invention: the present invention proposes a kind of method of controlling based on the auto magnetorheological semi-active suspension damping of wavelet decomposition, the utilization method of wavelet is come the decomposition vibration signal, its objective is the normal acceleration that reduces spring carried mass and nonspring carried mass simultaneously, improve the driving comfort and the road-holding property of automobile.

Technical solution of the present invention is as follows:

The overall plan of the inventive method is: the vibration signal that utilizes acceleration pick-up vehicle body in the line drawing running car, the utilization method of wavelet is decomposed the high low some frequency ranges that arrive of serving as reasons with vibration signal, determine by the method for energy statistics which frequency range current vibration signal mainly is distributed in, determine to regulate the suspension damping coefficient according to damping coefficient and axletree, body vibrations response curve, thereby make automobile obtain good ride comfort and road-holding property by the controllable damping shock absorber.

Key of the present invention is that the utilization method of wavelet is come the decomposition vibration signal, and in conjunction with Fig. 2, its step is as follows:

1, establish: x (n) expression filter is at n input constantly, X ₀=[x (n), x (n-1) ... x (1)] ^TThe input vector that expression is formed by x (n); DWT represents that the wavelet transform of vibration signal decomposes; X _jExpression burst X ₀The low frequency signal sequence of j level after decomposing; D _jExpression burst X ₀J level high-frequency signal sequence; σ _jBe the mean effective value of j level detail signal sequence, σ _J+1Be the mean effective value of j level approximation signal sequence, mean effective value vector: P=[σ ₁, σ ₂..., σ _J+1]; w _jBe the cooresponding output coefficient of weight of j level detail signal sequence, w _J+1Be the cooresponding output coefficient of weight of j level approximation signal sequence, output weight vector: W=[w ₁, w ₂..., w _J+1] ^T

The calculating of controller output F:

F＝kPW (1)

K is an output gain in the formula.

2, the wavelet decomposition of vibration signal, in conjunction with Fig. 3:

If: G and H are respectively small echo low-pass filter and small echo high-pass filter, X _jAnd D _jThe approximation signal (low frequency) and detail signal (high frequency) transformation for mula that are the decomposition of j level are as follows:

$X_j\left(m\right)=\underset{k}{\mathrm{\Σ}}h(k-2m)X_{j-1}\left(k\right)---\left(2\right)$

$D_j\left(m\right)=\underset{k}{\mathrm{\Σ}}g(k-2m)X_{j-1}\left(k\right)---\left(3\right)$

Wherein h (n) and g (n) represent the shock response sequence of this filters H and G, and h (n) and g (n) have following relation:

g(n)＝(-1) ^1-nh(1-n) (4)

Have symmetric wavelet filter h (n) in order to improve the real-time of algorithm, to select, make that h (n)=h (n)

If l=k-2m, k=2m+l, if the length of shock response sequence h (n) is 2t+1,

-(t-1)/2≤l≤(t-1)/2, so formula (1) can be write as:

$X_j\left(m\right)=\underset{l=-t}{\overset{t}{\mathrm{\Σ}}}h\left(l\right)X_{j-1}(2m+l)---\left(5\right)$

Wushu (5) is launched into bilateral form:

$X_j\left(m\right)=\underset{l=0}{\overset{t}{\mathrm{\Σ}}}h\left(l\right)X_{j-1}(2m+l)+\underset{l=0}{\overset{t}{\mathrm{\Σ}}}h(-l)X_{j-1}(2m-l)h\left(0\right)X_{j-1}\left(2m\right)---\left(6\right)$

Because h (l)=h (l), so:

$X_j\left(m\right)=\underset{l=0}{\overset{t}{\mathrm{\Σ}}}h\left(l\right)[X_{j-1}(2m-l)+X_{j-1}(2m+l)]-h\left(0\right)X_{j-1}\left(2m\right)---\left(7\right)$

In like manner, formula (3) is write as following form

$D_j\left(m\right)=\underset{l=0}{\overset{t}{\mathrm{\Σ}}}{(-1)}^l[X_{j-1}(2m+1-l)+X_{j-1}(2m+1+l)]h\left(l\right)---\left(8\right)$

$-h\left(0\right)X_{j-1}(2m-1)$

Therefore utilize the relation of filters H and filter G and the symmetry of filters H, can simplify, reduce the part calculated amount, improve the real-time of algorithm the signal decomposition computing formula.

3, the adjustment of output weight vector:

The output weight vector is after off-line is adjusted, and output weight vector W does not generally need to adjust, when the control effect is not good enough, and can be by output weight vector regulating control to adjusting.Adjustment algorithm is as follows:

If the m time controller is output as F _m, export weight vector and mean effective value vector accordingly and be W _m=[w _{M, 1}, w _{M, 2}..., w _{M, j+1}] and P _m=[P _{M, 1}, P _{M, 2}..., P _{M, j+1}].

Instrument error vector: E _m=P _m-P _M-1=[e _{M, 1}, e _{M, 2}..., e _{M, j+1}];

The output weight is regulated vector: Δ W=[Δ w _{M, 1}, Δ w _{M, 2}..., Δ w _{M, j+1}];

$\mathrm{\&Delta;}{w}_{m,j}=\left\{\begin{array}{cc}{(-1)}^r\frac{e_{m,i}}{e_{m,i-1}}\mathrm{\&Delta;}{w}_{m-1,i}& (e_{m-1,i}>\mathrm{\&epsiv;and}|\frac{e_{m,i}}{e_{m-1,i}}|\&GreaterEqual;R)\\ 0& (e_{m,i}<\mathrm{\&epsiv;})\\ {(-1)}^r\mathrm{R\&Delta;}{w}_{m-1,i}& (e_{m,i}>\mathrm{\&epsiv;and}|\frac{e_{m,i}}{e_{m-1,i}}|<R)\end{array}\right.---\left(9\right)$

ε is the control threshold value in the formula, and r gets 0 or 1 according to the place frequency range, extinguishing coefficient R ∈ (0,0.618).

W _m+1＝W _m+ΔW _m (10)

4. by wavelet decomposition, vibration signal just can be decomposed into some frequency ranges from high to low, determines by the method for energy statistics which frequency range current vibration signal mainly is distributed in, and concrete steps are as follows:

(1) vibration signals at different levels that wavelet decomposition is obtained carry out energy statistics, the signal vector D after decomposing for the j level _i=[d _i(1) ..., d _i(n-1), d _i(n)] ^T, (i=1,2 ..., j, j+1; D _J+1=X _i), ask each vectorial mean effective value

${\mathrm{\&sigma;}}_i=\sqrt{\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{d}_i^2\left(k\right)},(i=\mathrm{1,2},...,j,j+1)$

(2) set up the mean effective value vector: P=[σ ₁, σ ₂..., σ _J+1] middle element ordering, choose two maximum element σ _mAnd σ _i

(3) according to the corresponding relation of wavelet decomposition signal and frequency, determine σ _mAnd σ _iThe frequency limit of place frequency range.If the highest frequency of original signal is f _Max, can calculate 7 grades of wavelet decomposition after the shared concrete frequency band of each layer be:

X ₇：(0～0.0078125)×f _max

D ₇：(0.0078125～0.015625)×f _max

D ₆：(0.015625～0.03125)×f _max

D ₅：(0.03125～0.0625)×f _max

D ₄：(0.0625～0.125)×f _max

D ₃：(0.125～0.25)×f _max

D ₂：(0.25～0.5)×f _max

D ₁：(0.5～1)×f _max

Advantage of the present invention:

This control method utilization method of wavelet is decomposed the high low some frequency ranges that arrive of serving as reasons with vibration signal, and the body vibrations signal is carried out multiresolution analysis, and high-frequency signal part resolution is low, and low frequency signal is differentiated high, meets the characteristics of automobile vibration signal.Because algorithm only need utilize the vibration signal of acceleration pick-up vehicle body in the line drawing running car, need not predict the road surface, can greatly reduce system cost, improve system reliability.Determine by the method for energy statistics which frequency range current vibration signal mainly is distributed in, determine to regulate the suspension damping coefficient according to damping coefficient and axletree, body vibrations response curve by the controllable damping shock absorber, take into account suspension low-frequency vibration and high-frequency vibration, thereby make automobile obtain good ride comfort and road-holding property.This control method realizes easily, can be applied to easily on the existing automotive suspension, realizes control in real time.

Description of drawings

Fig. 1 is the auto magnetorheological semi-active suspension system scheme drawing that adopts the wavelet frequency domain control method;

Fig. 2 is the FB(flow block) of wavelet frequency domain control algorithm;

Fig. 3 is the diagram of circuit of the multistage decomposition of signal;

The frequency range of Fig. 4 suspension response is divided, and wherein Fig. 4 A is a spring carried mass acceleration/accel transmission characteristic, and Fig. 4 B is a tire live load transmission characteristic;

The multistage decomposition of Fig. 5 vibration signal;

Fig. 6 floor compartment vertical shake control effect (40km/h);

Fig. 7 axletree vertical shake control effect (40km/h).

The specific embodiment

As shown in Figure 1, degree of will speed up sensor 1 is installed on vehicle body barycenter place, gather the bouncing of automobile body acceleration signal, input as wavelet frequency domain controller 6, after controller carries out wavelet decomposition to incoming signal, referring to Fig. 2 and Fig. 3, determine by the method for energy statistics which frequency range current vibration signal mainly is distributed in, according to damping coefficient and axletree, the body vibrations response curve is determined to regulate the suspension damping coefficient by the controllable damping shock absorber, the output control signal is regulated the damping force characteristics of actr (as the electric/magnetic rheological damper) 2-5 of controllable damping, finishes a Control Circulation.

When Fig. 5 had provided automobile and travels with 40km/h speed, 7 fractions of the acceleration signal of floor compartment vertical shake were separated the result, wherein X ₀Be the original signal that sensor acquisition arrives, X ₇Be the 7th grade of approximation signal (low frequency part), D ₁~ D ₇Be 1～7 grade of detail signal (HFS), as we can see from the figure, each layer wavelet transformation be logical or low-pass filter as band, and by wavelet transformation, the signal of each frequency range is clearly shown.The highest frequency of original signal is 50Hz, can calculate the shared concrete frequency band of each layer to be:

X ₇：0～0.39Hz

D ₇：0.39~0.78Hz

D ₆：0.78～1.56Hz

D ₅：1.56～3.13Hz

D ₄：3.125～6.25Hz

D ₃：6.25～12.5Hz

D ₂：12.5～25Hz

D ₁：25～50Hz

By the signal energy statistics, the mean effective value of vibration signals at different levels is: P=[σ ₁, σ ₂..., σ _J+1]=[0.002,0.005,0.004,0.01,0.09,0.23,0.16,0.03].Hence one can see that, and vibrational energy concentrates on D ₆: 0.78～1.56Hz and D ₇: these two frequency ranges of 0.39 ~ 0.78Hz.Therefore exporting dumping force should get smaller value, establishes output gain k=1000;

Frequency range according to suspension response shown in Figure 4 is divided as can be known, should be the low resistance state in [0 ~ 1.1Hz] frequency range, [1.1 ~ 9Hz] frequency range should be the high damping state, and 9 ~ 100Hz frequency range should be medium damping state, therefore the output weight vector can be set is W=[0,0,0.5,1,1,0.8,0.5,0.5].

Output damping force F=k*P*W=380N

Control effect when Fig. 6 has shown automobile with 40km/h speed travelling on the B level road surface.Black line is the spring carried mass-floor compartment normal acceleration power spectrum of passive suspension among the figure, and red line records floor compartment normal acceleration power spectrum after adopting the automotive semi-active suspension dumping force control method based on wavelet decomposition.Learn that relatively the control method that this patent adopted can effectively reduce the normal acceleration of the 4-12.5Hz of human body sensitivity, improve the automobile driving comfort.Fig. 7 has shown the normal acceleration power spectrum of nonspring carried mass-axletree, and as can be seen from the figure, by control, the vertical shake of axletree has also obtained better inhibition, has also improved road-holding property when improving vehicle ride comfort.

Claims (2)

1, a kind of method of controlling based on the automotive semi-active suspension damping of wavelet decomposition, its step is as follows:

(1) utilizes the vibration signal of acceleration pick-up vehicle body in the line drawing running car;

(2) the utilization method of wavelet is decomposed the high low some frequency ranges that arrive of serving as reasons with vibration signal;

(3) determine by the method for energy statistics which frequency range current vibration signal mainly is distributed in;

(4) determine to regulate the suspension damping coefficient according to damping coefficient and axletree, body vibrations response curve, thereby make automobile obtain good ride comfort and road-holding property by the controllable damping shock absorber;

Wherein: the step of utilization method of wavelet decomposition vibration signal is as follows:

(1) establish: x (n) expression filter is at n input constantly, X ₀=[x (n), x (n-1) ... x (1)] ^TThe input vector that expression is formed by x (n); DWT represents that the wavelet transform of vibration signal decomposes; X _jExpression burst X ₀The low frequency signal sequence of j level after decomposing; D _jExpression burst X ₀J level high-frequency signal sequence; σ _jBe the mean effective value of j level high-frequency signal sequence, σ _J+1Be the mean effective value of j level low frequency signal sequence, mean effective value vector: P=[σ ₁, σ ₂..., σ _J+1]; w _jBe the cooresponding output coefficient of weight of j level detail signal sequence, w _J+1Be the cooresponding output coefficient of weight of j level approximation signal sequence, output weight vector: W=[w ₁, w ₂..., w _J+1] ^T

The calculating of controller output F:

F＝kPW (1)

K is an output gain in the formula;

(2) wavelet decomposition of vibration signal:

Represent small echo low-pass filter and small echo high-pass filter, X respectively with G and H _jAnd D _jThe low frequency approximation signal and the high frequency detail signal transformation for mula that are the decomposition of j level are as follows:

$X_j\left(m\right)\text{=}\underset{k}{\mathrm{\&Sigma;}}h(k-2m)X_{j-1}\left(k\right)---\left(2\right)$

$D_j\left(m\right)\text{=}\underset{k}{\mathrm{\&Sigma;}}g(k-2m)X_{j-1}\left(k\right)---\left(3\right)$

Wherein h (n) and g (n) represent the shock response sequence of this filters H and G, and h (n) and g (n) have following relation:

g(n)＝(-1) ^1-nh(1-n) (4)

Have symmetric wavelet filter h (n) in order to improve the real-time of algorithm, to select, make that h (n)=h (n)

If l=k-2m, k=2m+l, if the length of shock response sequence h (n) is 2t+1 ,-(t-1)/2≤l≤(t-1)/2, so formula (1) can be write as:

$X_j\left(m\right)=\underset{l=-t}{\overset{t}{\mathrm{\&Sigma;}}}h\left(l\right)X_{j-1}(2m+l)---\left(5\right)$

Wushu (5) is launched into bilateral form:

$X_j\left(m\right)=\underset{l=0}{\overset{t}{\mathrm{\&Sigma;}}}h\left(l\right)X_{j-1}(2m+l)+\underset{l=0}{\overset{t}{\mathrm{\&Sigma;}}}h(-l)X_{j-1}(2m-l)-h\left(0\right)X_{j-1}\left(2m\right)---\left(6\right)$

Because h (l)=h (l), so:

$X_j\left(m\right)=\underset{l=0}{\overset{t}{\mathrm{\&Sigma;}}}h\left(l\right)[X_{j-1}(2m-l)+X_{j-1}(2m+l)]-h\left(0\right)X_{j-1}\left(2m\right)---\left(7\right)$

In like manner, formula (3) is write as following form

$D_j\left(m\right)=\underset{l=0}{\overset{t}{\mathrm{\&Sigma;}}}{(-1)}^l[X_{j-1}(2m+1-l)+X_{j-1}(2m+1+l)]h\left(l\right)--\left(8\right)$

$-h\left(0\right)X_{j-1}(2m-1)$

(3) adjusting of output weight vector:

The output weight vector is after off-line is adjusted, and output weight vector W does not generally need to adjust, when the control effect is not good enough, can be adjusted by output weight vector regulating control, and control method is as follows:

If the m time controller is output as F _m, export weight vector and mean effective value vector accordingly and be W _m=[w _{M, 1}, w _{M, 2}..., w _{M, j+1}] and P _m=[P _{M, 1}, P _{M, 2}..., P _{M, j+1}];

Instrument error vector: E _m=P _m-P _M-1=[e _{M, 1}, e _{M, 2}..., e _{M, j+1}];

The output weight is regulated vector: Δ W=[Δ w _{M, 1}, Δ w _{M, 2}..., Δ w _{M, j+1}];

$\mathrm{\&Delta;}{w}_{m,i}=\left\{\begin{array}{cc}{(-1)}^r\frac{e_{m,i}}{e_{m,i-1}}\mathrm{\&Delta;}{w}_{m-1,i}& (e_{m-1,i}>\mathrm{\&epsiv;and}|\frac{e_{m,i}}{e_{m-1,i}}|\&GreaterEqual;R)\\ 0& (e_{m,i}<\mathrm{\&epsiv;})\\ {(-1)}^r\mathrm{R\&Delta;}{w}_{m-1,i}& (e_{m,i}>\mathrm{\&epsiv;and}|\frac{e_{m,i}}{e_{m-1,i}}|<R)\end{array}\right.---\left(9\right)$

ε is the control threshold value in the formula, and r gets 0 or 1 according to the place frequency range, extinguishing coefficient R ∈ (0,0.618)

W _m+1＝W _m+ΔW _m (10)；

It is as follows that described step (3) determines that by the method for energy statistics current vibration signal mainly is distributed in the concrete steps of which frequency range:

A, the signal vector D after decomposing for the j level _i=[d _i(1) ..., d _i(n-1), d _i(n)] ^T, (i=1,2 ..., j, j+1; D _J+1=X _i), ask each vectorial mean effective value ${\mathrm{\&sigma;}}_i=\sqrt{\frac{1}{n}\underset{k=1}{\overset{n}{\mathrm{\&Sigma;}}}{d}_i^2\left(k\right)},$ (i＝1，2，...，j，j+1)；

B, mean effective value vector: P=[σ ₁, σ ₂..., σ _J+1] middle element ordering, choose two maximum element σ _mAnd σ _l

C, according to the corresponding relation of wavelet decomposition signal and frequency, determine σ _mAnd σ _lThe frequency limit of place frequency range is if the highest frequency of original signal is f _Max, calculate 7 grades of wavelet decomposition after the shared concrete frequency band of each layer be:

X ₇：(0～0.0078125)×f _max

D ₇：(0.0078125～0.015625)×f _max

D ₆：(0.015625～0.03125)×f _max

D ₅：(0.03125～0.0625)×f _max

D ₄：(0.0625～0.125)×f _max

D ₃：(0.125～0.25)×f _max

D ₂：(0.25～0.5)×f _max

D ₁：(0.5～1)×f _max。

2. method according to claim 1, it is characterized in that: according to damping coefficient and axletree, the body vibrations response curve determines that regulating the suspension damping coefficient by the controllable damping shock absorber specifically is, according to spring carried mass acceleration/accel under the different damping coefficient with respect to the intersection point of the transfer function amplitude-versus-frequency curve of road surface speed input and tire live load intersection point for the transfer function amplitude-versus-frequency curve of road surface speed input, divide the frequency of 0.1-100Hz into from low to high three frequency range A, B, C, use big damping coefficient at frequency range A, use little damping coefficient at frequency range B, for frequency range C, select medium damping coefficient for use, make the mean effective value of spring carried mass acceleration/accel and tire live load all remain on a lower level, A, B, the value of three Frequency points of C should obtain in conjunction with concrete vehicle test.

Images