A method for identifying rotor damping ratio based on key phase sensing measurement

Abstract

This invention discloses a rotor damping ratio identification method based on key phase sensing measurement, belonging to the technical field of rotating machinery. The method includes: bandpass filtering of the key phase signal; calculating the self-power spectral density and then implementing vectorized Hampel filtering to construct a window matrix to detect and replace outliers caused by harmonics; reconstructing the complex spectrum and inversely transforming it to generate a time-domain pulse decay signal; extracting eigenvalues ​​using an inverse autoregressive model and automatically obtaining stability parameters using a clustering algorithm. This method analyzes the modal parameters of the rotor through key phase signals, expanding the application of key phase signals in rotor dynamics analysis. During signal analysis, the power spectral density function is treated as a special "time series," and vectorized Hampel filtering achieves excellent harmonic interference effects. Furthermore, replacing the cyclic sliding window with vectorized operations significantly improves computational efficiency, providing a reliable technical means for real-time monitoring of rotating machinery.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of rotating machinery, and particularly relates to a rotor damping ratio identification method based on key phase sensing measurement. BACKGROUND

[0002] Rotating power machinery is the core power device in the industrial fields of petrochemical, power and metallurgy, and its operation stability directly affects the reliability of the industrial production system. With the evolution of modern industrial equipment towards high power density, supercritical parameters and limit flow, the accurate identification of the dynamic characteristics of the rotor system faces severe challenges. Engineering practice shows that constructing an early warning mechanism by real-time monitoring of key parameters such as vibration spectrum, pressure pulsation and temperature field distribution is an effective means to realize equipment state prediction. However, the existing monitoring system has some shortcomings. On the one hand, the data processing method is difficult to realize real-time analysis and early warning response of massive monitoring data, resulting in significant lag in fault early warning; on the other hand, the feature extraction process of multi-source signals is easily disturbed by background noise, causing the cumulative amplification of modal parameter identification error, which seriously affects the accuracy of early warning. Modal analysis as a key technology can effectively identify the dynamic characteristics of the structure, and has extremely important engineering value for structure optimization design, fault diagnosis, damage identification and health monitoring. However, most of the traditional OMA modal analysis is carried out in the laboratory under high signal-to-noise ratio conditions, which is a challenge for traditional working modal parameter identification methods in the actual engineering operation process of engines, turbines, compressors and other equipment. The system structure is often excited by non-white noise such as harmonic excitation.

[0003] Therefore, optimizing the data processing process to reduce the amount of data, solving the problem of harmonic interference, improving the identification efficiency and accuracy, will help to realize the efficient, stable and accurate modal identification of rotating power machinery. It has great engineering significance for the stable operation of the equipment. SUMMARY

[0004] The technical purpose of the present application is to provide a rotor damping ratio identification method based on key phase sensing measurement, which combines industrial data processing and dynamic mechanism in rotating machinery state monitoring, identifies rotor system modal parameters through time domain signals collected by key phase sensors in real time, realizes harmonic interference suppression by using an improved Hampel filter algorithm, and optimizes the signal processing procedure based on vector calculation method. The rotor damping ratio identification model based on key phase sensor signal analysis is constructed, which expands the application scope of key phase signals in rotor dynamics parameter identification. The Hampel filter is introduced to suppress non-Gaussian noise of the original signal, and the harmonic interference components are effectively removed by dynamically adjusting the threshold of the sliding window, solving the modal aliasing problem caused by periodic process disturbance in traditional industrial data processing. The matrix vectorization calculation method is used to reconstruct the filter algorithm in parallel, and the sliding window filter algorithm is reconstructed as a two-dimensional matrix block operation, optimizing the industrial data processing procedure and improving the overall efficiency. The present method effectively solves the problems of insufficient real-time performance and limited modal parameter identification accuracy in existing monitoring technology.

[0005] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is a rotor damping ratio identification method based on key phase sensing measurement, comprising:

[0006] S1, obtaining the key phase signal of the rotating machinery vibration measurement plane key phase sensor, performing fast Fourier transform (FFT) on the key phase signal, calculating the modulus square of the frequency domain complex result, and obtaining the key phase signal self-power spectral density function representing the vibration characteristics of the rotor system through one-sided spectrum normalization.

[0007] S2, performing Hampel filtering processing on the key phase signal self-power spectral density function. Firstly, according to the frequency band distribution characteristics of harmonic interference in industrial data, a dynamic sliding window is constructed on the key phase signal self-power spectral density function; the width of the dynamic sliding window is dynamically adjusted according to the rotor speed, wherein the width of the dynamic sliding window is inversely proportional to the speed; secondly, the vectorization processing method is used to convert the spectral data in the dynamic sliding window into a two-dimensional matrix block, and the median and the absolute median deviation (MAD) in each window are calculated based on the statistical mechanism; finally, according to the preset threshold, the power spectral density function is filtered to remove harmonic interference and obtain the power spectral density function after Hampel filtering.

[0008] S3, performing inverse Fourier transform on the power spectral density function after Hampel filtering, and normalizing the time domain decay signal after inverse Fourier transform to ensure the numerical stability of subsequent parameter identification, and obtaining the time domain pulse decay signal after Hampel filtering.

[0009] S4, constructing a reverse auto-regressive model according to the Hampel filtered time-domain pulse decay signal, and calculating corresponding stability parameters of each order of the rotor system, including damping ratio and natural frequency; according to the statistical distribution characteristics of industrial data, using the 3σ criterion to remove abnormal data of the stability parameters, and taking the mean value of the normal working condition data as the dynamic characteristic identification result of the rotor system.

[0010] Further, the S1 comprises the following steps:

[0011] S1.1, synchronously collecting a key phase signal on a measurement plane of a rotating machine key phase sensor at a sampling frequency of not less than 2 kHz; and performing preliminary filtering processing on the collected key phase signal, and removing high-frequency noise and part of low-frequency interference by using a band-pass filter, and setting a cut-off frequency of the band-pass filter according to frequency spectrum distribution characteristics of the key phase signal.

[0012] S1.2, performing fast Fourier transform (FFT) on the key phase signal, calculating a modulus square of a frequency domain complex result, and obtaining a key phase signal self-power spectrum density function representing vibration characteristics of the rotor system by single-sided spectrum normalization.

[0013] Further, the S2 comprises the following steps:

[0014] S2.1, supplementing a specific number of data points (usually zero value filling or mirror symmetry continuation) at a starting and ending position of the key phase signal self-power spectrum density function, taking each original data point (non-supplemented data point) as a center to form a dynamic sliding window, and representing each dynamic sliding window as each row of a matrix to form a window matrix of the key phase signal self-power spectrum density function.

[0015] S2.2, obtaining a median of each dynamic sliding window of the power spectrum density window matrix of the key phase signal by sorting and taking a middle value, and calculating an absolute difference value between each element in the window and the median, and taking a median of the absolute difference values as a median absolute deviation (MAD).

[0016] S2.3, according to a preset threshold, comparing a difference between each element in the sliding window and the median and a product of the absolute median deviation (MAD) and the threshold, performing an abnormal value judgment on the power spectrum density window matrix of the key phase signal, and replacing the abnormal point with a median of the corresponding window, so as to batch eliminate harmonic interference in the key phase signal, and obtaining a de-harmonic power spectrum density function.

[0017] Further, the S4 comprises the following steps:

[0018] S4.1, respectively, reverse autoregressive modal parameter identification is performed on the time domain pulse attenuation signal to obtain characteristic roots of a discrete model of each order, and based on the characteristic roots, an inherent frequency and a damping ratio corresponding to each order in the measured key belief signal are calculated.

[0019] S4.2, a 3σ clustering algorithm is used to automatically cluster the inherent frequencies and damping ratios of different orders in the key belief signal. Through clustering analysis, similar dynamic characteristic parameters are classified into a class, thereby realizing efficient identification and classification of the dynamic characteristics of the rotor system. The mean value of the normal working condition data is taken as the identification result of the dynamic characteristics of the rotor system, ensuring the reliability and accuracy of the result.

[0020] The present application has the following advantages:

[0021] The rotor stability parameter identification method based on key belief signal anti-harmonic interference proposed in the present application has significant innovation and practicality. The technology expands the use of key belief signals in rotor dynamics analysis. Compared with the key belief signals in the prior art which are only used for speed extraction and phase-based synchronous sampling, the present method analyzes the modal parameters of the rotor by constructing a key belief signal power spectrum function. In the signal processing link, the Hampel filtering algorithm is introduced. The signal processed by the Hampel filtering algorithm greatly restores the modal characteristics of the rotor system in the power spectrum, which simplifies the subsequent modal parameter extraction process. Compared with the traditional signal processing method, the calculation complexity of the Hampel filtering algorithm is relatively low. This advantage makes it particularly suitable for real-time or near-real-time data processing needs. In the key fields of online monitoring and fault diagnosis in industrial production, the timeliness of data processing is crucial. Furthermore, the Hampel filtering method is further vectorized in the present application. Through vectorization, the calculation speed of the method is significantly improved. In the online monitoring scenario, fast calculation speed means that subtle changes in the running state of the rotor system can be captured more timely, thereby providing more efficient and accurate operation data for the system. This has immeasurable value for realizing real-time monitoring and fault warning of the rotor system. BRIEF DESCRIPTION OF DRAWINGS

[0022] The present application will be further described in detail below with reference to the accompanying drawings and specific implementation methods. The accompanying drawings described herein are used to provide further understanding of the present application, constitute a part of the present application, and do not constitute an improper limitation on the present application.

[0023] Figure 1 is a flowchart for implementing the embodiments of the present application;

[0024] Figure 2 is a graph of analyzing the amplitude-frequency curve of the signal power spectrum density function in the embodiments of the present application;

[0025] Figure 3 A flowchart of the vectorized Hampel filter method in the embodiment of the present application;

[0026] Figure 4 A magnitude-frequency curve diagram of the power spectral density function after de-harmonic interference in the embodiment of the present application;

[0027] Figure 5 A time-domain pulse decay signal curve diagram before filtering in the embodiment of the present application;

[0028] Figure 6 A time-domain pulse decay signal curve diagram after filtering in the embodiment of the present application;

[0029] Figure 7 A two-dimensional steady-state diagram of the fitted order and natural frequency in the embodiment of the present application;

[0030] Figure 8 A two-dimensional steady-state diagram of the fitted order and damping ratio in the embodiment of the present application;

[0031] Figure 9 A steady-state diagram data automatic clustering result diagram corresponding to the decay signal in the embodiment of the present application. DETAILED DESCRIPTION

[0032] The technical solutions of the present application will be further described in detail below with the aid of the accompanying drawings and embodiments. The description of the exemplary embodiments is merely illustrative, and by no means as any limitation on the present disclosure and its applications or uses. The present disclosure can be implemented in many different forms, not limited to the embodiments described herein. These embodiments are provided to make the present disclosure thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0033] The embodiments of the present application will be described in detail below, please refer to Figure 1 The present application discloses a rotor damping ratio identification method based on key phase sensing measurement, comprising:

[0034] S1, acquiring a key phase signal of a rotating machinery vibration measurement plane key phase sensor, synchronously collecting the signal at a sampling frequency not lower than 2 kHz; performing fast Fourier transform (FFT) on the key phase signal, calculating the modulus square of the frequency domain complex result, and obtaining a self-power spectral density function representing the vibration characteristics of the rotor system through single-sided spectrum normalization.

[0035] Acquisition of the key phase signal of the rotor: a small convex metal sheet or a groove is pasted on the shaft surface as a key phase, and an eddy current displacement sensor is used to measure the pulse signal of the key phase position.

[0036] S1.1 Collecting the key phase signal of the rotating machinery on the measurement plane where the key phase sensor is located at a sampling frequency of no less than 2 kHz; performing preliminary filtering on the collected key phase signal, and removing high-frequency noise and part of low-frequency interference by using a band-pass filter, the cut-off frequency of which is set according to the frequency spectrum distribution characteristics of the key phase signal.

[0037] The collected key phase signal z(t) is input into a band-pass filter, the passband range of which is determined according to the frequency characteristics of the key phase signal. The passband range [0.8 f0, 1.2(M+1) f0] (f0 is the real-time rotational frequency, and M is the harmonic order of interest, and by default M=5), removes non-synchronous vibration interference;

[0038] For example, if the rotational frequency f0 measured according to the key phase signal of the rotating machinery is 50 Hz, and the harmonic order of interest M is 5, then the passband of the band-pass filter is set to 40 Hz~360 Hz, so as to ensure that the effective signal is retained, and at the same time, high-frequency and low-frequency interference beyond the range is filtered out.

[0039] S1.2 Performing fast Fourier transform (FFT) on the key phase signal, calculating the modulus square of the complex result in the frequency domain, and obtaining the self-power spectral density function representing the vibration characteristics of the rotor system through one-sided spectrum normalization.

[0040] Performing fast Fourier transform (FFT) on the key phase signal z(t) to obtain Z(w), and then calculating the modulus square of the complex amplitude after Fourier transform to obtain the self-power spectrum function of the key phase signal. Specifically, the power spectral density function expression is:

[0041]

[0042] Wherein, G(w) represents the power spectral density at the angular frequency w, w is the angular frequency; Z(w) is the Fourier transform of the key phase signal z(t), that is, , t is time, j is the imaginary unit (engineering symbol commonly used), and e is the natural constant in mathematics (the base of natural logarithm, approximately equal to 2.71828); is the modulus square of the complex amplitude after Fourier transform.

[0043] In an embodiment, the key phase signals of a group of rotors with flexible foundations at a rotational speed of 2000 r / min and a sampling rate of fs=20 kHz are analyzed. First, the power spectral density of the key phase signal z(t) of the rotor system with flexible foundation is analyzed, and the power spectral density function G(w) of the key phase signal is obtained as shown in Figure 2 Figure 2 ​It can be seen that, in addition to the rotational frequency (f0=33.3Hz) representing the normal operation of the rotor system, there are also a large number of harmonic and sub-harmonic components. Near these frequency components, there are also a large number of spike signals, which are distributed at different frequency points, i.e., the so-called harmonic interference.

[0044] S2. Hampel filtering is applied to the power spectral density function of the key phase signal. First, based on the frequency band distribution characteristics of harmonic interference in industrial data, a dynamic sliding window is constructed on the power spectral density function of the key phase signal. The window width is dynamically adjusted according to the rotor speed (the window width is inversely proportional to the speed). Second, a vectorization method is used to convert the spectral data within the sliding window into two-dimensional matrix blocks. The median and absolute median difference (MAD) of each window are calculated based on statistical mechanisms. Finally, the power spectral density function is filtered according to a preset threshold to remove harmonic interference and obtain the Hampel-filtered power spectral density function.

[0045] In one embodiment, please refer to Figure 3 The vectorized Hampel filtering method includes:

[0046] S2.1 Supplement a specific number of data points (usually zero-value filling or mirror-symmetric extension) at the start and end positions of the self-power spectral density function of the bond phase signal, and form a dynamic sliding window with each original data point (non-supplemented data point) as the center. Each dynamic sliding window is represented as each row of a matrix, forming a window matrix of the self-power spectral density function of the bond phase signal.

[0047] Set the dynamic sliding window length L = 2k + 1, where k is the half-length of the window. L is dynamically adjusted according to the rotor speed. When L < 3, it is forced to be set to L = 3 (i.e., the half-length of the window k = 1) to avoid the filter from failing due to the window being too small.

[0048] The preset window length is L1. When L1 is even, the window length is L = L1 - 1; when L1 is odd, the window length is L = L1. Specifically, the preset window length is expressed as:

[0049]

[0050] Among them, f s The sampling frequency (Hz) of the key phase signal of the planar key phase sensor for rotating machinery vibration measurement is determined by the data acquisition system hardware; f r The rotor fundamental frequency (Hz) is calculated using the pulse interval of the key phase signal; ⌊⋅⌋ represents the floor function.

[0051] The power spectral density function of the G(w) bond phase signal is padded with k (half the window length) zeros before and after to ensure sufficient data at the boundary during sliding. If the original signal length is n, the extended signal length is n+2k. Specifically, the extended bond phase signal power spectral density function is expressed as:

[0052]

[0053] in Let G1, G2, ..., G be the power spectral density function of the bond phase signal after signal expansion; G1, G2, ..., G be the power spectral density function of the bond phase signal. n Let n be the element in G(w), and n be the signal length of the power spectral density function of the bond phase signal in G(w).

[0054] Using the elements G in G(w) i Centered on a point, k neighboring points are selected in front and behind to form a dynamic sliding window Z of fixed length. i ={G i-k G i-k+1 G i-k+2 ,…, G i , …, G i+k ,}, where i=1,2,3,……,n; a series of dynamic sliding windows are generated by traversing the extended power spectral density function of the bond phase signal. Each sliding window is represented as a row of a window matrix, and finally a power spectral density window matrix of shape n×L is constructed, specifically represented as:

[0055]

[0056] S2.2 For the window matrix of the power spectral density of the bond phase signal, the median value of each dynamic sliding window is obtained by sorting and taking the median value, and the absolute difference between each element in the window and the median value in the window is calculated. The median absolute deviation (MAD) is taken as the median absolute deviation.

[0057] Calculate the median within each dynamic sliding window. After sorting the data within each dynamic sliding window, the k+1th position of each window is the median, specifically expressed as follows:

[0058]

[0059] Where m is the row midpoint vector of the power signal window matrix; For the i-th dynamic sliding window Z i The (k+1)th element after sorting the internal elements, where Z i For Z G The i-th row of the window matrix.

[0060] For each dynamic sliding window, calculate the absolute difference between each element and the value in the window, and take the median absolute deviation (MAD) as the median absolute deviation. The specific expression is as follows:

[0061]

[0062] Z G ij Z is the power spectral density window matrix of the bond phase signal. G The element in the i-th row and the j-th element; For row MAD vectors , m is the absolute deviation of the median of the i-th dynamic sliding window; i It is the i-th element of the value vector in the power signal window matrix.

[0063] S2.3 Based on a preset threshold, outlier determination is performed on the power spectral density window matrix of the bond phase signal by comparing the product of the difference between each element in the sliding window and the median, and the absolute median difference (MAD), with the threshold. Outliers are replaced with the median of the corresponding window, thereby eliminating harmonic interference in the batch of bond phase signals and obtaining a deharmonicized power spectral density function.

[0064] The absolute value deviation of the median is calculated and multiplied by a set threshold (usually a multiple of MAD). Each element of the power spectral density window matrix is ​​then compared. If the absolute value of the difference between an element and the median exceeds the product of MAD and the threshold, that element is marked as an outlier. Outliers are then replaced in batches using Boolean indexing to obtain the Hampel-filtered power spectral density function G. d (w). Specifically, outlier detection is represented as:

[0065] If the current point Z G ij satisfy: Then determine Z G ij This is an outlier.

[0066] Where t0 is the outlier detection threshold, which is usually set to 3.

[0067] If the discrete amplitude signal in the power spectral density function is regarded as a special "time series", then the steep peaks formed by harmonics in the key phase signal can be regarded as "outliers" because they are far from the group. By using the Hampel filter to detect and remove outliers, an excellent harmonic interference removal effect can be achieved.

[0068] For example, regarding Figure 2 The power spectral density plot G(w) of the key phase signal is shown, with a sampling rate fs = 20 kHz and a rotor fundamental frequency f. r=f0=33.3Hz, the calculated signal window length L=299, then the half-length of the window k=149; a vectorized Hampel filter is used to process the power spectral density function of the key phase signal. When setting the parameters, the threshold t0 is set to 3. After processing by this filter, the following is obtained: Figure 4 The results are shown. Comparing the images before and after processing reveals that... Figure 2 The original harmonic interference has disappeared, and the entire spectrum has become smoother, more accurately reflecting the true characteristics of the compressor vibration signal.

[0069] S3. Perform an inverse Fourier transform on the power spectral density function after Hampel filtering, and normalize the complex signal after the inverse transform to ensure the numerical stability of subsequent parameter identification, thereby obtaining the time-domain pulse attenuation signal after Hampel filtering.

[0070] After filtering the power spectral density function, in order to further recover the complete characteristics of the signal, it is necessary to further analyze G... d The (w) function supplements the zero-phase constructs a complex spectrum Z. d (w).

[0071] The inverse Fourier transform is used to restore the complex spectrum in the frequency domain to a time-domain pulse decay signal. Specifically, the time-domain pulse decay signal z1(t) is expressed as:

[0072]

[0073] After obtaining the time-domain signal z1(t), to ensure the stability of the numerical calculation and maintain the consistency of the signal's dimensions, the signal is normalized using maximum absolute value normalization. The specific normalization is expressed as follows:

[0074]

[0075] For example, regarding Figure 2 and Figure 4 The power spectral density function shown is the power spectral density function G(w) without filtering and the power spectral density function G(w) after filtering. d (w) Perform an inverse Fourier transform, and then normalize the maximum amplitude to obtain the result. Figure 5 The time-domain pulse decay signal corresponding to the power spectral density function before filtering is shown. Figure 6 The filtered power spectral density function shown corresponds to the time-domain pulse attenuation signal. (Comparison) Figure 5 and Figure 6 The time-decayed signal, without filtering. Figure 5 The time-domain signal is frequency-transformed and approximated as a cosine signal; it is then processed by spectral domain filtering. Figure 6The time-domain signal effectively eliminates frequency conversion and harmonic interference, and better represents the original attenuation characteristics.

[0076] S4. Based on the time-domain pulse attenuation signal after Hampel filtering, construct an inverse autoregressive model to calculate the corresponding stability parameters (damping ratio and natural frequency) for each order of the rotor system. Based on the statistical distribution characteristics of industrial data, use the 3σ criterion to remove abnormal data from the stability parameters, and use the mean of the normal operating data as the identification result of the dynamic characteristics of the rotor system.

[0077] S4.1 Perform inverse autoregressive mode parameter identification on the filtered time-domain pulse attenuation signal to obtain the characteristic roots of the discrete model of each order; and calculate the natural frequency and damping ratio corresponding to each order in the measured key phase signal based on the characteristic roots.

[0078] Specifically, a single-input single-output inverse autoregressive discrete model is used for the S(t) time-domain pulse decay signal:

[0079]

[0080] Where S is the time-domain attenuated signal after Hampel filtering; N is the signal sequence discretized from the time-domain attenuated signal; p is the order of the inverse autoregressive model; b i Here are the system characteristic parameters; W is a white noise sequence with a mean of 0 and a sampling interval of . ;

[0081] The sequence of the time-domain decayed signal S after Hampel filtering is 1, 2, ..., N-1. A system of linear equations is constructed based on the above inverse autoregressive model:

[0082]

[0083] Where S[p] simplifies the above equation to:

[0084]

[0085] In the formula .

[0086] Performing a z-transform on the above equation yields:

[0087]

[0088] In the formula z is the discrete transform factor; S(z) is the output signal S [N] The z-transform; W(z) is the output signal W [N] z-transform;

[0089] In real-world signals, there is a certain amount of noise interference. Therefore, SVD is used to decompose the noise in signal A.

[0090]

[0091] In the formula, A is the original matrix of size Np×p; U is an Np×Np orthogonal matrix (left singular value vector matrix), U r To preserve the true signal direction, U r+ The direction of higher-order noise; V is the transpose of a p×p orthogonal matrix (right singular vector matrix). To preserve the transpose of the right singular vector matrix of the first r principal components, S is the transpose of the right singular vector matrix of the higher-order noise component after decomposition; S is an Np×p diagonal matrix (singular value matrix). The retained portion after SVD decomposition has an order of r×r. For the higher-order noise component; the minimum noise reduction order r needs to satisfy the following conditions:

[0092]

[0093] Therefore, the denoised matrix A is represented as:

[0094]

[0095] Based on the aforementioned system of linear equations, the parameter vector b of the model can be estimated as follows:

[0096]

[0097] The characteristic equation of the system This allows us to obtain the eigenvalues ​​corresponding to each mode of the system. ;

[0098]

[0099] This allows us to obtain the natural frequency, damping ratio, and logarithmic decay rate of the rotor system.

[0100]

[0101] S4.2 usage The clustering algorithm automatically clusters the natural frequencies and damping ratios of different orders in the key phase signal. By grouping similar dynamic characteristic parameters into one category through cluster analysis, efficient identification and classification of the rotor system's dynamic characteristics are achieved. The mean value of normal operating condition data is used as the identification result of the rotor system's dynamic characteristics to ensure the reliability and accuracy of the results.

[0102] After obtaining the natural frequency and damping ratio from the time-domain pulse attenuation signal processing, an automatic statistical clustering algorithm is used to further eliminate outliers for automatic monitoring and display.

[0103] Specifically, use Clustering methods are used to cluster the results of inverse autoregressive parameter identification at different orders.

[0104]

[0105] In the formula For the natural frequency, The damping ratio; and This represents the average of the natural frequency and the damping ratio. and The standard deviation is denoted as .

[0106] After multiple iterations and outlier removal, the mean values ​​of the natural frequency and damping ratio after statistical clustering are calculated. , This allows us to obtain an estimate of the system's stability identification results.

[0107] For example, for Figure 6 The damping ratio and natural frequency of the time-domain decay signal of the key phase signal are identified. An inverse autoregressive model is used to identify stability parameters, with the order of the inverse autoregressive model set to p=300. The natural frequency and damping ratio of the rotor system on the flexible foundation are calculated, such as... Figure 7 , 8 As shown, the horizontal axis represents the order, and the vertical axis represents the damping ratio and natural frequency, respectively. The final result obtained by clustering, such as Figure 9 As shown, the center of the ellipse represents the final parameter estimate ( = (49.28Hz, 0.75%), which is the result of rotor system stability identification based on key phase signal.

[0108] The above description is a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art can make appropriate adjustments without departing from the principle of the present invention, and these adjustments should all be included within the scope of protection of the present invention.

Claims

1. A method for rotor damping ratio identification based on keyphasor sensing measurements, characterized by, It comprises the following steps: S1, acquiring the key phase signal of the rotating machinery vibration measurement plane key phase sensor; performing fast Fourier transform (FFT) on the key phase signal, calculating the modulus square of the frequency domain complex result, and obtaining the key phase signal self-power spectrum density function representing the vibration characteristics of the rotor system through single-sided spectrum normalization; S2, Hampel filtering processing is performed on the key phase signal self-power spectrum density function: first, according to the frequency band distribution characteristics of harmonic interference in industrial data, a dynamic sliding window is constructed on the key phase signal self-power spectrum density function; the width of the dynamic sliding window is dynamically adjusted according to the rotor speed, wherein the width of the dynamic sliding window is inversely proportional to the speed; second, the spectral data in the dynamic sliding window is converted into a two-dimensional matrix block by using the vectorization processing method, and the median and the absolute median deviation (MAD) in each window are calculated based on the statistical mechanism; finally, according to the preset threshold, the power spectrum density function is filtered to remove harmonic interference and obtain the Hampel filtered power spectrum density function; S3, inverse Fourier transform is performed on the Hampel filtered power spectrum density function, and the time domain decay signal after inverse Fourier transform is normalized to ensure the numerical stability of subsequent parameter identification, and the Hampel filtered time domain pulse decay signal is obtained; S4, an inverse autoregressive model is constructed according to the Hampel filtered time domain pulse decay signal, and the corresponding stability parameters of each order of the rotor system are calculated, including damping ratio and natural frequency; according to the statistical distribution characteristics of industrial data, the 3σ criterion is used to remove abnormal data of the stability parameters, and the mean value of the normal working condition data is taken as the dynamic characteristic identification result of the rotor system.

2. The method for identifying the rotor damping ratio based on the key-phase sensing measurement according to claim 1, characterized in that, The S1 comprises the following steps: S1.1, synchronously collecting the key phase signal of the rotating machinery key phase sensor at a sampling frequency of not less than 2kHz; performing preliminary filtering processing on the collected key phase signal, and removing high frequency noise and part of low frequency interference by using a band pass filter; the cutoff frequency of the band pass filter is set according to the spectral distribution characteristics of the key phase signal; S1.2, performing fast Fourier transform (FFT) on the key phase signal, calculating the modulus square of the frequency domain complex result, and obtaining the key phase signal self-power spectrum density function representing the vibration characteristics of the rotor system through single-sided spectrum normalization.

3. The method of claim 1, wherein, The S2 comprises the following steps: S2.1, supplementing a certain number of data points at the starting and ending positions of the key phase signal self-power spectrum density function to form a dynamic sliding window centered on each original data point, and each dynamic sliding window is represented as each row of the matrix to form a window matrix of the key phase signal self-power spectrum density function; S2.2, obtaining the median of each dynamic sliding window by sorting the power spectrum density window matrix of the key phase signal and taking the middle value, and calculating the absolute difference between each element in the window and the window median, and taking the median of these absolute differences as the median absolute deviation (MAD); S2.3, according to a preset threshold, performing outlier determination on the power spectrum density window matrix of the keyphasel signal by comparing the difference between each element in the sliding window and the median with the product of the absolute median deviation MAD and the threshold; determining that the abnormal point is replaced by the median of the corresponding window, thereby batch eliminating the harmonic interference in the keyphasel signal to obtain a deharmonic power spectrum density function.

4. The method of claim 1, wherein, The S4 includes the following steps: S4.1, performing inverse autoregressive modal parameter identification on the time-domain pulse decay signal respectively to obtain characteristic roots of a discrete model of each order, and calculating an inherent frequency and a damping ratio corresponding to each order in the measured keyphasel signal based on the characteristic roots; S4.2, performing automatic clustering operation on the inherent frequencies and the damping ratios of different orders in the keyphasel signal by using a 3σ clustering algorithm; and grouping similar dynamic characteristic parameters into a class by clustering analysis, thereby realizing efficient identification and classification of the dynamic characteristics of the rotor system.

Images