Phase modifier bearing fault diagnosis method and system, and medium

Abstract

The invention discloses a phase modifier bearing fault diagnosis method, which belongs to the technical field of power equipment fault diagnosis, and comprises the following steps of: analyzing independent components, performing blind source separation on a received phase modifier bearing vibration signal, and extracting three types of independent source signals of impact, abrasion and noise; the gradient driving window length is self-adaptive, and the window length of short-time Fourier transform is dynamically adjusted based on the instantaneous frequency gradient of the independent source signal; wavelet packet frequency band energy screening: performing wavelet packet decomposition on the signal after window length adaptive processing, screening a fault characteristic frequency band based on an energy contribution rate, and reconstructing the signal; and enhancing stochastic resonance, and inputting the reconstructed signal into a stochastic resonance system. According to the method, independent component analysis, gradient driving window length self-adaption, wavelet packet frequency band energy screening and stochastic resonance enhanced fourth-order diagnosis chain are constructed, so that multi-stage cooperative processing of phase modifier bearing faults is realized, and the technical problems of time-frequency resolution contradiction, insufficient feature decoupling and weak generalization ability are effectively solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power equipment fault diagnosis, and particularly relates to a phase modifier bearing fault diagnosis method and system and a medium. BACKGROUND

[0002] With the expansion of the scale of UHVDC power transmission and new energy grid connection, as the core reactive power compensation equipment of the power grid, the bearing reliability of the phase modifier directly relates to the stability of the power system. The bearing of the phase modifier bears high speed (0-3000 rpm), variable working conditions (speed fluctuation ±20%), compound faults (oil film failure and mechanical damage concurrent rate exceeding 60%) and the bearing reliability of its auxiliary equipment directly affects the stability of the system. The early fault characteristics are weak (such as the impact energy of the inner ring crack is only 0.02% of the normal vibration), and the traditional threshold alarm has a 5-8 hour lag, which cannot meet the safety requirements of the power grid. Especially in the power grid structure of "strong direct and weak alternating", bearing failure may cause the sudden shutdown of the phase modifier, causing the collapse of the power grid voltage, so it is urgent to develop a real-time diagnosis method suitable for variable speed and anti-noise interference.

[0003] The current mainstream diagnosis methods can be divided into three categories:

[0004] Signal analysis method: such as random subspace identification (SSI) combined with multi-kernel support vector machine (MSVM), which realizes fault recognition through vibration signal feature extraction and classification, but the fixed window STFT has a missing detection rate of more than 40% for high-frequency impact characteristics when the speed suddenly changes, and is sensitive to noise.

[0005] Physical model method: such as ultrasonic oil film thickness monitoring, which calculates the oil film thickness through the resonance frequency, but a single parameter cannot distinguish between oil film failure and mechanical damage, and needs to rely on multiple sensors such as temperature and pressure.

[0006] Data-driven method: such as random forest (RF) and deep learning model (CNN / LSTM), which uses vibration spectrum to classify fault types, but the accuracy rate drops by 25% when the noise is greater than 10dB, and more than 2000 labeled samples are needed for training, which has high engineering cost.

[0007] Retrieved literature: Publication (announcement) No. CN108444709B, a rolling bearing fault diagnosis method combining VMD and FastICA, the steps are: using existing data acquisition equipment to collect the original vibration signal of the rolling bearing; the collected original vibration signal of the rolling bearing is decomposed by VMD; after VMD, the original vibration signal is decomposed into k modal components, and continuous 3 modal components are combined as a sequence to perform FastICA analysis to obtain a reconstructed fault signal.

[0008] Publication (announcement) number: CN112082793A, a rotating machinery coupling fault diagnosis method based on SCA and FastICA, can effectively filter out pulse noise and white noise, reduce noise and improve signal-to-noise ratio, and realize effective extraction of fault characteristic signals.

[0009] Publication (announcement) number: CN110146291A, a rolling bearing fault feature extraction method based on CEEMD and FastICA, belongs to the field of fault diagnosis technology and signal processing analysis technology. The method first decomposes the vibration signal into several IMF components of different frequencies by using the CEEMD algorithm, then selects the corresponding IMF component according to the kurtosis criterion to obtain the observation signal, and the remaining IMF component is reconstructed to obtain the virtual noise channel signal; the observation signal and the virtual noise channel signal are processed by using the FastICA algorithm to realize the demixing and denoising.

[0010] The search literature (such as CN108444709B, CN112082793A, CN110146291A) uses "two-order" or "three-order" combination (such as VMD+FastICA, CEEMD+FastICA), which is mainly to improve the signal decomposition and blind source separation. The defects of the prior art have not been solved. First, the time-frequency resolution is contradictory, that is, the fixed window short-time Fourier transform cannot consider the feature extraction of high-frequency impact (which requires a short window) and low-frequency wear (which requires a long window); second, the feature decoupling is insufficient, and a single signal source is difficult to separate the coupling effect of complex faults; third, the generalization ability is weak, and the traditional threshold method relies on artificial experience, while the data-driven model is sensitive to sample distribution deviation. SUMMARY

[0011] The purpose of the present application is to overcome the shortcomings of the prior art and provide a phase-modulator bearing fault diagnosis method, which solves the technical problems in the above background art.

[0012] The purpose of the application is achieved in that the method for diagnosing the phase modulator bearing fault comprises the following steps: an independent component analysis step, blind source separation is performed on the received phase modulator bearing vibration signal to extract three types of independent source signals of impact, wear and noise; a gradient-driven window length self-adaptive step, the window length of short-time Fourier transform is dynamically adjusted based on the instantaneous frequency gradient of the independent source signal; a wavelet packet frequency band energy screening step, the signal processed by the window length self-adaptive step is decomposed by wavelet packet, and the fault characteristic frequency band is screened based on the energy contribution rate and the signal is reconstructed; a stochastic resonance enhancement step, the reconstructed signal is input into the stochastic resonance system, and the weak fault characteristics are enhanced by using the nonlinear effect. The gradient-driven window length self-adaptive can dynamically adjust the analysis window length according to the instantaneous frequency gradient, and improve the frequency resolution of the low frequency band while maintaining good time resolution in the high frequency band. The gradient window link adaptively selects the STFT window length (high frequency short window, low frequency long window) by real-time calculation of the signal instantaneous frequency gradient, effectively solving the contradiction between time and frequency resolution. The method of the application realizes the multi-level cooperative processing of the phase modulator bearing fault by constructing the four-order diagnosis chain of independent component analysis, gradient-driven window length self-adaptation, wavelet packet frequency band energy screening and stochastic resonance enhancement, effectively solving the technical problems of time-frequency resolution contradiction, feature decoupling deficiency and weak generalization ability.

[0013] Further, the independent component analysis step comprises: performing mean removal processing on the observation signal, calculating the covariance matrix of the mean-removed signal, performing eigenvalue decomposition on the covariance matrix to obtain the eigenvector and the eigenvalue diagonal matrix; performing whitening processing on the signal by using the eigenvector and the eigenvalue diagonal matrix, so that the covariance matrix of the whitened signal is a unit matrix; solving the signal separation vector by an iterative method of approximately calculating the negative entropy and maximizing it, so as to complete the separation of the independent source signal. Through matrix operation and optimization algorithm, a set of separation vectors are found, so that the statistical independence of the output signal (source signal) is maximized. The FastICA algorithm has fast convergence speed and stable calculation. The specific process ensures the effectiveness and efficiency of blind source separation, and provides high-quality input signal for the subsequent steps.

[0014] Further, the gradient-driven window length adaptive step includes: obtaining the instantaneous phase of the signal by Hilbert transform, and then obtaining the instantaneous frequency by derivation of the instantaneous phase; performing differential operation on the instantaneous frequency to obtain the rate of change of the instantaneous frequency; comparing the rate of change with a dynamic threshold set according to the maximum rate of change of the current signal; when the rate of change is greater than the dynamic threshold, a shorter window length is selected for time-frequency analysis to capture high-frequency impact characteristics; when the rate of change is less than or equal to the dynamic threshold, a longer window length is selected for time-frequency analysis to refine low-frequency wear characteristics. The dynamic threshold is set to thirty percent of the maximum rate of change of the instantaneous frequency of the current signal. The shorter window length is set to one hundred and twenty-eight data points, and the longer window length is set to five hundred and twelve data points.

[0015] The degree of change of the signal is calculated in real time through signal processing (Hilbert transform, derivation, difference), and the most suitable analysis window length is automatically selected according to the preset and optimized rules (threshold, window length). The adaptive and intelligent adjustment of the window length is realized, and the dependence on artificial experience is eliminated. The threshold of 30% is an optimal value verified by experiments, and the best balance can be achieved under most working conditions.

[0016] Further, the wavelet packet frequency band energy screening step includes: using Db6 wavelet basis function to perform five-layer complete decomposition on the signal to obtain thirty-two uniformly distributed sub-bands; calculating the energy of each sub-band signal and the percentage of the total energy; screening out sub-bands with energy percentage exceeding five percent; reconstructing the screened key sub-band signals to obtain signals with energy concentrated in fault characteristics. Db6 wavelet achieves a good balance between smoothness and compactness, which is suitable for analyzing mechanical vibration signals. Five-layer decomposition can divide the signal into fine enough frequency bands to capture fault characteristics, and the energy threshold of 5% can effectively screen out key frequency bands most related to faults and exclude interference. The wavelet packet decomposes the signal into different frequency bands in a tree-like manner, and through calculation of energy distribution and setting of a threshold, automatic identification of fault characteristic frequency bands and signal purification are realized.

[0017] Further, in the stochastic resonance enhancement step, the dynamic behavior of the nonlinear bistable potential well system is described by a nonlinear differential equation containing a linear restoring force, a nonlinear restoring force, an input signal, and a noise term. The particle swarm optimization algorithm is used to adaptively optimize the structural parameters of the nonlinear bistable potential well system, and the signal-to-noise ratio of the system output signal is maximized as the optimization goal. The signal is input into the bistable potential well model, and the optimal parameters (such as the potential barrier height) of the system are automatically searched by the intelligent optimization algorithm (particle swarm optimization algorithm), so that the noise energy is maximized to signal energy, thereby obtaining significantly enhanced periodic fault features at the output end. The particle swarm optimization algorithm is used for adaptive parameter optimization, which overcomes the difficulty of manual parameter adjustment, so that the system can automatically achieve the best enhancement effect under different working conditions, greatly improving the applicability and effect of the method.

[0018] A phase modulator bearing fault diagnosis system is used to implement the above method, comprising: an independent component analysis module for blind source separation of the received phase modulator bearing vibration signal to extract three types of independent source signals of impact, wear and noise; a gradient-driven window length adaptive module for dynamically adjusting the window length of short-time Fourier transform based on the instantaneous frequency gradient of the independent source signal; a wavelet packet frequency band energy screening module for wavelet packet decomposition of the signal processed by the window length adaptive module, screening of the fault feature frequency band based on the energy contribution rate and reconstruction of the signal; a stochastic resonance enhancement module for inputting the reconstructed signal into a stochastic resonance system to enhance the weak fault features by using nonlinear effects.

[0019] A computer readable storage medium storing a computer program, which is executed by a processor to implement the steps of the above method.

[0020] The method of the present application realizes multi-level collaborative processing of phase modulator bearing faults by constructing an independent component analysis, gradient-driven window length adaptation, wavelet packet frequency band energy screening, and stochastic resonance enhancement four-order diagnosis chain, effectively solving the technical problems of time-frequency resolution contradiction, feature decoupling deficiency, and weak generalization ability. Through matrix operation and optimization algorithm, a set of separation vectors are found to maximize the statistical independence of the output signal (source signal). The FastICA algorithm has fast convergence speed and stable calculation. The specific process ensures the effectiveness and efficiency of blind source separation, providing high-quality input signals for subsequent steps. The particle swarm optimization algorithm is used for adaptive parameter optimization, which overcomes the difficulty of manual parameter adjustment, so that the system can automatically achieve the best enhancement effect under different working conditions, greatly improving the applicability and effect of the method. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a method flowchart of the present application. DETAILED DESCRIPTION

[0022] The present application will be further described in detail below with reference to the accompanying drawings, it should be pointed out that all orientation words do not limit the present application, but only for more clearly illustrating and explaining the present application.

[0023] As shown in Figure 1 The application discloses a phase modifier bearing fault diagnosis method, and proposes a "FastICA-gradient window-wavelet packet-stochastic resonance" fourth-order diagnosis chain, which comprises the following steps: an independent component analysis step, which performs blind source separation on received phase modifier bearing vibration signals, and extracts three types of independent source signals, namely impact, wear and noise; a gradient-driven window length self-adaptive step, which dynamically adjusts the window length of short-time Fourier transform based on the instantaneous frequency gradient of the independent source signals; a wavelet packet frequency band energy screening step, which performs wavelet packet decomposition on the signals processed through the window length self-adaptive step, screens a fault characteristic frequency band based on an energy contribution rate, and reconstructs a signal; and a stochastic resonance enhancement step, which inputs the reconstructed signal into a stochastic resonance system, and enhances weak fault characteristics by using nonlinear effects. The gradient-driven window length self-adaptive step can dynamically adjust the analysis window length according to the instantaneous frequency gradient, thereby improving the frequency resolution of the low-frequency band while maintaining good time resolution in the high-frequency band. The gradient window link adaptively selects the STFT window length (high-frequency short window and low-frequency long window) by calculating the instantaneous frequency gradient of the signal in real time, and effectively solves the contradiction between time-frequency resolution. The method of the present application realizes multi-level collaborative processing of phase modifier bearing faults by constructing an independent component analysis, gradient-driven window length self-adaptive, wavelet packet frequency band energy screening, and stochastic resonance enhancement fourth-order diagnosis chain, and effectively solves the technical problems of contradiction between time-frequency resolution, insufficient characteristic decoupling, and weak generalization ability.

[0024] The independent component analysis step comprises the following steps: performing mean removal processing on an observation signal, calculating the covariance matrix of the mean-removed signal, performing eigenvalue decomposition on the covariance matrix to obtain an eigenvector and an eigenvalue diagonal matrix; performing whitening processing on the signal by using the eigenvector and the eigenvalue diagonal matrix, so that the covariance matrix of the whitened signal is a unit matrix; and solving a signal separation vector by using an iterative method of approximately calculating negative entropy and maximizing the negative entropy, so as to complete separation of the independent source signals. By using matrix operation and an optimization algorithm, a group of separation vectors are found, so that the statistical independence of the output signal (source signal) is maximized. The FastICA algorithm has fast convergence speed and stable calculation. The specific process ensures the effectiveness and efficiency of blind source separation, and provides high-quality input signals for subsequent steps.

[0025] The gradient-driven window length adaptive step includes: obtaining the instantaneous phase of the signal through Hilbert transform, and then differentiating the instantaneous phase to obtain the instantaneous frequency; performing a difference operation on the instantaneous frequency to obtain the rate of change of the instantaneous frequency; comparing the rate of change with a dynamic threshold set according to the maximum rate of change of the current signal; when the rate of change is greater than the dynamic threshold, selecting a shorter window length for time-frequency analysis to capture high-frequency impact characteristics; when the rate of change is less than or equal to the dynamic threshold, selecting a longer window length for time-frequency analysis to refine low-frequency wear characteristics. The dynamic threshold is set to 30% of the maximum instantaneous frequency change rate of the current signal. The shorter window length is set to 128 data points, and the longer window length is set to 512 data points.

[0026] The system calculates the magnitude of signal changes in real time through signal processing (Hilbert transform, differentiation, and differencing), and automatically selects the most suitable analysis window length based on preset and optimized rules (threshold, window length). This achieves adaptive and intelligent adjustment of the window length, eliminating reliance on manual experience. The 30% threshold is an experimentally verified optimal value that achieves the best balance under most operating conditions.

[0027] The wavelet packet frequency band energy screening step includes: performing a five-level complete decomposition of the signal using the Db6 wavelet basis function to obtain thirty-two uniformly distributed sub-frequency bands; calculating the energy of each sub-frequency band signal and its percentage of the total energy; screening out sub-frequency bands with an energy percentage exceeding 5%; and reconstructing the selected key sub-frequency band signals to obtain signals with energy concentrated in fault characteristics. The Db6 wavelet achieves a good balance between smoothness and compact support, making it suitable for analyzing mechanical vibration signals. The five-level decomposition can divide the signal into sufficiently fine frequency bands to capture fault characteristics, and the 5% energy threshold can effectively screen out the key frequency bands most relevant to the fault, eliminating interference. The wavelet packet decomposes the signal into different frequency bands in a tree-like manner, and by calculating the energy distribution and setting thresholds, automatic identification and signal purification of fault characteristic frequency bands are achieved.

[0028] In the stochastic resonance enhancement step, the dynamic behavior of the nonlinear bistable potential trap system is described by a nonlinear differential equation containing linear restoring force, nonlinear restoring force, input signal, and noise terms. A particle swarm optimization algorithm is used to adaptively optimize the structural parameters of the nonlinear bistable potential trap system, with the goal of maximizing the signal-to-noise ratio of the system output signal. The signal is input into the bistable potential trap model, and the intelligent optimization algorithm (particle swarm optimization) automatically searches for the optimal parameters of the system (such as the barrier height), maximizing the conversion of noise energy into signal energy, thereby achieving a significantly enhanced periodic fault characteristic at the output. Using particle swarm optimization for adaptive parameter optimization overcomes the difficulties of manual parameter tuning, enabling the system to automatically achieve the best enhancement effect under different operating conditions, greatly improving the applicability and effectiveness of the method.

[0029] The calculation process of the method of this invention is as follows:

[0030] FastICA dominant component extraction:

[0031] Let the received observed signal X(t) be a linear mixture of the source signals S(t) = [s1(t), s2(t), ..., sn(t)]. Then the digital signal model is: X(t) = AS(t). When there is only one observed signal, A is a 1×n row vector, and X(t) is a one-dimensional row vector. This can also be extended to multi-channel signals. The signal needs to satisfy the following properties: statistical independence, non-Gaussianity, linear instantaneous mixing, and no linear convolutional mixing.

[0032] First, the matrix needs to be whitened. The first step is to obtain the mean-removed result X. centered E represents the mean operator, which then calculates the covariance matrix C. T This represents the transpose operation. Then, the eigenvectors E and eigenvalue diagonal matrix D of the covariance matrix are calculated, finally yielding the whitened matrix Z. The whitened matrix E[zz]... T ] = I.

[0033] Xcentered = X - EX], (1)

[0034]

[0035] C = EDE T (3)

[0036] Z = D -1 / 2 E T X centered (4)

[0037] Theoretically, the greater the negative entropy, the more ordered the signal, and the more likely it is to be an independent source. Therefore, by maximizing negative entropy, we can separate the three types of source signals: impact, wear, and noise, and avoid the mode mixing problem of EMD-type methods.

[0038] The following section calculates the negative entropy of the signal. Formulas 5 and 6 are the formulas for calculating entropy and negative entropy, respectively.

[0039] H(y)=-∫p(y)logp(y)dy, (5)

[0040] J(y)=H(y gauss )-H(y), (6)

[0041] However, calculating negative entropy is complex, so an approximation is needed. The G function is used as an approximation method. Equation 7 shows the method for calculating negative entropy using the G function, and Equation 8 shows the function chosen in this method for approximating negative entropy: J(y)∝[E{G(y)}-E{G(v)}]. 2 ,v~N(0,1),(7)

[0042] G(y) = logcosh(y), (8)

[0043] The subsequent step is to iteratively derive the separating vector w, as shown in the following formula, where g and g' represent the first and second derivatives of G, respectively:

[0044]

[0045] H(w)=E{g′(w T z)zz T}, (11)

[0046] Since z is the whitened matrix, H(w) ≈ E{g′(w)} T z)}I≡βI

[0047] Gradient-driven adaptive window length:

[0048] Gradient-driven adaptive window length is the core component for achieving high-resolution time-frequency analysis. Its core principle is to dynamically adjust the window length of the short-time Fourier transform (STFT) based on the instantaneous frequency gradient of the signal in order to resolve the time-frequency resolution contradiction caused by a fixed window length.

[0049] After the dominant component of the signal is extracted using FastICA, its instantaneous frequency gradient needs to be calculated to characterize the rate of frequency change. Let the dominant component be s(t), and its instantaneous phase φ(t) be obtained through Hilbert transform, where... This is the Hilbert transform operator.

[0050]

[0051] The instantaneous frequency f(t) is defined as the derivative of the phase, and the instantaneous frequency gradient is... This gradient, calculated using first-order difference or numerical differentiation, quantifies the local rate of change of the signal frequency: the gradient is large in the high-frequency region (such as bearing impact pulses) and small in the low-frequency region (such as wear vibrations).

[0052]

[0053] A window length adjustment strategy is designed based on the instantaneous frequency gradient to balance time and frequency resolution.

[0054]

[0055] High frequency region ( Large: A short window (128 points) is used to improve the time resolution to capture transient impacts (such as bearing crack pulses), retain the steepness of the impact pulse leading edge, and avoid leading edge distortion (error <0.1μs).

[0056] Low frequency region ( Small: Employs a long window (512 points) to improve frequency resolution and refine wear harmonic characteristics (accuracy up to 0.1Hz), reducing characteristic frequency ambiguity.

[0057] Gradient threshold selection The reason is that this threshold covers most frequency conversion scenarios (such as ±20% fluctuation in the bearing speed of a synchronous condenser). When the gradient exceeds the threshold, it indicates that the signal is in a high-frequency transient state, requiring a short window to suppress spectral leakage; conversely, a long window is needed to enhance frequency concentration. In addition, in engineering, it is also necessary to avoid boundary effects, so the signal needs to be preprocessed with mirror extension to ensure that the window function remains intact at the signal boundaries.

[0058] Wavelet packet band energy screening:

[0059] Wavelet Packet Transform (WPT) leverages its multi-resolution analysis capabilities to perform deeper, more complete sub-band division of the signal after gradient window adaptive time-frequency analysis, thereby achieving fine energy focusing. Specifically, it employs the db6 wavelet basis function, which has tight support and moderate regularity, to perform a 5-level complete decomposition of the dynamic time-frequency spectrum, generating 32 mutually orthogonal sub-band nodes that uniformly cover the analysis frequency range. The energy contribution rate E of each sub-band node is then calculated. k

[0060]

[0061] It enables accurate identification and extraction of sensitive frequency bands containing key fault information. During this process, an empirical threshold of energy contribution rate greater than 5% is set, thereby retaining key frequency bands such as the 2.5-3.2kHz band related to inner ring crack features and the 0.1-0.5kHz band related to oil film failure, and performing signal reconstruction. This effectively suppresses interference from non-stationary noise, resulting in a high concentration of signal energy near the fault characteristic frequency. Due to WPT's adaptive matching capability for non-stationary signals, its energy focusing accuracy is improved compared to traditional linear noise reduction methods such as singular value decomposition (SVD).

[0062] Stochastic resonance enhancement:

[0063] The signal component s(t), after wavelet packet filtering and reconstruction, is input into a stochastic resonance (SR) system. Stochastic resonance is a physical phenomenon and method that utilizes the synergistic effect of nonlinear systems and noise to enhance weak periodic signals. Its core is a bistable system, typically described by the Langevin equations:

[0064]

[0065] Where a and b are the structural parameters of the bistable potential well, s(t) is the noisy fault signal to be enhanced, and n(t) represents the noise. Under specific parameters, the weak periodic signal s(t) can overcome the potential barrier with the help of noise energy, thus being significantly amplified at the output. To achieve the optimal resonance effect, the particle swarm optimization (PSO) algorithm is used to adaptively optimize the key system parameters a and b to adapt to different input signals and noise levels. PSO takes maximizing the system output signal-to-noise ratio as the objective function, as shown in the following formula.

[0066]

[0067] The drive system reaches a resonant state, thereby converting some noise energy into signal energy and significantly amplifying the potential periodic fault impact component. Even under harsh conditions with an input signal-to-noise ratio as low as -10dB, this strategy can still enhance the amplitude of the target's impact characteristics, fundamentally breaking through the bottleneck of early fault detection under strong background noise and providing a high signal-to-noise ratio input signal for subsequent fault feature extraction and classification.

[0068] A synchronous condenser bearing fault diagnosis system, used to implement the above-mentioned method, includes: an independent component analysis module for performing blind source separation on the received synchronous condenser bearing vibration signal and extracting three independent source signals: impact, wear, and noise; a gradient-driven window length adaptive module for dynamically adjusting the window length of the short-time Fourier transform based on the instantaneous frequency gradient of the independent source signals; a wavelet packet frequency band energy screening module for performing wavelet packet decomposition on the signal after window length adaptive processing, screening fault feature frequency bands based on energy contribution rate, and reconstructing the signal; and a stochastic resonance enhancement module for inputting the reconstructed signal into a stochastic resonance system to enhance weak fault features using nonlinear effects.

[0069] A computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the above-described method.

[0070] This invention's diagnostic chain achieves independent component separation of vibration signals through FastICA, effectively decoupling multi-source interference. The gradient window adaptively adjusts its length based on instantaneous frequency changes, maintaining good time resolution in the high-frequency band while improving frequency resolution in the low-frequency band. Wavelet packets further refine the frequency band division, focusing on fault-related sub-bands and suppressing irrelevant components. Stochastic resonance utilizes nonlinear effects to enhance the output of weak fault features from strong noise. Fourth-order collaborative step-by-step signal purification significantly improves the detectability of early weak faults. This multi-level signal processing mechanism collaboratively solves the technical challenges in diagnosing camera bearing faults. Adaptive window length control dynamically resolves the time-frequency resolution contradiction. The method of this invention is designed for scenarios where instantaneous frequency gradient is used as the standard; in other scenarios, other indicators and thresholds are needed for better results.

[0071] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing a fault of a phase modulator bearing, characterized by, The method comprises the following steps: an independent component analysis step for performing blind source separation on the received modulator bearing vibration signal to extract three types of independent source signals of impact, wear and noise; a gradient-driven window length self-adaption step for dynamically adjusting the window length of short-time Fourier transform based on the instantaneous frequency gradient of the independent source signals; a wavelet packet frequency band energy screening step for performing wavelet packet decomposition on the signal processed by the window length self-adaption, screening a fault characteristic frequency band based on energy contribution rate and reconstructing the signal; a stochastic resonance enhancement step for inputting the reconstructed signal into a stochastic resonance system to enhance the weak fault characteristics by using nonlinear effects.

2. The method of claim 1, wherein, The independent component analysis step comprises: performing mean removal processing on the observation signal, calculating the covariance matrix of the signal after mean removal, performing eigenvalue decomposition on the covariance matrix to obtain an eigenvector and an eigenvalue diagonal matrix; performing whitening processing on the signal by using the eigenvector and the eigenvalue diagonal matrix, so that the covariance matrix of the signal after whitening is a unit matrix; solving a signal separation vector by an iterative method of approximately calculating negative entropy and maximizing it, so as to complete separation of the independent source signals.

3. The method of claim 1, wherein, The gradient-driven window length self-adaption step comprises: obtaining the instantaneous phase of the signal by Hilbert transform, and then obtaining the instantaneous frequency by derivation of the instantaneous phase; performing difference operation on the instantaneous frequency to obtain the rate of change of the instantaneous frequency; and comparing the rate of change with a dynamic threshold value set according to the maximum rate of change of the current signal; when the rate of change is greater than the dynamic threshold value, a shorter window length is selected for time-frequency analysis to capture high-frequency impact characteristics; and when the rate of change is less than or equal to the dynamic threshold value, a longer window length is selected for time-frequency analysis to refine low-frequency wear characteristics.

4. The method of claim 3, wherein, The dynamic threshold value is set to thirty percent of the maximum value of the instantaneous frequency rate of change of the current signal.

5. The method of claim 4, wherein, The shorter window length is set to one hundred and twenty-eight data points, and the longer window length is set to five hundred and twelve data points.

6. The method of claim 1, wherein, The wavelet packet frequency band energy screening step comprises: performing five-layer complete decomposition on the signal by using a Db6 wavelet basis function to obtain thirty-two uniformly distributed sub-bands; calculating the energy of each sub-band signal and the percentage of the energy in the total energy; screening the sub-bands with an energy percentage of more than five percent; and reconstructing the key sub-band signals screened to obtain a signal with energy concentrated in fault characteristics.

7. The method of claim 1, wherein, In the stochastic resonance enhancement step, the dynamic behavior of a nonlinear bistable potential well system is described by a nonlinear differential equation containing a linear restoring force, a nonlinear restoring force, an input signal and a noise term.

8. The method of claim 7, wherein, A particle swarm optimization algorithm is used to adaptively optimize the structural parameters of the nonlinear bistable potential well system, and the maximum signal-to-noise ratio of the system output signal is used as the optimization objective.

9. A phase modulator bearing fault diagnosis system for implementing the method of any one of claims 1-8, characterized by, The method comprises: an independent component analysis module for performing blind source separation on the received modulator bearing vibration signal to extract three types of independent source signals of impact, wear and noise; a gradient-driven window length self-adaption module for dynamically adjusting the window length of short-time Fourier transform based on the instantaneous frequency gradient of the independent source signals; A wavelet packet frequency band energy screening module is configured to perform wavelet packet decomposition on the signal after the window length adaptive processing, screen the fault feature frequency band based on the energy contribution rate, and reconstruct the signal; A stochastic resonance enhancement module is configured to input the reconstructed signal into a stochastic resonance system, and enhance the weak fault feature by using nonlinear effect.

10. A computer readable storage medium storing a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the method of any one of claims 1-8.

Images