CN115406630A - Method for detecting faults of wind driven generator blades through passive acoustic signals based on machine learning - Google Patents
Method for detecting faults of wind driven generator blades through passive acoustic signals based on machine learning Download PDFInfo
- Publication number
- CN115406630A CN115406630A CN202110586205.2A CN202110586205A CN115406630A CN 115406630 A CN115406630 A CN 115406630A CN 202110586205 A CN202110586205 A CN 202110586205A CN 115406630 A CN115406630 A CN 115406630A
- Authority
- CN
- China
- Prior art keywords
- fault detection
- audio
- training
- frame
- signals
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000010801 machine learning Methods 0.000 title abstract description 5
- 238000001514 detection method Methods 0.000 claims abstract description 333
- 230000005236 sound signal Effects 0.000 claims abstract description 136
- 230000008569 process Effects 0.000 claims abstract description 33
- 230000004044 response Effects 0.000 claims abstract description 14
- 238000012549 training Methods 0.000 claims description 138
- 238000012545 processing Methods 0.000 claims description 64
- 238000001228 spectrum Methods 0.000 claims description 40
- 238000004590 computer program Methods 0.000 claims description 28
- 238000001914 filtration Methods 0.000 claims description 21
- 238000010248 power generation Methods 0.000 claims description 13
- 238000004458 analytical method Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 6
- 230000007423 decrease Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 31
- 238000005516 engineering process Methods 0.000 description 14
- 238000013473 artificial intelligence Methods 0.000 description 12
- 230000000007 visual effect Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 238000007726 management method Methods 0.000 description 10
- 210000002569 neuron Anatomy 0.000 description 6
- 230000004913 activation Effects 0.000 description 5
- 230000006872 improvement Effects 0.000 description 5
- 238000011202 physical detection method Methods 0.000 description 5
- 102100032202 Cornulin Human genes 0.000 description 4
- 101000920981 Homo sapiens Cornulin Proteins 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000009432 framing Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000003203 everyday effect Effects 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000000053 physical method Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000037433 frameshift Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 230000005654 stationary process Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D17/00—Monitoring or testing of wind motors, e.g. diagnostics
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/24—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being the cepstrum
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/72—Wind turbines with rotation axis in wind direction
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Acoustics & Sound (AREA)
- Human Computer Interaction (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Sustainable Energy (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Combustion & Propulsion (AREA)
- Chemical & Material Sciences (AREA)
- Sustainable Development (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The embodiment of the invention discloses a method for detecting faults of a wind driven generator blade by using passive acoustic signals based on machine learning, wherein the method comprises the following steps: in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates; extracting mel frequency cepstrum coefficients of the audio signal; and calling a fault detection model to analyze and process the Mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade. By adopting the embodiment of the invention, the fault detection can be conveniently and accurately carried out on the wind driven generator blade.
Description
Technical Field
The present application relates to the field of artificial intelligence, and in particular, to a method, an apparatus, a device, and a storage medium for fault detection.
Background
Wind energy is a clean renewable resource, and the development of new energy sources such as wind energy and the like is concerned by more and more people. Therefore, wind power generation is one of the fastest growing energy sources in the world today. Wind power generation is that wind power drives blades to rotate, and then the rotating speed is increased through a speed increaser to promote a generator to generate power, and wind power generators are generally built in high-altitude areas, and the blades are easy to age, damage or damage due to severe operating environments. Damaged or iced blades may cause blade angle asymmetry or yaw asymmetry, thereby changing the aerodynamic performance of the wind turbine, resulting in a wind turbine that is not operating properly. Therefore, how to detect the fault of the wind driven generator is a hot issue of research.
Disclosure of Invention
The embodiment of the invention provides a method for detecting the fault of a wind driven generator blade by using a passive acoustic signal based on machine learning, which can accurately detect the fault of the wind driven generator.
In one aspect, an embodiment of the present invention provides a fault detection method, including:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates;
extracting mel frequency cepstrum coefficients of the audio signal;
and calling a fault detection model to analyze and process the mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In one aspect, an embodiment of the present invention provides a fault detection apparatus, where the apparatus includes:
the system comprises an acquisition unit, a control unit and a control unit, wherein the acquisition unit is used for responding to a trigger event for carrying out fault detection on a wind power generation blade and adopting an audio signal generated when the wind power generation blade rotates;
an extraction unit for extracting mel-frequency cepstrum coefficients of the audio signal;
and the processing unit is used for calling a fault detection model to analyze and process the mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In one embodiment, the extracting unit performs the following steps when extracting mel-frequency spectrum cepstrum coefficients of the audio signal:
pre-emphasis processing is carried out on the audio signal, and frame division processing is carried out on the audio signal after the pre-emphasis processing, so that multi-frame audio sub-signals are obtained; overlapping signals with preset duration exist between two adjacent frame audio sub-signals;
performing short-time Fourier transform operation on each frame of audio sub-signals in the multi-frame audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals;
and filtering the frequency spectrum corresponding to each frame of audio sub-signal, and obtaining a Mel frequency spectrum cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering.
In an embodiment, when the extracting unit performs a short-time fourier transform operation on each frame of audio sub-signals in the multiple frames of audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals, the extracting unit performs the following steps:
adding a Hamming window to each frame of audio sub-signal;
and inputting each frame of audio sub-signal added with the Hamming window into a short-time Fourier transform rule for operation to obtain a frequency spectrum corresponding to each frame of audio sub-signal.
In one embodiment, the fault detection model is trained based on a training set, the training set includes a plurality of training audio data and a supervision label corresponding to each of the plurality of training audio data, the plurality of training audio data are obtained by preprocessing collected training audio signals, and the training audio signals are obtained by collecting sounds generated when the wind turbine blades rotate through a microphone array; the supervision label corresponding to any training audio data is used for indicating that when the wind driven generator blade rotates to generate any training audio data, the wind driven generator blade is in a target fault state or a non-fault state, and the target fault state is any one or more of the following fault states: loss of any blade and damage to any blade.
In one embodiment, the processing unit is further configured to:
extracting a Mel frequency spectrum cepstrum coefficient corresponding to each training audio data, and inputting the Mel frequency spectrum cepstrum coefficient corresponding to each training audio data into the fault detection model for analysis processing to obtain a prediction result corresponding to each training audio data;
determining a target loss function according to the prediction result corresponding to each training audio data and the supervision label corresponding to each training audio data;
optimizing the fault detection model in a direction to detect the value of the objective loss function.
In one embodiment, the triggering event refers to that the current time meets a preset time condition for fault detection of the wind driven generator; or the triggering event refers to the fact that a detection instruction for detecting the fault of the wind driven generator is received.
In one aspect, an embodiment of the present invention provides a fault detection device, including: a processor adapted to implement one or more computer programs; and a computer storage medium storing one or more computer programs adapted to be loaded and executed by the processor to:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates;
extracting mel frequency cepstrum coefficients of the audio signal;
and calling a fault detection model to analyze and process the mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In one aspect, an embodiment of the present invention provides a computer storage medium, where a computer program is stored in the computer storage medium, and when executed by a processor, the computer program is configured to perform the following steps:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates;
extracting mel frequency cepstrum coefficients of the audio signal;
and calling a fault detection model to analyze and process the Mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In one aspect, an embodiment of the present invention provides a computer program product or a computer program, where the computer program product includes a computer program, and the computer program is stored in a computer-readable storage medium; a processor of the fault detection device reads the computer program from the computer storage medium, the processor executing the computer program to cause the fault detection device to perform:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates;
extracting mel frequency cepstrum coefficients of the audio signal;
and calling a fault detection model to analyze and process the mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In the embodiment of the invention, when a trigger event for carrying out fault detection on the blade of the wind driven generator exists, an audio signal generated when the blade of the wind driven generator rotates is acquired; further, a Mel frequency cepstrum coefficient of the audio signal is extracted, a fault detection model is called to analyze and process the Mel frequency cepstrum coefficient, and a detection result of fault detection of the wind driven generator blade is obtained. Therefore, in the fault detection process of the embodiment of the invention, the fault detection model is called to carry out fault detection according to the audio signal, so that the fault detection of the wind driven generator is conveniently carried out; and the fault detection model obtains a detection result by analyzing the Mel frequency cepstrum coefficient of the audio signal, and the detection result can accurately reflect whether the blade of the wind driven generator has a fault.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is an application scenario diagram of a fault detection method according to an embodiment of the present invention;
fig. 2 is a schematic flow chart of a fault detection method according to an embodiment of the present invention;
fig. 3 is a schematic flow chart of another fault detection method provided in the embodiment of the present invention;
FIG. 4 is a schematic diagram of a fault detection model provided by an embodiment of the invention;
FIG. 5 is a diagram illustrating a comparison of accuracy of fault detection provided by an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a fault detection apparatus according to an embodiment of the present invention;
fig. 7 is a schematic structural diagram of a fault detection device according to an embodiment of the present invention.
Detailed Description
The technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
With the research and progress of artificial intelligence technology, the artificial intelligence technology is developed and applied in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, smart customer service, and the like.
Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.
The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence base technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.
In addition, wind energy is a clean renewable resource, and the development of new energy sources such as wind energy and the like is paid more and more attention, so that wind power generation is one of the fastest-developing energy sources in the world. Wind power generation is that wind power drives blades to rotate, and then the rotating speed is increased through a speed increaser to promote a generator to generate electricity, and wind power generators are generally built in high-altitude areas, and the blades are easy to age, damage or damage due to severe operating environments. The damaged blade or the iced blade can cause the blade angle asymmetry and the yaw asymmetry, so that the aerodynamic performance of the wind driven generator is changed, the wind driven generator cannot normally operate, the power generation capacity of the generator is reduced, and other parts can be damaged after the wind driven generator operates for a long time, so that greater economic loss is caused. The reduction of the operation and maintenance cost through early fault detection is an important way for increasing the yield of the wind power plant, and the early fault of the wind driven generator can be detected through monitoring the operation state of the blades, so that equipment maintenance can be carried out in advance.
There are many methods for wind turbine blade fault detection based on physical characteristics, such as ultrasonic waves, active acoustic emission signals, vibration signals, visual image signals, and the like. The method of active acoustic emission signal for early blade fault detection is not economical to install because of the wide frequency range of its emission source, the difficult reduction of interference noise, and the need for expensive active transmitters and receivers; with the development of image processing technology, the visual-based blade fault detection method has good performance, but the acquisition of the visual image of the blade in operation is not an easy task, and the algorithm adopting the visual method has high complexity and low efficiency; therefore, the methods are difficult to be applied to the actual wind farm blade fault detection. How to reduce the fault detection cost, reduce the calculation complexity, improve the detection robustness under the complicated and changeable environment becomes a problem to be solved urgently.
Based on the above, the embodiment of the invention applies the artificial intelligence technology to the fault detection of the wind power generation blade, and provides a new fault detection scheme, specifically: firstly, training a fault detection model which can be used for carrying out fault detection according to a passive sound signal through a large amount of training audio data; then, when a triggering event for carrying out fault detection on the blade of the wind driven generator exists, acquiring an audio signal generated when the blade of the wind driven generator rotates, and extracting a Mel frequency cepstrum coefficient of the audio signal; furthermore, the fault detection model is called to analyze and process the audio signal, so that the detection result of the wind driven generator blade can be obtained. Compared with the prior art that a physical method is adopted for fault detection, the method adopting the passive acoustic signal in the embodiment of the invention can greatly reduce the detection cost and is less influenced by environmental interference noise; compared with the fault detection method adopting a visual method in the prior art, the fault detection method provided by the embodiment of the invention has the advantages of lower complexity and higher efficiency. And the fault detection model has strong characteristic learning capability, and the trained fault detection model can effectively identify fault information, so that the accuracy of fault detection of the wind driven generator can be improved.
Referring to fig. 1, a scene diagram of an application of the fault detection method provided in the embodiment of the present invention is shown. 101 denotes a wind turbine blade. When a triggering event for fault detection of a wind turbine blade is detected, an audio signal generated when the wind turbine blade rotates may be collected as shown at 102 in FIG. 1; then extracting mel-frequency cepstrum coefficients of the audio signal as shown by 103 in fig. 1; finally, the mel frequency cepstrum coefficient is input into the fault detection model 104 for analysis and processing, and a detection result of fault detection on the wind driven generator blade is output.
Based on the above fault detection scheme and the application scenario of the fault detection method, an embodiment of the present invention provides a fault detection method, and referring to fig. 2, a flow diagram of the fault detection method provided by the embodiment of the present invention is shown. The fault detection method described in fig. 2 may be performed by the fault detection device, in particular by a processor of the fault detection device. The fault detection device may be a terminal or a server. The terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart sound box, a smart watch, a smart car and the like; the server can be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and can also be a cloud server for providing basic cloud computing services such as cloud service, a cloud database, cloud computing, cloud functions, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, big data and artificial intelligence platform and the like. The fault detection method shown in fig. 2 may include the steps of:
step S201, in response to a trigger event for detecting the fault of the blade of the wind driven generator, acquiring an audio signal generated when the blade of the wind driven generator rotates.
In one embodiment, the triggering event may refer to a current time satisfying a preset time condition for fault detection of the wind turbine blade. As an alternative embodiment, the time condition may refer to that the detection of the wind turbine blades at a certain target time is preset, such as 12 pm every day, or 2 pm every day; as another example, 12 noon every friday or XX 15 every month, etc. In this case, the current time meeting the preset time condition means that the current time is the time indicated in the preset time condition.
As another alternative, the time condition may also refer to that a detection period is preset, for example, detection is performed every several hours, for example, detection is performed every 5 hours or 10 hours. In this case, the current time satisfying the preset time condition means that the current time reaches the detection period indicated in the time condition.
In other embodiments, the triggering event may also refer to receiving a detection instruction for fault detection of the wind turbine blade. The detection instruction can be sent to the fault detection device by a management person through the management device. For example, a fault management interface is displayed in a management device used by a manager, and the fault management interface comprises a starting detection option; and when the manager selects the detection starting option, the management equipment sends a detection instruction to the fault detection equipment. Alternatively, the detection instruction may be generated by directly triggering, by a manager, a start detection option displayed in the fault detection device.
When a triggering event for fault detection of the wind driven generator blade is detected, an audio signal of a product generated when the wind driven generator blade rotates is collected. Alternatively, the audio signal may be picked up by a microphone array.
Step S202, extracting a Mel frequency cepstrum coefficient of the audio signal.
In one embodiment, after the audio signal is acquired, in order to determine whether the blade of the wind turbine has a fault through analysis of the audio signal, features of the audio signal need to be extracted, and fault detection is performed based on the features of the audio signal. The features of the audio signal extracted in the embodiments of the present invention are mel-frequency cepstrum coefficients.
In one embodiment, the extracting mel-frequency cepstrum coefficients of the audio signal may include: pre-emphasis processing is carried out on the audio signal, and frame division processing is carried out on the audio signal after the pre-emphasis processing, so that multi-frame audio sub-signals are obtained; overlapping signals with preset duration exist between two adjacent frame audio sub-signals; performing short-time Fourier transform operation on each frame of audio sub-signals in the multi-frame audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals; and filtering the frequency spectrum corresponding to each frame of audio sub-signal, and obtaining a Mel frequency spectrum cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering.
The pre-emphasis processing is performed on the audio signal to amplify high-frequency components in the audio signal, so that a subsequently obtained frequency spectrum can be flatter. Optionally, in the embodiment of the present invention, a pre-emphasis filter may be used to perform pre-emphasis processing on the audio signal. For example, pre-emphasis processing of an audio signal using a first-order filter can be expressed by the following equation (1):
y(t)＝x(t)-αx(t-1) (1)
in formula (1), x (t) represents an audio signal, y (t) represents a pre-emphasis processed audio signal, and a represents a coefficient of a first-order filter, and in general, a may be 0.95 or 0.97.
It should be understood that in the study of audio signals it was found that audio signals can be considered as short-time stationary, typically in the range of 10-40 milliseconds, the spectral characteristics and some physical characteristic parameters of the audio signal remain substantially unchanged. The processing method and theory of stationary processes can be introduced into the short-time processing of audio signals. Specifically, when the audio signal is processed, the audio signal needs to be subjected to framing processing, so as to ensure that the frame length of each frame of audio sub-signal after the framing processing is a stable signal, that is, the frame length of each frame of audio sub-signal is between 10 milliseconds and 40 milliseconds.
In the embodiment of the invention, after audio signals are subjected to pre-emphasis processing, the audio signals are subjected to framing processing by adopting the frame length of 20-40 milliseconds to obtain multi-frame audio sub-signals, in other words, the frame length of each frame of audio sub-signals in the multi-frame audio sub-signals subjected to framing processing is 20-40 milliseconds. Preferably, the frame length of each frame of the audio sub-signal can be set to be 25 milliseconds. A frame overlap (alternatively referred to as a frame shift) of a preset duration, such as 15 ms, may be set between each frame of the audio sub-signal.
Further, after obtaining the multi-frame audio sub-signals, short-time fourier transform operation may be performed on each frame of audio sub-signals in the multi-frame audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals. Optionally, before performing the short-time fourier transform operation on each frame of audio sub-signals in the multiple frames of audio sub-signals, in order to prevent the spectrum leakage, a hamming window may be added to each frame of audio sub-signals, and then the short-time fourier transform operation of 512 points is performed on each frame of audio sub-signals to which the hamming window is added. Alternatively, the window function can be expressed as the following equation (2):
in formula (2), ω [ N ] represents a window function, N represents a time variable, and N is a window length.
And then, after calculating the frequency spectrum of each frame of audio sub-signal, filtering the frequency spectrum corresponding to each frame of audio sub-signal, and obtaining the Mel frequency cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering.
In one embodiment, the fault detection device may perform a filtering process on a frequency spectrum corresponding to each frame of the audio sub-signal based on a mel (mel) filter. The mel filter is typically a set of triangular filter banks, typically comprising 40 filters, each filter in the filter bank being triangular with a response of 1 at the center frequency and linearly decreasing towards 0 until reaching a response of 0 at the center frequency of two adjacent filters. This process can be expressed by the following formula (3):
wherein the content of the first and second substances,
in the formula (3), H m (k) Is the coefficient of the filter, m is the mth filter, f is the frequency, and k is the abscissa.
In an embodiment, after filtering the frequency spectrum corresponding to each frame of audio sub-signal, the mel-frequency cepstrum coefficient of the audio signal may be obtained according to each frame of audio sub-signal after filtering. In a specific implementation, the obtaining of the mel-frequency cepstrum coefficient of the audio signal according to each frame of filtered audio sub-signal includes: and performing discrete cosine transform processing on each frame of audio sub-signal after filtering processing to obtain a Mel frequency spectrum cepstrum coefficient of the audio signal.
The discrete cosine transform processing is performed on each frame of filtered sub-signals in order to correlate the coefficients of the filter bank and obtain the compressed representation of the filter bank, so as to obtain the mel-frequency cepstrum coefficient (MFCC).
And S203, calling a fault detection model to analyze and process the Mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
After the mel-frequency cepstrum coefficient is obtained in step S202, the mel-frequency cepstrum coefficient may be input into a fault detection model for analysis processing, so as to obtain a detection result. The fault detection model may be a Convolutional Neural Network (CRNN) model. In the embodiment of the invention, the fault detection model is obtained by training based on a training set, the training set comprises a plurality of training audio data and a supervision label corresponding to each training audio data in the plurality of training audio data, the plurality of training audio data are obtained by preprocessing collected training audio signals, and the training audio signals are obtained by collecting sounds generated when the blades of the wind driven generator rotate through a microphone array; the supervision label corresponding to any training audio data is used for indicating that when the wind driven generator blade rotates to generate any training audio data, the wind driven generator blade is in a fault state or a non-fault state, and the fault state is any one or more of the following fault states: loss of blades and damage to blades.
In one embodiment, the fault detection device may output a detection result after acquiring the detection result of the fault detection of the wind turbine blade. Optionally, the outputting the detection result may refer to sending the detection result to a management device of a management person, and the management device displays the detection result, so that the management person can manage the wind turbine generator according to the detection result. Specifically, if the detection result shows that the blade of the wind driven generator is in a fault state, the manager can timely maintain the wind driven generator so as to quickly recover the normal work of the wind driven generator; if the detection result shows that the wind driven generator is in a non-fault state, the manager does not need to maintain the wind driven generator.
In the embodiment of the invention, when a triggering event for carrying out fault detection on the blade of the wind driven generator exists, an audio signal generated when the blade of the wind driven generator rotates is collected; further, a Mel frequency cepstrum coefficient of the audio signal is extracted, a fault detection model is called to analyze and process the Mel frequency cepstrum coefficient, and a detection result of fault detection on the wind driven generator blade is obtained. Therefore, in the fault detection process of the embodiment of the invention, the fault detection model is called to perform fault detection according to the audio signal, and compared with a physical detection method and a visual detection method in the prior art, the fault detection method has the advantages of low implementation cost and improvement of convenience of fault detection; moreover, the fault detection model is obtained based on a large amount of data training and has strong learning capacity, the Mel frequency cepstrum coefficient of the audio signal is analyzed through the fault detection model to obtain a detection result, and the detection result can reflect whether the wind driven generator blade has a fault or not more accurately.
Based on the above fault detection method, an embodiment of the present invention further provides another fault detection method, and referring to fig. 3, a flowchart of the another fault detection method provided in the embodiment of the present invention is shown. The fault detection method shown in fig. 3 may be performed by the fault detection device, in particular by a processor of the fault detection device. The fault detection device can be a terminal or a server, wherein the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart sound box, a smart watch, a smart vehicle-mounted device and the like; the server can be an independent physical server, can also be a server cluster or distributed system formed by a plurality of physical servers, and can also be a cloud server for providing basic cloud computing services such as cloud service, a cloud database, cloud computing, cloud functions, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN (content delivery network) and big data and artificial intelligence platforms. The fault detection method shown in fig. 3 may include the steps of:
step S301, a training set is obtained, wherein the training set comprises a plurality of training audio data and a supervision label corresponding to each training audio data in the plurality of training audio data.
In one embodiment, the plurality of training audio data included in the training set may be obtained by preprocessing the acquired training audio signal, and the preprocessing may be referred to as a partition processing. The training audio signal may include: the microphone array collects audio signals generated when the blades of the wind driven generator rotate. The term aerogenerator blade rotation as used herein may refer to rotation of an aerogenerator blade when in a non-faulted state, and may also include rotation of an aerogenerator blade when in a faulted state. When the wind driven generator blade rotates in a non-fault state, the obtained training audio signal can be regarded as a positive sample; when the wind turbine blade is rotating in a fault state, the obtained training audio signal can be regarded as a negative sample. The fault detection model is trained by the positive sample and the negative sample at the same time, so that the accuracy of the model can be improved.
In one embodiment, assuming that the training audio signal is represented as X (t), which is composed of M data points, the training audio signal is preprocessed to obtain a plurality of training audio data, which can be represented by the following formula (5):
X(t)＝[x 1 …x N ] (5)
in the formula (4), x 1 Representing the first training audio data, x N Represents the nth training audio data, that is, the number of the plurality of training audio data is N, wherein the calculation manner of N can be obtained by formula (6):
N＝{M/L} (6)
in equation (6), L represents the size of each audio piece.
In one embodiment, in order to improve the accuracy of the fault detection model, in the embodiment of the invention, supervised learning is adopted when the fault detection model is trained, and the supervised learning is to add a supervision tag to each training audio data. And the supervision label corresponding to each training audio data is used for indicating whether the wind driven generator blade is in a fault state or a non-fault state when the training audio data is obtained.
Step S302, extracting a Mel frequency cepstrum coefficient corresponding to each training audio data, inputting the Mel frequency cepstrum coefficient corresponding to each training audio data into a fault detection model for analysis processing, and obtaining a prediction result corresponding to each training audio data.
After the training set for training is obtained in step S301, the fault detection model may be trained based on the training set, so that the fault detection model may have the capability of determining whether the wind turbine blade has a fault according to the audio signal. In a specific implementation, model training may be performed through step S302 and step S303.
In an embodiment, the implementation of step S302 is the same as the implementation of step S202 in the embodiment of fig. 2, and reference may be specifically made to the description in the embodiment of fig. 2, which is not repeated herein.
In one embodiment, a schematic structural diagram of a fault detection model in an embodiment of the present invention is shown in fig. 4. The fault detection model shown in fig. 4 may include three convolutional layers, each convolutional layer corresponds to one pooling layer, the size of the convolutional kernel in the first convolutional layer 400 may be 3x3, the number of channels is 64, the size of the convolutional kernel in the second convolutional layer 401 is 3x3, the number of channels is 128, the size of the convolutional kernel in the third convolutional layer 402 is 3x3, and the number of channels is 256. The Mel frequency cepstrum coefficient of any training audio data is input into a fault detection model, each convolution layer of the fault detection model processes the Mel frequency cepstrum coefficient of the training audio data, and the input of the next convolution layer is the result of convolution processing of the previous convolution layer. One convolution layer may include a plurality of neurons, and the convolution layer performs convolution operations, actually the neurons perform the convolution operations, and the convolution operation process in one convolution layer may be expressed by the following formula: z = a 1 w 1 +…+a k w k + b, wherein b is a parameter, a i Data representing input of any one neuron, i is 1 or more and k or less, and w i And a parameter value representing the ith neuron, wherein i is greater than or equal to 1 and less than or equal to k.
In one embodiment, the fault detection model shown in fig. 4 further includes a feature processing module 403, where the feature processing module 403 may include a plurality of processing units 4030, and the feature processing module 403 is configured to perform feature extraction again on the feature map output by the last convolutional layer by using the correlation of the time series. Alternatively, as shown in 404 in fig. 4, the feature map input by the last convolutional layer may be converted into a feature sequence 405, and the feature sequence 405 is input to the feature processing module 403, so that the feature processing module 403 performs further feature extraction.
Finally, the feature processing module 403 outputs the processed feature sequence to the full connection layer 406 of the fault detection model, and 406 obtains the probability that the input training audio data has faults, and outputs the probability as a prediction result.
Optionally, in the fault check model shown in fig. 4, each convolution layer corresponds to an activation function, and the activation function may be represented as: f (x) = min (0, x + N (0, 1)), the activation function may be used to improve the non-linear capability of all convolutional layers in the fault detection model. And adding a normalization layer between each convolution layer and the corresponding activation function to normalize the input of the activation function. In the fault detection model shown in fig. 4, L2 regularization (penalty factor λ = 0.008) and dropout are used to prevent overfitting during model training.
Step S303, determining a target loss function according to the prediction result corresponding to each training audio data and the supervision label corresponding to each training audio data.
In one embodiment, the determining the target loss function according to the prediction result corresponding to each piece of training audio data and the supervised label corresponding to each piece of training audio data may include: calculating the deviation between the prediction result corresponding to each training audio data and the supervision label corresponding to the corresponding training audio data; an objective loss function is constructed from the plurality of deviations. For example, the target loss function can be expressed by the following formula (7):
in formula (6), loss represents the target Loss function, N represents the total number of the plurality of training audio data, y i Representing a supervision label corresponding to training audio data i, wherein i is greater than or equal to 1 and less than or equal to N; p is a radical of formula i Representing the probability that the predicted training audio data i corresponds to. It should be understood that the larger the value of the target loss function is, the worse the performance of the fault detection model is, and the fault detection model is not optimal; conversely, the smaller the value of the target loss function is, the better the performance of the fault detection model is.
Step S304, optimizing the fault detection model according to the direction of reducing the value of the target loss function.
To minimize the loss, embodiments of the present invention optimize the fault detection model in a direction that reduces the value of the objective loss function. Specifically, parameter optimization can be performed by adopting an Adam optimizer, an Adam algorithm designs independent self-adaptive learning rates for different parameters by calculating first-order matrix and second-order matrix estimation of gradient, bias derivation is performed on the weight values of the neurons of the fault detection model by reverse propagation, and the weight values of the network neurons are continuously updated under the optimal solution.
Step S305, in response to a trigger event for detecting the fault of the blade of the wind driven generator, acquiring an audio signal generated when the blade of the wind driven generator rotates.
And S306, extracting the Mel frequency cepstrum coefficient of the audio signal, calling the optimized fault detection model to analyze and process the Mel frequency cepstrum coefficient, and obtaining a detection result of fault detection on the wind driven generator blade.
After the training of the fault detection model is finished, in response to a trigger event for carrying out fault detection on the blade of the wind driven generator, the microphone collects an audio signal when the blade of the wind driven generator is currently running, a Mel frequency cepstrum coefficient of the audio signal is extracted, the Mel frequency cepstrum coefficient of the audio signal is input into the trained fault detection model for detection, and the probability that the blade of the wind driven generator has a fault at present is output.
In a specific implementation, reference may be made to the description of the relevant steps in the embodiment of fig. 2 for specific implementation of step S305 and step S306, and details are not repeated here.
In one embodiment, in order to verify the accuracy of the trained fault detection model during fault detection, the accuracy of the fault detection model may be verified before the fault detection model is put into use. In a specific implementation, a verification data set may be used to verify the fault detection model. Then, the verified accuracy and the trained accuracy are compared, the specific comparison result can be seen in fig. 5, 501 in fig. 5 represents the trained accuracy, 502 represents the verified accuracy, and thus the difference between the two is small, and in this case, it is determined that the fault detection model can be put into use.
In an embodiment, in order to verify that the accuracy of the fault detection model constructed by using the CRNN in the embodiment of the present invention is higher than that of the fault detection model constructed based on other Neural Networks, the embodiment of the present invention compares the accuracy of the fault detection model constructed based on the CRNN with the accuracy of the fault detection model constructed based on the Convolutional Neural Network (CNN) in fault detection. The accuracy rate of the fault detection model constructed based on the CRNN can reach 99.6 percent when fault detection is carried out, and the accuracy rate of the fault detection model constructed based on the CNN can reach 98.5 percent when fault detection is carried out.
In the embodiment of the invention, a plurality of training audio data and a supervision label corresponding to each training audio data in the plurality of training audio data are obtained; then, extracting a Mel frequency cepstrum coefficient corresponding to each training audio data, and inputting the Mel frequency cepstrum coefficient corresponding to each training audio data into a fault detection model for analysis processing; obtaining a prediction result corresponding to each training audio data; further, a target loss function is determined according to the prediction result corresponding to each piece of training audio data and the supervision label corresponding to each piece of training audio data, and the fault detection model is optimized according to the direction of reducing the value of the target loss function. It should be appreciated that training the fault detection model with a large amount of training data may improve the accuracy of the fault detection model.
After the fault detection model converges, the audio signal of the current wind driven generator blade during operation can be collected, and the Mel frequency cepstrum coefficient corresponding to the audio signal is input into the converged fault detection model, so that the fault detection model analyzes and processes the Mel frequency cepstrum coefficient of the audio signal, and outputs a detection result for fault detection of the current wind driven generator blade. Compared with the prior art that a physical method is adopted for fault detection, the method adopting the passive acoustic signal in the embodiment of the invention can greatly reduce the detection cost and is less influenced by environmental interference noise; compared with the fault detection method adopting a visual method in the prior art, the fault detection method provided by the embodiment of the invention has the advantages of lower complexity and higher efficiency. And the fault detection model has strong characteristic learning capability, and the trained fault detection model can effectively identify fault information, so that the accuracy of fault detection of the wind driven generator can be improved.
Based on the above method embodiment, the embodiment of the present invention provides a fault detection apparatus. Fig. 6 is a schematic structural diagram of a fault detection apparatus according to an embodiment of the present invention. The fault detection arrangement shown in fig. 6 may operate as follows:
the acquisition unit 601 is used for responding to a trigger event for fault detection of the wind power generation blade and adopting an audio signal generated when the wind power generation blade rotates;
an extracting unit 602, configured to extract mel-frequency cepstrum coefficients of the audio signal;
and the processing unit 603 is configured to invoke a fault detection model to analyze and process the mel-frequency cepstrum coefficient, so as to obtain a detection result of performing fault detection on the wind turbine blade.
In one embodiment, the extracting unit 602 performs the following steps when extracting mel-frequency spectrum cepstrum coefficients of the audio signal:
pre-emphasis processing is carried out on the audio signal, and frame division processing is carried out on the audio signal after the pre-emphasis processing, so that multi-frame audio sub-signals are obtained; overlapping signals with preset duration exist between two adjacent frame audio frequency sub-signals;
carrying out short-time Fourier transform operation on each frame of audio sub-signals in the multi-frame audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals;
and filtering the frequency spectrum corresponding to each frame of audio sub-signal, and obtaining a Mel frequency cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering.
In an embodiment, when the extracting unit 602 performs a short-time fourier transform operation on each frame of audio sub-signals in the multiple frames of audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals, the following steps are performed:
adding a Hamming window to each frame of audio sub-signal;
and inputting each frame of audio sub-signal added with the Hamming window into a short-time Fourier transform rule for operation to obtain a frequency spectrum corresponding to each frame of audio sub-signal.
In one embodiment, when obtaining the mel-frequency spectrum cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering processing, the extracting unit 602 performs the following steps:
and carrying out discrete cosine transform processing on each frame of audio sub-signal after filtering processing to obtain a Mel frequency spectrum cepstrum coefficient of the audio signal.
In one embodiment, the fault detection model is trained based on a training set, the training set includes a plurality of training audio data and a supervised label corresponding to each of the plurality of training audio data, the plurality of training audio data is obtained by preprocessing collected training audio signals, and the training audio signals include audio signals collected by a microphone array and generated when the wind turbine blades rotate; the supervision label corresponding to any training audio data is used for indicating that when the wind driven generator blade rotates to generate any training audio data, the wind driven generator blade is in a fault state or a non-fault state, and the fault state is any one or more of the following fault states: blade loss and blade damage.
In one embodiment, the processing unit 603 is further configured to:
acquiring a training set, wherein the training set comprises a plurality of training audio data and a supervision label corresponding to each training audio data in the plurality of training audio data;
extracting a Mel frequency cepstrum coefficient corresponding to each training audio data, and inputting the Mel frequency cepstrum coefficient corresponding to each training audio data into the fault detection model for analysis processing to obtain a prediction result corresponding to each training audio data;
determining a target loss function according to the prediction result corresponding to each training audio data and the supervision label corresponding to each training audio data;
optimizing the fault detection model in a direction that decreases a value of the objective loss function.
In one embodiment, the triggering event refers to that the current time meets a preset time condition for fault detection of the wind turbine blade; or the triggering event is the reception of a detection instruction for fault detection of the wind turbine blade unit.
According to an embodiment of the present invention, the steps involved in the fault detection methods shown in fig. 2 and 3 may be performed by units in the fault detection apparatus shown in fig. 6. For example, step S201 shown in fig. 2 may be performed by the acquisition unit 601 in the fault detection apparatus shown in fig. 6, step S202 may be performed by the extraction unit 602 in the fault detection apparatus shown in fig. 6, and step S203 may be performed by the processing unit 603 in the fault detection apparatus shown in fig. 6. As another example, steps S301 to S304 shown in fig. 3 may be performed by the processing unit 603 in the failure detection apparatus shown in fig. 6, step S305 may be performed by the acquisition unit 601 and the extraction unit 602 in the failure detection apparatus shown in fig. 6, and step S306 may be performed by the processing unit 603 in the failure detection apparatus shown in fig. 6.
According to another embodiment of the present invention, the units in the fault detection apparatus shown in fig. 6 may be respectively or entirely combined into one or several other units to form another unit, or some unit(s) may be further split into multiple functionally smaller units to form another unit, which may implement the same operation without affecting the implementation of the technical effect of the embodiment of the present invention. The units are divided based on logic functions, and in practical application, the functions of one unit can be realized by a plurality of units, or the functions of a plurality of units can be realized by one unit. In other embodiments of the present invention, the fault detection device may also include other units, and in practical applications, these functions may also be implemented by assistance of other units, and may be implemented by cooperation of a plurality of units.
According to another embodiment of the present invention, the fault detection apparatus as shown in fig. 6 may be constructed by running a computer program (including program codes) capable of executing the steps involved in the respective methods as shown in fig. 2 and 3 on a general-purpose computing device, such as a computer, including a Central Processing Unit (CPU), a random access storage medium (RAM), a read-only storage medium (ROM), and the like, processing elements and storage elements, and implementing the fault detection method of the embodiment of the present invention. The computer program may be, for example, embodied on a computer-readable storage medium, and loaded into and executed in the above-described computing apparatus via the computer-readable storage medium.
In the embodiment of the invention, when a triggering event for carrying out fault detection on the blade of the wind driven generator exists, an audio signal generated when the blade of the wind driven generator rotates is collected; further, a Mel frequency cepstrum coefficient of the audio signal is extracted, a fault detection model is called to analyze and process the Mel frequency cepstrum coefficient, and a detection result of fault detection on the wind driven generator blade is obtained. Therefore, in the fault detection process of the embodiment of the invention, the fault detection model is called to perform fault detection according to the audio signal, and compared with a physical detection method and a visual detection method in the prior art, the fault detection method has the advantages of low implementation cost and improvement on the convenience of fault detection; and the fault detection model is obtained based on a large amount of data training, has strong learning capacity, and analyzes the Mel frequency cepstrum coefficient of the audio signal through the fault detection model to obtain a detection result which can more accurately reflect whether the blade of the wind driven generator has a fault.
Based on the above method embodiment and apparatus embodiment, an embodiment of the present invention further provides a fault detection device, and referring to fig. 7, a schematic structural diagram of a fault detection device provided in an embodiment of the present invention is shown. The fault detection device shown in fig. 7 may include at least a processor 701, an input interface 702, an output interface 703, and a computer storage medium 704. The processor 701, the input interface 702, the output interface 703, and the computer storage medium 704 may be connected by a bus or other means.
A computer storage medium 704 may be stored in the memory of the fault detection device, said computer storage medium 704 being adapted to store a computer program comprising a program computer program, said processor 701 being adapted to execute the program computer program stored by said computer storage medium 704. The processor 701 (or CPU) is a computing core and a control core of the fault detection device, and is adapted to implement one or more computer programs, and specifically adapted to load and execute:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates; extracting mel frequency cepstrum coefficients of the audio signal; and calling a fault detection model to analyze and process the Mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In the embodiment of the invention, when a trigger event for carrying out fault detection on the blade of the wind driven generator exists, an audio signal generated when the blade of the wind driven generator rotates is acquired; further, a Mel frequency cepstrum coefficient of the audio signal is extracted, a fault detection model is called to analyze and process the Mel frequency cepstrum coefficient, and a detection result of fault detection of the wind driven generator blade is obtained. Therefore, in the fault detection process of the embodiment of the invention, the fault detection model is called to perform fault detection according to the audio signal, and compared with a physical detection method and a visual detection method in the prior art, the fault detection method has the advantages of low implementation cost and improvement of convenience of fault detection; moreover, the fault detection model is obtained based on a large amount of data training and has strong learning capacity, the Mel frequency cepstrum coefficient of the audio signal is analyzed through the fault detection model to obtain a detection result, and the detection result can reflect whether the wind driven generator blade has a fault or not more accurately.
The embodiment of the invention also provides a computer storage medium (Memory), which is a Memory device in the fault detection device and is used for storing programs and data. It is understood that the computer storage medium herein may include a built-in storage medium of the failure detection device, and may also include an extended storage medium supported by the failure detection device. The computer storage medium provides a memory space in which an operating system of the fault detection device is stored and in which one or more computer programs, which may be one or more computer programs (including program code), are stored that are adapted to be loaded and executed by the processor 701. The computer storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory; and optionally at least one computer storage medium located remotely from the processor.
In one embodiment, the computer storage medium when loaded by processor 701 performs the following: in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates; extracting mel frequency cepstrum coefficients of the audio signal; and calling a fault detection model to analyze and process the Mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In one embodiment, the processor 701, when extracting mel-frequency spectrum cepstrum coefficients of the audio signal, performs the following steps:
pre-emphasis processing is carried out on the audio signal, and frame division processing is carried out on the audio signal after the pre-emphasis processing, so that multi-frame audio sub-signals are obtained; overlapping signals with preset duration exist between two adjacent frame audio sub-signals;
carrying out short-time Fourier transform operation on each frame of audio sub-signals in the multi-frame audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals;
and filtering the frequency spectrum corresponding to each frame of audio sub-signal, and obtaining a Mel frequency cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering.
In an embodiment, when the processor 701 performs a short-time fourier transform operation on each frame of audio sub-signals in the multiple frames of audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals, the following steps are performed:
adding a Hamming window to each frame of audio sub-signal;
and inputting each frame of audio sub-signal added with the Hamming window into a short-time Fourier transform rule for operation to obtain a frequency spectrum corresponding to each frame of audio sub-signal.
In one embodiment, when obtaining the mel-frequency spectrum cepstrum coefficient of the audio signal according to each frame of audio sub-signal after the filtering process, the processor 701 performs the following steps:
and carrying out discrete cosine transform processing on each frame of audio sub-signal after filtering processing to obtain a Mel frequency spectrum cepstrum coefficient of the audio signal.
In one embodiment, the fault detection model is trained based on a training set, the training set includes a plurality of training audio data and a supervision tag corresponding to each of the plurality of training audio data, the plurality of training audio data is obtained by preprocessing collected training audio signals, and the training audio signals include audio signals collected by a microphone array and generated when the wind turbine blades rotate; the supervision label corresponding to any training audio data is used for indicating that when the wind driven generator blade rotates to generate any training audio data, the wind driven generator blade is in a fault state or a non-fault state, and the fault state is any one or more of the following fault states: blade loss and blade damage.
In one embodiment, the processor 701 is further configured to:
acquiring a training set, wherein the training set comprises a plurality of training audio data and a supervision label corresponding to each of the plurality of training audio data;
extracting a Mel frequency cepstrum coefficient corresponding to each training audio data, and inputting the Mel frequency cepstrum coefficient corresponding to each training audio data into the fault detection model for analysis processing to obtain a prediction result corresponding to each training audio data;
determining a target loss function according to the prediction result corresponding to each training audio data and the supervision label corresponding to each training audio data;
optimizing the fault detection model in a direction that decreases a value of the objective loss function.
In one embodiment, the trigger event refers to that the current time meets a preset time condition for fault detection of the wind turbine blade; or the triggering event refers to the fact that a detection instruction for detecting the fault of the wind power generation blade machine is received.
In the embodiment of the invention, when a triggering event for carrying out fault detection on the blade of the wind driven generator exists, an audio signal generated when the blade of the wind driven generator rotates is collected; further, a Mel frequency cepstrum coefficient of the audio signal is extracted, a fault detection model is called to analyze and process the Mel frequency cepstrum coefficient, and a detection result of fault detection on the wind driven generator blade is obtained. Therefore, in the fault detection process of the embodiment of the invention, the fault detection model is called to perform fault detection according to the audio signal, and compared with a physical detection method and a visual detection method in the prior art, the fault detection method has the advantages of low implementation cost and improvement of convenience of fault detection; moreover, the fault detection model is obtained based on a large amount of data training and has strong learning capacity, the Mel frequency cepstrum coefficient of the audio signal is analyzed through the fault detection model to obtain a detection result, and the detection result can reflect whether the wind driven generator blade has a fault or not more accurately.
According to an aspect of the present application, the embodiment of the present invention further provides a computer program product or a computer program, the computer program product comprising a computer program, the computer program being stored in a computer readable storage medium. The processor 701 reads the computer program from the computer-readable storage medium, and the processor 701 executes the computer program, so that the fault detection apparatus performs the fault detection method described in fig. 2 and 3, specifically:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates; extracting mel frequency cepstrum coefficients of the audio signal; and calling a fault detection model to analyze and process the Mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
In the embodiment of the invention, when a trigger event for carrying out fault detection on the blade of the wind driven generator exists, an audio signal generated when the blade of the wind driven generator rotates is acquired; further, a Mel frequency cepstrum coefficient of the audio signal is extracted, a fault detection model is called to analyze and process the Mel frequency cepstrum coefficient, and a detection result of fault detection on the wind driven generator blade is obtained. Therefore, in the fault detection process of the embodiment of the invention, the fault detection model is called to perform fault detection according to the audio signal, and compared with a physical detection method and a visual detection method in the prior art, the fault detection method has the advantages of low implementation cost and improvement on the convenience of fault detection; moreover, the fault detection model is obtained based on a large amount of data training and has strong learning capacity, the Mel frequency cepstrum coefficient of the audio signal is analyzed through the fault detection model to obtain a detection result, and the detection result can reflect whether the wind driven generator blade has a fault or not more accurately.
Claims (10)
1. A method of fault detection, comprising:
in response to a triggering event for fault detection of a wind turbine blade, acquiring an audio signal generated when the wind turbine blade rotates;
extracting mel frequency cepstrum coefficients of the audio signal;
and calling a fault detection model to analyze and process the mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
2. The method of claim 1, wherein said extracting mel-frequency spectrum cepstrum coefficients of the audio signal comprises:
pre-emphasis processing is carried out on the audio signal, and frame division processing is carried out on the audio signal after the pre-emphasis processing, so that multi-frame audio sub-signals are obtained; overlapping signals with preset duration exist between two adjacent frame audio sub-signals;
performing short-time Fourier transform operation on each frame of audio sub-signals in the multi-frame audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals;
and filtering the frequency spectrum corresponding to each frame of audio sub-signal, and obtaining a Mel frequency cepstrum coefficient of the audio signal according to each frame of audio sub-signal after filtering.
3. The method as claimed in claim 2, wherein said performing a short-time fourier transform operation on each frame of audio sub-signals in the plurality of frames of audio sub-signals to obtain a frequency spectrum corresponding to each frame of audio sub-signals comprises:
adding a Hamming window to each frame of audio sub-signal;
and inputting each frame of audio sub-signal added with the Hamming window into a short-time Fourier transform rule for operation to obtain a frequency spectrum corresponding to each frame of audio sub-signal.
4. The method as claimed in claim 2, wherein said deriving mel-frequency spectrum cepstrum coefficients of said audio signal from each frame of said filtered audio sub-signals comprises:
and carrying out discrete cosine transform processing on each frame of audio sub-signal after filtering processing to obtain a Mel frequency spectrum cepstrum coefficient of the audio signal.
5. The method of claim 1, wherein the fault detection model is trained based on a training set, the training set including a plurality of training audio data and a supervised label corresponding to each of the plurality of training audio data, the plurality of training audio data being obtained by preprocessing acquired training audio signals, the training audio signals including audio signals generated when the wind turbine blades are rotated and acquired by a microphone array; the supervision label corresponding to any training audio data is used for indicating that when the wind driven generator blade rotates to generate any training audio data, the wind driven generator blade is in a fault state or a non-fault state, and the fault state is any one or more of the following fault states: loss of blades and damage to blades.
6. The method of claim 1, wherein the method further comprises:
acquiring a training set, wherein the training set comprises a plurality of training audio data and a supervision label corresponding to each of the plurality of training audio data;
extracting a Mel frequency cepstrum coefficient corresponding to each training audio data, and inputting the Mel frequency cepstrum coefficient corresponding to each training audio data into the fault detection model for analysis processing to obtain a prediction result corresponding to each training audio data;
determining a target loss function according to the prediction result corresponding to each training audio data and the supervision label corresponding to each training audio data;
optimizing the fault detection model in a direction that decreases a value of the objective loss function.
7. The method according to claim 1, wherein the triggering event is that a current time meets a preset time condition for fault detection of the wind turbine blade; or the triggering event is the reception of a detection instruction for fault detection of the wind turbine blade unit.
8. A fault detection device, comprising:
the system comprises an acquisition unit, a control unit and a control unit, wherein the acquisition unit is used for responding to a trigger event for carrying out fault detection on a wind power generation blade and adopting an audio signal generated when the wind power generation blade rotates;
an extraction unit for extracting mel-frequency cepstrum coefficients of the audio signal;
and the processing unit is used for calling a fault detection model to analyze and process the mel frequency cepstrum coefficient to obtain a detection result of fault detection of the wind driven generator blade.
9. A fault detection device, comprising:
a processor adapted to implement one or more computer programs; and
computer storage medium storing one or more computer programs adapted to be loaded by the processor and to perform the fault detection method according to any of claims 1-6.
10. A computer-readable storage medium, characterized in that the computer storage medium stores a computer program for performing the fault detection method according to any one of claims 1-6 when the computer program is executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110586205.2A CN115406630A (en) | 2021-05-27 | 2021-05-27 | Method for detecting faults of wind driven generator blades through passive acoustic signals based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110586205.2A CN115406630A (en) | 2021-05-27 | 2021-05-27 | Method for detecting faults of wind driven generator blades through passive acoustic signals based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115406630A true CN115406630A (en) | 2022-11-29 |
Family
ID=84155010
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110586205.2A Pending CN115406630A (en) | 2021-05-27 | 2021-05-27 | Method for detecting faults of wind driven generator blades through passive acoustic signals based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115406630A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116403605A (en) * | 2023-06-08 | 2023-07-07 | 宁德时代新能源科技股份有限公司 | Equipment fault prediction method, stacker fault prediction method and related devices |
| CN116403605B (en) * | 2023-06-08 | 2024-06-07 | 宁德时代新能源科技股份有限公司 | Stacker fault prediction method and related device |
-
2021
- 2021-05-27 CN CN202110586205.2A patent/CN115406630A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116403605A (en) * | 2023-06-08 | 2023-07-07 | 宁德时代新能源科技股份有限公司 | Equipment fault prediction method, stacker fault prediction method and related devices |
| CN116403605B (en) * | 2023-06-08 | 2024-06-07 | 宁德时代新能源科技股份有限公司 | Stacker fault prediction method and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112783940B (en) | Multi-source time sequence data fault diagnosis method and medium based on graph neural network | |
| CN111325095B (en) | Intelligent detection method and system for equipment health state based on acoustic wave signals | |
| Pashazadeh et al. | Data driven sensor and actuator fault detection and isolation in wind turbine using classifier fusion | |
| US11418029B2 (en) | Method for recognizing contingencies in a power supply network | |
| Cao et al. | Fault diagnosis of wind turbine gearbox based on deep bi-directional long short-term memory under time-varying non-stationary operating conditions | |
| KR100918313B1 (en) | Method for diagnosis and analysis of electric power quality using artificial intelligence | |
| CN108444589B (en) | Hydroelectric generating set state monitoring signal processing method based on frequency domain feature extraction | |
| CN113670434B (en) | Method and device for identifying sound abnormality of substation equipment and computer equipment | |
| Afrasiabi et al. | Wind turbine fault diagnosis with generative-temporal convolutional neural network | |
| CN107644231A (en) | A kind of generator amature method for diagnosing faults and device | |
| CN113639993A (en) | Gearbox fault diagnosis method of multi-mode multi-task convolutional neural network | |
| CN115409052A (en) | Fault diagnosis method and system for wind generating set bearing under data imbalance | |
| CN115406630A (en) | Method for detecting faults of wind driven generator blades through passive acoustic signals based on machine learning | |
| CN113746780A (en) | Abnormal host detection method, device, medium and equipment based on host image | |
| FR2831981A1 (en) | METHOD AND DEVICE FOR ANALYZING ALARMS FROM A COMMUNICATION NETWORK | |
| CN110020685A (en) | A kind of preprocess method based on adaptive-filtering and limited Boltzmann machine, terminal and readable storage medium storing program for executing | |
| CN116825131A (en) | Power plant equipment state auditory monitoring method integrating frequency band self-downward attention mechanism | |
| CN115296933A (en) | Industrial production data risk level assessment method and system | |
| US20220187789A1 (en) | Equipment failure diagnosis apparatus, equipment failure diagnosis method, smart factory system and application agent | |
| CN113919525A (en) | Power station fan state early warning method, system and application thereof | |
| CN114121025A (en) | Voiceprint fault intelligent detection method and device for substation equipment | |
| Bhattacharya et al. | Energy efficient fault detection and classification using hyperparameter-tuned machine learning classifiers with sensors | |
| Chacon et al. | Ensembles Learning Algorithms with K-Fold Cross Validation to Detect False Alarms in Wind Turbines | |
| CN116624342A (en) | Method and system for diagnosing and early warning state of generator of wind turbine generator | |
| CN117725402B (en) | Method and system for determining abnormal state of equipment based on equipment operation audio |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |