CN103822786A - Wind turbine generator set mechanical equipment state diagnosis method based on multivariate statistical analysis - Google Patents

Abstract

The invention discloses a wind turbine generator set mechanical equipment state diagnosis method based on multivariate statistical analysis. The method is characterized by utilizing a sensor to collect state information generated by wind power equipment; performing feature extraction, signal analysis and state identification on the state information based on the multivariate statistical analysis; with low-dimension principal component feature expression technology expressing and classifying wind turbine generator set mechanical state, establishing an average correlation law to assess the ability, for describing the wind turbine generator set mechanical state, of each principal component; and selecting low-dimension principal component feature to express the comprehensiveness for the wind power equipment state features and the diagnosis of the wind power equipment state is realized. According to the wind turbine generator set mechanical equipment state diagnosis method based on the multivariate statistical analysis, early failure of the wind power equipment can be found and failure conditions can be accurately judged, utilization rate of the wind turbine generator set is improved, and cycle period and financial costs of maintenance and service are reduced as possible as one could; and the method can ensure safe, stable and reliable operation of the wind turbine generator set, and has great acceleration effect.

Description

Wind-powered electricity generation unit mechanical equipment state diagnostic method based on multivariate statistical analysis

Technical field

The invention belongs to wind power equipment status fault diagnostic method, particularly a kind of wind-powered electricity generation unit mechanical equipment state diagnostic method based on multivariate statistical analysis.

Background technology

Wind energy is a kind of clean regenerative resource, worldwide fast development.According to data: Chinese adding new capacity 18927.99MW in 2010, increase by 73.3% on a year-on-year basis, accumulative total installed capacity reaches 44733.29MW, and two indexs all occupy the first in the world.The newly-increased blower fan installed capacity of China in 2011 is 20665MW (2066.5 ten thousand kilowatts), increases by 9.18% than 2010, continues to keep first place, the world.Because wind energy turbine set is generally arranged in the environment of inclement condition, operating mode is extremely unstable, therefore the running status of wind-powered electricity generation unit is carried out to on-line monitoring and fault diagonosing, understand in real time the running status of wind-powered electricity generation unit key components and parts, the symptom of a trend of early detection accident, in the time that the fault of equipment is in the wearing and tearing initial stage, by early stage maintenance and replacing, improving the measures such as lubricating status eliminates fault in bud, avoid early sending out fault and cause even more serious fault to be destroyed, to guaranteeing that aerogenerator unit safe, stable, reliability service have great facilitation.

In wind power equipment actual moving process, all inevitably produce vibration, vibration is the external expression of the dynamics of plant equipment own, and vibration information is the carrier of research mechanical equipment state.So according to the analysis of mechanical equipment vibration signal and processing, just can obtain the running status of plant equipment, obtain the state change information that system parts causes due to reasons such as wearing and tearing, fatigues, and identify thus the fault of plant equipment or parts.For other mechanical equipment state treatment technologies, method is simple for analysis of vibration signal, and technology maturation is the main method using in current engineering.

Current domestic shortage is the research with feature extracting method to the processing of wind generating set vibration signal.Wind-powered electricity generation unit moves under variable speed operating mode, and fault characteristic frequency dyscalculia acquires a certain degree of difficulty the accurate location of fault, needs research to be applicable to the signal analysis method that wind-powered electricity generation unit is diagnosed, and promotes the ability of wind-powered electricity generation unit intelligent diagnostics.Wind power equipment status monitoring adopts various measurements and supervision method exactly, and record and display device running status, make warning to abnormality, for the fault analysis of equipment provides data and information.Status information shows as vibration, noise, pressure, temperature, strain, acoustic emission, liquid sample, electromagnetic field of this plant equipment etc., but the reflection degree of the state of different status information types to wind power equipment is different, even some information type is the status flag that can not reflect some equipment.Therefore the selection of information type is very crucial to the fault diagnosis of wind power equipment.The more reliable information of acquisition that develops into of measuring technology and sensor performance provides the foundation.The research of the extracting method of status flag is the key of wind power equipment fault diagnosis technology.The feature extraction that contains the plant equipment operating state signal that enriches status information is to be based upon on the basis of the processing of signal.Traditional signal processing method comprises filtering technique, spectrum analysis.The conventional feature extracting method being based upon on signal processing basis has Model in Time Domain analysis, fast fourier transform, short time discrete Fourier transform, time series analysis, cepstral analysis, Winger distributional analysis, time frequency analysis, wavelet analysis, Hilbert-Huang transform analysis, rank than methods such as analysis of spectrum, higher-order spectrum analysis, fractal chaos analysis.The object of signal analysis and processing is by the processing to operating state signal, determines the characteristic quantity that can characterize well wind power equipment running status.The application of status signal processing new technology in fault diagnosis promoted the development of fault diagnosis technology greatly, from short time discrete Fourier transform to time frequency analysis, arrive wavelet analysis, Hilbert-Huang transform analysis, the Non-Stationary Analysis method of signal processing is improved gradually again.Data when Multielement statistical analysis method only needs acquisition system normally to move, these data are projected to low dimensional feature space from high-dimensional data space, reflect the integrated information of multiple variablees with a small amount of variable, there is powerful data mining ability, show one's talent with its unique advantage.

In engineering application, vibration signal is the coupling of various vibration signals, adopt analysis of vibration signal to carry out failure diagnosis information type many, magnitude variations scope is wide, people always extract a variety of vibration signal characteristics parameters and carry out the assessment of wind power equipment health status, but the state regularity that they reflect, susceptibility and not identical in cluster, the separability of model space, along with some variations can occur different operation operating modes.So the method for using multivariate statistical analysis on primitive character basis is such as principal component analysis technology is extracted better regularity, new feature that susceptibility is strong carries out effective equipment fault diagnosis and seems and be starved of.

Wind power equipment fault diagnosis is the information obtaining according to status monitoring, the architectural characteristic of bonding apparatus and operation information and all previous maintenance record, to occurring or contingent diagnosing malfunction, analysis and forecast, to determine classification, position, degree and the reason of fault, propose maintenance game, make equipment return to normal condition.The fault diagnosis technology of application of advanced, not only can find initial failure, avoids the generation of serious accident, can also fundamentally solve the problem that maintenance is not enough and surplus is keeped in repair in equipment periodic maintenance.Therefore, how to ensure the safe and reliable efficient operation of wind power equipment, prevention key equipment breaks down, and stops great and generation catastrophic failure, has become one of important subject of current science and technology and wind energy development.

Summary of the invention

The object of the present invention is to provide a kind of wind-powered electricity generation unit mechanical equipment state diagnostic method based on multivariate statistical analysis, can improve the reliability of wind power equipment and the management level of maintainability and equipment, guarantee product quality, avoid the generation of major accident.

For achieving the above object, the technical solution adopted in the present invention is: a kind of wind-powered electricity generation unit mechanical equipment state diagnostic method based on multivariate statistical analysis, application sensors gathers the status information that wind-powered electricity generation unit plant equipment produces, described status information is carried out to feature extraction with multivariate statistical analysis, signal analysis, the analyzing and processing such as state identification, for multivariate statistics characteristic evaluating and On The Choice, take principal component analysis (PCA) technology as representative, express and the wind-powered electricity generation unit machine performance of classifying with low-dimensional principal component character representation technology, set up a kind of average degree of correlation rule and assessed each principal component and depict the ability of wind-powered electricity generation unit machine performance, and chosen low-dimensional principal component character representation comprehensive to wind power equipment status flag, the regularity having had and strong susceptibility.Described multivariate statistical analysis comprises that principal component analysis, independent component analysis, the main component analysis of core and blind source separate.

Technical solution of the present invention is the applied research in the diagnosis of wind-powered electricity generation unit mechanical equipment state based on multivariate statistical analysis, just separates (BSS) four kinds of Multivariates with blind source in developing principal component analysis (PCA), independent component analysis (ICA), core principal component analysis (KPCA) and be applied in this field by analyzing, research is in this respect unified in to higher-order statistics extraction, the fusion of polynary redundancy feature, multidimensional measure signal and separates under these three subsystems.Higher-order statistics extracts the higher-order statistics that can extract multidimensional or one-dimensional measurement signal, and this statistical information has embodied the essential attribute of wind-powered electricity generation unit mechanical equipment state pattern, embodiment device running status that can be more responsive than original measurement signal.Polynary redundancy feature merges can extract the statistical framework making new advances from polynary primitive character, and these new features are than primitive character is more sane, cluster is better, has realized the more Precise Diagnosis to wind-powered electricity generation unit plant equipment running status.Multidimensional measure signal separate can be from multidimensional measure signal separation reflect the signal content of certain critical component, can judge more easily accordingly the failure condition of these parts, this is a kind of new wind-powered electricity generation unit mechanical equipment state analysis approach in fact.The vibration signal of wind-powered electricity generation unit plant equipment has always shown the mixing of hyperchannel transmission of signal, and the signal that each sensor obtains not is the actual vibration signal of surveyed location component, but the coupling of multiple vibration source signals.Mixed signal is separated and analyzing and processing and expressing and the wind-powered electricity generation unit machine performance pattern of classifying extracting more effectively new statistical nature in multi-dimensional signal by Multielement statistical analysis methods such as principal component analysis (PCA), independent component analysis (ICA) and core principal component analysiss (KPCA), thereby realize the fault diagnosis to wind power equipment state.

Described status information is vibration information, noise information and the rotary speed information of wind-powered electricity generation unit plant equipment.

Described status information is the bearing of wind-powered electricity generation unit plant equipment and vibration information and the rotary speed information of gearbox parts.

The absolute mean of described vibration information involving vibrations, peak-peak, effective value, root amplitude, gradient variance, kurtosis, shape factor, peak factor.

Described principal component analysis comprises the wind power equipment status information based on low-dimensional principal component character representation, set up a kind of average degree of correlation rule and assess each principal component and depict the ability of wind-powered electricity generation unit machine performance, and chosen low-dimensional principal component character representation comprehensive to wind power equipment status flag.

Described vibration information is gathered by vibration transducer.

Technical solution of the present invention can be found wind power equipment initial failure accurate failure judgement situation, improve the availability of wind-powered electricity generation unit, reduce as much as possible cycle and the financial cost of care and maintenance, can also fundamentally solve the problem that maintenance is not enough and surplus is keeped in repair in equipment periodic maintenance.Safe and reliable efficient operation that can support equipment, prevention key equipment breaks down, and stops great and generation catastrophic failure.

Accompanying drawing explanation

Fig. 1 is based on low-dimensional principal component character representation wind power equipment status fault diagnosis scheme.

Similarity distance between test sample book and three state class that Fig. 2 calculates with one-dimensional characteristic: (a) normal condition (b) mild wear (c) catastrophic failure state.

Embodiment

Below in conjunction with embodiment, the present invention is made to further explaination:

A wind power equipment status fault diagnostic method based on multivariate statistical analysis, is characterized in that, comprises that principal component analysis, independent component analysis, the main component analysis of core and blind source separate.

Multivariate statistics feature has reflected the essential statistical framework of measured data, these structures can be portrayed the pattern of wind-powered electricity generation unit plant equipment effectively, multivariate statistical analysis analytical approach is a lot, comprise the methods such as principal component analysis (PCA), independent component analysis (ICA), core principal component analysis (KPCA), these analytical approachs have all demonstrated good effect in the diagnosis of wind power equipment status fault.

Wind-powered electricity generation unit mechanical equipment vibration or noise signal are a kind of stochastic processes of complexity, are difficult to represent with definite function of time.While carrying out the monitoring of wind-powered electricity generation unit machine performance and fault diagnosis, need to carry out signature analysis, extraction can reflect the pattern feature of wind-powered electricity generation unit mechanical equipment state information.In order to eliminate running environment, the impact of sensor factor on data, make signal analysis result have objective, a just standard, must carry out pre-service to signal-under-test.Therefore adopt Mean-Variance Standardization Act, following surface analysis, pretreated signal has the standard of zero-mean and unit variance.

The signal of the wind-powered electricity generation unit mechanical equipment state that gathers converts one-dimensional discrete data { y to ₁, y ₂, Λ, y _n, N is sample length, surveys data sample average as shown in the formula (1):

$\stackrel{\‾}{y}=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{y}_i---\left(1\right)$

The variance of sample is as shown in the formula (2):

$S=\frac{1}{N}\left(\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i-\stackrel{\‾}{y})}^2\right)---\left(2\right)$

Show that sample data after Mean-Variance standardization is as shown in the formula (3):

${y_i}^{\′}=\frac{y_i-\stackrel{\‾}{y}}{\sqrt{S}}---\left(3\right)$

After pre-service, be subject to the impact of neighbourhood noise, the wind-powered electricity generation unit mechanical equipment state signal of different conditions generally in time domain difference little, temporal analysis is generally as the preliminary anticipation of mechanical fault, only adopts Time-domain Statistics feature to be difficult to distinguish different state models.Frequency-domain analysis has amplitude and the phase information of former frequency of time domain signal composition.Energy distribution on frequency domain can reflect the difference between state to a certain extent, by analyzing the performance of different conditions signal on frequency domain, can find some characteristic feature frequency bands, has reflected the development of wind power equipment fault in various degree.Therefore based on time frequency analysis and frequency-domain analysis, extract time domain and the frequency domain statistical nature index of unit device signal and carry out its pattern feature of comprehensive representation, the temporal signatures such as the absolute mean of these feature involving vibrations, peak-peak, effective value, root amplitude, gradient variance, kurtosis, shape factor, peak factor, and 8 relative spectrum energy parameters of frequency domain.So altogether from measuring-signal, extract 16 features, be expressed as eigenvector

{F _j|j＝1，2，A，16} (4)

Principal component analysis (PCA) is a kind of effectively decorrelation instrument.Its basic thought is that some principal components of finding high dimensional data represent, these components have maximum variance, represent that with them former data have minimum square error.Principal component analysis feature has reflected the second-order statistics structure between diverse characteristics.

Independent component analysis (ICA) more enters one deck than principal component analysis (PCA), and it does not only require decorrelation, but also requires each variable mutual statistical independent.ICA can carry out effective blind source and separate, and also can be used for extracting the absolute construction feature of polytomy variable.

Core principal component analysis (KPCA) is a kind of nonlinear stretch of principal component analysis (PCA), and the mode of its ingenious utilization kernel function has realized the nonlinear characteristic structure extraction of PCA.On core pivot element analysis (KPCA) theoretical essence, by nonlinear transformation, the input space is transformed to higher dimensional space exactly, then carry out pivot analysis at the feature space of higher-dimension.

No matter second-order statistics structure, statistics absolute construction, or nonlinear organization, all reflected the essential distinction between wind-powered electricity generation unit plant equipment running status from the statistical significance.Each primary statistics characteristic index of plant equipment has also all reflected the essential statistical nature of unit equipment state, just they have reflected the not susceptibility of ipsilateral, obviously comprehensive these not the multivariate statistics feature structure of ipsilateral susceptibility can effectively characterize the sensitive mode feature of wind-powered electricity generation unit plant equipment.In the time of wind-powered electricity generation unit machine performance fault diagnosis, these statistical frameworks, compared with primary statistics feature, can be realized sensitiveer diagnostic result on very low dimension.

The ability of the general multivariate statistics feature representation wind-powered electricity generation unit equipment mode extracting is not identical yet, for example, the first dimension principal component character representation has maximum eigenwert, have the maximum variance of data variation, thereby can depict the typical module of primitive character, other principal component features can only reflect the local feature of wind-powered electricity generation unit plant equipment pattern; Isolated component feature has different absolute construction, also all different to the expressive ability of wind-powered electricity generation unit mechanical equipment state; Each non-linear principal component feature also has different nonlinear characteristics.For can effectively expressing and sorting device fault mode can reduce again computation complexity, can only select most representative characteristic component to set up a sub spaces.A kind of traditional feature selection approach is accumulation contribution rate method (ACR), and to front m principal component feature, its contribution rate of accumulative total is defined as:

$R_m={\mathrm{\Σ}}_{i=1}^m{\mathrm{\λ}}_i/{\mathrm{\Σ}}_{i=1}^d{\mathrm{\λ}}_i---\left(5\right)$

R _mrepresent the number percent of front m principal component feature in the whole variances of observational characteristic.Choosing by threshold value of contribution rate of accumulative total setting of number of features m determined.But the theory of this threshold value is determined and is still acquired a certain degree of difficulty, and is all generally determine according to experiment value.The degree of correlation based on multivariate statistical analysis and primary statistics feature is measured and is proposed a kind of characteristic evaluating and select new method.To adopting principal component analysis (PCA) to carry out the method theoretical description, this can also have reference to other multivariate statistical methods below.

If PC _i(i=1, Λ, d) is i principal component character representation of wind-powered electricity generation unit plant equipment sample of signal, F _j(j=1, Λ, d) is j dimension raw mode proper vector, and the degree of correlation between them is calculated with the absolute value of related coefficient:

${\mathrm{\ρ}}_{\mathrm{ij}}=\left|\frac{E\left\{\right[{\mathrm{PC}}_i-E\left({\mathrm{PC}}_i\right)\left]\right[F_j-E\left(F_j\right)\left]\right\}}{\sqrt{D\left({\mathrm{PC}}_I\right)}\·\sqrt{D\left(F_j\right)}}\right|---\left(6\right)$

Wherein E () represents to ask expectation value, and D () represents to ask variance.Definition correlation matrix ρ is wherein ρ _ifor associated vector, be defined as { ρ _i1, ρ _i2, Λ, ρ _id, it has reflected PC _idegree of association power with all primary statistics features.ρ _ijlarger, i principal component character representation is just more similar to i dimension statistical nature.In addition correlation coefficient ρ, _ijreact PC _iextract the degree of wind-powered electricity generation unit machine performance change information from j dimension statistical nature.Therefore also to may be thought of as be the weight of extracting new feature from primary statistics feature to above-mentioned coefficient.According to the feature of PCA technology, principal component above has larger variance, this show they than principal Component Extraction below more useful wind-powered electricity generation unit plant equipment pattern information.This phenomenon can be used average correlation

describe, be defined as follows:

$\stackrel{\‾}{\mathrm{\ρ}}=\frac{1}{d}\underset{j=1}{\overset{d}{\mathrm{\Σ}}}{\mathrm{\ρ}}_{\mathrm{ij}}---\left(7\right)$

reflect that i principal component character representation portray the ability of wind-powered electricity generation unit plant equipment pattern.

larger, principal component PC _iwill express and the mechanical equipment state of classifying goodly.Conventionally maximum, first principal component is expressed and is had best performance in other words.In addition, the average degree of correlation also can be by comparing ρ _iin correlation, for selecting responsive primitive character.Correlation coefficient ρ _ijlarger, j dimension statistical nature F _jjust more responsive.

During to wind-powered electricity generation unit machine performance fault diagnosis, a controversial problem selects how many principal component features to catch better the variation of unit plant equipment health status.More character representation can be useful to wind-powered electricity generation unit machine performance fault diagnosis in theory.But adopt too many feature can increase computation complexity, and the most responsive feature is just more conducive to pattern classification.Therefore, should from whole principal component features, select the principal component of proper number.Can well portray the performance of principal component feature in view of the average degree of correlation, can decide optimum intrinsic dimensionality with it.The following on average relevant rule (MCR) of principal component of selecting is proposed:

$M_C>{\mathrm{\ρ}}_{\mathrm{th}}^d---\left(8\right)$

Wherein M _crepresent the average relevance degree of selecting, it is a default thresholding system.The principal component that only has average relevance degree to be greater than this thresholding system is chosen as representational principal component, other abandon.But MCR method is used need to solve default thresholding problem.Good thresholding should have two advantages: stability and validity.

The pattern-recognition when essential problem of wind power equipment status fault diagnosis.By analyzing the sign ability of principal component feature to wind-powered electricity generation unit mechanical equipment state, select effective low-dimensional principal component character representation with this, propose following wind power equipment status fault diagnosis scheme.

Suppose that the dimension of selecting is 1, training sample x _j1 dimensional pattern in primitive character space is expressed as { f _1j, Λ, f _ij.1 dimension PC of all training samples expresses and can be expressed as wherein

be i PC eigenvector, length is M.Suppose that training sample set is categorized into c class { C _i| i=1, Λ, c}, be expressed as (xn, cn) | n=1, Λ, Ni}, wherein cn represents the class label of pattern xn, N; It is the number of samples of class Cj., to each class, the average of the PC feature representation of training sample is:

${\mathrm{MEAN}}_i^{\mathrm{train}}=\{\frac{1}{N}\underset{x_i\∈C_k}{\mathrm{\Σ}}{f}_{1f},\mathrm{\Λ},\frac{1}{N}\underset{x_i\∈C_k}{\mathrm{\Σ}}{f}_{\mathrm{lf}\}---\left(9\right)}$

These averages can be for expressing corresponding class, and appropriate fault category is assigned to in a sample area for input.Countermeasure is that sample also extracts corresponding 1 dimension principal component character representation, and it projects to training sample set mode matrix R by test sample book _xfront 1 eigenvalue of maximum characteristic of correspondence vector on obtain, they are expressed as $F_k^{\mathrm{test}}=\{{\mathrm{<figure-callout\; id="1"\; label="f\; k"\; filenames="HSA00000809122700012.png"\; state="\{\{state\}\}">f</figure-callout>}}_k,\mathrm{\&Lambda;},f_{\mathrm{kl}}\},$ Be calculated as follows:

$f_{\mathrm{kl}}=v_i^T(x_k-\stackrel{\&OverBar;}{x}),i=1,\mathrm{\&Lambda;},l---\left(10\right)$

Wherein

the mean vector of training sample, therefore come compare test sample and of all categories between similarity, with principal component feature and of all categories between Euclidean distance calculate, be:

${\mathrm{\&delta;}}_{\mathrm{kj}}\left(x\right)=\left|\right|F_k^{\mathrm{test}}-{\mathrm{MEAN}}_l^{\mathrm{train}}\left|\right|---\left(11\right)$

The relatively size of these distances, test sample book is assigned to the class with minor increment, with apart from minimum state class has maximum similarity.

Based on above analysis, as shown in Figure 1, its main overview of steps is as follows for the wind power equipment condition diagnosing scheme of proposition based on low-dimensional principal component character representation:

(1) f of computation and measurement sample dimension statistical nature, the data matrix X of tissue training's sample set.

(2) f that extracts training sample ties up PC character representation, extracts test sample book 1 and ties up PC character representation.

(3) based on propose MCR method with training sample select optimum dimension 1 (1≤f).

(4) the 1 dimension PC character representation based on training sample compute classes average.

(5) calculate test sample book x _kto the similarity distance of each class average, test sample book is categorized into the class with minimum similarity distance.

The corresponding similarity distance of certain test sample book while selecting as shown in Figure 2 one dimension principal component.Can find out, select effective principal component feature and set up a kind of average degree of correlation rule to assess each principal component and can effectively depict the ability of wind-powered electricity generation unit machine performance.

Described fault diagnosis is that bearing to wind power equipment and vibration information and the rotary speed information of gearbox parts carry out fault diagnosis.

Described vibration information and rotary speed information, gathered by vibration transducer and speed probe.

Technical solution of the present invention is the applied research in wind-powered electricity generation unit equipment state fault diagnosis based on multivariate statistical analysis, just separates four kinds of Multivariates such as (BSS) with blind source in developing principal component analysis (PCA), independent component analysis (ICA), core principal component analysis (KPCA) and be applied in this field by analyzing, research is in this respect unified in to higher-order statistics extraction, the fusion of polynary redundancy feature, multidimensional measure signal and separates under these three subsystems.Higher-order statistics extracts the higher-order statistics that can extract multidimensional or one-dimensional measurement signal, and this statistical information has embodied the essential attribute of wind-powered electricity generation unit mechanical equipment state pattern, embodiment device running status that can be more responsive than original measurement signal.Polynary redundancy feature merges can extract the statistical framework making new advances from polynary primitive character, and these new features are than primitive character is more sane, cluster is better, has realized the more Precise Diagnosis to wind-powered electricity generation unit plant equipment running status.Multidimensional measure signal separate can be from multidimensional measure signal sub-department reflect the signal content of certain critical component, can judge more easily accordingly the failure condition of these parts, this is a kind of new wind-powered electricity generation unit mechanical equipment state analysis approach in fact.The vibration signal of wind power equipment has always shown the mixing of hyperchannel transmission of signal, and the signal that each sensor obtains not is the actual vibration signal of surveyed location component, but the coupling of multiple vibration source signals.Express and the wind-powered electricity generation unit machine performance pattern of classifying to mixed signal separation and analyzing and processing and to extracting more effectively new statistical nature in multi-dimensional signal by Multielement statistical analysis methods such as principal component analysis (PCA), independent component analysis (ICA), core principal component analysiss (KPCA), thereby realize the fault diagnosis to wind power equipment state.

Technical solution of the present invention can be found wind power equipment initial failure accurate failure judgement situation, improve the availability of wind-powered electricity generation unit, reduce as much as possible cycle and the financial cost of care and maintenance, can also fundamentally solve the problem that maintenance is not enough and surplus is keeped in repair in equipment periodic maintenance.Can ensure the safe and reliable efficient operation of wind power equipment, prevention key equipment breaks down, and stops great and generation catastrophic failure.

Claims (6)

1. the wind-powered electricity generation unit mechanical equipment state diagnostic method based on multivariate statistical analysis, it is characterized in that, application sensors gathers the status information that wind-powered electricity generation unit plant equipment produces, with multivariate statistical analysis, described status information is processed, described multivariate statistical analysis comprises that principal component analysis, independent component analysis, the main component analysis of core and blind source separate; Described multivariate statistical analysis process is for to carry out feature extraction, signal analysis, state identification to described status information.

2. the wind power equipment method for diagnosing status based on multivariate statistical analysis as claimed in claim 1, is characterized in that, described status information is vibration information, noise information and the rotary speed information of wind-powered electricity generation unit plant equipment.

3. the wind power equipment method for diagnosing status based on multivariate statistical analysis as claimed in claim 1, is characterized in that, described status information is the bearing of wind-powered electricity generation unit plant equipment and vibration information and the rotary speed information of gearbox parts.

4. the wind power equipment method for diagnosing status based on multivariate statistical analysis as claimed in claim 2, it is characterized in that the absolute mean of described vibration information involving vibrations, peak-peak, effective value, root amplitude, gradient variance, kurtosis, shape factor, peak factor.

5. the wind power equipment method for diagnosing status based on multivariate statistical analysis as claimed in claim 1, it is characterized in that, described principal component analysis comprises the wind power equipment status information based on low-dimensional principal component character representation, set up a kind of average degree of correlation rule and assess each principal component and depict the ability of wind-powered electricity generation unit machine performance, and chosen low-dimensional principal component character representation comprehensive to wind power equipment status flag.

6. the wind power equipment method for diagnosing status based on multivariate statistical analysis as claimed in claim 2, is characterized in that described vibration information is gathered by vibration transducer.

Images