CN110985310A

CN110985310A - Wind driven generator blade fault monitoring method and device based on acoustic sensor array

Info

Publication number: CN110985310A
Application number: CN201911295468.7A
Authority: CN
Inventors: 胡晓宇; 代金良
Original assignee: Dalian Sailing Technology Co Ltd
Current assignee: Dalian Sailing Technology Co Ltd
Priority date: 2019-12-16
Filing date: 2019-12-16
Publication date: 2020-04-10
Anticipated expiration: 2039-12-16
Also published as: CN110985310B

Abstract

The invention relates to a method and equipment for monitoring the fault of a wind driven generator blade based on an acoustic sensor array, S1, deploying an acquisition array, and fixedly installing the acquisition array at the bottom of a support rod of the wind driven generator; s2, enhancing the blade sound signals picked up by the acquisition array; constructing a CNN fault judgment model; s3, positioning the fault position of the fault blade; s4, restraining the strong interference direction and returning to the step S3; and S5, positioning the wind blade with the fault after the fault position is positioned. The collecting array is arranged at the bottom of the wind driven generator, and the wind driven blade is subjected to non-contact sound signal collection, so that the arrangement, operation and maintenance costs can be effectively reduced.

Description

Wind driven generator blade fault monitoring method and device based on acoustic sensor array

Technical Field

The invention relates to the field of blade fault monitoring, in particular to a method and equipment for monitoring the fault of a wind driven generator blade based on an acoustic sensor array.

Background

Along with the development of society, people pay more and more attention to environmental pollution, the traditional power generation mode based on chemical energy (coal, petroleum and natural gas) is more and more limited, the nuclear power generation has the problem of large investment in the early stage, and the wind power generation and the solar power generation are more and more paid more attention. The wind power generation is renewable clean energy, has small environmental pollution and high automation degree, is easy to realize remote control, and is very suitable for solving the power supply requirements of regions with rare population and difficult access to a power grid, so the wind power generation has important economic and social benefits.

The blade is a key component of a wind generating set (hereinafter referred to as a "fan"), the energy generated by the fan is from the work done when the wind power pushes the blade to rotate, the stress of the blade changes in the process of rotating around a tower, particularly when the blade rotates from the upper part to the lower part, the stress changes alternately in the rotating process, the wind power condition has obvious instability, the factors can cause uneven stress of the blade and form vibration, and in addition, the blade is blown by wind, sun and rain, the material is aged and worn, and the physical damage can be brought to the blade, which can seriously affect the operating efficiency of the whole fan, so the monitoring of the working condition of the blade is very necessary.

However, in the past, an efficient blade working condition monitoring scheme is lacked in the industry, and a scheme of mounting a sensor on a blade is proposed in recent years, but the scheme has the problems of high deployment, operation and maintenance cost, low management efficiency and the like.

Disclosure of Invention

The invention aims to provide a method and equipment for monitoring the fault of a wind driven generator blade based on an acoustic sensor array, so as to solve the problems of high operation and maintenance cost and low management efficiency in monitoring the working condition of the blade.

The invention solves the technical problems through the following technical means:

the wind driven generator blade fault monitoring method based on the acoustic sensor array comprises a wind driven generator, wherein the wind driven generator comprises a plurality of wind driven blades, and the wind driven blades rotate around a tower, and the method further comprises the following steps:

s1, deploying an acquisition array, and fixedly mounting the acquisition array at the bottom of a support rod of the wind driven generator;

s2, enhancing the blade sound signals picked up by the acquisition array; constructing a CNN fault judgment model;

s3, positioning the fault position of the fault blade;

when the fault noise has a different frequency component than the normal operation noise, or the frequency of the fault noise is higher than a preset value a, or the energy of the fault noise is higher than that of the normal operation noise, performing step S31;

s31, determining a fault area, and determining a fault position through a fault azimuth angle;

when the fault noise is lower than the preset value a, performing step S32;

s32, judging whether each direction sound source is a fault sound source by using a CNN fault judgment model, and determining the fault position; if the energy of the sound source in a certain direction exceeds the threshold value B and the sound source is judged to be a non-fault sound source, the strong interference sound source exists in the direction, and the step S4 is executed, otherwise, the step S5 is directly executed;

s4, restraining the strong interference direction, and returning to the step S3;

and S5, positioning the wind blade with the fault after the fault position is positioned.

The collection array is deployed at the bottom of the wind driven generator, and the wind driven blade is subjected to non-contact sound signal collection, so that the deployment, operation and maintenance cost can be effectively reduced; meanwhile, the sound signals of the blades are enhanced, so that the pickup distance of the array and the robustness of environmental noise can be effectively increased.

As a further scheme of the invention: step S1 includes that the collection array is in the same plane as the plane of rotation of the wind blade; and meanwhile, when the perpendicular bisector of the line segment where the wind blade is located passes through the midpoint of the acquisition array, the acquisition array is parallel to the wind blade.

As a further scheme of the invention: said step S2 includes;

the wind blade is divided into m areas according to n meters as a unit, a direction is determined in each area by the acquisition array, m directions are obtained, and signal enhancement is carried out on the m directions.

As a further scheme of the invention: the CNN fault judgment model comprises a plurality of layers of convolution layers and a full-connection network which are sequentially connected in series, wherein after a time frequency distribution diagram is input to a first layer of convolution layer, the first layer of convolution layer is subjected to feature extraction, extracted features are input to a next layer of convolution layer for pooling, and dimensionality is reduced through pooling of the plurality of layers of convolution layers; after the last convolution layer is subjected to pooling, splitting and splicing a plurality of groups of finally output two-dimensional extraction features into vectors, and inputting the vectors into a full-connection network; processing and outputting a required result through the full-connection network;

the output of the CNN fault judgment model has at least 4 states, namely a normal state, a blade cracking fault, a large abrasion fault and other unknown faults, and the blade cracking fault, the large abrasion fault and the other unknown faults jointly form a fault state corresponding to four nodes output by the fully-connected network;

and identifying the wind blade sound signal picked up by the target direction area by using the CNN fault judgment model, and when the state output by the CNN fault judgment model is a fault state, determining that the current wind blade has a fault, namely the wind blade sound signal is a fault sound source, and if the output is normal, determining that the wind blade sound signal is a non-fault sound source.

As a further scheme of the invention: the step S31 includes the steps of,

respectively carrying out enhanced denoising processing on the selected m directions by the acquisition array receiving signals to obtain noises of m regions; comparing the noise obtained by enhancement with the statistical result of the normal working noise, and selecting the area with the largest difference as a suspected fault area; inputting the enhancement result of the suspected fault area into a CNN fault judgment model which is trained in advance for further verification; if the judgment result is that the fault exists, the area is a fault area;

after the fault area is determined, the fault position changes along a circular track along with the rotation of the wind blade, two rays are led out from the center of the acquisition array, one ray passes through the fault position, the other ray passes through the circle center of the rotating surface of the wind blade, the fault azimuth angle is formed at the moment, and when the fault azimuth angle is the largest, the fault position is determined.

As a further scheme of the invention: said step S32 includes; inputting the enhancement result of each direction into a trained CNN fault judgment model, and outputting the probability of generating faults;

if the maximum value of the fault probability in each output direction is larger than the threshold value, the fault exists, and the area with the maximum fault probability is regarded as a fault area;

and when the state output by the CNN fault judgment model is a fault state, the current blade is considered to have a fault, and the sound signal comprises fault noise and normal working noise.

And if the energy of the sound source in a certain direction exceeds the threshold value B and the sound source is judged to be a non-fault sound source, the direction is considered to have a strong interference sound source and the step S4 is executed, otherwise, the step S5 is directly executed.

As a further scheme of the invention: the suppression of the strong interference direction is to suppress the interference sound source in the direction by adopting a wave beam nulling technology.

As a further scheme of the invention: said step S5 includes;

s51, when the fault azimuth reaches the maximum, the fault azimuth is tangent to the circular track of the fault position; determining a faulty wind blade position;

and S52, calculating the rotating speed of the blade by recording the time interval of the wind blade with more than two faults passing through the position, thereby predicting the moment when the fault blade rotates vertically downwards, and lighting a fault indicator lamp on the acquisition equipment at the moment, namely, when the fault indicator lamp is lighted, the wind blade at the lowest position is the fault wind blade.

As a further scheme of the invention: further comprising step S6 and step S7;

s6, archiving abnormal state sound data, and establishing a blade fault type database;

s7, performing model increment training on the CNN fault judgment model; the incremental training has two modes, namely sample incremental training and output class incremental learning;

and (3) sample increment training: under the condition of ensuring the existing knowledge, extracting new knowledge through incremental learning of a CNN fault judgment model on a new sample, using the new sample to obtain a new sample, training the CNN fault judgment model, and realizing updating;

output category incremental learning: copying the CNN fault judgment model to obtain another model, wherein one model is not modified, the other model increases output to the original fault and the new fault, retraining to obtain a new model, and adjusting the LOSS function of the new model to obtain: LOSS1+ LOSS 2;

wherein LOSS1 refers to new class detection; LOSS2 refers to the difference in the output of data on the old model and the new model.

The monitoring equipment based on the wind driven generator blade fault monitoring method based on the acoustic sensor array comprises a wind driven generator, a sensor array and a control system, wherein the wind driven generator comprises a plurality of wind driven blades which rotate around a tower;

the deployment module is used for deploying the acquisition array and fixedly mounting the acquisition array at the bottom of a support rod of the wind driven generator;

the enhancement module is used for enhancing the blade sound signals picked up by the acquisition array and constructing a CNN fault judgment model;

the fault positioning module is used for positioning the fault position of the faulted blade;

the suppression module is used for suppressing the strong interference direction and then positioning the fault position after suppression;

and the blade positioning module is used for positioning the wind blade with the fault.

The invention has the advantages that:

1. the collection array is deployed at the bottom of the wind driven generator, and the wind driven blade is subjected to non-contact sound signal collection, so that the deployment, operation and maintenance cost can be effectively reduced; meanwhile, the sound signals of the blades are enhanced, the pickup distance of the array and the robustness of environmental noise can be effectively increased, and the CNN fault judgment model can realize the preparation identification and the fault type identification of the blade conditions, so that operation and maintenance workers can remotely obtain the specific condition of the fault without arriving at the site, and the operation and maintenance efficiency is improved.

2. The method and the device perform positioning through the fault azimuth, overcome the problem that workers need to check the fault position in a short distance in the traditional scheme, quickly position the fault blade, facilitate checking and greatly reduce the maintenance time.

Drawings

Fig. 1 is a schematic position diagram of a collecting array and a wind turbine provided in an embodiment of the present invention.

FIG. 2 provides a schematic illustration of the division of a wind blade into several regions according to an embodiment of the invention.

Fig. 3 is a structural diagram of a CNN fault determination model provided in an embodiment of the present invention.

Fig. 4 is a schematic diagram of a fault azimuth provided by an embodiment of the present invention.

Fig. 5 is a flow chart provided by an embodiment of the present invention.

Fig. 6 is a schematic structural diagram provided in the embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Example 1

Referring to fig. 1 and 5, fig. 1 is a schematic position diagram of a collecting array and a wind turbine provided in an embodiment of the present invention; FIG. 5 is a block flow diagram provided by an embodiment of the present invention; the method for monitoring the fault of the wind driven generator blade based on the acoustic sensor array comprises a wind driven generator, wherein the wind driven generator comprises a plurality of wind driven blades, the wind driven blades rotate around a tower, in the embodiment, three wind driven blades are provided, such as a wind driven blade 1, a wind driven blade 2 and a wind driven blade 3 in fig. 1, and the wind driven blade 1, the wind driven blade 2 and the wind driven blade 3 rotate around the tower, and the method further comprises the following steps:

s1, deploying a collecting array, and picking up a blade sound signal through the collecting array;

as shown in fig. 1, the collecting array is fixedly arranged at the bottom of a support rod of the wind driven generator through a bolt and is positioned in the same plane with the rotating surface of the wind driven blade; as shown in fig. 1, when the position of point a is the perpendicular bisector of the line segment where the wind blade is located passes through the midpoint of the collecting array, that is, the point a is passed through, so that the collecting array is parallel to the wind blade.

Further, the height of the collecting array is between 1/12 and 1/5 of the supporting rod, in this embodiment, the height of the collecting array is preferably 1/7 of the supporting rod; the acquisition array is a pickup array of a 16-microphone linear array.

When any one wind blade rotates to a certain position, the perpendicular bisector of the line segment where the wind blade is located passes through the point A, if the straight line where the collecting array is located is parallel to the wind blade, the best fault location accuracy of the tested sound source is achieved for the wind blade, therefore, in the embodiment, the collecting array is obliquely placed, a certain included angle ∠ 1 is formed between the collecting array and the horizontal line in the rotating surface of the wind blade, the midpoint of the line segment where the wind blade is located is set to be the point B, and the point of the center of the rotating surface of the wind blade is set to be the point O, so that ∠ 1 and ∠ BOA are the same, and the size of ∠ BOA can be determined through the lengths of BO and AO, and the collecting array is arranged.

When the wind driven generator works, the wind blades can be approximately regarded as straight lines, so that fault sound sources are only distributed in one dimension in a certain time interval, and the embodiment adopts a straight line acquisition array which is positioned on the same plane as the rotating surface of the wind blades for acquisition and positioning; further, in this embodiment, the upper limit frequency of the fault sound source is 16kHz, the sampling rate of the acquisition array may be set to 32kHz, and the distance between the sensors on the acquisition array is set to 0.1 m; and the closer the fault of the tested sound source on the wind blade and the acquisition array are to the vertical direction, the higher the positioning precision is.

S2, enhancing the picked blade sound signal through an array signal processing technology; constructing a CNN fault judgment model;

because the wind blade rotates continuously and has symmetry, in the embodiment, the acquisition array only scans the direction of the sound source in the single-side area of the rotating surface of the wind blade, such as the angle range covered by ∠ BAO in FIG. 1.

Further, the step S2 specifically includes the following steps;

in order to meet the requirement of positioning accuracy, the wind blade is divided into a plurality of areas according to the unit of n meters;

FIG. 2, is a schematic view of an embodiment of the invention providing a division of a wind blade into several regions; in this embodiment, the total length of the wind blade is 60 meters, the wind blade is divided into 30 regions by taking 2 meters as a unit, 30 directions from the array to the fan blade are divided according to the relative position relationship between the acquisition array and the first blade 1 in fig. 1, each direction points to the central position of a small region, and after the 30 directions are determined, the sound signal enhancement in the 30 directions is respectively performed on each frame of received signals by adopting the beam forming technology.

In this embodiment, a CNN (convolutional neural networks) fault determination model is used to perform voice signal identification, so as to implement accurate fan blade fault diagnosis; and adopting a method similar to image recognition to judge the sound fault, firstly carrying out Fourier transform on the sound sampling signal, and taking the obtained time-frequency distribution graph of the sound signal as the input of a CNN fault judgment model.

The CNN fault determination model adopted in the present invention is shown in fig. 3, and fig. 3 is a structural diagram of the CNN fault determination model provided in the embodiment of the present invention; the CNN fault judgment model comprises a plurality of layers of convolution layers and a full-connection network which are sequentially connected in series, wherein after a time frequency distribution diagram is input to a first layer of convolution layer, the first layer of convolution layer is subjected to feature extraction, extracted features are input to a next layer of convolution layer for pooling, and dimensionality is reduced through pooling of the plurality of layers of convolution layers; after the last convolution layer is subjected to pooling, splitting and splicing a plurality of groups of finally output two-dimensional extraction features into vectors, and inputting the vectors into a full-connection network; and processing and outputting the required result through the full-connection network.

Preferably, in this embodiment, the convolutional layers each include three steps of convolution, activation, and pooling, where the convolutional layers are four layers, and are respectively a first convolutional layer, a second convolutional layer, a third convolutional layer, and a fourth convolutional layer, where the first convolutional layer 16 × 16, the number of the convolutional layers is 64, the step size is 2, and the activation function of the first convolutional layer uses ReLu (rectifiedlireunit, linear rectification function); pooling of the first layer convolutional layer uses 2x2, step 2; a second layer of convolutional layer convolutional kernels 8x8, number 128, step 2; ReLu is also used as the activation function of the second convolutional layer; pooling of the second convolutional layer also uses 2x2, step 2; convolution kernels of the third convolutional layer 4x4, the number of which is 128, and the step size of which is 1; ReLu is also used as the activation function of the convolution layer at the third layer; the third convolutional layer pooling also uses 2x2, step 2; a fourth layer of convolutional layer convolutional kernels 2x2, the number of which is 128, and the step size of which is 1; the activation function of the fourth convolutional layer is ReLu as well; the fourth convolutional layer pooling uses 2x2, step 2;

the full-connection network is composed of two full-connection layers which are sequentially connected in series, and 2048 neurons are used in the full-connection layers;

the convolution operation formula is as follows:

wherein x is_i,jRepresenting the input-frequency distribution pattern participating in the convolution, c_i,jRepresenting a convolution kernel.

The convolution kernel size is calculated by a back propagation algorithm in a training stage, in the embodiment, the convolution kernel size of the first convolution layer is 16 × 16, the convolution kernel size of the second convolution layer is 8 × 8, the convolution kernel size of the third convolution layer is 4 × 4, and the convolution kernel size of the fourth convolution layer is 2 × 2;

S3, positioning the fault position of the fault blade by using the enhanced blade sound signal; the blade sound signal comprises fault noise and normal working noise;

when the fault noise has a different frequency component compared to the normal operation noise, or the frequency of the fault noise is higher than the preset value a, or the energy of the fault noise is higher than that of the normal operation noise, step S31 is performed,

the preset value a of this embodiment may be set according to actual conditions, and this value is not a fixed value, and the preset value a of this embodiment is preferably: twice the normal operating noise frequency;

when the frequency of the fault noise is lower than the preset value a (usually, the frequency of the fault noise coincides with or is close to the frequency of the normal operation noise, and in this embodiment, the close frequency is two times lower than the frequency of the normal operation noise), step S32 is executed;

s31, respectively carrying out enhanced denoising processing on the selected 30 directions by the acquisition array receiving signals to obtain 30 region noises; comparing the noise obtained by enhancement with the statistical result of the normal working noise, and selecting the area with the largest difference as a suspected fault area; inputting the enhancement result of the suspected fault area into a CNN fault judgment model which is trained in advance for further verification; if a failure is determined, the area is a failure area (i.e., failure direction).

The enhanced denoising process is the prior art, and will not be described in detail here.

However, since the wind blade is rotating and the current position of the wind blade is unknown to the collecting array, although the fault direction can be determined, the position of the fault on the wind blade cannot be determined, so the fault position is determined by determining the fault azimuth angle, specifically as follows:

as shown in fig. 4, along with the rotation of the wind blade, the fault position on the wind blade changes along a circular track, two rays are led out from the center of the acquisition array, one ray passes through the fault position, the other ray passes through the center of the rotating surface of the wind blade, and the fault azimuth angle is formed at the moment, as shown in the figure, a point U is the fault position, O is the center of the rotating surface of the wind blade, and a is the center point of the acquisition array;

it is emphasized that the magnitude of the fault azimuth does not need to be measured manually, and the acquisition array automatically records the magnitude of the fault azimuth when the fault position is detected.

S32, when the denoising processing is enhanced for each direction, the enhancement result of each direction is input into the trained CNN fault judgment model, and the probability of generating faults is output; and if the maximum value of the fault probability in each output direction is greater than the threshold value, determining that the fault exists in the direction, determining that the region with the maximum fault probability is the fault direction, and determining the position of the fault on the blade by adopting the method for determining the fault azimuth angle.

Preferably, in this embodiment, the threshold is 50%.

If the energy of the sound source in a certain direction exceeds the threshold value B and the sound source is judged to be a non-fault sound source, the strong interference sound source exists in the direction, and the step S4 is executed, otherwise, the step S5 is directly executed;

it should be noted that the threshold B in this embodiment may be set according to an actual situation, and preferably, in this embodiment, the threshold B is a numerical value in which the sound energy received in a certain direction is 5 times higher than the background noise energy;

in this embodiment, a beam nulling technique is used to suppress the interfering sound source in the direction, and the process returns to step S3.

The sound signal enhancement method and the beam nulling technique are prior art, and will not be described in detail here.

S5, positioning the wind blade with the fault after positioning the fault position;

referring to fig. 4, fig. 4 is a schematic diagram of a fault azimuth angle provided by an embodiment of the invention, a wind driven generator has three wind driven blades, and it is required to determine on which wind driven blade a fault occurs, when the fault azimuth angle ∠ MAU reaches the maximum, the fault azimuth angle is tangent to a circular track of a fault position, and a position of the fault wind driven blade can be determined.

in order to be able to accurately diagnose blade faults, a more complex and accurate blade fault type database needs to be established; the blade fault type database comprises key information such as common fault types and fault generating positions of the blades. The establishment of the blade fault type database needs to classify and label various blade faults of different types based on collected sound signals in the long-term working process of the blade and by combining abundant operation and maintenance personnel experience so as to increase the discrimination and establish an accurate fault type description relationship. In addition, for different blade faults, the characteristics of sound signals are different, so that various fault audio information needs to be recorded and labeled; especially, when the CNN fault judgment model outputs a suspected unknown fault type, the collected sound signal is marked as an actual fault type after being manually confirmed.

S7 model increment training;

there are two ways of incremental training, Sample Incremental Training (SIT) and output class incremental learning (CIT), where,

and (3) sample increment training: and under the condition of ensuring the existing knowledge, extracting new knowledge through incremental learning of the CNN fault judgment model on the new sample, using the new sample to train the CNN fault judgment model, and realizing updating.

Output category incremental learning: copying the CNN fault judgment model to obtain another model, wherein one model is not modified, the other model is added with new faults on the basis of outputting the original faults, retraining to obtain a new model, and simultaneously considering the prediction performance of the new model on new fault categories and the response difference of the original fault categories on the new model and the old model, so that the LOSS function of the new model is adjusted to obtain: LOSS1+ LOSS 2;

wherein LOSS1 refers to new class detection; LOSS2 refers to the difference in the output of data on the old model and the new model; LOSS2 ensures that the output on the old and new models is similar, so that the new model does not change the prediction effect of the old model.

Example 2

the deployment module is used for deploying the acquisition array;

also includes; the acquisition array and the rotating surface of the wind blade are positioned in the same plane; meanwhile, when the perpendicular bisector of the line segment where the wind blade is located passes through the midpoint of the acquisition array, the acquisition array is parallel to the wind blade;

also includes; dividing the wind blade into m regions according to n meters as a unit, determining one direction from the acquisition array to each region to obtain m directions, and performing signal enhancement on the m directions;

the CNN fault judgment model comprises a plurality of layers of convolution layers and a full-connection network which are sequentially connected in series, wherein after a time frequency distribution diagram is input to a first layer of convolution layer, the first layer of convolution layer is subjected to feature extraction, extracted features are input to a next layer of convolution layer for pooling, and dimensionality is reduced through pooling of the plurality of layers of convolution layers; after the last convolution layer is subjected to pooling, splitting and splicing a plurality of groups of finally output two-dimensional extraction features into vectors, and inputting the vectors into a full-connection network; processing and outputting a required result through the full-connection network;

and identifying the wind blade sound signal picked up by the target direction area by using the CNN fault judgment model, wherein the wind blade sound signal is a fault sound source when the output state of the CNN fault judgment model is a fault state, and the wind blade sound signal is a non-fault sound source if the output state is normal.

also includes; when the fault noise and the normal working noise have different frequency components, or the frequency of the fault noise is higher than a preset value A, or the energy of the fault noise is higher than the energy of the normal working noise, the collected array received signals are subjected to enhanced denoising processing on the selected m directions respectively to obtain the noise of m areas; comparing the noise obtained by enhancement with the statistical result of the normal working noise, and selecting the area with the largest difference as a suspected fault area; inputting the enhancement result of the suspected fault area into a CNN fault judgment model which is trained in advance for further verification; if the judgment result is that the fault exists, the area is a fault area;

after the fault area is determined, two rays are led out from the center of the collecting array, one ray passes through the fault position, the other ray passes through the circle center of the rotating surface of the wind blade, the fault azimuth angle is formed at the moment, and when the fault azimuth angle is the largest, the fault position is determined;

when the fault noise is lower than a preset value A, inputting the enhancement result of each direction into a trained CNN fault judgment model, and outputting the probability of generating faults; if the maximum value of the fault probabilities in the output directions is larger than the threshold value, the fault is determined to exist, and the region with the maximum fault probability is a fault region; then positioning the fault position through a fault azimuth angle;

if the energy of a sound source in a certain direction exceeds a threshold value B and a non-fault sound source is judged, the module is inhibited to work if a strong interference sound source exists in the direction, and otherwise, the blade positioning module works;

the blade positioning module is used for positioning the wind blade with the fault; further comprising:

when the fault azimuth reaches the maximum, the fault azimuth is tangent to the circular track of the fault position; determining a faulty wind blade position;

the blade rotating speed is calculated by recording the time interval of the wind blade with more than two faults passing through the position, so that the moment when the fault blade rotates to the vertical downward direction is estimated, the fault indicating lamp on the collecting device is lightened, namely, when the fault indicating lamp is lightened, the wind blade at the lowest position is the fault wind blade.

The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. The method for monitoring the fault of the wind driven generator blade based on the acoustic sensor array comprises a wind driven generator, wherein the wind driven generator comprises a plurality of wind driven blades, and the wind driven blades rotate around a tower, and is characterized by further comprising the following steps:

s3, positioning the fault position of the fault blade;

when the fault noise is lower than the preset value a, performing step S32;

s4, restraining the strong interference direction and returning to the step S3;

2. The method for monitoring the fault of the wind driven generator blade based on the acoustic sensor array according to the claim 1, wherein the step S1 includes that the collection array is in the same plane with the rotating surface of the wind driven blade; and meanwhile, when the perpendicular bisector of the line segment where the wind blade is located passes through the midpoint of the acquisition array, the acquisition array is parallel to the wind blade.

3. The method for monitoring the fault of the wind driven generator blade based on the acoustic sensor array according to the claim 1, wherein the enhancing the blade sound signal picked up by the collecting array comprises;

4. The method for monitoring the blade fault of the wind driven generator based on the acoustic sensor array is characterized in that the CNN fault judgment model comprises a plurality of convolution layers and a full-connection network which are sequentially connected in series, after a time frequency distribution diagram is input to the first convolution layer, the first convolution layer is subjected to feature extraction, the extracted features are input to the next convolution layer for pooling, and dimensionality is reduced through pooling of the plurality of convolution layers; after the last convolution layer is subjected to pooling, splitting and splicing a plurality of groups of finally output two-dimensional extraction features into vectors, and inputting the vectors into a full-connection network; processing and outputting a required result through the full-connection network;

5. The method for monitoring blade fault of wind power generator based on acoustic sensor array according to claim 1, wherein said step S31 includes,

and after the fault area is determined, two rays are led out from the center of the acquisition array, one ray passes through the fault position, the other ray passes through the center of the rotating surface of the wind blade, the fault azimuth angle is formed at the moment, and when the fault azimuth angle is the largest, the fault position is determined.

6. The method for monitoring the fault of the wind turbine blade based on the acoustic sensor array according to claim 5, wherein the step S32 includes; inputting the enhancement result of each direction into a trained CNN fault judgment model, and outputting the probability of generating faults; if the maximum value of the fault probabilities in the output directions is larger than the threshold value, the fault is determined to exist, and the region with the maximum fault probability is a fault region; then positioning the fault position through a fault azimuth angle;

7. The method for monitoring the fault of the wind driven generator blade based on the acoustic sensor array according to claim 6, wherein the strong interference direction is suppressed to; and adopting a wave beam nulling technology to suppress the interference sound source in the direction.

8. The method for monitoring the fault of the wind turbine blade based on the acoustic sensor array according to any one of claims 6 to 7, wherein the step S5 includes;

9. The method for monitoring the fault of the wind driven generator blade based on the acoustic sensor array according to claim 1, further comprising steps S6 and S7;

and (3) sample increment training: extracting new knowledge through incremental learning of the CNN fault judgment model on a new sample, using the new sample to obtain a new sample, and training the CNN fault judgment model to realize updating;

output category incremental learning: copying the CNN fault judgment model to obtain another model, adding a new fault to the copied model on the basis of outputting the original fault, retraining to obtain a new model, and adjusting the LOSS function of the new model to obtain: LOSS1+ LOSS 2;

10. The monitoring device of the method for monitoring the fault of the wind turbine blade based on the acoustic sensor array according to any one of claims 1 to 9, comprising a wind turbine, the wind turbine comprising a plurality of wind blades, the wind blades rotating around a tower, and further comprising;