Resonance excitation assisted wind power disc bearing acoustic emission fault diagnosis method and system

Abstract

This invention discloses a method and system for diagnosing acoustic emission faults in wind turbine slewing bearings using resonant excitation, relating to the field of condition monitoring and non-destructive testing technology for key components of wind power equipment. This invention induces the load-bearing structure into a high-dynamic response state through resonant excitation, improving the observability of microscopic contact events of defects. It can obtain more sufficient and stable effective acoustic emission events under low-speed micro-oscillation conditions, enhancing the identification capability of weak faults such as early micro-cracks and early spalling. Because it uses frequency sweeping to identify the resonant frequency and performs fixed-frequency excitation, the enhancement process is repeatable, which is beneficial for the consistency of threshold setting and diagnostic results. The method of this invention does not rely on the high rotational speed of the bearing itself to generate excitation, making it suitable for typical low-speed heavy-load and micro-oscillation scenarios of wind turbine slewing bearings, and can be used for experimental verification and engineering monitoring applications.

Claims

1. A wind turbine turntable bearing acoustic emission fault diagnosis system assisted by resonance excitation, characterized in that: It includes a working condition simulation unit, a resonance excitation unit, an acoustic emission acquisition unit, and a synchronization control unit; The operating condition simulation unit is used to provide low-speed heavy load and slight oscillation conditions for the slewing bearing and to maintain structural installation stability. The resonant excitation unit includes a signal generator, a power amplifier, and an exciter. The exciter is fixed to a structural part that is rigidly associated with the raceway load path via a connector to ensure effective input of the excitation force. The structural part that is rigidly associated with the raceway load path includes at least a bearing housing and an outer ring flange. The acoustic emission acquisition unit includes an acoustic emission sensor, a coupling medium, a fixing fixture, a preamplifier, a filtering module, and a data acquisition and analysis module; The synchronization control unit is used to synchronously trigger and align the acquisition of excitation signals, operating parameters and acoustic emission signals.

2. A method for diagnosing acoustic emission faults in wind turbine turntable bearings with resonant excitation assistance, based on the acoustic emission fault diagnosis system for wind turbine turntable bearings with resonant excitation assistance as described in claim 1, characterized in that: At least the following steps are included: S1: System preparation and data acquisition. Acoustic emission sensors are deployed in the defect-sensitive area of ​​the wind turbine slewing bearing to simulate the low-speed, heavy-load, micro-oscillation condition of the wind turbine slewing bearing, and the raw acoustic emission signals of the bearing under this condition are collected. Simultaneously acquire excitation reference channel signals and operating parameters, and establish a background noise baseline; S2: Resonance frequency identification and fixed-frequency excitation enhancement. A sweep frequency excitation is applied to the bearing structure of the wind turbine turntable bearing, the structural response signal is collected and the resonant frequency of the structure is identified. Then, a fixed-frequency excitation matching the resonant frequency is applied to make the bearing structure enter a steady-state resonance state, and the enhanced acoustic emission signal under the resonance state is collected. S3: Signal preprocessing and sample construction. The enhanced acoustic emission signal is detrended, and a sample sequence is formed by sliding segmentation with a fixed window length. Each sample segment is then normalized using Min-Max. S4: MVMD mode decomposition and reconstruction. Multiscale variational mode decomposition is performed on the normalized time-domain samples to obtain multiple intrinsic mode components. The effective mode set is obtained through mode screening. The effective modes are superimposed and reconstructed to obtain the reconstructed samples. S5: FFT spectrum construction, perform fast Fourier transform on the reconstructed sample to construct the frequency domain complex spectrum, extract the single-sided amplitude spectrum and perform normalization processing to obtain the frequency domain feature vector; S6: Feature enhancement is performed using the CBAM attention module. The normalized frequency domain features are input into the CBAM attention module, and channel attention weighting and time-series attention weighting are performed sequentially to output the attention-enhanced spectral feature vector. S7: TCN Feature Extraction and Fault Classification. The attention-enhanced spectral feature vector is input into the temporal convolutional network to extract multi-scale temporal features. The extracted features are then input into the classifier to output the fault category of the wind turbine turntable bearing.

3. The method for diagnosing acoustic emission faults in wind turbine turntable bearings with resonant excitation assistance according to claim 2, characterized in that: S1 includes at least the following steps: S1.1: Install the wind turbine turntable bearing in the working condition simulation unit, control the bearing to be in a low-speed rotation or slight oscillation state, and apply at least one of the following loads: axial load, radial load, and overturning moment to simulate yaw or pitch working conditions. S1.2: Arrange acoustic emission sensors on the outer flange, bearing housing, or metal surface rigidly associated with the load transfer path of the wind turbine turntable bearing; S1.3: Synchronously acquire the raw acoustic emission signal through a data acquisition device. Excitation signal reference channel and operating parameters; all acquired signals are triggered and aligned with a unified clock. S1.4: Acquire background sound emission data under conditions where no resonant excitation is applied or the excitation amplitude is zero, statistically analyze the amplitude distribution and event rate of background noise, and determine the threshold and filtering frequency band for signal processing.

4. The method for diagnosing acoustic emission faults in wind turbine turntable bearings assisted by resonant excitation according to claim 3, characterized in that: S2 includes at least the following steps: S2.1: Frequency sweep excitation, the frequency sweep signal is output by the signal generator, and the exciter is driven by the power amplifier to apply external excitation to the bearing bearing structure. The frequency sweep signal is a linear frequency sweep signal or a logarithmic frequency sweep signal. S2.2: Response measurement: During the frequency sweep process, the structural response signal is acquired. The energy response of the acoustic emission signal, the acceleration sensor signal, or the change in the input current or voltage of the exciter are used as the response characterization. The frequency response function or amplitude-frequency curve is calculated. S2.3: Resonance identification, identifying the resonant frequency corresponding to the peak amplitude in the amplitude-frequency curve. And determine the candidate interval A fine scan is performed in the candidate region to obtain stability. ; S2.4: Fixed-frequency excitation enhancement, switching the excitation to a frequency equal to or close to... A fixed-frequency sinusoidal signal is used to induce a steady-state resonance in the structure, and the enhanced acoustic emission signal is then acquired under this state. .

5. The method for diagnosing acoustic emission faults in wind turbine turntable bearings assisted by resonant excitation according to claim 2, characterized in that: The S3 includes at least the following steps: S3.1: Detrending, for the enhanced acoustic emission signal Perform mean and trend removal processing, and limit the amplitude of abnormal saturation points in the signal; S3.2: Window segmentation, with a fixed window length L. Sliding segmentation forms sample sequences The sample in segment k is: in, This is the index of the starting sampling point of the k-th segment; S3.3: Min-Max normalization, performing normalization on each sample segment to obtain... : in, To prevent division by zero of constants; and These are the minimum and maximum values ​​of the current segment sample, respectively.

6. The method for diagnosing acoustic emission faults in wind turbine turntable bearings assisted by resonant excitation according to claim 5, characterized in that: The normalized time-domain sample obtained by S4 from S3 Input the MVMD decomposition module to obtain k intrinsic mode components, and obtain samples through mode filtering and reconstruction. This is used for subsequent FFT spectrum construction; The S4 includes at least the following steps: S4.1: Set MVMD decomposition parameters: number of modes K, penalty factor Time step Convergence threshold Maximum number of iterations and the initial center frequency setting method; S4.2: Perform MVMD decomposition on the samples. Variational mode decomposition yields a set of modal components: in, For the first One modal component; For residual terms; S4.3: Screen the modes to avoid introducing low-frequency trends or random noise modes that are unrelated to defects into subsequent features, and obtain the effective mode set Ω; S4.4: The selected effective modes are superimposed and reconstructed to obtain the reconstructed sample: Reconstructed While maintaining the time-domain structure of the defect impact, random noise and irrelevant trend terms are suppressed, making the spectral peaks and frequency band energy distribution of the subsequent FFT more stable and separable.

7. The method for diagnosing acoustic emission faults in wind turbine turntable bearings with resonant excitation assistance according to claim 6, characterized in that: The methods for screening modes in S4.3 include at least the correlation coefficient criterion, kurtosis criterion, energy concentration criterion, and center frequency constraint.

8. The method for diagnosing acoustic emission faults in wind turbine turntable bearings with resonant excitation assistance according to claim 6, characterized in that: The S5 includes at least the following steps: S5.1: FFT spectrum construction for reconstructed samples Perform a fast Fourier transform to obtain the frequency domain complex spectrum. : Where m is the frequency domain discrete frequency index; L is the length of each sample segment; These are Fourier basis functions; S5.2: Extraction of one-sided amplitude spectrum, calculation and normalization of two-sided amplitude spectrum: Extract the single-sided amplitude spectrum as a sample feature: And a doubling correction is applied to the amplitudes other than the DC and Nyquist components: This yields the frequency domain feature vector corresponding to each sample segment. Its dimensions are ; S5.3: Frequency domain normalization. Min-Max normalization is performed to obtain frequency domain features, which are then used as input for subsequent deep networks. One-sided amplitude spectrum The numerical values ​​are linearly mapped to the [0,1] interval, eliminating the difference in spectral amplitude scale between different samples and ensuring the consistency and stability of subsequent deep network inputs.

9. The method for diagnosing acoustic emission faults in wind turbine turntable bearings assisted by resonance excitation according to claim 8, characterized in that: S6 includes at least the following steps: S6.1: Input representation construction, using a one-dimensional spectral feature vector The feature tensor is mapped to a multi-channel feature tensor through one-dimensional convolution. Where C is the number of channels and T is the time length; S6.1: For Global average pooling and global max pooling are performed in the time dimension to obtain the channel description vector: in, This is the channel description vector obtained through global average pooling; This is the channel description vector obtained through global max pooling; For global average pooling; This is for global max pooling; By inputting both into a shared MLP, the channel weight vector is obtained: in, It is a multilayer perceptron; Using the Sigmoid activation function, compress the output to... ; And by weighting the channels, we get: Used to highlight key frequency bands related to defect events; S6.3: For Perform average pooling and max pooling along the channel dimension and concatenate the results to obtain the frequency location description: in, Average pooling is performed along the channel vitamin C; Max pooling is performed along the channel C; For splicing operations; The positional weights are obtained through one-dimensional convolution and Sigmoid: in, It is a one-dimensional convolution; Position-weighted summation yields: Used to highlight the critical time and location of the impact; S6.4: Output spectral characteristics, By mapping back to one-dimensional spectral features through global pooling or flattening, we obtain the attention-enhanced spectral feature vector: in, This represents global average pooling and linear mapping operations.

10. The method for diagnosing acoustic emission faults in wind turbine turntable bearings assisted by resonant excitation according to claim 9, characterized in that: The S7 enhances the attention-enhanced spectral features output by the S6. Inputting into a TCN network, multi-layer causal dilated convolution is used to extract multi-scale temporal features, and the output feature vector is... ; Then Input classifier outputs fault categories To achieve state recognition, the state recognition includes at least normal, wear pitting, and crack; The TCN network employs a multi-layer residual network composed of causal convolution and dilated convolution to ensure temporal causality and expand the receptive field. Will Considering a one-dimensional sequence of length T as input, the TCN... The layer uses causal dilated convolution: in, The kernel length is [length]. For the first The first layer of convolution kernel One learnable weight parameter, The position index within the convolution kernel; For the first The output features of the current layer are used as the input of the current layer. For the first Layer expansion factor; ; Stable training is achieved through residual connections and nonlinear activation. Perform global average pooling or extract the features from the last time step of the TCN network's output to obtain the feature vector: Perform classification and output, Input a classifier to obtain the probability of each category and the predicted fault category, thereby realizing state identification; the classifier includes at least a Softmax classifier, an SVM classifier, or a fully connected layer classifier.

Images