Fan gearbox low-frequency fault diagnosis and early warning method based on optical fiber F-P microcavity

Abstract

The invention provides a fan gearbox low-frequency fault diagnosis early warning method based on an optical fiber F-P microcavity, and relates to the technical field of wind driven generator equipment state detection.The fan gearbox low-frequency fault diagnosis early warning method comprises the steps that the vibration response of a fan gearbox is collected, and a vibration signal of the fan gearbox is obtained through a displacement optical conversion mechanism; performing time sequence analysis on the vibration signal, and extracting high-order features in the vibration signal based on an LSTM algorithm and in combination with an attention mechanism to obtain a high-order feature vector; constructing a meta learning task, calculating a fault category prototype through a prototype network in a meta learner, carrying out the classification of the fault category prototype based on probability distribution, and carrying out the diagnosis of the fault category of the fan gearbox according to a classification result. According to the method, diagnosis and early warning of five early-stage gearbox low-frequency fault types such as shaft misalignment and rotor imbalance are achieved, further deterioration of the faults is avoided, and the reliability and economic benefits of wind power generation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine equipment condition detection. Specifically, it particularly relates to a low-frequency fault diagnosis and early warning method for a wind turbine gearbox based on an optical fiber F-P microcavity. Background Technique

[0002] With the transformation of the global energy structure towards cleaner energy, wind power, as an important part of renewable energy, has seen a continuous and rapid increase in its installed capacity. As a key transmission component of a wind turbine, the gearbox is subject to complex alternating loads and extreme environmental impacts for a long time. Statistical data shows that its failure rate is as high as over 30%, and the maintenance cost can reach 15%-20% of the total machine maintenance cost. Low-frequency faults in the wind turbine gearbox (such as mechanical looseness, local gear damage, etc.) have the characteristic of progressive development. Early characteristic signals are often submerged in strong background noise, and traditional detection means are difficult to detect in time, and are extremely likely to evolve into catastrophic failures.

[0003] Currently, in most of the existing technologies, the vibration signals of the gearbox are detected by piezoelectric acceleration sensors to reflect its operating conditions. However, limited by the inherent frequency characteristics of the sensors, their sensitivity significantly decreases in the low-frequency band below 200 Hz, and their electromagnetic interference tolerance is poor, which easily leads to signal distortion and results in low detection accuracy under low-frequency faults. To overcome the above problems, an optical fiber sensing method based on Fabry-Perot (F-P) is introduced. Optical fiber F-P microcavity sensors have gradually become one of the effective ways for vibration signal detection due to their advantages such as compact structure, excellent weather resistance, and strong anti-electromagnetic interference ability. However, most studies focus on the detection of broadband vibration signals, and there are relatively few application studies on the vibration detection of low-frequency faults in wind turbine gearboxes. Their vibration signal detection sensitivity in the low-frequency region needs to be improved urgently.

[0004] Regarding the problems in the related technologies, no effective solutions have been proposed yet. Summary of the Invention

[0005] In view of this, the present invention provides a low-frequency fault diagnosis and early warning method for a wind turbine gearbox based on an optical fiber F-P microcavity to solve the above-mentioned problems.

[0006] To solve the above problems, the specific technical solutions adopted by the present invention are as follows:

[0007] A low-frequency fault diagnosis and early warning method for a wind turbine gearbox based on an optical fiber F-P microcavity, comprising:

[0008] S1. Collect the vibration response of the wind turbine gearbox through an optical fiber F-P microcavity sensor pre-installed above the planetary gear set in the wind turbine gearbox, and obtain the vibration signal of the wind turbine gearbox by using a displacement optical conversion mechanism;

[0009] S2. Conduct a time series analysis on the vibration signal, extract high-order features in the vibration signal based on the LSTM algorithm and combined with the attention mechanism to obtain a high-order feature vector;

[0010] S3. Based on the meta-learning strategy, use the high-order feature vector to construct a meta-learning task, calculate the fault category prototype through the prototype network in the meta-learner, and classify the fault category prototype based on the probability distribution, and diagnose the fault category of the wind turbine gearbox according to the classification result.

[0011] Preferably, collecting the vibration response of the wind turbine gearbox through the fiber optic F-P microcavity sensor pre-installed above the planetary gear set in the wind turbine gearbox, and obtaining the vibration signal of the wind turbine gearbox by using the displacement optical conversion mechanism includes:

[0012] S11. Transmit a laser signal source to the fiber optic coupler through a narrowband laser, use the fiber optic coupler to perform coupling processing on the laser signal source, and input the coupled laser signal source into the fiber optic F-P microcavity sensor in an equally divided manner;

[0013] S12. Use the fiber optic F-P microcavity sensor to respond to the vibration of the wind turbine gearbox, and reflect the coupled laser signal source to the photodetector through the fiber optic circulator;

[0014] S13. Use the photodiode in the photodetector to receive the reflected laser signal source, calculate the change in the reflected light intensity in combination with the displacement optical conversion mechanism, and perform signal conversion processing on the change in the reflected light intensity to obtain an electrical signal;

[0015] S14. Use the low-pass filter in the photodetector to perform noise filtering processing on the electrical signal, and use the signal amplifier in the photodetector to amplify the electrical signal after noise filtering processing to obtain the vibration signal of the wind turbine gearbox.

[0016] Preferably, the fiber optic F-P microcavity sensor includes: a circular diaphragm, a base, a single-mode optical fiber and an inertial mass block;

[0017] The circular diaphragm is used to be installed on the sensitive end of the base and serve as a vibration sensing structure, and a plurality of equal-strength beams are symmetrically arranged on the circular diaphragm;

[0018] The base is used for installing the circular diaphragm and the single-mode optical fiber;

[0019] The single-mode optical fiber is used to be installed on the fixed end of the base;

[0020] The inertial mass block is used as a vibration loading sensitive element, and will generate a radial micro-displacement when affected by the vibration of the wind turbine gearbox, so that the F-P microcavity in the inertial mass block changes.

[0021] Preferably, the calculation formula for calculating the change in the intensity of the reflected light by combining the displacement optical conversion mechanism is:

[0022]

[0023] In the formula, I R represents the change value of the reflected light intensity, R represents the reflectivity, λ represents the incident wavelength, L represents the cavity length of the F-P microcavity, and I0 represents the intensity of the incident light.

[0024] Preferably, the transfer function expression of the low-pass filter is:

[0025]

[0026] ω c = 2πf;

[0027] In the formula, H(s) represents the transfer function of the low-pass filter, ω c represents the cut-off angular frequency, f represents the cut-off frequency, and s represents the complex frequency variable.

[0028] Preferably, the vibration signal is subjected to time series analysis, based on the LSTM algorithm, and combined with the attention mechanism to extract high-order features in the vibration signal, and the high-order feature vector obtained includes:

[0029] S21. Based on the data acquisition card, use the analog-to-digital converter to convert the vibration signal of the fan gearbox into a digital signal;

[0030] S22. Divide the digital signal into several subsequences according to the preset number of sampling points, and perform normalization processing on the divided subsequences;

[0031] S23. Assign corresponding fault category labels to each normalized subsequence according to the fault type;

[0032] S24. Use the LSTM algorithm to perform time series modeling on the normalized subsequences, obtain time series dependence features, and perform weighted aggregation on the hidden states of each time step of the LSTM algorithm through the attention mechanism to generate a high-order feature vector.

[0033] Preferably, the calculation formula for weighted aggregation of the hidden states of each time step of the LSTM algorithm through the attention mechanism is:

[0034] α t = Softmax(W a h t );

[0035] Wherein, z represents the high-order feature vector, T represents the total number of time steps of the LSTM algorithm, and h t represents the hidden state of the LSTM algorithm at time step t, and α t represents the attention weight at the time step, and W a represents the attention weight matrix.

[0036] Preferably, based on the meta-learning strategy, a meta-learning task is constructed using the high-order feature vector, the fault category prototype is calculated through the prototype network in the meta-learner, and the fault category prototype is classified based on the probability distribution. The diagnosis of the fault category of the wind turbine gearbox includes:

[0037] S31. Divide the extracted high-order feature vectors into multiple meta-tasks, and simulate the fault sample learning scenario based on the fault category labels to generate a fault type set;

[0038] S32. Randomly select several samples from each type of fault in the fault type set, and construct a support set and a query set for the meta-learning task;

[0039] S33. Input the support set and the query set into the meta-learner, and calculate the fault category prototype through the prototype network in the meta-learner to obtain a prototype vector;

[0040] S34. According to the prototype vector, calculate the probability distribution of the query sample belonging to the fault category, and select the fault category corresponding to the maximum value in the probability distribution as the prediction label of the query sample to realize the diagnosis of the fault category of the wind turbine gearbox.

[0041] Preferably, the calculation formula for calculating the fault category prototype through the prototype network in the meta-learner to obtain the prototype vector is:

[0042]

[0043] Wherein, p c represents the prototype vector of the fault category c, K represents the sample size of the taken support set, z i represents the sample feature quantity in the query set, y i represents the fault category label corresponding to the sample z i , and S c represents the support set sample of the fault category c.

[0044] Preferably, the calculation formula for calculating the probability distribution of the query sample belonging to the fault category according to the prototype vector is:

[0045]

[0046] Wherein, P(y = c|z j ) represents the probability distribution of the query sample belonging to the fault category c, and zj represents the sample feature quantity in the query set, N represents the total number of fault categories, and p c represents the prototype vector of fault category c, and p c′ represents the prototype vector of fault category c'.

[0047] The beneficial effects of the present invention are as follows:

[0048] 1. By adopting the optical F-P microcavity sensing technology, the present invention realizes the high-sensitivity acquisition of vibration signals in the low-frequency region in the fan gearbox, and introduces a meta-learning strategy to learn the acquired signals, so as to realize the diagnosis and early warning of five early gearbox low-frequency fault types such as misalignment of shafts and rotor imbalance, avoid the further deterioration of faults, and improve the reliability and economic benefits of wind power generation.

[0049] 2. The present invention uses an optical fiber F-P microcavity sensor to realize the high-sensitivity detection of vibration signals in the low-frequency region. Compared with the traditional vibration detection method based on an acceleration sensor, the present invention focuses on providing high-resolution and high-sensitivity vibration signal detection in a lower frequency range, and has the advantages of being passive, easy to adjust, and strong anti-electromagnetic interference ability.

[0050] 3. The present invention also introduces a meta-learning strategy, which can complete model training with limited vibration data samples and has dynamic adaptability and continuous learning ability. Therefore, it can not only effectively detect the low-frequency vibration signals when the fan gearbox has low-frequency faults, but also perform signal training and analysis to realize the diagnosis and early warning of five types of low-frequency faults, reducing the costs brought by frequent maintenance or unplanned shutdowns. Description of the Drawings

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings. In the drawings:

[0052] Figure 1 is a flowchart of a method for diagnosing and early warning low-frequency faults in a fan gearbox based on an optical fiber F-P microcavity according to an embodiment of the present invention;

[0053] Figure 2 is a flowchart of the implementation of a method for diagnosing and early warning low-frequency faults in a fan gearbox based on an optical fiber F-P microcavity according to an embodiment of the present invention;

[0054] Figure 3 is a top view of an optical fiber F-P microcavity sensor in a method for diagnosing and early warning low-frequency faults in a fan gearbox based on an optical fiber F-P microcavity according to an embodiment of the present invention;

[0055] Figure 4 It is the front view of the fiber optic F-P microcavity sensor in a method for diagnosing and warning low-frequency faults of a fan gearbox based on a fiber optic F-P microcavity according to an embodiment of the present invention. Specific embodiments

[0056] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, rather than all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0057] According to an embodiment of the present invention, a method for diagnosing and warning low-frequency faults of a fan gearbox based on a fiber optic F-P microcavity is provided.

[0058] Now, the present invention will be further described in conjunction with the accompanying drawings and specific embodiments. As Figure 1-2 shown, the method for diagnosing and warning low-frequency faults of a fan gearbox based on a fiber optic F-P microcavity according to an embodiment of the present invention includes:

[0059] S1. Collect the vibration response of the fan gearbox through a fiber optic F-P microcavity sensor pre-installed above the planetary gear set in the fan gearbox, and use a displacement optical conversion mechanism to obtain the vibration signal of the fan gearbox;

[0060] Specifically, in order to achieve the diagnosis and warning of low-frequency faults of the fan gearbox, the following equipment is also required: a narrowband laser, an optical fiber coupler, an optical fiber circulator, a fiber optic F-P microcavity, a photodetector, a data acquisition card, and a computer upper computer.

[0061] Among them, by using 3M adhesive tape to stick two fiber optic F-P microcavity sensors tightly above the two planetary gear sets of the fan gearbox respectively to collect the micro-vibration signals of the entire gearbox. The fiber optic F-P microcavity sensor acts as a system low-frequency vibration sensor, responds to the vibration of the fan gearbox, and the vibration causes the mass block in the F-P microcavity to displace, resulting in a change in the cavity length and a change in the reflected light intensity. The reflected light enters the photodetector through the optical fiber circulator.

[0062] Specifically, fiber optic F-P microcavity sensors can also be installed directly above the outer casings of the four planetary gear sets and the three planetary gear sets of the fan gearbox. Connect the narrowband laser, optical fiber coupler, optical fiber circulator, fiber optic F-P microcavity, photodetector, and computer upper computer in sequence, and turn on the power switches of all equipment.

[0063] In addition, the narrowband laser used is the UNL-1550-1K-20-FA-M ultra-narrowband laser, with a working wavelength of 1550 nm, a working frequency of 1 KHz, and the power always maintained at 20 dBm.

[0064] In addition, an optical fiber coupler: The light emitted by the narrowband laser is coupled through a 50:50 optical fiber coupler, and the coupled laser signal source is evenly input into two fiber optic F-P microcavity sensors respectively.

[0065] As a preferred embodiment, collecting the vibration response of the fan gearbox through the fiber optic F-P microcavity sensor pre-installed above the planetary gear set in the fan gearbox, and obtaining the vibration signal of the fan gearbox by using the displacement optical conversion mechanism includes:

[0066] S11: Emitting a laser signal source from the narrowband laser to the optical fiber coupler, using the optical fiber coupler to perform coupling processing on the laser signal source, and inputting the coupled laser signal source into the fiber optic F-P microcavity sensor in an equally divided manner;

[0067] It should be noted that the power supply of the narrowband laser is turned on to provide a stable ultra-narrowband laser source for the entire detection system. By adjusting the power of the laser, it is ensured that the power of the laser signal emitted by the laser source meets the requirements of the detection system.

[0068] S12: Using the fiber optic F-P microcavity sensor to respond to the vibration of the fan gearbox, and reflecting the coupled laser signal source to the photodetector through the optical fiber circulator;

[0069] As a preferred embodiment, the fiber optic F-P microcavity sensor includes: a circular diaphragm, a base, a single-mode optical fiber, and an inertial mass block;

[0070] The circular diaphragm is used to be installed on the sensitive end of the base and serve as a vibration sensing structure, and a number of equal-strength beams are symmetrically arranged on the circular diaphragm;

[0071] The base is used for installing the circular diaphragm and the single-mode optical fiber;

[0072] The single-mode optical fiber is used to be installed on the fixed end of the base;

[0073] The inertial mass block is used as a vibration loading sensitive element, and will generate a radial micro-displacement when affected by the vibration of the fan gearbox, so that the F-P microcavity in the inertial mass block changes.

[0074] Specifically, as Figure 3-4As shown in the figure, the fiber optic F-P microcavity sensor consists of a circular diaphragm, a base, and a single-mode optical fiber. Among them, the circular diaphragm is regarded as a vibration sensing structure installed on the sensitive end of the base. Three equal-strength beams are symmetrically designed on the circular diaphragm. An inertial mass block located at the center of the diaphragm is supported by three equal-strength beams. The circular diaphragm and the single-mode optical fiber are respectively assembled on the sensitive end and the fixed end of the base. The inertial mass block serves as a vibration-loading sensitive element, and radial micro-displacements will be generated under the influence of vibration. The micro-displacement of the mass block causes a change in the cavity length, which in turn leads to a change in the intensity of the reflected light. The vibration signal is detected through intensity demodulation. Therefore, by converting the micro-displacement of the mass block into a change in the intensity of the reflected light, effective detection of micro-vibrations can be achieved.

[0075] Specifically, the fiber optic F-P microcavity sensor designed in the present invention has a frequency of 223 Hz, and the actual frequency response range is 1 - 150 Hz, which can accurately measure the low-frequency vibration of the fan gearbox.

[0076] For sensing the low-frequency vibration of the fan gearbox and generating an optical response, different fiber optic F-P microcavity structure size parameters will cause different resonance frequencies, which can be expressed as:

[0077]

[0078] In the formula, f represents the frequency at which resonance occurs, K eff represents the equivalent stiffness of the three equal-strength beams, b represents the width of the beam, h represents the thickness of the beam, E represents the elastic modulus of the material, l represents the length of the beam, and M represents the equivalent mass of the central mass block. The application frequency band range of the sensor should maintain a stable frequency-amplitude response in the low-frequency band. Since when the external excitation frequency approaches the resonance frequency, the output amplitude of the sensor will increase significantly, in order to avoid non-linear distortion caused by resonance, the frequency response range of the sensor should be lower than the resonance frequency.

[0079] When the central mass block is vibrationally excited, it vibrates radially along the interference cavity, resulting in a shortening / elongation of the cavity length, which in turn causes a change in the intensity I R of the reflected light.

[0080] S13. Use the photodiode in the photodetector to receive the reflected laser signal source, calculate the change in the intensity of the reflected light in combination with the displacement optical conversion mechanism, and perform signal conversion processing on the change in the intensity of the reflected light to obtain an electrical signal;

[0081] As a preferred embodiment, the calculation formula for calculating the change in the intensity of the reflected light in combination with the displacement optical conversion mechanism is:

[0082]

[0083] In the formula, I Rrepresents the change value of the reflected light intensity, R represents the reflectivity, λ represents the incident wavelength, L represents the cavity length of the F-P microcavity, and I0 represents the intensity of the incident light. The reflected light intensity I caused by external vibration excitation R will be converted into a change in the electrical signal when the subsequent reflected light enters the photodetector, and then the change in the vibration signal can be obtained through analysis on the upper computer.

[0084] S14. Use the low-pass filter in the photodetector to filter the noise of the electrical signal, and use the signal amplifier in the photodetector to amplify the electrical signal after noise filtering to obtain the vibration signal of the fan gearbox.

[0085] It should be noted that the photodetector includes a photodiode, a low-pass filter, and a signal amplifier. The optical signal output by the fiber F-P microcavity is detected by the photodiode and converted into an electrical signal. A low-pass filter with a cut-off frequency of 500 Hz is used to remove the high-frequency noise interference existing in the electrical signal. And the filtered weak electrical signal is amplified by the signal amplifier for subsequent acquisition and processing. Among them, the low-pass filter adopts the working principle of the Butterworth filter.

[0086] As a preferred implementation manner, the transfer function expression of the low-pass filter is:

[0087]

[0088] ω c = 2πf;

[0089] In the formula, H(s) represents the transfer function of the low-pass filter, ω c represents the cut-off angular frequency, f represents the cut-off frequency, and s represents the complex frequency variable, which is used to describe the dynamic characteristics of the system.

[0090] ω << ω c (low-frequency signal): |H(s)| ≈ 1, indicating that the signal passes through with almost no attenuation;

[0091] ω >> ω c (high-frequency signal): indicates that the signal is significantly attenuated.

[0092] S2. Perform time series analysis on the vibration signal, based on the LSTM algorithm, and combine the attention mechanism to extract the high-order features in the vibration signal to obtain the high-order feature vector;

[0093] As a preferred implementation manner, the performing time series analysis on the vibration signal, based on the LSTM algorithm, and combining the attention mechanism to extract the high-order features in the vibration signal to obtain the high-order feature vector includes:

[0094] S21. Based on the data acquisition card, use the analog-to-digital converter to convert the vibration signal of the fan gearbox into a digital signal;

[0095] Specifically, the data acquisition card performs dual-channel sampling, and uses the analog-to-digital converter to convert the amplified electrical signal into a digital signal that can be directly read by the computer upper computer.

[0096] S22. Divide the digital signal into several subsequences according to the preset number of sampling points, and perform normalization processing on the divided subsequences;

[0097] S23. Assign corresponding fault category labels to each normalized subsequence according to the fault type;

[0098] S24. Use the LSTM algorithm to perform temporal modeling on the normalized subsequences, obtain temporal dependence features, and perform weighted aggregation on the hidden states of each time step of the LSTM algorithm through the attention mechanism to generate high-order feature vectors.

[0099] It should be noted that the vibration digital signal is processed and analyzed in the computer upper computer. The meta-learning strategy is introduced. First, the original vibration signal is divided into subsequences with 512 sampling points per segment. The subsequences are normalized and assigned fault category labels, and then the LSTM algorithm is used to extract high-order features from the normalized temporal signal. Then, a meta-learning task is constructed. The extracted high-order feature vectors are divided into multiple meta-tasks to simulate the few-shot learning scenario. Randomly select 5 samples from each type of fault as the support set S and 15 samples as the query set Q, input them into the meta-learner of ProtoNets, calculate the class prototypes through the prototype network, and perform classification based on the distance to realize the diagnosis and early warning of the early faults of the fan gearbox.

[0100] Among them, the received vibration signal segment is The LSTM feature extraction process is as follows:

[0101] h t = LSTM(x t ,h t-1 ), t = 1, 2, …, T;

[0102] In the formula, represents the hidden state at time step t, and d is the dimension of the hidden layer. Further, in combination with the attention mechanism, all time steps are weighted and aggregated. x t represents the specific vibration signal value at the t-th time step in the subsequence, which is the input feature of the LSTM network.

[0103] As a preferred embodiment, the calculation formula for weighted aggregation of the hidden states of each time step of the LSTM algorithm through the attention mechanism is:

[0104] α t = Softmax(W a h t );

[0105] In the formula, z represents the high - order feature vector, also represents the time - series pattern of the vibration signal, T represents the total number of time steps of the LSTM algorithm, h t represents the hidden state of the LSTM algorithm at time step t, α t represents the time - step attention weight, which is used to quantify the contribution of each time - step feature to fault diagnosis, achieve key fault feature focusing, and W a represents the attention weight matrix, which is used to map the LSTM hidden state to the attention space and control the importance weights of different time - step features.

[0106] S3. Based on the meta - learning strategy, use the high - order feature vector to construct a meta - learning task, calculate the fault - class prototype through the prototype network in the meta - learner, and classify the fault - class prototype based on the probability distribution, and diagnose the fault class of the wind turbine gearbox according to the classification result.

[0107] As a preferred implementation manner, the method of using the high - order feature vector to construct a meta - learning task based on the meta - learning strategy, calculating the fault - class prototype through the prototype network in the meta - learner, classifying the fault - class prototype based on the probability distribution, and diagnosing the fault class of the wind turbine gearbox according to the classification result includes:

[0108] S31. Divide the extracted high - order feature vector into multiple meta - tasks, and simulate the fault sample learning scenario based on the fault - class label to generate a fault - type set;

[0109] S32. Randomly select several samples from each type of fault in the fault - type set, and construct the support set and query set of the meta - learning task;

[0110] S33. Input the support set and query set into the meta - learner, and calculate the fault - class prototype through the prototype network in the meta - learner to obtain the prototype vector;

[0111] S34. According to the prototype vector, calculate the probability distribution of the query sample belonging to the fault class, and select the fault class corresponding to the maximum value in the probability distribution as the prediction label of the query sample to achieve the diagnosis of the fault class of the wind turbine gearbox.

[0112] It should be noted that the roles in the meta - learning task include the support set S and the query set Q:

[0113]

[0114] In the formula, z iRepresents the high-order feature vector extracted by LSTM, y i Represents the sample z i The corresponding fault category label, usually a discrete integer value, z j Represents the sample feature quantity in the query set, y j Represents the sample z j The corresponding fault category label. In the present invention, 0 represents normal, 1 represents shaft misalignment, 2 represents rotor imbalance, 3 represents mechanical looseness, 4 represents local gear damage, and 5 represents outer ring loss of the bearing.

[0115] As a preferred embodiment, the calculation formula for obtaining the prototype vector by calculating the fault category prototype in the meta-learner is:

[0116]

[0117] In the formula, p c Represents the prototype vector of the fault category c, K represents the sample size of the selected support set, z i Represents the sample feature quantity in the query set, y i Represents the sample z i The corresponding fault category label, S c Represents the support set sample of the fault category c.

[0118] As a preferred embodiment, the calculation formula for calculating the probability distribution of the query sample belonging to the fault category according to the prototype vector is:

[0119]

[0120] In the formula, P(y = c∣z j ) represents the probability distribution that the query sample belongs to the fault category c, representing the normalization result of the relative distance, z j Represents the sample feature quantity in the query set, N represents the total number of fault categories. In the present invention, N takes 5, p c Represents the prototype vector of the fault category c, p c′ Represents the prototype vector of the fault category c′.

[0121] Specifically, in the above manner, it is judged whether the query sample z j is close to the prototype vector p c . The closer the distance, the more likely the fault belongs to the category c.

[0122] Specifically, if the distance between the query sample and the prototype of category A is 0.1 and the distance from category B is 1.0;

[0123] p(B)≈0.29;

[0124] At this time, the probability value of fault category A is 0.71 > the probability value of fault category B is 0.71, indicating that the query sample is closer to fault category A.

[0125] By performing time series analysis on the sampled vibration signals in the computer host, the diagnosis of five types of low-frequency faults is realized. The diagnosis results are displayed in the host, and a warning prompt is sent to the staff when a fault is detected.

[0126] The present invention can realize the diagnosis of five early low-frequency fault types of gearbox, namely misalignment of shafts, rotor imbalance, mechanical looseness, local damage of gears, and loss of the outer ring of bearings. The diagnosis results are displayed in the computer host, and a warning prompt is sent, so as to early warn of the deterioration of early faults, effectively extend the operation life of the wind turbine, and improve its operation efficiency and safety.

[0127] In summary, by means of the above technical solutions of the present invention, the present invention uses the optical F-P microcavity sensing technology to realize the high-sensitivity acquisition of vibration signals in the low-frequency region of the wind turbine gearbox, and introduces the meta-learning strategy to learn the acquired signals, realizing the diagnosis and warning of five early low-frequency fault types of gearbox such as misalignment of shafts and rotor imbalance, so as to avoid the further deterioration of faults and improve the reliability and economic benefits of wind power generation. The present invention uses an optical fiber F-P microcavity sensor to realize the high-sensitivity detection of vibration signals in the low-frequency region. Compared with the traditional vibration detection method based on an acceleration sensor, the present invention focuses on providing high-resolution and high-sensitivity vibration signal detection in a lower frequency range, and has the advantages of being passive, easy to adjust, and strong anti-electromagnetic interference ability. The present invention also introduces the meta-learning strategy, which can complete model training under limited vibration data samples, and has dynamic adaptability and continuous learning ability. Therefore, it can not only effectively detect the low-frequency vibration signals of the wind turbine gearbox when a low-frequency fault occurs, but also train and analyze the signals to realize the diagnosis and warning of five types of low-frequency faults, reducing the costs brought by frequent maintenance or unplanned shutdowns.

[0128] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program code.

[0129] The specific embodiments described above further elaborate on the object, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity, characterized in that, Including: S1. Collect the vibration response of the wind turbine gearbox through an optical fiber F-P microcavity sensor pre-installed above the planetary gear set in the wind turbine gearbox, and use the displacement optical conversion mechanism to obtain the vibration signal of the wind turbine gearbox; S2. Conduct time series analysis on the vibration signal, extract high-order features in the vibration signal based on the LSTM algorithm and combined with the attention mechanism, and obtain a high-order feature vector; S3. Based on the meta-learning strategy, use the high-order feature vector to construct a meta-learning task, calculate the fault category prototype through the prototype network in the meta-learner, classify the fault category prototype based on the probability distribution, and diagnose the fault category of the wind turbine gearbox according to the classification result.

2. The low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 1, wherein, The process of collecting the vibration response of the wind turbine gearbox through the optical fiber F-P microcavity sensor pre-installed above the planetary gear set in the wind turbine gearbox and using the displacement optical conversion mechanism to obtain the vibration signal of the wind turbine gearbox includes: S11. Transmit a laser signal source to the optical fiber coupler through a narrowband laser, use the optical fiber coupler to couple the laser signal source, and input the coupled laser signal source into the optical fiber F-P microcavity sensor in an equally divided manner; S12. Use the optical fiber F-P microcavity sensor to respond to the vibration of the wind turbine gearbox, and reflect the coupled laser signal source to the photodetector through the optical fiber circulator; S13. Use the photodiode in the photodetector to receive the reflected laser signal source, calculate the change in the reflected light intensity in combination with the displacement optical conversion mechanism, and perform signal conversion processing on the change in the reflected light intensity to obtain an electrical signal; S14. Use the low-pass filter in the photodetector to filter the noise of the electrical signal, and use the signal amplifier in the photodetector to amplify the electrical signal after noise filtering to obtain the vibration signal of the wind turbine gearbox.

3. A low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 2, wherein The optical fiber F-P microcavity sensor includes: a circular diaphragm, a base, a single-mode optical fiber, and an inertial mass block; The circular diaphragm is used to be installed on the sensitive end of the base and serve as a vibration sensing structure, and a number of equal-strength beams are symmetrically arranged on the circular diaphragm; The base is used for installing the circular diaphragm and the single-mode optical fiber; The single-mode optical fiber is used to be installed on the fixed end of the base; The inertial mass block is used as a vibration loading sensitive element, and will generate a radial micro-displacement when affected by the vibration of the wind turbine gearbox, so that the F-P microcavity in the inertial mass block changes.

4. A method for low-frequency fault diagnosis and early warning of a fan gearbox based on an optical fiber F-P microcavity according to claim 2, characterized in that, The calculation formula for calculating the change in the reflected light intensity in combination with the displacement optical conversion mechanism is: Where I R represents the change value of the reflected light intensity, R represents the reflectivity, λ represents the incident wavelength, L represents the cavity length of the F-P microcavity, and I0 represents the intensity of the incident light.

5. A low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 1, characterized in that The transfer function expression of the low-pass filter is: ω c = 2πf; where \(H(s)\) represents the transfer function of the low-pass filter, \(\omega\) c represents the cut-off angular frequency, \(f\) represents the cut-off frequency, and \(s\) represents the complex frequency variable.

6. The low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 1, wherein The process of conducting time series analysis on the vibration signal, extracting high-order features in the vibration signal based on the LSTM algorithm and combined with the attention mechanism, and obtaining a high-order feature vector includes: S21. Based on the data acquisition card, use the analog-to-digital converter to convert the vibration signal of the wind turbine gearbox into a digital signal; S22. Divide the digital signal into several subsequences according to the preset number of sampling points, and perform normalization processing on the divided subsequences; S23. Assign corresponding fault category labels to each normalized subsequence according to the fault type. S24. Use the LSTM algorithm to perform temporal modeling on the normalized subsequences, obtain temporal dependence features, and generate high-order feature vectors by weighted aggregation of the hidden states of each time step of the LSTM algorithm through an attention mechanism.

7. A low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 6, characterized in that The calculation formula for weighted aggregation of the hidden states of each time step of the LSTM algorithm through the attention mechanism is: α t = Softmax(W a h t ); where z represents the high-order feature vector, T represents the total number of time steps of the LSTM algorithm, and h t represents the hidden state of the LSTM algorithm at time step t, and α t represents the time step attention weight, and W a represents the attention weight matrix.

8. A low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 1, characterized in that, Based on the meta-learning strategy, using the high-order feature vectors to construct meta-learning tasks, calculating the fault category prototypes through the prototype network in the meta-learner, and classifying the fault category prototypes based on the probability distribution. The fault category diagnosis of the wind turbine gearbox according to the classification results includes: S31. Divide the extracted high-order feature vectors into multiple meta-tasks, and simulate the fault sample learning scenario based on the fault category labels to generate a fault type set; S32. Randomly select several samples from each type of fault in the fault type set, and construct the support set and query set of the meta-learning task; S33. Input the support set and query set into the meta-learner, and calculate the fault category prototypes through the prototype network in the meta-learner to obtain the prototype vectors; S34. According to the prototype vectors, calculate the probability distribution of the query samples belonging to the fault categories, and select the fault category corresponding to the maximum value in the probability distribution as the prediction label of the query samples to achieve the diagnosis of the fault categories of the wind turbine gearbox.

9. A low-frequency fault diagnosis and early warning method for a fan gearbox based on an optical fiber F-P microcavity according to claim 8, characterized in that The calculation formula for calculating the fault category prototypes through the prototype network in the meta-learner to obtain the prototype vectors is: where p c represents the prototype vector of the fault category c, K represents the sample size of the selected support set, z i represents the sample feature quantity in the query set, y i represents the fault category label corresponding to the sample z i , and S c represents the support set samples of the fault category c.

10. A method for low-frequency fault diagnosis and early warning of a fan gearbox based on an optical fiber F-P microcavity according to claim 8, characterized in that, The calculation formula for calculating the probability distribution of the query samples belonging to the fault categories according to the prototype vectors is: where, P(y = c|z j ) represents the probability distribution that the query sample belongs to the fault category c, z j represents the sample feature quantity in the query set, N represents the total number of fault categories, p c represents the prototype vector of the fault category c, p c′ represents the prototype vector of the fault category c'.

Images