CN103927761A - Fault weak signal feature extraction method based on sparse representation - Google Patents

Abstract

The invention provides a method used for implementing fault weak signal feature extraction based on sparse representation, and the purpose is to apply a sparse decomposition algorithm to the fault weak signal field. The method includes the following concrete steps that step1, an overcomplete atom dictionary corresponding to fault weak signals is built through historical data, an atom dictionary set partitioning method is adopted, the overcomplete dictionary is seen as a set, and time-frequency parameter scales, frequency, phase positions of atoms are determined to obtain a feature atom dictionary of the fault weak signals; step2, a rapid Fourier transform algorithm is combined with OMP sparse decomposition to obtain a series of atom parameters for describing signal features; step3, the fault signal features are extracted in an optimized mode through sparse principal component analysis (SPCA). Based on atom dictionary set partitioning and an FFT signal sparse decomposition OMP algorithm, complexity of the sparse decomposition algorithm is effectively reduced, speed of signal sparse decomposition is increased, the effect of signal sparse decomposition is improved, and optimizing extraction of signal features is achieved.

Description

A kind of fault feeble signal feature extracting method based on rarefaction representation

Technical field

The present invention relates to Technique of Weak Signal Detection, be specifically related to a kind of fault feeble signal feature extracting method based on rarefaction representation.

Background technology

Giant mechanical and electrical equipment is carried out in malfunction monitoring diagnostic procedure; its core component as: the fault signature of rotor, bearing and gear etc. is often very faint; if the signal processing method of energy uses advanced is identified the fault feeble signal feature of kernel component in time, exactly; will provide technical support for fault indication and evolution, life prediction and formulation maintenance policy; improve giant mechanical and electrical equipment overall operation safety and reliability, avoid hang-up and serious accident.The feature extraction of fault feeble signal has become study hotspot and the difficult point of fault diagnosis field.

The theory of Its Sparse Decomposition and applied research mainly concentrate on three aspects: the searching algorithm of Optimum Matching atom and improvement algorithm thereof, dictionary construction algorithm, the application of Its Sparse Decomposition algorithm in signal is processed.Matching pursuit algorithm is a simple basic Optimum Matching atom searching algorithm, it is by all atom computing inner products in residual signal and dictionary, and get peaked method and determine Optimum Matching atom, shortcoming is that speed of convergence is slower, can not determine in the iterative steps that is less than signal dimension and reaches convergence.

Orthogonal matching pursuit algorithm (Orthogonal Matching Pursuit, and improve algorithm and matching pursuit algorithm difference and be OMP), in iterative process, complete the orthogonalization to selecting atom, thereby make algorithm reach convergence in the step number that is less than signal dimension to be decomposed, and make the signal decomposition can Accurate Reconstruction.Match tracing, orthogonal matching pursuit and improvement algorithm thereof all belong to greedy algorithm, realize the search of Optimum Matching atom by the principle of traversal dictionary atom.

Due to the superperformance of signal Its Sparse Decomposition, signal Its Sparse Decomposition has caused many scholars' interest, and the Its Sparse Decomposition of signal has also been applied to many aspects of the signal processing such as denoising, compression, coding, parameter estimation, feature extraction, target identification.

The transform domain that develops into signal of Its Sparse Decomposition represents to provide new developing direction with feature extraction.The method, according to the feature of signal to be decomposed, selects to press close to most the atom of residual signals from over-complete dictionary of atoms, and what the atomic parameter that decomposition obtains characterized is the feature of representative waveform.But in actual applications,, there is the problem such as calculating degree complexity, Riming time of algorithm length in Its Sparse Decomposition algorithm, hardware realization and algorithm time have been proposed to high requirement.In actual environment, the noise that signal comprises in real work brings more complicated resolution to decomposition simultaneously.

Summary of the invention

In view of this, the object of the present invention is to provide a kind of fault feeble signal feature extracting method based on rarefaction representation, the method adopts based on former word bank set and divides and FFT signal Its Sparse Decomposition OMP algorithm, effectively reduce the complexity of Its Sparse Decomposition algorithm, improve the speed of signal Its Sparse Decomposition and the effect of signal Its Sparse Decomposition, realized the optimization of signal characteristic and extracted.

For achieving the above object, the invention provides following technical scheme:

A fault feeble signal feature extracting method based on rarefaction representation, comprises the following steps: step 1: set up the over-complete dictionary of atoms corresponding with fault feeble signal, complete former word bank set and divide; Step 2: utilize FFT to realize signal Its Sparse Decomposition, thereby obtain describing a series of atomic parameters of signal characteristic, decompose stop condition until meet; Step 3: decompose and finish, the atomic parameter obtaining is carried out to the feature extraction of non-negative rarefaction representation.

Further, in step 1, specifically comprise the following steps: 21: according to historical data and set resolution parameter, obtain yardstick, displacement, frequency, the amplitude of each atom, set up the over-complete dictionary of atoms corresponding with fault feeble signal; 22: identical to parameter yardstick, frequency, the amplitude of the atom i.e. atom of " waveform is identical " is divided into a class, realizes the set of former word bank and divide.

Further, in step 2, specifically comprise the following steps: 31: in the process of Its Sparse Decomposition, for an atom of former word bank, step down and pipette all possible value [0, N-1], improve the effect of signal Its Sparse Decomposition; 32: utilize fft algorithm, the residual error of atom and signal or signal is made to inner product <R N time ^kf,g _γ> is converted to R one time ^kf and g _γthe computing of simple crosscorrelation 33: adopt OMP algorithm to obtain signal and selected component and the residual component on atom at each, then by identical method decomposition residual component.

Further, in step 3, adopt non-negative sparse principal component analysis principle to extract the feature of fault feeble signal.

Beneficial effect of the present invention is: the present invention sets up the over-complete dictionary of atoms corresponding with fault feeble signal by historical data, the method that adopts former word bank set to divide, greatly reduce the computation complexity of signal Its Sparse Decomposition, and can represent more accurately the time-frequency characteristic of fault feeble signal; Fast fourier transform algorithm (FFT) is combined with OMP Its Sparse Decomposition, effectively reduces the complexity of Its Sparse Decomposition algorithm, improve the speed of signal Its Sparse Decomposition and the effect of signal Its Sparse Decomposition.Meanwhile, carry out feature extraction based on non-negative rarefaction representation, realized the optimization of signal characteristic and extracted.

Brief description of the drawings

In order to make the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, the present invention is described in further detail, wherein:

Fig. 1 is the process flow diagram of the fault feeble signal feature extracting method based on rarefaction representation of the present invention.

Embodiment

Below in conjunction with accompanying drawing, the preferred embodiments of the present invention are described in detail.

Fig. 1 is the process flow diagram of the method for the invention, and this method comprises the following steps:

S1: set up the over-complete dictionary of atoms corresponding with fault feeble signal, complete former word bank set and divide.According to historical data and set resolution parameter, by atom normalization, set up the over-complete dictionary of atoms corresponding with fault feeble signal.An atom g _γdetermined by 4 time and frequency parameter γ=(s, u, v, w), wherein s is contraction-expansion factor (scale factor), and u is the shift factor of atom, and v is atomic frequency, and w is the phase place of atom.

Make (s, v, w)=β, Γ _β={ β _i| i=1,2 ..., over-complete dictionary of atoms D={g _γ} _{γ ∈ Γ}decompose as formula (1):

$\left\{\begin{array}{c}D=D_{{\mathrm{\&beta;}}_1}\&cup;D_{{\mathrm{\&beta;}}_2}\&cup;D_{\mathrm{\&beta;}3}...\\ D_{\mathrm{\&beta;i}}\&cap;D_{{\mathrm{\&beta;}}_j}=\mathrm{\&phi;},i\&NotEqual;j\end{array}\right.---\left(1\right)$

Thereby in over-complete dictionary of atoms, each word bank of equal value is as formula (2):

$D_{{\mathrm{\&beta;}}_i}=\left\{g_{\mathrm{\&gamma;}}\right|\mathrm{\&gamma;}=(s,u,w,v)\&Element;\mathrm{\&Gamma;},(s,w,v)={\mathrm{\&beta;}}_i\}---\left(2\right)$

For each word bank of equal value , in signal Its Sparse Decomposition process, need only and generate and an atom of storage wherein γ _i=(s, u=N/2, v, w), (s, v, w)=β _i, other atoms in former word bank of equal value can pass through atom translation generates.

S2: utilize FFT to realize signal Its Sparse Decomposition, thereby obtain describing a series of atomic parameters of signal characteristic, decompose stop condition until meet, concrete steps are as follows:

S21: in the process of Its Sparse Decomposition, for an atom of former word bank, step down and pipette all possible value [0, N-1], improve the effect of signal Its Sparse Decomposition;

S22: utilize fft algorithm, the residual error of atom and signal or signal is made to inner product <R N time ^kf,g _γ> is converted to R one time ^kf and g _γthe computing of simple crosscorrelation

S23: adopt OMP algorithm to obtain signal and selected component and the residual component on atom at each, then by identical method decomposition residual component.

S3: decompose and finish, the atomic parameter obtaining is carried out to the feature extraction of non-negative rarefaction representation.Corresponding non-negative sparse principal component optimization is described as formula (3):

$W^{*}=\underset{W}{\arg }\left(\max {\left|\right|W^{T}X\left|\right|}_1\right),s.t.W^{T}W=I_m,\left|\right|{W\left|\right|}_1<t,W>0---\left(3\right)$

First Optimizing Search goes out the first factor (m=1), then carries out the Optimizing Search of all the other principal components by projection again, thereby realize, the atomic parameter obtaining is carried out to the feature extraction of non-negative rarefaction representation.

By above step, can realize the feature extraction to fault feeble signal.

Finally explanation is, above preferred embodiment is only unrestricted in order to technical scheme of the present invention to be described, although the present invention is described in detail by above preferred embodiment, but those skilled in the art are to be understood that, can make various changes to it in the form and details, and not depart from the claims in the present invention book limited range.

Claims (4)

1. the fault feeble signal feature extracting method based on rarefaction representation, is characterized in that: comprise the following steps:

Step 1: set up the over-complete dictionary of atoms corresponding with fault feeble signal, complete former word bank set and divide;

Step 2: utilize FFT to realize signal Its Sparse Decomposition, thereby obtain describing a series of atomic parameters of signal characteristic, decompose stop condition until meet;

Step 3: decompose and finish, the atomic parameter obtaining is carried out to the feature extraction of non-negative rarefaction representation.

2. a kind of fault feeble signal feature extracting method based on rarefaction representation according to claim 1, it is characterized in that: in step 1, specifically comprise the following steps: 21: according to historical data and set resolution parameter, obtain yardstick, displacement, frequency, the amplitude of each atom, set up the over-complete dictionary of atoms corresponding with fault feeble signal; 22: identical to parameter yardstick, frequency, the amplitude of the atom i.e. atom of " waveform is identical " is divided into a class, realizes the set of former word bank and divide.

3. a kind of fault feeble signal feature extracting method based on rarefaction representation according to claim 1, it is characterized in that: in step 2, specifically comprise the following steps: 31: in the process of Its Sparse Decomposition, for an atom of former word bank, step down and pipette all possible value [0, N-1], the effect of raising signal Its Sparse Decomposition; 32: utilize fft algorithm, the residual error of atom and signal or signal is made to inner product <R N time ^kf,g _γ> is converted to R one time ^kf and g _γthe computing of simple crosscorrelation 33: adopt OMP algorithm to obtain signal and selected component and the residual component on atom at each, then by identical method decomposition residual component.

4. a kind of fault feeble signal feature extracting method based on rarefaction representation according to claim 1, is characterized in that: in step 3, adopt non-negative sparse principal component analysis principle to extract the feature of fault feeble signal.