Wind power gear box variable speed fault diagnosis method based on LMSRCT and medium

Abstract

The invention discloses a wind power gear box variable speed fault diagnosis method based on LMSRCT and a medium, and belongs to the field of wind power fault detection.The method comprises the following steps that a vibration acceleration sensor is installed on a wind power gear box, an original vibration signal x (t) of the gear box in the running state is collected at the sampling frequency f s, and meanwhile a rotating speed pulse signal is collected; obtaining accurate rotation frequency f < r > (t) and shaft rotation angle information theta (t); preprocessing the collected original vibration signal x (t) to obtain x pre (t); the x pre (t) is converted to an angle domain through an LMSRCT algorithm, and a one-dimensional angle domain sequence signal s (theta) is generated; the s (theta) is segmented into a sample sequence with a fixed length L for embedded encoding, the sample sequence is input to a four-layer Transform encoder, and probability distribution of different fault types is output; and taking the fault type corresponding to the maximum probability value as a final diagnosis result. According to the method, the LMSRCT algorithm is adopted to convert the vibration signal from the time domain to the angle domain, the problem of diagnosis failure caused by spectrum aliasing is solved, and the accuracy of fault diagnosis is improved by combining the LMSRCT algorithm with a Transform model.

Description

Technical Field

[0001] This invention belongs to the field of wind power fault detection, specifically relating to a method and medium for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT. Background Technology

[0002] The wind turbine gearbox is a crucial component of a wind turbine generator set, and its operating status directly affects the performance and reliability of the generator set. Various faults can occur in wind turbine gearboxes during operation, such as bearing failure and gear failure. These faults often manifest as vibration signals, making vibration signal analysis and diagnosis key to wind turbine gearbox fault diagnosis. Currently, time-frequency analysis methods are widely used for vibration signal processing; however, traditional methods are susceptible to aliasing effects when dealing with large variations in rotational speed.

[0003] Patent document CN103712792A discloses a fault diagnosis method for wind turbine gearboxes. It involves performing order spectrum analysis on vibration signals and speed pulse signals during gearbox operation to obtain meshing vibration energy. Then, based on the quantitative relationship between meshing vibration energy and generator output power, the meshing vibration energy is fitted using least squares and its mean is normalized to eliminate interference from generator output power. However, least squares fitting struggles to capture complex nonlinear coupling or higher-order interaction effects, and without incorporating deep learning models, its ability to mine high-dimensional features is limited.

[0004] Patent document CN116124448A discloses a fault diagnosis system and method for wind turbine gearboxes. It extracts feature vectors from vibration, current, and temperature signals using a convolutional neural network, constructing a Gaussian mixture model to achieve fault detection in wind turbine gearboxes. However, this method requires simultaneous acquisition of multiple signal features, resulting in high acquisition costs and computational resource consumption. Summary of the Invention

[0005] To address the problems in the prior art, this invention proposes a method and medium for diagnosing wind turbine gearbox speed variation faults based on LMSRCT. The Local Maximum Synchronous Compression Redistribution Frequency Variation (LMSRCT) time-frequency analysis algorithm can effectively remove aliasing caused by speed variation. Combined with accurate feature extraction from the Transformer model and deep learning model training, it can effectively diagnose wind turbine gearbox faults.

[0006] The technical solution of the present invention is as follows: In a first aspect, the present invention provides a method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT, comprising the following steps: A triaxial acceleration sensor is installed on the high-speed shaft bearing housing and planetary carrier bearing housing of the wind turbine gearbox, with a sampling frequency of... f s Acquire raw vibration signals of the gearbox during operation. x ( t Simultaneously, speed pulse signals are acquired through an encoder or keyer to obtain accurate rotational frequency. f r ( t and axis rotation angle information i ( t ); The acquired raw vibration signals x ( t Preprocessing is performed, and the preprocessed signal is denoted as... x pre ( t ); Using the LMSRCT algorithm to x pre ( t Transform to the angle domain to generate a one-dimensional angle domain sequence signal. s ( i ); Will s ( i Divided into fixed lengths L sample sequence For the sequence Embedding encoding is performed, and the embedded sequence is input into a four-layer Transformer encoder for processing, outputting the probability distribution of different fault types; The fault type corresponding to the highest probability value is taken as the final diagnostic result.

[0007] Furthermore, the preprocessing includes noise reduction and normalization. The denoising process utilizes wavelet thresholding to suppress environmental noise and random interference. The normalization process normalizes the amplitude of the denoised vibration signal so that its mean is 0 and its standard deviation is 1.

[0008] Furthermore, the conversion to the angle domain specifically refers to: Preprocessed signal x pre ( t Inputting the LMSRCT algorithm yields the time-frequency representation. This improves time-frequency resolution and sharpens the time-frequency ridges of the signal; Based on the collected axis rotation angle information i ( t ), representing time and frequency Remapping from the time-frequency domain to the angle-frequency domain; By performing equal-interval resampling in the angle domain, the non-stationary time-domain signal is converted into a stationary angle-domain sequence signal.

[0009] Furthermore, the time-frequency representation The formula for calculation is: ; in, G ( t , f () represents the initial time-frequency representation; d () represents the Dirac delta function; t Use time coordinates; The time coordinates are redistributed when the instantaneous frequency falls within the ridge of a local maximum on the frequency axis; oh Frequency coordinates; The frequency coordinates are the redistributed values ​​when the instantaneous frequency falls within the ridge of the local maximum of the frequency axis.

[0010] Furthermore, the embedding encoding specifically includes: Each sampling point of the sample sequence is mapped to a linear projection layer. d model From a dimensional vector, we obtain a value embedding vector; A position encoding vector composed of sine and cosine functions is introduced and added to the value embedding vector to provide the model with position information for each sampling point.

[0011] Furthermore, each layer of the Transformer encoder contains two sub-layers: a multi-head self-attention layer and a feedforward neural network layer. The multi-head self-attention layer simultaneously focuses on different parts of the sample sequence, adaptively capturing long-range dependencies related to faults in the angle domain sequence; The feedforward neural network layer consists of a fully connected network and is used to perform nonlinear transformations on the vector at each position.

[0012] Furthermore, each sublayer is followed by residual connections and layer normalization to optimize the training results of the Transformer encoder.

[0013] Furthermore, the different fault types are analyzed using a Softmax classifier to output the probability distributions corresponding to each fault type. ; in, p n The probability of being the nth fault category; c This represents the number of fault categories.

[0014] Furthermore, the diagnostic results can be displayed, stored, or used to trigger warning signals through a human-machine interface.

[0015] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, the computer program being used to execute any of the above-described methods for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention combines LMSRCT time-frequency analysis technology with equal-angle resampling to convert the vibration signal from the time domain to the angle domain. In the angle domain, the fault characteristics are manifested as stable order components that are independent of the absolute value of the rotational speed, thereby solving the diagnostic failure problem caused by spectral aliasing in traditional methods under variable speed conditions. The time-frequency resolution provided by the LMSRCT algorithm can clearly characterize the instantaneous frequency changes of non-stationary signals, thereby enhancing the visibility of weak fault features. The Transformer model, with its powerful multi-head self-attention mechanism, can efficiently capture the long-range dependent features such as periodicity and harmonics of fault impacts in the angle domain sequence, realizing end-to-end intelligent diagnosis and avoiding dependence on complex manual feature engineering. By integrating signal processing techniques with deep learning models, the accuracy and robustness of various common faults are significantly improved. Attached Figure Description

[0017] Figure 1 A flowchart of a wind turbine gearbox variable speed fault diagnosis method based on LMSRCT; Figure 2 Here is a flowchart of the LMSRCT algorithm; Figure 3 This is a flowchart for fault diagnosis using the Transformer model. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Example 1 This embodiment provides a wind turbine gearbox speed change fault diagnosis method based on LMSRCT. The core of this method lies in converting the non-stationary time-domain vibration signal into a stable angle-domain sequence using the Local Maximum Synchronous Compression Redistribution Frequency Change (LMSRCT) algorithm to eliminate the influence of speed fluctuations. Subsequently, a Transformer encoder is used to perform deep feature extraction and fault classification on the angle-domain sequence. The specific process of this method is as follows: Figure 1 As shown, it includes the following steps: A triaxial acceleration sensor is installed on the high-speed shaft bearing housing and planetary carrier bearing housing of the wind turbine gearbox, with a sampling frequency of... f s Acquire raw vibration signals of the gearbox during operation. x ( t Simultaneously, speed pulse signals are acquired via an encoder or keyer to obtain accurate rotational frequency. f r ( t and axis rotation angle information i ( t ).

[0020] The acquired raw vibration signals x ( t Preprocessing is performed to improve signal quality. Preprocessing includes: denoising, using wavelet thresholding to suppress environmental noise and random interference; and normalization, which normalizes the amplitude of the denoised signal to a mean of 0 and a standard deviation of 1, to accelerate the training and convergence process of the subsequent model. The preprocessed signal is denoted as... x pre ( t ).

[0021] like Figure 2 and Figure 3 The image shows the LMSRCT-Transformer network structure. The preprocessed signal... x pre ( t The input to the LMSRCT algorithm yields a time-frequency representation with high time-frequency clustering. The LMSRCT algorithm can effectively overcome the insufficient resolution of traditional time-frequency analysis methods through synchronous compression and redistribution techniques, and can sharpen the time-frequency ridges of the signal; based on the acquired axis rotation angle information i ( t ), representing time and frequency The time-frequency domain signal is remapped to the angle-frequency domain, and then resampled at equal intervals in the angle domain (1024 points per revolution) to convert the non-stationary time-domain signal into a stationary angle-domain sequence signal, generating a one-dimensional angle-domain sequence signal. s ( i In the angular domain, the gear meshing frequency, fault characteristic frequency, etc., all become order components independent of the rotational speed, thus eliminating the spectral aliasing phenomenon caused by changes in rotational speed.

[0022] The LMS RCT algorithm is specifically as follows: First, the preprocessed vibration signal... x pre ( t Perform a short-time Fourier transform (STFT): ; in, G ( t , f ) represents the initial time-frequency representation, i.e. Figure 2 The fundamental time-frequency representation in [the text]; f For frequency; g () represents a window function; j The imaginary unit; t This refers to the position of the window function as it slides along the time axis; in this embodiment... g () Take the Gaussian window function; Calculate the partial derivative of the phase of the short-time Fourier transform to estimate the signal at each time frequency point. t , f The instantaneous frequency on ) is: ; in, for G ( t , f ) on time t The partial derivatives, through x pre ( t The time-domain derivative of the window function The result is obtained from the short-time Fourier transform. Synchronous compression transform (SST) transforms energy in the time-frequency plane along the frequency... f The axis is reassigned to a new frequency coordinate system determined by the estimated instantaneous frequency, calculated as follows: ; in, d () represents the Dirac delta function; oh These are the compressed frequency coordinates; Local maximum search is introduced as a guide for redistribution. For each fixed time t, the time spectrum after synchronization compression is... Above, we search for local maxima along the frequency ω-axis. These maxima are the ridge locations where the signal component is most likely to exist at that moment, and are denoted as the set: ; Local maximum redistribution can effectively filter out a large number of weak time-frequency points caused by noise, so that the redistribution process focuses only on the dominant signal component; A finer redistribution is performed at each ridge position. The redistribution frequency change (RCT) considers both the instantaneous frequency and the rate of change of frequency change. The rate of change of frequency change is the derivative of the instantaneous frequency, effectively tracking the trend of frequency change. The calculation formula is: ; Frequency modulation rate of change calculation x pre ( t The time-domain second derivative of the window function The result is obtained through short-time Fourier transform calculation. For each time frequency point ( t , f If its instantaneous frequency Falling on a certain ridge line oh k Within its neighborhood, its energy is redistributed to the new time-frequency coordinates. : ; ; in, The time coordinates after reallocation; The frequency coordinates are the reassigned coordinates; Indicates taking the real part; e To prevent positive numbers with a denominator of zero; The new time-frequency coordinates are fine-tuned based on the frequency modulation rate of change, so that energy is concentrated not only at the correct frequency, but also at the correct time. All time-frequency points that meet the conditions ( t , f ) energy According to the new time-frequency coordinates By accumulating the data, a highly clustered time-frequency representation of LMSRCT is obtained. : ; get Then, using the synchronously acquired speed signal i ( t ), creation time t Rotation angle with axis i The mapping relationship is achieved through equal angle differences, From the time-frequency domain Resampling to the angle-frequency domain One-dimensional angle domain sequence signal is generated by integrating key frequencies. s ( i ).

[0023] One-dimensional angle domain sequence signal s ( i As features, these are input into the Transformer encoder model for fault diagnosis, converting continuous angle domain sequence signals. s ( i Divided into fixed lengthsL sample sequence , as the input samples for the encoder; For sequence For embedding encoding, value embedding is first performed, and each sampling point is processed through a linear projection layer. s i Mapped to a d model The vector is then used for position encoding. To prevent the Transformer model from losing the sequence order information, a position encoding vector composed of sine and cosine functions is introduced and added to the value embedding vector to provide the model with the position information of each sampling point in the sequence. The embedded sequence is fed into a four-layer Transformer encoder for processing. Each encoder layer contains two sub-layers: a multi-head attention layer and a feedforward neural network layer. The multi-head attention mechanism allows the model to simultaneously focus on different parts of the sequence at different positions, thereby adaptively capturing long-range dependencies related to faults in the angular domain sequence. When a fault cycle occurs at any position in the sequence, the multi-head attention layer can learn the periodic pattern between fault impacts. The feedforward neural network layer is composed of a fully connected network and is used to perform nonlinear transformations on the vector at each position. Each sub-layer is followed by residual connections and layer normalization to promote the training of deep networks. Residual connections prevent feature loss during deep training, and layer normalization stabilizes the layer input distribution, making the training process of the four-layer Transformer encoder smoother and accelerating the convergence speed. The vector corresponding to the first position output by the Transformer encoder is taken as the feature representation of the entire sequence. Using a Softmax classifier, the probability distribution corresponding to different fault types is output. ,in c The number of fault categories (such as tooth surface wear, broken teeth, pitting, etc.).

[0024] Finally, based on the probability distribution P output by the Softmax classifier, the fault category corresponding to the maximum probability value is output as the final diagnostic result. This result can be displayed, stored, or used to trigger a warning signal through the human-machine interface.

[0025] Example 2 This embodiment provides a computer-readable storage medium, which is a non-volatile memory and stores computer-executable instructions. The computer-executable instructions are used to execute a wind turbine gearbox speed change fault diagnosis method based on LMSRCT. The specific method can be referred to the method described in Embodiment 1, and will not be repeated here for the sake of simplicity.

[0026] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure made using the contents of the present invention specification and drawings, or directly or indirectly applied to other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT, characterized in that, Includes the following steps: A triaxial acceleration sensor is installed on the high-speed shaft bearing housing and planetary carrier bearing housing of the wind turbine gearbox, with a sampling frequency of... f s Acquire raw vibration signals of the gearbox during operation. x ( t Simultaneously, the rotational speed pulse signal is acquired through an encoder or keyer to obtain the rotational frequency. f r ( t and axis rotation angle information θ ( t ); The acquired raw vibration signals x ( t Preprocessing is performed, and the preprocessed signal is denoted as... x pre ( t ); Using the LMSRCT algorithm to x pre ( t Transform to the angle domain to generate a one-dimensional angle domain sequence signal. s ( θ ); Will s ( θ Divided into fixed lengths L sample sequence For the sequence Embedding encoding is performed, and the embedded sequence is input into a four-layer Transformer encoder for processing, outputting the probability distribution of different fault types; The fault type corresponding to the highest probability value is taken as the final diagnostic result.

2. The method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 1, characterized in that, The preprocessing includes noise reduction and normalization. The denoising process utilizes wavelet thresholding to suppress environmental noise and random interference. The normalization process normalizes the amplitude of the denoised vibration signal so that its mean is 0 and its standard deviation is 1.

3. The method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 1, characterized in that, The conversion to the angle domain specifically refers to: Preprocessed signal x pre ( t Inputting the LMSRCT algorithm yields the time-frequency representation. This improves time-frequency resolution and sharpens the time-frequency ridges of the signal; Based on the collected axis rotation angle information θ ( t ), representing time and frequency Remapping from the time-frequency domain to the angle-frequency domain; By performing equal-interval resampling in the angle domain, the non-stationary time-domain signal is converted into a stationary angle-domain sequence signal.

4. The method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 3, characterized in that, The time-frequency representation The formula for calculation is: ； in, G ( t , f () represents the initial time-frequency representation; δ () represents the Dirac delta function; t Use time coordinates; The time coordinates are redistributed when the instantaneous frequency falls within the ridge of a local maximum on the frequency axis; ω Frequency coordinates; The frequency coordinates are the redistributed values ​​when the instantaneous frequency falls within the ridge of the local maximum of the frequency axis.

5. The method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 1, characterized in that, The embedding encoding is specifically as follows: Each sampling point of the sample sequence is mapped to a linear projection layer. d model From a dimensional vector, we obtain a value embedding vector; A position encoding vector composed of sine and cosine functions is introduced and added to the value embedding vector to provide the model with position information for each sampling point.

6. The method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 1, characterized in that, Each layer of the Transformer encoder contains two sub-layers: a multi-head self-attention layer and a feedforward neural network layer. The multi-head self-attention layer simultaneously focuses on different parts of the sample sequence, adaptively capturing long-range dependencies related to faults in the angle domain sequence; The feedforward neural network layer consists of a fully connected network and is used to perform nonlinear transformations on the vector at each position.

7. A method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 6, characterized in that, Each sublayer is followed by residual connections and layer normalization to optimize the training results of the Transformer encoder.

8. The method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 1, characterized in that, The different fault types are classified using a Softmax classifier, which outputs the probability distribution corresponding to each fault type. ; in, p n The probability of being the nth fault category; c This represents the number of fault categories.

9. A method for diagnosing variable speed faults in wind turbine gearboxes based on LMSRCT according to claim 1, characterized in that, The diagnostic results are displayed, stored, or used to trigger warning signals through a human-machine interface.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is used to execute the wind turbine gearbox variable speed fault diagnosis method based on any one of claims 1 to 9.

Images