CN104089778B - Water turbine vibration fault diagnosis method - Google Patents

Abstract

The invention provides a water turbine vibration fault diagnosis method. The water turbine vibration fault diagnosis method is characterized in that the water turbine vibration fault diagnosis method includes the steps that wavelet transformation is used for conducting self-adaptive denoising on collected water turbine vibration signals, and the complexity of data is reduced while noise influence is reduced; a local mean decomposition method is used for conducting self-adaptive decomposition on signals pre-processed through wavelets, and fault feature information in the water turbine vibration signals is extracted; according to the vibration principle of a water turbine, the extracted feature information such as frequency and energy is analyzed, and the operating state of the water turbine is diagnosed. Firstly, wavelet transformation is used for pre-processing the signals, the noise influence on the processed signals is reduced, the complexity of the signals is reduced, the entropy of the signals is reduced, and thus the fault feather information can be extracted more accurately and more effectively through local mean decomposition.

Description

A kind of hydraulic turbine vibrating failure diagnosis method

Technical field

The invention belongs to vibration equipment fault diagnosis technology field, is a kind of hydraulic turbine vibrating failure diagnosis method, more Say body, be a kind of hydraulic turbine vibrating failure diagnosis method decomposed based on wavelet pretreatment and local average.

Background technology

Hydraulic turbine operation in the environment such as very noisy, electromagnetic interference, when breaking down typically all can in vibration body Existing, its vibration signal has non-stationary, many modulation, multicomponent feature, using traditional method, often it is difficult to accurately extract The fault characteristic information of the hydraulic turbine, at present, the analyzing and processing means of signal are the big obstructions for restricting fault diagnosis technology development.

Fourier transformation is a kind of integral transformation, whole time-domain signal is moved and be analyzed on frequency domain, and it lacks right The description of signal detail, and fault message often embodies in detail, Fourier transformation lacks the amplification effect to detailed information.

Wavelet transformation, although its time frequency window is variable, but higher time domain and frequency domain resolution can not be obtained simultaneously, it is little The essence of wave conversion is to remove approximate signal to be analyzed with the linear combination of one group of wavelet basis function, and this essence is a kind of linear change Change, when nonlinear properties are processed, have time error very big.

Hilbert-Huang transform is a kind of adaptive Time-Frequency Analysis Method, but which is crossed envelope, owes envelope, end effect The problems such as, and the negative frequency problem of the edge effect when instantaneous amplitude and instantaneous frequency is calculated and generation without physical significance is still It is to be solved.

The content of the invention

The purpose of the present invention is the shortcoming for overcoming prior art, there is provided a kind of hydraulic turbine vibrating failure diagnosis method, application This method can effectively extract the fault characteristic information being included in hydraulic turbine vibration signal, and on this basis, diagnose The running status of the hydraulic turbine.

The purpose of the present invention is realized by technical scheme below：

A kind of hydraulic turbine vibrating failure diagnosis method, is characterized in that, it comprises the following steps：

1) wavelet transformation is adopted, to the hydraulic turbine vibration signal self-adaptive solution for collecting, the same of influence of noise is being reduced When, reduce the complexity of data；

2) adaptive decomposition is carried out to the signal after wavelet pretreatment using local average decomposition method, extract water wheels machine vibration Fault characteristic information in signal；

3) with reference to the vibration mechanism of the hydraulic turbine, frequency, the characteristic information such as energy for extracting is analyzed, the fortune of the hydraulic turbine is diagnosed Row state.

Step (1) is using the hydraulic turbine vibration signal self-adaptive solution that wavelet transformation is to collecting, wherein, adaptive Answer the selection course of threshold value as follows：

(1) n high frequency coefficient D for decomposing this layer sorts by modulus value size, sets initial threshold, by threshold value following coefficient Zero setting, D=(D₁,D₂...D_k, 0...0), wherein D₁To D_kFor the wavelet coefficient for retaining；

(2) energy for calculating the wavelet coefficient for retaining accounts for signal energy percent η_i=∑ (d_j)²/∑(d_j)²+∑(d_i)² And the percent ξ shared by the number that wavelet coefficient is zero_i=(n-k)/n, wherein, d_jFor the wavelet coefficient for retaining, d_iFor zero setting Wavelet coefficient；

(3) abs (η are selected_i-ξ_i) corresponding minimum coefficient is used as threshold value T_i；

(4) do soft-threshold quantification treatment:

The method have technical effect that：Pretreatment is carried out to signal by wavelet transformation first, the signal after process is made an uproar Sound shadow is rung to be reduced, and the complexity of signal is reduced, and the entropy of signal reduces, and so decomposes through local average again, can be more accurate Really, effectively extract fault characteristic information.

Description of the drawings

Fig. 1 is a kind of hydraulic turbine vibrating failure diagnosis method flow chart of the invention；

Fig. 2 is hydraulic turbine machine cap horizontal vibration time domain beamformer；

Fig. 3 is the hydraulic turbine machine cap horizontal vibration oscillogram after wavelet pretreatment；

Fig. 4 is local average decomposition result figure；

Fig. 5 is the time-frequency energy profile built by each PF components.

Specific embodiment

With reference to Fig. 1, a kind of hydraulic turbine vibrating failure diagnosis method is comprised the following steps：

1) wavelet transformation is adopted, to the hydraulic turbine vibration signal self-adaptive solution for collecting, the same of influence of noise is being reduced When, reduce the complexity of data；

2) adaptive decomposition is carried out to the signal after wavelet pretreatment using local average decomposition method, extract water wheels machine vibration Fault characteristic information in signal；

3) with reference to the vibration mechanism of the hydraulic turbine, frequency, the characteristic information such as energy for extracting is analyzed, the fortune of the hydraulic turbine is diagnosed Row state.

Below by taking the vibration of hydraulic turbine machine cap as an example, sensor, the signal for collecting db4 small echos are set at machine cap Adaptive Wavelet Thrinkage is carried out, the selection course of adaptive threshold is as follows：

(1) the high frequency coefficient D for decomposing this layer sorts by modulus value size, sets initial threshold, threshold value following coefficient is put Zero, k coefficient before retaining, D=(D₁,D₂...D_k,0...0).

(2) energy for calculating nonzero coefficient accounts for signal energy percent η_i=∑ (d_j)²/∑(d_j)²+∑(d_i)²And small echo Coefficient is the percent ξ shared by zero number_i, wherein, d_jFor the coefficient for retaining, d_iFor the coefficient of zero setting.

(3) abs (η are selected_i-ξ_i) corresponding minimum coefficient is used as threshold value T_i。

(4) soft-threshold quantification treatment:

Fig. 2 is for collecting hydraulic turbine machine cap horizontal vibration time domain beamformer, it can be seen that vibration is contained greatly The influence of noise of amount, after small echo self-adaptive solution, its vibrational waveform as shown in figure 3, influence of noise is substantially reduced, and Waveform before and after contrast denoising, has similarity, it is known that the feature of signal is retained in trend.

Signal after wavelet pretreatment is carried out into adaptive decomposition with local average decomposition method, each component obtained after decomposition With actual physical meaning, and the characteristic information of primary signal is contained, catabolic process is as follows：

(1) all Local Extremums n of primary signal x (t) are determined_i, obtain the average of all adjacent Local Extremums Value m_i：

By all of adjacent mean point m_iCoupled together with straight line, be then smoothed with moving average method, obtained To local mean value function m₁₁(t)；

(2) obtain envelope estimated value

By all two neighboring envelope estimated values a_iConnected with straight line, be then smoothed using moving average method, Obtain envelope estimation function a₁₁(t)；

(3) by local mean value function m₁₁T () is separated from primary signal x (t), obtain removing low frequency signal part h₁₁(t),

h₁₁(t)=x (t)-m₁₁(t) (3)

(4) utilize h₁₁T () is divided by envelope estimation function a₁₁T () is with to a₁₁T () is demodulated, obtain FM Function s₁₁ (t),

s₁₁(t)=h₁₁(t)/a₁₁(t) (4)

(5) with s₁₁T () replaces primary signal x (t) repeat the above steps (1)-(4) just can obtain s₁₁T the envelope of () is estimated Function a₁₂T (), judges a₁₂T whether () be equal to 1, if equal to 1 goes to step (6), otherwise, with s_1nT () replaces x (t), weight Multiple step (1) arrives (4), until s_1n(t) be a pure FM signal, i.e. s_1nThe envelope estimation function a of (t)_1(n+1)(t)=1, In practical engineering application, stopping criterion for iteration can be a_1nT () ≈ 1, each component produced in iterative process are as follows,

In formula, m_1nT () is FM Function s_1nThe local mean value function of (t), a_1nT () is s_1(n-1)T the envelope of () estimates letter Number；

(6) all envelope estimation functions produced in iterative process mutually can be obtained envelope signal (instantaneous amplitude at convenience Function)

(7) by envelope signal a₁(t) and pure FM signal s_1nT () mutually can obtain first PF point of primary signal at convenience Amount：

PF₁(t)=a₁(t)s_1n(t) (8)

It contains highest frequency content in primary signal, is the FMAM signal of a simple component, its wink When amplitude be exactly envelope signal a₁(t), its instantaneous frequency f₁T () can be by pure FM signal s_1nT () is obtained, i.e.,：

(8) by first PF component PF₁T () is separated from primary signal x (t), obtain a new signal u₁(t)；

(9) by u₁T () replaces x (t) repeat step (1)-(8), circulate k time, until u_kT () is a monotonic function till,

In formula, PF_kT () is to be decomposed each PF components for obtaining, u by primary signal_kT () is residual components；

Primary signal x (t) can be by all of PF components and u_kT () reconstructs, i.e.,：

Referring to Fig. 4, local average decomposition method, that is, the decomposition result of LMD, LMD each component have actual physical meaning, The each PF components obtained by decomposition ask for the instantaneous frequency and amplitude of signal, build three-dimensional time-frequency energy profile, see Fig. 5, from As can be seen that containing the unit rotational frequency of 5hz in the vibration information of the hydraulic turbine in figure, and 1/4 faint frequency multiplication, 2-4 Frequency multiplication, and stronger 50hz frequency contents.From hydraulic turbine vibration mechanism, 50hz compositions therein are system acquisition data When the influence of noise that produces because of power plant's imperfect earth, 1/4 frequency multiplication composition is that the low frequency vortex rope under this load causes.Through many It is secondary analysis test, after guide vane opening is adjusted to specified aperture, 2-4 frequencys multiplication composition disappear, thus can analyze, this 2-4 frequency multiplication into Point be due to waterpower it is uneven caused by, it is this kind of it may be the case that the blade openings of stator are unbalanced, linear discontinuities, or due to Seal ring circularity not enough, causes runner to go out stream circumferentially uneven.The phenomenon will not be caused to the normal operation of the hydraulic turbine Too much influence, it is proposed that avoid this load area and run.As can be seen here, the local average decomposition method based on wavelet pretreatment, The characteristic information in hydraulic turbine vibration signal under strong noise background can be extracted, and can effectively diagnose the running status of unit.

Claims (1)

1. a kind of hydraulic turbine vibrating failure diagnosis method, it include in have：Using wavelet transformation, to the hydraulic turbine for collecting Vibration signal self-adaptive solution, while influence of noise is reduced, reduces the complexity of data；Using local average decomposition method pair Signal after wavelet pretreatment carries out adaptive decomposition, extracts the fault characteristic information in hydraulic turbine vibration signal；With reference to water wheels The vibration mechanism of machine, analyzes frequency, the energy feature information extracted, diagnoses the running status of the hydraulic turbine, it is characterized in that：It is described Using wavelet transformation, the threshold selection method adopted by the hydraulic turbine vibration signal self-adaptive solution to collecting is as follows：

(1) n high frequency coefficient D for decomposing this layer sorts by modulus value size, sets initial threshold, threshold value following coefficient is put Zero, D=(d₁, d₂..., d_k, 0 ..., 0), wherein d₁To d_kFor the wavelet coefficient for retaining；

(2) energy for calculating the wavelet coefficient for retaining accounts for signal energy percent η_i=Σ (d_j)²/Σ(d_j)²+Σ(d_m)²And small echo Coefficient is the percent ξ shared by zero number_i=(n-k)/n, wherein, when subscript i is as k wavelet coefficient is retained, η's and ξ Mark, d_jFor the wavelet coefficient for retaining, d_mFor the wavelet coefficient of zero setting；

(3) abs (η are selected_i-ξ_i) corresponding minimum coefficient is used as threshold value T；

(4) do soft-threshold quantification treatment:

$d_i=\left\{\begin{array}{cc}|d_i-T|\&CenterDot;sign\left(d_i\right)& \left|d_i\right|>T\\ 0& \left|d_i\right|<T\end{array}\right..$