Single-sensing rotor blade frequency identification method based on projection method

Abstract

The invention discloses a single-sensing rotor blade frequency identification method based on a projection method, and belongs to the field of rotor blade non-contact state monitoring. The method comprises the following steps: acquiring a signal based on a leaf tip timing single-sensor sampling mechanism, performing aliasing mechanism analysis of a leaf tip timing single-sensor frequency spectrum on the acquired signal, and determining whether undersampling occurs or not, a spectral line inevitably exists at a corresponding position of a real frequency in the frequency spectrum in the signal; variable rate sampling is carried out through blade tip timing single sensing, and an original signal is obtained; performing time-frequency analysis on original signals obtained by single-sensing variable-rate sampling through a multiple signal classification method to obtain a time-frequency matrix; and projecting each pseudo spectrum of the whole time-frequency matrix to a frequency domain direction through a projection method to obtain a projection frequency spectrum, and carrying out priori-free blade inherent frequency identification based on projection by combining a conclusion of existence of a spectral line to obtain blade kinetic parameters. According to the method, the frequency identification problem of timing under-sampling signals of the lower blade end by only adopting a single sensor is solved.

Description

Technical Field

[0001] The invention belongs to the field of non-contact state monitoring of rotor blades, and in particular relates to a single-sensor rotor blade frequency identification method based on a projection method. Background Art

[0002] Rotor blades are key components in aircraft engines and gas turbines. Since rotor blades are in direct contact with the outside air, foreign object damage often occurs, which seriously affects the safety of high-speed rotor equipment. Therefore, it is very important to carry out rotor blade condition monitoring. Blade tip timing technology is a non-contact blade testing method. Compared with the traditional strain gauge and adhesive acceleration sensor testing method, it has the following three advantages: 1) A single sensor can monitor all blades at the same time; 2) Non-contact measurement does not affect the performance of the blade itself; 3) No wires are required on the blades or rotors, which will not cause motion interference or affect equipment performance. The main principle of blade tip timing technology is to collect the arrival pulses of the blades and speed marks during the rotation of the rotor, record the arrival time of the pulses, and convert the arrival time into blade vibration displacement in combination with the speed. The blade condition is monitored by extracting the characteristics of the blade vibration displacement.

[0003] In blade tip timing technology, accurate identification of rotor blade frequency is crucial to assessing the health status of the blades. Blade frequency reflects the working status and potential structural problems of the blades, and can provide an important basis for early fault detection. However, the difficulty of frequency identification mainly comes from the non-uniformity of blade tip timing sampling, spectrum analysis aliasing and noise interference, which greatly increases the complexity of accurately extracting the blade's natural frequency. A single blade tip timing sensor can achieve the lowest system installation cost but also corresponds to the most serious signal undersampling problem. Research on frequency identification methods for single sensors is still in its infancy, and there are problems of relying on frequency domain prior information and redundant analysis strategies.

[0004] Therefore, the present invention studies the frequency identification of the blade tip timing signal under a single sensor, and proposes a single-sensor rotor blade frequency identification method based on the projection method. Without the need for frequency domain prior information, the time-frequency analysis and frequency domain projection method under variable rate sampling are utilized to realize the frequency identification of the under-sampled signal under a single blade tip timing sensor without prior information. Summary of the invention

[0005] Technical issues to be solved:

[0006] In order to avoid the shortcomings of the prior art, the present invention provides a single-sensor rotor blade frequency identification method based on the projection method. Without the need for frequency domain prior information, the method uses time-frequency analysis and frequency domain projection under variable rate sampling to achieve frequency identification of undersampled signals under a single blade tip timing sensor without prior information. This method solves the frequency identification problem of blade tip timing undersampled signals using only a single sensor. Compared with the existing methods, it does not require complex optimization solutions and transformations, has strong scalability, can achieve non-contact monitoring of key state parameters of rotor blades, and has certain practical value.

[0007] The technical solution of the present invention is: a single sensor rotor blade frequency identification method based on projection method, the specific steps are as follows:

[0008] Based on the blade-end timing single sensor sampling mechanism, the signal is acquired, and the aliasing mechanism of the blade-end timing single sensor spectrum is analyzed for the acquired signal to determine whether undersampling occurs or not, there must be a spectral line at the corresponding position of the real frequency in the signal in the spectrum;

[0009] The original signal is obtained by performing variable rate sampling through the blade end timing single sensor;

[0010] The original signal obtained by single sensor variable rate sampling is subjected to time-frequency analysis by multiple signal classification method to obtain a time-frequency matrix.

[0011] Each pseudo-spectrum of the entire time-frequency matrix is ​​projected into the frequency domain through the projection method to obtain the projection spectrum. Combined with the conclusion of the existence of the spectrum line, the blade natural frequency is identified without prior knowledge based on the projection to obtain the blade dynamic parameters.

[0012] A further technical solution of the present invention is that the blade tip timing single sensor sampling mechanism is dual rate sampling or variable rate sampling.

[0013] A further technical solution of the present invention is that the aliasing mechanism analysis method of the blade tip timing single sensor spectrum is a least squares fitting method, and the expression is as follows:

[0014] x=A T (AA T ) -1 y

[0015] Among them, x is the frequency coefficient obtained by least squares, y is the blade end timing sampling signal, A is the overcomplete basis matrix set according to the sampling time, frequency range and frequency resolution, and each frequency point corresponds to two columns of sine and cosine vectors in A.

[0016] A further technical solution of the present invention is: the constraint conditions of the blade tip timing single sensor sampling are:

[0017] Ignoring blade vibration, the ratio of rotor circumference to the number of blades is expressed as follows:

[0018] d<<2πR / N b

[0019] Where d represents the blade vibration displacement, R represents the rotor radius, N b Indicates the number of rotor blades;

[0020] The rotor frequency is less than the blade natural frequency, and the expression is as follows:

[0021] f r <f n

[0022] Among them, f r Indicates the rotor frequency, f n represents the blade natural frequency.

[0023] A further technical solution of the present invention is: the steps of the multiple signal classification method are:

[0024] Taking a single signal point in the original signal as the time delay of each snapshot, a snapshot matrix of the signal is established, and singular value decomposition is performed on it to obtain singular values ​​and singular vectors;

[0025] Arrange the decomposed singular values ​​from large to small, and perform subspace division according to the set subspace division threshold, where the singular values ​​greater than the subspace division threshold are divided into signal subspace, and those less than the subspace division threshold are divided into noise subspace;

[0026] The steering vector is set according to the actual sampling time, the orthogonality between the steering vector and the noise subspace is calculated, and the pseudo amplitude is obtained. The formula is as follows:

[0027]

[0028] Among them, p(f) is the pseudo amplitude corresponding to f, c(f) is the steering vector corresponding to f, and G is the noise subspace obtained according to the set subspace partition threshold after singular value decomposition.

[0029] A further technical solution of the present invention is: the time-frequency analysis is an overlapping sliding window analysis, and an overlapping sliding window analysis is performed on the original signal, and the middle sampling moment of each analysis segment is used as the corresponding moment of the pseudo-spectrum of the segment, and the pseudo-spectra obtained at different time points are stacked according to time to form a time-frequency matrix.

[0030] A further technical solution of the present invention is: the pseudo spectrum is constructed by setting the frequency domain resolution and the upper frequency limit to obtain a frequency traversal range, and the pseudo spectrum is constructed by the pseudo amplitudes of all frequency points within the range.

[0031] A further technical solution of the present invention is: the expression of the projection spectrum is as follows:

[0032]

[0033] Where P is the projected spectrum, p n is the spectrum of the nth signal segment, and N is the total number of signal segments.

[0034] A further technical solution of the present invention is: the method for identifying the natural frequency is: a maximum value search method, based on the conclusion that there is always a spectrum line at any sampling frequency of the true frequency, the true frequency, that is, the aliasing frequency corresponding to the natural frequency should be the maximum value after projection accumulation; the expression is:

[0035]

[0036] Among them, f n is the blade natural frequency.

[0037] A single-sensor rotor blade frequency identification system based on a projection method includes a single-sensor sampling module, a multiple signal classification module, a time-frequency analysis module and a projection module;

[0038] The single sensor sampling module performs timed and variable rate sampling on the blade end to obtain the original signal;

[0039] The original signal is divided into signal subspace and noise subspace through the multiple signal classification module, and the pseudo amplitude is obtained by calculation to construct the pseudo spectrum;

[0040] The original signal is analyzed with overlapping sliding windows through the time-frequency analysis module to obtain the time-frequency matrix;

[0041] The projection spectrum is obtained through the projection module, and then the blade natural frequency is identified and the blade dynamic parameters are obtained.

[0042] Beneficial Effects

[0043] The beneficial effects of the present invention are as follows: in the present invention, the aliasing mechanism of the blade tip timing single sensor spectrum analysis is first analyzed, and the occurrence position law of the aliasing frequency is obtained under the classical uniform sampling theory, and the conclusion that the aliasing frequency always exists at the real frequency is obtained by changing different sampling rates; this mechanism explores the aliasing law under different sampling frequencies, and can theoretically provide a basis for the signal analysis of single or multiple blade tip timing sensors, thereby reducing the difficulty of signal analysis.

[0044] Then, combined with the condition that the sampling frequency of the blade tip timing is equal to the rotation frequency under the single sensor layout, the variable rate sampling condition is constructed based on the variable speed condition, so that the theoretical aliasing mechanism is mapped to the actual blade tip timing sampling; this step builds a bridge between the theoretical aliasing mechanism and the actual blade tip timing sampling, and only uses a single sensor, which minimizes the equipment installation and manufacturing costs.

[0045] The blade tip timing signal collected by a single sensor is taken as the analysis object. The sampling signal under variable speed is processed by sliding window, and multiple signal classification analysis is carried out on each signal segment to obtain the spectrum, and the spectrum is further organized into a time-frequency matrix. This step converts the blade signal into an easily observable frequency domain through the strategy of time-frequency analysis, so as to realize the horizontal comparison of the aliasing change law under different sampling rates.

[0046] Finally, according to the acquired time-frequency matrix, each spectrum vector is superimposed and projected onto the frequency axis, and the peak value of the superimposed projection is determined as the natural frequency of the blade to realize the extraction of the blade vibration parameters. This step only requires an accumulation operation, and only the maximum value needs to be paid attention to when extracting the frequency. The calculation strategy is simple and the complexity is low, and it can realize the anti-aliasing extraction of the blade natural frequency to reflect the blade state.

[0047] The method of the present invention can realize the frequency identification of rotor blades through time-frequency analysis of multiple signal classification method under variable rate, based on sampling of single blade tip timing sensor. This method has three advantages: first, only a single blade tip timing sensor is required, which minimizes the cost of monitoring equipment and installation cost; second, only time-frequency analysis and projection operation are required for the blades, and the frequency domain analysis is performed by the multiple signal classification method that has been more maturely applied in this field. The projection operation is relatively simple, so the calculation complexity is low, the method is easy to understand, and it is convenient to use in engineering; finally, no prior information about the blade frequency is required, and the variable rate sampling aliasing mechanism is introduced, and the blade frequency can be obtained to reflect the blade status by only changing the working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a method flow chart of a single-sensor rotor blade frequency identification method based on a projection method in an embodiment of the present invention;

[0049] Figure 2 It is a blade tip timing sensor installation position diagram of a single sensor rotor blade frequency identification method based on a projection method in an embodiment of the present invention;

[0050] Figure 3 It is a time domain diagram of a blade tip timing single sensor sampling signal of a single sensor rotor blade frequency identification method based on a projection method in an embodiment of the present invention;

[0051] Figure 4It is a time-frequency distribution diagram of a sampling signal of a blade tip timing single sensor of a single sensor rotor blade frequency identification method based on a projection method in an embodiment of the present invention;

[0052] Figure 5 It is a projection diagram corresponding to the sampling signal of a blade tip timing single sensor in a single sensor rotor blade frequency identification method based on a projection method in an embodiment of the present invention.

[0053] Explanation of the reference numerals: 1. Blade tip timing sensor, 2. Key phase sensor. DETAILED DESCRIPTION

[0054] The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0055] Based on the problem that a large number of sensors, complex algorithms and frequency domain priors are required when identifying blade tip timing under-sampling signals in the prior art, the present invention provides a single-sensor rotor blade frequency identification method based on a projection method, and the specific steps are as follows:

[0056] Step 1: Acquire the signal based on the blade-end timing single sensor sampling mechanism, and perform the aliasing mechanism analysis of the blade-end timing single sensor spectrum on the acquired signal to determine whether undersampling occurs or not, there must be a spectral line at the corresponding position of the real frequency in the signal in the spectrum;

[0057] Assume that the blade vibration displacement is negligible compared to the ratio of the rotor circumference to the number of blades, as shown below:

[0058] d<<2πR / N b

[0059] Where d represents the blade vibration displacement, R represents the rotor radius, and N b represents the number of rotor blades, and assumes that the rotor speed f r Less than the blade natural frequency f n ;

[0060] Specifically, the blade end timing single sensor sampling mechanism is dual rate sampling or variable rate sampling.

[0061] Specifically, the aliasing mechanism analysis method of the blade tip timing single sensor spectrum is the least squares fitting method, and the expression is as follows:

[0062] x=A T (AA T )- 1 y

[0063] Among them, x is the frequency coefficient obtained by least squares, y is the blade tip timing sampling signal, A is the overcomplete basis matrix set according to the sampling time, frequency range and frequency resolution, and each frequency point corresponds to two columns of sine and cosine vectors in A. At this time, it can be determined that there are always spectral lines at the real frequency in the aliased spectrum at different sampling rates.

[0064] Step 2: Perform variable rate sampling through a blade end timing single sensor to obtain an original signal; the variable rate sampling may be irregular speed variation;

[0065] Step 3: Perform time-frequency analysis on the original signal obtained by single sensor variable rate sampling through multiple signal classification method to obtain a time-frequency matrix;

[0066] The frequency domain analysis of the multiple signal classification method is as follows: a single signal point is used as the time delay of each snapshot, a snapshot matrix of the signal is established, and singular value decomposition is performed on it to obtain singular values ​​and singular vectors; the decomposed singular values ​​are arranged from large to small, and the subspace is divided by the set subspace division threshold. In this embodiment, the signal subspace and the noise subspace are divided according to the size of the singular value at a ratio of 95%, and the steering vector is set according to the actual sampling time, and the orthogonality of the steering vector and the noise subspace is calculated to obtain the pseudo amplitude, as shown in the following formula:

[0067]

[0068] Among them, p(f) and c(f) are the pseudo amplitude and steering vector corresponding to f, and G is the noise subspace obtained according to the set threshold after singular value decomposition;

[0069] The time-frequency analysis strategy is an overlapping sliding window analysis, the step size of each sliding window in the time domain is 50% of the number of analysis points each time, the middle sampling moment of each analysis segment is used as the corresponding moment of the pseudo-spectrum of the segment, and the pseudo-spectra obtained at different time points are stacked according to time to form a time-frequency matrix. The pseudo-spectrum construction method is to set the frequency domain resolution and the upper frequency limit to obtain the range of frequency traversal, and the pseudo-spectrum is composed of the pseudo-amplitudes of all frequency points within the range.

[0070] Step 4: According to the time-frequency diagram of the multiple signal classification method, a projection-based method for identifying blade natural frequencies without prior knowledge is established. Each spectrum of the entire time-frequency matrix is ​​projected into the frequency domain. Mathematically, it can be directly obtained by accumulating the spectra at different times obtained after the time-frequency analysis of the variable rate signal, as shown in the following formula:

[0071]

[0072] Where P is the projected spectrum, p n is the spectrum of the nth signal segment, and N is the total number of segments;

[0073] Then, through the maximum value search method, according to the conclusion that there is always a spectrum line at any sampling frequency, the true frequency, that is, the aliasing frequency corresponding to the natural frequency, should be the maximum value after projection accumulation; the expression is as follows:

[0074]

[0075] The following is combined with Figure 1-5 The technical solution of the present invention is described in further detail.

[0076] Reference Figure 1 As shown, the present embodiment provides a single sensor rotor blade frequency identification method based on a projection method, comprising the following steps:

[0077] S1: Based on the blade tip timing sampling mechanism, the aliasing mechanism of blade tip timing single sensor spectrum analysis is analyzed;

[0078] In this embodiment, the blade disk material is Al 6061, the diameter is 136mm, the blade height is 48mm, the width is 20mm, and the thickness is 1mm. According to the results of finite element analysis and the experimental results of multiple sensors in the past, it can be known that the natural frequency of the blade exceeds 300Hz, and the corresponding required sampling frequency is 600Hz. In the case of a single sensor, it is converted to a speed of 36000rpm, and the actual speed is much lower than this speed. Therefore, aliasing is bound to occur, and it is necessary to utilize it through the aliasing mechanism.

[0079] S2: Propose a variable rate sampling strategy for blade tip timing single sensor;

[0080] In this embodiment, if Figure 2 As shown in the figure, it is the position arrangement diagram of the sampling of a single blade tip timing sensor, where the blade tip timing sensor monitors the vibration of all blades at the same time, and the sampling displacement is once per rotation. Therefore, for the analyzed blade, the sampling rate is equal to the rotation frequency; Figure 3 As shown, the black line is the frequency change line, the speed sampling is set to linear speed change, the starting frequency is 62.5Hz, the ending frequency is 78Hz, and the duration is 40s; the blue line is the blade vibration displacement sampled by sensor No. 1, during which the blade experiences a resonance at about 35s.

[0081] S3: Establish a multiple signal classification time-frequency analysis method for single sensor sampling signals;

[0082] In this embodiment, the window length for time-frequency analysis is set to be, that is, the number of signal segments analyzed each time, the overlap of two adjacent signal segments is, for the multiple signal classification method to analyze the spectrum, the snapshot vector length is set to 600, the number of snapshots is set to 100, the frequency resolution is set to 1 Hz, the upper limit of frequency analysis is set to 500 Hz, and a pseudo spectrum is obtained. The time-frequency analysis results are output in the form of a matrix, and each column represents the spectrum of each signal segment. The obtained time-frequency diagram is as follows Figure 4 shown.

[0083] S4: According to the time-frequency diagram of the multiple signal classification method, a projection-based and non-a priori blade natural frequency identification method is established to obtain the blade dynamic parameters;

[0084] In this embodiment, the spectrum is accumulated along time according to the time-frequency matrix obtained in the previous time-frequency analysis. In this example, the average operation is performed after the accumulation, which only affects the vertical axis height of the spectrum and does not affect the identification result of the blade frequency. The final identified blade frequency is 338 Hz, which corresponds to the signals collected by multiple sensors and the finite element simulation results.

[0085] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and intent of the present invention.

Claims

1. A single sensor rotor blade frequency identification method based on projection method, characterized in that The specific steps are as follows: Based on the blade-end timing single sensor sampling mechanism, the signal is acquired, and the aliasing mechanism of the blade-end timing single sensor spectrum is analyzed for the acquired signal to determine whether undersampling occurs or not, there must be a spectral line at the corresponding position of the real frequency in the signal in the spectrum; The original signal is obtained by performing variable rate sampling through the blade end timing single sensor; The original signal obtained by single sensor variable rate sampling is subjected to time-frequency analysis by multiple signal classification method to obtain a time-frequency matrix. Each pseudo-spectrum of the entire time-frequency matrix is ​​projected into the frequency domain through the projection method to obtain the projection spectrum. Combined with the conclusion of the existence of the spectrum line, the blade natural frequency is identified without prior knowledge based on the projection to obtain the blade dynamic parameters.

2. According to claim 1, a single sensor rotor blade frequency identification method based on projection method is characterized by: The blade end timing single sensor sampling mechanism is dual rate sampling or variable rate sampling.

3. The single sensor rotor blade frequency identification method based on projection method according to claim 2 is characterized by: The aliasing mechanism analysis method of the blade tip timing single sensor spectrum is the least squares fitting method, and the expression is as follows: x=A T (AA T )- 1 y Among them, x is the frequency coefficient obtained by least squares, y is the blade end timing sampling signal, A is the overcomplete basis matrix set according to the sampling time, frequency range and frequency resolution, and each frequency point corresponds to two columns of sine and cosine vectors in A.

4. The single sensor rotor blade frequency identification method based on projection method according to claim 3 is characterized by: The constraint conditions for the blade tip timing single sensor sampling are: Ignoring blade vibration, the ratio of rotor circumference to the number of blades is expressed as follows: d<2πR / N Where d represents the blade vibration displacement, R represents the rotor radius, N b Indicates the number of rotor blades; The rotor frequency is less than the blade natural frequency, and the expression is as follows: f r <f n Among them, f r Indicates the rotor frequency, f n represents the blade natural frequency.

5. The single sensor rotor blade frequency identification method based on projection method according to claim 1 is characterized by: The steps of the multiple signal classification method are: Taking a single signal point in the original signal as the time delay of each snapshot, a snapshot matrix of the signal is established, and singular value decomposition is performed on it to obtain singular values ​​and singular vectors; Arrange the decomposed singular values ​​from large to small, and perform subspace division according to the set subspace division threshold, where the singular values ​​greater than the subspace division threshold are divided into signal subspace, and those less than the subspace division threshold are divided into noise subspace; The steering vector is set according to the actual sampling time, the orthogonality between the steering vector and the noise subspace is calculated, and the pseudo amplitude is obtained. The formula is as follows: Among them, p(f) is the pseudo amplitude corresponding to f, c(f) is the steering vector corresponding to f, and G is the noise subspace obtained according to the set subspace partition threshold after singular value decomposition.

6. The single sensor rotor blade frequency identification method based on projection method according to claim 5 is characterized by: The time-frequency analysis is an overlapping sliding window analysis, which performs overlapping sliding window analysis on the original signal, takes the middle sampling moment of each analysis segment as the corresponding moment of the pseudo-spectrum of the segment, and stacks the pseudo-spectra obtained at different time points according to time to form a time-frequency matrix.

7. The single sensor rotor blade frequency identification method based on projection method according to claim 6 is characterized by: The pseudo spectrum is constructed by setting the frequency domain resolution and the upper frequency limit to obtain a frequency traversal range, and the pseudo spectrum is constructed by the pseudo amplitudes of all frequency points within the range.

8. The single sensor rotor blade frequency identification method based on projection method according to claim 7 is characterized by: The expression of the projection spectrum is as follows: Where P is the projected spectrum, p n is the spectrum of the nth signal segment, and N is the total number of signal segments.

9. The single sensor rotor blade frequency identification method based on projection method according to claim 8, characterized in that: The method for identifying the natural frequency is the maximum value search method. According to the conclusion that there is always a spectrum line at any sampling frequency of the true frequency, the true frequency, that is, the aliasing frequency corresponding to the natural frequency, should be the maximum value after projection accumulation; the expression is: Among them, f n is the blade natural frequency.

10. A single-sensor rotor blade frequency identification system based on projection method, used to implement the single-sensor rotor blade frequency identification method based on projection method according to any one of claims 1 to 9; characterized in that: It includes a single sensor sampling module, a multiple signal classification module, a time-frequency analysis module and a projection module; The single sensor sampling module performs timed and variable rate sampling on the blade end to obtain the original signal; The original signal is divided into signal subspace and noise subspace through the multiple signal classification module, and the pseudo amplitude is obtained by calculation to construct the pseudo spectrum; The original signal is analyzed with overlapping sliding windows through the time-frequency analysis module to obtain the time-frequency matrix; The projection spectrum is obtained through the projection module, and then the blade natural frequency is identified and the blade dynamic parameters are obtained.

Images