CN102252748A - Cavitation noise modulation feature extraction method based on empirical mode - Google Patents

Abstract

The invention provides a cavitation noise modulation feature extraction method based on an empirical mode. The method comprises the following steps: firstly standardizing a short cavitation noise signal; carrying out bandpass filtering on the standardized signal to obtain the bandpass signal of cavitation noise; carrying out envelope detection on the bandpass signal to obtain an envelope signal; carrying out lowpass filtering on the envelope signal to obtain a low-frequency envelope signal; decomposing the low-frequency envelope signal into a plurality of intrinsic mode functions (IMFs) by using empirical mode decomposition analysis; selecting the optimum IMF through evaluation; carrying out Hilbert transformation on the optimum IMF to obtain a Hilbert spectrum of the optimum IMF; and calculating the instantaneous frequency at every moment by using the Hilbert spectrum, so as to complete cavitation noise modulation feature extraction. According to the method provided by the invention, based on the adaptability of empirical mode decomposition and high resolution of Hilbert-Huang transformation, the disadvantage of the traditional modulation feature extraction method that modulation feature extraction is difficultly carried out on short-time and non-stably modulated cavitation noise data can be overcome.

Description

Cavitation noise modulation signature extracting method based on empirical modal

Technical field:

The present invention relates to the hydroacoustic noise signal Processing, relate in particular to a kind of cavitation noise modulation signature extracting method based on empirical modal.

Background technology:

Cavitation is that the pressure of liquid stream is reduced to a kind of fluid phenomenon that this state vaporizing liquid pressure takes place when following, and cavity is crumbled and fall and produce shock wave in liquid, and microjet impact solid wall surface then produces vibration.When screw propeller produced cavitation in water, it was the wide range signal that the vibration that bubble breaks and water impact causes produces high-frequency impulse, often modulated by it by screw propeller tool rotation effect, so the modulation signature of cavitation noise has reflected the important information of screw propeller.By can extract the rotating speed of screw propeller to the demodulation analysis of cavitation noise.Conventional demodulation analysis method is that noise signal is carried out bandpass filtering, envelope detection, power spectrumanalysis, extracts modulation signature from the power spectrum of noise signal envelope detection signal.The modulating frequency resolution that the Modulation analysis of conventional cavitation noise obtains depends on the duration of cavitation noise, the resolution precision that is difficult to obtain under the situation of noise signal duration weak point.Under the situation of modulation source job insecurity, also can cause the performance of conventional method to descend serious.

" the ship noise signal DEMON based on the modern signal processing technology analyzes ", acoustic technique, 2006,25 (1): 71-74, studied the improvement to conventional rectification method such as the modern signal processing technology of utilizing higher-order spectrum analysis, wavelet analysis and svd, these methods have certain performance to improve than conventional rectification method, but can't solve the low problem of modulating frequency resolution of short signal." the Underwater Acoustic Object Broad-band Modulated Signal based on High-order Spectrum Purification detects ", commander's control and emulation, 2007,29 (1): 103-106 provides a kind of Xi Er of utilization baud conversion and separates the method that the diagonal angle section of mediation high-order cumulative amount purifies demodulation spectra.The modulation signature of the cavitation noise of low signal-to-noise ratio and low frequency signal serious interference extracted better effects, but can not solve the problem of the modulating frequency low precision of cavitation noise in short-term.

It is the important composition step of Hilbert-Huang conversion that empirical modal decomposes.Be different from the method for traditional use solid form window for the boundary substrate, the basis function of empirical modal boundary be the eigenmode state function that extracted in self-adaptive obtains from signal (Intrinsic Mode Function, IMF).It utilizes the variation of signal internal time yardstick to carry out the parsing of energy and frequency, signal is launched into the IMF:1 that satisfies following condition) the extreme point number of function equates with the zero crossing number or differs one; 2) be zero by the defined envelope average of local extremum envelope at any time.Satisfy the IMF of above-mentioned two conditions and the hilbert conversion spectrum of IMF has been constituted a kind of, obtain in recent years to use widely effective adaptive processing method non-linear, non-stationary signal.

Summary of the invention:

In order to overcome the deficiency that classic method is extracted the modulation of non-stationary cavitation noise, a kind of cavitation noise modulation signature extracting method based on empirical modal has been proposed, it utilizes the adaptivity of empirical modal decomposition and the high resolving power of Hilbert-Huang conversion, at cavitation noise data in short-term, evaluation and selection by mode, utilize the hilbert spectrum of optimal modal, obtain each instantaneous modulating frequency constantly, the modulation signature of finishing cavitation noise extracts.

The object of the present invention is achieved like this: a kind of cavitation noise modulation signature extracting method based on empirical modal, it is characterized in that: at cavitation noise signal in short-term, carry out signal normalization earlier, normalized signal is carried out the bandpass signal that bandpass filtering obtains cavitation noise, bandpass signal is carried out envelope detection obtain envelope signal, envelope signal is carried out low-pass filtering obtain low frequency envelope signal, utilizing empirical modal to decompose is decomposed into low frequency envelope signal a plurality of eigenmode state function IMF and estimates the IMF that chooses optimum, optimum IMF is carried out the Hilbert conversion obtain its Hilbert spectrum, utilize the Hilbert spectrum to calculate each instantaneous frequency constantly, finish the cavitation noise modulation signature and extract, comprise the steps:

A. gathering propeller for vessels cavitation noise burst by nautical receiving set is s (n), n=0, and 1 .., N-1, the sample frequency of cavitation noise burst is f _s, N 〉=f _s, data s (n) is carried out standardization,

E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n);

B. by bandpass filter, to s ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n);

C. to bandpass signal s ₂(n) carry out detection, obtain envelope signal s ₃(n);

D. to envelope signal s ₃(n) carry out low-pass filtering, obtain low frequency envelope signal s ₄(n);

E. to low frequency envelope signal s ₄(n) carry out decomposing with empirical modal, obtain k IMF component, step is as follows:

E.1 make r (n)=s ₄(n), k=0;

E.2 make h (n)=r (n), standard deviation SD=1;

Whether the extreme value number of E.3 judging h (n) is greater than 2, if E.9 execution if not, carries out next step;

E.4 find out all maximum points and the minimum point of h (n) respectively, utilize cubic spline interpolation, calculate the upper and lower envelope h of h (n) _Max(n) and h _Min(n);

E.5 calculate the equal value sequence of envelope,

E.6 make h _Pre(n)=and h (n), h (n)=h (n)-n (n);

E.7 according to formula

Basis of calculation difference SD;

If SD＞0.2 E.8, then E.3 redirect is carried out;

E.9 h (n) is saved as IMF as single order IMF _k(n), k=k+1;

E.10r(n)＝r(n)-h(n)；

If E.11 E.2 then carry out limit number＞2 of r (n), otherwise arrive E.12;

E.12 mode is decomposed end, obtains k IMF component IMF _i(n), i=0,1 ..., k-1;

F. select optimum IMF in k IMF, step is as follows:

F.1 calculate and ask each rank IMF _i(n) power spectrum obtains power spectrum P _i(f), i=0,1 ..., k-1, f=0,1 ..., N-1/2;

F.2 calculate all P _i(f) the quality coefficient Q in _i,

F.3 find out all quality coefficient Q _iMiddle maximal value Q _m, 0≤m≤k-1, IMF at this moment _m(n) be optimum IMF;

G.. utilize the IMF of Hilbert transformation calculations optimum _m(n) analytic signal z (n) calculates each instantaneous frequency F (l) constantly by analytic signal z (n), l=0, and 1 ..., N-2,, F (l) is the instantaneous modulating frequency of obtaining cavitation noise, and concrete steps are as follows:

G.1 to IMF _m(n) carry out the Hilbert conversion, obtain

G.2 construct analytic signal At this moment, $a\left(n\right)=\sqrt{{\mathrm{IMF}}_m^2\left(n\right)+{\widehat{\mathrm{IMF}}}_m^2\left(n\right)},\mathrm{\θ}\left(n\right)=\arctan \frac{{\widehat{\mathrm{IMF}}}_m\left(n\right)}{{\mathrm{IMF}}_m\left(n\right)};$

G.3 calculate each moment instantaneous frequency F (l) of analytic signal z (n), F (l)=f _s(θ (l+1)-θ (l)), l=0,1 ..., N-2, F (l) is the modulation signature that this method will be extracted.

Description of drawings

Fig. 1 is a process flow diagram of the present invention;

Fig. 2 carries out the process flow diagram that mode is decomposed to envelope signal among the present invention;

Fig. 3 is the process flow diagram that optimal modal is chosen among the present invention;

Fig. 4 is the time domain data of the embodiment of the invention;

Fig. 5 is the envelope signal that the embodiment of the invention obtains

Fig. 6 is the 5 rank IMF that the embodiment of the invention obtains

Fig. 7 is the modulating frequency curve that the embodiment of the invention obtains, the final modulation signature that obtains for this method.

Embodiment

Referring to Fig. 1-3, the present invention is directed to cavitation noise signal in short-term, carry out signal normalization earlier, normalized signal is carried out the bandpass signal that bandpass filtering obtains cavitation noise, bandpass signal is carried out envelope detection obtain envelope signal, envelope signal is carried out low-pass filtering obtain low frequency envelope signal, utilizing empirical modal to decompose is decomposed into low frequency envelope signal a plurality of IMF and estimates the IMF that chooses optimum, utilize the analytic signal of the optimum IMF of Hilbert transition structure, calculate each instantaneous frequency constantly by the analytic signal of optimum IMF, finish the cavitation noise modulation signature and extract, method comprises following process:

A. gathering propeller for vessels cavitation noise burst by nautical receiving set is s (n), n=0, and 1 .., N-1, the sampling rate of cavitation noise burst is f _s, data are carried out standardization,

E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n),

B. by bandpass filter, to s ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n),

C. to bandpass signal s ₂(n) carry out detection, obtain envelope signal s ₃(n)

D. to envelope signal s ₃(n) carry out low-pass filtering, obtain low frequency envelope signal s ₄(n)

E. to low frequency envelope signal s ₄(n) carry out decomposing with empirical modal, obtain k IMF component, concrete steps are as follows:

E.1 make r (n)=s ₄(n), k=0,

E.2 make h (n)=r (n), standard deviation SD=1,

Whether the extreme value number of E.3 judging h (n) is greater than 2, if carry out E.9.If not, continue to carry out,

E.4 find out all maximum points and the minimum point of h (n) respectively, utilize cubic spline interpolation, calculate the upper and lower envelope of h (n), h _Max(n) and h _Min(n),

E.5 calculate the equal value sequence of envelope,

E.6 make h _Pre(n)=and h (n), h (n)=h (n)-m (n),

E.7 according to formula

Basis of calculation difference SD,

If SD＞0.2 E.8, then E.3 redirect is carried out,

E.9 h (n) is saved as IMF as single order IMF _k(n), k=k+1,

E.10r(n)＝r(n)-h(n)，

If E.11 E.2 then carry out limit number＞2 of r (n), otherwise arrive E.12,

E.12 mode is decomposed end, obtains k IMF component IMF this moment _i(n), i=0,1 ..., k-1,

F. select optimum IMF in k IMF, step is as follows:

F.1 calculate and ask each rank IMF _i(n) power spectrum obtains P _i(f), i=0,1 ..., k-1, f=0,1 ..., N-1/2,

F.2 calculate all P _i(f) the quality coefficient Q in _i, method is

F.3 find out all quality coefficient Q _iMiddle maximal value Q _m, 0≤m≤k-1, IMF at this moment _mBe optimum IMF,

G. utilize the IMF of Hilbert transformation calculations optimum _m(n) analytic signal z (n) calculates each instantaneous frequency F (l) constantly by analytic signal z (n), l=0, and 1 ..., N-2,, F (l) is the instantaneous modulating frequency of obtaining cavitation noise, and concrete steps are as follows,

G.1 to IMF _m(n) carry out the Hilbert conversion, obtain

G.2 construct analytic signal

At this moment, $a\left(n\right)=\sqrt{{\mathrm{IMF}}_m^2\left(n\right)+{\widehat{\mathrm{IMF}}}_m^2\left(n\right)},\mathrm{\θ}\left(n\right)=\arctan \frac{{\widehat{\mathrm{IMF}}}_m\left(n\right)}{{\mathrm{IMF}}_m\left(n\right)},$

G.3 calculate each moment instantaneous frequency F (l) of analytic signal z (n), F (l)=f _s(θ (l+1)-θ (l)), l=0,1 ..., N-2, F (l) is the modulation signature that this method will be extracted.

Embodiment:

Gather the non-stationary modulated signal sequences s (n) in 5 seconds, n=0 wherein, 1 ..., 49999, sample frequency is f _s=10000Hz, its waveform as shown in Figure 4.

According to above-mentioned A step, data s (n) is carried out standardization,

E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n).According to the B step, select 32 rank FIR bandpass filter for use, band connection frequency is that 10Hz is to 1000Hz, to s ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n), again according to the C step to s ₂(n) carry out quadratic detection,

As shown in Figure 5.

According to above-mentioned D step, select 32 rank FIR low-pass filters for use, cutoff frequency is 100Hz, to envelope signal s ₃(n) carry out low-pass filtering, obtain low frequency envelope signal s ₄(n), according to above-mentioned E step, to low frequency envelope signal s ₄(n) carry out decomposing, obtain 5 rank IMF, as shown in Figure 6 with empirical modal.

According to above-mentioned F step, in k IMF, select optimum IMF, at this moment IMF ₁Be optimal modal.According to the G step, to IMF ₁Calculate and obtain modulation signature F (l), as shown in Figure 7.

Claims (1)

1. cavitation noise modulation signature extracting method based on empirical modal, it is characterized in that: at cavitation noise signal in short-term, carry out signal normalization earlier, normalized signal is carried out the bandpass signal that bandpass filtering obtains cavitation noise, bandpass signal is carried out envelope detection obtain envelope signal, envelope signal is carried out low-pass filtering obtain low frequency envelope signal, utilizing empirical modal to decompose is decomposed into low frequency envelope signal a plurality of eigenmode state function IMF and estimates the IMF that chooses optimum, optimum IMF is carried out the Hilbert conversion obtain its Hilbert spectrum, utilize the Hilbert spectrum to calculate each instantaneous frequency constantly, finish the cavitation noise modulation signature and extract, comprise the steps:

A. gathering propeller for vessels cavitation noise burst by nautical receiving set is s (n), n=0, and 1 .., N-1, the sample frequency of cavitation noise burst is f _s, N 〉=f _s, data s (n) is carried out standardization,

E{s (n) } be the average of s (n), Std{s (n) } be the standard deviation of s (n);

B. by bandpass filter, to s ₁(n) carry out bandpass filtering, obtain bandpass signal s ₂(n);

C. to bandpass signal s ₂(n) carry out detection, obtain envelope signal s ₃(n);

D. to envelope signal s ₃(n) carry out low-pass filtering, obtain low frequency envelope signal s ₄(n);

E. to low frequency envelope signal s ₄(n) carry out decomposing with empirical modal, obtain k IMF component, step is as follows:

E.1 make r (n)=s ₄(n), k=0;

E.2 make h (n)=r (n), standard deviation SD=1;

Whether the extreme value number of E.3 judging h (n) is greater than 2, if E.9 execution if not, carries out next step;

E.4 find out all maximum points and the minimum point of h (n) respectively, utilize cubic spline interpolation, calculate the upper and lower envelope h of h (n) _Max(n) and h _Min(n);

E.5 calculate the equal value sequence of envelope,

E.6 make h _Pre(n)=and h (n), h (n)=h (n)-m (n);

E.7 according to formula Basis of calculation difference SD;

If SD＞0.2 E.8, then E.3 redirect is carried out;

E.9 h (n) is saved as IMF as single order IMF _k(n), k=k+1;

E.10r(n)＝r(n)-h(n)；

If E.11 E.2 then carry out limit number＞2 of r (n), otherwise arrive E.12;

E.12 mode is decomposed end, obtains k IMF component IMF _i(n), i=0,1 ..., k-1;

F. select optimum IMF in k IMF, step is as follows:

F.1 calculate and ask each rank IMF _i(n) power spectrum obtains power spectrum P _i(f), i=0,1 ..., k-1, f=0,1 ..., N-1/2;

F.2 calculate all P _i(f) the quality coefficient Q in _i,

F.3 find out all quality coefficient Q _iMiddle maximal value Q _m, 0≤m≤k-1, IMF at this moment _m(n) be optimum IMF;

G. utilize the IMF of Hilbert transformation calculations optimum _m(n) analytic signal z (n) calculates each instantaneous frequency F (l) constantly by analytic signal z (n), l=0, and 1 ..., N-2,, F (l) is the instantaneous modulating frequency of obtaining cavitation noise, and concrete steps are as follows,

G.1 to IMF _m(n) carry out the Hilbert conversion, obtain

G.2 construct analytic signal

At this moment, $a\left(n\right)=\sqrt{{\mathrm{IMF}}_m^2\left(n\right)+{\widehat{\mathrm{IMF}}}_m^2\left(n\right)},\mathrm{\θ}\left(n\right)=\arctan \frac{{\widehat{\mathrm{IMF}}}_m\left(n\right)}{{\mathrm{IMF}}_m\left(n\right)};$

G.3 calculate each moment instantaneous frequency F (l) of analytic signal z (n), F (l)=f _s(θ (l+1)-θ (l)), l=0,1 ..., N-2, F (l) is the modulation signature that this method will be extracted.

Images