Fault diagnosis method for wind-turbine drive chain based on time-frequency plane expectation maximization

Abstract

The present invention relates to the technical field of fault diagnosis, and provides a fault diagnosis method for a wind-turbine drive train based on time-frequency plane expectation maximization. The method comprises the following steps: configuring a cylindrical MEMS acceleration sensor on a drive chain of a wind turbine, and acquiring a vibration signal of the drive chain of the wind turbine via the cylindrical MEMS acceleration sensor; adopting a spectrogram-zero-based unsupervised classification method, acquiring a vibration signal and a rotational speed signal of the wind turbine during operation, generating a time-frequency representation of the vibration signal by using a short-time Fourier transform, and extracting spectrogram zeros and performing unsupervised classification to implement denoising processing of the vibration signal; on the basis of a multi-component signal estimation method under time-frequency plane expectation maximization, accurately estimating an instantaneous frequency and an instantaneous amplitude of a multi-component signal on a time-domain plane of the denoised signal; and performing order spectrum analysis to identify vibration signal characteristics of the drive chain of the wind turbine, so as to achieve fault diagnosis of the drive chain of the wind turbine under variable rotational speeds.

Description

A Fault Diagnosis Method for Wind Turbine Drivetrain Based on Time-Frequency Plane Expectation Maximization Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method for diagnosing wind turbine drivetrain faults based on time-frequency plane expectation maximization. Background Technology

[0002] Wind turbines are crucial equipment in modern renewable energy utilization and are widely used in wind farms worldwide. With the increasing global demand for clean energy, the number and scale of wind turbines are also growing rapidly. Wind turbines are typically installed in high-altitude, harsh environments, making the reliability and lifespan of key components in the turbine drivetrain particularly important. Under complex operating conditions such as variable speeds, the blades of wind turbines are subjected to brief but frequent impacts, subjecting the drivetrain to alternating loads and significantly affecting its service life. Therefore, effectively monitoring the operating status of the wind turbine drivetrain can improve operational reliability and prevent major safety accidents. The wind turbine drivetrain mainly consists of rotating components such as the main shaft and support bearings (direct drive type) and the speed-increasing gearbox (doubly fed type). Currently, fault diagnosis of wind turbine drivetrain faces two prominent problems: 1) Due to the continuous changes in speed, the fault characteristics have typical time-varying and non-stationary characteristics; 2) Due to the low speed of the main shaft, the fault characteristic frequency of key components in the drivetrain is in a low frequency band, and weak fault signals are easily submerged by strong noise, making it difficult to identify early weak faults and extract features of variable speed and weak faults.

[0003] Therefore, it is essential to provide a wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization, which collects vibration signals of the wind turbine drivetrain under variable speed conditions, accurately characterizes the service status of key components of the wind turbine drivetrain, accurately obtains the instantaneous amplitude and instantaneous frequency of different component signals, and realizes early detection and fault diagnosis of wind turbine drivetrain faults. Summary of the Invention

[0004] In view of this, the present invention proposes a wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization, which uses a cylindrical MEMS sensor to collect vibration signals of wind turbine drivetrain under variable speed conditions, extracts spectral zero points and performs unsupervised classification to achieve signal denoising preprocessing, and accurately obtains the instantaneous amplitude and instantaneous frequency of different component signals.

[0005] This invention provides a wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization, comprising the following steps:

[0006] S1: A cylindrical MEMS accelerometer is installed in the transmission chain of the wind turbine to collect vibration signals of the wind turbine transmission chain.

[0007] S2: An unsupervised classification method based on spectrum zero points is adopted to collect vibration and speed signals of wind turbines during operation. The vibration signals are represented by time-frequency using short-time Fourier transform, spectrum zero points are extracted and unsupervised classification is performed to achieve noise reduction of vibration signals.

[0008] S3: A multi-component signal estimation method based on expectation maximization in the time-frequency plane accurately estimates the instantaneous frequency and instantaneous amplitude of the multi-component signal in the time-domain plane of the denoised vibration signal;

[0009] S4: Perform order spectrum analysis to identify the vibration signal characteristics of the wind turbine drive train and realize fault diagnosis of the wind turbine drive train under variable speed.

[0010] Based on the above technical solution, preferably, the cylindrical MEMS accelerometer includes a cylindrical base, a mass block, and several electrode plates. A groove is formed on the surface of the base, extending along the axial direction of the base. The mass block is embedded in the groove and slidably connected to the inner surface of the groove. Anchor points are respectively provided at both ends of the mass block. One end of the anchor point is fixedly connected to the mass block, and the other end extends away from the mass block, fitting against the surface of the base and slidably connected to the base. Several comb teeth are symmetrically arranged at the non-end positions on both sides of the mass block, with the comb teeth spaced along the axial direction of the base. The fan drive chain is designed with several comb teeth, one end of which is fixedly connected to a mass block, and the other end of which extends away from the mass block. The comb teeth are slidably connected to the surface of the base. Adjacent comb teeth and between comb teeth and anchor points are spaced apart. Several electrode plates are fixedly set on the base surface between adjacent comb teeth, and the position of the electrode plates relative to the base remains unchanged. When the fan drive chain generates acceleration, the mass block drives the anchor points and several comb teeth to slide along the axial direction of the base, changing the distance between the comb teeth and adjacent electrode plates, thereby generating a current signal. The magnitude of the current signal is sampled and converted into a vibration signal of the fan drive chain under variable speed conditions.

[0011] Preferably, the noise reduction processing of the vibration signal in step S2 specifically includes:

[0012] The vibration signal of the wind turbine drive chain is converted into a time-frequency representation. The vibration signal is segmented using a Gaussian window function. The window slides across the vibration signal to obtain the spectral information corresponding to each time window. The spectral information corresponding to each window is combined to obtain the time-frequency spectrum of the entire vibration signal.

[0013] Find the zeros in the time-frequency spectrum and classify them into SS, SN and NN classes based on their properties.

[0014] After zero-point classification is completed, Delaunay triangulation is performed on SS-class and SN-class zeros in the time-frequency plane to obtain the signal domain estimate in the time-frequency plane. The vibration signal in the signal domain estimate is reconstructed using inverse short-time Fourier transform to obtain the estimated value of the noise-free vibration signal.

[0015] A further preferred approach involves identifying zeros in the time-frequency spectrum, classifying these zeros based on their properties, and setting thresholds. Any point in the time-frequency spectrum where the spectral value is below a threshold is considered a zero point. The method for determining it is as follows: ,in The mean of the spectral values. The standard deviation of the spectral values. It is an adjustment factor; based on the properties of zeros, zeros are divided into three categories: 1) SS class: signal-signal zeros; 2) SN class: signal-noise zeros; 3) NN class: noise-noise zeros; assuming N noise simulations are performed, resulting in N time-frequency spectra, the position of zero i in different noise simulations is defined as... , For each zero point i, calculate the change in the position of the zero point in different noise simulations. : ,in Together they represent the average position of zero point i in the Nth noise simulation, and are respectively taken as and Change in position The threshold, To cover the right offset of the first peak of the signal-to-signal zero-point position change, To cover the left offset of the second peak of the noise-noise zero-point position change, the zero-point classification rule is as follows: , as well as .

[0016] Further optimized adjustment factors The value is 2.5.

[0017] More preferably, the step of obtaining the signal domain estimate in the time-frequency plane involves reconstructing the vibration signal within the estimated signal domain using the inverse short-time Fourier transform to obtain an estimate of the noise-free vibration signal. Specifically, after zero-point classification, Delaunay triangulation is performed on SS-class and SN-class zeros in the time-frequency plane. The distribution of zeros and the set characteristics of the corresponding triangles are used to identify the signal domain, and the side length of the triangulation is... The calculation formula is: , The table shows two different zero points. and These are the positions of the two zeros; a weighting factor is introduced. This allows the triangulation to simultaneously consider the density and geometric distance of the zero points. ,in It is the local density of the zero point on the time-frequency plane, and the weighted edge length of each edge of the triangulation is obtained through weighting factors. , The effective length of each edge is dynamically adjusted; for each triangle formed by triangulation, if the weighted side length of any edge exceeds a preset side length threshold... If the triangle is larger than the side length threshold, then it is considered to belong to the signal domain; The triangles cluster together to obtain an estimate of the signal domain in the time-frequency plane: Based on the obtained signal domain estimate in the time-frequency plane, the signal in the signal domain is reconstructed using the inverse short-time Fourier transform, and the estimated value of the noise-free signal is calculated. .

[0018] Further preferred, step S3 includes:

[0019] S31, Expected Step: Calculate the expected value of the vibration signal, and express the estimated value of the noise-free signal with multiple components as follows: , , It is the instantaneous amplitude of the k-th component. It is the instantaneous phase. It is noise; then Monte Carlo sampling is used to estimate the posterior distribution, which is generated from the prior distribution. Sample Here, the superscript i represents different times, assuming the observed signal at each time point The above is independent for each sample. Calculate the estimated value of the noise-free signal. likelihood function value : , Then calculate the weight of each sample. Then, using a weighted average method, the expected value of the posterior distribution is estimated: ;

[0020] S32, Maximization Step: Obtaining the desired value that conforms to the characteristics of the actual vibration signal. Next, the estimated values ​​of instantaneous frequency and amplitude need to be updated. First, substitute the expected values ​​of the vibration signal characteristics obtained above, maximize the log-likelihood function of the expected value, and find the optimal instantaneous frequency and amplitude: ,in Indicates the first The parameter estimates after the next iteration are used to update the instantaneous frequency and amplitude parameters using the result of maximizing the expectation: , ,in It is the first Instantaneous phase after the next iteration Let be the instantaneous amplitude of the k components after the t-th iteration. The learning rate;

[0021] S33. To ensure the stability and accuracy of parameter estimation, the expectation step S31 and the maximization step S32 need to be repeated until the parameter estimation converges. The convergence condition for parameter estimation is: , It is the first Parameter estimation after the next iteration The convergence threshold is used; finally, the instantaneous frequency of each signal component is extracted. and instantaneous amplitude This serves as the input condition for the next step of order spectral analysis.

[0022] More preferably, step S4 involves: after extracting the instantaneous frequency and instantaneous amplitude of the vibration signal components, obtaining the instantaneous frequencies for signal feature classification and rotational speed feature classification. and Utilizing instantaneous frequency and Feature extraction is performed on the variable speed signal, and fault diagnosis of the wind turbine drivetrain is realized based on the results of the feature extraction.

[0023] A further optimized approach utilizes instantaneous frequency. and Feature extraction of variable speed signals employs the order ratio analysis method, defining the order ratio. It is a multiple of the rotational speed. Based on the magnitude of the order ratio, different fault types and fault locations in the wind turbine drive train can be characterized.

[0024] Preferably, the gap between adjacent comb teeth is greater than the sliding distance of the mass block in the groove.

[0025] The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization provided by this invention has the following advantages compared with the prior art:

[0026] (1) This invention collects vibration signals of the wind turbine drive chain under variable speed conditions to accurately characterize the service status of key components of the wind turbine drive chain, providing effective data support for subsequent time-frequency analysis. Secondly, it proposes an unsupervised classification method based on spectrum zero points. By collecting vibration and speed signals during wind turbine operation, it uses short-time Fourier transform to generate time-frequency representation, extracts spectrum zero points, and performs unsupervised classification to achieve signal denoising preprocessing. Then, it proposes a multi-component signal estimation method under the time-frequency plane expectation-maximization algorithm to accurately obtain the instantaneous amplitude and instantaneous frequency of different component signals, achieving high-resolution time-frequency representation of the time-frequency characteristics of multi-component vibration signals under non-stationary conditions. Finally, it extracts the time-frequency ridges of rotation frequency and fault feature components from the highly readable time-frequency plane obtained by the above time-frequency analysis method, and uses order spectrum analysis to achieve early detection and fault diagnosis of wind turbine drive chain faults.

[0027] (2) A cylindrical MEMS sensor structure is provided. Compared with the planar structure of existing sensors, it can restrict the degree of freedom of the mass block, so that it can only move along the cylindrical axis and restrict the movement of the mass block in other directions, thereby eliminating measurement interference. Multiple annular plates and annular comb teeth are attached to the surface of the cylinder. The comb teeth follow the mass block to move up and down on the cylindrical surface, ensuring that the capacitance changes only due to the change of the electrode spacing. At the same time, the influence of the change of the covering area on the measurement is avoided, ensuring the accuracy and stability of the measurement. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 is a flowchart of the steps of the wind turbine drive train fault diagnosis method based on time-frequency plane expectation maximization of the present invention.

[0030] Figure 2 is a perspective view of the cylindrical MEMS accelerometer sensor used in the wind turbine drive train fault diagnosis method based on time-frequency plane expectation maximization according to the present invention.

[0031] Figure 3 is a perspective view of the base of the cylindrical MEMS accelerometer of the wind turbine drive train fault diagnosis method based on time-frequency plane expectation maximization according to the present invention.

[0032] Figure 4 is a three-dimensional view of the combined state of the mass block, anchor point, and comb teeth of the cylindrical MEMS accelerometer of the wind turbine transmission chain fault diagnosis method based on time-frequency plane expectation maximization of the present invention.

[0033] Figure 5 is a perspective view of the electrode plate of the cylindrical MEMS accelerometer of the wind turbine drive train fault diagnosis method based on time-frequency plane expectation maximization according to the present invention. Embodiments of the present invention

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this disclosure. All other embodiments obtained by those skilled in the art based on the implementation methods of this invention without creative effort are within the scope of protection of this invention.

[0035] Existing methods for detecting wind turbine drivetrains have the following drawbacks: 1. Using planar MEMS accelerometers to measure vibration signals is susceptible to interference from gravity and centrifugal force when installed on rotating equipment, leading to decreased measurement accuracy. When faced with complex vibration modes, it cannot effectively distinguish signals from different vibration sources, especially when vibration signal frequencies are close or overlap, easily causing misjudgments. 2. In signal denoising, threshold-based wavelet denoising algorithms are widely used, but they have the following drawbacks: wavelet transform itself relies on pre-selected wavelet bases, which do not always perfectly match the signal characteristics, leading to the loss of signal details or the generation of artifacts. 3. Synchronous compression transform rearranges the time-frequency graph obtained by traditional continuous wavelet transform or short-time Fourier transform, concentrating signal energy at the instantaneous frequency. However, it is essentially a post-processing method for time-frequency analysis, which is sensitive to noise, especially when the frequency components of the signal are dense or overlap. Noise can significantly interfere with the rearrangement process, leading to increased estimation errors, making it unsuitable for industrial applications, particularly when processing large-scale data or real-time signals. In view of this, as shown in Figure 1, the present invention provides a wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization, which specifically includes the following steps:

[0036] S1; A cylindrical MEMS accelerometer is configured in the transmission chain of the wind turbine to collect vibration signals of the wind turbine transmission chain.

[0037] As shown in Figure 2-5, the cylindrical MEMS accelerometer includes a cylindrical base 1, a mass block 2, and several electrode plates 3. A groove is formed on the surface of the base 1, extending along the axial direction of the base 1. The mass block 2 is embedded in the groove and slidably connected to the inner surface of the groove. Anchor points 2.1 are respectively provided at both ends of the mass block 2. One end of the anchor point 2.1 is fixedly connected to the mass block 2, and the other end of the anchor point 2.1 extends away from the mass block 2, fitting against the surface of the base 1 and slidably connected to the base 1. Several comb teeth 2.2 are symmetrically arranged at the non-end positions on both sides of the mass block 2, spaced apart along the axial direction of the base 1. One end of the comb 2 is fixedly connected to the mass block 2, and the other ends of several comb teeth 2.2 extend away from the mass block 2. The comb teeth 2.2 are slidably connected to the surface of the base 1, and are spaced apart between adjacent comb teeth 2.2 and between comb teeth 2.2 and anchor points 2.1. Several electrode plates 3 are fixedly installed on the surface of the base 1 between adjacent comb teeth 2.2, and the position of the electrode plates 3 relative to the base 1 remains unchanged. When the fan drive chain generates acceleration, the mass block 2 drives the anchor point 2.1 and several comb teeth 2.2 to slide along the axial direction of the base 1, changing the distance between the comb teeth 2.2 and the adjacent electrode plates 3, thereby generating a current signal. The magnitude of the current signal is sampled and converted into a vibration signal of the fan drive chain under variable speed conditions. The main function of the electrode plates 3 is to cooperate with the comb teeth on the mass block to facilitate the sensing of the displacement of the mass block. Since the distance between the comb teeth and the adjacent electrode plates 3 can change, the capacitance between the output and the electrode plates also changes accordingly, thereby generating a detection signal corresponding to the acceleration generated by the vibration. To avoid the comb teeth 2.2 from overlapping with the electrode plate 3, the gap between adjacent comb teeth 2.2 can be set to be greater than the sliding distance of the mass block 2 in the groove.

[0038] In this embodiment of the wind turbine drivetrain, the bearings are the key components bearing load and vibration, and the gearbox is the core component transmitting power. Therefore, the sensor is installed above the outer shell of the gearbox bearing housing, ensuring that the axial direction of the sensor is aligned with the expected vibration direction to accurately measure vibration in the vertical direction. This effectively monitors the vibration state of the wind turbine drivetrain and identifies potential faults. As shown in Figures 2-5, both the comb teeth 2.2 and the electrode plate 3 are arc-shaped structures that fit the base 1. The groove has a wedge-shaped cross-section, restricting the movement of the mass block in other directions, thereby reducing measurement errors caused by lateral movement. The comb tooth structure can greatly expand the sensing area, improve measurement accuracy, and reduce signal processing difficulty.

[0039] S2: An unsupervised classification method based on spectrum zero points is adopted to collect vibration and speed signals of wind turbines during operation. The vibration signals are represented by time-frequency using short-time Fourier transform, and spectrum zero points are extracted and unsupervised classification is performed to achieve noise reduction of vibration signals.

[0040] The specific content of step S2 includes:

[0041] The vibration signal of the wind turbine drivetrain is converted into a time-frequency representation. A Gaussian window function is used to segment the vibration signal, with the window sliding across the vibration signal to obtain the spectral information corresponding to each time window. The spectral information corresponding to each window is combined to obtain the time-frequency spectrum of the entire vibration signal. This step utilizes the Short-Time Fourier Transform (STFT) to generate a cylindrical MEMS accelerometer to measure the vibration signal. Time-frequency (TF) representation, i.e. This method allows for the conversion of a one-dimensional time signal into a two-dimensional time-frequency spectrum, enabling observation of the frequency variation of the measured vibration signal over time. Next, a Gaussian window function is used to segment the signal. In the short-time Fourier transform, the window function slides across the signal, processing only a short time window at a time, and then performing a Fourier transform on these short time segments. This yields the spectral information corresponding to each time window, which, when combined, forms the entire signal's time-frequency spectrum.

[0042] Zeros are located in the time-frequency spectrum and classified into SS, SN, and NN classes based on their properties; specifically, a threshold is set. Any point in the time-frequency spectrum where the spectral value is below a threshold is considered a zero point. The method for determining it is as follows: ,in The mean of the spectral values. The standard deviation of the spectral values. It is an adjustment factor; based on the nature of the zeros, zeros are divided into three categories: 1) SS type: signal-signal zeros; generated by interference between signal components, usually the most stable. 2) SN type: signal-noise zeros; generated by interference between signal and noise, their stability is between SS and NN types. 3) NN type: noise-noise zeros; generated only by noise, the least stable.

[0043] Assuming N noise simulations are performed, resulting in N time-frequency spectra, the position of zero i in different noise simulations is defined as follows: , For each zero point i, calculate the change in the position of the zero point in different noise simulations. : ,in Together they represent the average position of zero point i in the Nth noise simulation, and are respectively taken as and Change in position The threshold, To cover most signals—the offset value to the right of the first peak of the signal zero-point position change— To cover the left offset of the second peak of most noise-noise zero-point position changes, the zero-point classification rule is as follows: , as well as .

[0044] This embodiment provides an adjustment factor through cross-validation. The value is 2.5.

[0045] After zero-point classification, Delaunay triangulation is performed on SS-class and SN-class zeros in the time-frequency plane to obtain an estimate of the signal domain in the time-frequency plane. The vibration signal within the estimated signal domain is then reconstructed using the inverse short-time Fourier transform to obtain an estimate of the noise-free vibration signal. Specifically, after zero-point classification, Delaunay triangulation is performed on SS-class and SN-class zeros in the time-frequency plane, ignoring NN-class zeros to reduce the influence of noise. The distribution of zeros and the set characteristics of the corresponding triangles are used to identify the signal domain, and the side length of the triangulation is determined. The calculation formula is: , The table shows two different zero points. and These are the positions of the two zeros; a weighting factor is introduced. This allows the triangulation to simultaneously consider the density and geometric distance of the zero points. ,in It is the local density of the zero point on the time-frequency plane, and the weighted edge length of each edge of the triangulation is obtained through weighting factors. , The effective length of each edge is dynamically adjusted; for each triangle formed by triangulation, if the weighted side length of any edge exceeds a preset side length threshold... If the triangle is larger than the side length threshold, then it is considered to belong to the signal domain; The triangles cluster together to obtain an estimate of the signal domain in the time-frequency plane: Based on the obtained signal domain estimation in the time-frequency plane, the inverse short-time Fourier transform (ISTFT) is used to estimate the signal in the signal domain. Perform reconstruction and calculate the estimated value of the noise-free signal. .

[0046] Traditional methods may be insufficient to effectively distinguish between signals and noise in certain situations, especially when signal components are densely packed or noise is strong. This invention introduces a weighting factor to more accurately reflect the actual importance and distribution characteristics of zeros, thereby improving denoising performance. Introducing the weighting factor allows triangulation to consider not only geometric distance but also zero density. High-density regions may contain more noise zeros, while low-density regions are more likely to be in the signal domain. In traditional methods, regions with high noise levels are easily misclassified as signal domain. By considering the local density of zeros and introducing the weighting factor, this misclassification can be effectively reduced, improving the overall denoising effect. This improvement makes the algorithm more robust when dealing with densely noisy regions, reducing misclassification. This method effectively utilizes the stability characteristics of zeros to accurately distinguish between signals and noise, achieving the goal of signal denoising.

[0047] S3: A multi-component signal estimation method based on expectation maximization in the time-frequency plane accurately estimates the instantaneous frequency and instantaneous amplitude of the multi-component signal in the time-domain plane of the denoised vibration signal.

[0048] After the above denoising preprocessing, the noise in the wind turbine vibration signal is effectively suppressed. However, in order to further accurately extract the instantaneous frequency and instantaneous amplitude of different signal components (such as rotational frequency and fault characteristic frequency) in the wind turbine drivetrain vibration signal under variable speed, further processing is required, namely, extracting the time-frequency ridges corresponding to the rotational frequency and fault characteristic frequency in the two-dimensional time-frequency plane. This method uses an expectation-maximization algorithm to perform deeper processing on the denoised signal to accurately estimate the instantaneous frequency and instantaneous amplitude of the multi-component signal, thus laying the foundation for the next step of order spectrum analysis under variable speed.

[0049] The specific content of step S3 is as follows:

[0050] S31, Expected Step: Calculate the expected value of the vibration signal, and express the estimated value of the noise-free signal with multiple components as follows: , , It is the instantaneous amplitude of the k-th component. It is the instantaneous phase. It is noise; then Monte Carlo sampling is used to estimate the posterior distribution, which is generated from the prior distribution. Sample Here, the superscript i represents different times, assuming the observed signal at each time point The above is independent for each sample. Calculate the estimated value of the noise-free signal. likelihood function value : , Then calculate the weight of each sample. Then, using a weighted average method, the expected value of the posterior distribution is estimated: ;

[0051] S32, Maximization Step: Obtaining the desired value that conforms to the characteristics of the actual vibration signal. Next, the estimated values ​​of instantaneous frequency and amplitude need to be updated. First, substitute the expected values ​​of the vibration signal characteristics obtained above, maximize the log-likelihood function of the expected value, and find the optimal instantaneous frequency and amplitude: ,in Indicates the first The parameter estimates after the next iteration are used to update the instantaneous frequency and amplitude parameters using the result of maximizing the expectation: , ,in It is the first Instantaneous phase after the next iteration Let be the instantaneous amplitude of the k components after the t-th iteration. The learning rate;

[0052] S33. To ensure the stability and accuracy of parameter estimation, the expectation step S31 and the maximization step S32 need to be repeated until the parameter estimation converges. The convergence condition for parameter estimation is: , It is the first Parameter estimation after the next iteration The convergence threshold is used; finally, the instantaneous frequency of each signal component is extracted. and instantaneous amplitude This serves as the input condition for the next step of order spectral analysis.

[0053] S4: Perform order spectrum analysis to identify the vibration signal characteristics of the wind turbine drive train and realize fault diagnosis of the wind turbine drive train under variable speed.

[0054] Step S4 involves: after extracting the instantaneous frequency and instantaneous amplitude of the vibration signal components, obtaining the instantaneous frequencies for signal feature classification and rotational speed feature classification. and Utilizing instantaneous frequency and Feature extraction is performed on variable speed signals, and fault diagnosis of the wind turbine drivetrain is achieved based on the results. If the equipment speed is not uniform or fluctuates significantly, various characteristic signals will not appear at equal intervals, rendering traditional Fourier transform-based spectrum analysis methods inapplicable. In such cases, the order ratio analysis method demonstrates greater advantages. Defined as a multiple of the rotational speed, i.e. Based on the magnitude of the order ratio, different fault types and fault locations in the wind turbine drivetrain are characterized. Order ratio =1 indicates that the signal frequency is consistent with the frequency conversion; order ratio =2 indicates that the signal frequency is twice the rotational frequency. It can be seen that the order ratio is also a method of frequency representation, equivalent to a multiple of the rotational frequency. For rotating machinery such as wind turbines, their vibration signals are closely related to the rotational speed. The order ratio can better represent this correspondence and can eliminate the influence of speed variations. If a specific order ratio is calculated, such as an integer multiple of 0.6×Z corresponding to an inner ring fault, or an integer multiple of 0.4×Z corresponding to an outer ring fault, where Z represents the number of rolling elements in the rolling bearing, then it can be used to characterize different fault types and fault locations, thereby enabling fault diagnosis of wind turbines under variable speed conditions.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for fault diagnosis of wind turbine drivetrain based on time-frequency plane expectation maximization, characterized in that, Includes the following steps: S1: A cylindrical MEMS accelerometer is installed in the transmission chain of the wind turbine to collect vibration signals of the wind turbine transmission chain. S2: An unsupervised classification method based on spectrum zero points is adopted to collect vibration and speed signals of wind turbines during operation. The vibration signals are represented by time-frequency using short-time Fourier transform, spectrum zero points are extracted and unsupervised classification is performed to achieve noise reduction of vibration signals. S3: A multi-component signal estimation method based on expectation maximization in the time-frequency plane accurately estimates the instantaneous frequency and instantaneous amplitude of the multi-component signal in the time-domain plane of the denoised signal. S4: Perform order spectrum analysis to identify the vibration signal characteristics of the wind turbine drive train and realize fault diagnosis of the wind turbine drive train under variable speed.

2. The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization according to claim 1, characterized in that, The cylindrical MEMS accelerometer includes a cylindrical base (1), a mass block (2), and several electrode plates (3). A groove is formed on the surface of the base (1), extending along the axial direction of the base (1). The mass block (2) is embedded in the groove and slidably connected to the inner surface of the groove. Anchor points (2.1) are provided at both ends of the mass block (2). One end of each anchor point (2.1) is fixedly connected to the mass block (2), and the other end extends away from the mass block (2). The anchor point (2.1) fits against the surface of the base (1) and slidably connects to the base (1). Several comb teeth (2.2) are symmetrically arranged at the non-end positions on both sides of the mass block (2). These comb teeth (2.2) are spaced apart along the axial direction of the base (1). One end is fixedly connected to the mass block (2), and the other end of several comb teeth (2.2) extends away from the mass block (2). The comb teeth (2.2) are slidably connected to the surface of the base (1). The comb teeth (2.2) are spaced apart from each other and from the anchor point (2.1). Several pole plates (3) are fixedly set on the surface of the base (1) between the adjacent comb teeth (2.2). The position of the pole plates (3) relative to the base (1) remains unchanged. When the fan drive chain generates acceleration, the mass block (2) drives the anchor point (2.1) and several comb teeth (2.2) to slide along the axial direction of the base (1), changing the distance between the comb teeth (2.2) and the adjacent pole plates (3), thereby generating a current signal. The magnitude of the current signal is sampled and converted into a vibration signal of the fan drive chain under variable speed conditions.

3. The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization according to claim 2, characterized in that, The noise reduction processing of the vibration signal described in step S2 specifically includes: The vibration signal of the wind turbine drive chain is converted into a time-frequency representation. The vibration signal is segmented using a Gaussian window function. The window slides across the vibration signal to obtain the spectral information corresponding to each time window. The spectral information corresponding to each window is combined to obtain the time-frequency spectrum of the entire vibration signal. Find the zeros in the time-frequency spectrum and classify them into SS, SN and NN classes based on their properties. After zero-point classification is completed, Delaunay triangulation is performed on SS-class and SN-class zeros in the time-frequency plane to obtain the signal domain estimate in the time-frequency plane. The vibration signal in the signal domain estimate is reconstructed using inverse short-time Fourier transform to obtain the estimated value of the noise-free vibration signal.

4. The wind turbine drive train fault diagnosis method based on time-frequency plane expectation maximization of claim 3, wherein, Finding zeros in the time-frequency spectrum, classifying zeros based on their properties, and setting thresholds are all part of the process. Any point in the time-frequency spectrum where the spectral value is below a threshold is considered a zero point. The method for determining it is as follows: ,in The mean of the spectral values. The standard deviation of the spectral values. It is an adjustment factor; based on the properties of zeros, zeros are divided into three categories: 1) SS class: signal-signal zeros; 2) SN class: signal-noise zeros; 3) NN class: noise-noise zeros; assuming N noise simulations are performed, resulting in N time-frequency spectra, the position of zero i in different noise simulations is defined as... , For each zero point i, calculate the change in the position of the zero point in different noise simulations. : ,in Together they represent the average position of zero point i in the Nth noise simulation, and are respectively taken as and Change in position The threshold, To cover the right offset of the first peak of the signal-to-signal zero-point position change, To cover the left offset of the second peak of the noise-noise zero-point position change, the zero-point classification rule is as follows: , as well as .

5. The wind turbine drive train fault diagnosis method based on time-frequency plane expectation maximization of claim 4, wherein, adjustment factor is 2.

5.

6. The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization according to claim 4, characterized in that, The process of obtaining an estimate of the signal domain in the time-frequency plane involves reconstructing the vibration signal within the estimated signal domain using the inverse short-time Fourier transform to obtain an estimate of the noise-free vibration signal. Specifically, after zero-point classification, Delaunay triangulation is performed on SS-class and SN-class zeros in the time-frequency plane. The distribution of zeros and the set characteristics of the corresponding triangles are used to identify the signal domain. The side length of the triangulation is determined by... The calculation formula is: , The table shows two different zero points. and These are the positions of the two zeros; a weighting factor is introduced. This allows the triangulation to simultaneously consider the density and geometric distance of the zero points. ,in It is the local density of the zero point on the time-frequency plane, and the weighted edge length of each edge of the triangulation is obtained through weighting factors. , The effective length of each edge is dynamically adjusted; for each triangle formed by triangulation, if the weighted side length of any edge exceeds a preset side length threshold... If the triangle is larger than the side length threshold, then it is considered to belong to the signal domain; The triangles cluster together to obtain an estimate of the signal domain in the time-frequency plane: Based on the obtained signal domain estimate in the time-frequency plane, the signal in the signal domain is reconstructed using the inverse short-time Fourier transform, and the estimated value of the noise-free signal is calculated. .

7. The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization according to claim 6, characterized in that, The content of step S3 is as follows: S31, Expected Step: Calculate the expected value of the vibration signal, and express the estimated value of the noise-free signal with multiple components as follows: ， ， It is the instantaneous amplitude of the k-th component. It is the instantaneous phase. It's noise; Then, Monte Carlo sampling is used to estimate the posterior distribution, which is generated from the prior distribution. Sample Here, the superscript i represents different times, assuming the observed signal at each time point The above is independent for each sample. Calculate the estimated value of the noise-free signal. likelihood function value ： ， Then calculate the weight of each sample. Then, using a weighted average method, the expected value of the posterior distribution is estimated: ； S32, Maximization Step: Obtaining the desired value that conforms to the characteristics of the actual vibration signal. Next, the estimated values ​​of instantaneous frequency and amplitude need to be updated. First, substitute the expected values ​​of the vibration signal characteristics obtained above, maximize the log-likelihood function of the expected value, and find the optimal instantaneous frequency and amplitude: ,in Indicates the first The parameter estimates after the next iteration are used to update the instantaneous frequency and amplitude parameters using the result of maximizing the expectation: ， ,in is the first Instantaneous phase after the next iteration Let be the instantaneous amplitude of the k components after the t-th iteration. The learning rate; S33. To ensure the stability and accuracy of parameter estimation, the expectation step S31 and the maximization step S32 need to be repeated until the parameter estimation converges. The convergence condition for parameter estimation is: ， It is the first Parameter estimation after the next iteration The convergence threshold is used; finally, the instantaneous frequency of each signal component is extracted. and instantaneous amplitude This serves as the input condition for the next step of order spectral analysis.

8. The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization according to claim 7, characterized in that, Step S4 involves: after extracting the instantaneous frequency and instantaneous amplitude of the vibration signal components, obtaining the instantaneous frequencies for signal feature classification and rotational speed feature classification. and Utilizing instantaneous frequency and Feature extraction is performed on the variable speed signal, and fault diagnosis of the wind turbine drivetrain is realized based on the results of the feature extraction.

9. The wind turbine drivetrain fault diagnosis method based on time-frequency plane expectation maximization according to claim 8, characterized in that, Using instantaneous frequency and Feature extraction of variable speed signals employs the order ratio analysis method, defining the order ratio. It is a multiple of the rotational speed. Based on the magnitude of the order ratio, different fault types and fault locations in the wind turbine drive train can be characterized.

10. A method for fault diagnosis of wind turbine drivetrain based on time-frequency plane expectation maximization according to claim 2, characterized in that, The gap between adjacent comb teeth (2.2) is greater than the sliding distance of the mass block (2) in the groove.

Images