WO2022142213A1 - 叶片故障诊断方法、装置、系统及存储介质 - Google Patents

叶片故障诊断方法、装置、系统及存储介质 Download PDF

Info

Publication number
WO2022142213A1
WO2022142213A1 PCT/CN2021/103439 CN2021103439W WO2022142213A1 WO 2022142213 A1 WO2022142213 A1 WO 2022142213A1 CN 2021103439 W CN2021103439 W CN 2021103439W WO 2022142213 A1 WO2022142213 A1 WO 2022142213A1
Authority
WO
WIPO (PCT)
Prior art keywords
audio
blade
wind noise
blade rotation
fault diagnosis
Prior art date
Application number
PCT/CN2021/103439
Other languages
English (en)
French (fr)
Inventor
赵勇
李新乐
牛馨苑
Original Assignee
北京金风科创风电设备有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 北京金风科创风电设备有限公司 filed Critical 北京金风科创风电设备有限公司
Priority to CA3203703A priority Critical patent/CA3203703A1/en
Priority to EP21912969.9A priority patent/EP4254261A4/en
Priority to KR1020237022434A priority patent/KR20230113384A/ko
Priority to US18/260,075 priority patent/US20240052810A1/en
Priority to AU2021415086A priority patent/AU2021415086B2/en
Publication of WO2022142213A1 publication Critical patent/WO2022142213A1/zh

Links

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
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H3/00—Measuring characteristics of vibrations by using a detector in a fluid
    • G01H3/04—Frequency
    • G01H3/08—Analysing frequencies present in complex vibrations, e.g. comparing harmonics present
    • 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
    • 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
    • F03D17/005—Monitoring or testing of wind motors, e.g. diagnostics using computation methods, e.g. neural networks
    • 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
    • F03D17/009—Monitoring or testing of wind motors, e.g. diagnostics characterised by the purpose
    • F03D17/024—Monitoring or testing of wind motors, e.g. diagnostics characterised by the purpose for monitoring sound or acoustic signals
    • 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
    • F03D17/027—Monitoring or testing of wind motors, e.g. diagnostics characterised by the component being monitored or tested
    • F03D17/028—Blades
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H15/00—Measuring mechanical or acoustic impedance
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M5/00—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings
    • G01M5/0016—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings of aircraft wings or blades
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M5/00—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings
    • G01M5/0066—Investigating the elasticity of structures, e.g. deflection of bridges or air-craft wings by exciting or detecting vibration or acceleration
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M99/00—Subject matter not provided for in other groups of this subclass
    • G01M99/005—Testing of complete machines, e.g. washing-machines or mobile phones
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/10—Pre-processing; Data cleansing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05B—INDEXING SCHEME RELATING TO WIND, SPRING, WEIGHT, INERTIA OR LIKE MOTORS, TO MACHINES OR ENGINES FOR LIQUIDS COVERED BY SUBCLASSES F03B, F03D AND F03G
    • F05B2260/00—Function
    • F05B2260/80—Diagnostics
    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
    • F05B—INDEXING SCHEME RELATING TO WIND, SPRING, WEIGHT, INERTIA OR LIKE MOTORS, TO MACHINES OR ENGINES FOR LIQUIDS COVERED BY SUBCLASSES F03B, F03D AND F03G
    • F05B2260/00—Function
    • F05B2260/84—Modelling or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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
    • 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

Definitions

  • the present application belongs to the technical field of wind power generation, and in particular, relates to a blade fault diagnosis method, device, system and storage medium.
  • the blades of the wind turbine will emit sound when they rotate.
  • the audio frequency of the blade rotation will change to a certain extent. Therefore, when monitoring the blade status of the wind turbine, an implementation is as follows: Collect the audio of blade rotation, and then judge whether the blade of the wind turbine is faulty according to the audio signal.
  • the inventor found that since the audio signal changes little when the blade fails, and the audio signals of different blades are slightly different, the existing method of diagnosing whether the blade is faulty through audio may cause misdiagnosis.
  • Embodiments of the present application provide a blade fault diagnosis method, device, system, and storage medium, which can respectively diagnose whether a corresponding blade is faulty according to audio clips of different blades, thereby improving the accuracy of diagnosis results.
  • an embodiment of the present application provides a method for diagnosing blade faults.
  • the method includes: acquiring the blade rotation audio collected by an audio collection device during the operation of the wind turbine; preprocessing the blade rotation audio based on a wind noise filtering algorithm, Obtain the blade rotation audio after filtering out wind noise; Divide the blade rotation audio after filtering out wind noise to obtain the audio segment corresponding to each blade; According to each audio segment, diagnose whether the blade corresponding to each audio segment is faulty .
  • segmenting the blade rotation audio from which wind noise has been filtered out to obtain an audio segment corresponding to each blade including: processing the blade rotation audio after filtering out wind noise through short-time Fourier transform processing to obtain a first eigenvalue.
  • the first eigenvalue is used to characterize the frequency domain feature of the blade rotation audio after filtering out wind noise; the first eigenvalue is input into the blade recognition model to obtain the segmentation time point of the blade rotation audio after filtering out the wind noise, wherein , the blade recognition model is a pre-trained model, which is used to identify the switching time points of the rotation sounds of different blades according to the first eigenvalue; the blade rotation audio after filtering out the wind noise is divided according to the segmentation time point, so as to obtain the corresponding sound of each blade. audio clip.
  • diagnosing whether the corresponding blade is faulty according to each audio segment includes: processing each audio segment through Fourier transform to obtain a second eigenvalue of each audio segment, where the second eigenvalue is used to represent The frequency domain feature of each audio segment; the second feature value of each audio segment is input into the blade fault diagnosis model respectively to obtain the first fault diagnosis result of each blade, and the blade fault diagnosis model is pre-trained and used according to The second eigenvalue of the audio segment identifies the model for whether the corresponding blade is faulty.
  • diagnosing whether the corresponding blade is faulty includes: counting the duration of the audio clip corresponding to each blade; For the faulty blade, a second fault diagnosis result is obtained; combining the first fault diagnosis result and the second fault diagnosis result of each blade, it is judged whether there is a faulty blade.
  • the method before diagnosing whether the corresponding blade is faulty according to each audio clip, the method further includes: acquiring environmental parameters of the wind turbine, wherein the environmental parameters are used to represent seasons and/or weather; Among multiple candidate fault diagnosis models for identifying different fault types, a model for identifying fault types corresponding to environmental parameters is determined, so as to determine a fault diagnosis model to be used.
  • acquiring the blade rotation audio collected by the audio collection device during the operation of the wind turbine includes: acquiring the blade rotation audio collected at multiple locations; preprocessing the blade rotation audio based on a wind noise filtering algorithm to obtain The blade rotation audio after filtering out wind noise includes: performing wind noise filtering algorithm processing on each blade rotation audio to obtain a plurality of blade rotation audio after filtering wind noise; after obtaining a plurality of blade rotation audio after filtering wind noise After rotating the audio frequency, the method further includes: calculating the wind noise parameters of each blade rotation audio after filtering out the wind noise through the wind noise identification model, wherein the wind noise parameters are used to represent the wind noise in the audio frequency, and the wind noise identification model It is a pre-trained model for evaluating the wind noise parameters of audio; the blade rotation audio after filtering out wind noise is segmented to obtain the audio segment corresponding to each blade, including: In the noise of the blade rotation audio, select the blade rotation audio with the smallest wind noise after filtering out the wind noise for segmentation, and obtain the audio segment corresponding to
  • calculating the wind noise parameters of each blade rotation audio after filtering out the wind noise through the wind noise identification model includes: processing each blade rotation audio after filtering out the wind noise by Fourier transform, to obtain each blade rotation audio.
  • a third eigenvalue of the blade rotation audio after filtering out wind noise, and the third eigenvalue is used to represent the corresponding frequency domain feature of the blade rotation audio after filtering out wind noise;
  • the third eigenvalue of the rotating audio is input into the wind noise identification model, and the wind noise parameters of each blade rotating audio after filtering out the wind noise are obtained.
  • an embodiment of the present application provides a blade fault diagnosis device, the device includes: an acquisition module for acquiring the blade rotation audio collected by the audio acquisition device during the operation of the wind turbine; a preprocessing module, which uses Based on the preprocessing of the blade rotation audio based on the wind noise filtering algorithm, the blade rotation audio after filtering out the wind noise is obtained; the segmentation module is used to segment the blade rotation audio after filtering the wind noise, and obtain the audio segment corresponding to each blade. ; The diagnosis module is used for diagnosing whether the blade corresponding to each audio clip is faulty according to each audio clip.
  • segmenting the blade rotation audio from which wind noise has been filtered out to obtain an audio segment corresponding to each blade including: processing the blade rotation audio after filtering out wind noise through short-time Fourier transform processing to obtain a first eigenvalue.
  • the first eigenvalue is used to characterize the frequency domain feature of the blade rotation audio after filtering out wind noise; the first eigenvalue is input into the blade recognition model to obtain the segmentation time point of the blade rotation audio after filtering out the wind noise, wherein , the blade recognition model is a pre-trained model, which is used to identify the switching time points of the rotation sounds of different blades according to the first eigenvalue; the blade rotation audio after filtering out the wind noise is divided according to the segmentation time point, so as to obtain the corresponding sound of each blade. audio clip.
  • diagnosing whether the corresponding blade is faulty according to each audio segment includes: processing each audio segment through Fourier transform to obtain a second eigenvalue of each audio segment, where the second eigenvalue is used to represent The frequency domain feature of each audio segment; the second feature value of each audio segment is input into the blade fault diagnosis model respectively to obtain the first fault diagnosis result of each blade, and the blade fault diagnosis model is pre-trained and used according to The second eigenvalue of the audio segment identifies the model for whether the corresponding blade is faulty.
  • diagnosing whether the corresponding blade is faulty includes: counting the duration of the audio clip corresponding to each blade; For the faulty blade, a second fault diagnosis result is obtained; combining the first fault diagnosis result and the second fault diagnosis result of each blade, it is judged whether there is a faulty blade.
  • the method before diagnosing whether the corresponding blade is faulty according to each audio clip, the method further includes: acquiring environmental parameters of the wind turbine, wherein the environmental parameters are used to represent seasons and/or weather; Among multiple candidate fault diagnosis models for identifying different fault types, a model for identifying fault types corresponding to environmental parameters is determined, so as to determine a fault diagnosis model to be used.
  • acquiring the blade rotation audio collected by the audio collection device during the operation of the wind turbine includes: acquiring the blade rotation audio collected at multiple locations; preprocessing the blade rotation audio based on a wind noise filtering algorithm to obtain The blade rotation audio after filtering out wind noise includes: performing wind noise filtering algorithm processing on each blade rotation audio to obtain a plurality of blade rotation audio after filtering wind noise; after obtaining a plurality of blade rotation audio after filtering wind noise After rotating the audio frequency, the method further includes: calculating the wind noise parameters of each blade rotation audio after filtering out the wind noise through the wind noise identification model, wherein the wind noise parameters are used to represent the wind noise in the audio frequency, and the wind noise identification model It is a pre-trained model for evaluating the wind noise parameters of audio; the blade rotation audio after filtering out wind noise is segmented to obtain the audio segment corresponding to each blade, including: In the noise of the blade rotation audio, select the blade rotation audio with the smallest wind noise after filtering out the wind noise for segmentation, and obtain the audio segment corresponding to
  • calculating the wind noise parameters of each blade rotation audio after filtering out the wind noise through the wind noise identification model includes: processing each blade rotation audio after filtering out the wind noise by Fourier transform, to obtain each blade rotation audio.
  • a third eigenvalue of the blade rotation audio after filtering out wind noise, and the third eigenvalue is used to represent the corresponding frequency domain feature of the blade rotation audio after filtering out wind noise;
  • the third eigenvalue of the rotating audio is input into the wind noise identification model, and the wind noise parameters of each blade rotating audio after filtering out the wind noise are obtained.
  • an embodiment of the present application provides a blade fault diagnosis system, the system includes: an audio collection device, including an audio sensor, wherein at least one audio sensor is disposed at the main wind direction position or the leeward direction of the tower of the wind turbine. Position; processor, arranged inside the tower, connected with the audio sensor, used to receive the blade rotation audio collected by the audio sensor, preprocess the blade rotation audio based on the wind noise filtering algorithm, and obtain the blade rotation after filtering the wind noise Audio; segment the blade rotation audio after filtering out wind noise to obtain audio clips corresponding to each blade; according to each audio clip, diagnose whether the blade corresponding to each audio clip is faulty.
  • an audio collection device including an audio sensor, wherein at least one audio sensor is disposed at the main wind direction position or the leeward direction of the tower of the wind turbine.
  • processor arranged inside the tower, connected with the audio sensor, used to receive the blade rotation audio collected by the audio sensor, preprocess the blade rotation audio based on the wind noise filtering algorithm, and obtain the blade rotation after filtering the wind noise Audio
  • the system further includes: a server, connected to the processor, for obtaining a result of the processor diagnosing whether the blade is faulty, and prompting when the blade is faulty.
  • an embodiment of the present application provides a storage medium, and when the computer program instructions are executed by a processor, the blade fault diagnosis method according to the embodiment of the present application is implemented.
  • the blade fault diagnosis method, device, system, and storage medium can process the blade rotation audio with a wind noise filtering algorithm to obtain the blade rotation audio after filtering the wind noise, thereby removing the interference of wind noise in the blade rotation audio , obtain the audio with more accurate and stronger blade rotation sound; further, divide the blade rotation audio after filtering out wind noise to obtain the audio clip corresponding to each blade, and diagnose each audio clip according to each audio clip. Whether the blades corresponding to the audio clips are faulty can be diagnosed according to the audio clips of different blades respectively, which improves the accuracy of the diagnosis results.
  • FIG. 1 is a schematic diagram of a blade fault diagnosis system provided by an embodiment of the present application
  • FIG. 2 is a schematic diagram of a blade fault diagnosis system provided by another embodiment of the present application.
  • FIG. 3 is a schematic flowchart of a blade fault diagnosis method provided by an embodiment of the present application.
  • FIG. 4 is a schematic flowchart of a training model of a blade fault diagnosis method provided by an embodiment of the present application
  • FIG. 5 is a schematic flowchart of identifying wind noise through a wind noise identification model in a blade fault diagnosis method provided by an embodiment of the present application
  • FIG. 6 is a schematic flowchart of obtaining audio clips through a blade identification model in a blade fault diagnosis method provided by an embodiment of the present application
  • FIG. 7 is a schematic flowchart of diagnosing a blade fault by using a blade fault diagnosis model in a blade fault diagnosis method provided by an embodiment of the present application;
  • FIG. 8 is a schematic flowchart of a blade fault diagnosis method provided by another embodiment of the present application.
  • FIG. 9 is a schematic diagram of a blade fault diagnosis system provided by another embodiment of the present application.
  • FIG. 10 is a schematic diagram of a blade fault diagnosis system provided by another embodiment of the present application.
  • FIG. 11 is a schematic diagram of a blade fault diagnosis apparatus provided by an embodiment of the present application.
  • embodiments of the present application provide a blade fault diagnosis method, device, system, and storage medium.
  • FIG. 1 a schematic structural diagram of an example of a blade fault diagnosis system provided in an embodiment of the application, the system includes: an audio collection device 110 and a processor 120 , wherein the audio collection device 110 includes at least one audio sensor 111 .
  • At least one audio sensor 111 is disposed at the main wind direction position or the leeward direction position of the tower of the wind turbine.
  • the audio sensor 111 is used to collect the audio signal of the blade rotation of the wind turbine.
  • the blades of the wind turbine will emit sound when rotating, and when the blades fail, the sound of the blades rotating will change. Therefore, the audio signal collected by the audio sensor 111 can be used to diagnose the blade failure.
  • the audio sensor 111 may be a pickup.
  • the audio collection device 110 may include a signal processing module, wherein the signal processing module of the audio collection device 110 may be used to convert the blade rotation audio signal collected by the audio sensor 111 into blade rotation audio.
  • the preprocessing of the blade rotation audio may also be further performed.
  • the audio sensor 111 of the audio collection device 110 can be arranged on the outer wall of the tower of the wind turbine, and other hardware parts of the audio collection device 110 can be arranged inside the tower of the wind turbine, and the hardware inside the tower can communicate by wire or wirelessly.
  • the communication method communicates with the audio sensor 111 .
  • the hardware of the audio collection device 110 disposed inside the tower can be connected to the audio sensor 111 through an optical fiber communication cable to receive audio signals collected by the audio sensor 111 .
  • FIG. 2 is a schematic diagram of the installation of an exemplary audio sensor.
  • the audio sensor 111 in FIG. 1 may include the audio sensor 141 and the audio sensor 142 in FIG. 2 .
  • the audio sensor 141 is arranged at the position of the main wind direction 151 of the tower 130 of the wind turbine (shown as a cross section of the wind turbine tower 130 in FIG. 2 ), and the audio sensor 142 is arranged at the tower door 131 of the wind turbine. Above, that is, the direction with an angle of 90° with the main wind direction 151 or the leeward direction 152 .
  • the audio sensor may be fixed on the outer tower wall of the bottom of the tower of the wind turbine by means of magnetic attraction or the like.
  • the audio sensor is set in the main wind direction or the leeward position, because the main wind direction and the leeward direction are relatively small wind speed and wind noise is relatively small, therefore, the wind noise interference in the collected blade rotation audio is small, and the fault diagnosis results are more accurate.
  • the wind speed is higher, and the collected audio has a greater impact on wind noise, making it difficult to record clear blade rotation audio.
  • Installed above the tower door (90° with the main wind direction), so the tower door is the area with the fastest wind speed in the airflow field of the wind, and the effect of wind noise in the collected audio is also greater.
  • the wind speed in the main wind direction or the leeward direction in the gas flow field is relatively small, and the collected audio is relatively less affected by wind noise, and relatively clear blade rotation audio can be collected.
  • the audio collection device 110 includes more than two audio sensors 111
  • at least one audio sensor 111 is arranged on the tower of the wind turbine, at the position of the main wind direction or the leeward direction, and the rest of the audio sensors 111 can be arranged at other positions
  • it can be set at a position where the angle between the tower and the main wind direction is 90°, or, two or more audio sensors 111 can be evenly distributed around the circumference of the tower of the wind turbine, and ensure that at least one of the audio sensors 111
  • An audio sensor 111 is provided at the main wind direction position or the leeward direction position.
  • more than two audio sensors 111 may be disposed on the same cross section of the tower, or may be disposed on different cross sections of the tower.
  • the processor 120 can be arranged inside the tower of the wind turbine, connected with the audio sensor 111, and used to receive the blade rotation audio collected by the audio sensor 111, and preprocess the blade rotation audio based on the wind noise filtering algorithm to obtain the filtered wind.
  • the blade rotation audio after noise is divided; the blade rotation audio after filtering out wind noise is divided to obtain the audio clip corresponding to each blade; according to each audio clip, diagnose whether the blade corresponding to each audio clip is faulty.
  • the processor 120 may execute a program to diagnose whether the blade is faulty according to the rotation audio of the blade.
  • the system may further include a server, which is connected to the processor 120 and may be used to obtain a result of the processor 120 diagnosing whether the blade is faulty, and give a prompt when the blade is faulty.
  • the server may be arranged in the central control room of the wind farm, and communicate with the processor 120 arranged in the tower of the wind turbine through a switch.
  • An embodiment of the present application also provides a method for diagnosing a blade fault, which may be executed by a processor in the blade fault diagnosis system provided by the embodiment of the present application.
  • a processor in the blade fault diagnosis system provided by the embodiment of the present application.
  • FIG. 3 shows a schematic flowchart of a blade fault diagnosis method provided by an embodiment of the present application. As shown in FIG. 3 , the method includes the following steps 201 to 204:
  • Step 201 Acquire the blade rotation audio collected by the audio collection device during the operation of the wind turbine.
  • the audio collection device can be used to collect the blade rotation audio. During the operation of the wind turbine, the blade rotation will emit sound. The sound signal collected by the audio collection device in real time is used to generate the blade rotation audio.
  • the audio collection device may communicate with the executor of the blade fault diagnosis method provided by the embodiment of the present application through wired communication such as optical fiber communication cables, or wireless communication, so as to send the blade rotation audio collected by the audio collection device to the the executor.
  • the executor of the blade fault diagnosis method may be a server provided in the central control room, or the executor may also be the processor 120 of the blade fault diagnosis system in the embodiment shown in FIG. 1 .
  • the processor 120 It may also be set inside the tower of the wind turbine. After the processor 120 executes the blade fault diagnosis method provided in the embodiment of the present application to obtain the blade fault diagnosis result, the processor 120 sends the diagnosis result to the server in the central control room.
  • the collection time period may be a preset time period, for example, it is preconfigured to collect 1 minute of blade rotation audio each time.
  • Step 202 preprocessing the blade rotation audio based on the wind noise filtering algorithm to obtain the blade rotation audio after the wind noise has been filtered out.
  • the wind noise filtering algorithm is used to filter out wind noise in the sound of blade rotation. Since the audio collection device is located in a wind field, the collected audio includes wind noise, that is, wind noise. In order to obtain clearer blade rotation audio, wind noise in the blade rotation audio can be eliminated through a wind noise filtering algorithm.
  • the wind noise filtering algorithm may be a Kalman filtering algorithm, a median filtering algorithm, an arithmetic mean filtering algorithm, a moving average filtering algorithm, etc., which is not limited in this embodiment of the present application.
  • the frequency of wind noise can be counted according to the historically collected audio records, and the frequency filter of the audio frequency can be designed to reduce the components in some frequency bands of the blade rotation audio.
  • An exemplary implementation is that, after acquiring the blade rotation audio, the blade rotation audio is converted to the frequency domain through Fourier transform, and then, the designed frequency filter (wind noise filtering algorithm) is used in the frequency domain, Therefore, the component of the frequency band where the wind noise is located is reduced, and the blade rotation audio after the wind noise is filtered out is obtained.
  • the designed frequency filter wind noise filtering algorithm
  • Step 203 Segment the blade rotation audio after filtering out the wind noise to obtain an audio segment corresponding to each blade.
  • the rotational sound intensity (unit can be decibel) of each blade is from small to large, and then to small. Therefore, this feature can be used to segment the sound of blade rotation to obtain the audio segment corresponding to each blade.
  • the sound intensity of the corresponding blade accounts for the main component of the sound intensity in the audio segment.
  • step 203 divides the blade rotation audio that has been filtered out of wind noise to obtain an audio segment corresponding to each blade, which may include performing the following steps:
  • Step 2031 through short-time Fourier transform processing, the blade rotation audio after the wind noise is filtered out, to obtain a first eigenvalue.
  • the first feature value is used to represent the frequency domain feature of the blade rotation audio after filtering out the wind noise.
  • the short-time Fourier transform uses a fixed window function, and then moves the window function to calculate the power spectrum at different times, that is, to obtain the first eigenvalue.
  • the window length of the window function used in the short-time Fourier transform may be a preset value, for example, Fourier transform is performed on the blade turning audio with a window of 10 seconds.
  • Fourier transform is performed on the blade turning audio with a window of 10 seconds.
  • Step 2032 Input the first feature value into the blade identification model to obtain the segmentation time point of the blade rotation audio after filtering out the wind noise.
  • the first eigenvalue is input to the blade identification model.
  • the above-mentioned blade identification model is a pre-trained model, which is used to identify the switching time points of the rotating sounds of different blades according to the first characteristic value.
  • the leaf identification model may be an algorithm model based on machine learning, for example, a model based on a support vector machine (Support Vector Machine, SVM for short), or a k-nearest neighbor algorithm, or a convolutional neural network (Convolutional Neural Networks, abbreviated as SVM). CNN) model, etc., which are not limited in the embodiments of the present application.
  • a support vector machine Serial Vector Machine, SVM for short
  • a k-nearest neighbor algorithm or a convolutional neural network (Convolutional Neural Networks, abbreviated as SVM).
  • CNN convolutional Neural Networks
  • Support vector machine is a kind of supervised learning method, which is a generalized linear classifier that can perform binary classification of data.
  • plane maximal-margin hyperplane.
  • Supervised learning is the process of using a set of samples of known classes to adjust the parameters of a classifier to achieve the required performance.
  • a hyperplane is a linear subspace in an n-dimensional Euclidean space with a co-dimension equal to one, that is, it must be (n-1) dimension.
  • Step 2033 Segment the blade rotation audio after the wind noise has been filtered out according to the segmentation time point, so as to obtain an audio segment corresponding to each blade.
  • the output result of the blade recognition model is the segmentation time point, and the audio segment corresponding to each blade can be cut out from the blade rotation audio according to the segmentation time point.
  • the number of clipped audio clips is the same as the number of blades of the wind turbine.
  • the obtained segmentation time points are 1 second, 2.1 seconds, 3.08 seconds, and 4.13 seconds, which are used to represent multiple audio clips obtained by segmenting the blade rotation audio, including: audio within a period of 1 to 2.1 seconds Fragment 1, Audio Fragment 2 from 2.1 to 3.08 seconds, and Audio Fragment 3 from 3.08 to 4.13 seconds.
  • each audio clip cannot be distinguished which blade corresponds to in reality, and the audio clips corresponding to different blades can be distinguished by different identifiers. For example, audio clip 1 corresponds to blade 1, and audio clip 2 corresponds to Blade 2, Audio Clip 3 corresponds to Blade 3.
  • Step 204 diagnose whether the blade corresponding to each audio segment is faulty.
  • each audio segment is the sound of rotation of the corresponding blade, according to the characteristics of the audio segment in the time domain and/or frequency domain, it is possible to diagnose whether the blade corresponding to each audio segment is faulty.
  • the blade is faulty, such as breaking, icing, etc., the sound of rotation will change, resulting in changes in sound intensity and frequency. Therefore, it is possible to identify whether the blade is faulty by analyzing the audio clips.
  • step 204 diagnoses whether the corresponding blade is faulty, which may include the following steps:
  • Step 2041 Process each audio segment through Fourier transform to obtain a second feature value of each audio segment, and the second feature value is used to characterize the frequency domain feature of each audio segment.
  • the amplitude of the audio segment at each frequency can be obtained, which can represent the characteristics of the audio segment in the frequency domain, wherein the amplitude of each frequency is Indicates the signal power of the audio segment at the corresponding frequency.
  • Step 2042 Input the second feature value of each audio segment into the blade fault diagnosis model respectively, so as to obtain the first fault diagnosis result of each blade.
  • the blade fault diagnosis model is a pre-trained model for identifying whether the corresponding blade is faulty according to the second feature value of the audio clip.
  • the blade fault diagnosis model may also be an algorithm model based on machine learning, such as a Support Vector Machine (Support Vector Machine, SVM for short) model, or a k-nearest neighbor algorithm.
  • SVM Support Vector Machine
  • the fault diagnosis can also be performed based on the correlation of the audio segment durations of multiple blades.
  • the duration of the audio clips for each blade should be similar. After there is a faulty blade, the corresponding blade rotation sound may become louder, that is, the sound intensity becomes louder, so that in the When the blade rotation audio is segmented according to the sound intensity, the duration of the corresponding blade will increase, the duration of other blades will decrease, and the difference in duration will become larger.
  • step 204 may further include the following steps:
  • Step 2043 count the duration of the audio clip corresponding to each blade
  • Step 2044 according to whether the difference between the durations of each two audio clips exceeds a preset threshold, determine whether there is a faulty blade, and obtain a second fault diagnosis result;
  • a threshold of duration can be preset. If the duration difference between any two audio clips exceeds the preset threshold, the wind turbine has a faulty blade.
  • steps 2042 and 2044 it is determined whether there is a faulty blade in combination with the first fault diagnosis result and the second fault diagnosis result of each blade.
  • the first fault diagnosis result is a result obtained by independently performing fault diagnosis through the blade fault identification model according to each audio segment.
  • a corresponding first fault diagnosis result can be obtained, for example,
  • the corresponding blades can be represented by the state values of 0 and 1, respectively, that there is no fault and that there is a fault.
  • the second fault diagnosis result is a result of judging whether there is a faulty blade according to whether the duration of the audio clips is consistent.
  • the blades may fail due to environmental reasons. For example, lightning strikes are prone to occur during the lightning season in summer, and the wind speed is too high in the strong wind season in autumn, which can easily lead to failures such as blade breakage, and ice may be covered on the blades in winter, and so on.
  • an appropriate blade fault diagnosis model can be selected according to different environments to perform fault diagnosis on the collected blade rotation audio.
  • an exemplary implementation is to obtain environmental parameters of the wind turbine before diagnosing whether the corresponding blade is faulty according to each audio clip, wherein the environmental parameters are used to represent seasons and/or weather, and then , among multiple candidate fault diagnosis models for identifying different fault types respectively, select a model for identifying fault types corresponding to environmental parameters, so as to determine the fault diagnosis model to be used. That is, each environmental parameter corresponds to a fault type, so a blade fault diagnosis model corresponding to the fault type is selected for fault diagnosis.
  • the blade rotation audio frequency can be processed by the wind noise filtering algorithm to obtain the blade rotation audio frequency after the wind noise has been filtered out, so as to remove the interference of the wind noise in the blade rotation audio frequency, and obtain more accurate, Stronger audio frequency of blade rotation sound; further, the blade rotation audio after filtering out wind noise is divided to obtain the audio clip corresponding to each blade, and according to each audio clip, diagnose whether the blade corresponding to each audio clip is If there is a fault, whether the corresponding blade is faulty can be diagnosed according to the audio clips of different blades, which improves the accuracy of the diagnosis result.
  • the audio collection device may acquire the blade rotation audio collected at multiple locations, and further, may select a blade with the smallest wind noise and the best audio quality from the blade rotation audio at multiple locations Rotate the audio and perform the next steps 203 and 204.
  • acquiring the blade rotation audio collected by the audio collection device during the operation of the wind turbine may be acquiring blade rotation audio collected at multiple locations;
  • step 202 the blade rotation audio is preprocessed based on the wind noise filtering algorithm to obtain the blade rotation audio after filtering the wind noise.
  • the wind noise parameters of each blade rotation audio after filtering out wind noise can be calculated separately through the wind noise identification model; the wind noise identification model is pre-trained for evaluation
  • the audio wind noise parameter model for example, the blade fault diagnosis model may also be an algorithm model based on machine learning, for example, an SVM model, a k-nearest neighbor algorithm, a CNN algorithm, etc., which is not limited in this embodiment of the present application.
  • the blade rotation audio after filtering out wind noise with the smallest wind noise is selected for segmentation, and the audio segment corresponding to each blade is obtained.
  • the blade rotation audio with the smallest wind noise can be optimally selected, which can increase the probability that the blade rotation audio can be used, and improve the real-time performance of the blade fault diagnosis method.
  • an exemplary implementation is as follows: The blade rotation audio after wind noise is obtained, and the third eigenvalue of each blade rotation audio after filtering wind noise is obtained, wherein the third eigenvalue is used to represent the corresponding frequency domain feature of the blade rotation audio after wind noise filtering .
  • the third eigenvalue of each blade rotation audio after the wind noise has been filtered out is respectively input into the wind noise identification model to obtain the wind noise parameters of each blade rotation audio after the wind noise has been filtered out.
  • the blade fault diagnosis model, wind noise identification model, and blade identification model described in the embodiments of the present application may be trained models based on machine learning, such as SVM models.
  • An example model training method is shown in Figure 4. The following takes the training of the blade recognition model as an example to describe the flow of the training method shown in FIG. 4 . It should be noted that other models described in the embodiments of the present application may also be trained based on the training method shown in FIG. 4 .
  • each training audio file is a blade rotation audio with a preset duration (such as 10 seconds) collected.
  • a waveform diagram of the sound intensity of the blade rotation audio can be displayed. Since the blade rotates from far to near to far, correspondingly, the sound intensity of the blade rotation will also be from small to large and then small, then the blade rotation audio can be segmented according to the sound intensity, wind power generation
  • the unit usually has three blades, then, three audio segments are divided in the blade rotation audio, corresponding to one rotation period.
  • the labels can be 1 second, 2.1 seconds, 3.08 seconds, and 4.13 seconds, which are used to represent multiple audio segments obtained by segmenting the blade rotation audio 1, including: audio within a period of 1 to 2.1 seconds Fragment 1, Audio Fragment 2 from 2.1 to 3.08 seconds, and Audio Fragment 3 from 3.08 to 4.13 seconds.
  • each audio clip cannot be distinguished which blade corresponds to in reality, and the audio clips corresponding to different blades can be distinguished by different identifiers.
  • audio clip 1 corresponds to blade 1
  • audio clip 2 corresponds to Leaf 2.
  • a part of the blade rotation audio with the split time labels is used as the training set, and the other part is used as the test set.
  • the leaf recognition model is trained on the training set. Obtain a blade rotation audio in the training set each time, identify the segmentation time point through the blade recognition model, compare the results obtained by the model with the labels, modify the parameters of the model according to the preset parameter modification algorithm, and use the updated model Identify the next training audio file in the training set again.
  • the model After the model is trained using all the files in the training set, use the model to identify the test set and evaluate the accuracy of the recognition. If qualified, another multiple new blade rotation audios can be used to generate a training set and a test set, the model is further revised, and the final model is obtained after multiple revisions. If it fails, you can adjust the label on the file and continue to use the current training set to retrain the model until the accuracy rate is qualified.
  • FIG. 5 A schematic flowchart of an example of identifying wind noise in blade rotation audio through a wind noise recognition model is shown in Figure 5.
  • First read a blade rotation audio and perform data preprocessing, such as preprocessing through a wind noise filtering algorithm.
  • Fourier transform is performed on the blade rotation audio to obtain eigenvalues.
  • the wind noise parameters are calculated according to the characteristic value of the blade rotation audio through the wind noise identification model. The larger the wind noise, the larger the wind noise parameter, and the smaller the wind noise, the smaller the wind noise parameter.
  • the wind noise parameter exceeds the threshold, the wind noise is too large, and the corresponding blade rotation audio can be abandoned for fault diagnosis.
  • the blade rotation audio frequency with the smallest wind noise parameter may be selected as the audio frequency for diagnosing blade faults.
  • FIG. 6 A schematic flowchart of an example of segmenting the blade rotation audio through the blade recognition model is shown in Figure 6.
  • a blade rotation audio is read, and data preprocessing, such as denoising and other algorithms, is performed.
  • short-time Fourier transform is performed on the blade rotation audio according to the preset window length to obtain the first eigenvalue.
  • the first eigenvalue may be a sequence composed of amplitude values corresponding to multiple frequencies.
  • the first eigenvalue is input to the blade identification model, and the blade identification model outputs the time point of the blade segmentation.
  • the blade rotation audio can be segmented according to the segmentation time point, and the audio segment corresponding to each blade in one rotation period can be cut out.
  • FIG. 7 A schematic flowchart of an example of diagnosing a blade fault in combination with a blade fault diagnosis model may be shown in FIG. 7 .
  • the fault diagnosis of each audio clip can be performed separately through the blade fault model to determine the status of the blade.
  • Fourier transform may be performed on each audio segment to obtain the second eigenvalue of the corresponding audio segment, and then the second eigenvalue of each audio segment is input into the blade fault diagnosis model corresponding to the value to obtain the corresponding Troubleshooting results for audio clips.
  • 0 can be used to represent normal
  • 1 can be used to represent abnormal
  • the state of the corresponding leaf can be represented by outputting 0 or 1.
  • the blade state decider may combine the first fault diagnosis result of each blade and the consistency detection result (second fault diagnosis result) according to the audio duration to determine and output the state of the blade.
  • FIG. 8 a schematic flowchart of the blade fault diagnosis model provided in the embodiment of the present application is shown in FIG. 8 .
  • the delay waits for other equipment of the wind turbine to start.
  • the server set in the central control room of the wind farm. If the connection fails, wait for a further delay.
  • the initialization file can be loaded for initial configuration.
  • the initial configuration may include configuring parameters such as the number of audio sensors, the duration of audio data collection, and the duration of audio data storage.
  • the audio sensor can send the audio data acquired in real time to the blade fault diagnosis system.
  • the wind noise identification model is used to determine whether there is wind noise.
  • the wind noise identification model is used to identify whether there is noise. If there is still noise, one of the status values of the fault diagnosis result is returned: Blade not identified. If no noise is identified by the wind noise identification model, one blade rotation audio with the smallest wind noise can be selected from the multiple blade rotation audios according to the wind noise parameters obtained by the wind noise identification model.
  • the selected blade rotation audio is segmented by a blade recognition algorithm to obtain an audio segment corresponding to each blade. Then, whether there is a fault can be diagnosed by a fault diagnosis algorithm as shown in FIG. 7 . If there is a fault, a state value of the fault diagnosis result can be returned: there is a blade fault, and a file of the faulty blade rotation audio can be saved to save the record. If there is no fault, it can return the status value: Blade OK. Next, after a delay for a period of time, the real-time streaming data collected by the audio sensor can be obtained again, and then fault diagnosis can be performed to achieve the effect of real-time monitoring of the blades.
  • a wind farm (wind farm 1) includes a plurality of wind turbines. Each fan is provided with an audio sensor, which is used to collect the sound signal of the rotation of the fan blade and send it to the edge processor, wherein the edge processor can be arranged inside the fan tower.
  • the edge processor can process the sound signal collected by the audio sensor into audio to obtain the blade rotation audio, and obtain the diagnosis result through the blade fault diagnosis method provided in any one of the optional embodiments of FIGS. 3 to 8 .
  • the fault diagnosis result of the blade can be sent to the central control room core fiber switch in the ring network of the first area of the wind farm through the switch at the bottom of the wind turbine tower through the cable, and then to the server in the central control room.
  • the server can display the display interface of the WEB front end, display the fault diagnosis of the blade through the interface program, and issue an alarm when the blade is faulty.
  • FIG. 10 Another schematic diagram of the blade fault diagnosis system provided by the embodiment of the present application may be shown in FIG. 10 .
  • the server may be configured with a variety of candidate blade fault diagnosis models, and the corresponding blade fault is selected based on the environmental parameters selected by the user at the front end. Diagnose the model and inform the edge processor.
  • the edge processor configures the edge processor program to perform fault diagnosis according to the blade fault diagnosis model selected by the server.
  • the edge processor of each fan may include an audio processing unit, which may be used for collecting real-time audio stream data, performing fault diagnosis, saving fault data (which may include fault diagnosis results and fault audio), and server communication.
  • the data when performing fault diagnosis, the data can be cached, and if there is a fault, the fault data can be saved. After the fault diagnosis result is obtained, the data stored in the cache area can be uploaded to the upper computer of the server.
  • the edge processor can also be controlled by the server host computer, for example, perform initial configuration according to the server, configure the number of audio sensors, data storage duration, maximum data buffer space, etc., exemplarily, can also receive the server according to environmental parameters. Selection of blade fault diagnosis models.
  • the server algorithm program in the central control room can be set up with functions such as unit connection status verification, unit blade status communication, control configuration file delivery, fault data reception, etc.; Display, unit fault data playback and other functions.
  • the blade fault diagnosis method, device, and system provided by the embodiments of the present application can monitor the status of fan blades in real time by processing audio data. It can also be monitored in real time in the dark, and it is not affected by foggy weather. It can effectively monitor the blade status for a long time, so as to identify the blade abnormality caused by its leading edge corrosion, lightning strike, fracture, crack and other reasons at the first time. The noise problem, and then repaired in time, can avoid the major failure loss of the blade caused by the cumulative failure of the blade.
  • the embodiments of the present application further provide a blade fault diagnosis apparatus, which can be used to execute the blade fault diagnosis method provided by the embodiments of the present application.
  • a blade fault diagnosis apparatus which can be used to execute the blade fault diagnosis method provided by the embodiments of the present application.
  • the blade fault diagnosis apparatus 10 provided in the embodiment of the present application includes an acquisition module 11 , a preprocessing module 12 , a segmentation module 13 and a diagnosis module 14 .
  • the acquisition module 11 is used to acquire the blade rotation audio collected by the audio collection device during the operation of the wind turbine; the preprocessing module 12 is used to preprocess the blade rotation audio based on the wind noise filtering algorithm to obtain the blade after filtering the wind noise. Rotation audio; the segmentation module 13 is used to segment the blade rotation audio after filtering out the wind noise to obtain the audio segment corresponding to each blade; the diagnosis module 14 is used to diagnose whether the corresponding blade of each audio segment is based on each audio segment. There is a malfunction.
  • the segmentation module 13 is further configured to: process the blade rotation audio after filtering out the wind noise through short-time Fourier transform processing to obtain a first eigenvalue, where the first eigenvalue is used to represent the blade rotation after filtering out the wind noise.
  • Frequency domain features of audio input the first feature value into the blade identification model to obtain the segmentation time point of the blade rotation audio after filtering out wind noise, wherein the blade identification model is a pre-trained model, which is used for according to the first feature value, identify the switching time points of the rotation sounds of different blades; segment the blade rotation audio after filtering out the wind noise according to the segmentation time points, so as to obtain the audio segment corresponding to each blade.
  • the diagnosis module 14 is further configured to: process each audio segment respectively through Fourier transform to obtain a second feature value of each audio segment, where the second feature value is used to characterize the frequency domain feature of each audio segment; Input the second eigenvalue of each audio clip into the blade fault diagnosis model respectively to obtain the first fault diagnosis result of each blade.
  • the blade fault diagnosis model is pre-trained and used to identify the corresponding model of whether the blade is faulty.
  • the diagnosis module 14 is further configured to: count the duration of the audio clips corresponding to each blade; according to whether the difference in duration of each two audio clips exceeds a preset threshold, determine whether there is a faulty blade, and obtain the second fault.
  • Diagnosis result Combine the first fault diagnosis result and the second fault diagnosis result of each blade to determine whether there is a faulty blade.
  • the blade fault diagnosis device 10 further includes:
  • an obtaining unit configured to obtain environmental parameters of the wind turbine before diagnosing whether the corresponding blades are faulty according to each audio clip, wherein the environmental parameters are used to represent seasons and/or weather;
  • the determining unit is configured to determine the model for identifying the fault type corresponding to the environmental parameter among the multiple candidate fault diagnosis models for identifying different fault types, so as to determine the fault diagnosis model to be used.
  • the acquisition module 11 is further configured to acquire the blade rotation audio collected at multiple locations; the blade fault diagnosis device 10 is further configured to perform wind noise filtering algorithm processing for each blade rotation audio, to obtain a plurality of filtered wind noises.
  • the noise of the blade rotation audio; the device also includes: a computing unit, after obtaining a plurality of blade rotation audio after filtering the wind noise, through the wind noise recognition model to calculate each blade rotation audio after filtering the wind noise.
  • the wind noise parameter is used to represent the wind noise in the audio
  • the wind noise identification model is a pre-trained model for evaluating the wind noise parameter of the audio
  • the segmentation module 13 is also used for, according to the wind noise parameter, Among the plurality of blade rotation audios after filtering out wind noise, the blade rotation audio after filtering out wind noise with the smallest wind noise is selected for segmentation, and an audio segment corresponding to each blade is obtained.
  • the computing unit is further configured to: process each blade rotation audio after filtering out wind noise through Fourier transform, to obtain a third eigenvalue of each blade rotation audio after filtering out wind noise, the third characteristic The value is used to characterize the frequency domain features of the corresponding blade rotation audio after filtering out wind noise; input the third eigenvalue of each blade rotation audio after filtering wind noise into the wind noise recognition model, and obtain each filter Wind noise parameters of blade rotation audio after wind noise.
  • the blade fault diagnosis device provided by the embodiment of the present application can obtain the blade rotation audio after filtering the wind noise by performing the wind noise filtering algorithm processing on the blade rotation audio, so as to remove the interference of the wind noise in the blade rotation audio, and obtain a more accurate, Stronger audio frequency of blade rotation sound; further, the blade rotation audio after filtering out wind noise is divided to obtain the audio clip corresponding to each blade, and according to each audio clip, diagnose whether the blade corresponding to each audio clip is If there is a fault, whether the corresponding blade is faulty can be diagnosed according to the audio clips of different blades, which improves the accuracy of the diagnosis result.
  • An embodiment of the present application further provides a storage medium, and when the computer program instructions are executed by the processor, the blade fault diagnosis method as provided by an embodiment of the present application can be implemented.
  • the functional blocks shown in the above-described structural block diagrams may be implemented as hardware, software, firmware, or a combination thereof.
  • it When implemented in hardware, it may be, for example, an electronic circuit, an application specific integrated circuit (ASIC), suitable firmware, a plug-in, a function card, or the like.
  • ASIC application specific integrated circuit
  • elements of the present application are programs or code segments used to perform the required tasks.
  • the program or code segments may be stored in a machine-readable medium or transmitted over a transmission medium or communication link by a data signal carried in a carrier wave.
  • "Machine-readable medium" may include any medium that can store or transmit information.
  • machine-readable media examples include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like.
  • the code segments may be downloaded via a computer network such as the Internet, an intranet, or the like.
  • processors may be, but are not limited to, general purpose processors, special purpose processors, application specific processors, or field programmable logic circuits. It will also be understood that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can also be implemented by special purpose hardware that performs the specified functions or actions, or that special purpose hardware and/or A combination of computer instructions is implemented.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Sustainable Development (AREA)
  • Sustainable Energy (AREA)
  • Chemical & Material Sciences (AREA)
  • Combustion & Propulsion (AREA)
  • Mechanical Engineering (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Acoustics & Sound (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Excavating Of Shafts Or Tunnels (AREA)
  • Polyurethanes Or Polyureas (AREA)

Abstract

本申请公开了一种叶片故障诊断方法、装置、系统及存储介质，所述方法包括：获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频；基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频；将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。本申请能够分别根据不同叶片的音频片段诊断对应叶片是否发生故障，提高了诊断结果的准确度。

Description

叶片故障诊断方法、装置、系统及存储介质
相关申请的交叉引用
本申请要求享有于2020年12月30日提交的名称为“叶片故障诊断方法、装置、系统及存储介质”的中国专利申请202011622722.2的优先权，该申请的全部内容通过引用并入本文中。
技术领域
本申请属于风力发电技术领域，尤其涉及一种叶片故障诊断方法、装置、系统及存储介质。
背景技术
风力发电机组的叶片在转动时会发出声音，在叶片受到损伤或存在缺陷时，叶片转动的音频会发生一定的变化，因此，在对风力发电机组的叶片状态进行监控时，一种实施方式为采集叶片转动的音频，进而根据音频信号判断风力发电机组的叶片是否存在故障。
但是，发明人发现，由于叶片发生故障时音频信号的改变较小，而不同叶片的音频信号存在细微的区别，可能导致现有的通过音频诊断叶片是否发生故障的方法发生误诊断。
发明内容
本申请实施例提供一种叶片故障诊断方法、装置、系统及存储介质，能够分别根据不同叶片的音频片段诊断对应叶片是否发生故障，提高了诊断结果的准确度。
一方面，本申请实施例提供一种叶片故障诊断方法，该方法包括：获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频；基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频； 将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
示例性地，将滤除风噪的叶片转动音频进行分割，得到每支叶片对应的音频片段，包括：通过短时傅里叶变换处理滤除风噪后的叶片转动音频，得到第一特征值，第一特征值用于表征滤除风噪后的叶片转动音频的频域特征；将第一特征值输入至叶片识别模型，以得到滤除风噪后的叶片转动音频的分割时间点，其中，叶片识别模型为预先训练的模型，用于根据第一特征值，识别不同叶片的转动声音的切换时间点；根据分割时间点分割滤除风噪后的叶片转动音频，以得到每支叶片对应的音频片段。
示例性地，根据每个音频片段，诊断对应的叶片是否存在故障，包括：通过傅里叶变换分别处理每个音频片段，得到每个音频片段的第二特征值，第二特征值用于表征每个音频片段的频域特征；将每个音频片段的第二特征值分别输入至叶片故障诊断模型，以得到每支叶片的第一故障诊断结果，叶片故障诊断模型为预先训练的用于根据音频片段的第二特征值识别对应的叶片是否存在故障的模型。
示例性地，根据每个音频片段，诊断对应的叶片是否存在故障，包括：统计每支叶片对应的音频片段的时长；根据每两个音频片段的时长的差值是否超过预设阈值，判断是否存在故障的叶片，得到第二故障诊断结果；结合每支叶片的第一故障诊断结果和第二故障诊断结果，判断是否存在故障叶片。
示例性地，在根据每个音频片段，诊断对应的叶片是否存在故障之前，该方法还包括：获取风力发电机组的环境参数，其中，环境参数用于表示季节和/或天气；在分别用于识别不同故障类型的多种候选故障诊断模型中，确定用于识别与环境参数对应的故障类型的模型，以确定使用的故障诊断模型。
示例性地，获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频，包括：获取多个位置处采集到的叶片转动音频；基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频，包括：分别针对每个叶片转动音频进行风噪滤波算法处理，得到多个滤除风噪后 的叶片转动音频；在得到多个滤除风噪后的叶片转动音频之后，方法还包括：通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数，其中，风噪参数用于表示音频中的风噪大小，风噪识别模型为预先训练的用于评估音频的风噪参数的模型；将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段，包括：根据风噪参数，在多个滤除风噪后的叶片转动音频中，选择风噪最小的滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段。
示例性地，通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数，包括：通过傅里叶变换分别处理每个滤除风噪后的叶片转动音频，得到每个滤除风噪后的叶片转动音频的第三特征值，第三特征值用于表征对应的滤除风噪后的叶片转动音频的频域特征；分别将每个滤除风噪后的叶片转动音频的第三特征值输入至风噪识别模型，得到每个滤除风噪后的叶片转动音频的风噪参数。
另一方面，本申请实施例提供了一种叶片故障诊断装置，该装置包括：获取模块，用于获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频；预处理模块，用于基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频；分割模块，用于将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；诊断模块，用于根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
示例性地，将滤除风噪的叶片转动音频进行分割，得到每支叶片对应的音频片段，包括：通过短时傅里叶变换处理滤除风噪后的叶片转动音频，得到第一特征值，第一特征值用于表征滤除风噪后的叶片转动音频的频域特征；将第一特征值输入至叶片识别模型，以得到滤除风噪后的叶片转动音频的分割时间点，其中，叶片识别模型为预先训练的模型，用于根据第一特征值，识别不同叶片的转动声音的切换时间点；根据分割时间点分割滤除风噪后的叶片转动音频，以得到每支叶片对应的音频片段。
示例性地，根据每个音频片段，诊断对应的叶片是否存在故障，包括：通过傅里叶变换分别处理每个音频片段，得到每个音频片段的第二特征值，第二特征值用于表征每个音频片段的频域特征；将每个音频片段的第二特 征值分别输入至叶片故障诊断模型，以得到每支叶片的第一故障诊断结果，叶片故障诊断模型为预先训练的用于根据音频片段的第二特征值识别对应的叶片是否存在故障的模型。
示例性地，根据每个音频片段，诊断对应的叶片是否存在故障，包括：统计每支叶片对应的音频片段的时长；根据每两个音频片段的时长的差值是否超过预设阈值，判断是否存在故障的叶片，得到第二故障诊断结果；结合每支叶片的第一故障诊断结果和第二故障诊断结果，判断是否存在故障叶片。
示例性地，在根据每个音频片段，诊断对应的叶片是否存在故障之前，该方法还包括：获取风力发电机组的环境参数，其中，环境参数用于表示季节和/或天气；在分别用于识别不同故障类型的多种候选故障诊断模型中，确定用于识别与环境参数对应的故障类型的模型，以确定使用的故障诊断模型。
示例性地，获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频，包括：获取多个位置处采集到的叶片转动音频；基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频，包括：分别针对每个叶片转动音频进行风噪滤波算法处理，得到多个滤除风噪后的叶片转动音频；在得到多个滤除风噪后的叶片转动音频之后，方法还包括：通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数，其中，风噪参数用于表示音频中的风噪大小，风噪识别模型为预先训练的用于评估音频的风噪参数的模型；将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段，包括：根据风噪参数，在多个滤除风噪后的叶片转动音频中，选择风噪最小的滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段。
示例性地，通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数，包括：通过傅里叶变换分别处理每个滤除风噪后的叶片转动音频，得到每个滤除风噪后的叶片转动音频的第三特征值，第三特征值用于表征对应的滤除风噪后的叶片转动音频的频域特征；分别将每个滤除风噪后的叶片转动音频的第三特征值输入至风噪识别模型，得到每个滤除 风噪后的叶片转动音频的风噪参数。
再一方面，本申请实施例提供了一种叶片故障诊断系统，该系统包括：音频采集设备，包括音频传感器，其中，至少一个音频传感器设置于风力发电机组的塔筒的主风向位置或背风向位置；处理器，设置于塔筒的内部，与音频传感器连接，用于接收音频传感器采集到的叶片转动音频，基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频；将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
示例性地，该系统还包括：服务器，与处理器连接，用于获取处理器诊断叶片是否存在故障的结果，并在叶片存在故障的情况下进行提示。
再一方面，本申请实施例提供了一种存储介质，计算机程序指令被处理器执行时实现如本申请实施例的叶片故障诊断方法。
本申请实施例的叶片故障诊断方法、装置、系统及存储介质，可以通过对叶片转动音频进行风噪滤波算法处理，得到滤除风噪后的叶片转动音频，从而去除叶片转动音频中风噪的干扰，得到携带更准确、更强的叶片转动声音的音频；进而，再将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段，并根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障，能够分别根据不同叶片的音频片段诊断对应叶片是否发生故障，提高了诊断结果的准确度。
附图说明
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例中所需要使用的附图作简单的介绍，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一个实施例提供的叶片故障诊断系统的示意图；
图2是本申请另一个实施例提供的叶片故障诊断系统的示意图；
图3是本申请一个实施例提供的叶片故障诊断方法的流程示意图；
图4是本申请一个实施例提供的叶片故障诊断方法训练模型的流程示意图；
图5是本申请一个实施例提供的叶片故障诊断方法中通过风噪识别模型识别风噪的流程示意图；
图6是本申请一个实施例提供的叶片故障诊断方法中通过叶片识别模型得到音频片段的流程示意图；
图7是本申请一个实施例提供的叶片故障诊断方法中通过叶片故障诊断模型诊断叶片故障的流程示意图；
图8是本申请另一个实施例提供的叶片故障诊断方法的流程示意图；
图9是本申请另一个实施例提供的叶片故障诊断系统的示意图；
图10是本申请另一个实施例提供的叶片故障诊断系统的示意图；
图11是本申请一个实施例提供的叶片故障诊断装置的示意图。
具体实施方式
下面将详细描述本申请的各个方面的特征和示例性实施例，为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及具体实施例，对本申请进行进一步详细描述。应理解，此处所描述的具体实施例仅意在解释本申请，而不是限定本申请。对于本领域技术人员来说，本申请可以在不需要这些具体细节中的一些细节的情况下实施。下面对实施例的描述仅仅是为了通过示出本申请的示例来提供对本申请更好的理解。
需要说明的是，在本文中，诸如第一和第二等之类的关系术语仅仅用来将一个实体或者操作与另一个实体或操作区分开来，而不一定要求或者暗示这些实体或操作之间存在任何这种实际的关系或者顺序。而且，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。
为了解决现有技术问题，本申请实施例提供了一种叶片故障诊断方法、装置、系统及存储介质。
下面首先对本申请实施例所提供的叶片故障诊断系统进行介绍。
如图1所示，为本申请实施例提供的叶片故障诊断系统一种示例的结构示意图，该系统包括：音频采集设备110和处理器120，其中，音频采集设备110包括至少一个音频传感器111。
至少一个音频传感器111设置于风力发电机组的塔筒的主风向位置或背风向位置。音频传感器111用于采集风力发电机组的叶片转动音频信号。风力发电机组的叶片在转动时会发出声音，而在叶片发生故障时，叶片转动的声音会发生改变，因此，音频传感器111采集到的音频信号，可以用于诊断叶片故障。示例性地，音频传感器111可以是拾音器。示例性地，除音频传感器111之外，音频采集设备110可以包括信号处理模块，其中，音频采集设备110的信号处理模块可以用于将音频传感器111采集到的叶片转动音频信号转换为叶片转动音频，示例性地，也可以进一步对叶片转动音频的预处理等。音频采集设备110的音频传感器111可以设置在风力发电机组的塔筒外壁上，音频采集设备110的其它硬件部分可以设置于风力发电机组的塔筒内部，塔筒内部的硬件可以通过有线通信或无线通信方式与音频传感器111通信，例如，音频采集设备110设置在塔筒内部的硬件可以通过光纤通信线缆与音频传感器111连接，以接收音频传感器111采集到的音频信号。
如图2所示为一种示例的音频传感器的安装示意图，图1中的音频传感器111可以包括图2的音频传感器141和音频传感器142。其中，音频传感器141设置在风力发电机组的塔筒130(图2示出为风力发电机组塔筒130的横截面)的主风向151的位置，音频传感器142设置在风力发电机组的塔筒门131的上方，也即与主风向151或背风向152夹角90°的方向。示例性地，音频传感器可以是通过磁吸等固定方式，设置在风力发电机组塔筒底部的外部塔壁。
音频传感器设置在主风向位置或背风向位置上，是由于主风向和背风向上风速较小，风力噪声比较小，因此，采集到的叶片转动音频中风噪干扰较小，故障诊断结果更准确。与安装在风力发电机组的机舱上，或者安装在塔底的塔筒门上方相比较，安装在机舱附近，风速较大，采集到的音 频中风噪影响较大，难以录制出清晰的叶片转动音频；安装在塔筒门上方(与主风向夹角90°方向)，因此塔筒门上为风的气流流场中风速最快的区域，采集到的音频中风噪影响也较大。而主风向或背风向在气体流场中的风速较小，采集到的音频受风噪影响相对较少，可以采集到比较清晰的叶片转动音频。
在音频采集设备110包括两个以上的音频传感器111时，至少将一个音频传感器111设置于风力发电机组的塔筒上，主风向或背风向的位置，其余的音频传感器111，可以设置于其它位置，例如，可以设置在塔筒的与主风向夹角为90°的位置，或者，可以将两个以上的音频传感器111均匀的分布围绕设置在风力发电机组的塔筒的一周，且保证其中至少一个音频传感器111设置在主风向位置或背风向位置。示例性地，两个以上的音频传感器111可以设置在塔筒的同一横截面上，也可以设置在塔筒的不同横截面上。
处理器120可以设置于风力发电机组的塔筒的内部，与音频传感器111连接，用于接收音频传感器111采集到的叶片转动音频，基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频；将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。处理器120可以通过执行程序，以实现根据叶片转动音频对叶片是否存在故障的诊断。
示例性地，该系统还可以包括服务器，服务器与处理器120连接，可以用于获取处理器120诊断叶片是否存在故障的结果，并在叶片存在故障的情况下进行提示。服务器可以设置在风场的中控室中，通过交换机与设置在风力发电机组塔筒内的处理器120进行通信。
本申请实施例还提供的一种叶片故障诊断方法，可以由本申请实施例提供的叶片故障诊断系统中的处理器执行。对于本申请实施例提供的叶片故障诊断系统中处理器执行的步骤未详述的部分，可以参考对本申请实施例提供的叶片故障诊断方法的说明。
下面对本申请实施例所提供的叶片故障诊断方法进行介绍。
图3示出了本申请一个实施例提供的叶片故障诊断方法的流程示意图。 如图3所示，该方法包括如下步骤201～步骤204：
步骤201，获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频。
音频采集设备可以用于采集叶片转动音频，在发风力发电机组的运行过程中，叶片转动会发出声音，通过音频采集设备实时采集到的声音信号，生成叶片转动音频。
音频采集设备可以通过光纤通信线缆等有线通信方式、或无线通信方式，与本申请实施例提供的叶片故障诊断方法的执行方进行通信，以将音频采集设备采集到的叶片转动音频发送至所述执行方。例如，所述叶片故障诊断方法的执行方可以是设置于中控室的服务器，或者，所述执行方也可以是由如图1所示实施例的叶片故障诊断系统的处理器120，处理器120也可以设置于风力发电机组的塔筒内部，处理器120在执行本申请实施例提供的叶片故障诊断方法得到叶片故障诊断结果之后，将诊断结果发送至中控室的服务器。
示例性地，采集的时长可以是预设时长，例如，预先配置好每次采集1分钟的叶片转动音频。步骤202，基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频。
风噪滤波算法用于滤除叶片转动音频中的风噪。由于音频采集设备处于风场中，因此，采集到的音频中包括风噪，也就是风的噪声。为了获取更清晰的叶片转动音频，可以通过风噪滤波算法消除叶片转动音频中的风噪。
可选定，风噪滤波算法可以是卡尔曼滤波算法、中位值滤波算法、算术平均滤波算法、滑动平均滤波算法等，本申请实施例对此不作限定。
由于风噪是一种白噪声，可以根据历史采集音频的记录统计风噪的频率，设计音频的频率滤波器，减小叶片转动音频中部分频段内的分量。
一种示例的实施方式为，在获取到叶片转动音频之后，将叶片转动音频通过傅里叶变换转换到频域，进而，在频域上使用设计好的频率滤波器(风噪滤波算法)，从而减少风噪所在频段的分量，得到滤除风噪后的叶片转动音频。
步骤203，将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段。
由于音频采集设备的音频传感器是固定的，因此，在风力发电机组多支叶片的一个转动周期内的，每支叶片的转动声音强度(单位可以是分贝)是由小到大、再到小的过程，因此，可以利用这个特性对叶片转动音频进行分割，得到每支叶片对应的音频片段，在每支叶片对应的音频片段中，对应叶片的声音强度占音频片段中声音强度的主要分量。通过对叶片转动音频进行划分，可以提高接下来诊断叶片故障的准确度。
示例性地，步骤203将滤除风噪的叶片转动音频进行分割，得到每支叶片对应的音频片段，可以包括执行如下步骤：
步骤2031，通过短时傅里叶变换处理滤除风噪后的叶片转动音频，得到第一特征值。
第一特征值用于表征滤除风噪后的叶片转动音频的频域特征。
短时傅里叶变换使用一个固定的窗函数，然后移动窗函数，计算出各个不同时刻的功率谱，也即得到第一特征值。
示例性地，短时傅里叶变换使用的窗函数的窗口长度可以是预设的数值，例如，以10秒时长为窗口对叶片转动音频进行傅里叶变换。通过短时傅里叶变换可以更好的得到每个窗口内的音频频域特征，以使得后续根据短时傅里叶变换的结果识别分割时间点的结果更准确。
步骤2032，将第一特征值输入至叶片识别模型，以得到滤除风噪后的叶片转动音频的分割时间点。
在得到叶片转动音频的短时傅里叶变换结果(第一特征值)之后，将第一特征值输入至叶片识别模型。
上述叶片识别模型为预先训练的模型，用于根据第一特征值，识别不同叶片的转动声音的切换时间点。
示例性地，该叶片识别模型可以是基于机器学习的算法模型，例如，基于支持向量机(Support Vector Machine，简称SVM)的模型，或k近邻算法，或者卷积神经网络(Convolutional Neural Networks，简称CNN)模型等，本申请实施例对此不作限定。
支持向量机是一类按监督学习(supervised learning)的方法，是一种可以对数据进行二元分类的广义线性分类器(generalized linear classifier)，其决策边界是对学习样本求解的最大边距超平面(maximum-margin hyperplane)。监督学习是利用一组已知类别的样本调整分类器的参数，使其达到所要求性能的过程。超平面是n维欧氏空间中余维度等于一的线性子空间，也就是必须是(n-1)维度。
步骤2033，根据分割时间点分割滤除风噪后的叶片转动音频，以得到每支叶片对应的音频片段。
叶片识别模型的输出结果为分割时间点，能够根据分割时间点在叶片转动音频中截取出每支叶片对应的音频片段。示例性地，截取出一个转动周期内每支叶片的音频片段即可，也即，截取出的音频片段的数量与风力发电机组的叶片数量相同。
例如，对于叶片转动音频，得到分割时间点为1秒，2.1秒，3.08秒，4.13秒，用于表示对叶片转动音频进行分割得到的多个音频片段，包括：1～2.1秒时段内的音频片段1，2.1～3.08秒的音频片段2，和3.08～4.13秒的音频片段3。需要说明的是，每个音频片段无法分辨出对应于实际中的哪个叶片，可以通过不同的标识来区分不同的叶片对应的音频片段，例如，音频片段1对应于叶片1，音频片段2对应于叶片2，音频片段3对应于叶片3。
步骤204，根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
由于每个音频片段为对应叶片的转动声音，根据音频片段在时域和/或频域上的特性，可以诊断每个音频片段对应的叶片是否存在故障。由于叶片在存在故障时，例如断裂、覆冰等，转动的声音会发生变化，产生声音强度、频率的变化，因此，可以通过对音频片段的分析，识别叶片是否故障。
示例性地，步骤204根据每个音频片段，诊断对应的叶片是否存在故障，可以包括如下步骤：
步骤2041，通过傅里叶变换分别处理每个音频片段，得到每个音频片 段的第二特征值，第二特征值用于表征每个音频片段的频域特征。
在对音频片段进行傅里叶变换之后，能够得到音频片段在每个频率的幅值，也即第二特征值，能够表示出音频片段在频域上的特征，其中，每个频率的幅值表示音频片段在对应频率上的信号功率。
步骤2042，将每个音频片段的第二特征值分别输入至叶片故障诊断模型，以得到每支叶片的第一故障诊断结果。
叶片故障诊断模型为预先训练的用于根据音频片段的第二特征值识别对应的叶片是否存在故障的模型。示例性地，叶片故障诊断模型也可以是基于机器学习的算法模型，例如，支持向量机(Support Vector Machine，简称SVM)的模型，或k近邻算法等。
示例性地，除了上述根据每个音频片段的频域特征进行故障诊断，还可以根据多支叶片的音频片段时长的相关性进行故障诊断。在叶片没有故障的情况下，每个叶片的音频片段时长应该是相近的，在其中存在有故障的叶片之后，对应的叶片转动声音可能会变大，也即，声音强度变大，从而，在根据声音强度对叶片转动音频进行分割时，对应叶片的时长会增加，其它叶片的时长会减少，时长的差距变大。
基于上述的原理，步骤204还可以包括如下步骤：
步骤2043，统计每支叶片对应的音频片段的时长；
步骤2044，根据每两个音频片段的时长的差值是否超过预设阈值，判断是否存在故障的叶片，得到第二故障诊断结果；
如果某一个叶片存在故障，那么该叶片的转动声音强度会发生变化，通常是故障叶片的转动声音变大，这样会导致在截取每支叶片的音频片段时，故障叶片的音频片段时长较长，与其它正常叶片的音频片段时长存在比较明显的差异。根据上述的原理，可以预设一个时长的阈值(预设阈值)，如果存在任意两个音频片段的时长差值超过预设阈值，则风力发电机组存在有故障的叶片。
在执行完步骤2042和步骤2044之后，结合每支叶片的第一故障诊断结果和第二故障诊断结果，判断是否存在故障叶片。
第一故障诊断结果是分别根据每个音频片段，独立的通过叶片故障识 别模型进行故障诊断得到的结果，针对每支叶片对应的音频片段，都可以得到一个对应的第一故障诊断结果，例如，可以通过状态值0和1分别表示对应叶片无故障和有故障。
而第二故障诊断结果是根据音频片段的时长是否一致，判断是否存在故障叶片的结果。
结合第一故障诊断结果和第二故障诊断结果进行判断，可以更准确的判断是否存在故障叶片。
在不同的季节或天气，叶片可能由于环境的原因导致故障，例如，夏季雷电时节容易发生雷击，秋季大风季节风速过大，都容易导致叶片断裂等故障，冬季叶片上可能覆冰，等等。
因此，可以根据不同的故障类型，训练不同的叶片故障诊断模型，进而，根据不同的环境选择适用的叶片故障诊断模型，对采集到的叶片转动音频进行故障诊断。具体而言，一种示例的实施方式为，在根据每个音频片段，诊断对应的叶片是否存在故障之前，获取风力发电机组的环境参数，其中，环境参数用于表示季节和/或天气，进而，在分别用于识别不同故障类型的多种候选故障诊断模型中，选择用于识别与环境参数对应的故障类型的模型，以确定使用的故障诊断模型。也即，每种环境参数对应于一种故障类型，从而，选择对应故障类型的叶片故障诊断模型进行故障诊断。
本申请实施例提供的叶片故障诊断方法，可以通过对叶片转动音频进行风噪滤波算法处理，得到滤除风噪后的叶片转动音频，从而去除叶片转动音频中风噪的干扰，得到携带更准确、更强的叶片转动声音的音频；进而，再将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段，并根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障，能够分别根据不同叶片的音频片段诊断对应叶片是否发生故障，提高了诊断结果的准确度。
在一种示例的实施方式中，音频采集设备可以获取多个位置处采集到的叶片转动音频，进而，可以在多个位置的叶片转动音频中，选择一个风噪最小、音频质量最好的叶片转动音频，执行接下来的步骤203和步骤204。
相应的，步骤201中获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频，可以是获取多个位置处采集到的叶片转动音频；
步骤202中基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频，可以分别针对每个叶片转动音频进行风噪滤波算法处理，得到多个滤除风噪后的叶片转动音频；
在得到多个滤除风噪后的叶片转动音频之后，可以通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数；风噪识别模型为预先训练的用于评估音频的风噪参数的模型，示例性地，叶片故障诊断模型也可以是基于机器学习的算法模型，例如，SVM模型，或k近邻算法，CNN算法等，本申请实施例对此不作限制。接着，根据风噪参数，在多个滤除风噪后的叶片转动音频中，选择风噪最小的滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段。由此可以得到择优选择风噪最小的叶片转动音频，能够增加叶片转动音频可使用的概率，提高叶片故障诊断方法的实时性。
示例性地，在通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数时，一种示例的实施方式为，首先，通过傅里叶变换分别处理每个滤除风噪后的叶片转动音频，得到每个滤除风噪后的叶片转动音频的第三特征值，其中，第三特征值用于表征对应的滤除风噪后的叶片转动音频的频域特征。接着，分别将每个滤除风噪后的叶片转动音频的第三特征值输入至风噪识别模型，得到每个滤除风噪后的叶片转动音频的风噪参数。
本申请实施例中所述的叶片故障诊断模型、风噪识别模型、叶片识别模型，可以是训练好的基于机器学习的模型，例如，可以是SVM模型等。一种示例的模型训练方法如图4所示。下面以训练叶片识别模型为例，说明图4所示训练方法的流程。需要说明的是，本申请实施例中其它所述的模型也可以基于图4所示的训练方法进行训练。
首先，获取多个训练音频文件，训练音频文件是用于训练模型所使用的音频文件。接着，对每个训练音频文件打上标签。例如，为了训练叶片 识别模型，每个训练音频文件是一个采集了预设时长(如10秒钟)的叶片转动音频，在得到叶片转动音频之后，可以显示出叶片转动音频的声音强度的波形示意图，由于叶片转动时会由远到近再到远，相应的，叶片转动的声音强度也会由小到大再到小，那么可以根据声音强度的大小，对叶片转动音频进行分段，风力发电机组通常具有三支叶片，那么，在叶片转动音频中分割出三个音频片段，对应于一个转动周期。
例如，对于叶片转动音频1，标签可以是1秒，2.1秒，3.08秒，4.13秒，用于表示对叶片转动音频1进行分割得到的多个音频片段，包括：1～2.1秒时段内的音频片段1，2.1～3.08秒的音频片段2，和3.08～4.13秒的音频片段3。需要说明的是，每个音频片段无法分辨出对应于实际中的哪个叶片，可以通过不同的标识来区分不同的叶片对应的音频片段，例如，音频片段1对应于叶片1，音频片段2对应于叶片2。
在打上标签之后，将一部分打好分割时间标签的叶片转动音频作为训练集，另一部分作为测试集。通过训练集对叶片识别模型进行训练。每次获取训练集中的一个叶片转动音频，通过叶片识别模型识别分割时间点，并将模型得到的结果与标签进行对比，根据预设的参数修改算法对模型的参数进行修改，并使用更新的模型再次对训练集中的下一个训练音频文件进行识别。
当使用训练集中的所有文件对模型训练完毕之后，使用模型对测试集进行识别，并评估识别的准确率。如果合格，则可以使用另外的多个新的叶片转动音频生成训练集和测试集，对模型进行进一步的修正，多次修正之后得到最终的模型。如果不合格，则可以调整对文件打上的标签，继续使用当前的训练集对模型重新进行训练，直至准确率合格。
通过风噪识别模型识别叶片转动音频中风噪的一种示例的流程示意图如图5所示，首先，读取一个叶片转动音频，进行数据预处理，如通过风噪滤波算法进行预处理。接着，对叶片转动音频进行傅里叶变换，得到特征值。接着，通过风噪识别模型根据叶片转动音频的特征值计算风噪参数。风噪越大，风噪参数越大，风噪越小，风噪参数越小。在风噪参数超过阈值的情况下，风噪过大，可以放弃使用对应的叶片转动音频进行故障诊断。 在风噪参数未超过阈值的情况下，可以在未超过阈值的多个叶片转动音频中，选择风噪参数最小的一个叶片转动音频，作为诊断叶片故障的音频。
通过叶片识别模型对叶片转动音频进行分割的一种示例的流程示意图可以如图6所示，首先，读取一个叶片转动音频，进行数据预处理，如去噪等算法。接着，对叶片转动音频按照预设窗口长度进行短时傅里叶变换，得到第一特征值。第一特征值可以是多个频率对应的幅值构成的序列。在得到第一特征值之后，将第一特征值输入至叶片识别模型，叶片识别模型会输出叶片分割的时间点。然后，可以根据分割时间点对叶片转动音频进行分割，截取出在一个转动周期内每支叶片对应的音频片段。
结合叶片故障诊断模型诊断叶片故障的一种示例的流程示意图可以如图7所示。在分割得到三支叶片的音频片段之后，可以通过叶片故障模型分别对每个音频片段进行故障诊断，确定叶片的状态。具体的，可以分别对每个音频片段进行傅里叶变换，得到对应音频片段的第二特征值，接着，分别将每个音频片段的第二特征值输入值对应的叶片故障诊断模型，得到对应音频片段的故障诊断结果。例如，可以用0表示正常，用1表示异常，针对每个音频片段，通过输出0或1来表示对应叶片的状态。此外，在得到分割的每支叶片的音频片段之后，可以分别根据音频片段的时长，判断是否存在异常叶片，得到第二故障诊断结果。例如，如果音频片段1和音频片段2的时长相差为0.1秒，小于阈值0.2秒，而音频片段1和音频片段3的时长相差为0.4秒，音频片段2和音频片段3的时长相差为0.3秒，大于阈值0.2秒，那么，确定音频片段1和2的状态是一致的，音频片段3的状态是不一致的。进而，叶片状态决策器可以结合每个叶片的第一故障诊断结果，以及根据音频时长的一致性检测结果(第二故障诊断结果)，确定叶片的状态并输出。
作为一种示例的具体实施方式，本申请实施例提供的叶片故障诊断模型的流程示意图如图8所示。在叶片故障诊断系统上电之后，延时等待风力发电机组的其它设备启动。延时预设时长之后，与风场中控室内设置的服务器进行通信，如果连接失败，进一步延时等待，如果连接成功，则可以加载初始化文件进行初始化配置。初始化的配置可以包括对音频传感器 个数、音频数据采集的时长、音频数据存储的时长等参数进行配置。接着，音频传感器可以将实时获取到的音频数据发送至叶片故障诊断系统。接着，通过风噪识别模型判断是否有风噪。如果有噪声，通过风噪滤波算法进行滤波，接着再通过风噪识别模型识别是否有噪声。如果仍然有噪声，则返回故障诊断结果的一种状态值：未识别出叶片。如果通过风噪识别模型识别出无噪声，则可以根据风噪识别模型得到的风噪参数，在多个叶片转动音频中选择一个风噪最小的叶片转动音频。
接着，将选择出的叶片转动音频通过叶片识别算法进行分割，得到每支叶片对应的音频片段。然后，可以通过如图7所示的故障诊断算法诊断是否存在故障。如果存在故障，则可以返回故障诊断结果的一种状态值：存在叶片故障，并可以保存故障的叶片转动音频的文件，进行记录的保存。如果不存在故障，则可以返回状态值：叶片正常。接下来，可以延时一段时间之后，再次获取音频传感器采集的实时流数据，接着进行故障诊断，以实现对叶片进行实时监测的效果。
本申请还提供了另一种叶片故障诊断系统的实施例，如图9所示，在一个风场(风场一)中，包括多个风机。每个风机设置有音频传感器，用于采集风机叶片转动的声音信号，并发送至边缘处理器，其中，边缘处理器可以设置于风机塔筒内部。边缘处理器可以将音频传感器采集到的声音信号处理为音频，得到叶片转动音频，并通过如图3～8任一个可选实施例提供的叶片故障诊断方法，得到诊断结果。进而，可以将叶片故障诊断结果通过风机塔底的交换机，经由线缆发送至风场一区环网的中控室核心光纤交换机，进而发送至中控室的服务器。服务器可以显示WEB前端的显示界面，通过界面程序显示叶片故障诊断情况，并在叶片存在故障时发出报警提示。本申请实施例提供的叶片故障诊断系统的另一种示意图可以如图10所示，服务器可以配置有多种候选的叶片故障诊断模型，基于用户在前端所选择的环境参数，选择对应的叶片故障诊断模型，并告知边缘处理器。边缘处理器在接收服务器选择的叶片故障诊断模型之后，配置边缘处理器程序根据服务器选择的叶片故障诊断模型进行故障诊断。
示例性地，每个风机的边缘处理器中可以包括音频处理单元，可以用于采集实时的音频流数据、进行故障诊断、保存故障数据(可以包括故障诊断结果和故障音频)、服务器通信。其中，在进行故障诊断时，可以缓存数据，如果存在故障，则保存故障数据。在得到故障诊断结果之后，可以将缓存区存储的数据上传至服务器上位机。此外，边缘处理器还可以接受服务器上位机的控制，例如，根据服务器进行初始化配置，配置音频传感器个数、数据存储时长、数据最大缓存空间等，示例性地，还可以接收服务器根据环境参数对叶片故障诊断模型的选择。
中控室的服务器算法程序可以设置有机组连接状态验证、机组叶片状态通信、控制配置文件下发、故障数据接收等功能；中控室的服务器的WEB前端界面可以设置有机组连接状态显示、机组叶片状态显示、机组故障数据播放等功能。
本申请实施例提供的叶片故障诊断方法、装置、系统，可以实时的通过对音频进行数据处理的方式来监测风机叶片的状态，相比于通过视频进行监控的方法，优点在于不受光照影响，在黑夜也可以实时监测，也不受大雾天气影响，能长期有效地监测叶片状态，从而在第一时间识别出叶片由于本身的前缘腐蚀、雷击、断裂、裂纹等原因，产生的叶片异响问题，进而及时维修，可以避免叶片累积故障造成的叶片重大故障损失。
本申请实施例还提供了一种叶片故障诊断装置，可以用于执行本申请实施例提供的叶片故障诊断方法。在本申请实施例提供的叶片故障诊断装置中未详述的部分，可以参考对本申请实施例提供的叶片故障诊断方法的说明，在此不再赘述。
如图11所示，本申请实施例提供的叶片故障诊断装置10包括获取模块11，预处理模块12，分割模块13和诊断模块14。
获取模块11用于获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频；预处理模块12用于基于风噪滤波算法对叶片转动音频预处理，得到滤除风噪后的叶片转动音频；分割模块13用于将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；诊断模块14用 于根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
示例性地，分割模块13还用于：通过短时傅里叶变换处理滤除风噪后的叶片转动音频，得到第一特征值，第一特征值用于表征滤除风噪后的叶片转动音频的频域特征；将第一特征值输入至叶片识别模型，以得到滤除风噪后的叶片转动音频的分割时间点，其中，叶片识别模型为预先训练的模型，用于根据第一特征值，识别不同叶片的转动声音的切换时间点；根据分割时间点分割滤除风噪后的叶片转动音频，以得到每支叶片对应的音频片段。
示例性地，诊断模块14还用于：通过傅里叶变换分别处理每个音频片段，得到每个音频片段的第二特征值，第二特征值用于表征每个音频片段的频域特征；将每个音频片段的第二特征值分别输入至叶片故障诊断模型，以得到每支叶片的第一故障诊断结果，叶片故障诊断模型为预先训练的用于根据音频片段的第二特征值识别对应的叶片是否存在故障的模型。
示例性地，诊断模块14还用于：统计每支叶片对应的音频片段的时长；根据每两个音频片段的时长的差值是否超过预设阈值，判断是否存在故障的叶片，得到第二故障诊断结果；结合每支叶片的第一故障诊断结果和第二故障诊断结果，判断是否存在故障叶片。
示例性地，叶片故障诊断装置10还包括：
获取单元，用于在根据每个音频片段，诊断对应的叶片是否存在故障之前，获取风力发电机组的环境参数，其中，环境参数用于表示季节和/或天气；
确定单元，用于在分别用于识别不同故障类型的多种候选故障诊断模型中，确定用于识别与环境参数对应的故障类型的模型，以确定使用的故障诊断模型。
示例性地，获取模块11还用于获取多个位置处采集到的叶片转动音频；叶片故障诊断装置10还用于分别针对每个叶片转动音频进行风噪滤波算法处理，得到多个滤除风噪后的叶片转动音频；该装置还包括：计算单元，用于在得到多个滤除风噪后的叶片转动音频之后，通过风噪识别模型分别计算每个滤除风噪后的叶片转动音频的风噪参数，其中，风噪参数用于表 示音频中的风噪大小，风噪识别模型为预先训练的用于评估音频的风噪参数的模型；分割模块13还用于根据风噪参数，在多个滤除风噪后的叶片转动音频中，选择风噪最小的滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段。
示例性地，计算单元还用于：通过傅里叶变换分别处理每个滤除风噪后的叶片转动音频，得到每个滤除风噪后的叶片转动音频的第三特征值，第三特征值用于表征对应的滤除风噪后的叶片转动音频的频域特征；分别将每个滤除风噪后的叶片转动音频的第三特征值输入至风噪识别模型，得到每个滤除风噪后的叶片转动音频的风噪参数。
本申请实施例提供的叶片故障诊断装置，可以通过对叶片转动音频进行风噪滤波算法处理，得到滤除风噪后的叶片转动音频，从而去除叶片转动音频中风噪的干扰，得到携带更准确、更强的叶片转动声音的音频；进而，再将滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段，并根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障，能够分别根据不同叶片的音频片段诊断对应叶片是否发生故障，提高了诊断结果的准确度。
本申请一个实施例还提供了一种存储介质，计算机程序指令被处理器执行时，可以实现如本申请一个实施例提供的叶片故障诊断方法。
需要明确的是，本申请并不局限于上文所描述并在图中示出的特定配置和处理。为了简明起见，这里省略了对已知方法的详细描述。在上述实施例中，描述和示出了若干具体的步骤作为示例。但是，本申请的方法过程并不限于所描述和示出的具体步骤，本领域的技术人员可以在领会本申请的精神后，作出各种改变、修改和添加，或者改变步骤之间的顺序。
以上所述的结构框图中所示的功能块可以实现为硬件、软件、固件或者它们的组合。当以硬件方式实现时，其可以例如是电子电路、专用集成电路(ASIC)、适当的固件、插件、功能卡等等。当以软件方式实现时，本申请的元素是被用于执行所需任务的程序或者代码段。程序或者代码段可以存储在机器可读介质中，或者通过载波中携带的数据信号在传输介质或者通信链路上传送。“机器可读介质”可以包括能够存储或传输信息的 任何介质。机器可读介质的例子包括电子电路、半导体存储器设备、ROM、闪存、可擦除ROM(EROM)、软盘、CD-ROM、光盘、硬盘、光纤介质、射频(RF)链路，等等。代码段可以经由诸如因特网、内联网等的计算机网络被下载。
还需要说明的是，本申请中提及的示例性实施例，基于一系列的步骤或者装置描述一些方法或系统。但是，本申请不局限于上述步骤的顺序，也就是说，可以按照实施例中提及的顺序执行步骤，也可以不同于实施例中的顺序，或者若干步骤同时执行。
上面参考根据本公开的实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本公开的各方面。应当理解，流程图和/或框图中的每个方框以及流程图和/或框图中各方框的组合可以由计算机程序指令实现。这些计算机程序指令可被提供给通用计算机、专用计算机、或其它可编程数据处理装置的处理器，以产生一种机器，使得经由计算机或其它可编程数据处理装置的处理器执行的这些指令使能对流程图和/或框图的一个或多个方框中指定的功能/动作的实现。这种处理器可以是但不限于是通用处理器、专用处理器、特殊应用处理器或者现场可编程逻辑电路。还可理解，框图和/或流程图中的每个方框以及框图和/或流程图中的方框的组合，也可以由执行指定的功能或动作的专用硬件来实现，或可由专用硬件和计算机指令的组合来实现。
以上所述，仅为本申请的具体实施方式，所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，上述描述的系统、模块和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。应理解，本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到各种等效的修改或替换，这些修改或替换都应涵盖在本申请的保护范围之内。

Claims (10)

  1. 一种叶片故障诊断方法，包括：
    获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频；
    基于风噪滤波算法对所述叶片转动音频预处理，得到滤除风噪后的叶片转动音频；
    将所述滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；
    根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
  2. 根据权利要求1所述的叶片故障诊断方法，其中，所述将滤除风噪的所述叶片转动音频进行分割，得到每支叶片对应的音频片段，包括：
    通过短时傅里叶变换处理所述滤除风噪后的叶片转动音频，得到第一特征值，所述第一特征值用于表征所述滤除风噪后的叶片转动音频的频域特征；
    将所述第一特征值输入至叶片识别模型，以得到所述滤除风噪后的叶片转动音频的分割时间点，其中，所述叶片识别模型为预先训练的模型，用于根据所述第一特征值，识别不同叶片的转动声音的切换时间点；
    根据所述分割时间点分割所述滤除风噪后的叶片转动音频，以得到每支叶片对应的音频片段。
  3. 根据权利要求1所述的叶片故障诊断方法，其中，所述根据每个音频片段，诊断对应的叶片是否存在故障，包括：
    通过傅里叶变换分别处理每个音频片段，得到每个音频片段的第二特征值，所述第二特征值用于表征每个音频片段的频域特征；
    将每个音频片段的第二特征值分别输入至叶片故障诊断模型，以得到每支叶片的第一故障诊断结果，所述叶片故障诊断模型为预先训练的用于根据所述音频片段的第二特征值识别对应的叶片是否存在故障的模型。
  4. 根据权利要求3所述的叶片故障诊断方法，其中，所述根据每个音频片段，诊断对应的叶片是否存在故障，包括：
    统计每支叶片对应的音频片段的时长；
    根据每两个音频片段的时长的差值是否超过预设阈值，判断是否存在故障的叶片，得到第二故障诊断结果；
    结合每支叶片的第一故障诊断结果和所述第二故障诊断结果，判断是否存在故障叶片。
  5. 根据权利要求3所述的叶片故障诊断方法，其中，在根据每个音频片段，诊断对应的叶片是否存在故障之前，所述方法还包括：
    获取所述风力发电机组的环境参数，其中，所述环境参数用于表示季节和/或天气；
    在分别用于识别不同故障类型的多种候选故障诊断模型中，确定用于识别与所述环境参数对应的故障类型的模型，以确定使用的所述故障诊断模型。
  6. 根据权利要求1-5任一项所述的叶片故障诊断方法，其中，所述获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频，包括：获取多个位置处采集到的叶片转动音频；
    所述基于风噪滤波算法对所述叶片转动音频预处理，得到滤除风噪后的叶片转动音频，包括：分别针对每个叶片转动音频进行风噪滤波算法处理，得到多个所述滤除风噪后的叶片转动音频；
    在得到多个所述滤除风噪后的叶片转动音频之后，所述方法还包括：通过风噪识别模型分别计算每个所述滤除风噪后的叶片转动音频的风噪参数，其中，所述风噪参数用于表示音频中的风噪大小，所述风噪识别模型为预先训练的用于评估音频的风噪参数的模型；
    所述将所述滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段，包括：根据所述风噪参数，在多个所述滤除风噪后的叶片转动音频中，选择风噪最小的所述滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段。
  7. 根据权利要求6所述的叶片故障诊断方法，其中，所述通过风噪识别模型分别计算每个所述滤除风噪后的叶片转动音频的风噪参数，包括：
    通过傅里叶变换分别处理每个所述滤除风噪后的叶片转动音频，得到 每个所述滤除风噪后的叶片转动音频的第三特征值，所述第三特征值用于表征对应的所述滤除风噪后的叶片转动音频的频域特征；
    分别将每个所述滤除风噪后的叶片转动音频的第三特征值输入至所述风噪识别模型，得到每个所述滤除风噪后的叶片转动音频的所述风噪参数。
  8. 一种叶片故障诊断装置，包括：
    获取模块，用于获取音频采集设备在风力发电机组的运行过程中采集到的叶片转动音频；
    预处理模块，用于基于风噪滤波算法对所述叶片转动音频预处理，得到滤除风噪后的叶片转动音频；
    分割模块，用于将所述滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；
    诊断模块，用于根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
  9. 一种叶片故障诊断系统，包括：
    音频采集设备，包括音频传感器，其中，至少一个所述音频传感器设置于风力发电机组的塔筒的主风向位置或背风向位置；
    处理器，设置于所述塔筒的内部，与所述音频传感器连接，用于接收所述音频传感器采集到的叶片转动音频，基于风噪滤波算法对所述叶片转动音频预处理，得到滤除风噪后的叶片转动音频；将所述滤除风噪后的叶片转动音频进行分割，得到每支叶片对应的音频片段；根据每个音频片段，诊断每个音频片段对应的叶片是否存在故障。
  10. 一种存储介质，所述存储介质上存储有计算机程序指令，所述计算机程序指令被处理器执行时实现如权利要求1-7任意一项所述的叶片故障诊断方法。
PCT/CN2021/103439 2020-12-30 2021-06-30 叶片故障诊断方法、装置、系统及存储介质 WO2022142213A1 (zh)

Priority Applications (5)

Application Number Priority Date Filing Date Title
CA3203703A CA3203703A1 (en) 2020-12-30 2021-06-30 Blade fault diagnosis method, apparatus and system, and storage medium
EP21912969.9A EP4254261A4 (en) 2020-12-30 2021-06-30 METHOD, DEVICE AND SYSTEM FOR DIAGNOSIS OF BLADE FAULTS AND STORAGE MEDIUM
KR1020237022434A KR20230113384A (ko) 2020-12-30 2021-06-30 블레이드 고장 진단 방법, 장치, 시스템 및 저장 매체
US18/260,075 US20240052810A1 (en) 2020-12-30 2021-06-30 Blade fault diagnosis method, apparatus and system, and storage medium
AU2021415086A AU2021415086B2 (en) 2020-12-30 2021-06-30 Blade fault diagnosis method, apparatus and system, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011622722.2 2020-12-30
CN202011622722.2A CN114764570A (zh) 2020-12-30 2020-12-30 叶片故障诊断方法、装置、系统及存储介质

Publications (1)

Publication Number Publication Date
WO2022142213A1 true WO2022142213A1 (zh) 2022-07-07

Family

ID=82260162

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/103439 WO2022142213A1 (zh) 2020-12-30 2021-06-30 叶片故障诊断方法、装置、系统及存储介质

Country Status (7)

Country Link
US (1) US20240052810A1 (zh)
EP (1) EP4254261A4 (zh)
KR (1) KR20230113384A (zh)
CN (1) CN114764570A (zh)
AU (1) AU2021415086B2 (zh)
CA (1) CA3203703A1 (zh)
WO (1) WO2022142213A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116398378A (zh) * 2023-04-18 2023-07-07 中国长江三峡集团有限公司 一种风电机组叶片多维状态监测装置及方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5448645A (en) * 1994-02-28 1995-09-05 Raymond Guerci International, Inc. Active fan blade noise cancellation system
CN105508149A (zh) * 2015-12-31 2016-04-20 北京金风科创风电设备有限公司 用于风力发电机组的故障检测方法及装置
CN110792563A (zh) * 2019-11-04 2020-02-14 北京天泽智云科技有限公司 基于卷积生成对抗网络的风电机组叶片故障音频监测方法
CN110838302A (zh) * 2019-11-15 2020-02-25 北京天泽智云科技有限公司 基于信号能量尖峰识别的音频分割方法
CN112067701A (zh) * 2020-09-07 2020-12-11 国电电力新疆新能源开发有限公司 基于声学诊断的风机叶片远程听诊方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104101652B (zh) * 2014-07-10 2017-02-15 南京航空航天大学 一种基于音频信号的风电叶片损伤监测方法及监测系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5448645A (en) * 1994-02-28 1995-09-05 Raymond Guerci International, Inc. Active fan blade noise cancellation system
CN105508149A (zh) * 2015-12-31 2016-04-20 北京金风科创风电设备有限公司 用于风力发电机组的故障检测方法及装置
CN110792563A (zh) * 2019-11-04 2020-02-14 北京天泽智云科技有限公司 基于卷积生成对抗网络的风电机组叶片故障音频监测方法
CN110838302A (zh) * 2019-11-15 2020-02-25 北京天泽智云科技有限公司 基于信号能量尖峰识别的音频分割方法
CN112067701A (zh) * 2020-09-07 2020-12-11 国电电力新疆新能源开发有限公司 基于声学诊断的风机叶片远程听诊方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4254261A4 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116398378A (zh) * 2023-04-18 2023-07-07 中国长江三峡集团有限公司 一种风电机组叶片多维状态监测装置及方法
CN116398378B (zh) * 2023-04-18 2024-04-19 中国长江三峡集团有限公司 一种风电机组叶片多维状态监测装置及方法

Also Published As

Publication number Publication date
EP4254261A4 (en) 2024-05-29
EP4254261A1 (en) 2023-10-04
AU2021415086B2 (en) 2024-05-09
CA3203703A1 (en) 2022-07-07
KR20230113384A (ko) 2023-07-28
US20240052810A1 (en) 2024-02-15
CN114764570A (zh) 2022-07-19
AU2021415086A1 (en) 2023-07-20

Similar Documents

Publication Publication Date Title
CN108985279B (zh) 多功能车辆总线mvb波形的故障诊断方法及装置
JP7199608B2 (ja) 風力タービン翼を検査するための方法および装置、ならびにそれらの機器および記憶媒体
CN110259648B (zh) 一种基于优化K-means聚类的风机叶片故障诊断方法
CN111370027B (zh) 一种离线嵌入式异常声音检测系统和方法
CN111161756B (zh) 风机叶片扫风声音信号中异常哨声轮廓的提取及识别方法
CN109209783A (zh) 一种基于噪声检测叶片的雷击损伤的方法及装置
CN112504673B (zh) 基于机器学习的托辊故障诊断方法、系统及存储介质
US20190032641A1 (en) Method for monitoring a wind turbine
EP2208981B1 (en) Monitoring of rotating machines
WO2022142213A1 (zh) 叶片故障诊断方法、装置、系统及存储介质
CN115640503A (zh) 风电机组叶片异常检测方法和装置
CN115467787A (zh) 一种基于音频分析的电机状态检测系统及方法
KR20210006832A (ko) 기계고장 진단 방법 및 장치
KR102306244B1 (ko) 디바이스의 결함 검출 모델 생성 방법 및 장치
CN116907029A (zh) 检测室外机内风扇异常的方法、控制装置以及空调室外机
CN113139430A (zh) 用于故障检测的声信号分析方法、故障检测的方法、装置
JP2008097361A (ja) 異常監視装置
CN117435908A (zh) 一种用于旋转机械的多种故障特征提取方法
CN109026555B (zh) 用于识别叶片的失速的方法和装置
CN112660746B (zh) 基于大数据技术的托辊故障诊断方法、系统及存储介质
CN114036973A (zh) 基于动态在线序列的极限学习机的串联电弧故障识别方法
CN110905735A (zh) 一种基于声音周期性的叶片故障诊断方法
Wißbrock et al. Discussion of Features for Acoustic Anomaly Detection under Industrial Disturbing Noise in an End-of-Line Test of Geared Motors
KR102418118B1 (ko) 주파수 합성을 이용한 딥러닝 기반 설비 진단 장치 및 방법
CN111456915A (zh) 风机机舱内部部件的故障诊断装置及方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21912969

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 3203703

Country of ref document: CA

WWE Wipo information: entry into national phase

Ref document number: 18260075

Country of ref document: US

ENP Entry into the national phase

Ref document number: 20237022434

Country of ref document: KR

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 2021912969

Country of ref document: EP

Effective date: 20230630

REG Reference to national code

Ref country code: BR

Ref legal event code: B01A

Ref document number: 112023013102

Country of ref document: BR

ENP Entry into the national phase

Ref document number: 2021415086

Country of ref document: AU

Date of ref document: 20210630

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 112023013102

Country of ref document: BR

Kind code of ref document: A2

Effective date: 20230629

NENP Non-entry into the national phase

Ref country code: DE