A method and system for extracting early fault features of rolling bearings suitable for liquid ring pumps

Abstract

The application discloses a kind of early fault feature extraction method and system suitable for liquid ring pump rolling bearing, obtain the operation data of liquid ring pump rolling bearing, and operation data includes the instantaneous angular acceleration signal of bearing;Instantaneous angular acceleration signal is pretreated, adopts adaptive noise cancellation technology, estimates and removes periodic speed fluctuation noise in signal by time domain adaptive filter, obtains preprocessed signal;The preprocessed signal is processed to enhance fault feature, for minimum entropy deconvolution processing to multiple preprocessed signals, the spectral kurtosis of output signal is maximized by iterative optimization filter coefficient, to enhance the periodic transient impact response corresponding to bearing fault feature frequency, obtain enhanced signal;The optimal demodulation frequency band selection is carried out to enhanced signal, the node signal containing fault feature is extracted by optimal demodulation frequency band selection to enhanced signal;Spectral analysis is included to node signal, obtain fault feature frequency, complete early fault feature extraction.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a method and system for extracting early fault features of rolling bearings suitable for liquid ring pumps. Background Technology

[0002] Rolling bearings are among the most critical components in rotating machinery, accounting for approximately 30%-40% of motor failures. As a type of fluid machinery, liquid ring pumps operate in harsh environments with their rolling bearings, which not only bear mechanical loads but are also subject to fluid impact interference. Vibration signals are often drowned out by strong background noise, making it difficult to extract early fault characteristics.

[0003] Traditional bearing fault diagnosis is mostly based on vibration acceleration sensor analysis. However, this method has the following drawbacks: (1) The transmission path of vibration signal from the fault source to the sensor is long, which leads to a decrease in signal-to-noise ratio; (2) The vibration energy caused by early faults is concentrated in the high frequency band, which requires a high sampling rate of tens of kHz, resulting in high hardware costs; (3) The non-Gaussian noise generated by fluid impact in the liquid ring pump will interfere with the calculation of kurtosis index, causing the traditional frequency band selection method to fail.

[0004] Therefore, how to effectively extract early failure characteristics of rolling bearings in liquid ring pumps is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, a method for extracting early fault features of rolling bearings in liquid ring pumps is provided. By using instantaneous angular acceleration signals as input signals and performing aligned analysis, the method can capture minute speed fluctuations caused by bearing faults earlier, thus solving the problem that existing technologies cannot effectively detect early faults in rolling bearings of liquid ring pumps.

[0006] This application also provides a rolling bearing early fault feature extraction system suitable for liquid ring pumps.

[0007] To address the above problems, this invention provides a method for extracting early fault features of rolling bearings suitable for liquid ring pumps, the method comprising the following steps:

[0008] Obtain the operating data of the rolling bearing of the liquid ring pump, the operating data including the instantaneous angular acceleration signal of the bearing;

[0009] The instantaneous angular acceleration signal is preprocessed by using an adaptive noise cancellation technique. The periodic rotational speed fluctuation noise in the signal is estimated and removed by a time-domain adaptive filter to obtain the preprocessed signal.

[0010] The preprocessed signal is subjected to fault feature enhancement processing, which is used to perform minimum entropy deconvolution processing on multiple preprocessed signals, and maximize the spectral kurtosis of the output signal by iteratively optimizing the filter coefficients, so as to enhance the periodic transient impact response corresponding to the bearing fault characteristic frequency and obtain the enhanced signal.

[0011] The enhanced signal is subjected to optimal demodulation frequency band selection, and node signals containing fault characteristics are extracted.

[0012] The node signals are subjected to spectral analysis to obtain fault characteristic frequencies and complete the early fault feature extraction.

[0013] Furthermore, the instantaneous angular acceleration signal is acquired by a liquid annular angular accelerometer.

[0014] Furthermore, the instantaneous angular acceleration signal contains discrete frequency noise, and the preprocessing using adaptive noise cancellation technology specifically involves: The collected instantaneous angular acceleration signal and its delayed signal As a reference input adaptive filter, where... The delay step size; the adaptive filter adjusts the filter coefficients using the least mean square algorithm or the recursive least squares algorithm to make the filter output... Approaching The periodic speed fluctuation component in the; from Subtract The preprocessed signal is obtained. Wherein, the delay step size The selection satisfies and Relevant conditions for the periodic speed fluctuation component.

[0015] Furthermore, the minimum entropy deconvolution method maximizes the spectral kurtosis of the signal by iteratively optimizing the filter coefficients, specifically by iteratively inverse filtering. To make its output signal Maximize the spectral kurtosis, where kurtosis is defined as: ; The iterative formula is: ; Where R is the autocorrelation matrix of the input signal x, It is the cross-correlation vector of the output signal y in the k-th iteration; the iteration continues until the kurtosis value converges.

[0016] Furthermore, the optimal demodulation frequency band selection for the enhanced signal specifically employs an improved Autogram method. This improved Autogram method includes at least: decomposing the enhanced signal using maximum overlap discrete wavelet packet transform to obtain multiple frequency band nodes; calculating the autocorrelation of the squared envelope of each node signal to enhance periodic impact information and reduce the impact of fluid impact random noise on the evaluation index; and using the power spectral entropy of the autocorrelation function as the frequency band selection index, with the node having the minimum entropy value as the optimal demodulation frequency band. Through these improvements, frequency band selection no longer solely depends on kurtosis, avoiding distortion of the kurtosis index caused by non-Gaussian fluid impact in the liquid ring pump. Specifically, this includes:

[0017] S41 Maximum Overlap Discrete Wavelet Packet Transform Decomposition; The maximum overlap discrete wavelet packet transform method is used to enhance the signal. Perform J-level decomposition to obtain Each node signal; where the number of decomposition layers J is determined based on the sampling frequency and the frequency range of bearing fault characteristics;

[0018] S42 Calculate the autocorrelation function of the squared envelope of the signal at each node; for each node signal... (j=1,..., The analytic signal is obtained through the Hilbert transform:

[0019] ;

[0020] Where H[·] denotes the Hilbert transform; the squared envelope signal is obtained by taking the square of the modulus of the analytic signal. ;

[0021] Calculate the autocorrelation function of the squared envelope signal. :

[0022] ;

[0023] in, for The mean, For time delay, The signal length;

[0024] S43 Calculate the autocorrelation function The power spectral entropy, for the autocorrelation function Perform a fast Fourier transform to obtain the power spectral density. Normalize it to obtain the probability distribution :

[0025] ;

[0026] Power spectral entropy Defined as:

[0027] ;

[0028] Power spectral entropy It reflects the degree of concentration of signal energy in the frequency domain: if the signal contains obvious periodic impulses, the power spectrum of its autocorrelation function will have a peak at the characteristic frequency, the energy is concentrated, and the entropy value is small; if the signal is random noise or non-Gaussian noise generated by fluid impulses, its power spectrum is uniformly distributed and the entropy value is large.

[0029] S44 Select the optimal demodulation frequency band and compare the power spectral entropy of signals at all nodes. The node with the smallest entropy value is selected as the optimal demodulation frequency band.

[0030] Furthermore, a Hilbert transform is performed on the node signal selected by S4 to obtain an envelope signal, and a Fourier transform is performed on the envelope signal to obtain an envelope spectrum. Spectral peaks corresponding to the bearing fault characteristic frequencies are identified in the envelope spectrum to complete the extraction of fault features.

[0031] The present invention also provides a system for extracting early fault features of rolling bearings suitable for liquid ring pumps, the system comprising:

[0032] The signal acquisition module is used to acquire the instantaneous angular acceleration signal of the rolling bearing of the liquid ring pump;

[0033] The signal preprocessing module is communicatively connected to the signal acquisition module and is used to perform adaptive noise cancellation processing on the instantaneous angular acceleration signal to separate periodic rotational speed fluctuation noise and obtain the preprocessed signal.

[0034] The feature enhancement module is communicatively connected to the signal preprocessing module. It is used to perform minimum entropy deconvolution processing on multiple preprocessed signals, and maximize the spectral kurtosis of the output signal by iteratively optimizing the filter coefficients, so as to enhance the periodic transient impact response corresponding to the bearing fault characteristic frequency and obtain the enhanced signal.

[0035] The frequency band selection module is communicatively connected to the feature enhancement module and is used to perform optimal demodulation frequency band selection on the enhanced signal and extract node signals containing fault features.

[0036] The feature extraction module is communicatively connected to the frequency band selection module and is used to perform envelope spectrum analysis on the node, obtain fault feature frequencies, and complete early fault feature extraction.

[0037] Furthermore, the signal acquisition module includes a liquid annular angular accelerometer and a wireless transmission unit.

[0038] Furthermore, the frequency band selection module includes a maximum overlap discrete wavelet packet transform unit and a power spectral entropy calculation unit.

[0039] A computer-readable storage medium storing at least one computer instruction, which is loaded and executed by a processor to implement the fault feature extraction method.

[0040] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0041] 1. This application uses a liquid annular angular accelerometer to directly acquire the instantaneous angular acceleration signal of the bearing, shortening the transmission path from the fault source to the sensor and reducing signal attenuation; at the same time, it uses adaptive noise cancellation technology to separate the periodic speed fluctuation noise in the signal, thereby effectively improving the signal-to-noise ratio of the early fault signal;

[0042] 2. This application uses the minimum entropy deconvolution method on the preprocessed signal, and maximizes the spectral kurtosis of the output signal by iteratively optimizing the filter coefficients, thereby enhancing the periodic transient impact response corresponding to the bearing fault characteristic frequency and highlighting the weak fault characteristics.

[0043] 3. This application adopts an improved Autogram method, using power spectral entropy instead of kurtosis as the frequency band selection index. It utilizes the energy concentration characteristics of the power spectrum of the autocorrelation function of the periodic impact signal to suppress non-Gaussian noise interference generated by fluid impact in the liquid ring pump, thereby improving the accuracy of frequency band selection and robustness under complex working conditions.

[0044] 4. This application uses instantaneous angular acceleration signals for fault diagnosis. Angular acceleration signals are sensitive to speed fluctuations and fault impact information is concentrated in the low-to-mid frequency band. This allows for the effective extraction of early fault features under low sampling rate conditions, thereby reducing hardware costs. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method described in the embodiments of this application;

[0046] Figure 2 This is a block diagram of the system structure described in the embodiments of this application. Detailed Implementation

[0047] The technical solution of the present invention will be further described clearly and in detail below with reference to the embodiments and accompanying drawings.

[0048] Example 1

[0049] like Figure 1As shown in the embodiment of this application, a practical data processing flow applicable to the early fault feature extraction of rolling bearings in liquid ring pumps is provided. Taking a certain type of liquid ring vacuum pump as the object, the rated speed of the liquid ring vacuum pump is 1450 r / min, the test bearing is a deep groove ball bearing type 6205, with 9 rolling elements, a rolling element diameter of 7.94 mm, a pitch circle diameter of 39.04 mm, and a contact angle of 0°.

[0050] Step S1: Obtain the operating data of the rolling bearing of the liquid ring pump. Specifically, obtain the operating data of the rolling bearing of the liquid ring pump. Specifically, a liquid annular angular accelerometer is used to collect the instantaneous angular acceleration signal of the bearing. The liquid annular angular accelerometer is fixedly installed on the axial end face near the outer side of the bearing housing at the pump end, and is kept in close contact with the bearing housing by a rigid clamp to shorten the transmission path from the fault source to the sensor and reduce signal attenuation. The liquid annular angular accelerometer has a range of ±500 rad / s², a sensitivity of 100 mV / (rad / s²), and a frequency response range of 0.5 Hz to 5 kHz. The sampling frequency is set to 12.8 kHz, the sampling duration is 10 s, and the corresponding number of sampling points is N = 12800 × 10 = 128000 points. The collected instantaneous angular acceleration signal is denoted as... , as input signals for subsequent steps;

[0051] The instantaneous angular acceleration signal It contains the following ingredients:

[0052] (1) The periodic speed fluctuation component is caused by the impeller eccentricity of the liquid ring pump, uneven thickness of the liquid ring and hydrodynamic imbalance. The frequency is concentrated at 24.17 Hz and its integer multiples.

[0053] (2) The non-stationary transient impact component caused by early bearing failure is a periodic transient impact response caused by local defects in the inner ring, outer ring or rolling element of the bearing, and its frequency is related to the fault characteristic frequency.

[0054] (3) Fluid impact noise, which is broadband non-Gaussian random noise generated by the working fluid in the liquid ring pump impacting the blades and the flow channel wall;

[0055] (4) Background white noise, which is broadband Gaussian noise introduced by the sensor and acquisition system.

[0056] Among them, the periodic speed fluctuation component and fluid impact noise have strong energy and can overwhelm the weak transient impact caused by early faults. Therefore, noise suppression and feature enhancement are required through subsequent steps.

[0057] The time-domain waveform of the instantaneous angular acceleration signal acquired.

[0058] Based on the bearing geometry and rotational speed, the theoretical values ​​of each fault characteristic frequency are calculated as follows:

[0059] Frequency conversion: fr = 1450 / 60 = 24.17 Hz

[0060] Outer ring fault characteristic frequencies:

[0061] BPFO = (n / 2) × fr × [1 - (d / D) × cosα]

[0062] = (9 / 2) × 24.17 × [1 - (7.94 / 39.04) × cos0°]

[0063] = 86.63 Hz

[0064] Inner ring fault characteristic frequency:

[0065] BPFI = (n / 2) × fr × [1 + (d / D) × cosα]

[0066] = (9 / 2) × 24.17 × [1 + (7.94 / 39.04) × cos0°]

[0067] = 130.87 Hz

[0068] Rolling element failure characteristic frequencies:

[0069] BSF = (D / 2d) × fr × [1 - (d / D × cosα)²]

[0070] = (39.04 / 15.88) × 24.17 × [1 - (7.94 / 39.04)²]

[0071] = 56.95 Hz

[0072] Cage failure characteristic frequencies:

[0073] FTF = (fr / 2) × [1 - (d / D) × cosα]

[0074] = (24.17 / 2) × [1 - (7.94 / 39.04)]

[0075] = 9.63 Hz

[0076] Where n is the number of rolling elements, fr is the rotational frequency, d is the diameter of the rolling element, D is the pitch circle diameter, and α is the contact angle. These theoretical values ​​will serve as a reference for peak identification in subsequent envelope spectral analysis.

[0077] Step S2 involves performing adaptive noise cancellation preprocessing on the acquired instantaneous angular acceleration signal. Specifically, the delayed instantaneous angle...

[0078] The acceleration signal d(n) = x(n - Δ) is used as the reference input, and the delay step Δ is set to 529 points, which is approximately equal to 1 / 2 of the rotational frequency period (1 / (2×24.17) ≈ 0.0207 s, corresponding to 0.0207×12800 ≈ 265 sampling points). The actual value is taken to be close to an integer multiple of the rotational frequency period to ensure that the periodic rotational speed fluctuation component in x(n) and d(n) satisfies the strong correlation condition.

[0079] The adaptive filter uses the least mean square (LMS) algorithm, with a filter order of 128 and a step size parameter μ of 0.01. The filter coefficients are updated to make the filter output y(n) approximate the periodic rotational speed fluctuation component in x(n).

[0080] The periodic speed fluctuation component is then subtracted from the original signal to obtain the preprocessed signal e(n) = x(n) - y(n). This process is used to reduce the interference of periodic speed fluctuations during stable operation of the liquid ring pump, while retaining the non-stationary transient impact components caused by early faults.

[0081] Step S3 involves performing minimum entropy deconvolution feature enhancement on the preprocessed signal. Specifically, the initial length of the inverse filter is set to 64 points. Using the preprocessed signal as input, the inverse filter coefficients are iteratively calculated, and the kurtosis convergence of the output signal is used as the stopping condition. When the kurtosis change between two adjacent iterations is less than a preset threshold or the maximum number of iterations is reached, the enhanced signal is output. Through this processing, the periodic transient impacts generated by local defects in the bearing are compressed and enhanced, facilitating subsequent frequency band selection and envelope spectrum analysis.

[0082] Step S4 involves selecting the optimal demodulation frequency band for the enhanced signal using an improved Autogram method. Specifically, the enhanced signal is decomposed into four levels using Maximum Overlap Discrete Wavelet Packet Transform (MODWPT), with Daubechies8 (db8) chosen as the wavelet basis function, resulting in 2... 4 = 16 node signals, each node has a bandwidth of fs / 2 4= 12800 / 16 = 800 Hz. A Hilbert transform is performed on the signal at each node to obtain the analytic signal. The squared envelope signal is obtained by taking the modulus and squaring it. The autocorrelation function is calculated, and then a Fast Fourier Transform (1024-point window) is performed on the autocorrelation function. After normalization, the power spectral entropy is calculated. The power spectral entropy values ​​of all 16 nodes are compared, and the node with the smallest entropy value is selected as the optimal demodulation frequency band. The periodic impact energy is more concentrated in the frequency band corresponding to this node, which can more effectively reflect the early fault characteristics of the bearing.

[0083] Step S5: Perform Hilbert envelope analysis on the selected node signal and obtain the envelope spectrum by Fourier transform of the envelope signal. Calculate the fault characteristic frequencies of the inner ring, outer ring, rolling elements, and cage based on the bearing geometry and rotational speed. Match the spectral peak positions in the envelope spectrum with the aforementioned fault characteristic frequencies and their harmonics. When a spectral peak corresponds to a certain fault characteristic frequency and its harmonic components, output the corresponding fault characteristic frequency and fault type, thereby completing the early fault feature extraction of the rolling bearing.

[0084] As a comparative experimental example, the same set of instantaneous angular acceleration data was processed using traditional envelope spectrum analysis, the traditional Autogram frequency band selection method, and the improved Autogram method described in this application, respectively. The comparative results show that in traditional envelope spectrum analysis, fault characteristic frequencies are easily submerged by the fluid impact noise of the liquid ring pump; the traditional Autogram method, due to the influence of non-Gaussian impact noise on the kurtosis index, tends to select frequency bands that deviate from the main fault impact frequency band; the method in this application selects the optimal demodulation frequency band through squared envelope autocorrelation and power spectral entropy evaluation, resulting in clearer fault characteristic frequencies and their harmonic components, thus improving the stability of early fault feature extraction.

[0085] Example 2

[0086] This embodiment also provides a rolling bearing early fault feature extraction system suitable for liquid ring pumps, such as... Figure 2 As shown, it includes:

[0087] The signal acquisition module is used to acquire vibration signals from the bearing area of ​​the liquid ring pump using a liquid annular angular accelerometer, and to preprocess the vibration signals by removing the mean and trend to obtain a preprocessed signal. The liquid annular angular accelerometer is mounted on the outer wall of the bearing housing using a rigid clamp.

[0088] The signal preprocessing module is communicatively connected to the signal acquisition module and is used to perform adaptive noise cancellation processing on the preprocessed signal based on the LMS algorithm. By setting the delay step size Δ and the filter order N, the filter weight coefficients are iteratively updated to eliminate the periodic interference components in the signal and obtain the residual signal.

[0089] The feature enhancement module is communicatively connected to the signal preprocessing module. It is used to perform minimum entropy deconvolution processing on the residual signal, and maximize the kurtosis of the output signal by iteratively optimizing the FIR filter coefficients. When the kurtosis change between two adjacent iterations is less than a preset convergence threshold or the maximum number of iterations is reached, the iteration stops and the enhanced signal is output.

[0090] The frequency band selection module is communicatively connected to the feature enhancement module. It is used to perform 4-level wavelet packet decomposition on the enhanced signal to obtain 16 node signals, calculate the power spectral entropy of each node, select the node with the smallest power spectral entropy as the optimal demodulation frequency band, and output the corresponding node signal.

[0091] The feature extraction module is communicatively connected to the frequency band selection module. It is used to perform envelope spectrum analysis on the node signal, calculate the fault characteristic frequency based on the bearing geometric parameters, and compare the spectral peaks in the envelope spectrum with the fault characteristic frequency and its harmonics to identify the bearing fault type.

[0092] The above modules can be implemented using hardware circuits, embedded processors, or general-purpose computers in conjunction with corresponding software programs. Data is transferred between modules via a data bus or communication interface.

[0093] This embodiment also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of the method for extracting early fault features of rolling bearings for liquid ring pumps as described in Embodiment 1 above, namely, sequentially performing signal acquisition and preprocessing, adaptive noise cancellation, minimum entropy deconvolution, optimal demodulation frequency band selection based on improved Autogram, and envelope spectrum analysis and fault feature identification.

[0094] The computer-readable storage media include, but are not limited to: read-only memory (ROM), random access memory (RAM), magnetic disk, optical disk, flash memory, etc.

[0095] Finally, it should be pointed out that the above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting early fault features of rolling bearings in liquid ring pumps, characterized in that, The method includes the following steps: Obtain the operating data of the rolling bearing of the liquid ring pump, the operating data including the instantaneous angular acceleration signal of the bearing; The instantaneous angular acceleration signal is preprocessed by using an adaptive noise cancellation technique. The periodic rotational speed fluctuation noise in the signal is estimated and removed by a time-domain adaptive filter to obtain the preprocessed signal. The preprocessed signal is subjected to fault feature enhancement processing, which is used to perform minimum entropy deconvolution processing on multiple preprocessed signals, and maximize the spectral kurtosis of the output signal by iteratively optimizing the filter coefficients, so as to enhance the periodic transient impact response corresponding to the bearing fault characteristic frequency and obtain the enhanced signal. The enhanced signal is subjected to optimal demodulation frequency band selection, and node signals containing fault characteristics are extracted. The node signals are subjected to spectral analysis to obtain fault characteristic frequencies and complete the early fault feature extraction.

2. The method for extracting early fault features of rolling bearings in liquid ring pumps according to claim 1, characterized in that, The instantaneous angular acceleration signal is acquired by a liquid annular angular accelerometer.

3. The method for extracting early fault features of rolling bearings in liquid ring pumps according to claim 1, characterized in that, The instantaneous angular acceleration signal contains discrete frequency noise, and the preprocessing using adaptive noise cancellation technology specifically involves: The collected instantaneous angular acceleration signal and its delayed signal As a reference input adaptive filter, where, The delay step size; the adaptive filter adjusts the filter coefficients using the least mean square algorithm or the recursive least squares algorithm to make the filter output... Approaching The periodic speed fluctuation component in the; from Subtract The preprocessed signal is obtained. Wherein, the delay step size The selection satisfies and Relevant conditions for the periodic speed fluctuation component.

4. The method for extracting early fault features of rolling bearings in liquid ring pumps according to claim 1, characterized in that, The method employs minimum entropy deconvolution, which maximizes the spectral kurtosis of the signal by iteratively optimizing the filter coefficients. Specifically, this involves iteratively inverting the filter... to make its output signal Maximize the spectral kurtosis, where kurtosis is defined as: ； The iterative formula is: ; Where R is the autocorrelation matrix of the input signal x, It is the cross-correlation vector of the output signal y in the k-th iteration; the iteration continues until the kurtosis value converges.

5. The method for extracting early fault features of rolling bearings in liquid ring pumps according to claim 1, characterized in that, The optimal demodulation frequency band selection for the enhanced signal is specifically achieved by using an improved Autogram method, which includes the following improvements: using maximum overlap discrete wavelet packet transform to perform multi-level decomposition on the enhanced signal to obtain node signals in different frequency bands; Autocorrelation processing is performed on the squared envelope of each node's signal to highlight periodic impulse components and suppress the influence of random noise. The power spectral entropy of the autocorrelation function replaces the kurtosis index in the traditional autogram as the frequency band selection metric, determining the node signal with the minimum power spectral entropy as the optimal demodulation frequency band. Specifically, this includes: S41 Maximum Overlap Discrete Wavelet Packet Transform Decomposition; The maximum overlap discrete wavelet packet transform method is used to enhance the signal. Perform J-level decomposition to obtain Each node signal; where the number of decomposition layers J is determined based on the sampling frequency and the frequency range of bearing fault characteristics; S42 Calculate the autocorrelation function of the squared envelope of the signal at each node; for each node signal... (j=1,..., The analytic signal is obtained through the Hilbert transform: ； Where H[·] denotes the Hilbert transform; the squared envelope signal is obtained by taking the square of the modulus of the analytic signal. ; Calculate the autocorrelation function of the squared envelope signal. : ； in, for The mean, For time delay, The signal length; S43 Calculate the autocorrelation function The power spectral entropy, for the autocorrelation function Perform a fast Fourier transform to obtain the power spectral density. Normalize it to obtain the probability distribution : ； Power spectral entropy Defined as: ； Power spectral entropy It reflects the degree of concentration of signal energy in the frequency domain: if the signal contains obvious periodic impulses, the power spectrum of its autocorrelation function will have a peak at the characteristic frequency, the energy is concentrated, and the entropy value is small; if the signal is random noise or non-Gaussian noise generated by fluid impulses, its power spectrum is uniformly distributed and the entropy value is large. S44 Select the optimal demodulation frequency band and compare the power spectral entropy of signals at all nodes. The node with the smallest entropy value is selected as the optimal demodulation frequency band.

6. The method for extracting early fault features of rolling bearings in a liquid ring pump according to claim 5, characterized in that, The Hilbert transform of the node signal selected by S4 is performed to obtain the envelope signal, and the Fourier transform of the envelope signal is performed to obtain the envelope spectrum. The spectral peaks corresponding to the bearing fault characteristic frequencies are identified in the envelope spectrum to complete the extraction of fault features.

7. A rolling bearing early fault feature extraction system suitable for liquid ring pumps, characterized in that, The system includes: The signal acquisition module is used to acquire the instantaneous angular acceleration signal of the rolling bearing of the liquid ring pump; The signal preprocessing module is communicatively connected to the signal acquisition module and is used to perform adaptive noise cancellation processing on the instantaneous angular acceleration signal to separate periodic rotational speed fluctuation noise and obtain the preprocessed signal. The feature enhancement module is communicatively connected to the signal preprocessing module. It is used to perform minimum entropy deconvolution processing on multiple preprocessed signals, and maximize the spectral kurtosis of the output signal by iteratively optimizing the filter coefficients, so as to enhance the periodic transient impact response corresponding to the bearing fault characteristic frequency and obtain the enhanced signal. The frequency band selection module is communicatively connected to the feature enhancement module and is used to perform optimal demodulation frequency band selection on the enhanced signal and extract node signals containing fault features. The feature extraction module is communicatively connected to the frequency band selection module and is used to perform envelope spectrum analysis on the node, obtain fault feature frequencies, and complete early fault feature extraction.

8. The system according to claim 7, characterized in that, The signal acquisition module includes a liquid annular angular accelerometer and a wireless transmission unit.

9. The system according to claim 7, characterized in that, The frequency band selection module includes a maximum overlap discrete wavelet packet transform unit and a power spectral entropy calculation unit.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one computer instruction, which is loaded and executed by a processor to implement the fault feature extraction method according to any one of claims 1-6.

Images