A method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation
By using selective trend estimation and frequency domain analysis methods, noise and natural frequencies in rotating machinery signals are eliminated, sideband characteristics are significantly enhanced, and the problem of noise and natural frequency contamination in traditional methods is solved, thus achieving efficient fault diagnosis and equipment health monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG MINGYANG WIND POWER IND GRP CO LTD
- Filing Date
- 2026-01-23
- Publication Date
- 2026-06-02
AI Technical Summary
Traditional signal processing methods struggle to effectively extract target sideband information from rotating machinery, and are susceptible to environmental noise and inherent frequency contamination, resulting in low fault diagnosis efficiency.
By employing selective trend estimation, bandpass filtering, frequency domain detrending, and envelope spectrum analysis, noise and inherent frequency interference are eliminated through selective trend estimation, significantly enhancing sideband characteristics.
It achieves accurate extraction of sideband features, improves the reliability and accuracy of equipment health monitoring and fault diagnosis, and is suitable for complex noise and inherent frequency interference environments.
Smart Images
Figure CN122132679A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of condition monitoring of rotating machinery, and in particular to a method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation. Background Technology
[0002] Rotating machinery is an important component of industrial systems, and the frequency domain characteristics of its vibration signals are crucial for monitoring equipment health and diagnosing potential faults. However, in actual operating conditions, signals are often severely contaminated by environmental noise and inherent frequency components, making it difficult for traditional signal processing methods to accurately extract target sideband information and resulting in low efficiency in fault diagnosis.
[0003] Traditional signal processing methods, such as time-domain denoising or simple frequency-domain filtering, struggle to effectively address local contamination characteristics in the frequency domain—that is, while some sub-bands within the target frequency band are interfered with by noise, others retain clear sidebands. While wavelet filtering has been applied in multi-scale analysis, its rigid thresholding—directly setting components below a threshold to zero—is ill-suited to this local non-stationarity, often weakening key sideband features. Furthermore, intrinsic frequency components introduce spurious peaks in envelope analysis, masking true modulation sidebands and severely limiting diagnostic accuracy.
[0004] Therefore, there is an urgent need for a method for extracting the frequency domain sidebands of rotating machinery, which can effectively suppress noise and natural frequencies by intelligently distinguishing signal trends and characteristics, thus providing an effective solution for the condition monitoring of rotating machinery. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a method for extracting sidebands in the frequency domain of rotating machinery based on selective trend estimation. Through selective trend estimation, bandpass filtering, frequency domain detrending, and envelope spectrum analysis, noise and inherent frequency interference are accurately eliminated, significantly enhancing sideband features and further achieving accurate extraction of these features, thus providing a reliable basis for equipment health monitoring and fault diagnosis.
[0006] The objective of this invention is achieved through the following technical solution: a method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation, comprising the following steps:
[0007] S1. Obtain the raw vibration signal of the rotating machinery;
[0008] S2. Perform bandpass filtering on the original vibration signal to obtain the filtered vibration signal;
[0009] S3. Perform a Fast Fourier Transform (FFT) on the filtered vibration signal to obtain the frequency domain amplitude spectrum and phase spectrum of the vibration signal.
[0010] S4. Calculate the local spectral kurtosis based on the frequency domain amplitude spectrum;
[0011] S5. Set the spectral kurtosis threshold. Based on the spectral kurtosis threshold, the local spectral kurtosis is divided into peak region and background region. The region where the local spectral kurtosis is less than the spectral kurtosis threshold is the background region, and the region where the local spectral kurtosis is greater than the spectral kurtosis threshold is the peak region. The peak region represents the fault vibration signal, and the background region represents the noise and natural frequency in the vibration signal.
[0012] S6. Perform selective trend estimation on the local spectral kurtosis of the peak region and the background region respectively, and generate selective trend estimation curves;
[0013] S7. Subtract the selective trend estimation curve from the frequency domain amplitude spectrum to generate the detrended spectrum;
[0014] S8. Perform non-negativity constraint correction on the detrended spectrum;
[0015] S9. Perform symmetrical processing on the detrended amplitude spectrum and combine it with the phase spectrum obtained in step S3 to reconstruct the complex spectrum. Convert the vibration signal into a time-domain vibration signal by inverse fast Fourier transform (IFFT) on the complex spectrum and remove the DC component.
[0016] S10. Perform Hilbert transform on the time-domain vibration signal to extract the envelope signal and generate an envelope spectrum with significantly enhanced frequency domain sideband features.
[0017] Furthermore, the rotating machinery includes a wind turbine, a gearbox, and a compressor.
[0018] Furthermore, step S2 includes:
[0019] The original vibration signal of the rotating machinery is bandpass filtered to determine the target frequency band of the original vibration signal and filter out irrelevant low-frequency and high-frequency components.
[0020] Furthermore, step S4 includes:
[0021] Calculate the local spectral kurtosis of the frequency domain amplitude spectrum based on the frequency domain amplitude spectrum. :
[0022] ;
[0023] Where W is the window width, the window W is calculated from the starting frequency of the frequency domain amplitude spectrum, and slides towards higher frequencies in one step, traversing the entire target frequency band; It is a window Amplitude data points within, It is a window The average value of the internal amplitude data.
[0024] Furthermore, step S6 includes:
[0025] Selective trend estimation is achieved using nonparametric trend filtering or low-order fitting methods, including moving median filtering, moving quantile filtering, Savitzky-Golay filtering, smoothing prior methods, and weighted low-order polynomial fitting.
[0026] A rotating machinery frequency domain sideband extraction system based on selective trend estimation, used to implement the aforementioned rotating machinery frequency domain sideband extraction method based on selective trend estimation, includes:
[0027] A signal acquisition module is used to acquire the raw vibration signals of the rotating machinery, which includes a wind turbine, a gearbox, and a compressor.
[0028] The signal processing module is used to perform bandpass filtering on the original vibration signal of the rotating machinery, determine the target frequency band of the original vibration signal and filter out irrelevant low-frequency and high-frequency components, and perform Fast Fourier Transform (FFT) on the bandpass-filtered vibration signal to obtain the frequency domain amplitude spectrum and phase spectrum of the vibration signal.
[0029] The local spectral kurtosis calculation module calculates the local spectral kurtosis based on the frequency domain amplitude spectrum.
[0030] The local spectral kurtosis segmentation module sets a spectral kurtosis threshold and divides the local spectral kurtosis into peak regions and background regions based on the spectral kurtosis threshold.
[0031] The selective trend estimation module performs selective trend estimation on the local spectral kurtosis of the peak region and the background region, respectively, and generates selective trend estimation curves.
[0032] The detrended spectrum generation module subtracts the selective trend estimation curve from the frequency domain amplitude spectrum to generate a detrended spectrum.
[0033] The detrending spectrum correction module performs non-negativity constraint correction on the detrending spectrum;
[0034] The complex spectrum reconstruction and inverse transform module is used to perform symmetrical processing on the detrended amplitude spectrum and, combined with the phase spectrum obtained by the signal processing module, reconstruct the complex spectrum. The complex spectrum is then converted into a time-domain vibration signal by inverse fast Fourier transform (IFFT) and the DC component is removed.
[0035] The envelope spectrum analysis module is used to perform Hilbert transform on time-domain vibration signals, extract the envelope signal, and generate an envelope spectrum with significantly enhanced frequency domain sideband features.
[0036] Furthermore, the local spectral kurtosis calculation module includes:
[0037] Calculate the local spectral kurtosis of the frequency domain amplitude spectrum based on the frequency domain amplitude spectrum. :
[0038] ;
[0039] Where W is the window width, the window W is calculated from the starting frequency of the frequency domain amplitude spectrum, and slides towards higher frequencies in one step, traversing the entire target frequency band; It is a window Amplitude data points within, It is a window The average value of the internal amplitude data.
[0040] Furthermore, the selective trend estimation module includes:
[0041] Different execution strengths are applied to the local spectral kurtosis of the peak region and the background region to perform selective trend estimation, generating selective trend estimation curves;
[0042] Selective trend estimation is achieved using nonparametric trend filtering or low-order fitting methods, including moving median filtering, moving quantile filtering, Savitzky-Golay filtering, smoothing prior methods, and weighted low-order polynomial fitting.
[0043] A non-transitory computer-readable medium storing instructions that, when executed by a processor, perform the steps of the rotating machinery frequency domain sideband extraction method described above based on selective trend estimation.
[0044] A computing device includes a processor and a memory for storing a processor-executable program, wherein when the processor executes the program stored in the memory, it implements the above-described method for extracting the frequency domain sidebands of rotating machinery based on selective trend estimation.
[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0046] 1. Breakthrough noise reduction capability: In response to the local contamination characteristics in the frequency domain, the selective trend estimation described in this invention can intelligently remove noise and inherent frequency interference, retain core sideband features, and achieve accurate modeling of the background baseline.
[0047] 2. Significant sideband enhancement: Selective trend estimation significantly enhances sideband information within the target frequency band by capturing the overall trend of the spectrum rather than sideband peaks, thereby improving signal clarity.
[0048] 3. Applicability: This invention can be widely applied to vibration signal analysis of rotating machinery such as wind turbines and gearboxes, and is adaptable to complex noise and natural frequency interference environments.
[0049] 4. Highly efficient diagnosis: By removing noise and natural frequency trend terms, and combining inverse Fourier transform and envelope spectrum analysis, sideband information is accurately extracted, providing reliable data for equipment health monitoring and fault diagnosis.
[0050] 5. Robustness and Engineering Applicability: Compared with the traditional global minimum error method, the present invention can effectively prevent the disappearance of sidebands caused by excessive smoothing, and has higher robustness and engineering applicability. Attached Figure Description
[0051] Figure 1 This is a flowchart of a method for extracting frequency domain sidebands from rotating machinery.
[0052] Figure 2 This is a schematic diagram of the original amplitude spectrum of the axial vibration signal of a semi-direct drive wind turbine gearbox.
[0053] Figure 3 This is a schematic diagram of the local spectral kurtosis of the axial vibration signal of a semi-direct drive wind turbine gearbox.
[0054] Figure 4 This is a comparison chart of the original amplitude spectrum and the selectively estimated noise trend line of the axial vibration signal of a semi-direct drive wind turbine gearbox.
[0055] Figure 5 This is a comparison of the detrended spectrum and the original amplitude spectrum of the axial vibration signal of a semi-direct drive wind turbine gearbox.
[0056] Figure 6 This is a schematic diagram of the non-negative constraint correction of the detrending spectrum of the axial vibration signal of a semi-direct drive wind turbine gearbox.
[0057] Figure 7 This is a schematic diagram of the envelope spectrum of the axial vibration signal of a semi-direct drive wind turbine gearbox. Detailed Implementation
[0058] The present invention will be further described below with reference to specific embodiments.
[0059] Example 1
[0060] See Figure 1 As shown, taking the axial vibration signal of a semi-direct drive wind turbine gearbox as an example, with a sampling rate of 12800 Hz, the frequency domain sideband extraction method for rotating machinery based on selective trend estimation provided in this embodiment includes the following steps:
[0061] S1. Obtain the original axial vibration signal of the semi-direct drive wind turbine gearbox;
[0062] S2. Bandpass filter the original axial vibration signal, selecting the 4600–5700 Hz frequency band to isolate the target frequency components, and obtain the filtered vibration signal.
[0063] S3. Perform a Fast Fourier Transform (FFT) on the filtered vibration signal to obtain the frequency domain amplitude spectrum and phase spectrum of the vibration signal; see [link / reference]. Figure 2 As shown, the frequency domain amplitude spectrum is a positive frequency amplitude spectrum, showing regular sideband components, but the sideband spacing is difficult to extract due to noise and natural frequency interference.
[0064] S4. Calculate the local spectral kurtosis of the frequency domain amplitude spectrum based on the frequency domain amplitude spectrum. :
[0065] ;
[0066] Where W is the window width, the window W is calculated from the starting frequency of the frequency domain amplitude spectrum, and slides towards higher frequencies in one step, traversing the entire target frequency band; It is a window Amplitude data points within, It is a window The average value of the internal amplitude data.
[0067] S5. Set the spectral kurtosis threshold. Based on the spectral kurtosis threshold, divide the local spectral kurtosis into peak regions and background regions. See [link / reference] Figure 3 As shown. Using kurtosis calculation, the calculated local spectral kurtosis is around 3 for relatively flat background areas, and >3 for sharp peak areas. In order to highlight the peak areas and avoid interference from the background areas, a spectral kurtosis threshold of 6 is selected. Areas with local spectral kurtosis less than the spectral kurtosis threshold are considered background areas, and areas with local spectral kurtosis greater than the spectral kurtosis threshold are considered peak areas.
[0068] S6. Selectively estimate the selective trend by setting different parameters for the local spectral kurtosis of the peak region and the background region, and generate a selective trend estimation curve; see [link / reference] Figure 4 As shown, a non-parametric trend filtering method is used to estimate the low-amplitude background trend of the background region, that is, to selectively estimate the trend of noise; wherein, the non-parametric trend filtering method includes:
[0069] 1) Moving median filter: Estimates the trend using the median (0.5 quantile), suitable for medium noise scenarios.
[0070] 2) Moving quantile filtering: Focuses the target baseline by using quantiles.
[0071] 3) Savitzky-Golay filter: It achieves trend estimation through local polynomial smoothing, obtaining the baseline trend while retaining the lower sideband peaks.
[0072] 4) Smoothing prior method: Estimating background trends based on the prior statistical characteristics of the signal.
[0073] 5) Weighted low-order polynomial fitting: By using linear fitting and weight adjustment, the fitting to the peak value is reduced.
[0074] The above methods all achieve selective trend estimation through parameter settings, such as window size, quantiles, and order, and are used for scenarios with different noise levels and natural frequencies, and can be further extended to adaptive filtering.
[0075] S7. Subtract the selective trend estimation curve from the frequency domain amplitude spectrum to generate the detrended spectrum. (See below) Figure 5 As shown.
[0076] S8. Negative values may appear after detrending. A non-negative correction constraint method (Max(value, 0)) is used to process the detrended spectrum, setting negative values to zero to ensure the detrended spectrum is positive. The corrected detrended spectrum retains sideband peaks while significantly reducing the noise baseline. See also... Figure 6 As shown, this embodiment uses median filtering, and the detrended spectrum after non-negative constraint correction clearly retains the sideband components and significantly suppresses noise.
[0077] S9. The detrended amplitude spectrum is symmetrically processed. In this embodiment, the detrended amplitude spectrum is a positive frequency amplitude spectrum. After symmetrical processing, a negative frequency amplitude spectrum is generated. Combined with the phase spectrum obtained in step S3, the complex spectrum is reconstructed. The complex spectrum is converted into a time-domain vibration signal by inverse fast Fourier transform (IFFT) and the DC component is removed.
[0078] S10. Perform Hilbert transform on the time-domain vibration signal to extract the envelope signal and generate an envelope spectrum with significantly enhanced frequency domain sideband features. (See below) Figure 7 As shown, the envelope spectrum obtained by median filtering in this embodiment clearly shows the sideband components and their harmonics, and the sideband spacing is consistent with the theoretical value within the frequency band. Compared with traditional methods, this method significantly improves the robustness of sideband extraction.
[0079] This invention identifies peak and background regions in the frequency domain amplitude spectrum based on Local Spectrum Kurtosis (LSK). LSK is significantly higher in peak regions and closer to the Gaussian baseline in background regions, thus serving as an important basis for protecting fault sidebands. To achieve highly reliable sideband feature preservation, trend estimation uses different execution strengths for different regions after spectral kurtosis thresholding, avoiding fitting fault-related local peaks (such as modulation sidebands). Compared to traditional global minimum error fitting, this invention effectively prevents sideband disappearance caused by over-smoothing, exhibiting higher robustness and engineering versatility. It achieves the dual objectives of accurate background baseline modeling and complete preservation of fault features.
[0080] Example 2
[0081] This embodiment provides a rotating machinery frequency domain sideband extraction system based on selective trend estimation, used to implement the rotating machinery frequency domain sideband extraction method based on selective trend estimation described in Embodiment 1, including:
[0082] A signal acquisition module is used to acquire the original vibration signals of rotating machinery, including a wind turbine, a gearbox, and a compressor.
[0083] The signal processing module is used to perform bandpass filtering on the original vibration signal of the rotating machinery, determine the target frequency band of the original vibration signal and filter out irrelevant low-frequency and high-frequency components, and perform Fast Fourier Transform (FFT) on the bandpass-filtered vibration signal to obtain the frequency domain amplitude spectrum and phase spectrum of the vibration signal.
[0084] The local spectral kurtosis calculation module calculates the local spectral kurtosis based on the frequency domain amplitude spectrum; after obtaining the amplitude spectrum, it calculates the local spectral kurtosis of the frequency domain amplitude spectrum. :
[0085] ;
[0086] Where W is the window width, the window W is calculated starting from the starting frequency of the spectrum, and slides towards higher frequencies in one step, traversing the entire target frequency band; It is a window Amplitude data points within, It is a window The average value of the internal amplitude data.
[0087] The local spectral kurtosis segmentation module sets a spectral kurtosis threshold and divides the local spectral kurtosis into peak regions and background regions based on the spectral kurtosis threshold.
[0088] The selective trend estimation module sets different parameters for the local spectral kurtosis of the peak region and the background region to perform selective trend estimation and generate a selective trend estimation curve. Selective trend estimation is achieved using non-parametric trend filtering or low-order fitting methods, including moving median filtering, moving quantile filtering, Savitzky-Golay filtering, smoothing prior methods, and weighted low-order polynomial fitting.
[0089] The detrended spectrum generation module subtracts the selective trend estimation curve from the frequency domain amplitude spectrum to generate a detrended spectrum.
[0090] The detrending spectrum correction module performs non-negativity constraint correction on the detrending spectrum;
[0091] The complex spectrum reconstruction and inverse transform module is used to perform symmetrical processing on the detrended amplitude spectrum and, combined with the phase spectrum obtained by the signal processing module, reconstruct the complex spectrum. The complex spectrum is then converted into a time-domain vibration signal by inverse fast Fourier transform (IFFT) and the DC component is removed.
[0092] The envelope spectrum analysis module is used to perform Hilbert transform on time-domain vibration signals, extract the envelope signal, and generate an envelope spectrum with significantly enhanced frequency domain sideband features.
[0093] Example 3
[0094] This embodiment discloses a non-transitory computer-readable medium storing instructions that, when executed by a processor, perform the steps of the rotating machinery frequency domain sideband extraction method based on selective trend estimation as described in Embodiment 1.
[0095] In this embodiment, the non-transitory computer-readable medium can be a disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), USB flash drive, portable hard drive, etc.
[0096] Example 4
[0097] This embodiment discloses a computing device, including a processor and a memory for storing processor-executable programs. When the processor executes the program stored in the memory, it implements the rotating machinery frequency domain sideband extraction method based on selective trend estimation described in Embodiment 1.
[0098] The computing device described in this embodiment may be a desktop computer, laptop computer, smartphone, PDA handheld terminal, tablet computer, programmable logic controller (PLC), or other terminal device with processor function.
[0099] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation, characterized in that, Includes the following steps: S1. Obtain the raw vibration signal of the rotating machinery; S2. Perform bandpass filtering on the original vibration signal to obtain the filtered vibration signal; S3. Perform a Fast Fourier Transform (FFT) on the filtered vibration signal to obtain the frequency domain amplitude spectrum and phase spectrum of the vibration signal. S4. Calculate the local spectral kurtosis based on the frequency domain amplitude spectrum; S5. Set the spectral kurtosis threshold. Based on the spectral kurtosis threshold, the local spectral kurtosis is divided into peak region and background region. The region where the local spectral kurtosis is less than the spectral kurtosis threshold is the background region, and the region where the local spectral kurtosis is greater than the spectral kurtosis threshold is the peak region. The peak region represents the fault vibration signal, and the background region represents the noise and natural frequency in the vibration signal. S6. Perform selective trend estimation on the local spectral kurtosis of the peak region and the background region respectively, and generate selective trend estimation curves; S7. Subtract the selective trend estimation curve from the frequency domain amplitude spectrum to generate the detrended spectrum; S8. Perform non-negativity constraint correction on the detrended spectrum; S9. Perform symmetrical processing on the detrended amplitude spectrum and combine it with the phase spectrum obtained in step S3 to reconstruct the complex spectrum. Convert the vibration signal into a time-domain vibration signal by inverse fast Fourier transform (IFFT) on the complex spectrum and remove the DC component. S10. Perform Hilbert transform on the time-domain vibration signal to extract the envelope signal and generate an envelope spectrum with significantly enhanced frequency domain sideband features.
2. The method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation according to claim 1, characterized in that, The rotating machinery includes a wind turbine, a gearbox, and a compressor.
3. The method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation according to claim 1, characterized in that, Step S2 includes: The original vibration signal of the rotating machinery is bandpass filtered to determine the target frequency band of the original vibration signal and filter out irrelevant low-frequency and high-frequency components.
4. The method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation according to claim 3, characterized in that, Step S4 includes: Calculate the local spectral kurtosis of the frequency domain amplitude spectrum based on the frequency domain amplitude spectrum. : ； Where W is the window width, the window W is calculated from the starting frequency of the frequency domain amplitude spectrum, and slides towards higher frequencies in one step, traversing the entire target frequency band; It is a window Amplitude data points within, It is a window The average value of the internal amplitude data.
5. The method for extracting frequency domain sidebands of rotating machinery based on selective trend estimation according to claim 1, characterized in that, Step S6 includes: Selective trend estimation is achieved using nonparametric trend filtering or low-order fitting methods, including moving median filtering, moving quantile filtering, Savitzky-Golay filtering, smoothing prior methods, and weighted low-order polynomial fitting.
6. A rotating machinery frequency domain sideband extraction system based on selective trend estimation, characterized in that, The method for extracting the frequency domain sidebands of rotating machinery based on selective trend estimation as described in any one of claims 1-5 includes: A signal acquisition module is used to acquire the raw vibration signals of the rotating machinery, which includes a wind turbine, a gearbox, and a compressor. The signal processing module is used to perform bandpass filtering on the original vibration signal of the rotating machinery, determine the target frequency band of the original vibration signal and filter out irrelevant low-frequency and high-frequency components, and perform Fast Fourier Transform (FFT) on the bandpass-filtered vibration signal to obtain the frequency domain amplitude spectrum and phase spectrum of the vibration signal. The local spectral kurtosis calculation module calculates the local spectral kurtosis based on the frequency domain amplitude spectrum. The local spectral kurtosis segmentation module sets a spectral kurtosis threshold and divides the local spectral kurtosis into peak regions and background regions based on the spectral kurtosis threshold. The selective trend estimation module performs selective trend estimation on the local spectral kurtosis of the peak region and the background region, respectively, and generates selective trend estimation curves. The detrended spectrum generation module subtracts the selective trend estimation curve from the frequency domain amplitude spectrum to generate a detrended spectrum. The detrending spectrum correction module performs non-negativity constraint correction on the detrending spectrum; The complex spectrum reconstruction and inverse transform module is used to perform symmetrical processing on the detrended amplitude spectrum and, combined with the phase spectrum obtained by the signal processing module, reconstruct the complex spectrum. The complex spectrum is then converted into a time-domain vibration signal by inverse fast Fourier transform (IFFT) and the DC component is removed. The envelope spectrum analysis module is used to perform Hilbert transform on time-domain vibration signals, extract the envelope signal, and generate an envelope spectrum with significantly enhanced frequency domain sideband features.
7. The rotating machinery frequency domain sideband extraction system based on selective trend estimation according to claim 6, characterized in that, The local spectral kurtosis calculation module includes: Calculate the local spectral kurtosis of the frequency domain amplitude spectrum based on the frequency domain amplitude spectrum. : ； Where W is the window width, the window W is calculated from the starting frequency of the frequency domain amplitude spectrum, and slides towards higher frequencies in one step, traversing the entire target frequency band; It is a window Amplitude data points within, It is a window The average value of the internal amplitude data.
8. A rotating machinery frequency domain sideband extraction system based on selective trend estimation according to claim 6, characterized in that, The selective trend estimation module includes: Different execution strengths are applied to the local spectral kurtosis of the peak region and the background region to perform selective trend estimation, generating selective trend estimation curves; Selective trend estimation is achieved using nonparametric trend filtering or low-order fitting methods, including moving median filtering, moving quantile filtering, Savitzky-Golay filtering, smoothing prior methods, and weighted low-order polynomial fitting.
9. A non-transitory computer-readable medium storing instructions, characterized in that, When the instruction is executed by the processor, the steps of the rotating machinery frequency domain sideband extraction method based on selective trend estimation according to any one of claims 1-5 are performed.
10. A computing device, comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the rotating machinery frequency domain sideband extraction method based on selective trend estimation as described in any one of claims 1-5.