CN102944416A - Multi-sensor signal fusion technology-based fault diagnosis method for wind turbine blades - Google Patents

Abstract

The invention discloses a multi-sensor signal fusion technology-based fault diagnosis method for wind turbine blades. According to the method, the problems of lack of fault information and the like caused by the insufficiency of sensors is solved by adopting a plurality of sensors. An independent classifier is used for performing primary diagnosis on information acquired by each sensor, so as to determine the possibility that to-be-diagnosed faults belong to different faults; and the decision fusion diagnosis is performed by adopting a fuzzy integral fusion technology based on that the importance degree of information output by each classifier is adequately considered. According to the fault diagnosis method disclosed by the invention, classified results of all the classifiers are integrated, and the importance degree of each classifier is considered, thus effectively improving the accuracy of the fault diagnosis on the wind turbine blades.

Description

Wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology

Technical field

The invention belongs to the on-line monitoring and fault diagonosing technical field, especially a kind of method of the wind power generation unit blade fault diagnosis based on the multiple sensor signals integration technology.

Background technology

In recent years, since the deterioration of the shortage of resource and environment to make countries in the world begin to pay attention to development and utilization renewable and without the energy of discharging.Wind resource more and more obtains people's attention as the resource of a kind of green, environmental protection.In the world, the operation of a large amount of wind-powered electricity generation units is so that the safe and stable operation of wind-powered electricity generation unit causes showing great attention to of people.Because the wind turbine group leader phase is operated in the field, be exposed to the sun and the rugged surroundings such as thunderstorm in, the wind field wind regime is complicated and changeable, very easily causes various faults, therefore, the on-line monitoring and fault diagonosing of wind-powered electricity generation unit has become requisite link.The blade fault type of wind-powered electricity generation unit comprises leaf quality imbalance fault, the pneumatic imbalance of blade, driftage and disconnected blade etc., because wind power generation unit blade is expensive, difficult in maintenance after damaging, therefore the condition monitoring and fault diagnosis of blade seemed particularly important.In time find fault at the initial stage that blade fault occurs, before problem worse affects unit operation, in time process, can greatly reduce blade maintenance, upkeep cost and difficulty.

In the process of wind power generation unit blade fault diagnosis, the data of processing all collect by sensor.Because the diagnosis object operating condition is complicated, influence factor is numerous, the same fault often has different performances, the same symptom usually is again the coefficient result of several faults, in fact strict, between detection limit and the fault signature, all be a kind of Nonlinear Mapping between fault signature and the source of trouble, the fault signature amount that only relies on single-sensor to obtain generally can't be finished fault diagnosis effectively, and one of effective means that addresses the above problem just adopts the multiple sensor signals integration technology.

The mode of information fusion is generally in sensor layer, characteristic layer and decision-making level, and what often use is to merge in decision-making level.The information fusion technology of decision-making level is two or more sorters to be carried out integrated, adopts certain blending algorithm to diagnose.Existing blending algorithm mainly contains bayes method, D-S evidential reasoning method and fuzzy integral method.Bayes method needs prior imformation, and this prior imformation often is difficult to obtain in actual applications; And require the decision-making Element of a set separate, this requires too harsh.D-S evidential reasoning bill requires employed evidence necessary separate, generally is difficult to satisfy, and in addition, also shot array, events conflict etc. can occur.Fuzzy set can be described indeterminacy phenomenon well, and therefore the fusion method based on fuzzy integrals theory is a kind of instrument that is most widely used.Fuzzy integral is a kind of non-linear Decision fusion method based on fog-density, and integral process is the classification results of comprehensive each sorter not only, also considers the significance level of each sorter, and fuzzy integral is applied to fault diagnosis, can reach accurate location fault.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, provide a kind of simple, cost is low, can Effective Raise wind power generation unit blade safety, the wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology of reliability.

The technical solution used in the present invention is: a kind of wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology, at the wind-powered electricity generation unit multisensor is installed, the failure message shortcoming problem of bringing owing to sensor is not enough by adopting multisensor to solve, adopt sorter that the information of each sensor collection is carried out tentative diagnosis, determine that fault to be diagnosed is under the jurisdiction of the possibility of different faults, on the basis of the correlation degree that takes into full account each sorter and different faults type, adopt the fuzzy integral fusion method to carry out the Decision fusion diagnosis, the concrete steps of its diagnostic method are:

(1) measures the vibration signal of blade under normal and typical fault type by the acceleration transducer that is installed in wind generator set main shaft seat horizontal direction and vertical direction, the corresponding sorter of each acceleration transducer;

(2) utilize empirical mode decomposition respectively the horizontal direction of collection and the vibration signal on the vertical direction to be decomposed, different fault signatures is reacted to different intrinsic mode functions;

(3) energy of several intrinsic mode functions before the calculating forms the characteristic information that reacts fault, and the fault characteristic information normalized to extracting obtains fault feature vector, as training sample and test sample book;

(4) utilize training sample and test sample book on horizontal direction and the vertical direction respectively two sorters to be carried out training and testing;

(5) determine fog-density and fuzzy mearue; $g_j^i=g_j\left(\right\{y_i\left\}\right),i=\mathrm{1,2},\·\·\·c,j=\mathrm{1,2},\·\·\·t,$

The fog-density that represents j information of i sorter, t are the numbers of fault type, and c is the number of sorter; Make Y={y ₁, y ₂... y _cBe the set that c sorter consists of, A _i={ y ₁, y ₂... y _i, utilize each sorter in test to the correct recognition rata of each fault as the correlation degree of this sorter to each fault type, namely then fog-density determines fuzzy mearue according to fog-density;

(6) vibration signals measured is carried out empirical mode decomposition and extract fault feature vector, utilize sorter to carry out respectively the primary fault diagnosis, the result who obtains is p _j=(p _j(y ₁), p _j(y ₂) ... p _j(y _c)), p wherein _j(y _i) presentation class device y _iThe possibility that example to be diagnosed is divided into the j class;

(7) utilize the Choquet fuzzy integral to do the fusion operator primary fault diagnostic result is carried out the fault type that fusion treatment is determined wind power generation unit blade; According to formula

Calculate fuzzy integral value e _j, e _jBe the likelihood of failure index that comprehensive diagnos goes out, then forming fault may index set E={e ₁, e ₂... e _t, according to its failure judgement type, the corresponding classification of maximal value is the fault type of this example among the E.

As preferably, the concrete account form of described step (5) fuzzy mearue is:

According to formula

Determine λ _j, then according to formula g _j(A ₁)=g _j({ y ₁) and formula g _j(A _i)=g _j({ y _i)+g _j(A _I-1)+λ _jg _j( _yI}) g _j(A _I-1), i=1,2 ... c asks for fuzzy mearue g _j(A _i); λ _jIt is an intermediate variable.

As preferably, in the described step (6), the likelihood probability of support vector machine output is under the jurisdiction of the possibility of certain fault as fault to be diagnosed in the primary diagnosis, and with its result as the primary fault diagnosis, likelihood probability is calculated as follows

$p_j\left(y_i\right)=\left\{\begin{array}{cc}\frac{1}{1+\exp \left(\ln \left(\left(\mathrm{\ϵ}\right)f_j\left(i\right)\right)\right)}& \left|f_j\left(i\right)\right|\≤1\\ 1& \left|f_j\left(i\right)\right|>1\end{array}\right.$

In the formula: f _j(i) be the output of j support vector machine in i the sorter; p _j(y _i) likelihood probability of j fault in i sorter of expression, likelihood probability calculating is carried out for just identifying the sample class; Work as f _j(i)〉0 o'clock, it was p that sample is assigned to the degree of membership that just identifies class _j(y _i) (p _j(y _i)＞0.5); Work as f _j(i)＜0 o'clock, it is p that sample is assigned to the degree of membership that just identifies class _j(y _i) (p _j(y _i)＜0.5); ε is arbitrarily small positive integer.

Beneficial effect: 1, the present invention adopts based on empirical mode decomposition and extracts fault feature vector, can improve the resolution of fault.

2, the present invention adopts the not only classification results of comprehensive each sorter of Fuzzy Integral Fusion, has also considered the significance level of each sorter, Effective Raise the accuracy of system diagnostics.

3, the present invention can carry out the wind power generation unit blade localization of fault exactly, shortens maintenance and searches the time, improves the efficient of maintenance maintenance.

4, the present invention is simple, the diagnosis cost is low, be a kind of can Effective Raise wind power generation unit blade safety, the wind power generation unit blade method for diagnosing faults of reliability.

Description of drawings

Fig. 1 is based on the block scheme of the wind power generation unit blade fault diagnosis of multiple sensor signals integration technology among the present invention;

Fig. 2 is apoplexy group of motors synoptic diagram of the present invention;

Fig. 3 is empirical mode decomposition block scheme among the present invention;

Fig. 4 is the block scheme that extracts based on the fault feature vector of empirical mode decomposition among the present invention.

Embodiment

The present invention will be further described below in conjunction with the drawings and specific embodiments.

As shown in Figure 1, a kind of wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology, at the wind-powered electricity generation unit multisensor is installed, the failure message shortcoming problem of bringing owing to sensor is not enough by adopting multisensor to solve, adopt sorter that the information of each sensor collection is carried out tentative diagnosis, determine that fault to be diagnosed is under the jurisdiction of the possibility of different faults, on the basis of the correlation degree that takes into full account each sorter and different faults type, adopt the fuzzy integral fusion method to carry out the Decision fusion diagnosis, concrete steps are as follows:

(1) as shown in Figure 2, the wind wheel 1 of wind-powered electricity generation unit is connected with main shaft, main shaft is installed in the spindle drum 2, main shaft is connected with generator 4 by shaft coupling 3, measure the vibration signal of blade under normal and typical fault type by two acceleration transducers that are installed in wind generator set main shaft seat 2 horizontal directions and vertical direction, the corresponding sorter of each acceleration transducer.

(2) utilize empirical mode decomposition (EMD) respectively the horizontal direction of collection and the vibration signal on the vertical direction to be decomposed, different fault signatures is reacted to different intrinsic mode functions (IMF).

Empirical mode decomposition is a kind of method of processing non-stationary, nonlinear properties.The method is decomposed into intrinsic mode functions (IMF) and a residual volume sum of a plurality of different frequency ranges with burst, can give prominence to well the local feature of signal, and its computation process as shown in Figure 3.After the calculating through series, original signal x (t) can decompose as follows

$x\left(t\right)={\mathrm{\Σ}}_{i=1}^n{c}_i\left(t\right)+r_n\left(t\right)---\left(1\right)$

Therefore, can resolve into n IMF and a remaining component sum, intrinsic mode functions c to any one signal x (t) ₁(t), c ₂(t), c ₃(t) ..., c _n(t) distinguish the from high to low composition of different frequency range of representation signal, the frequency content that each frequency range comprises is different, and can change remaining function r along with the variation of vibration signal x (t) _n(t) average tendency of representation signal.

(3) energy of several IMF before the calculating forms the characteristic signal that reacts fault, and the fault characteristic information normalized to extracting obtains fault feature vector, as training sample and test sample book.

Fig. 4 is that it is specially based on the block scheme of the fault feature vector extraction of empirical mode decomposition:

Step 1: the vibration signal sequence is carried out EMD decompose, obtain the IMF component, the IMF component number of different vibration signals is different, and m IMF component is as research object before selecting.

Step 2: the energy of m IMF before calculating:

$E_i={\∫}_{-\∞}^{+\∞}{\left|c_i\left(t\right)\right|}^2\mathrm{dt}$ (i=1,2…,m)(2)

Step 3: the structure of proper vector:

T＝[E ₁，E ₂，…E _m](3)

Because the energy of some IMF is larger, for the ease of analyzing and processing T is carried out normalization.Proper vector is

T′＝[E ₁/E，E ₂/E，…E _m/E](4)

In the formula:

(4) utilize training sample and test sample book on horizontal direction and the vertical direction respectively two sorters to be carried out training and testing.

SVM asks the optimal classification face to propose under the linear separability situation.So-called optimal classification lineoid is exactly that requirement classification plane not only can be faultless separately with two class samples, and will makes the cluster between two classes maximum.For two classes classification situation, its objective function is:

$\max W\left(\mathrm{\α}\right)=-\frac{1}{2}\underset{i=1}{\overset{1}{\mathrm{\Σ}}}\underset{j=1}{\overset{1}{\mathrm{\Σ}}}{\mathrm{\α}}_i{\mathrm{\α}}_j{y}_i{y}_{j}K(x_i,x_j)+\underset{i=1}{\overset{1}{\mathrm{\Σ}}}{\mathrm{\α}}_i$

$s.t.{\mathrm{\&Sigma;}}_{i=1}^1{\mathrm{<figure-callout\; id="1"\; label="y"\; filenames="HDA00002534776200021.png,HDA00002534776200031.png"\; state="\{\{state\}\}">y</figure-callout>}}_1{\mathrm{\&alpha;}}_i=0---\left(5\right)$

0≤α _i≤C，i＝1，2，…1

And decision function is

$f\left(x\right)=\mathrm{sign}({\mathrm{\&Sigma;}}_{i=1}^1{\mathrm{\&alpha;}}_i{y}_{i}K(x_i,x)+b)---\left(6\right)$

For the multicategory classification problem, adopt one-against-all Combination of Methods two class support vector machines structure multicategory classification device among the present invention, two sorters adopt identical combined method.Specific as follows:

To with a t class problem, the one-against-all method needs t two class support vector machines, namely adopt t classification lineoid to classify, with the tag location of i class+1, then the tag location of the sample that all the other are all-1 utilizes training sample and test sample book respectively sorter to be carried out training and testing.

(5) determine fog-density and fuzzy mearue. $g_j^i=g_j\left(\right\{y_i\left\}\right),i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;c,j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;t,$

The fog-density that represents j information of i sorter, t are the numbers of fault type, and c is the number of sorter.Make Y={y ₁, y ₂... y _cBe the set that c sorter consists of, A _i={ y ₁, y ₂... y _i, c is 2, utilize each sorter in test to the correct recognition rata of each fault as the correlation degree of this sorter to each fault type, namely then fog-density determines fuzzy mearue according to fog-density.

According to formula

Determine λ _j, then according to formula g _j(A ₁)=g _j({ y ₁) and formula g _j(A _i)=g _j({ y _i)+g _j(A _I-1)+λ _jg _j({ y _i) g _j(A _I-1), i=1,2 ... c asks for fuzzy mearue g _j(A _i).λ _jIt is an intermediate variable.

(6) vibration signals measured is carried out empirical mode decomposition and extract fault feature vector, utilize sorter to carry out respectively the primary fault diagnosis, the result who obtains is p _j=(p _j(y ₁), p _j(y ₂) ... p _j(y _c)), p wherein _j(y _i) presentation class device y _iExample to be diagnosed is divided into the possibility of j class, wherein

$p_j\left(y_i\right)=\left\{\begin{array}{cc}\frac{1}{1+\exp \left(\ln \left(\left(\mathrm{\&epsiv;}\right)f_j\left(i\right)\right)\right)}& \left|f_j\left(i\right)\right|\&le;1\\ 1& \left|f_j\left(i\right)\right|>1\end{array}\right.---\left(7\right)$

In the formula: f _j(i) be the output of j support vector machine in i the sorter; p _j(y _i) likelihood probability of j fault in i sorter of expression, likelihood probability calculating is carried out for just identifying the sample class.Work as f _j(i)＞0 o'clock, it is p that sample is assigned to the degree of membership that just identifies class _j(y _i) (p _j(y _i)＞0.5); Work as f _j(i)＜0 o'clock, it is p that sample is assigned to the degree of membership that just identifies class _j(y _i) (p _j(y _i)＜0.5); ε is arbitrarily small positive integer.

(7) utilize the Choquet fuzzy integral to do the fusion operator primary fault diagnostic result is carried out the fault type that fusion treatment is determined wind power generation unit blade.According to formula

Calculate fuzzy integral value e _i, e _jBe the likelihood of failure index that comprehensive diagnos goes out, then forming fault may index set E={e ₁, e ₂... e _t, according to its failure judgement type, the corresponding classification of maximal value is the fault type of this example among the E.

Should be pointed out that for those skilled in the art under the prerequisite that does not break away from the principle of the invention, can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.In the present embodiment not clear and definite each ingredient all available prior art realized.

Claims (3)

1. wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology, it is characterized in that: at the wind-powered electricity generation unit multisensor is installed, adopt sorter that the information of each sensor collection is carried out tentative diagnosis, determine that fault to be diagnosed is under the jurisdiction of the possibility of different faults, on the basis of the correlation degree that takes into full account each sorter and different faults type, adopt the fuzzy integral fusion method to carry out the Decision fusion diagnosis, the concrete steps of its diagnostic method are:

(1) measures the vibration signal of blade under normal and typical fault type by the acceleration transducer that is installed in wind generator set main shaft seat horizontal direction and vertical direction, the corresponding sorter of each acceleration transducer;

(2) utilize empirical mode decomposition respectively the horizontal direction of collection and the vibration signal on the vertical direction to be decomposed, different fault signatures is reacted to different intrinsic mode functions;

(3) energy of several intrinsic mode functions before the calculating forms the characteristic information that reacts fault, and the fault characteristic information normalized to extracting obtains fault feature vector, as training sample and test sample book;

(4) utilize training sample and test sample book on horizontal direction and the vertical direction respectively two sorters to be carried out training and testing;

(5) determine fog-density and fuzzy mearue; $g_j^i=g_j\left(\right\{y_i\left\}\right),i=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;c,j=\mathrm{1,2},\&CenterDot;\&CenterDot;\&CenterDot;t,$

The fog-density that represents j information of i sorter, t are the numbers of fault type, and c is the number of sorter; Make Y={y ₁, y ₂... y _cBe the set that c sorter consists of, A _i={ y ₁, y ₂... y _i, utilize each sorter in test to the correct recognition rata of each fault as the correlation degree of this sorter to each fault type, namely then fog-density determines fuzzy mearue according to fog-density;

(6) vibration signals measured is carried out empirical mode decomposition and extract fault feature vector, utilize sorter to carry out respectively the primary fault diagnosis, the result who obtains is p _j=(p _j(y ₁), p _j(y ₂) ... p _j(y _c)), p wherein _j(y _i) presentation class device y _iThe possibility that example to be diagnosed is divided into the j class;

(7) utilize the Choquet fuzzy integral to do the fusion operator primary fault diagnostic result is carried out the fault type that fusion treatment is determined wind power generation unit blade; According to formula

Calculate fuzzy integral value e _j, e _jBe the likelihood of failure index that comprehensive diagnos goes out, then forming fault may index set E={e ₁, e ₂... e _t, according to its failure judgement type, the corresponding classification of maximal value is the fault type of this example among the E.

2. the wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology according to claim 1, it is characterized in that: the concrete account form of described step (5) fuzzy mearue is:

According to formula

Determine λ _j, then according to formula g _j(A ₁)=g _j({ y ₁) and formula g _j(A _i)=g _j({ y _i)+g _j(A _I-1)+λ _ig _j({ y _i) g _j(A _I-1), i=1,2 ... c asks for fuzzy mearue g _j(A _i); λ _jIt is an intermediate variable.

3. the wind power generation unit blade method for diagnosing faults based on the multiple sensor signals integration technology according to claim 1, it is characterized in that: in the described step (6), the likelihood probability of support vector machine output is under the jurisdiction of the possibility of certain fault as fault to be diagnosed in the primary diagnosis, with its result as the primary fault diagnosis, likelihood probability is calculated as follows

$p_j\left(y_i\right)=\left\{\begin{array}{cc}\frac{1}{1+\exp \left(\ln \left(\left(\mathrm{\&epsiv;}\right)f_j\left(i\right)\right)\right)}& \left|f_j\left(i\right)\right|\&le;1\\ 1& \left|f_j\left(i\right)\right|>1\end{array}\right.$

In the formula: f _j(i) be the output of j support vector machine in i the sorter; p _j(y _i) likelihood probability of j fault in i sorter of expression, likelihood probability calculating is carried out for just identifying the sample class; Work as f _j(i)〉0 o'clock, it was p that sample is assigned to the degree of membership that just identifies class _j(y _i) (p _j(y _i)＞0.5); Work as f _j(i)＜0 o'clock, it is p that sample is assigned to the degree of membership that just identifies class _j(y _i) (p _j(y _i)＜0.5); ε is arbitrarily small positive integer.

Images