Fault identification method based on characteristic value sidelobe

Abstract

The invention relates to a fault identification method based on a characteristic value side lobe, which comprises the following steps of: 1, identifying a side lobe characteristic which is close to a characteristic frequency and is directly related to a secondary vibration fault by analyzing the spectrum distribution of a manikin model; 2, feature value sidelobe extraction: extracting feature value sidelobe information from the preprocessed signal by adopting a wavelet transform and Hilbert transform signal processing method; step 3, characteristic value side lobe analysis: carrying out quantitative analysis on the extracted characteristic value side lobe information; and step 4, fault identification: according to the analysis result of the characteristic value side lobe, performing fault identification on the equipment in combination with a fault characteristic library and an intelligent algorithm. According to the method, the fault identification precision can be improved, and the identification precision can be remarkably improved and the misjudgment rate can be reduced by performing fault identification by using the fine change of the characteristic value sidelobe; and the anti-interference capability is enhanced: the characteristic value sidelobe analysis has a relatively strong inhibition effect on noise and interference, and the method can stably work in a complex environment.

Description

Technical Field

[0001] The present invention relates to a signal processing and fault diagnosis method, in particular to a method for fault identification using signal eigenvalue sidelobe characteristics and an implementation system thereof, and is particularly suitable for fault diagnosis of mechanical equipment. Background Art

[0002] Among the existing technologies, feature-based fault diagnosis methods are widely used. They rely on various feature information collected from the equipment to identify, analyze and diagnose faults. These methods often rely on feature information such as signal amplitude and frequency for fault identification. However, these methods are easily affected by factors such as noise and interference in complex environments. There may be some side lobes near the main features of the vibration spectrum, making the main fault features unclear, resulting in a high misjudgment rate of fault diagnosis. The reason is that during the propagation of vibration signals, multiple propagation paths and reflection surfaces may be encountered. These paths and reflection surfaces will cause the signals to be superimposed and interfered when they reach the receiving end, forming a complex spectrum structure. Especially in complex environments, such as when there are obstacles or irregular surfaces, vibration signals are more likely to be reflected and scattered. These reflected and scattered signals will be superimposed with the original signal to form side lobes. Or there is an impact in the system. When the signal is truncated during signal processing, its spectrum will change, and the originally concentrated energy will be dispersed into two wider frequency bands to form side lobes. As an important part of the signal propagation characteristics, the vibration spectrum eigenvalue sidelobe can reflect the slight changes in the signal propagation path and vibration energy, so it has unique advantages in fault identification. However, there is no mature technology to apply eigenvalue sidelobe analysis to the field of fault identification. Summary of the invention

[0003] The purpose of the present invention is to provide a fault identification method based on eigenvalue sidelobe, process the eigenvalue sidelobe, extract the eigenvalue sidelobe information in the signal, and combine it with advanced signal processing algorithms to achieve accurate identification of mechanical equipment faults.

[0004] To achieve the above object, the technical solution of the present invention is: a fault identification method based on eigenvalue sidelobe, comprising the following steps:

[0005] Step 1: By analyzing the spectrum distribution of the phenomenological model, the sidelobe characteristics directly related to the sub-vibration fault near the characteristic frequency are identified. The vibration signal phenomenological modeling is a simple and effective mathematical model established by studying the system operation law, vibration characteristics and vibration information transmission mode combined with actual experience observation from the perspective of the vibration sensor fixed on the equipment;

[0006] Build a mathematical model that takes all operating conditions into account:

[0007]

[0008] In the formula, v com (t) represents the composite fault signal, μ represents the fault characteristic, μ = 1, 2..., q (1, 2..., Q) is the vibration harmonic order, v q and is the amplitude and phase of the qth order vibration; a dis (t), b dis (t) is the amplitude and frequency modulation function of the distributed fault, R μ (t) Periodic function; when there is no fault, R μ (t) = 0;

[0009] Step 2, eigenvalue sidelobe extraction: using wavelet transform and Hilbert transform signal processing methods to extract eigenvalue sidelobe information from the preprocessed signal. The eigenvalue sidelobe includes the sidelobe areas on both sides of the main lobe, and these sidelobe areas contain important information on the signal propagation path;

[0010] Step 3, eigenvalue sidelobe analysis: quantitatively analyze the extracted eigenvalue sidelobe information, including sidelobe intensity, distribution, and change trend, to reveal subtle changes in the signal propagation path; after the signal is decomposed, a time-frequency representation matrix is ​​obtained, which contains the key information of the signal: vibration amplitude and frequency change. The purpose of feature extraction is to identify patterns related to the main vibration characteristics of the device from these coefficients, and pay attention to the energy distribution of the coefficients, which is achieved by calculating the energy of each scaling layer.

[0011] E(s)=∫|CWT(t,s)| 2 dt

[0012] In order to perform effective signal recognition and classification, a feature vector needs to be constructed. The feature vector extracts key features such as energy distribution and frequency peak based on eigenvalue sidelobe analysis. The feature vector is expressed as:

[0013] F=[F1,F2,...,F n ]

[0014] Where F1, F2, ..., Fn are different characteristic parameters, such as the energy value of a specific frequency band and the position of the energy peak;

[0015] Step 4, fault identification: According to the analysis results of the eigenvalue sidelobe, combined with the fault feature library and intelligent algorithm, the equipment fault is identified, and the contribution diagram of the commonly used fault identification diagram method is used. The contribution diagram based on the SPE statistic is defined as follows:

[0016]

[0017]

[0018] represents the contribution of each variable to the SPE statistic, ζ i represents the i-th column of the identity matrix I, t i is the projection of the data matrix onto the corresponding load vector, c is the projection matrix of the residual space of the sample data x,

[0019] Based on T 2 The contribution graph of is defined as follows:

[0020]

[0021]

[0022] Among them, T 2 It is a statistic that measures the change of variable values ​​in space, D = P T ∧ -1 P,ζ i represents the i-th column of the identity matrix I, P is the load matrix,

[0023] After a fault is detected, the contribution graph method is used for fault identification. The variable with a larger contribution value is considered to be the cause of the fault, which facilitates subsequent fault handling.

[0024] Further, the specific method of step 2 is as follows:

[0025] (a) The modal component v is obtained by HHT transformation k (t) Analyze the signal, calculate the one-sided spectrum, and convert v k The center band of (t) is moved to the corresponding baseband, that is: [(δ(t)+j / (πt)v k (t)]e -jwk , where δ(t) is the unit pulse function, v k is the kth modal component, and there is ∑ k v k (t) = f, ω k is the center frequency corresponding to the kth modal component;

[0026] (b) Calculate the square norm of the above equation and estimate the bandwidth f of each modal component;

[0027] (c) The Lagrange multiplication operator λ and the second-order penalty factor α are introduced to transform the constrained variational problem into an unconstrained variational problem;

[0028] (d) The optimal solution to the problem can be obtained by continuously updating each component and its center frequency using the alternating direction method of multiplication operator ADMM.

[0029]

[0030] Furthermore, a fault identification method based on characteristic value side lobes is applied to gear transmission. Gears will excite amplitude modulation and frequency modulation effects near the gear meshing frequency, resulting in fault characteristic modulation side lobes in the vibration signal spectrum. These characteristic modulation side lobes are extracted to realize gear fault diagnosis.

[0031] Firstly, the frequency spectrum distribution of the phenomenological model is used to identify the sidelobe characteristics directly related to the sub-vibration fault near the gear meshing characteristic frequency;

[0032] Then, the signal processing methods such as Hilbert transform are used to extract the eigenvalue sidelobe information from the preprocessed signal. The eigenvalue sidelobe includes the sidelobe areas on both sides of the main lobe. These sidelobe areas contain important information on the signal propagation path. The signal decomposition prevention based on the sidelobe characteristics can extract the fault characteristic sideband with a higher signal-to-noise ratio than the existing narrowband signal decomposition method. By adding the fault characteristic sidelobes, the gear meshing modulation component of the planetary gearbox fault can be accurately reconstructed.

[0033] Finally, a set eigenvector for evaluating the amplitude modulation intensity of the gear meshing sidelobe component is constructed, and the key features are extracted based on the eigenvalue sidelobe analysis, and the contribution graph method is used to realize the gear fault severity assessment.

[0034] Furthermore, after calculation and analysis, the characteristic recognition accuracy of gear meshing frequency was improved by 12.1% after adopting the fault identification method based on eigenvalue sidelobe.

[0035] The beneficial effects of the present invention are:

[0036] The present invention relates to the technical field of signal processing and fault diagnosis, and specifically to a method for fault identification using the sidelobe characteristics of a signal characteristic value and its implementation system, which is particularly suitable for fault diagnosis of mechanical equipment, and has the following specific beneficial effects:

[0037] (1) Improve fault identification accuracy: By using subtle changes in the eigenvalue sidelobes for fault identification, the identification accuracy can be significantly improved and the misjudgment rate can be reduced.

[0038] (2) Enhanced anti-interference capability: Eigenvalue sidelobe analysis has a strong suppression effect on noise and interference and can work stably in complex environments.

[0039] (3) Wide scope of application: The present invention is applicable to various types of equipment and systems and has strong versatility and practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a flow chart of a fault identification method based on eigenvalue side lobes of the present invention;

[0041] Figure 2 is a schematic diagram of the eigenvalue sidelobe. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0043] like Figure 1 , as shown in 2, a fault identification method based on eigenvalue sidelobe of the present invention comprises the following steps:

[0044] (1) By analyzing the spectrum distribution of the phenomenological model, the sidelobe characteristics directly related to the sub-vibration fault near the characteristic frequency are identified. Vibration signal phenomenological modeling is a simple and effective mathematical model established by studying the system operation rules, vibration characteristics and vibration information transmission mode, combined with actual experience observation, from the perspective of the vibration sensor fixed on the equipment.

[0045] Build a mathematical model that takes all operating conditions into account:

[0046]

[0047] In the formula, v com (t) represents the composite fault signal, μ represents the fault characteristic, μ = 1, 2..., q (1, 2..., Q) is the vibration harmonic order, v q and is the amplitude and phase of the qth order vibration; a dis (t), b dis (t) is the amplitude and frequency modulation function of the distributed fault, R μ (t) Periodic function; when there is no fault, R μ (t)=0.

[0048] (2) Eigenvalue sidelobe extraction: Using signal processing methods such as wavelet transform and Hilbert transform, the eigenvalue sidelobe information is extracted from the preprocessed signal. The eigenvalue sidelobe includes the sidelobe areas on both sides of the main lobe, which contain important information on the signal propagation path.

[0049] (a) The modal component v is obtained by HHT transformation k (t) Analyze the signal, calculate the one-sided spectrum, and convert v k The center band of (t) is moved to the corresponding baseband, that is: [(δ(t)+j / (πt)v k (t)]e -jwk , where δ(t) is the unit pulse function, v k is the kth modal component, and there is ∑ k v k (t) = f, ω k is the center frequency corresponding to the kth modal component.

[0050] (b) Calculate the square norm of the following equation and estimate the bandwidth f of each modal component.

[0051] (c) The Lagrange multiplication operator λ and the second-order penalty factor α are introduced to ensure the accuracy of signal reconstruction in Gaussian noise environment, and the constrained variational problem is transformed into an unconstrained variational problem.

[0052] (d) By using the alternating direction method of multiplication operator ADMM to continuously update each component and its center frequency, the optimal solution to the problem can be obtained.

[0053]

[0054] (3) Eigenvalue sidelobe analysis: Quantitative analysis of the extracted eigenvalue sidelobe information, including sidelobe intensity, distribution, and change trend, to reveal subtle changes in the signal propagation path. After the signal is decomposed, a time-frequency representation matrix is ​​obtained, which contains key signal information such as vibration amplitude and frequency changes. The purpose of feature extraction is to identify patterns related to the main vibration characteristics of the device from these coefficients and focus on the energy distribution of the coefficients, which can be achieved by calculating the energy of each scaling layer.

[0055] E(s)=∫|CWT(t,s)| 2 dt

[0056] In order to perform effective signal recognition and classification, a feature vector needs to be constructed. The feature vector extracts key features based on eigenvalue sidelobe analysis, such as energy distribution, frequency peak, etc. The feature vector can be expressed as:

[0057] F=[F1,F2,...,F n ]

[0058] Wherein F1, F2, ..., Fn are different characteristic parameters, such as the energy value of a specific frequency band, the position of an energy peak, etc.

[0059] Taking the gear vibration characteristic analysis as an example, when there is an impact-type fault such as broken teeth, the gear pair cannot mesh ideally at certain times, resulting in a periodic decrease in meshing stiffness, and the degree of reduction is related to the overlap of the gears and the degree of the impact-type fault. Assuming that only a single tooth has an impact-type fault, the energy method is used to calculate the meshing stiffness spectrum. In addition to the meshing frequency harmonics that exist when the gears are normally meshed, there are also sidelobe components distributed in the entire frequency band with the faulty gear rotation frequency as the interval. Under the modulation of the amplitude threshold, several energy components are added. This added energy mainly reflects the size of the amplitude threshold, or the strength of the fault; the interval width frequency of the amplitude modulation sideband reflects the rotation frequency of the faulty component.

[0060] (4) Fault identification: Based on the analysis results of the eigenvalue sidelobe, combined with the fault feature library and intelligent algorithms (such as machine learning, deep learning, etc.), equipment faults are identified.

[0061] The contribution diagram is a commonly used fault identification diagram method. The contribution diagram based on the SPE statistic is defined as follows:

[0062]

[0063]

[0064] represents the contribution of each variable to the SPE statistic, ζ i represents the i-th column of the identity matrix I. i is the projection of the data matrix onto the corresponding load vector, c is the projection matrix of the residual space of the sample data x, based on T 2 The contribution graph of is defined as follows:

[0065]

[0066]

[0067] Among them, T 2 It is a statistic that measures the change of variable values ​​in space, D = P T ∧ -1 P,ζ i represents the i-th column of the identity matrix I, and P is the loading matrix.

[0068] After a fault is detected, the contribution graph method is used for fault identification. The variable with a larger contribution value is considered to be the cause of the fault, which facilitates subsequent fault handling.

[0069] The present invention discloses a fault identification method based on characteristic value side lobes, which is applied to gear transmission. Gears will excite amplitude modulation and frequency modulation effects near the gear meshing frequency, thereby causing fault characteristic modulation side lobes to appear in the vibration signal spectrum. Extracting these characteristic modulation side lobes can realize gear fault diagnosis.

[0070] Firstly, the frequency spectrum distribution of the phenomenological model is used to identify the sidelobe characteristics directly related to the sub-vibration fault near the gear meshing characteristic frequency;

[0071] Then, the eigenvalue sidelobe information is extracted from the preprocessed signal using signal processing methods such as Hilbert transform. The eigenvalue sidelobe includes the sidelobe areas on both sides of the main lobe, which contain important information on the signal propagation path. Compared with the existing narrowband signal decomposition method, the signal decomposition prevention based on sidelobe characteristics can extract fault characteristic sidebands with higher signal-to-noise ratio. By adding the fault characteristic sidelobes, the gear meshing modulation component of the planetary gearbox fault can be accurately reconstructed;

[0072] Finally, a set eigenvector for evaluating the amplitude modulation intensity of the gear meshing sidelobe component is constructed, and the key features are extracted based on the eigenvalue sidelobe analysis, and the contribution graph method is used to realize the gear fault severity assessment.

[0073] After analysis and calculation, the accuracy of characteristic recognition of gear meshing frequency was improved by about 12.1% after adopting the fault identification method based on eigenvalue sidelobe.

Claims

1. A fault identification method based on eigenvalue sidelobe, characterized in that: It includes the following steps: Step 1: By analyzing the spectrum distribution of the phenomenological model, the sidelobe characteristics directly related to the sub-vibration fault near the characteristic frequency are identified. The vibration signal phenomenological modeling is a simple and effective mathematical model established by studying the system operation law, vibration characteristics and vibration information transmission mode combined with actual experience observation from the perspective of the vibration sensor fixed on the equipment; Build a mathematical model that takes all operating conditions into account: In the formula, v com (t) represents the composite fault signal, μ represents the fault characteristic, μ = 1, 2..., q (1, 2..., Q) is the vibration harmonic order, v q and is the amplitude and phase of the qth order vibration; a dis (t), b dis (t) is the amplitude and frequency modulation function of the distributed fault, R μ (t) Periodic function; when there is no fault, R μ (t) = 0; Step 2, eigenvalue sidelobe extraction: using wavelet transform and Hilbert transform signal processing methods to extract eigenvalue sidelobe information from the preprocessed signal. The eigenvalue sidelobe includes the sidelobe areas on both sides of the main lobe, and these sidelobe areas contain important information on the signal propagation path; Step 3, eigenvalue sidelobe analysis: quantitatively analyze the extracted eigenvalue sidelobe information, including sidelobe intensity, distribution, and change trend, to reveal subtle changes in the signal propagation path; after the signal is decomposed, a time-frequency representation matrix is ​​obtained, which contains the key information of the signal: vibration amplitude and frequency change. The purpose of feature extraction is to identify patterns related to the main vibration characteristics of the device from these coefficients, and pay attention to the energy distribution of the coefficients, which is achieved by calculating the energy of each scaling layer. E(s)=∫|CWT(t,s)| 2 dt In order to perform effective signal recognition and classification, a feature vector needs to be constructed. The feature vector extracts key features such as energy distribution and frequency peak based on eigenvalue sidelobe analysis. The feature vector is expressed as: F=[F1,F2,...,F n ] Where F1, F2, …, Fn are different characteristic parameters, such as the energy value of a specific frequency band and the position of the energy peak; Step 4, fault identification: According to the analysis results of the eigenvalue sidelobe, combined with the fault feature library and intelligent algorithm, the equipment fault is identified, and the contribution diagram of the commonly used fault identification diagram method is used. The contribution diagram based on the SPE statistic is defined as follows: represents the contribution of each variable to the SPE statistic, ζ i represents the i-th column of the identity matrix I, t i is the projection of the data matrix onto the corresponding load vector, c is the projection matrix of the residual space of the sample data x, Based on T 2 The contribution graph of is defined as follows: Among them, T 2 It is a statistic that measures the change of variable values ​​in space, D = P T ∧ -1 P,ζ i represents the i-th column of the identity matrix I, P is the load matrix, After a fault is detected, the contribution graph method is used for fault identification. The variable with a larger contribution value is considered to be the cause of the fault, which facilitates subsequent fault handling.

2. The fault identification method based on eigenvalue sidelobe according to claim 1, characterized in that: The specific method of step 2 is as follows: (a) The modal component v is obtained by HHT transformation k (t) Analyze the signal, calculate the one-sided spectrum, and convert v k The center band of (t) is moved to the corresponding baseband, that is: [(δ(t)+j / (πt)v k (t)]e -jwk , where δ(t) is the unit pulse function, v k is the kth modal component, and there is Σ k v k (t) = f, ω k is the center frequency corresponding to the kth modal component; (b) Calculate the square norm of the above equation and estimate the bandwidth f of each modal component; (c) The Lagrange multiplication operator λ and the second-order penalty factor α are introduced to transform the constrained variational problem into an unconstrained variational problem; (d) The optimal solution to the problem can be obtained by continuously updating each component and its center frequency using the alternating direction method of multiplication operator ADMM.

3. The fault identification method based on eigenvalue sidelobe according to claim 2, characterized in that: This method is applied to gear transmission. Gears will excite amplitude modulation and frequency modulation effects near the gear meshing frequency, which will lead to fault characteristic modulation side lobes in the vibration signal spectrum. These characteristic modulation side lobes are extracted to realize gear fault diagnosis. Firstly, the frequency spectrum distribution of the phenomenological model is used to identify the sidelobe characteristics directly related to the sub-vibration fault near the gear meshing characteristic frequency; Then, the signal processing methods such as Hilbert transform are used to extract the eigenvalue sidelobe information from the preprocessed signal. The eigenvalue sidelobe includes the sidelobe areas on both sides of the main lobe. These sidelobe areas contain important information on the signal propagation path. The signal decomposition prevention based on the sidelobe characteristics can extract the fault characteristic sideband with a higher signal-to-noise ratio than the existing narrowband signal decomposition method. By adding the fault characteristic sidelobes, the gear meshing modulation component of the planetary gearbox fault can be accurately reconstructed. Finally, a set eigenvector for evaluating the amplitude modulation intensity of the gear meshing sidelobe component is constructed, and the key features are extracted based on the eigenvalue sidelobe analysis, and the contribution graph method is used to realize the gear fault severity assessment.

4. The fault identification method based on eigenvalue sidelobe according to claim 3 is characterized in that: After calculation and analysis, the characteristic recognition accuracy of gear meshing frequency was improved by 12.1% after adopting the fault identification method based on eigenvalue sidelobe.

Images