CN118517379A - Fan fault monitoring method based on vibration and abnormal sound feature fusion and storage medium - Google Patents

Abstract

The invention relates to a fan fault monitoring method and a storage medium based on vibration and abnormal sound feature fusion, wherein the method comprises the following steps: the method comprises the steps of collecting field data through vibration sensors and listening devices distributed at different positions on a wind turbine generator; aligning vibration data acquired by a vibration sensor and bone conduction audio data acquired by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms; inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification; and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center. The wind turbine generator system can be maintained in a preventive manner when faults occur in early stages, the wind turbine generator system cannot be maintained in time in the early stages of fault development, and high maintenance cost and high power generation loss are brought after the wind turbine generator system develops into a malignant equipment damage accident.

Description

Fan fault monitoring method based on vibration and abnormal sound feature fusion and storage medium

Technical Field

The application relates to the technical field of fan fault monitoring, in particular to a fan fault monitoring method based on vibration and abnormal sound feature fusion and a storage medium.

Background

With the rapid development of wind power, the unit performances of different manufacturers, models and batches in the early stage of wind power development are greatly different. And with the continuous lapse of the running time of the equipment, the problems of design, manufacture, installation and the like left in the running stage are gradually exposed, and the problems of familiarity and sporadic property are relatively more, so that the wind power utilization hours are gradually reduced.

The components with higher failure rate of the wind turbine are concentrated on a gear box, a pitch system, a generator, an electrical system and a control system. The failure of the electrical and control system of the wind turbine is frequent, but the downtime of the wind turbine caused by maintaining the failure is relatively short; the main shaft of the transmission chain, the bearing, the gear box, the generator and other parts with lower failure rate have long maintenance time, wherein the shutdown time caused by the failure of the gear box is longest. The high failure rate of the wind turbine and key components thereof not only affects the safe and reliable operation of the wind turbine, but also increases the maintenance cost of the wind farm to a great extent. Therefore, it is necessary to make early predictions of wind turbine plant faults, such as: the intelligent recognition mechanism of the fan faults based on vibration and abnormal sounds can find defect characteristics causing unit failure in an early stage, and through early preventive maintenance, the situation that the wind turbine generator cannot be maintained in time in the early stage of fault development is avoided, and when the wind turbine generator is developed into a malignant equipment damage accident, high maintenance cost and high power generation loss are brought.

Disclosure of Invention

In view of the above problems, the application provides a fan fault monitoring method and a storage medium based on the fusion of vibration and abnormal sound characteristics, which solve the problems that a wind turbine cannot be maintained in time in the early stage of fault development, and high maintenance cost and high power generation loss are caused after the wind turbine is developed into a malignant equipment damage accident.

In order to achieve the above object, the present inventors provide a fan fault monitoring method based on the fusion of vibration and abnormal sound characteristics, including:

The method comprises the steps of collecting field data through vibration sensors and listening devices distributed at different positions on a wind turbine generator;

aligning vibration data acquired by a vibration sensor and bone conduction audio data acquired by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms;

Inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification;

and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center.

In some embodiments, the step of inputting the spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into the trained feature fusion network model to perform state recognition specifically includes the following steps:

Extracting features of a spectrogram corresponding to the input vibration sensor through a first feature extraction network;

Extracting features of a spectrogram corresponding to the input bone conduction audio data through a second feature extraction network;

the feature fusion layer fuses the features extracted by the first feature extraction network and the features extracted by the second feature extraction network to obtain fusion features;

and classifying the fusion features input by the feature fusion layer through a classification network to judge whether the state is abnormal.

In some embodiments, the first feature extraction network and the second feature extraction network of the feature fusion network model share weight parameters.

In some embodiments, the training method of the feature fusion network model is as follows:

randomly segmenting the working state spectrogram and the non-working state spectrogram into a training set and a testing set according to a preset proportion;

inputting the training set into a feature fusion network model for training;

And as the iteration times of the feature fusion network model are increased, when the accuracy of the feature fusion network model on the test set is kept stable and does not rise any more within the preset time, training the feature fusion network model is stopped, and a trained feature fusion network model is obtained.

In some embodiments, before the randomly segmenting the working state spectrogram and the non-working state spectrogram into the training set and the testing set according to the preset proportion, the method further includes the following steps:

And carrying out data normalization processing on the working state spectrogram and the non-working state spectrogram.

There is also provided another technical solution, a storage medium storing a computer program which, when executed by a processor, performs the steps of:

The method comprises the steps of collecting field data through vibration sensors and listening devices distributed at different positions on a wind turbine generator;

aligning vibration data acquired by a vibration sensor and bone conduction audio data acquired by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms;

Inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification;

and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center.

In some embodiments, the step of inputting the spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into the trained feature fusion network model to perform state recognition specifically includes the following steps:

Extracting features of a spectrogram corresponding to the input vibration sensor through a first feature extraction network;

Extracting features of a spectrogram corresponding to the input bone conduction audio data through a second feature extraction network;

the feature fusion layer fuses the features extracted by the first feature extraction network and the features extracted by the second feature extraction network to obtain fusion features;

and classifying the fusion features input by the feature fusion layer through a classification network to judge whether the state is abnormal.

In some embodiments, the first feature extraction network and the second feature extraction network of the feature fusion network model share weight parameters.

In some embodiments, the training method of the feature fusion network model is as follows:

randomly segmenting the working state spectrogram and the non-working state spectrogram into a training set and a testing set according to a preset proportion;

inputting the training set into a feature fusion network model for training;

And as the iteration times of the feature fusion network model are increased, when the accuracy of the feature fusion network model on the test set is kept stable and does not rise any more within the preset time, training the feature fusion network model is stopped, and a trained feature fusion network model is obtained.

In some embodiments, before the randomly segmenting the working state spectrogram and the non-working state spectrogram into the training set and the testing set according to the preset proportion, the method further includes the following steps:

And carrying out data normalization processing on the working state spectrogram and the non-working state spectrogram.

Different from the prior art, the technical scheme is characterized in that the vibration sensors and the listening devices distributed at different positions on the wind turbine generator are used for collecting the field data of each position on the wind turbine generator; then aligning the vibration data and bone conduction audio data collected by the vibration sensor and the audiometer in the same group along a time axis, respectively calculating corresponding spectrograms, and then equally-spaced dividing the obtained spectrograms; and then inputting the spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into a trained feature fusion network model for state recognition, and sending early warning information to a control center when the recognition state of the feature fusion network model is abnormal. When a certain position of the wind turbine generator fails, the data collected through the vibration sensor and the listening device are identified as abnormal by the feature fusion network model, so that after the control center receives the early warning information, maintenance personnel can be quickly arranged to go to the site for checking and maintaining, preventive maintenance can be performed when the wind turbine generator fails in early stage, the wind turbine generator is prevented from being maintained in time at the early stage of failure development, and high maintenance cost and high power generation loss are brought after the wind turbine generator is developed into a malignant equipment damage accident.

The foregoing summary is merely an overview of the present application, and may be implemented according to the text and the accompanying drawings in order to make it clear to a person skilled in the art that the present application may be implemented, and in order to make the above-mentioned objects and other objects, features and advantages of the present application more easily understood, the following description will be given with reference to the specific embodiments and the accompanying drawings of the present application.

Drawings

The drawings are only for purposes of illustrating the principles, implementations, applications, features, and effects of the present application and are not to be construed as limiting the application.

In the drawings of the specification:

FIG. 1 is a schematic flow chart of a fan fault monitoring method based on vibration and abnormal sound feature fusion according to an embodiment;

Fig. 2 is a schematic flow chart of step S130 according to the embodiment;

FIG. 3 is a schematic structural diagram of a feature fusion network model according to an embodiment;

FIG. 4 is a schematic flow chart of another fan fault monitoring method based on vibration and abnormal sound feature fusion according to an embodiment;

fig. 5 is a schematic structural view of a storage medium according to an embodiment.

Reference numerals referred to in the above drawings are explained as follows:

510. The storage medium may be a storage medium,

520. A processor.

Detailed Description

In order to describe the possible application scenarios, technical principles, practical embodiments, and the like of the present application in detail, the following description is made with reference to the specific embodiments and the accompanying drawings. The embodiments described herein are only for more clearly illustrating the technical aspects of the present application, and thus are only exemplary and not intended to limit the scope of the present application.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of the phrase "in various places in the specification are not necessarily all referring to the same embodiment, nor are they particularly limited to independence or relevance from other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the technical features mentioned in each embodiment may be combined in any manner to form a corresponding implementable technical solution.

Unless defined otherwise, technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application pertains; the use of related terms herein is for the purpose of describing particular embodiments only and is not intended to limit the application.

In the description of the present application, the term "and/or" is a representation for describing a logical relationship between objects, which means that three relationships may exist, for example a and/or B, representing: there are three cases, a, B, and both a and B. In addition, the character "/" herein generally indicates that the front-to-back associated object is an "or" logical relationship.

In the present application, terms such as "first" and "second" are used merely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual number, order, or sequence of such entities or operations.

Without further limitation, the use of the terms "comprising," "including," "having," or other like terms in this specification is intended to cover a non-exclusive inclusion, such that a process, method, or article of manufacture that comprises a list of elements does not include additional elements but may include other elements not expressly listed or inherent to such process, method, or article of manufacture.

As in the understanding of "review guidelines," the expressions "greater than", "less than", "exceeding" and the like are understood to exclude this number in the present application; the expressions "above", "below", "within" and the like are understood to include this number. Furthermore, in the description of embodiments of the present application, the meaning of "a plurality of" is two or more (including two), and similarly, the expression "a plurality of" is also to be understood as such, for example, "a plurality of" and the like, unless specifically defined otherwise.

In the description of embodiments of the present application, spatially relative terms such as "center," "longitudinal," "transverse," "length," "width," "thickness," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counter-clockwise," "axial," "radial," "circumferential," etc., are used herein as a basis for the description of the embodiments or as a basis for the description of the embodiments, and are not intended to indicate or imply that the devices or components referred to must have a particular position, a particular orientation, or be configured or operated in a particular orientation and therefore should not be construed as limiting the embodiments of the present application.

Unless specifically stated or limited otherwise, the terms "mounted," "connected," "affixed," "disposed," and the like as used in the description of embodiments of the application should be construed broadly. For example, the "connection" may be a fixed connection, a detachable connection, or an integral arrangement; the device can be mechanically connected, electrically connected and communicated; it can be directly connected or indirectly connected through an intermediate medium; which may be a communication between two elements or an interaction between two elements. The specific meaning of the above terms in the embodiments of the present application can be understood by those skilled in the art to which the present application pertains according to circumstances.

Referring to fig. 1, the embodiment provides a fan fault monitoring method based on vibration and abnormal sound feature fusion, including:

Step S110: the method comprises the steps of collecting field data through vibration sensors and listening devices distributed at different positions on a wind turbine generator;

Step S120: aligning vibration data acquired by a vibration sensor and bone conduction audio data acquired by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms;

step S130: inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification;

Step S140: and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center.

The method comprises the steps that field data of all positions on a wind turbine are collected through vibration sensors and audiometers distributed at different positions on the wind turbine; then aligning the vibration data and bone conduction audio data collected by the vibration sensor and the audiometer in the same group along a time axis, respectively calculating corresponding spectrograms, and then equally-spaced dividing the obtained spectrograms; and then inputting the spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into a trained feature fusion network model for state recognition, and sending early warning information to a control center when the recognition state of the feature fusion network model is abnormal. When a certain position of the wind turbine generator fails, the data collected through the vibration sensor and the listening device are identified as abnormal by the feature fusion network model, so that after the control center receives the early warning information, maintenance personnel can be quickly arranged to go to the site for checking and maintaining, preventive maintenance can be performed when the wind turbine generator fails in early stage, the wind turbine generator is prevented from being maintained in time at the early stage of failure development, and high maintenance cost and high power generation loss are brought after the wind turbine generator is developed into a malignant equipment damage accident.

Referring to fig. 2, in some embodiments, the step of inputting the segmented spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into the trained feature fusion network model to perform state recognition specifically includes the following steps:

step S210: extracting features of a spectrogram corresponding to the input vibration sensor through a first feature extraction network;

Step S220: extracting features of a spectrogram corresponding to the input bone conduction audio data through a second feature extraction network;

Step S230: the feature fusion layer fuses the features extracted by the first feature extraction network and the features extracted by the second feature extraction network to obtain fusion features;

Step S240: and classifying the fusion features input by the feature fusion layer through a classification network to judge whether the state is abnormal.

As shown in fig. 3, the feature fusion network model includes a feature extraction network, a feature fusion network and a classification network, where the feature extraction network is two: the first feature extraction network and the second feature extraction network respectively perform feature extraction on vibration data and audio data. The first feature extraction network and the second feature extraction network share weight parameters. Each feature extraction network consists of a 5-layer convolutional network and 1 fully-connected layer. The classification network employs an MLP classification network that employs three hidden layers, each hidden layer followed by a Relu activation function.

The received data includes different region vibration data and bone conduction audio data. First, both are converted from the time domain into a spectrogram. And respectively inputting the voice data spectrogram and the vibration data spectrogram into a feature extraction network and then outputting 512-dimensional feature vectors. And inputting the extracted features into an MLP classification network after feature fusion to judge whether the features are abnormal.

In some embodiments, the training method of the feature fusion network model is as follows:

randomly segmenting the working state spectrogram and the non-working state spectrogram into a training set and a testing set according to a preset proportion;

inputting the training set into a feature fusion network model for training;

And as the iteration times of the feature fusion network model are increased, when the accuracy of the feature fusion network model on the test set is kept stable and does not rise any more within the preset time, training the feature fusion network model is stopped, and a trained feature fusion network model is obtained.

And (3) the working state spectrogram and the non-working state spectrogram are combined according to a training set: test set = 9:1, randomly splitting. And finally, classifying the segmented data, and storing the segmented data for subsequent model training and evaluation. And training is carried out on the training set by using a feature fusion network, and the accuracy of the model on the testing set is continuously improved along with the increase of the iteration times of the model. When the model has reached an optimal state, the accuracy will tend to stabilize. Once the accuracy over the test set has stabilized for a period of time and is no longer significantly improved, the model can be considered to have converged. At this point, training should be stopped and the weights of the model saved.

In some embodiments, before the randomly segmenting the working state spectrogram and the non-working state spectrogram into the training set and the testing set according to the preset proportion, the method further includes the following steps:

And carrying out data normalization processing on the working state spectrogram and the non-working state spectrogram.

Data normalization is an important preprocessing step that scales data to fall within a particular interval. The accuracy, convergence speed and stability of the model can be improved through data normalization. Therefore, the segmented spectrogram is first normalized before further processing.

Referring to fig. 4, in some embodiments, a fan fault monitoring method based on the fusion of vibration and abnormal sound features includes the following steps:

step one: and deploying a sensor network to acquire field data.

And respectively arranging vibration sensors and audiometers near a wind wheel system, a hub variable-pitch control cabinet, a yaw system, yaw driving teeth, a main shaft, a low-speed end and a high-speed end of a gear box, a front end and a rear end of a generator, wherein the cabin control cabinet and the tower foundation control cabinet comprise a converter cabinet, a grid-connected cabinet and a control cabinet to acquire field data in real time. The data collected by the sensor and the listening device are transmitted to the edge gateway through the network cable, and the edge gateway is converged and then transmitted to the server.

Step two: the time spectrum of the vibration data and the audio data are calculated respectively.

The data received by the server includes vibration data and bone conduction audio data of different areas. For analysis, it is first necessary to align the two data along the time axis. Then, for each data type, a corresponding spectrogram is calculated separately. The time spectrum is a representation method for analyzing the change of the signal in time and frequency, which provides simultaneous analysis of the signal in time and frequency domains and can reflect the energy distribution of the signal in different times and frequencies. Finally, the time spectrum is cut at equal intervals and saved for later analysis. This effectively extracts the time-frequency characteristics of the signal for further analysis.

Step three: the training set and the test set are segmented.

Data normalization is an important preprocessing step that scales data to fall within a particular interval. The accuracy, convergence speed and stability of the model can be improved through data normalization. Therefore, the segmented spectrogram is first normalized before further processing. Then, the working state spectrogram and the non-working state spectrogram are combined according to a training set: test set = 9:1, randomly splitting. And finally, classifying the segmented data, and storing the segmented data for subsequent model training and evaluation.

Step four: and building a feature fusion network.

The received data includes different region vibration data and bone conduction audio data. First, both are converted from the time domain into a spectrogram. And respectively inputting the voice data spectrogram and the vibration data spectrogram into a feature extraction network and then outputting 512-dimensional feature vectors. The feature extraction network sharing weight parameter consists of a 5-layer convolution network and 1 full connection layer.

And inputting the extracted features into an MLP classification network after feature fusion to judge whether the features are abnormal. The MLP network employs three hidden layers, each hidden layer followed by a Relu activation function.

Step five: and (5) training a network.

And training is performed on the training set by using a feature fusion network, and the accuracy of the model on the verification set is continuously improved along with the increase of the iteration times of the model. When the model has reached an optimal state, the accuracy will tend to stabilize. Once the accuracy on the validation set has stabilized for a period of time and is no longer significantly improved, the model can be considered to have converged. At this point, training should be stopped and the weights of the model saved.

Step six: and (5) identifying states.

If the algorithm judges abnormality in the duration time, early warning information is sent to the control center in time, and a control center technician can further confirm whether the abnormality exists or not by referring to the sound data and the vibration data stored in real time.

By disposing vibration sensors and audiometers near the yaw system, the yaw driving teeth, the main shaft and the generator end side, on-site vibration and bone conduction data are obtained in real time, and the data are converted into a spectrogram from a time domain, so that fault characteristics are extracted by using a deep learning algorithm. The frequency spectrum features are used for representing the audio frequency and vibration features of the fan unit, so that the complexity is reduced, the model parameters are few and easy to deploy, the anti-interference capability is high, and the robustness is good.

Referring to fig. 5, in another embodiment, a storage medium 510 stores a computer program, which when executed by a processor 520 performs the steps of:

The method comprises the steps of collecting field data through vibration sensors and listening devices distributed at different positions on a wind turbine generator;

aligning vibration data acquired by a vibration sensor and bone conduction audio data acquired by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms;

Inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification;

and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center.

The method comprises the steps that field data of all positions on a wind turbine are collected through vibration sensors and audiometers distributed at different positions on the wind turbine; then aligning the vibration data and bone conduction audio data collected by the vibration sensor and the audiometer in the same group along a time axis, respectively calculating corresponding spectrograms, and then equally-spaced dividing the obtained spectrograms; and then inputting the spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into a trained feature fusion network model for state recognition, and sending early warning information to a control center when the recognition state of the feature fusion network model is abnormal. When a certain position of the wind turbine generator fails, the data collected through the vibration sensor and the listening device are identified as abnormal by the feature fusion network model, so that after the control center receives the early warning information, maintenance personnel can be quickly arranged to go to the site for checking and maintaining, preventive maintenance can be performed when the wind turbine generator fails in early stage, the wind turbine generator is prevented from being maintained in time at the early stage of failure development, and high maintenance cost and high power generation loss are brought after the wind turbine generator is developed into a malignant equipment damage accident.

In some embodiments, the step of inputting the spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into the trained feature fusion network model to perform state recognition specifically includes the following steps:

Extracting features of a spectrogram corresponding to the input vibration sensor through a first feature extraction network;

Extracting features of a spectrogram corresponding to the input bone conduction audio data through a second feature extraction network;

the feature fusion layer fuses the features extracted by the first feature extraction network and the features extracted by the second feature extraction network to obtain fusion features;

and classifying the fusion features input by the feature fusion layer through a classification network to judge whether the state is abnormal.

As shown in fig. 3, the feature fusion network model includes a feature extraction network, a feature fusion network and a classification network, where the feature extraction network is two: the first feature extraction network and the second feature extraction network respectively perform feature extraction on vibration data and audio data. The first feature extraction network and the second feature extraction network share weight parameters. Each feature extraction network consists of a 5-layer convolutional network and 1 fully-connected layer. The classification network employs an MLP classification network that employs three hidden layers, each hidden layer followed by a Relu activation function.

The received data includes different region vibration data and bone conduction audio data. First, both are converted from the time domain into a spectrogram. And respectively inputting the voice data spectrogram and the vibration data spectrogram into a feature extraction network and then outputting 512-dimensional feature vectors. And inputting the extracted features into an MLP classification network after feature fusion to judge whether the features are abnormal.

In some embodiments, the training method of the feature fusion network model is as follows:

randomly segmenting the working state spectrogram and the non-working state spectrogram into a training set and a testing set according to a preset proportion;

inputting the training set into a feature fusion network model for training;

And as the iteration times of the feature fusion network model are increased, when the accuracy of the feature fusion network model on the test set is kept stable and does not rise any more within the preset time, training the feature fusion network model is stopped, and a trained feature fusion network model is obtained.

And (3) the working state spectrogram and the non-working state spectrogram are combined according to a training set: test set = 9:1, randomly splitting. And finally, classifying the segmented data, and storing the segmented data for subsequent model training and evaluation. And training is carried out on the training set by using a feature fusion network, and the accuracy of the model on the testing set is continuously improved along with the increase of the iteration times of the model. When the model has reached an optimal state, the accuracy will tend to stabilize. Once the accuracy over the test set has stabilized for a period of time and is no longer significantly improved, the model can be considered to have converged. At this point, training should be stopped and the weights of the model saved.

In some embodiments, before the randomly segmenting the working state spectrogram and the non-working state spectrogram into the training set and the testing set according to the preset proportion, the method further includes the following steps:

And carrying out data normalization processing on the working state spectrogram and the non-working state spectrogram.

Data normalization is an important preprocessing step that scales data to fall within a particular interval. The accuracy, convergence speed and stability of the model can be improved through data normalization. Therefore, the segmented spectrogram is first normalized before further processing.

Finally, it should be noted that, although the embodiments have been described in the text and the drawings, the scope of the application is not limited thereby. The technical scheme generated by replacing or modifying the equivalent structure or equivalent flow by utilizing the content recorded in the text and the drawings of the specification based on the essential idea of the application, and the technical scheme of the embodiment directly or indirectly implemented in other related technical fields are included in the patent protection scope of the application.

Claims (10)

1. A fan fault monitoring method based on vibration and abnormal sound feature fusion is characterized by comprising the following steps:

collecting field data through sensor groups distributed at different positions on the wind turbine generator, wherein the sensor groups comprise vibration sensors and listening devices;

After aligning vibration data collected by vibration sensors in the same group and bone conduction audio data collected by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms;

Inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification;

and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center.

2. The fan fault monitoring method based on vibration and abnormal sound feature fusion according to claim 1, wherein the step of inputting the segmented spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data at the same position into the trained feature fusion network model for performing state recognition specifically comprises the following steps:

Extracting features of a spectrogram corresponding to the input vibration sensor through a first feature extraction network;

Extracting features of a spectrogram corresponding to the input bone conduction audio data through a second feature extraction network;

the feature fusion layer fuses the features extracted by the first feature extraction network and the features extracted by the second feature extraction network to obtain fusion features;

and classifying the fusion features input by the feature fusion layer through a classification network to judge whether the state is abnormal.

3. The fan fault monitoring method based on vibration and abnormal sound feature fusion according to claim 2, wherein the first feature extraction network and the second feature extraction network of the feature fusion network model share weight parameters.

4. The fan fault monitoring method based on vibration and abnormal sound feature fusion according to claim 1, wherein the training method of the feature fusion network model is as follows:

randomly segmenting the working state spectrogram and the non-working state spectrogram into a training set and a testing set according to a preset proportion;

inputting the training set into a feature fusion network model for training;

And as the iteration times of the feature fusion network model are increased, when the accuracy of the feature fusion network model on the test set is kept stable and does not rise any more within the preset time, training the feature fusion network model is stopped, and a trained feature fusion network model is obtained.

5. The fan fault monitoring method based on the combination of vibration and abnormal sound features according to claim 4, wherein before the working state spectrogram and the non-working state spectrogram are randomly segmented into the training set and the testing set according to a preset proportion, the method further comprises the following steps:

And carrying out data normalization processing on the working state spectrogram and the non-working state spectrogram.

6. A storage medium storing a computer program, characterized in that the computer program when run by a processor performs the steps of:

collecting field data through sensor groups distributed at different positions on the wind turbine generator, wherein the sensor groups comprise vibration sensors and listening devices;

After aligning vibration data collected by vibration sensors in the same group and bone conduction audio data collected by a listening device along a time axis, respectively calculating corresponding spectrograms, and equally-spaced dividing the calculated spectrograms;

Inputting a spectrogram corresponding to the vibration data of the same position after segmentation into a trained feature fusion network model for state identification;

and when the identification state of the feature fusion network model is abnormal, sending early warning information to the control center.

7. The storage medium of claim 6, wherein the inputting the segmented spectrogram corresponding to the vibration data and the spectrogram corresponding to the bone conduction audio data in the same location into the trained feature fusion network model for performing state recognition specifically comprises the following steps:

Extracting features of a spectrogram corresponding to the input vibration sensor through a first feature extraction network;

Extracting features of a spectrogram corresponding to the input bone conduction audio data through a second feature extraction network;

the feature fusion layer fuses the features extracted by the first feature extraction network and the features extracted by the second feature extraction network to obtain fusion features;

and classifying the fusion features input by the feature fusion layer through a classification network to judge whether the state is abnormal.

8. The storage medium of claim 7, wherein the first feature extraction network and the second feature extraction network of the feature fusion network model share weight parameters.

9. The storage medium of claim 6, wherein the training method of the feature fusion network model is as follows:

randomly segmenting the working state spectrogram and the non-working state spectrogram into a training set and a testing set according to a preset proportion;

inputting the training set into a feature fusion network model for training;

And as the iteration times of the feature fusion network model are increased, when the accuracy of the feature fusion network model on the test set is kept stable and does not rise any more within the preset time, training the feature fusion network model is stopped, and a trained feature fusion network model is obtained.

10. The storage medium of claim 9, wherein before randomly segmenting the active and inactive spectrograms into the training set and the test set according to a predetermined ratio, the method further comprises the steps of:

And carrying out data normalization processing on the working state spectrogram and the non-working state spectrogram.